graphics: rewrite screenshot and api capture image processing (#870 )

## Link to GitHub Issue or related Pull Request, if one exists #0 ## Description of change Significantly speeds up API screen capture and D3D9 screenshots saving. Two reasons for doing this: 1. We now have a 4K game (GITADORA) and existing capture code was taking multiple seconds. 2. Renewed user interest on streaming as we have a couple more companion apps in active development. **API screen capture (streaming), 1280x720:** 14.3ms -> 6.3ms per frame. Back buffer copies go to pooled `D3DPOOL_SYSTEMMEM` surfaces via `GetRenderTargetData` instead of allocating a lockable render target every frame, and TooJpeg is replaced with libjpeg-turbo (encode 9.8ms -> 3.0ms). MSAA remains unsupported **Screenshots for GITADORA arena model, across 4 screens with one of them 4K**: 4068ms -> 124ms. `D3DXSaveSurfaceToFileA` is replaced with fpng (encode 4043ms -> 76ms) and the screens encode in parallel. Dropping D3DX also removes the `d3dx9_43.dll` ... `d3dx9_24.dll` probing loop, so screenshots no longer fail outright on machines with no D3DX9 runtime installed. Screenshot surfaces are read on the present thread, so no D3D call reaches another thread for screenshots. This fixes a hang in DDR X2 introduced earlier in the branch: its device has no internal locking, and reading the surface on a pool thread while the present thread sat inside `GetRenderTargetData` left the game's own render thread deadlocked. ## Testing - **GITADORA** (arena model, D3D9Ex, 4K main plus three subscreens, windowed) with `-screenshotsub`: three sets of four screenshots, images verified correct. Completion order differs between sets, so the screens really are encoding in parallel. - **LovePlus** (KLP, plain D3D9, 768x1360): covers the inline path used by games whose image processing must not leave the present thread. - **API screen capture** through a companion app: live video correct throughout. - **Print Screen** bound as the screenshot key: the clipboard copy succeeded on every shot. - Quitting the game after capturing leaves no `IDirect3DDevice9` reference count warning, so the pooled readback surfaces are released along with the device.
misc: various clean up for diagnosing launch failures (#872 )
2026-10-06 01:58:13 -07:00 · 2026-08-18 00:22:45 -07:00 · 2026-08-17 22:41:07 -07:00 · 2026-08-16 21:27:03 -07:00 · 2026-08-16 21:23:41 -07:00 · 2026-08-16 04:24:14 -07:00
559 changed files with 153616 additions and 28861 deletions
@@ -0,0 +1,46 @@
+# Normalize every text file to LF in the repository.
+# The build runs under Linux/MinGW containers, so LF is also used in the
+# working tree; Windows editors and toolchains handle LF fine.
+* text=auto eol=lf
+
+# Windows-only files that must keep CRLF in the working tree.
+*.bat        text eol=crlf
+*.cmd        text eol=crlf
+*.sln        text eol=crlf
+*.vcproj     text eol=crlf
+*.vcxproj    text eol=crlf
+*.props      text eol=crlf
+*.filters    text eol=crlf
+
+# Files that must keep LF even if a Windows editor rewrites them.
+*.sh         text eol=lf
+*.in         text eol=lf
+*.cmake      text eol=lf
+*.mk         text eol=lf
+Makefile     text eol=lf
+Dockerfile   text eol=lf
+
+# Binary files - never touch the contents.
+*.bin        binary
+*.ico        binary
+*.ttf        binary
+*.otf        binary
+*.png        binary
+*.jpg        binary
+*.jpeg       binary
+*.gif        binary
+*.bmp        binary
+*.zip        binary
+*.7z         binary
+*.gz         binary
+*.dll        binary
+*.exe        binary
+*.lib        binary
+*.a          binary
+*.o          binary
+*.obj        binary
+*.pdb        binary
+
+# Vendored code is stored and checked out byte-for-byte as upstream ships it,
+# so re-importing a library never produces line-ending-only diffs.
+src/spice2x/external/** -text
@@ -11,9 +11,7 @@ assignees: ''
 *name of game, version of game*

 ## Version of spice2x
-*version of spice, this can be seen in the About tab, and in log.txt*
-
-*please remember to test the LATEST beta version before posting a bug about it**
+*please remember to test the LATEST beta version before posting a bug about it*

 ## Describe the issue
 *what's the issue?*
@@ -22,3 +20,7 @@ assignees: ''
 > [!WARNING]
 > Please make sure you remove any personally identifiable information from the log file.
 > Please set `-loglevel` to `all` before launching the game for more verbose logs.
+
+## A note on issue management
+
+`Closed as not planned` means exactly that - it's closed, because it's not on our plans to address them any time soon. It doesn't mean your issue is invalid, it just means it's not a priority, and we can always revisit them later.
@@ -1,32 +1,32 @@
-on: [push, pull_request]
-name: Continuous Integration
-concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
-  cancel-in-progress: true
-
-jobs:
-  fw-ci:
-    name: Build
-    runs-on: ubuntu-latest
-    defaults:
-      run:
-        working-directory: ./src/spice2x
-    steps:
-      - uses: actions/checkout@v4
-      - name: Set ccache environment variables
-        run: |
-          echo "CCACHE_DIR=${{ github.workspace }}/src/spice2x/.ccache" >> $GITHUB_ENV
-      - name: Install ccache
-        uses: hendrikmuhs/ccache-action@v1.2
-      - name: Calculate commit SHA
-        id: vars
-        run: |
-          calculatedSha=$(git rev-parse --short ${{ github.sha }})
-          echo "COMMIT_SHORT_SHA=$calculatedSha" >> $GITHUB_ENV
-      - name: Compile
-        run: ./build_docker.sh
-      - uses: actions/upload-artifact@v4
-        with:
-          name: spice2x-ci-${{ env.COMMIT_SHORT_SHA }}
-          path: src/spice2x/bin
+on: [push, pull_request]
+name: Continuous Integration
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  fw-ci:
+    name: Build
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: ./src/spice2x
+    steps:
+      - uses: actions/checkout@v5
+      - name: Set ccache environment variables
+        run: |
+          echo "CCACHE_DIR=${{ github.workspace }}/src/spice2x/.ccache" >> $GITHUB_ENV
+      - name: Install ccache
+        uses: hendrikmuhs/ccache-action@v1.2.23
+      - name: Calculate commit SHA
+        id: vars
+        run: |
+          calculatedSha=$(git rev-parse --short ${{ github.sha }})
+          echo "COMMIT_SHORT_SHA=$calculatedSha" >> $GITHUB_ENV
+      - name: Compile
+        run: ./build_docker.sh
+      - uses: actions/upload-artifact@v6
+        with:
+          name: spice2x-ci-${{ env.COMMIT_SHORT_SHA }}
+          path: src/spice2x/bin
          if-no-files-found: error
@@ -0,0 +1,49 @@
+name: Draft Release
+
+on:
+  workflow_dispatch:
+
+permissions:
+  contents: write
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  release:
+    name: Build and Draft Release
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: ./src/spice2x
+    steps:
+      - uses: actions/checkout@v5
+        with:
+          ref: main
+          fetch-depth: 0
+      - name: Clean leftover build artifacts
+        run: |
+          rm -rf .ccache dist bin cmake-build-*
+      - name: Compile
+        run: ./build_docker.sh
+      - name: Determine release name from dist filename
+        run: |
+          dist=$(basename "$(ls dist/spice2x-*.zip | grep -v -- '-full.zip')")
+          # strip the ".zip" to get the base name, e.g. spice2x-26-06-28
+          name="${dist%.zip}"
+          # the tag is the date portion, e.g. 26-06-28
+          tag="${name#spice2x-}"
+          echo "RELEASE_NAME=$name" >> $GITHUB_ENV
+          echo "RELEASE_TAG=$tag" >> $GITHUB_ENV
+      - name: Create draft release
+        uses: softprops/action-gh-release@v3
+        with:
+          draft: true
+          prerelease: true
+          tag_name: ${{ env.RELEASE_TAG }}
+          name: ${{ env.RELEASE_NAME }}
+          target_commitish: main
+          generate_release_notes: true
+          files: |
+            src/spice2x/dist/spice2x-*.zip
@@ -11,13 +11,17 @@ To contribute, fork the repo (just the main branch), make changes in your fork,
 * For a new game that was never supported (not a new version of a game, but rather a new series): please wait 1 year after official AC release in Japan.
 * For new version of an already supported game: proceed with caution and use generally-accepted community guidelines.

-### Forbidden "features" that will never be accepted
+### Pull requests that will not be accepted

-* Performing decryption of encrypted files
-* Including (bundling), distributing, or linking to game data, or making it easier to find/download game data
+* Performing decryption of encrypted files.
+* Including (bundling), distributing, or linking to game data, or making it easier to find/download game data.
 * Any features that [phone home](https://en.wikipedia.org/wiki/Phoning_home)
  * This includes automatic updaters.
  * Exception: we have allowed the "patch import from URL" feature, but this has only been done under very careful considerations and plenty of warnings to the user.
+* Built-in EA server that saves data. They should live in a separate project.
+* Any additions to built-in patches (modifications / removals are OK if we have a good reason).
+* Changes to the software license terms.
+* Localization / translation for UI text. It becomes too difficult to manage by the maintainers.

 ### Avoiding regressions

@@ -235,6 +235,13 @@ add_compile_definitions(
        _WIN32_IE=0x0400
 )

+# SPICE_XP: set by build_all.sh for the WinXP-compat toolchains. Gates out
+# the DX11 overlay backend
+option(SPICE_XP "Build for the WinXP-compat toolchain" OFF)
+if(SPICE_XP)
+    add_compile_definitions(SPICE_XP=1)
+endif()
+
 # acioemu log
 #set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DACIOEMU_LOG")

@@ -249,6 +256,28 @@ add_subdirectory(external/imgui EXCLUDE_FROM_ALL)
 add_subdirectory(external/minhook EXCLUDE_FROM_ALL)
 add_subdirectory(external/cpu_features EXCLUDE_FROM_ALL)

+# libjpeg-turbo's SIMD needs nasm; enable_language belongs at top level. Without
+# it the library still builds, just without SIMD.
+include(CheckLanguage)
+check_language(ASM_NASM)
+if(CMAKE_ASM_NASM_COMPILER)
+    enable_language(ASM_NASM)
+else()
+    message(WARNING
+            "nasm not found: libjpeg-turbo will be built without SIMD, "
+            "which makes api screen capture several times slower")
+endif()
+add_subdirectory(external/libjpeg-turbo EXCLUDE_FROM_ALL)
+
+# fpng's SIMD needs the whole unit built for SSE4.1, which its runtime CPU check
+# cannot undo, so keep it scalar rather than raising the CPU baseline
+set_source_files_properties(external/fpng/fpng.cpp PROPERTIES
+        COMPILE_DEFINITIONS "FPNG_NO_SSE=1")
+if(NOT MSVC)
+    set_source_files_properties(external/fpng/fpng.cpp PROPERTIES
+            COMPILE_OPTIONS "-fno-strict-aliasing")
+endif()
+
 # set link time optimizations (disabled for Debug builds for speed, disabled
 # for RelWithDebInfo builds due to "lto1: error: two or more sections for"
 # errors)
@@ -266,6 +295,7 @@ set_source_files_properties(cfg/manifest.rc PROPERTIES LANGUAGE RC)
 set_source_files_properties(cfg/icon.rc PROPERTIES LANGUAGE RC)
 set_source_files_properties(cfg/Win32D.rc PROPERTIES LANGUAGE RC)
 set_source_files_properties(build/manifest64.rc PROPERTIES LANGUAGE RC)
+set_source_files_properties(stubs/manifest.rc PROPERTIES LANGUAGE RC)

 # sources
 #########
@@ -324,6 +354,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
        api/modules/control.cpp
        api/modules/touch.cpp
        api/modules/iidx.cpp
+        api/modules/sdvx.cpp
        api/serial.cpp
        api/modules/drs.cpp
        api/modules/lcd.cpp
@@ -371,7 +402,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
        external/tinyxml2/tinyxml2.cpp
        external/http-parser/http_parser.c
        external/usbhidusage/usb-hid-usage.c
-        external/toojpeg/toojpeg.cpp
+        external/fpng/fpng.cpp
        external/scard/scard.cpp

        # games
@@ -382,6 +413,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
        games/shared/twtouch.cpp
        games/popn/popn.cpp
        games/popn/io.cpp
+        games/popn/bi3a_hook.cpp
        games/bbc/bbc.cpp
        games/bbc/io.cpp
        games/hpm/hpm.cpp
@@ -399,14 +431,19 @@ set(SOURCE_FILES ${SOURCE_FILES}
        games/iidx/mf_wrappers.cpp
        games/sdvx/bi2x_hook.cpp
        games/sdvx/sdvx.cpp
+        games/sdvx/sdvx_live2d.cpp
        games/sdvx/io.cpp
        games/sdvx/camera.cpp
+        games/jb/bi2x_hook.cpp
        games/jb/jb.cpp
+        games/jb/jb_touch.cpp
        games/jb/io.cpp
        games/nost/nost.cpp
        games/nost/io.cpp
        games/nost/poke.cpp
+        games/nost/touch_mode.cpp
        games/gitadora/gitadora.cpp
+        games/gitadora/asio.cpp
        games/gitadora/io.cpp
        games/gitadora/handle.cpp
        games/gitadora/j32d.cpp
@@ -420,6 +457,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
        games/rb/rb.cpp
        games/rb/io.cpp
        games/rb/touch.cpp
+        games/rb/touch_debug.cpp
        games/bs/bs.cpp
        games/bs/io.cpp
        games/rf3d/rf3d.cpp
@@ -484,16 +522,24 @@ set(SOURCE_FILES ${SOURCE_FILES}
        # hooks
        hooks/audio/acm.cpp
        hooks/audio/audio.cpp
+        hooks/audio/asio_driver_scan.cpp
+        hooks/audio/asio_proxy.cpp
        hooks/audio/buffer.cpp
        hooks/audio/mme.cpp
        hooks/audio/util.cpp
+        hooks/audio/xact.cpp
        hooks/audio/backends/dsound/dsound_backend.cpp
        hooks/audio/backends/mmdevice/audio_endpoint_volume.cpp
        hooks/audio/backends/mmdevice/device.cpp
        hooks/audio/backends/mmdevice/device_collection.cpp
        hooks/audio/backends/mmdevice/device_enumerator.cpp
+        hooks/audio/backends/mmdevice/null_device.cpp
+        hooks/audio/backends/mmdevice/null_discard_backend.cpp
        hooks/audio/backends/wasapi/audio_client.cpp
        hooks/audio/backends/wasapi/audio_render_client.cpp
+        hooks/audio/backends/wasapi/downmix.cpp
+        hooks/audio/backends/wasapi/resample.cpp
+        hooks/audio/backends/wasapi/shared.cpp
        hooks/audio/backends/wasapi/dummy_audio_client.cpp
        hooks/audio/backends/wasapi/dummy_audio_clock.cpp
        hooks/audio/backends/wasapi/dummy_audio_render_client.cpp
@@ -508,19 +554,32 @@ set(SOURCE_FILES ${SOURCE_FILES}
        hooks/devicehook.cpp
        hooks/graphics/graphics.cpp
        hooks/graphics/graphics_windowed.cpp
+        hooks/graphics/jpeg_encoder.cpp
+        hooks/graphics/nvapi_impl.cpp
        hooks/graphics/nvapi_hook.cpp
        hooks/graphics/nvenc_hook.cpp
        hooks/graphics/backends/d3d9/d3d9_backend.cpp
+        hooks/graphics/backends/d3d9/d3d9_readback.cpp
+        hooks/graphics/backends/d3d9/d3d9_screenshot.cpp
        hooks/graphics/backends/d3d9/d3d9_device.cpp
+        hooks/graphics/backends/d3d9/d3d9_gfdm.cpp
+        hooks/graphics/backends/d3d9/d3d9_live2d.cpp
        hooks/graphics/backends/d3d9/d3d9_fake_swapchain.cpp
        hooks/graphics/backends/d3d9/d3d9_swapchain.cpp
        hooks/graphics/backends/d3d9/d3d9_texture.cpp
+        hooks/graphics/backends/d3d11/d3d11_backend.cpp
+        hooks/graphics/backends/d3d11/d3d11_swapchain.cpp
+        hooks/graphics/backends/d3d11/d3d11_factory.cpp
+        hooks/graphics/backends/d3d11/d3d11_vtable_capture.cpp
+        hooks/graphics/backends/d3d11/d3d11_screenshot.cpp
        hooks/input/dinput8/fake_backend.cpp
        hooks/input/dinput8/fake_device.cpp
        hooks/input/dinput8/hook.cpp
        hooks/lang.cpp
        hooks/libraryhook.cpp
        hooks/networkhook.cpp
+        hooks/icmphook_net.cpp
+        hooks/icmphook_iphlpapi.cpp
        hooks/powrprof.cpp
        #hooks/rom.cpp
        hooks/setupapihook.cpp
@@ -543,7 +602,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
        misc/device.cpp
        misc/eamuse.cpp
        misc/extdev.cpp
-        misc/nativetouchhook.cpp
+        misc/hotkeys.cpp
        misc/sciunit.cpp
        misc/sde.cpp
        misc/wintouchemu.cpp
@@ -554,6 +613,7 @@ set(SOURCE_FILES ${SOURCE_FILES}

        # overlay
        overlay/overlay.cpp
+        overlay/notifications.cpp
        overlay/window.cpp
        overlay/imgui/extensions.cpp
        overlay/imgui/impl_spice.cpp
@@ -580,11 +640,27 @@ set(SOURCE_FILES ${SOURCE_FILES}
        overlay/windows/keypad.cpp
        overlay/windows/log.cpp
        overlay/windows/midi.cpp
+        overlay/windows/nostalgia_touch_piano.cpp
+        overlay/windows/obs.cpp
+        overlay/windows/obs_websocket.cpp
        overlay/windows/patch_manager.cpp
+        overlay/windows/popn_sub.cpp
        overlay/windows/wnd_manager.cpp

+        # patcher
+        patcher/config.cpp
+        patcher/lifecycle.cpp
+        patcher/loader.cpp
+        patcher/loader_group.cpp
+        patcher/runtime.cpp
+
+        # external
+        external/easywsclient/easywsclient.cpp
+
        # rawinput
        rawinput/rawinput.cpp
+        rawinput/rawinput_handles.cpp
+        rawinput/midi.cpp
        rawinput/sextet.cpp
        rawinput/piuio.cpp
        rawinput/touch.cpp
@@ -595,6 +671,8 @@ set(SOURCE_FILES ${SOURCE_FILES}
        rawinput/smxstage.h
        rawinput/smxdedicab.cpp
        rawinput/smxdedicab.h
+        rawinput/xinput.cpp
+        rawinput/xinput.h

        # reader
        reader/reader.cpp
@@ -602,12 +680,21 @@ set(SOURCE_FILES ${SOURCE_FILES}
        reader/structuredmessage.cpp
        reader/crypt.cpp

+        # sdk
+        sdk/sdk.cpp
+
        # stubs
        stubs/stubs.cpp

        # touch
+        touch/gdi_overlay.cpp
+        touch/native/inject.cpp
+        touch/native/inject_mouse.cpp
+        touch/native/inject_synthetic.cpp
+        touch/native/nativetouchhook.cpp
+        touch/native/transform.cpp
        touch/touch.cpp
-        touch/touch_indicators.cpp
+        touch/touch_gestures.cpp
        touch/win7.cpp
        touch/win8.cpp

@@ -628,6 +715,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
        util/time.cpp
        util/cpuutils.cpp
        util/netutils.cpp
+        util/precise_timer.cpp
        util/sysutils.cpp
        util/lz77.cpp
        util/tapeled.cpp
@@ -639,37 +727,74 @@ set(SOURCE_FILES ${SOURCE_FILES}

 source_group(TREE ${CMAKE_CURRENT_SOURCE_DIR} PREFIX "Source Files" FILES ${SOURCE_FILES})

+# guard against statically importing DLLs that must always be loaded dynamically:
+#   * DLLs users override via the modules directory (e.g. DXVK's d3d9.dll) - a
+#     static import loads the system copy at startup and preempts the override
+#     (issue #779).
+#   * Media Foundation DLLs (mf/mfplat/mfreadwrite) - a static import breaks
+#     Unity games.
+# the check runs objdump on each produced binary and fails the build if any
+# forbidden DLL is imported.
+set(SPICE_FORBIDDEN_STATIC_IMPORTS
+        d3d8.dll d3d9.dll d3d10core.dll d3d11.dll dxgi.dll opengl32.dll
+        mf.dll mfplat.dll mfreadwrite.dll)
+
+function(spice_guard_dll_imports target)
+    if(MSVC)
+        # objdump-based parsing assumes a GNU/LLVM toolchain
+        return()
+    endif()
+    add_custom_command(TARGET ${target} POST_BUILD
+            COMMAND ${CMAKE_COMMAND}
+                -DOBJDUMP=${CMAKE_OBJDUMP}
+                -DTARGET_FILE=$<TARGET_FILE:${target}>
+                "-DFORBIDDEN=${SPICE_FORBIDDEN_STATIC_IMPORTS}"
+                -P ${CMAKE_CURRENT_SOURCE_DIR}/cmake/check_no_static_dll_imports.cmake
+            VERBATIM
+            COMMENT "Checking ${target} for forbidden static DLL imports")
+endfunction()
+
+# spice.exe / spice_laa.exe shared objects
+###########################################
+# spice.exe and spice_laa.exe are compiled identically; the only difference is
+# the large-address-aware bit, which is set at link time. Compile the sources
+# once into a shared OBJECT library so spice_laa.exe only costs an extra link
+# instead of a full recompile.
+
+add_library(spicetools_spice_objs OBJECT ${SOURCE_FILES})
+target_link_libraries(spicetools_spice_objs
+        PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp
+        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static)
+target_link_libraries(spicetools_spice_objs PUBLIC winscard)
+
+if(NOT MSVC)
+    set_target_properties(spicetools_spice_objs PROPERTIES COMPILE_FLAGS "-m32")
+endif()
+
 # spice.exe
 ###########

 set(RESOURCE_FILES build/manifest.manifest build/manifest.rc build/icon.rc cfg/Win32D.rc)
-add_executable(spicetools_spice ${SOURCE_FILES} ${RESOURCE_FILES})
-target_link_libraries(spicetools_spice
-        PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp
-        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features)
-target_link_libraries(spicetools_spice PUBLIC winscard)
+add_executable(spicetools_spice ${RESOURCE_FILES})
+target_link_libraries(spicetools_spice PRIVATE spicetools_spice_objs)
 set_target_properties(spicetools_spice PROPERTIES PREFIX "")
 set_target_properties(spicetools_spice PROPERTIES OUTPUT_NAME "spice")

 IF(NOT MSVC)
-    set_target_properties(spicetools_spice PROPERTIES COMPILE_FLAGS "-m32" LINK_FLAGS "-m32")
+    set_target_properties(spicetools_spice PROPERTIES LINK_FLAGS "-m32")
 endif()

 # spice_laa.exe
 ###########

 set(RESOURCE_FILES build/manifest.manifest build/manifest.rc build/icon.rc cfg/Win32D.rc)
-add_executable(spicetools_spice_laa ${SOURCE_FILES} ${RESOURCE_FILES})
-target_link_libraries(spicetools_spice_laa
-        PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp
-        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features)
-target_link_libraries(spicetools_spice_laa PUBLIC winscard)
+add_executable(spicetools_spice_laa ${RESOURCE_FILES})
+target_link_libraries(spicetools_spice_laa PRIVATE spicetools_spice_objs)
 set_target_properties(spicetools_spice_laa PROPERTIES PREFIX "")
 set_target_properties(spicetools_spice_laa PROPERTIES OUTPUT_NAME "spice_laa")
-target_compile_definitions(spicetools_spice_laa PRIVATE SPICE32_LARGE_ADDRESS_AWARE=1)

 IF(NOT MSVC)
-    set_target_properties(spicetools_spice_laa PROPERTIES COMPILE_FLAGS "-m32" LINK_FLAGS "-m32 -Wl,--large-address-aware")
+    set_target_properties(spicetools_spice_laa PROPERTIES LINK_FLAGS "-m32 -Wl,--large-address-aware")
 endif()

 # spice_linux.exe
@@ -679,7 +804,7 @@ set(RESOURCE_FILES build/manifest.manifest build/manifest.rc build/icon.rc cfg/W
 add_executable(spicetools_spice_linux ${SOURCE_FILES} ${RESOURCE_FILES})
 target_link_libraries(spicetools_spice_linux
        PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp
-        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features)
+        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static)
 set_target_properties(spicetools_spice_linux PROPERTIES PREFIX "")
 set_target_properties(spicetools_spice_linux PROPERTIES OUTPUT_NAME "spice_linux")
 target_compile_definitions(spicetools_spice_linux PRIVATE NO_SCARD=1 PRIVATE SPICE_LINUX=1)
@@ -697,11 +822,15 @@ add_executable(spicetools_spice64 ${SOURCE_FILES} ${RESOURCE_FILES})
 # do NOT link against: mf, mfplat, mfreadwrite; otherwise unity games will break
 target_link_libraries(spicetools_spice64
        PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp mfuuid strmiids dxva2
-        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features)
+        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static)
 target_link_libraries(spicetools_spice64 PUBLIC winscard)
 set_target_properties(spicetools_spice64 PROPERTIES PREFIX "")
 set_target_properties(spicetools_spice64 PROPERTIES OUTPUT_NAME "spice64")
 target_compile_definitions(spicetools_spice64 PRIVATE SPICE64=1)
+if(NOT SPICE_XP)
+    target_link_libraries(spicetools_spice64 PUBLIC d3d11 dxgi)
+    target_compile_definitions(spicetools_spice64 PRIVATE SPICE_D3D11=1)
+endif()

 IF(NOT MSVC)
    set_target_properties(spicetools_spice64 PROPERTIES COMPILE_FLAGS "-m64" LINK_FLAGS "-m64")
@@ -716,11 +845,14 @@ add_executable(spicetools_spice64_linux ${SOURCE_FILES} ${RESOURCE_FILES})
 # do NOT link against: mf, mfplat, mfreadwrite; otherwise unity games will break
 target_link_libraries(spicetools_spice64_linux
        PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp mfuuid strmiids dxva2
-        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features)
+        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static)
 set_target_properties(spicetools_spice64_linux PROPERTIES PREFIX "")
 set_target_properties(spicetools_spice64_linux PROPERTIES OUTPUT_NAME "spice64_linux")
 target_compile_definitions(spicetools_spice64_linux PRIVATE SPICE64=1)
 target_compile_definitions(spicetools_spice64_linux PRIVATE NO_SCARD=1 PRIVATE SPICE_LINUX=1)
+# spice64_linux is never built under the WinXP toolchain, so dx11 is always on.
+target_link_libraries(spicetools_spice64_linux PUBLIC d3d11 dxgi)
+target_compile_definitions(spicetools_spice64_linux PRIVATE SPICE_D3D11=1)

 IF(NOT MSVC)
    set_target_properties(spicetools_spice64_linux PROPERTIES COMPILE_FLAGS "-m64" LINK_FLAGS "-m64")
@@ -733,8 +865,8 @@ set(SOURCE_FILES ${SOURCE_FILES} launcher/options.h launcher/options.cpp)
 set(RESOURCE_FILES cfg/manifest.manifest cfg/manifest.rc cfg/icon.rc cfg/Win32D.rc)
 add_executable(spicetools_cfg WIN32 ${SOURCE_FILES} ${RESOURCE_FILES})
 target_link_libraries(spicetools_cfg
-        PUBLIC ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp strmiids
-        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features)
+        PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp strmiids
+        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static)
 target_link_libraries(spicetools_cfg PUBLIC winscard)
 set_target_properties(spicetools_cfg PROPERTIES PREFIX "")
 set_target_properties(spicetools_cfg PROPERTIES OUTPUT_NAME "spicecfg")
@@ -751,8 +883,8 @@ set(SOURCE_FILES ${SOURCE_FILES} launcher/options.h launcher/options.cpp)
 set(RESOURCE_FILES cfg/manifest.manifest cfg/manifest.rc cfg/icon.rc cfg/Win32D.rc)
 add_executable(spicetools_cfg_linux WIN32 ${SOURCE_FILES} ${RESOURCE_FILES})
 target_link_libraries(spicetools_cfg_linux
-        PUBLIC ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp strmiids
-        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features)
+        PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp strmiids
+        PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static)
 set_target_properties(spicetools_cfg_linux PROPERTIES PREFIX "")
 set_target_properties(spicetools_cfg_linux PROPERTIES OUTPUT_NAME "spicecfg_linux")
 target_compile_definitions(spicetools_cfg_linux PRIVATE SPICETOOLS_SPICECFG_STANDALONE=1)
@@ -813,7 +945,8 @@ endif()

 # nvcuda.dll
 set(SOURCE_FILES stubs/nvcuda.cpp)
-add_library(spicetools_stubs_nvcuda SHARED ${SOURCE_FILES} stubs/nvcuda.def)
+set(RESOURCE_FILES stubs/manifest.rc)
+add_library(spicetools_stubs_nvcuda SHARED ${SOURCE_FILES} ${RESOURCE_FILES} stubs/nvcuda.def)
 set_target_properties(spicetools_stubs_nvcuda PROPERTIES PREFIX "")
 set_target_properties(spicetools_stubs_nvcuda PROPERTIES OUTPUT_NAME "nvcuda")

@@ -823,7 +956,8 @@ endif()

 # nvcuvid.dll
 set(SOURCE_FILES stubs/nvcuvid.cpp)
-add_library(spicetools_stubs_nvcuvid SHARED ${SOURCE_FILES} stubs/nvcuvid.def)
+set(RESOURCE_FILES stubs/manifest.rc)
+add_library(spicetools_stubs_nvcuvid SHARED ${SOURCE_FILES} ${RESOURCE_FILES} stubs/nvcuvid.def)
 set_target_properties(spicetools_stubs_nvcuvid PROPERTIES PREFIX "")
 set_target_properties(spicetools_stubs_nvcuvid PROPERTIES OUTPUT_NAME "nvcuvid")

@@ -833,7 +967,8 @@ endif()

 # nvEncodeAPI64.dll
 set(SOURCE_FILES stubs/nvEncodeAPI64.cpp)
-add_library(spicetools_stubs_nvEncodeAPI64 SHARED ${SOURCE_FILES} stubs/nvEncodeAPI64.def)
+set(RESOURCE_FILES stubs/manifest.rc)
+add_library(spicetools_stubs_nvEncodeAPI64 SHARED ${SOURCE_FILES} ${RESOURCE_FILES} stubs/nvEncodeAPI64.def)
 set_target_properties(spicetools_stubs_nvEncodeAPI64 PROPERTIES PREFIX "")
 set_target_properties(spicetools_stubs_nvEncodeAPI64 PROPERTIES OUTPUT_NAME "nvEncodeAPI64")

@@ -843,7 +978,8 @@ endif()

 # cpusbxpkm.dll (32 bit)
 set(SOURCE_FILES stubs/cpusbxpkm.cpp)
-add_library(spicetools_stubs_cpusbxpkm SHARED ${SOURCE_FILES} stubs/cpusbxpkm.def)
+set(RESOURCE_FILES stubs/manifest.rc)
+add_library(spicetools_stubs_cpusbxpkm SHARED ${SOURCE_FILES} ${RESOURCE_FILES} stubs/cpusbxpkm.def)
 set_target_properties(spicetools_stubs_cpusbxpkm PROPERTIES PREFIX "")
 set_target_properties(spicetools_stubs_cpusbxpkm PROPERTIES OUTPUT_NAME "cpusbxpkm")

@@ -851,6 +987,36 @@ if(NOT MSVC)
    set_target_properties(spicetools_stubs_cpusbxpkm PROPERTIES COMPILE_FLAGS "-m32" LINK_FLAGS "-m32")
 endif()

+# sdk_sample_v0_flat_c.dll (32 bit)
+set(SOURCE_FILES sdk/sample/v0/flat_c/v0_flat_c.c)
+add_library(spicetools_sdk_sample_v0_flat_c_32 SHARED ${SOURCE_FILES} ${RESOURCE_FILES} sdk/sample/v0/flat_c/v0_flat_c.def)
+set_target_properties(spicetools_sdk_sample_v0_flat_c_32 PROPERTIES PREFIX "")
+set_target_properties(spicetools_sdk_sample_v0_flat_c_32 PROPERTIES OUTPUT_NAME "sdk_sample_v0_flat_c")
+
+if(NOT MSVC)
+    set_target_properties(spicetools_sdk_sample_v0_flat_c_32 PROPERTIES COMPILE_FLAGS "-m32" LINK_FLAGS "-m32")
+endif()
+
+# sdk_sample_v0_flat_c.dll (64 bit)
+set(SOURCE_FILES sdk/sample/v0/flat_c/v0_flat_c.c)
+add_library(spicetools_sdk_sample_v0_flat_c_64 SHARED ${SOURCE_FILES} ${RESOURCE_FILES} sdk/sample/v0/flat_c/v0_flat_c.def)
+set_target_properties(spicetools_sdk_sample_v0_flat_c_64 PROPERTIES PREFIX "")
+set_target_properties(spicetools_sdk_sample_v0_flat_c_64 PROPERTIES OUTPUT_NAME "sdk_sample_v0_flat_c")
+
+if(NOT MSVC)
+    set_target_properties(spicetools_sdk_sample_v0_flat_c_64 PROPERTIES COMPILE_FLAGS "-m64" LINK_FLAGS "-m64")
+endif()
+
+# sdk_sample_v0_cpp.dll (64 bit)
+set(SOURCE_FILES sdk/sample/v0/cpp/v0_cpp.cpp)
+add_library(spicetools_sdk_sample_v0_cpp_64 SHARED ${SOURCE_FILES} ${RESOURCE_FILES} sdk/sample/v0/cpp/v0_cpp.def)
+set_target_properties(spicetools_sdk_sample_v0_cpp_64 PROPERTIES PREFIX "")
+set_target_properties(spicetools_sdk_sample_v0_cpp_64 PROPERTIES OUTPUT_NAME "sdk_sample_v0_cpp")
+
+if(NOT MSVC)
+    set_target_properties(spicetools_sdk_sample_v0_cpp_64 PROPERTIES COMPILE_FLAGS "-m64" LINK_FLAGS "-m64")
+endif()
+
 # output directories
 ####################

@@ -860,15 +1026,28 @@ set_target_properties(spicetools_cfg spicetools_cfg_linux
        RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/spicetools")

 # output 32bit
-set_target_properties(spicetools_spice spicetools_spice_laa spicetools_spice_linux spicetools_stubs_kbt spicetools_stubs_kld spicetools_stubs_cpusbxpkm
+set_target_properties(spicetools_spice spicetools_spice_laa spicetools_spice_linux spicetools_stubs_kbt spicetools_stubs_kld spicetools_stubs_cpusbxpkm spicetools_sdk_sample_v0_flat_c_32
        PROPERTIES
        ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/archive32"
        LIBRARY_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/spicetools/32"
        RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/spicetools/32")

 # output 64bit
-set_target_properties(spicetools_spice64 spicetools_spice64_linux spicetools_stubs_kbt64 spicetools_stubs_kld64 spicetools_stubs_nvcuda spicetools_stubs_nvcuvid spicetools_stubs_nvEncodeAPI64
+set_target_properties(spicetools_spice64 spicetools_spice64_linux spicetools_stubs_kbt64 spicetools_stubs_kld64 spicetools_stubs_nvcuda spicetools_stubs_nvcuvid spicetools_stubs_nvEncodeAPI64 spicetools_sdk_sample_v0_flat_c_64 spicetools_sdk_sample_v0_cpp_64
        PROPERTIES
        ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/archive64"
        LIBRARY_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/spicetools/64"
        RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}/spicetools/64")
+
+# forbidden static DLL import guard
+###################################
+# apply the check to every executable and DLL produced by this project.
+get_property(spice_all_targets DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTY BUILDSYSTEM_TARGETS)
+foreach(spice_target IN LISTS spice_all_targets)
+    get_target_property(spice_target_type ${spice_target} TYPE)
+    if(spice_target_type STREQUAL "EXECUTABLE"
+            OR spice_target_type STREQUAL "SHARED_LIBRARY"
+            OR spice_target_type STREQUAL "MODULE_LIBRARY")
+        spice_guard_dll_imports(${spice_target})
+    endif()
+endforeach()
@@ -267,6 +267,20 @@ which also means that your hex edits are applicable directly.
    - `Side Panel Right Inner`
    - `Side Panel Right`

+#### SDVX
+- tapeled_get(name: str, ...)
+  - returns a list containing a dict of the current tape LED states. The dict keys are:
+    - `Title`
+    - `Upper Left Speaker`
+    - `Upper Right Speaker`
+    - `Left Wing`
+    - `Right Wing`
+    - `Control Panel`
+    - `Lower Left Speaker`
+    - `Lower Right Speaker`
+    - `Woofer`
+    - `V Unit`
+
 #### LCD
 - info()
  - returns information about the serial LCD controller some games use
@@ -283,6 +297,16 @@ Python is the only one that is fully spec compliant.
 Other libraries may be missing features and contain bugs; please feel free to
 contribute code to fill the gaps if you work on a project using these libraries.

+## Spice SDK (for DLL hooks / plugins)
+
+As an alternative to Spice API, Spice SDK is available.
+
+This is a flat C, header-only library that can be included in your DLL. Once
+initialized, you can call directly into helper routines provided by Spicetools
+executable, making low-latency and high-throughput interactions possible.
+
+Check the documentation on the wiki for more details.
+
 ## License
 Unless otherwise noted, all files are licensed under the GPLv3.
 See the LICENSE file for the full license text.
@@ -4,6 +4,7 @@
 #include "launcher/launcher.h"
 #include "misc/wintouchemu.h"
 #include "rawinput/rawinput.h"
+#include "touch/native/nativetouchhook.h"

 // static stuff
 static int ACIO_WARMUP = 0;
@@ -140,8 +141,9 @@ static int __cdecl ac_io_update(int a1) {
    // flush device output
    RI_MGR->devices_flush_output();

-    // update wintouchemu
+    // update touch emulation
    wintouchemu::update();
+    nativetouch::refresh_contact_lifetime();

    return 1;
 }
@@ -35,24 +35,33 @@ struct ICCA_STATUS_LA9 {
 static_assert(sizeof(struct ICCA_STATUS) == 24, "ICCA_STATUS must be 24 bytes");

 enum ICCA_WORKFLOW {
-    STEP,
-    SLEEP,
-    START,
-    INIT,
-    READY,
-    GET_USERID,
-    ACTIVE,
-    EJECT,
-    EJECT_CHECK,
-    END,
-    CLOSE_EJECT,
-    CLOSE_E_CHK,
-    CLOSE_END,
-    ERR_GETUID = -2
+    STEP = 0,
+    SLEEP = 1,
+    START = 2,
+    INIT = 3,
+    READY = 4,
+    GET_USERID = 5,
+    ACTIVE = 6,
+    EJECT = 7,
+    EJECT_CHECK = 8,
+    END = 9,
+    CLOSE_EJECT = 10,
+    CLOSE_E_CHK = 11,
+    CLOSE_END = 12,
+    ERR_GETUID = -2,
+    ERR_REMOVED = -11
 };
+
+// used by RA as the status_code
+enum ICCA_STATUS_CODE {
+    EMPTY = 1,
+    ENGAGED = 2
+};
+
 struct ICCA_UNIT {
    struct ICCA_STATUS status {};
    enum ICCA_WORKFLOW state = STEP;
+    enum ICCA_WORKFLOW state_jdz = SLEEP;
    bool card_cmd_pressed = false;
    bool card_in = false;
    double card_in_time = 0.0;
@@ -75,6 +84,11 @@ static inline int icca_get_active_count() {
 }

 static inline int icca_get_unit_id(int unit_id) {
+    // on RA the unit_id is zero-indexed
+    if (avs::game::is_model("JDZ")) {
+        return unit_id;
+    }
+
    if (icca_get_active_count() < 2)
        return 1;
    else {
@@ -122,9 +136,15 @@ static inline void update_card(int unit_id) {
                    IS_LAST_CARD_FELICA = is_card_uid_felica(unit->status.uid);

                    unit->state = acio::ICCA_COMPAT_ACTIVE ? START : ACTIVE;
+                    unit->state_jdz = ACTIVE;
                    unit->status.error = 0;
+                    unit->status.front_sensor = 1;
+                    unit->status.rear_sensor = 1;
                } else {
                    unit->state = ERR_GETUID;
+                    unit->state_jdz = ERR_GETUID;
+                    unit->status.front_sensor = 1;
+                    unit->status.rear_sensor = 1;
                    memset(unit->status.uid, 0, 8);
                }
                unit->card_in = true;
@@ -132,10 +152,25 @@ static inline void update_card(int unit_id) {
            } else if (unit->state == EJECT_CHECK) {
                unit->state = SLEEP;
                unit->card_in = false;
+                unit->status.front_sensor = 0;
+                unit->status.rear_sensor = 0;
+            } else if (unit->state_jdz == READY) {
+                unit->state = GET_USERID;
+                unit->state_jdz = GET_USERID;
+                unit->status.error = 0;
+                unit->status.front_sensor = 1;
+                unit->status.rear_sensor = 1;
+                unit->card_in = true;
+                unit->card_in_time = get_performance_seconds();
            }
        } else {
            unit->state = acio::ICCA_COMPAT_ACTIVE ? START : ACTIVE;
+            if (unit->state_jdz == READY) {
+                unit->state_jdz = GET_USERID;
+            }
            unit->status.error = 0;
+            unit->status.front_sensor = 1;
+            unit->status.rear_sensor = 1;
            unit->card_in = true;
            unit->card_in_time = get_performance_seconds();
        }
@@ -144,6 +179,12 @@ static inline void update_card(int unit_id) {
        unit->state = CLOSE_EJECT;
        if (fabs(get_performance_seconds() - unit->card_in_time) > CARD_TIMEOUT) {
            unit->card_in = false;
+            memset(unit->status.uid, 0, 8);
+            if (unit->state_jdz == ACTIVE || unit->state_jdz == GET_USERID) {
+                unit->state_jdz = ERR_REMOVED;
+            }
+            unit->status.front_sensor = 0;
+            unit->status.rear_sensor = 0;
        }
    }

@@ -302,7 +343,11 @@ static char __cdecl ac_io_icca_get_status(void *a1, void *a2) {

    // copy state to output buffer
    ICCA_UNIT *unit = &ICCA_UNITS[unit_id];
-    unit->status.status_code = unit->state;
+    if (avs::game::is_model("JDZ")) {
+        unit->status.status_code = unit->card_in ? ENGAGED : EMPTY;
+    } else {
+        unit->status.status_code = unit->state;
+    }
    memcpy(status, &unit->status, sizeof(struct ICCA_STATUS));

    // funny workaround
@@ -387,6 +432,11 @@ static char __cdecl ac_io_icca_req_uid(int unit_id) {
 static int __cdecl ac_io_icca_req_uid_isfinished(int unit_id, DWORD *read_state) {
    unit_id = icca_get_unit_id(unit_id);
    ICCA_UNIT *unit = &ICCA_UNITS[unit_id];
+    if (avs::game::is_model("JDZ")) {
+        // hack for RA - only ever seen it as this
+        *read_state = 8;
+        return 1;
+    }
    if (unit->card_in) {
        if (fabs(get_performance_seconds() - unit->card_in_time) < CARD_TIMEOUT) {
            unit->state = END;
@@ -403,16 +453,61 @@ static int __cdecl ac_io_icca_send_keep_alive_packet(int a1, int a2, int a3) {
    return 0;
 }

+static int __cdecl ac_io_icca_workflow_jdz(int workflow, int unit_id) {
+
+    unit_id = icca_get_unit_id(unit_id);
+    ICCA_UNIT *unit = &ICCA_UNITS[unit_id];
+    switch (workflow) {
+        // RA uses "STEP" as more of a polling thing it seems
+        case STEP:
+            switch (unit->state_jdz) {
+                case CLOSE_EJECT:
+                    unit->state_jdz = CLOSE_E_CHK;
+                    break;
+                case CLOSE_E_CHK:
+                    unit->state_jdz = CLOSE_END;
+                    break;
+                case CLOSE_END:
+                    unit->state_jdz = SLEEP;
+                    break;
+                default:
+                    break;
+            }
+            break;
+        case SLEEP:
+            unit->state = SLEEP;
+            break;
+        case START:
+            unit->state_jdz = INIT;
+            break;
+        case INIT:
+        case READY:
+            unit->state_jdz = READY;
+            break;
+        case CLOSE_EJECT:
+            unit->state_jdz = CLOSE_EJECT;
+            return CLOSE_E_CHK;
+        case CLOSE_END:
+            unit->state_jdz = SLEEP;
+            break;
+        default:
+            // should probably log a warning here
+            break;
+    }
+    return unit->state_jdz;
+}
+
 static int __cdecl ac_io_icca_workflow(int workflow, int unit_id) {
+    // RA uses a different workflow process for the slotted readers; treat it differently
+    if (avs::game::is_model("JDZ")) {
+        return ac_io_icca_workflow_jdz(workflow, unit_id);
+    }

    unit_id = icca_get_unit_id(unit_id);
    ICCA_UNIT *unit = &ICCA_UNITS[unit_id];
    switch (workflow) {
        case STEP:
-            if (avs::game::is_model("JDZ"))
-                unit->state = SLEEP;
-            else
-                unit->state = STEP;
+            unit->state = STEP;
            break;
        case SLEEP:
            unit->state = SLEEP;
@@ -441,7 +536,6 @@ static int __cdecl ac_io_icca_workflow(int workflow, int unit_id) {
        default:
            break;
    }
-
    return unit->state;
 }

@@ -467,6 +561,14 @@ acio::ICCAModule::ICCAModule(HMODULE module, acio::HookMode hookMode) : ACIOModu
 void acio::ICCAModule::attach() {
    ACIOModule::attach();

+    // workaround for RA; ac_io_icca_cardunit_init is never called so treat like both are initialized
+    if (avs::game::is_model("JDZ")) {
+        ICCA_UNITS[0].initialized = true;
+        ICCA_UNITS[1].initialized = true;
+
+        CARD_TIMEOUT = 10.0; // keep the card in the slot for longer
+    }
+
    // hooks
    ACIO_MODULE_HOOK(ac_io_icca_cardunit_init);
    ACIO_MODULE_HOOK(ac_io_icca_cardunit_init_isfinished);
@@ -1,13 +1,16 @@
 #include "mdxf.h"
 #include "mdxf_poll.h"

+#include <mutex>
+
 #include "avs/game.h"
+#include "games/ddr/ddr.h"
 #include "games/ddr/io.h"
 #include "launcher/launcher.h"
 #include "rawinput/rawinput.h"
 #include "util/logging.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"
-#include <mutex>

 #define MDFX_DEBUG_VERBOSE 0

@@ -108,9 +111,10 @@ static void mdxf_thread_start() {
    MDXF_THREAD = std::thread([] {
        SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_BELOW_NORMAL);

+        timeutils::PreciseSleepTimer timer;
        while (MDXF_THREAD_RUNNING.load(std::memory_order_acquire)) {
            mdxf_poll(false);
-            std::this_thread::sleep_for(THREAD_PERIOD);
+            timer.sleep(THREAD_PERIOD);
        }
    });
 }
@@ -357,14 +361,17 @@ static bool __cdecl ac_io_mdxf_update_control_status_buffer_impl(int node, MDXFP

        // decide on button map
        const size_t *button_map = nullptr;
+        int player = 0;
        switch (node) {
            case 17:
            case 25:
                button_map = &buttons_p1[0];
+                player = 1;
                break;
            case 18:
            case 26:
                button_map = &buttons_p2[0];
+                player = 2;
                break;
        }
        
@@ -374,19 +381,27 @@ static bool __cdecl ac_io_mdxf_update_control_status_buffer_impl(int node, MDXFP
        if (source == EXTERNAL_POLL) {
            // get buttons
            auto &buttons = games::ddr::get_buttons();
+            
+            // get analogs
+            bool analog_left = false;
+            bool analog_right = false;
+            games::ddr::get_analog_x_axis(player, analog_left, analog_right);
+            bool analog_up = false;
+            bool analog_down = false;
+            games::ddr::get_analog_y_axis(player, analog_up, analog_down);
+            
            uint8_t up_down = 0;
            uint8_t left_right = 0;
-            
-            if (GameAPI::Buttons::getState(RI_MGR, buttons.at(button_map[0]))) {
+            if (GameAPI::Buttons::getState(RI_MGR, buttons.at(button_map[0])) || analog_up) {
                up_down |= 0xF0;
            }
-            if (GameAPI::Buttons::getState(RI_MGR, buttons.at(button_map[1]))) {
+            if (GameAPI::Buttons::getState(RI_MGR, buttons.at(button_map[1])) || analog_down) {
                up_down |= 0x0F;
            }
-            if (GameAPI::Buttons::getState(RI_MGR, buttons.at(button_map[2]))) {
+            if (GameAPI::Buttons::getState(RI_MGR, buttons.at(button_map[2])) || analog_left) {
                left_right |= 0xF0;
            }
-            if (GameAPI::Buttons::getState(RI_MGR, buttons.at(button_map[3]))) {
+            if (GameAPI::Buttons::getState(RI_MGR, buttons.at(button_map[3])) || analog_right) {
                left_right |= 0x0F;
            }
            current_state = (uint16_t(up_down) << 8) | left_right;
@@ -500,4 +515,4 @@ acio::MDXFModule::~MDXFModule() {
    if (IS_THREAD_NEEDED) {
        mdxf_thread_stop();
    }
-}
+}
@@ -3,6 +3,7 @@
 #include "rawinput/rawinput.h"
 #include "games/nost/io.h"
 #include "games/nost/nost.h"
+#include "games/nost/touch_mode.h"
 #include "util/logging.h"
 #include "avs/game.h"

@@ -16,6 +17,7 @@ using namespace GameAPI;
 // static stuff
 static uint8_t STATUS_BUFFER[277];
 static bool STATUS_BUFFER_FREEZE = false;
+static constexpr uint8_t NOST_TOUCH_PIANO_VELOCITY = 11;

 /*
 * Implementations
@@ -226,6 +228,8 @@ static bool __cdecl ac_io_panb_update_control_status_buffer() {
                games::nost::Analogs::Key27, games::nost::Analogs::Key28,
        };

+        const auto touch_key_state = games::nost::touch_mode::piano_key_state();
+
        // iterate pairs of keys
        for (size_t key_pair = 0; key_pair < 28 / 2; key_pair++) {

@@ -257,6 +261,10 @@ static bool __cdecl ac_io_panb_update_control_status_buffer() {
            if (button0 > 0) {
                state0 = button0;
            }
+            if ((touch_key_state & (UINT32_C(1) << (key_pair * 2))) &&
+                state0 < NOST_TOUCH_PIANO_VELOCITY) {
+                state0 = NOST_TOUCH_PIANO_VELOCITY;
+            }

            const auto button1 = panb_get_button_velocity(
                buttons.at(button_mapping[key_pair * 2 + 1]),
@@ -267,6 +275,10 @@ static bool __cdecl ac_io_panb_update_control_status_buffer() {
            if (button1 > 0) {
                state1 = button1;
            }
+            if ((touch_key_state & (UINT32_C(1) << (key_pair * 2 + 1))) &&
+                state1 < NOST_TOUCH_PIANO_VELOCITY) {
+                state1 = NOST_TOUCH_PIANO_VELOCITY;
+            }

            // build value
            uint8_t value = 0;
@@ -1,5 +1,6 @@
 #pragma once

+#include <cstdint>
 #include <string>

 #include "util/circular_buffer.h"
@@ -5,6 +5,7 @@
 #include "games/sdvx/sdvx.h"
 #include "misc/eamuse.h"
 #include "util/logging.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"

 using namespace acioemu;
@@ -46,11 +47,12 @@ ICCADevice::ICCADevice(bool flip_order, bool keypad_thread, uint8_t node_count)
    // keypad thread for faster polling
    if (keypad_thread) {
        this->keypad_thread = new std::thread([this]() {
+            timeutils::PreciseSleepTimer timer;
            while (this->cards) {
                for (int unit = 0; unit < this->node_count; unit++) {
                    this->update_keypad(unit, false);
                }
-                Sleep(7);
+                timer.sleep(7);
            }
        });
    }
@@ -237,7 +239,7 @@ bool ICCADevice::parse_msg(MessageData *msg_in,
            // SDVX Old cabinet mode
            if (avs::game::is_model("KFC") && avs::game::SPEC[0] != 'G' && avs::game::SPEC[0] != 'H')
                answer_type = 1;
-            if (avs::game::is_model("L44"))
+            if (avs::game::is_model({ "L44", "T44" }))
                answer_type = 2;

            // check answer type
@@ -0,0 +1,13 @@
+#pragma once
+
+#include <atomic>
+#include <cstdint>
+
+namespace api {
+
+    extern std::atomic_uint32_t CLIENT_COUNT;
+
+    inline bool has_clients() {
+        return CLIENT_COUNT.load(std::memory_order_relaxed) > 0;
+    }
+}
@@ -5,12 +5,15 @@

 #include <utility>

+#include "client.h"
 #include "cfg/configurator.h"
 #include "external/rapidjson/document.h"
 #include "util/crypt.h"
 #include "util/logging.h"
 #include "util/utils.h"

+#include "overlay/notifications.h"
+
 #include "module.h"
 #include "modules/analogs.h"
 #include "modules/buttons.h"
@@ -26,6 +29,7 @@
 #include "modules/lcd.h"
 #include "modules/lights.h"
 #include "modules/memory.h"
+#include "modules/sdvx.h"
 #include "modules/touch.h"
 #include "modules/resize.h"
 #include "request.h"
@@ -34,6 +38,8 @@
 using namespace rapidjson;
 using namespace api;

+std::atomic_uint32_t api::CLIENT_COUNT = 0;
+
 Controller::Controller(unsigned short port, std::string password, bool pretty)
    : port(port), password(std::move(password)), pretty(pretty)
 {
@@ -197,6 +203,9 @@ void Controller::connection_handler(api::ClientState client_state) {

    // log connection
    log_info("api", "client connected: {}", client_address_str);
+    overlay::notifications::add(
+            overlay::notifications::Severity::Success,
+            fmt::format("API client connected ({})", client_address_str));
    client_states_m.lock();
    client_states.emplace_back(&client_state);
    client_states_m.unlock();
@@ -277,6 +286,9 @@ void Controller::connection_handler(api::ClientState client_state) {

    // log disconnect
    log_info("api", "client disconnected: {}", client_address_str);
+    overlay::notifications::add(
+            overlay::notifications::Severity::Info,
+            fmt::format("API client disconnected ({})", client_address_str));
    client_states_m.lock();
    client_states.erase(std::remove(client_states.begin(), client_states.end(), &client_state));
    client_states_m.unlock();
@@ -403,8 +415,11 @@ void Controller::init_state(api::ClientState *state) {
    state->modules.push_back(new modules::LCD());
    state->modules.push_back(new modules::Lights());
    state->modules.push_back(new modules::Memory());
+    state->modules.push_back(new modules::SDVX());
    state->modules.push_back(new modules::Touch());
    state->modules.push_back(new modules::Resize());
+
+    CLIENT_COUNT.fetch_add(1, std::memory_order_relaxed);
 }

 void Controller::free_state(api::ClientState *state) {
@@ -416,6 +431,8 @@ void Controller::free_state(api::ClientState *state) {

    // free cipher
    delete state->cipher;
+
+    CLIENT_COUNT.fetch_sub(1, std::memory_order_relaxed);
 }

 void Controller::free_socket() {
@@ -6,6 +6,7 @@
 #include "launcher/launcher.h"
 #include "games/io.h"
 #include "util/utils.h"
+#include "acio/mdxf/mdxf_poll.h"

 using namespace std::placeholders;
 using namespace rapidjson;
@@ -104,6 +105,7 @@ namespace api::modules {
                for (auto &analog : *this->analogs) {
                    analog.override_enabled = false;
                }
+                mdxf_poll(true);
            }
            return;
        }
@@ -148,6 +150,7 @@ namespace api::modules {
            if (analog.getName() == name) {
                analog.override_state = CLAMP(state, 0.f, 1.f);
                analog.override_enabled = true;
+                mdxf_poll(true);
                return true;
            }
        }
@@ -167,6 +170,7 @@ namespace api::modules {
        for (auto &analog : *this->analogs) {
            if (analog.getName() == name) {
                analog.override_enabled = false;
+                mdxf_poll(true);
                return true;
            }
        }
@@ -6,6 +6,7 @@
 #include "launcher/launcher.h"
 #include "games/io.h"
 #include "util/utils.h"
+#include "acio/mdxf/mdxf_poll.h"

 using namespace std::placeholders;
 using namespace rapidjson;
@@ -107,6 +108,7 @@ namespace api::modules {
                for (auto &button : *this->buttons) {
                    button.override_enabled = false;
                }
+                mdxf_poll(true);
            }
            return;
        }
@@ -153,6 +155,7 @@ namespace api::modules {
                        GameAPI::Buttons::BUTTON_PRESSED : GameAPI::Buttons::BUTTON_NOT_PRESSED;
                button.override_velocity = CLAMP(state, 0.f, 1.f);
                button.override_enabled = true;
+                mdxf_poll(true);
                return true;
            }
        }
@@ -172,6 +175,7 @@ namespace api::modules {
        for (auto &button : *this->buttons) {
            if (button.getName() == name) {
                button.override_enabled = false;
+                mdxf_poll(true);
                return true;
            }
        }
@@ -1,5 +1,7 @@
 #include "capture.h"
 #include <functional>
+#include <mutex>
+#include <unordered_map>
 #include "external/rapidjson/document.h"
 #include "hooks/graphics/graphics.h"
 #include "util/crypt.h"
@@ -14,6 +16,56 @@ namespace api::modules {

    static thread_local std::vector<uint8_t> CAPTURE_BUFFER;

+    struct CachedFrame {
+        std::vector<uint8_t> jpeg;
+        uint64_t timestamp = 0;
+        int width = 0;
+        int height = 0;
+    };
+
+    static std::mutex FRAME_CACHE_M;
+    static std::unordered_map<int, CachedFrame> FRAME_CACHE;
+
+    static void add_jpeg_response(
+            int screen,
+            uint64_t timestamp,
+            int width,
+            int height,
+            const std::vector<uint8_t> &jpeg,
+            Response &res) {
+
+        auto encoded = crypt::base64_encode(jpeg.data(), jpeg.size());
+
+        Value data;
+        data.SetString(encoded.c_str(), encoded.length(), res.doc()->GetAllocator());
+        res.add_data(timestamp);
+        res.add_data(width);
+        res.add_data(height);
+        res.add_data(data);
+
+        std::lock_guard<std::mutex> lock(FRAME_CACHE_M);
+        FRAME_CACHE[screen] = {jpeg, timestamp, width, height};
+    }
+
+    static bool try_cached_response(int screen, Response &res) {
+        std::lock_guard<std::mutex> lock(FRAME_CACHE_M);
+        const auto pos = FRAME_CACHE.find(screen);
+        if (pos == FRAME_CACHE.end() || pos->second.jpeg.empty()) {
+            return false;
+        }
+
+        const auto &cached = pos->second;
+        auto encoded = crypt::base64_encode(cached.jpeg.data(), cached.jpeg.size());
+
+        Value data;
+        data.SetString(encoded.c_str(), encoded.length(), res.doc()->GetAllocator());
+        res.add_data(cached.timestamp);
+        res.add_data(cached.width);
+        res.add_data(cached.height);
+        res.add_data(data);
+        return true;
+    }
+
    Capture::Capture() : Module("capture") {
        functions["get_screens"] = std::bind(&Capture::get_screens, this, _1, _2);
        functions["get_jpg"] = std::bind(&Capture::get_jpg, this, _1, _2);
@@ -41,7 +93,6 @@ namespace api::modules {
     * reduce: uint for dividing image size
     */
    void Capture::get_jpg(Request &req, Response &res) {
-        CAPTURE_BUFFER.reserve(1024 * 128);

        // settings
        int screen = 0;
@@ -68,27 +119,17 @@ namespace api::modules {
        int width = 0;
        int height = 0;
        graphics_capture_trigger(screen);
-        bool success = graphics_capture_receive_jpeg(screen, [] (uint8_t byte) {
-            CAPTURE_BUFFER.push_back(byte);
-        }, true, quality, true, divide, &timestamp, &width, &height);
-        if (!success) {
+        bool success = graphics_capture_receive_jpeg(
+                screen, CAPTURE_BUFFER, quality, divide, &timestamp, &width, &height);
+
+        if (success) {
+            add_jpeg_response(screen, timestamp, width, height, CAPTURE_BUFFER, res);
+            CAPTURE_BUFFER.clear();
            return;
        }

-        // encode to base64
-        auto encoded = crypt::base64_encode(
-                CAPTURE_BUFFER.data(),
-                CAPTURE_BUFFER.size());
-
-        // clear buffer
+        // fall back to the last successful frame while the game is busy loading
        CAPTURE_BUFFER.clear();
-
-        // add data to response
-        Value data;
-        data.SetString(encoded.c_str(), encoded.length(), res.doc()->GetAllocator());
-        res.add_data(timestamp);
-        res.add_data(width);
-        res.add_data(height);
-        res.add_data(data);
+        try_cached_response(screen, res);
    }
 }
@@ -1,54 +1,54 @@
-#include "ddr.h"
-#include <functional>
-#include "external/rapidjson/document.h"
-#include "games/ddr/ddr.h"
-
-using namespace std::placeholders;
-using namespace rapidjson;
-
-namespace api::modules {
-
-    DDR::DDR() : Module("ddr") {
-        functions["tapeled_get"] = std::bind(&DDR::tapeled_get, this, _1, _2);
-    }
-
-    /**
-     * Allows fetching of the RGB LED strips that are gold cabinets, via SpiceAPI
-     */
-    void DDR::tapeled_get(Request &req, Response &res) {
-        static const char* device_names[11] = {
-            "p1_foot_up",
-            "p1_foot_right",
-            "p1_foot_left",
-            "p1_foot_down",
-            "p2_foot_up",
-            "p2_foot_right",
-            "p2_foot_left",
-            "p2_foot_down",
-            "top_panel",
-            "monitor_left",
-            "monitor_right"
-        };
-
-        Value response_object(kObjectType);
-
-        // Iterate through each device and dump its lights data into the response
-        for (size_t device = 0; device < 11; device++) {
-            size_t num_leds = 25;
-            if (device > 7)
-                num_leds = 50;
-
-            Value light_state(kArrayType);
-            light_state.Reserve(num_leds * 3, res.doc()->GetAllocator());
-            for (size_t led = 0; led < num_leds; led++) {
-                light_state.PushBack(games::ddr::DDR_TAPELEDS[device][led][0], res.doc()->GetAllocator());
-                light_state.PushBack(games::ddr::DDR_TAPELEDS[device][led][1], res.doc()->GetAllocator());
-                light_state.PushBack(games::ddr::DDR_TAPELEDS[device][led][2], res.doc()->GetAllocator());
-            }
-
-            response_object.AddMember(StringRef(device_names[device]), light_state, res.doc()->GetAllocator());
-        }
-
-        res.add_data(response_object);
-    }
-}
+#include "ddr.h"
+#include <functional>
+#include "external/rapidjson/document.h"
+#include "games/ddr/ddr.h"
+
+using namespace std::placeholders;
+using namespace rapidjson;
+
+namespace api::modules {
+
+    DDR::DDR() : Module("ddr") {
+        functions["tapeled_get"] = std::bind(&DDR::tapeled_get, this, _1, _2);
+    }
+
+    /**
+     * Allows fetching of the RGB LED strips that are gold cabinets, via SpiceAPI
+     */
+    void DDR::tapeled_get(Request &req, Response &res) {
+        static const char* device_names[11] = {
+            "p1_foot_up",
+            "p1_foot_right",
+            "p1_foot_left",
+            "p1_foot_down",
+            "p2_foot_up",
+            "p2_foot_right",
+            "p2_foot_left",
+            "p2_foot_down",
+            "top_panel",
+            "monitor_left",
+            "monitor_right"
+        };
+
+        Value response_object(kObjectType);
+
+        // Iterate through each device and dump its lights data into the response
+        for (size_t device = 0; device < 11; device++) {
+            size_t num_leds = 25;
+            if (device > 7)
+                num_leds = 50;
+
+            Value light_state(kArrayType);
+            light_state.Reserve(num_leds * 3, res.doc()->GetAllocator());
+            for (size_t led = 0; led < num_leds; led++) {
+                light_state.PushBack(games::ddr::DDR_TAPELEDS[device][led][0], res.doc()->GetAllocator());
+                light_state.PushBack(games::ddr::DDR_TAPELEDS[device][led][1], res.doc()->GetAllocator());
+                light_state.PushBack(games::ddr::DDR_TAPELEDS[device][led][2], res.doc()->GetAllocator());
+            }
+
+            response_object.AddMember(StringRef(device_names[device]), light_state, res.doc()->GetAllocator());
+        }
+
+        res.add_data(response_object);
+    }
+}
@@ -1,17 +1,17 @@
-#pragma once
-
-#include <vector>
-#include "api/module.h"
-#include "api/request.h"
-
-namespace api::modules {
-
-    class DDR : public Module {
-    public:
-        DDR();
-
-    private:
-        // function definitions
-        void tapeled_get(Request &req, Response &res);
-    };
-}
+#pragma once
+
+#include <vector>
+#include "api/module.h"
+#include "api/request.h"
+
+namespace api::modules {
+
+    class DDR : public Module {
+    public:
+        DDR();
+
+    private:
+        // function definitions
+        void tapeled_get(Request &req, Response &res);
+    };
+}
@@ -106,14 +106,16 @@ namespace api::modules {
    void IIDX::copy_tapeled_data(Response &res, Value &response_object, const tapeledutils::tape_led &mapping) {
        // Create an array for the light state
        Value light_state(kArrayType);
-        light_state.Reserve(mapping.data.capacity() * 3, res.doc()->GetAllocator());
+        light_state.Reserve(
+            static_cast<SizeType>(mapping.data.size() * 3),
+            res.doc()->GetAllocator());
        for (const auto [r, g, b] : mapping.data) {
            light_state.PushBack(r, res.doc()->GetAllocator());
            light_state.PushBack(g, res.doc()->GetAllocator());
            light_state.PushBack(b, res.doc()->GetAllocator());
        }

-        // Can't use StringRef here, turns some strings partially into null bytes for some reason
+        // can't use StringRef here, turns some strings partially into null bytes for some reason
        Value light_name(mapping.lightName.c_str(), res.doc()->GetAllocator());
        response_object.AddMember(light_name, light_state, res.doc()->GetAllocator());
    }
@@ -1,6 +1,8 @@
 #include "keypads.h"

+#include <chrono>
 #include <functional>
+#include <thread>

 #include <windows.h>

@@ -91,11 +93,11 @@ namespace api::modules {

            // set
            eamuse_set_keypad_overrides(keypad, state);
-            Sleep(sleep_time);
+            std::this_thread::sleep_for(std::chrono::milliseconds(sleep_time));

            // unset
            eamuse_set_keypad_overrides(keypad, 0);
-            Sleep(sleep_time);
+            std::this_thread::sleep_for(std::chrono::milliseconds(sleep_time));
        }
    }

@@ -58,7 +58,7 @@ namespace api::modules {

        // check if file exists in modules
        if (!fileutils::file_exists(dll_path)) {
-            return error(res, "Couldn't find " + dll_path.string());
+            return error(res, fmt::format("Couldn't find {}", dll_path));
        }

        // get module
@@ -118,7 +118,7 @@ namespace api::modules {

        // check if file exists in modules
        if (!fileutils::file_exists(dll_path)) {
-            return error(res, "Couldn't find " + dll_path.string());
+            return error(res, fmt::format("Couldn't find {}", dll_path));
        }

        // get module
@@ -196,7 +196,7 @@ namespace api::modules {

        // check if file exists in modules
        if (!fileutils::file_exists(dll_path)) {
-            return error(res, "Couldn't find " + dll_path.string());
+            return error(res, fmt::format("Couldn't find {}", dll_path));
        }

        // get module
@@ -0,0 +1,67 @@
+#include "sdvx.h"
+
+#include <functional>
+
+using namespace std::placeholders;
+using namespace rapidjson;
+
+namespace api::modules {
+
+    SDVX::SDVX() : Module("sdvx") {
+        functions["tapeled_get"] = std::bind(&SDVX::tapeled_get, this, _1, _2);
+
+        for (auto &light : games::sdvx::TAPELED_MAPPING) {
+            lights_by_names.emplace(light.lightName, light);
+        }
+    }
+
+    /**
+     * tapeled_get()
+     * tapeled_get(name: str, ...)
+     */
+    void SDVX::tapeled_get(Request &req, Response &res) {
+        Value response_object(kObjectType);
+
+        // all tape leds
+        if (req.params.Size() == 0) {
+            // iterate through each device and dump its lights data into the response
+            for (const auto &mapping : games::sdvx::TAPELED_MAPPING) {
+                copy_tapeled_data(res, response_object, mapping);
+            }
+        } else {
+            // specified light names
+            for (Value &param : req.params.GetArray()) {
+                // check params
+                if (!param.IsString()) {
+                    error_type(res, "name", "string");
+                    return;
+                }
+
+                const auto name = param.GetString();
+                if (const auto &it = lights_by_names.find(name); it != lights_by_names.end()) {
+                    copy_tapeled_data(res, response_object, it->second.get());
+                }
+            }
+        }
+
+        res.add_data(response_object);
+    }
+
+    void SDVX::copy_tapeled_data(Response &res, Value &response_object,
+            const tapeledutils::tape_led &mapping)
+    {
+        Value light_state(kArrayType);
+        light_state.Reserve(
+            static_cast<SizeType>(mapping.data.size() * 3),
+            res.doc()->GetAllocator());
+        for (const auto [r, g, b] : mapping.data) {
+            light_state.PushBack(r, res.doc()->GetAllocator());
+            light_state.PushBack(g, res.doc()->GetAllocator());
+            light_state.PushBack(b, res.doc()->GetAllocator());
+        }
+
+        // can't use StringRef here, turns some strings partially into null bytes for some reason
+        Value light_name(mapping.lightName.c_str(), res.doc()->GetAllocator());
+        response_object.AddMember(light_name, light_state, res.doc()->GetAllocator());
+    }
+}
@@ -0,0 +1,24 @@
+#pragma once
+
+#include <functional>
+#include <string>
+
+#include "api/module.h"
+#include "api/request.h"
+#include "external/robin_hood.h"
+#include "games/sdvx/sdvx.h"
+
+namespace api::modules {
+
+    class SDVX : public Module {
+    public:
+        SDVX();
+
+    private:
+        robin_hood::unordered_map<std::string, std::reference_wrapper<tapeledutils::tape_led>> lights_by_names;
+
+        void tapeled_get(Request &req, Response &res);
+        void copy_tapeled_data(Response &res, rapidjson::Value &response_object,
+                const tapeledutils::tape_led &mapping);
+    };
+}
@@ -1,5 +1,7 @@
 #include "touch.h"

+#include <algorithm>
+#include <cstdint>
 #include <functional>

 #include "external/rapidjson/document.h"
@@ -8,7 +10,11 @@
 #include "misc/eamuse.h"
 #include "launcher/launcher.h"
 #include "touch/touch.h"
+#include "touch/native/inject.h"
+#include "touch/native/nativetouchhook.h"
+#include "touch/native/transform.h"
 #include "util/utils.h"
+#include "games/gitadora/gitadora.h"
 #include "games/iidx/iidx.h"

 using namespace std::placeholders;
@@ -17,6 +23,54 @@ using namespace rapidjson;

 namespace api::modules {

+    static bool inject_native_touch(
+            const std::vector<TouchPoint> &touch_points, int canvas_w, int canvas_h) {
+        if (touch_points.empty()) {
+            // no active touches remain; release the synthetic contact
+            return nativetouch::inject::inject_synthetic_touch({ 0, 0 }, false);
+        }
+
+        // only a single synthetic contact is supported, so just use the first touch point
+        const auto &touch = touch_points.front();
+        POINT position { touch.x, touch.y };
+        if (canvas_w > 0 && canvas_h > 0) {
+            return nativetouch::inject::inject_synthetic_touch_from_canvas(
+                position,
+                { canvas_w, canvas_h },
+                true);
+        }
+        return nativetouch::inject::inject_synthetic_touch(position, true);
+    }
+
+    // map API coordinates onto the touch space SDVX reads, which depends on how it is displayed
+    static void sdvx_touch_errata(
+            int &x, int &y, bool use_native, int canvas_w, int canvas_h) {
+
+        // windowed coordinates already match the sub screen window they land on
+        if (GRAPHICS_WINDOWED) {
+            return;
+        }
+
+        // landscape mode: native injection hands the game these coordinates
+        // unchanged, so apply the rotation the touchscreen gets, while wintouchemu instead
+        // rotates them later through the subscreen overlay
+        if (GRAPHICS_FS_ORIENTATION_SWAP) {
+            if (use_native) {
+                POINT position { x, y };
+                if (nativetouch::transform::sdvx_landscape_rotate(&position, canvas_w, canvas_h)) {
+                    x = position.x;
+                    y = position.y;
+                }
+            }
+            return;
+        }
+
+        // rotate into the portrait touch space
+        const int x_raw = x;
+        x = canvas_w - y;
+        y = x_raw;
+    }
+
    Touch::Touch() : Module("touch") {
        is_sdvx = avs::game::is_model("KFC");

@@ -26,6 +80,32 @@ namespace api::modules {
            is_tdj_fhd = false;
        }

+        use_native = nativetouch::is_hooked();
+
+        // resolution the API/companion sends native touch coordinates in (after errata)
+        native_canvas_w = 0;
+        native_canvas_h = 0;
+        if (is_sdvx) {
+            // windowed API coordinates land on the sub screen window as-is; fullscreen
+            // coordinates are rotated into the game's touch space by apply_touch_errata
+            const bool landscape_coordinates =
+                GRAPHICS_WINDOWED || GRAPHICS_FS_ORIENTATION_SWAP;
+            native_canvas_w = landscape_coordinates ? 1920 : 1080;
+            native_canvas_h = landscape_coordinates ? 1080 : 1920;
+        } else if (avs::game::is_model("LDJ")) {
+            // TDJ subscreen; FHD models are upscaled to 1080p by apply_touch_errata
+            native_canvas_w = is_tdj_fhd ? 1920 : 1280;
+            native_canvas_h = is_tdj_fhd ? 1080 : 720;
+        } else if (avs::game::is_model("M39")) {
+            // pop'n music API touch surface
+            native_canvas_w = 1280;
+            native_canvas_h = 800;
+        } else if (games::gitadora::is_arena_model()) {
+            // GITADORA arena SMALL subscreen, either in its own window or in the overlay
+            native_canvas_w = games::gitadora::ARENA_SUBSCREEN_WIDTH;
+            native_canvas_h = games::gitadora::ARENA_SUBSCREEN_HEIGHT;
+        }
+
        functions["read"] = std::bind(&Touch::read, this, _1, _2);
        functions["write"] = std::bind(&Touch::write, this, _1, _2);
        functions["write_reset"] = std::bind(&Touch::write_reset, this, _1, _2);
@@ -99,7 +179,11 @@ namespace api::modules {
        }

        // apply touch points
-        touch_write_points(&touch_points);
+        if (use_native) {
+            write_native(touch_points);
+        } else {
+            touch_write_points(&touch_points);
+        }
    }

    /**
@@ -123,23 +207,54 @@ namespace api::modules {
        }

        // remove all IDs
-        touch_remove_points(&touch_point_ids);
+        if (use_native) {
+            write_reset_native(touch_point_ids);
+        } else {
+            touch_remove_points(&touch_point_ids);
+        }
+    }
+
+    void Touch::write_native(const std::vector<TouchPoint> &touch_points) {
+        if (touch_points.empty()) {
+            return;
+        }
+
+        std::lock_guard<std::mutex> lock(native_touch_mutex);
+        const auto touch_id = touch_points.front().id;
+        if (native_touch_id && *native_touch_id != touch_id) {
+            if (!inject_native_touch({}, native_canvas_w, native_canvas_h)) {
+                return;
+            }
+            native_touch_id.reset();
+        }
+        if (inject_native_touch(touch_points, native_canvas_w, native_canvas_h)) {
+            native_touch_id = touch_id;
+        }
+    }
+
+    void Touch::write_reset_native(const std::vector<DWORD> &touch_point_ids) {
+        std::lock_guard<std::mutex> lock(native_touch_mutex);
+        if (!native_touch_id ||
+                std::find(
+                    touch_point_ids.begin(),
+                    touch_point_ids.end(),
+                    *native_touch_id) == touch_point_ids.end()) {
+            return;
+        }
+        if (inject_native_touch({}, native_canvas_w, native_canvas_h)) {
+            native_touch_id.reset();
+        }
    }

    void Touch::apply_touch_errata(int &x, int &y) {
-        int x_raw = x;
-        int y_raw = y;
-
        if (is_tdj_fhd) {
            // deal with TDJ FHD resolution mismatch (upgrade 720p to 1080p)
            // we don't know what screen is being shown on the companion and the API doesn't specify
            // the target of the touch events so just assume it's the sub screen
-            x = x_raw * 1920 / 1280;
-            y = y_raw * 1080 / 720;
+            x = x * 1920 / 1280;
+            y = y * 1080 / 720;
        } else if (is_sdvx) {
-            // for exceed gear, they are both 1080p screens, but need to apply transformation
-            x = 1080 - y_raw;
-            y = x_raw;
+            sdvx_touch_errata(x, y, use_native, native_canvas_w, native_canvas_h);
        }
    }
 }
@@ -1,8 +1,12 @@
 #pragma once

+#include <cstdint>
+#include <mutex>
+#include <optional>
 #include <vector>
 #include "api/module.h"
 #include "api/request.h"
+#include "touch/touch.h"

 namespace api::modules {

@@ -13,11 +17,21 @@ namespace api::modules {
    private:
        bool is_sdvx;
        bool is_tdj_fhd;
+        bool use_native;
+        std::mutex native_touch_mutex;
+        std::optional<uint32_t> native_touch_id;
+
+        // resolution the API/companion sends native touch coordinates in (after errata);
+        // used to normalize before mapping into the native injection window
+        int native_canvas_w;
+        int native_canvas_h;

        // function definitions
        void read(Request &req, Response &res);
        void write(Request &req, Response &res);
        void write_reset(Request &req, Response &res);
+        void write_native(const std::vector<TouchPoint> &touch_points);
+        void write_reset_native(const std::vector<DWORD> &touch_point_ids);
        void apply_touch_errata(int &x, int &y);

    };
@@ -7,6 +7,7 @@ from .coin import *
 from .control import *
 from .exceptions import *
 from .iidx import *
+from .sdvx import *
 from .info import *
 from .keypads import *
 from .lights import *
@@ -0,0 +1,12 @@
+from .connection import Connection
+from .request import Request
+
+
+def sdvx_tapeled_get(con: Connection, *light_names):
+    req = Request("sdvx", "tapeled_get")
+
+    for light_name in light_names:
+        req.add_param(light_name)
+
+    res = con.request(req)
+    return res.get_data()
@@ -1,7 +1,9 @@
 #include "controller.h"
 #include "serial.h"

+#include <chrono>
 #include <string>
+#include <thread>
 #include <utility>

 #include "util/logging.h"
@@ -160,7 +162,7 @@ namespace api {

                // slow down on reconnect
                if (this->running) {
-                    Sleep(retry_time);
+                    std::this_thread::sleep_for(std::chrono::milliseconds(retry_time));
                }
            }
        });
@@ -5,6 +5,7 @@
 #include "util/utils.h"
 #include "util/rc4.h"
 #include "util/logging.h"
+#include "overlay/notifications.h"
 #include "controller.h"

 using namespace headsocket;
@@ -91,12 +92,18 @@ namespace api {

        // log connection
        log_info("api::websocket", "client connected");
+        overlay::notifications::add(
+                overlay::notifications::Severity::Success,
+                "API websocket client connected");
    }

    void WebSocketClient::on_disconnect() {

        // log disconnection
        log_info("api::websocket", "client disconnected");
+        overlay::notifications::add(
+                overlay::notifications::Severity::Info,
+                "API websocket client disconnected");

        // get pointer to server
        auto srv = reinterpret_cast<WebSocketServer *>(server().get());
@@ -7,6 +7,7 @@
 #include <stdint.h>

 #include "external/robin_hood.h"
+#include "games/popn/popn.h"
 #include "launcher/logger.h"
 #include "launcher/signal.h"
 #include "util/deferlog.h"
@@ -1113,6 +1114,7 @@ namespace avs {
 #endif
            // jubeat
            {"jubeat.dll", 0x2000000},
+            {"jubeat2019.dll", 0x10000000},

            // MUSECA
            {"museca.dll", 0xC000000},
@@ -1167,7 +1169,7 @@ namespace avs {
            // Ongaku Paradise
            {"arkjc9.dll", 0xA00000},

-            // KAMUNITY 
+            // KAMUNITY
            {"kamunity.dll", 33554432},
        };

@@ -1179,6 +1181,15 @@ namespace avs {
                return;
            }

+            // hack, both hello! popn and popn hc using popn.dll
+            if (games::popn::is_pikapika_model()) {
+                auto old_size = HEAP_SIZE;
+                HEAP_SIZE = 0x20000000;
+                DEFAULT_HEAP_SIZE_SET = true;
+                log_info("avs-core", "updated heap size: {} -> {}", old_size, avs::core::HEAP_SIZE);
+                return;
+            }
+
            if (HEAP_SIZE_DEFAULTS.find(dll_name) != HEAP_SIZE_DEFAULTS.end()) {
                auto old_size = HEAP_SIZE;

@@ -1506,7 +1517,7 @@ namespace avs {
                        "    * It's also possible that you have incomplete game data\n"
                        "    * Do NOT copy over random DLLs from another game installation; DLL must match game version\n"
                        "\n"
-                    , DLL_NAME, MODULE_PATH.string()) };
+                    , DLL_NAME, MODULE_PATH) };
                    log_warning("avs-ea3", "{}", info_str);
                    log_fatal("avs-ea3", "Failed to find critical avs DLL on disk (avs2-core.dll OR {})", DLL_NAME);
                }
@@ -1747,9 +1758,9 @@ namespace avs {
        {
            deferredlogs::defer_error_messages({
                "AVS filesystem initialization failure was previously detected during boot!",
-                fmt::format("    ERROR: directory could not be created: {}", src_path.string().c_str()),
-                fmt::format("    if you see a crash, it may have been caused by bad <mounttable> contents in {}", avs::core::CFG_PATH.c_str()),
-                "    fix the XML file and try again"  
+                fmt::format("    ERROR: directory could not be created: {}", src_path),
+                fmt::format("    if you see a crash, it may have been caused by bad <mounttable> contents in {}", avs::core::CFG_PATH),
+                "    fix the XML file and try again"
            });
        }

@@ -1770,14 +1781,14 @@ namespace avs {
            auto created = std::filesystem::create_directories(real_path, err);

            if (created) {
-                log_info("avs-core", "created '{}' at '{}'", avs_path, real_path.string());
+                log_info("avs-core", "created '{}' at '{}'", avs_path, real_path);
            }

            if (err) {
                avs_dir_err(real_path);
                log_warning("avs-core", "failed to create '{}' folder at '{}': {}",
                        avs_path,
-                        real_path.string(),
+                        real_path,
                        err.message());
            }
        }
@@ -1911,6 +1922,15 @@ namespace avs {
            // create nvram and raw directories if possible for mounttable configurations
            create_avs_config_fs_table(config, config_node);

+#if !SPICE64
+            // sdvx4 bad log config fix
+            if (avs::game::DLL_NAME == "soundvoltex.dll" &&  // it's too early for avs::game::is_model
+                property_search(config, config_node, "/log/enable_console")) {
+                log_info("avs-core", "applying SDVX4 avs-config.xml fix for <log><enable_console>");
+                property_search_remove_safe(config, config_node, "/log/enable_console");
+            }
+#endif
+
            // set log level
            if (!LOG_LEVEL_CUSTOM.empty()) {
                property_search_remove_safe(config, config_node, "/log/level");
@@ -151,8 +151,8 @@ namespace avs {
                        "    * It's also possible that you have incomplete game data\n"
                        "    * Do NOT copy over random DLLs from another game installation; DLL must match game version\n"
                        "\n"
-                    , DLL_NAME, MODULE_PATH.string()) };
-                    
+                    , DLL_NAME, MODULE_PATH) };
+
                    log_warning("avs-ea3", "{}", info_str);
                    log_fatal("avs-ea3", "Failed to find critical ea3 DLL on disk (avs2-ea3.dll OR {})", DLL_NAME);
                }
@@ -217,7 +217,25 @@ namespace avs {
                ea3_config_name = "prop/eamuse-config.xml";
            }

-            log_info("avs-ea3", "booting (using {})", ea3_config_name);
+            if (fileutils::file_exists(ea3_config_name)) {
+                log_info("avs-ea3", "found {} on disk", ea3_config_name);
+            } else if (CFG_PATH.size()) {
+                log_fatal("avs-ea3", "user-specified ea3 config file is missing: {}", ea3_config_name);
+            } else {
+                log_warning("avs-ea3", "looked for the following files in order:");
+                log_warning("avs-ea3", "  * prop/ea3-config.xml");
+                log_warning("avs-ea3", "  * prop/ea3-cfg.xml");
+                if (avs::game::DLL_NAME == "beatstream1.dll") {
+                    log_warning("avs-ea3", "  * prop/ea3-config-1.xml");
+                }
+                if (avs::game::DLL_NAME == "beatstream2.dll") {
+                    log_warning("avs-ea3", "  * prop/ea3-config-2.xml");
+                }
+                log_warning("avs-ea3", "  * prop/eamuse-config.xml");
+                log_warning("avs-ea3", "none of the files above were found, game will fail to boot");
+                log_warning("avs-ea3", "usually, this means your game data is incomplete");
+                log_fatal("avs-ea3", "no ea3 config file found in prop directory; check log messages");
+            }

            // remember new config path
            CFG_PATH = ea3_config_name;
@@ -372,8 +390,8 @@ namespace avs {

            // fall back to default PCBID if node is not found
            if (!EA3_PCBID[0] && PCBID_CUSTOM.empty()) {
-                log_warning("avs-ea3", "no PCBID set, falling back to default PCBID value (04040000000000000000)");
-                PCBID_CUSTOM = "04040000000000000000";
+                log_warning("avs-ea3", "no PCBID set, falling back to default PCBID value (01201000000000010101)");
+                PCBID_CUSTOM = "01201000000000010101";
            }

            // custom PCBID
@@ -554,9 +572,17 @@ namespace avs {
            auto app_param = avs::core::property_search_safe(app_config, nullptr, "/param");

            // call the game init
-            log_info("avs-ea3", "calling entry init (init code = {})", init_code_str);
+            log_info("avs-ea3", "calling dll_entry_init (init code = {})", init_code_str);
            if (!avs::game::entry_init(init_code_str.data(), app_param)) {
-                log_fatal("avs-ea3", "entry init failed :(");
+                for (size_t i = 0; i < 255; i++) {
+                    log_warning("avs-ea3", "dll_entry_init failed :(");
+                }
+                deferredlogs::defer_error_messages({
+                    "dll_entry_init failed",
+                    "    init code: " + init_code_str,
+                    "    this is a fatal error that causes boot failure",
+                    "    most likely, your prop files have errors"
+                });
            }

            // accommodate changes to soft id code
@@ -565,6 +591,7 @@ namespace avs {
            // for proper reporting of Omnimix and other song packs
            if (_stricmp(EA3_MODEL, "LDJ") == 0 ||
                _stricmp(EA3_MODEL, "L44") == 0 ||
+                _stricmp(EA3_MODEL, "T44") == 0 ||
                _stricmp(EA3_MODEL, "M39") == 0 ||
                _stricmp(EA3_MODEL, "KFC") == 0)
            {
@@ -73,11 +73,10 @@ namespace avs {

            // load game instance
            const auto dll_path = MODULE_PATH / DLL_NAME;
-            const auto dll_path_s = dll_path.string();
-            log_info("avs-game", "DLL path: {}", dll_path_s.c_str());
+            log_info("avs-game", "DLL path: {}", dll_path);

            // MAX_PATH is 260
-            if (130 <= dll_path_s.length()) {
+            if (const auto dll_path_len = dll_path.wstring().length(); 130 <= dll_path_len) {
                log_warning(
                    "avs-game",
                    "PATH TOO LONG WARNING\n\n"
@@ -93,7 +92,7 @@ namespace avs {
                    "long, often resulting in random crashes. Move the game contents to\n"
                    "a directory with shorter path.\n"
                    "-------------------------------------------------------------------\n\n",
-                    dll_path_s, dll_path_s.length());
+                    dll_path, dll_path_len);
            }

            // ddr gamemdx.dll user error
@@ -112,7 +111,7 @@ namespace avs {

            // file not found on disk
            if (!fileutils::file_exists(dll_path)) {
-                log_warning("avs-game", "game DLL could not be found on disk: {}", dll_path.string().c_str());
+                log_warning("avs-game", "game DLL could not be found on disk: {}", dll_path);
                log_warning("avs-game", "double check -exec and -modules parameters; unless you know what you're doing, leave them blank");
            }

@@ -39,8 +39,8 @@
            <button name="VEFX" vkey="10" analogtype="0" devid="Player 1" invert="false" debounce_up="0" debounce_down="0" velocity_threshold="0"/>
        </buttons>
        <analogs>
-            <analog name="Turntable P1" devid="Player 1" index="4" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Turntable P2" devid="Player 2" index="4" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
+            <analog name="Turntable P1" devid="Player 1" index="4" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Turntable P2" devid="Player 2" index="4" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
        </analogs>
        <lights>
            <light name="P1 1" devid="Player 1" index="0"/>
@@ -117,8 +117,8 @@
        <keypad_buttons/>
        <!-- Assumes analog mode. Need to invert direction since this FW is designed for INF. -->
        <analogs>
-            <analog name="Turntable P1" devid="arcin (1p)" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="true" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Turntable P2" devid="arcin (2p)" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="true" smoothing="false" multiplier="1" relative="false" delay="0"/>
+            <analog name="Turntable P1" devid="arcin (1p)" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="true" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Turntable P2" devid="arcin (2p)" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="true" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
        </analogs>
        <lights>
            <light name="P1 1" devid="arcin (1p)" index="0"/>
@@ -144,34 +144,34 @@
        <buttons/>
        <keypad_buttons/>
        <analogs>
-            <analog name="Key 1" devid="Analogs key 01-28" index="4" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 2" devid="Analogs key 01-28" index="5" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 3" devid="Analogs key 01-28" index="6" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 4" devid="Analogs key 01-28" index="7" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 5" devid="Analogs key 01-28" index="8" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 6" devid="Analogs key 01-28" index="9" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 7" devid="Analogs key 01-28" index="10" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 8" devid="Analogs key 01-28" index="11" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 9" devid="Analogs key 01-28" index="12" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 10" devid="Analogs key 01-28" index="13" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 11" devid="Analogs key 01-28" index="14" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 12" devid="Analogs key 01-28" index="15" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 13" devid="Analogs key 01-28" index="16" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 14" devid="Analogs key 01-28" index="17" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 15" devid="Analogs key 01-28" index="18" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 16" devid="Analogs key 01-28" index="19" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 17" devid="Analogs key 01-28" index="20" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 18" devid="Analogs key 01-28" index="21" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 19" devid="Analogs key 01-28" index="22" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 20" devid="Analogs key 01-28" index="23" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 21" devid="Analogs key 01-28" index="24" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 22" devid="Analogs key 01-28" index="25" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 23" devid="Analogs key 01-28" index="26" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 24" devid="Analogs key 01-28" index="27" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 25" devid="Analogs key 01-28" index="28" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 26" devid="Analogs key 01-28" index="29" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 27" devid="Analogs key 01-28" index="30" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Key 28" devid="Analogs key 01-28" index="31" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
+            <analog name="Key 1" devid="Analogs key 01-28" index="4" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 2" devid="Analogs key 01-28" index="5" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 3" devid="Analogs key 01-28" index="6" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 4" devid="Analogs key 01-28" index="7" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 5" devid="Analogs key 01-28" index="8" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 6" devid="Analogs key 01-28" index="9" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 7" devid="Analogs key 01-28" index="10" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 8" devid="Analogs key 01-28" index="11" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 9" devid="Analogs key 01-28" index="12" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 10" devid="Analogs key 01-28" index="13" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 11" devid="Analogs key 01-28" index="14" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 12" devid="Analogs key 01-28" index="15" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 13" devid="Analogs key 01-28" index="16" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 14" devid="Analogs key 01-28" index="17" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 15" devid="Analogs key 01-28" index="18" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 16" devid="Analogs key 01-28" index="19" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 17" devid="Analogs key 01-28" index="20" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 18" devid="Analogs key 01-28" index="21" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 19" devid="Analogs key 01-28" index="22" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 20" devid="Analogs key 01-28" index="23" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 21" devid="Analogs key 01-28" index="24" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 22" devid="Analogs key 01-28" index="25" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 23" devid="Analogs key 01-28" index="26" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 24" devid="Analogs key 01-28" index="27" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 25" devid="Analogs key 01-28" index="28" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 26" devid="Analogs key 01-28" index="29" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 27" devid="Analogs key 01-28" index="30" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Key 28" devid="Analogs key 01-28" index="31" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
        </analogs>
        <lights>
            <light name="Key 1 R" devid="Lights 01-14" index="0"/>
@@ -277,9 +277,9 @@
        </buttons>
        <keypad_buttons/>
        <analogs>
-            <analog name="Guitar P1 Wail X" devid="Gitaller P1" index="1" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Guitar P1 Wail Y" devid="Gitaller P1" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="Guitar P1 Wail Z" devid="Gitaller P1" index="5" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
+            <analog name="Guitar P1 Wail X" devid="Gitaller P1" index="1" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Guitar P1 Wail Y" devid="Gitaller P1" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="Guitar P1 Wail Z" devid="Gitaller P1" index="5" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
        </analogs>
        <lights/>
    <!-- Gen 1 FAUCETWO in HID mode; should also work for later gens -->
@@ -296,8 +296,8 @@
        </buttons>
        <keypad_buttons/>
        <analogs>
-            <analog name="VOL-L" devid="Buttons (MI_01)" index="1" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
-            <analog name="VOL-R" devid="Buttons (MI_01)" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay="0"/>
+            <analog name="VOL-L" devid="Buttons (MI_01)" index="1" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
+            <analog name="VOL-R" devid="Buttons (MI_01)" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1" relative="false" delay_ms="0"/>
        </analogs>
        <lights>
            <light name="BT-A" devid="Lights (MI_02)" index="24"/>
@@ -6,6 +6,7 @@
        <asmv3:windowsSettings xmlns="http://schemas.microsoft.com/SMI/2005/WindowsSettings">
            <dpiAwareness xmlns="http://schemas.microsoft.com/SMI/2016/WindowsSettings">PerMonitorV2</dpiAwareness>
            <dpiAware>true</dpiAware>
+            <activeCodePage xmlns="http://schemas.microsoft.com/SMI/2019/WindowsSettings">Legacy</activeCodePage>
        </asmv3:windowsSettings>
    </asmv3:application>
    <dependency>
@@ -23,7 +24,7 @@
    <trustInfo xmlns="urn:schemas-microsoft-com:asm.v2">
        <security>
            <requestedPrivileges>
-                <requestedExecutionLevel level="requireAdministrator" uiAccess="false"/>
+                <requestedExecutionLevel level="asInvoker" uiAccess="false"/>
            </requestedPrivileges>
        </security>
    </trustInfo>
@@ -1,4 +1,4 @@
-#!/bin/bash
+#!/usr/bin/env bash

 # set to EN-US for consistency
 LANG=en_us_8859_1
@@ -43,10 +43,16 @@ GIT_BRANCH=$(git rev-parse --abbrev-ref HEAD 2> /dev/null || echo "none")
 GIT_HEAD=$(git rev-parse HEAD || echo "none")
 TOOLCHAIN_32="${TOOLCHAIN_32:-"/usr/share/mingw/toolchain-i686-w64-mingw32.cmake"}"
 TOOLCHAIN_64="${TOOLCHAIN_64:-"/usr/share/mingw/toolchain-x86_64-w64-mingw32.cmake"}"
+TOOLCHAIN_WINXP_32="${TOOLCHAIN_WINXP_32:-"/opt/llvm-mingw-xp/toolchain-files/cmake/i686-mingw32-clang.cmake"}"
+TOOLCHAIN_WINXP_64="${TOOLCHAIN_WINXP_64:-"/opt/llvm-mingw-xp/toolchain-files/cmake/x86_64-mingw32-clang.cmake"}"
 BUILDDIR_32_RELEASE="./cmake-build-release-32"
 BUILDDIR_32_DEBUG="./cmake-build-debug-32"
 BUILDDIR_64_RELEASE="./cmake-build-release-64"
 BUILDDIR_64_DEBUG="./cmake-build-debug-64"
+BUILDDIR_WINXP_32_RELEASE="./cmake-build-release-winxp-32"
+BUILDDIR_WINXP_32_DEBUG="./cmake-build-debug-winxp-32"
+BUILDDIR_WINXP_64_RELEASE="./cmake-build-release-winxp-64"
+BUILDDIR_WINXP_64_DEBUG="./cmake-build-debug-winxp-64"
 DEBUG=0
 OUTDIR="./bin/spice2x"
 OUTDIR_EXTRAS="./bin/spice2x/extras"
@@ -62,21 +68,44 @@ DIST_FOLDER="./dist"
 DIST_NAME="spice2x-$(date +%y)-$(date +%m)-$(date +%d).zip"
 DIST_NAME_EXTRAS="spice2x-$(date +%y)-$(date +%m)-$(date +%d)-full.zip"
 DIST_COMMENT=${DIST_NAME}$'\n'"$GIT_BRANCH - $GIT_HEAD"$'\nThank you for playing.'
-TARGETS_32="spicetools_stubs_kbt spicetools_stubs_kld spicetools_cfg spicetools_cfg_linux spicetools_spice spicetools_spice_laa spicetools_spice_linux spicetools_stubs_cpusbxpkm"
-TARGETS_64="spicetools_stubs_kbt64 spicetools_stubs_kld64 spicetools_stubs_nvEncodeAPI64 spicetools_stubs_nvcuvid spicetools_stubs_nvcuda spicetools_spice64 spicetools_spice64_linux"
+TARGETS_32="spicetools_stubs_kbt spicetools_stubs_kld spicetools_cfg spicetools_cfg_linux spicetools_spice spicetools_spice_laa spicetools_spice_linux spicetools_stubs_cpusbxpkm spicetools_sdk_sample_v0_flat_c_32"
+TARGETS_64="spicetools_stubs_kbt64 spicetools_stubs_kld64 spicetools_stubs_nvEncodeAPI64 spicetools_stubs_nvcuvid spicetools_stubs_nvcuda spicetools_spice64 spicetools_spice64_linux spicetools_sdk_sample_v0_flat_c_64 spicetools_sdk_sample_v0_cpp_64"
+TARGETS_XP32="spicetools_cfg spicetools_spice"
+TARGETS_XP64="spicetools_spice64"

 # determine build type
 BUILD_TYPE="Release"
 BUILDDIR_32=${BUILDDIR_32_RELEASE}
 BUILDDIR_64=${BUILDDIR_64_RELEASE}
+BUILDDIR_WINXP_32=${BUILDDIR_WINXP_32_RELEASE}
+BUILDDIR_WINXP_64=${BUILDDIR_WINXP_64_RELEASE}
 if ((DEBUG > 0))
 then
 	BUILD_TYPE="Debug"
 	BUILDDIR_32=${BUILDDIR_32_DEBUG}
 	BUILDDIR_64=${BUILDDIR_64_DEBUG}
+	BUILDDIR_WINXP_32=${BUILDDIR_WINXP_32_DEBUG}
+	BUILDDIR_WINXP_64=${BUILDDIR_WINXP_64_DEBUG}
 	DIST_NAME=$(echo "${DIST_NAME}" | sed 's/\.[^.]*$/-dbg&/')
 fi

+# is the XP-compatible toolchain installed?
+XP_MUST_BUILD=0
+# 64-bit WinXP builds are disabled by default; set to 1 to opt in
+BUILD_XP_64_ENABLE=0
+BUILD_XP_32=0
+BUILD_XP_64=0
+if [ -f "$TOOLCHAIN_WINXP_32" ]; then
+	BUILD_XP_32=1;
+elif ((XP_MUST_BUILD > 0))
+then
+	echo "WinXP 32bit toolchain not available, aborting"
+	exit 1
+fi
+if ((BUILD_XP_64_ENABLE > 0)) && [ -f "$TOOLCHAIN_WINXP_64" ]; then
+	BUILD_XP_64=1;
+fi
+
 # determine number of cores
 CORES=$(nproc)

@@ -90,6 +119,18 @@ echo "Git Branch: $GIT_BRANCH"
 echo "Git Head: $GIT_HEAD"
 echo "Toolchain for 32bit targets: $TOOLCHAIN_32"
 echo "Toolchain for 64bit targets: $TOOLCHAIN_64"
+if ((BUILD_XP_32 > 0))
+then
+	echo "Toolchain for WinXP 32bit targets: $TOOLCHAIN_WINXP_32"
+else
+	echo "WinXP 32bit toolchain not available, skipping WinXP 32bit builds"
+fi
+if ((BUILD_XP_64 > 0))
+then
+	echo "Toolchain for WinXP 64bit targets: $TOOLCHAIN_WINXP_64"
+else
+	echo "WinXP 64bit builds disabled, skipping WinXP 64bit builds"
+fi
 echo "Distribution Name: $DIST_NAME"
 echo "Build Type: $BUILD_TYPE"
 echo "Cores: $CORES"
@@ -130,11 +171,84 @@ time (
 	cmake -G "Ninja" -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_64} -DCMAKE_BUILD_TYPE=${BUILD_TYPE} "$OLDPWD" && ninja ${TARGETS_64}
 	popd > /dev/null

+	if ((BUILD_XP_32 > 0))
+	then
+		# 32 bit Windows XP
+		echo ""
+		echo "Building 32bit targets (WinXP toolchain)..."
+		echo "========================="
+		if ((CLEAN_BUILD > 0))
+		then
+			rm -rf ${BUILDDIR_WINXP_32}
+		fi
+		mkdir -p ${BUILDDIR_WINXP_32}
+		pushd ${BUILDDIR_WINXP_32} > /dev/null
+		cmake -G "Ninja" -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_WINXP_32} -DCMAKE_BUILD_TYPE=${BUILD_TYPE} -DSPICE_XP=ON "$OLDPWD" && ninja ${TARGETS_XP32}
+		popd > /dev/null
+	else
+		echo ""
+		echo "Skipping WinXP 32bit builds, toolchain not specified"
+	fi
+
+	if ((BUILD_XP_64 > 0))
+	then
+		# 64 bit Windows XP
+		echo ""
+		echo "Building 64bit targets (WinXP toolchain)..."
+		echo "========================="
+		if ((CLEAN_BUILD > 0))
+		then
+			rm -rf ${BUILDDIR_WINXP_64}
+		fi
+		mkdir -p ${BUILDDIR_WINXP_64}
+		pushd ${BUILDDIR_WINXP_64} > /dev/null
+		cmake -G "Ninja" -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_WINXP_64} -DCMAKE_BUILD_TYPE=${BUILD_TYPE} -DSPICE_XP=ON "$OLDPWD" && ninja ${TARGETS_XP64}
+		popd > /dev/null
+	else
+		echo ""
+		echo "Skipping WinXP 64bit builds"
+	fi
+
 	echo ""
 	echo "Compilation process done :)"
 	echo "==========================="
 )

+echo ""
+echo "Checking Win7 compatibility..."
+echo "============================="
+if ! command -v windows_dll_compat_checker &> /dev/null; then
+	echo "WARNING: windows_dll_compat_checker not found, skipping Win7 compatibility check"
+else
+	windows_dll_compat_checker -s PREMADE/konami_win7_museca_x86_64.ini \
+		${BUILDDIR_64}/spicetools/64/spice64.exe
+	windows_dll_compat_checker -s PREMADE/konami_win7_museca_x86_64_32bit_dlls.ini \
+		${BUILDDIR_32}/spicetools/spicecfg.exe \
+		${BUILDDIR_32}/spicetools/32/spice.exe
+fi
+
+if ((BUILD_XP_32 > 0)) || ((BUILD_XP_64 > 0))
+then
+	echo ""
+	echo "Checking XP compatibility..."
+	echo "============================="
+	if ! command -v windows_dll_compat_checker &> /dev/null; then
+		echo "WARNING: windows_dll_compat_checker not found, skipping XP compatibility check"
+	else
+		if ((BUILD_XP_64 > 0))
+		then
+			windows_dll_compat_checker -s PREMADE/winxp_x86_64.ini \
+				${BUILDDIR_WINXP_64}/spicetools/64/spice64.exe
+		fi
+		if ((BUILD_XP_32 > 0))
+		then
+			windows_dll_compat_checker -s PREMADE/konami_winxp_jubeat_i686.ini \
+				${BUILDDIR_WINXP_32}/spicetools/spicecfg.exe \
+				${BUILDDIR_WINXP_32}/spicetools/32/spice.exe
+		fi
+	fi
+fi
+
 # generate PDBs
 if false  # ((DEBUG > 0))
 then
@@ -191,6 +305,13 @@ rm -rf ${OUTDIR_EXTRAS}
 mkdir -p ${OUTDIR_EXTRAS}
 mkdir -p ${OUTDIR_EXTRAS}/largeaddressaware
 mkdir -p ${OUTDIR_EXTRAS}/linux
+mkdir -p ${OUTDIR_EXTRAS}/sdk/samples/32
+mkdir -p ${OUTDIR_EXTRAS}/sdk/samples/64
+mkdir -p ${OUTDIR_EXTRAS}/updater
+if ((BUILD_XP_32 > 0)) || ((BUILD_XP_64 > 0))
+then
+mkdir -p ${OUTDIR_EXTRAS}/winxp
+fi

 if false # ((DEBUG > 0))
 then
@@ -199,16 +320,8 @@ then
    cp ${BUILDDIR_32}/spicetools/spicecfg-pdb.pdb ${OUTDIR} 2>/dev/null
    cp ${BUILDDIR_32}/spicetools/32/spice-pdb.exe ${OUTDIR} 2>/dev/null
    cp ${BUILDDIR_32}/spicetools/32/spice-pdb.pdb ${OUTDIR} 2>/dev/null
-    #cp ${BUILDDIR_32}/spicetools/32/kbt.dll ${OUTDIR}/stubs/32 2>/dev/null
-    #cp ${BUILDDIR_32}/spicetools/32/kld.dll ${OUTDIR}/stubs/32 2>/dev/null
    cp ${BUILDDIR_64}/spicetools/64/spice64-pdb.exe ${OUTDIR} 2>/dev/null
    cp ${BUILDDIR_64}/spicetools/64/spice64-pdb.pdb ${OUTDIR} 2>/dev/null
-    #cp ${BUILDDIR_64}/spicetools/64/kbt.dll ${OUTDIR}/stubs/64 2>/dev/null
-    #cp ${BUILDDIR_64}/spicetools/64/kld.dll ${OUTDIR}/stubs/64 2>/dev/null
-    #cp ${BUILDDIR_64}/spicetools/64/nvEncodeAPI64.dll ${OUTDIR}/stubs/64 2>/dev/null
-    #cp ${BUILDDIR_64}/spicetools/64/nvcuda.dll ${OUTDIR}/stubs/64 2>/dev/null
-    #cp ${BUILDDIR_64}/spicetools/64/nvcuvid.dll ${OUTDIR}/stubs/64 2>/dev/null
-	#cp ${BUILDDIR_32}/spicetools/32/cpusbxpkm.dll ${OUTDIR}/stubs/32 2>/dev/null
 else
    # release files
    cp ${BUILDDIR_32}/spicetools/spicecfg.exe ${OUTDIR} 2>/dev/null
@@ -216,16 +329,24 @@ else
    cp ${BUILDDIR_32}/spicetools/32/spice.exe ${OUTDIR} 2>/dev/null
    cp ${BUILDDIR_32}/spicetools/32/spice_laa.exe ${OUTDIR_EXTRAS}/largeaddressaware/spice.exe 2>/dev/null
    cp ${BUILDDIR_32}/spicetools/32/spice_linux.exe ${OUTDIR_EXTRAS}/linux/spice.exe 2>/dev/null
-    #cp ${BUILDDIR_32}/spicetools/32/kbt.dll ${OUTDIR}/stubs/32 2>/dev/null
-    #cp ${BUILDDIR_32}/spicetools/32/kld.dll ${OUTDIR}/stubs/32 2>/dev/null
    cp ${BUILDDIR_64}/spicetools/64/spice64.exe ${OUTDIR} 2>/dev/null
    cp ${BUILDDIR_64}/spicetools/64/spice64_linux.exe ${OUTDIR_EXTRAS}/linux/spice64.exe 2>/dev/null
-    #cp ${BUILDDIR_64}/spicetools/64/kbt.dll ${OUTDIR}/stubs/64 2>/dev/null
-    #cp ${BUILDDIR_64}/spicetools/64/kld.dll ${OUTDIR}/stubs/64 2>/dev/null
    cp ${BUILDDIR_64}/spicetools/64/nvEncodeAPI64.dll ${OUTDIR}/stubs/64 2>/dev/null
    cp ${BUILDDIR_64}/spicetools/64/nvcuda.dll ${OUTDIR}/stubs/64 2>/dev/null
    cp ${BUILDDIR_64}/spicetools/64/nvcuvid.dll ${OUTDIR}/stubs/64 2>/dev/null
    cp ${BUILDDIR_32}/spicetools/32/cpusbxpkm.dll ${OUTDIR}/stubs/32 2>/dev/null
+	cp ${BUILDDIR_32}/spicetools/32/sdk_sample_v0_flat_c.dll ${OUTDIR_EXTRAS}/sdk/samples/32/v0_flat_c.dll 2>/dev/null
+	cp ${BUILDDIR_64}/spicetools/64/sdk_sample_v0_flat_c.dll ${OUTDIR_EXTRAS}/sdk/samples/64/v0_flat_c.dll 2>/dev/null
+	cp ${BUILDDIR_64}/spicetools/64/sdk_sample_v0_cpp.dll ${OUTDIR_EXTRAS}/sdk/samples/64/v0_cpp.dll 2>/dev/null
+	if ((BUILD_XP_32 > 0))
+	then
+		cp ${BUILDDIR_WINXP_32}/spicetools/spicecfg.exe ${OUTDIR_EXTRAS}/winxp 2>/dev/null
+		cp ${BUILDDIR_WINXP_32}/spicetools/32/spice.exe ${OUTDIR_EXTRAS}/winxp 2>/dev/null
+	fi
+	if ((BUILD_XP_64 > 0))
+	then
+		cp ${BUILDDIR_WINXP_64}/spicetools/64/spice64.exe ${OUTDIR_EXTRAS}/winxp 2>/dev/null
+	fi
 fi

 # pack source files to output directory
@@ -238,22 +359,28 @@ then
 		echo "WARNING: Couldn't get git to create the archive. Is this a git repository?"
 fi

+# sdk headers
+cp -r ./sdk/include ${OUTDIR_EXTRAS}/sdk
+
 # copy resources
 rm -rf ${OUTDIR_EXTRAS}/api
 mkdir ${OUTDIR_EXTRAS}/api
 find ./api/resources/python -name "__pycache__" -exec rm -rf {} +
 cp -r ./api/resources/* ${OUTDIR_EXTRAS}/api

+# generate the standalone updater scripts from the shared template and drop them
+bash ./build_updaters.sh ${OUTDIR_EXTRAS}/updater
+
 # build distribution archive
 if ((DIST_ENABLE > 0))
 then
 	echo "Building dist..."
 	mkdir -p ${DIST_FOLDER}
 	rm -rf "${DIST_FOLDER}/${DIST_NAME}"
-	pushd ${OUTDIR}/.. > /dev/null
-	zip -qrXT9 "$OLDPWD/${DIST_FOLDER}/${DIST_NAME}" spice2x -x "spice2x/extras/*" -z <<< "$DIST_COMMENT"
+	pushd ${OUTDIR} > /dev/null
+	zip -qrXT9 "$OLDPWD/${DIST_FOLDER}/${DIST_NAME}" . -x "extras/*" -z <<< "$DIST_COMMENT"
 	echo "Building extras..."
-	zip -qrXT9 "$OLDPWD/${DIST_FOLDER}/${DIST_NAME_EXTRAS}" spice2x -z <<< "$DIST_COMMENT"
+	zip -qrXT9 "$OLDPWD/${DIST_FOLDER}/${DIST_NAME_EXTRAS}" . -z <<< "$DIST_COMMENT"
 	popd > /dev/null
 fi

@@ -1,7 +1,7 @@
 docker build --pull external/docker -t spicetools/deps
 docker build --build-context gitroot=%cd%/../../.git . -t spicetools/spice
-docker run --rm -it -v %cd%/dist:/src/src/spice2x/dist -v %cd%/bin:/src/src/spice2x/bin -v %cd%/.ccache:/src/src/spice2x/.ccache spicetools/spice %*
+docker run --rm -i -v %cd%/dist:/src/src/spice2x/dist -v %cd%/bin:/src/src/spice2x/bin -v %cd%/.ccache:/src/src/spice2x/.ccache spicetools/spice %*
@REM to generate PDBs, set DEBUG to 1 in build_all.sh, place cv2pdb in external\cv2pdb, and run below
@REM external\cv2pdb\cv2pdb.exe bin\spice2x\spicecfg.exe bin\spice2x\spicecfg-pdb.exe bin\spice2x\spicecfg-pdb.pdb
@REM external\cv2pdb\cv2pdb.exe bin\spice2x\spice.exe bin\spice2x\spice-pdb.exe bin\spice2x\spice-pdb.pdb
-@REM external\cv2pdb\cv2pdb.exe bin\spice2x\spice64.exe bin\spice2x\spice64-pdb.exe bin\spice2x\spice64-pdb.pdb
+@REM external\cv2pdb\cv2pdb.exe bin\spice2x\spice64.exe bin\spice2x\spice64-pdb.exe bin\spice2x\spice64-pdb.pdb
@@ -1,4 +1,12 @@
-#!/bin/bash
+#!/usr/bin/env bash
+
+set -eu
+
 docker build --pull "$PWD/external/docker" -t spicetools/deps --platform linux/x86_64
 docker build --build-context gitroot="$PWD/../../.git" . -t spicetools/spice:latest
-docker run --rm -v "$PWD/dist:/src/src/spice2x/dist" -v "$PWD/bin:/src/src/spice2x/bin" -v "$PWD/.ccache:/src/src/spice2x/.ccache" spicetools/spice "$@"
+
+# Interactive TTY if available, so docker build can be Ctrl+C'd
+DOCKER_FLAGS=""
+[ -t 0 ] && DOCKER_FLAGS="-it"
+
+docker run $DOCKER_FLAGS --rm -v "$PWD/dist:/src/src/spice2x/dist" -v "$PWD/bin:/src/src/spice2x/bin" -v "$PWD/.ccache:/src/src/spice2x/.ccache" spicetools/spice "$@"
@@ -0,0 +1,57 @@
+#!/usr/bin/env bash
+
+# builds the standalone updater scripts from the shared PowerShell logic in
+# update_spice.ps1. each output is a self-contained polyglot .bat: a small batch
+# header (with the release channel baked in) followed by the shared PowerShell
+# body after the marker line.
+#
+# usage: build_updaters.sh OUT_DIR
+#   OUT_DIR  directory to write the generated .bat files into
+
+set -eu
+
+if [ $# -lt 1 ]; then
+	echo "usage: build_updaters.sh OUT_DIR" >&2
+	exit 1
+fi
+
+SRC_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PS_BODY="${SRC_DIR}/update_spice.ps1"
+OUT_DIR="$1"
+
+[ -f "$PS_BODY" ] || { echo "error: $PS_BODY not found" >&2; exit 1; }
+
+MARKER='#___PS___'
+
+gen() {
+	channel="$1"; label="$2"; out="$3"
+	{
+		cat <<HEADER
+@echo off
+setlocal
+REM ============================================================================
+REM  spice2x updater  -  updates spice.exe, spice64.exe and spicecfg.exe in
+REM  this folder to the latest spice2x release from GitHub.
+REM
+REM  release channel: ${label}
+REM  make sure spice/spicecfg are NOT running before updating.
+REM ============================================================================
+
+REM expose the script folder + channel to the embedded PowerShell, then run the
+REM PowerShell part: re-read this file and execute everything after the marker.
+set "SPICE_DIR=%~dp0"
+set "SPICE_CHANNEL=${channel}"
+powershell -NoProfile -ExecutionPolicy Bypass -Command "\$m='#___'+'PS'+'___'; \$code=((Get-Content -LiteralPath '%~f0' -Raw) -split \$m)[-1]; Invoke-Expression \$code"
+set "RC=%ERRORLEVEL%"
+endlocal & exit /b %RC%
+
+${MARKER}
+HEADER
+		cat "$PS_BODY"
+	} | sed -e 's/\r$//' -e 's/$/\r/' > "${OUT_DIR}/${out}"
+	echo "  generated ${OUT_DIR}/${out}"
+}
+
+echo "Generating updater scripts from $(basename "$PS_BODY")..."
+gen "" "stable (pre-releases excluded)" "update_spice.bat"
+gen "beta" "beta (newest, pre-releases included)" "update_spice_beta.bat"
@@ -90,6 +90,10 @@ std::string Analog::getDisplayString(rawinput::RawInputManager *manager) {
                return "MIDI Unknown Index " + indexString + " (" + device->desc + ")";
            }
        }
+        case rawinput::XINPUT_GAMEPAD:
+            return fmt::format("{} ({})",
+                xinput::get_analog_string(static_cast<xinput::XInputAnalogEnum>(index)),
+                device->desc);
        case rawinput::DESTROYED:
            return "Device unplugged (" + indexString + ")";
        default:
@@ -203,22 +207,38 @@ float Analog::applyMultiplier(float value) {
 }

 float Analog::normalizeAnalogValue(float value) {
-    // effectively the same as fmodf(value, 1.f)
-    // for small values, this is MUCH faster than fmodf.
-    float new_value = value;
-    while (new_value > 1.f) {
-        new_value -= 1.f;
+    if (getType() == GameAPI::Analogs::AnalogType::Circular) {
+        // effectively the same as fmodf(value, 1.f)
+        // for small values, this is MUCH faster than fmodf.
+        float new_value = value;
+        while (new_value > 1.f) {
+            new_value -= 1.f;
+        }
+        while (new_value < 0.f) {
+            new_value += 1.f;
+        }
+        return new_value;
+
+    } else {
+        // clamp to [0, 1] range
+        return std::clamp(value, 0.f, 1.f);
    }
-    while (new_value < 0.f) {
-        new_value += 1.f;
-    }
-    return new_value;
 }

 float Analog::applyDeadzone(float raw_value) {
    float value = raw_value;
-    const auto deadzone = this->getDeadzone();
-    if (deadzone > 0) {
+    auto deadzone = this->getDeadzone();
+
+    // in the past, positive deadzone applied in the center, negative deadzone applied to 0
+    // after each analog value received a type (circular/linear) this has been simpliifed to
+    // positive values only since we can figure out where the rest value is
+    // for back compat, treat negative value as positive
+    if (deadzone < 0.f) {
+        deadzone = -deadzone;
+    }
+
+    // relative mode assumes that user is using a stick, so center is neutral regardless of analog type
+    if (getType() != GameAPI::Analogs::AnalogType::LinearPositive || isRelativeMode()) {

        // calculate values
        const auto delta = value - 0.5f;
@@ -251,7 +271,7 @@ float Analog::applyDeadzone(float raw_value) {
            }
        }

-    } else if (deadzone < 0) {
+    } else {

        // invert for mirror
        if (this->getDeadzoneMirror()) {
@@ -259,8 +279,8 @@ float Analog::applyDeadzone(float raw_value) {
        }

        // deadzone from minimum value
-        if (deadzone > -1 && value > -deadzone) {
-            value = std::min(1.f, (value + deadzone) / (1.f + deadzone));
+        if (deadzone < 1.f && deadzone < value) {
+            value = std::max(0.f, (value - deadzone) / (1.f - deadzone));
        } else {
            value = 0.f;
        }
@@ -271,4 +291,75 @@ float Analog::applyDeadzone(float raw_value) {
        }
    }
    return value;
-}
+}
+
+float Analog::getRelativeModeValue(float raw_value) {
+    const auto now = get_performance_seconds();
+    auto delta_time = now - this->rel_mode_last_read_time_s;
+
+    if (this->rel_mode_last_read_time_s != 0.f) {
+        // some heuristics to prevent huge jumps:
+        //  * if we went for more than 250ms without polls, discard it (e.g., during loading screens)
+        //  * cap the delta at 100ms to prevent huge jumps in case of very infrequent polling
+        if (delta_time < 0.f || 0.25f < delta_time) {
+            delta_time = 0.f;
+        } else if (delta_time > 0.1f) {
+            delta_time = 0.1f;
+        }
+
+        // scale [0, 1] to [-1, 1] to simplify calculations
+        const auto delta_raw_value = (raw_value - 0.5f) * 2.f;
+
+        // target is one revolution per second at max speed at 1.0 sensitivity
+        auto adjusted_delta_value = delta_raw_value * delta_time;
+
+        // multiplier / divisor
+        if (this->getMultiplier() > 1) {
+            adjusted_delta_value *= this->getMultiplier();
+        } else if (this->getMultiplier() < -1) {
+            adjusted_delta_value /= -this->getMultiplier();
+        }
+
+        // sensitivity
+        if (this->isSensitivitySet()) {
+            adjusted_delta_value *= this->getSensitivity();
+        }
+
+        // calculate the new absolute value
+        this->rel_mode_absolute_value += adjusted_delta_value;
+    }
+
+    // update for next poll
+    this->rel_mode_last_read_time_s = now;
+    this->rel_mode_absolute_value = normalizeAnalogValue(this->rel_mode_absolute_value);
+    return this->rel_mode_absolute_value;
+}
+
+float Analog::getDelayedValue(float raw_value) {
+    const double delay_ms = static_cast<double>(this->getDelayMs());
+    if (delay_ms == 0.0) {
+        return raw_value;
+    }
+
+    // always push a new value
+    const auto now = get_performance_milliseconds();
+    this->delayed_inputs.emplace(now, raw_value);
+    
+    // drain the queue down to reasonable length to prevent unconstrained growth
+    // this would accommodate 1 second at ~1000Hz which is overkill
+    // (UI only allows for 250ms of delay)
+    while (this->delayed_inputs.size() > 1024) {
+        this->delayed_inputs.pop();
+    }
+
+    // pop until we find the oldest value still inside the delay window
+    while (this->delayed_inputs.size() > 1) {
+        const auto delta_t = now - this->delayed_inputs.front().time_in_ms;
+        if (delta_t <= delay_ms) {
+            break;
+        }
+        this->delayed_inputs.pop();
+    }
+
+    return this->delayed_inputs.front().value;
+}
@@ -1,5 +1,6 @@
 #pragma once

+#include <cstdint>
 #include <array>
 #include <string>
 #include <cmath>
@@ -13,12 +14,32 @@ namespace rawinput {
    class RawInputManager;
 }

+namespace GameAPI::Analogs {
+    enum class AnalogType {
+        // default; values wrap around (below 0 turns into 1, over 1 is 0)
+        // knobs, turntables
+        Circular = 0,
+
+        // typical joystick that rests at the center and caps at [0, 1]
+        LinearCentered = 1,
+
+        // one-directional value (sliders and instruments like piano/drum velocity)
+        // starts at 0 and goes up to 1
+        LinearPositive = 2,
+    };
+}
+
 struct AnalogMovingAverage {
    double time_in_ms;
    float sine;
    float cosine;
 };

+struct AnalogDelayEntry {
+    double time_in_ms;
+    float value;
+};
+
 class Analog {
 private:
    std::string name;
@@ -31,10 +52,11 @@ private:
    float last_state = 0.5f;
    bool sensitivity_set = false;
    bool deadzone_set = false;
+    GameAPI::Analogs::AnalogType type = GameAPI::Analogs::AnalogType::Circular;
    
    // smoothing function
    bool smoothing = false;
-    std::array<AnalogMovingAverage, ANALOG_HISTORY_CNT> vector_history;
+    std::array<AnalogMovingAverage, ANALOG_HISTORY_CNT> vector_history = {};
    int vector_history_index = 0;
    float smoothed_last_state = 0.f;

@@ -48,12 +70,13 @@ private:
    unsigned short divisor_region = 0;

    // relative input mode
-    float absolute_value_for_rel_mode = 0.5f;
+    float rel_mode_absolute_value = 0.5f;
+    float rel_mode_last_read_time_s = 0.f;
    bool relative_mode = false;

-    // circular buffer (delayed input)
-    int delay_buffer_depth = 0;
-    std::queue<float> delay_buffer;
+    // delay
+    uint32_t delay_ms = 0;
+    std::queue<AnalogDelayEntry> delayed_inputs;

    float calculateAngularDifference(float old_rads, float new_rads);
    float normalizeAngle(float rads);
@@ -66,7 +89,8 @@ public:
    float override_state = 0.5f;

    explicit Analog(std::string name) : name(std::move(name)) {
-        vector_history.fill({0.0, 0.f, 0.f});
+    };
+    explicit Analog(std::string name, GameAPI::Analogs::AnalogType type) : name(std::move(name)), type(type) {
    };

    std::string getDisplayString(rawinput::RawInputManager* manager);
@@ -74,6 +98,8 @@ public:
    float applyAngularSensitivity(float raw_rads);
    float applyMultiplier(float raw_value);
    float applyDeadzone(float raw_value);
+    float getRelativeModeValue(float raw_value);
+    float getDelayedValue(float raw_value);

    inline bool isSet() {
        if (this->override_enabled) {
@@ -90,8 +116,8 @@ public:
        deadzone_mirror = false;
        setMultiplier(1);
        setRelativeMode(false);
-        setDelayBufferDepth(0);
        setLastState(0.5f);
+        setDelayMs(0);
    }

    inline void clearBindings() {
@@ -193,25 +219,27 @@ public:
    }

    inline void setRelativeMode(bool relative_mode) {
+        if (relative_mode) {
+            this->smoothing = false;
+        }
        this->relative_mode = relative_mode;
-        this->absolute_value_for_rel_mode = 0.5f;
+        this->rel_mode_absolute_value = 0.5f;
+        this->rel_mode_last_read_time_s = 0.f;
    }

-    inline float getAbsoluteValue(float relative_delta) {
-        this->absolute_value_for_rel_mode =
-            normalizeAnalogValue(this->absolute_value_for_rel_mode + relative_delta);
-        return this->absolute_value_for_rel_mode;
+    inline GameAPI::Analogs::AnalogType getType() const {
+        return this->type;
    }

-    inline int getDelayBufferDepth() const {
-        return this->delay_buffer_depth;
+    inline void setType(GameAPI::Analogs::AnalogType type) {
+        this->type = type;
    }

-    inline void setDelayBufferDepth(int depth) {
-        this->delay_buffer_depth = depth;
+    inline uint32_t getDelayMs() const {
+        return this->delay_ms;
    }

-    inline std::queue<float> &getDelayBuffer() {
-        return this->delay_buffer;
+    inline void setDelayMs(uint32_t delay_ms) {
+        this->delay_ms = delay_ms;
    }
 };
@@ -1,8 +1,11 @@
 #include "api.h"

+#include <cassert>
 #include <optional>

+#include "games/io.h"
 #include "launcher/superexit.h"
+#include "misc/eamuse.h"
 #include "rawinput/rawinput.h"
 #include "rawinput/piuio.h"
 #include "util/time.h"
@@ -61,7 +64,43 @@ std::vector<Button> GameAPI::Buttons::sortButtons(
    return sorted;
 }

-GameAPI::Buttons::State GameAPI::Buttons::getState(rawinput::RawInputManager *manager, Button &_button, bool check_alts) {
+namespace GameAPI::Buttons {
+    static State get_button_state(
+        rawinput::RawInputManager *manager,
+        Button &button,
+        bool check_alts,
+        bool check_modifiers);
+
+    static bool modifiers_pressed(rawinput::RawInputManager *manager, Button &button);
+}
+
+bool GameAPI::Buttons::modifiers_pressed(rawinput::RawInputManager *manager, Button &button) {
+    const auto modifier_mask = button.getModifierMask();
+    if (modifier_mask == 0) {
+        return true;
+    }
+
+    auto *modifier_buttons = games::get_buttons_modifiers(eamuse_get_game());
+    if (!modifier_buttons) {
+        return false;
+    }
+
+    for (uint8_t index = 0; index < games::ModifierButtons::Size; index++) {
+        if ((modifier_mask & (UINT8_C(1) << index)) != 0 &&
+            (index >= modifier_buttons->size() ||
+             get_button_state(manager, modifier_buttons->at(index), true, false) !=
+                 GameAPI::Buttons::BUTTON_PRESSED)) {
+            return false;
+        }
+    }
+    return true;
+}
+
+GameAPI::Buttons::State GameAPI::Buttons::get_button_state(
+        rawinput::RawInputManager *manager,
+        Button &_button,
+        bool check_alts,
+        bool check_modifiers) {

    // check override
    if (_button.override_enabled) {
@@ -75,6 +114,23 @@ GameAPI::Buttons::State GameAPI::Buttons::getState(rawinput::RawInputManager *ma
    std::optional<bool> window_has_focus;
    while (true) {

+        // skip bindings whose required modifiers are not held
+        //
+        // note that modifiers cannot process MIDI as the logic is written
+        // below, since MIDI buttons are event-based (on event, off event, etc)
+        // and cannot be correctly handled by a simple early return
+        // there is no explicit check for MIDI here, but the UI should have
+        // prevented it
+        if (check_modifiers && !modifiers_pressed(manager, *current_button)) {
+            button_count++;
+            if (!alternatives || alternatives->empty() ||
+                button_count - 1 >= alternatives->size()) {
+                return BUTTON_NOT_PRESSED;
+            }
+            current_button = &alternatives->at(button_count - 1);
+            continue;
+        }
+
        // naive behavior
        if (current_button->isNaive()) {
            GameAPI::Buttons::State state;
@@ -198,9 +254,19 @@ GameAPI::Buttons::State GameAPI::Buttons::getState(rawinput::RawInputManager *ma
                            if (vKey < value_states->size()) {
                                auto value = value_states->at(vKey);
                                if (current_button->getAnalogType() == BAT_POSITIVE) {
-                                    state = value > 0.6f ? BUTTON_PRESSED : BUTTON_NOT_PRESSED;
+                                    float threshold = 0.6f;
+                                    if (current_button->getBatThreshold() > 0) {
+                                        threshold = std::clamp(current_button->getBatThreshold() / 100.f, 0.01f, 0.99f);
+                                    }
+                                    state = value > threshold ? BUTTON_PRESSED : BUTTON_NOT_PRESSED;
+
                                } else if (current_button->getAnalogType() == BAT_NEGATIVE) {
-                                    state = value < 0.4f ? BUTTON_PRESSED : BUTTON_NOT_PRESSED;
+                                    float threshold = 0.4f;
+                                    if (current_button->getBatThreshold() > 0) {
+                                        threshold = std::clamp((100 - current_button->getBatThreshold()) / 100.f, 0.01f, 0.99f);
+                                    }
+                                    state = value < threshold ? BUTTON_PRESSED : BUTTON_NOT_PRESSED;
+
                                } else {
                                    state = value > 0.01f ? BUTTON_PRESSED : BUTTON_NOT_PRESSED;
                                }
@@ -389,7 +455,23 @@ GameAPI::Buttons::State GameAPI::Buttons::getState(rawinput::RawInputManager *ma
                }
                case rawinput::PIUIO_DEVICE: {
                    state = device->piuioDev->IsPressed(vKey) ? BUTTON_PRESSED : BUTTON_NOT_PRESSED;
+                    break;
                }
+                case rawinput::XINPUT_GAMEPAD: {
+                    assert(reinterpret_cast<uintptr_t>(device->handle) < XUSER_MAX_COUNT);
+                    const auto player = static_cast<uint8_t>(reinterpret_cast<uintptr_t>(device->handle));
+                    state = manager->XINPUT_MGR->is_button_pressed(
+                        player,
+                        static_cast<xinput::XInputButtonEnum>(vKey)) ?
+                        BUTTON_PRESSED : BUTTON_NOT_PRESSED;
+                    break;
+                }
+                case rawinput::DESTROYED:
+                    // device was unplugged. release the button instead of leaving
+                    // state at getLastState(), which would latch a held button on
+                    // forever (e.g. a MIDI note held while the device is removed)
+                    state = BUTTON_NOT_PRESSED;
+                    break;
                default:
                    break;
            }
@@ -442,6 +524,10 @@ GameAPI::Buttons::State GameAPI::Buttons::getState(rawinput::RawInputManager *ma
    }
 }

+Buttons::State Buttons::getState(rawinput::RawInputManager *manager, Button &button, bool check_alts) {
+    return get_button_state(manager, button, check_alts, true);
+}
+
 Buttons::State Buttons::getState(std::unique_ptr<rawinput::RawInputManager> &manager, Button &button, bool check_alts) {
    if (manager) {
        return getState(manager.get(), button, check_alts);
@@ -457,17 +543,13 @@ static float getVelocityHelper(rawinput::RawInputManager *manager, Button &butto
        return button.override_velocity;
    }

-    // naive behavior
-    if (button.isNaive()) {
-        if (button.getInvert()) {
-            return (GetAsyncKeyState(button.getVKey()) & 0x8000) ? 0.f : 1.f;
-        } else {
-            return (GetAsyncKeyState(button.getVKey()) & 0x8000) ? 1.f : 0.f;
-        }
-    }
-
    // get button state
-    Buttons::State button_state = Buttons::getState(manager, button, false);
+    const auto button_state = Buttons::getState(manager, button, false);
+
+    // naive bindings report their digital state as full or zero velocity
+    if (button.isNaive()) {
+        return button_state == Buttons::BUTTON_PRESSED ? 1.f : 0.f;
+    }

    // check if button isn't being pressed
    if (button_state != Buttons::BUTTON_PRESSED) {
@@ -586,7 +668,7 @@ float Buttons::getVelocity(std::unique_ptr<rawinput::RawInputManager> &manager,
    }
 }

-float GameAPI::Analogs::getState(rawinput::Device *device, Analog &analog) {
+float GameAPI::Analogs::getState(rawinput::RawInputManager *manager, rawinput::Device *device, Analog &analog) {
    float value = 0.5f;
    if (!device) {
        return value;
@@ -658,81 +740,98 @@ float GameAPI::Analogs::getState(rawinput::Device *device, Analog &analog) {
            }

            // deadzone
-            if (analog.isDeadzoneSet()) {
+            // do not apply deadzone to circular analogs since it doesn't make sense (except in relative mode)
+            if (analog.isDeadzoneSet() &&
+                (analog.getType() != AnalogType::Circular || analog.isRelativeMode())) {
                value = analog.applyDeadzone(value);
            }

-            if (analog.isRelativeMode()) {
-                float relative_delta = value - 0.5f;
-                // built-in scaling to make values reasonable
-                relative_delta /= 80.f;
+            if (analog.getType() == AnalogType::Circular) {

-                // integer multiplier/divisor
-                const auto mult = analog.getMultiplier();
-                if (mult < -1) {
-                    relative_delta /= -mult;
-                } else if (1 < mult) {
-                    relative_delta *= mult;
-                }
+                if (analog.isRelativeMode()) {
+                    value = analog.getRelativeModeValue(value);

-                // sensitivity (ranges from 0.0 to 4.0)
-                if (analog.isSensitivitySet()) {
-                    relative_delta *= analog.getSensitivity();
-                }
+                } else {
+                    // integer multiplier
+                    value = analog.applyMultiplier(value);

-                // translate relative movement to absolute value
-                value = analog.getAbsoluteValue(relative_delta);
+                    // smoothing/sensitivity
+                    if (analog.getSmoothing() || analog.isSensitivitySet()) {
+                        float rads = value * (float) M_TAU;

-            } else {
-                // integer multiplier
-                value = analog.applyMultiplier(value);
+                        // smoothing
+                        if (analog.getSmoothing()) {

-                // smoothing/sensitivity
-                if (analog.getSmoothing() || analog.isSensitivitySet()) {
-                    float rads = value * (float) M_TAU;
+                            // preserve direction
+                            if (rads >= M_TAU) {
+                                rads -= 0.0001f;
+                            }

-                    // smoothing
-                    if (analog.getSmoothing()) {
-
-                        // preserve direction
-                        if (rads >= M_TAU) {
-                            rads -= 0.0001f;
+                            // calculate angle
+                            rads = analog.getSmoothedValue(rads);
                        }

-                        // calculate angle
-                        rads = analog.getSmoothedValue(rads);
+                        // sensitivity
+                        if (analog.isSensitivitySet()) {
+                            rads = analog.applyAngularSensitivity(rads);
+                        }
+
+                        // apply to value
+                        value = rads * (float) M_1_TAU;
+                    }
+                }
+            } else {
+                // sensitivity
+                if (analog.isSensitivitySet()) {
+                    // adjust curve
+                    // values < 1.f : less sensitive around neutral
+                    // values > 1.f : more sensitive around neutral
+                    float curve = analog.getSensitivity();
+                    if (curve <= 0.f) {
+                        curve = 0.01f;
+                    }
+                    curve = 1.f / curve;
+
+                    if (analog.getType() == AnalogType::LinearCentered) {
+                        // convert 0.0..1.0 to -1.0..+1.0
+                        float signed_raw = (value - 0.5f) * 2.0f;
+                        // apply curve
+                        float sign = signed_raw < 0.0f ? -1.0f : 1.0f;
+                        float magnitude = fabsf(signed_raw);
+                        float curved = sign * powf(magnitude, curve);
+                        // convert back to 0.0..1.0
+                        value = curved * 0.5f + 0.5f;
+                    } else {
+                        value = powf(value, curve);
                    }

-                    // sensitivity
-                    if (analog.isSensitivitySet()) {
-                        rads = analog.applyAngularSensitivity(rads);
+                    value = std::clamp(value, 0.f, 1.f);
+                }
+
+                // multiplier / divisor
+                if (analog.getMultiplier() < -1) {
+                    if (analog.getType() == AnalogType::LinearCentered) {
+                        value = (value - 0.5f) / (-analog.getMultiplier()) + 0.5f;
+                    } else {
+                        value /= -analog.getMultiplier();
                    }

-                    // apply to value
-                    value = rads * (float) M_1_TAU;
+                    value = std::clamp(value, 0.f, 1.f);
+
+                } else if (analog.getMultiplier() > 1) {
+                    if (analog.getType() == AnalogType::LinearCentered) {
+                        value = (value - 0.5f) * analog.getMultiplier() + 0.5f;
+                    } else {
+                        value *= analog.getMultiplier();
+                    }
+
+                    value = std::clamp(value, 0.f, 1.f);
                }
            }

            // delay
-            if (0 < analog.getDelayBufferDepth()) {
-                auto& queue = analog.getDelayBuffer();
-
-                // ensure the queue isn't too long; drop old values
-                while (analog.getDelayBufferDepth() <= (int)queue.size()) {
-                    queue.pop();
-                }
-
-                // always push new value
-                queue.push(value);
-
-                // get a new value to return
-                if ((int)queue.size() < analog.getDelayBufferDepth()) {
-                    // not enough in the queue, stall for now, shouldn't happen often
-                    value = analog.getLastState();
-                } else {
-                    value = queue.front();
-                    queue.pop();
-                }
+            if (analog.getDelayMs() > 0) {
+                value = analog.getDelayedValue(value);
            }

            break;
@@ -767,6 +866,18 @@ float GameAPI::Analogs::getState(rawinput::Device *device, Analog &analog) {
            }
            break;
        }
+        case rawinput::XINPUT_GAMEPAD: {
+            assert(reinterpret_cast<uintptr_t>(device->handle) < XUSER_MAX_COUNT);
+            const auto player = static_cast<uint8_t>(reinterpret_cast<uintptr_t>(device->handle));
+            value = manager->XINPUT_MGR->get_analog_state(
+                player,
+                static_cast<xinput::XInputAnalogEnum>(index));
+            // invert value
+            if (inverted) {
+                value = 1.f - value;
+            }
+            break;
+        }
        default:
            break;
    }
@@ -806,6 +917,42 @@ std::vector<Analog> GameAPI::Analogs::sortAnalogs(
    return sorted;
 }

+static std::vector<Analog> sortAnalogsWithTypeInternal(
+    std::vector<Analog> &analogs,
+    const std::initializer_list<GameAPI::Analogs::AnalogWithType> list) {
+
+    std::vector<Analog> sorted;
+
+    bool analog_found;
+    for (auto &a : list) {
+        analog_found = false;
+
+        for (auto &analog : analogs) {
+            if (a.name == analog.getName()) {
+                analog_found = true;
+                analog.setType(a.type);
+                sorted.push_back(analog);
+                break;
+            }
+        }
+
+        if (!analog_found) {
+            sorted.emplace_back(a.name, a.type);
+        }
+    }
+
+    return sorted;
+}
+
+void GameAPI::Analogs::sortAnalogsWithType(
+    std::vector<Analog> *analogs,
+    const std::initializer_list<AnalogWithType> list) {
+
+    if (analogs) {
+        *analogs = sortAnalogsWithTypeInternal(*analogs, list);
+    }
+}
+
 float GameAPI::Analogs::getState(rawinput::RawInputManager *manager, Analog &analog) {

    // check override
@@ -822,7 +969,7 @@ float GameAPI::Analogs::getState(rawinput::RawInputManager *manager, Analog &ana
        return analog.getLastState();
    }

-    float state = getState(device, analog);
+    float state = getState(manager, device, analog);
    analog.setLastState(state);

    return state;
@@ -904,7 +1051,7 @@ void GameAPI::Lights::sortLightsWithCategory(
 }


-void GameAPI::Lights::writeLight(rawinput::Device *device, int index, float value) {
+void GameAPI::Lights::writeLight(rawinput::RawInputManager *manager, rawinput::Device *device, int index, float value) {

    // check device
    if (!device) {
@@ -991,6 +1138,18 @@ void GameAPI::Lights::writeLight(rawinput::Device *device, int index, float valu
            }
            break;
        }
+        case rawinput::XINPUT_GAMEPAD: {
+            assert(reinterpret_cast<uintptr_t>(device->handle) < XUSER_MAX_COUNT);
+            const auto player = static_cast<uint8_t>(reinterpret_cast<uintptr_t>(device->handle));
+            if (index < static_cast<int>(xinput::XInputOutputEnum::COUNT)) {
+                manager->XINPUT_MGR->set_output_state(
+                    player, static_cast<xinput::XInputOutputEnum>(index), value);
+                // no need to set output_pending; xinput output handled immediately
+            } else {
+                log_warning("api", "invalid xinput light index: {}", index);
+            }
+            break;
+        }
        default:
            break;
    }
@@ -1016,9 +1175,9 @@ void GameAPI::Lights::writeLight(rawinput::RawInputManager *manager, Light &ligh

        // write state
        if (light.override_enabled) {
-            writeLight(device, light.getIndex(), light.override_state);
+            writeLight(manager, device, light.getIndex(), light.override_state);
        } else {
-            writeLight(device, light.getIndex(), value);
+            writeLight(manager, device, light.getIndex(), value);
        }
    }

@@ -85,14 +85,15 @@ namespace GameAPI {
                const std::vector<Analog> &analogs,
                const std::vector<std::string> &analog_names);

-        template<typename T>
-        void sortAnalogs(std::vector<Analog> *analogs, T t) {
-            const std::vector<std::string> analog_names { t };
+        struct AnalogWithType {
+            const std::string name;
+            const AnalogType type;

-            if (analogs) {
-                *analogs = GameAPI::Analogs::sortAnalogs(*analogs, analog_names);
-            }
-        }
+            AnalogWithType(std::string name, AnalogType type) :
+                name(std::move(name)), type(type) {}
+        };
+
+        void sortAnalogsWithType(std::vector<Analog> *analogs, const std::initializer_list<AnalogWithType> list);

        template<typename T, typename... Rest>
        void sortAnalogs(std::vector<Analog> *analogs, T t, Rest... rest) {
@@ -103,7 +104,7 @@ namespace GameAPI {
            }
        }

-        float getState(rawinput::Device *device, Analog &analog);
+        float getState(rawinput::RawInputManager *manager, rawinput::Device *device, Analog &analog);
        float getState(rawinput::RawInputManager *manager, Analog &analog);
        float getState(std::unique_ptr<rawinput::RawInputManager> &manager, Analog &analog);
    }
@@ -135,7 +136,7 @@ namespace GameAPI {

        void sortLightsWithCategory(std::vector<Light> *lights, const std::initializer_list<LightAndCategory> list);

-        void writeLight(rawinput::Device *device, int index, float value);
+        void writeLight(rawinput::RawInputManager *manager, rawinput::Device *device, int index, float value);
        void writeLight(rawinput::RawInputManager *manager, Light &light, float value);
        void writeLight(std::unique_ptr<rawinput::RawInputManager> &manager, Light &light, float value);

@@ -278,6 +278,8 @@ std::string Button::getVKeyString() {
            return "Left Alt";
        case 0xA5:
            return "Right Alt";
+        case INVALID_VKEY:
+            return "Invalid";
        default:

            // check win API
@@ -297,14 +299,37 @@ std::string Button::getMidiNoteString() {
    return fmt::format("{}{}", note_names[index % 12], ((index / 12) - 1));
 }

-std::string Button::getDisplayString(rawinput::RawInputManager* manager) {
+std::string Button::getDisplayString(rawinput::RawInputManager *manager) {
+    auto binding_display = this->getBindingDisplayString(manager);
+    auto modifiers = this->getModifierMask();
+    if (binding_display.empty() || modifiers == 0) {
+        return binding_display;
+    }
+
+    std::string modifier_display;
+    for (unsigned int index = 1; modifiers != 0; index++, modifiers >>= 1) {
+        if ((modifiers & UINT8_C(1)) == 0) {
+            continue;
+        }
+        if (!modifier_display.empty()) {
+            modifier_display += "+";
+        }
+        modifier_display += fmt::format("Mod{}", index);
+    }
+    return modifier_display + "+" + binding_display;
+}
+
+std::string Button::getBindingDisplayString(rawinput::RawInputManager *manager) {

    // get VKey string
    auto vKey = (uint16_t) this->getVKey();
    std::string vKeyString = fmt::format("{:#x}", vKey);

    // device must be existing
-    if (this->device_identifier.empty() && vKey == INVALID_VKEY) {
+    if (this->isNaive() && vKey == INVALID_VKEY) {
+        if (this->getInvert()) {
+            return "(always on)";
+        }
        return "";
    }

@@ -415,6 +440,10 @@ std::string Button::getDisplayString(rawinput::RawInputManager* manager) {
            }
            case rawinput::PIUIO_DEVICE:
                return "PIUIO " + vKeyString;
+            case rawinput::XINPUT_GAMEPAD:
+                return fmt::format("{} ({})",
+                    xinput::get_button_string(static_cast<xinput::XInputButtonEnum>(vKey)),
+                    device->desc);
            case rawinput::DESTROYED:
                return "Device unplugged (" + vKeyString + ")";
            default:
@@ -49,6 +49,7 @@ private:
    std::string name;
    std::string device_identifier = "";
    unsigned short vKey = INVALID_VKEY;
+    uint8_t modifier_mask = 0;

    // default bindings are always naive
    unsigned short vKey_default = INVALID_VKEY;
@@ -56,6 +57,7 @@ private:
    ButtonAnalogType analog_type = BAT_NONE;
    double debounce_up = 0.0;
    double debounce_down = 0.0;
+    int bat_threshold = 0; // for positive/negative only, value in percentage, zero is default
    bool invert = false;
    bool is_temporary = false;

@@ -64,6 +66,7 @@ private:
    unsigned short velocity_threshold = 0;

    std::string getMidiNoteString();
+    std::string getBindingDisplayString(rawinput::RawInputManager *manager);

 public:
    std::string getVKeyString();
@@ -83,12 +86,20 @@ public:
        if (this->override_enabled) {
            return true;
        }
+
+        // has a device
+        if (!this->device_identifier.empty()) {
+            return true;
+        }
+
+        // naive, and has a valid vKey
        if (this->vKey != INVALID_VKEY) {
            return true;
        }

        for (auto &alternative : this->alternatives) {
-            if (alternative.vKey != INVALID_VKEY) {
+            if (!alternative.getDeviceIdentifier().empty() ||
+                alternative.getVKey() != INVALID_VKEY) {
                return true;
            }
        }
@@ -101,6 +112,8 @@ public:
        alternatives.clear();
        device_identifier = "";
        analog_type = BAT_NONE;
+        bat_threshold = 0;
+        modifier_mask = 0;
    }

    std::string getDisplayString(rawinput::RawInputManager* manager);
@@ -113,6 +126,14 @@ public:
        return this->name;
    }

+    inline uint8_t getModifierMask() const {
+        return this->modifier_mask;
+    }
+
+    inline void setModifierMask(uint8_t new_modifier_mask) {
+        this->modifier_mask = new_modifier_mask & UINT8_C(0x0F);
+    }
+
    inline const std::string &getDeviceIdentifier() const {
        return this->device_identifier;
    }
@@ -161,6 +182,14 @@ public:
        this->debounce_down = debounce_time_down;
    }

+    inline void setBatThreshold(int bat_threshold) {
+        this->bat_threshold = bat_threshold;
+    }
+
+    inline int getBatThreshold() const {
+        return this->bat_threshold;
+    }
+
    inline bool getInvert() const {
        return this->invert;
    }
@@ -28,7 +28,7 @@ Config::Config() {
    this->status = false;
    if (CONFIG_PATH_OVERRIDE.length() > 0) {
        this->configLocation = CONFIG_PATH_OVERRIDE;
-        log_info("cfg", "using custom config file: {}", this->configLocation.string());
+        log_info("cfg", "using custom config file: {}", this->configLocation);
    } else {
        this->configLocation = std::filesystem::path(_wgetenv(L"APPDATA")) / L"spicetools.xml";
        // avoids logging the expanded appdata path as it contains user name
@@ -57,7 +57,7 @@ Config::Config() {
                this->firstFillConfigFile();
                break;
            case tinyxml2::XMLError::XML_ERROR_FILE_COULD_NOT_BE_OPENED:
-                log_fatal("cfg", "could not open config file: {}", this->configLocation.string());
+                log_fatal("cfg", "could not open config file: {}", this->configLocation);
                break;
            case tinyxml2::XMLError::XML_ERROR_FILE_NOT_FOUND:
                this->createConfigFile();
@@ -72,7 +72,7 @@ Config::Config() {
            case tinyxml2::XMLError::XML_ERROR_PARSING_UNKNOWN:
            case tinyxml2::XMLError::XML_ERROR_MISMATCHED_ELEMENT:
            case tinyxml2::XMLError::XML_ERROR_PARSING:
-                log_warning("cfg", "Couldn't read config file: {}", this->configLocation.string());
+                log_warning("cfg", "Couldn't read config file: {}", this->configLocation);
                this->createConfigFile();
                this->firstFillConfigFile();
                break;
@@ -98,7 +98,7 @@ bool Config::getStatus() {
    return this->status;
 }

-bool Config::addGame(Game &game) {
+bool Config::addGame(Game &game, bool save) {
    tinyxml2::XMLNode *rootNode = this->configFile.LastChild();
    tinyxml2::XMLElement *gameNodes = rootNode->FirstChildElement("game");

@@ -162,15 +162,19 @@ bool Config::addGame(Game &game) {
                    auto analogType = (int) BAT_NONE;
                    double debounce_up = 0.0;
                    double debounce_down = 0.0;
+                    int bat_threshold = 0;
                    int velocity_threshold = 0;
                    bool invert = false;
+                    unsigned int modifier_mask = 0;
                    tinyxml2::XMLError attrError = gameButtonNode->QueryIntAttribute("vkey", &vKey);
                    const char *devid = gameButtonNode->Attribute("devid");
                    gameButtonNode->QueryIntAttribute("analogtype", &analogType);
                    gameButtonNode->QueryDoubleAttribute("debounce_up", &debounce_up);
                    gameButtonNode->QueryDoubleAttribute("debounce_down", &debounce_down);
+                    gameButtonNode->QueryIntAttribute("bat_threshold", &bat_threshold);
                    gameButtonNode->QueryIntAttribute("velocity_threshold", &velocity_threshold);
                    gameButtonNode->QueryBoolAttribute("invert", &invert);
+                    gameButtonNode->QueryUnsignedAttribute("modifiers", &modifier_mask);
                    if (attrError != tinyxml2::XMLError::XML_SUCCESS) {
                        gameButtonsNode->DeleteChild(gameButtonNode);
                        gameButtonNode = this->configFile.NewElement("button");
@@ -180,16 +184,20 @@ bool Config::addGame(Game &game) {
                        gameButtonNode->SetAttribute("devid", button->getDeviceIdentifier().c_str());
                        gameButtonNode->SetAttribute("debounce_up", debounce_up);
                        gameButtonNode->SetAttribute("debounce_down", debounce_down);
+                        gameButtonNode->SetAttribute("bat_threshold", bat_threshold);
                        gameButtonNode->SetAttribute("velocity_threshold", velocity_threshold);
                        gameButtonNode->SetAttribute("invert", invert);
+                        gameButtonNode->SetAttribute("modifiers", button->getModifierMask());
                        gameButtonsNode->InsertEndChild(gameButtonNode);
                    } else {
                        button->setVKey(static_cast<unsigned short int>(vKey));
                        button->setAnalogType((ButtonAnalogType) analogType);
                        button->setDebounceUp(debounce_up);
                        button->setDebounceDown(debounce_down);
+                        button->setBatThreshold(bat_threshold);
                        button->setVelocityThreshold(velocity_threshold);
                        button->setInvert(invert);
+                        button->setModifierMask(static_cast<uint8_t>(modifier_mask));
                        if (devid) {
                            button->setDeviceIdentifier(devid);
                        }
@@ -207,8 +215,10 @@ bool Config::addGame(Game &game) {
                gameButtonNode->SetAttribute("analogtype", (int) it.getAnalogType());
                gameButtonNode->SetAttribute("debounce_up", it.getDebounceUp());
                gameButtonNode->SetAttribute("debounce_down", it.getDebounceDown());
+                gameButtonNode->SetAttribute("bat_threshold", it.getBatThreshold());
                gameButtonNode->SetAttribute("velocity_threshold", it.getVelocityThreshold());
                gameButtonNode->SetAttribute("invert", it.getInvert());
+                gameButtonNode->SetAttribute("modifiers", it.getModifierMask());
                gameButtonNode->SetAttribute("devid", it.getDeviceIdentifier().c_str());
                gameButtonsNode->InsertEndChild(gameButtonNode);
            }
@@ -248,7 +258,7 @@ bool Config::addGame(Game &game) {
                bool smoothing = false;
                int multiplier = 1;
                bool relative_mode = false;
-                int delay_buffer_depth = 0;
+                uint32_t delay = 0;
                tinyxml2::XMLError err1 = gameAnalogNode->QueryIntAttribute("index", &index);
                gameAnalogNode->QueryFloatAttribute("sensivity", &sensitivity);
                gameAnalogNode->QueryFloatAttribute("deadzone", &deadzone);
@@ -257,7 +267,7 @@ bool Config::addGame(Game &game) {
                gameAnalogNode->QueryBoolAttribute("smoothing", &smoothing);
                gameAnalogNode->QueryIntAttribute("multiplier", &multiplier);
                gameAnalogNode->QueryBoolAttribute("relative", &relative_mode);
-                gameAnalogNode->QueryIntAttribute("delay", &delay_buffer_depth);
+                gameAnalogNode->QueryUnsignedAttribute("delay_ms", &delay);
                const char *devid = gameAnalogNode->Attribute("devid");

                if (err1 != tinyxml2::XMLError::XML_SUCCESS || !devid) {
@@ -273,7 +283,7 @@ bool Config::addGame(Game &game) {
                    gameAnalogNode->SetAttribute("smoothing", it.getSmoothing());
                    gameAnalogNode->SetAttribute("multiplier", it.getMultiplier());
                    gameAnalogNode->SetAttribute("relative", it.isRelativeMode());
-                    gameAnalogNode->SetAttribute("delay", it.getDelayBufferDepth());
+                    gameAnalogNode->SetAttribute("delay_ms", it.getDelayMs());
                    gameAnalogsNode->InsertEndChild(gameAnalogNode);
                } else {
                    it.setIndex(static_cast<unsigned short int>(index));
@@ -285,7 +295,7 @@ bool Config::addGame(Game &game) {
                    it.setSmoothing(smoothing);
                    it.setMultiplier(multiplier);
                    it.setRelativeMode(relative_mode);
-                    it.setDelayBufferDepth(delay_buffer_depth);
+                    it.setDelayMs(delay);
                }
            } else {
                gameAnalogNode = this->configFile.NewElement("analog");
@@ -298,7 +308,7 @@ bool Config::addGame(Game &game) {
                gameAnalogNode->SetAttribute("smoothing", it.getSmoothing());
                gameAnalogNode->SetAttribute("multiplier", it.getMultiplier());
                gameAnalogNode->SetAttribute("relative", it.isRelativeMode());
-                gameAnalogNode->SetAttribute("delay", it.getDelayBufferDepth());
+                gameAnalogNode->SetAttribute("delay_ms", it.getDelayMs());
                gameAnalogNode->SetAttribute("devid", it.getDeviceIdentifier().c_str());
                gameAnalogsNode->InsertEndChild(gameAnalogNode);
            }
@@ -432,8 +442,10 @@ bool Config::addGame(Game &game) {
            gameButtonNode->SetAttribute("analogtype", it.getAnalogType());
            gameButtonNode->SetAttribute("debounce_up", it.getDebounceUp());
            gameButtonNode->SetAttribute("debounce_down", it.getDebounceDown());
+            gameButtonNode->SetAttribute("bat_threshold", it.getBatThreshold());
            gameButtonNode->SetAttribute("velocity_threshold", it.getVelocityThreshold());
            gameButtonNode->SetAttribute("invert", it.getInvert());
+            gameButtonNode->SetAttribute("modifiers", it.getModifierMask());
            gameButtonNode->SetAttribute("devid", it.getDeviceIdentifier().c_str());
            gameButtonsNode->InsertEndChild(gameButtonNode);
        }
@@ -452,7 +464,7 @@ bool Config::addGame(Game &game) {
            gameAnalogNode->SetAttribute("smoothing", it.getSmoothing());
            gameAnalogNode->SetAttribute("multiplier", it.getMultiplier());
            gameAnalogNode->SetAttribute("relative", it.isRelativeMode());
-            gameAnalogNode->SetAttribute("delay", it.getDelayBufferDepth());
+            gameAnalogNode->SetAttribute("delay_ms", it.getDelayMs());
            gameAnalogsNode->InsertEndChild(gameAnalogNode);
        }

@@ -480,13 +492,20 @@ bool Config::addGame(Game &game) {
        rootNode->InsertEndChild(gameNode);
    }

-    // save config
-    this->saveConfigFile();
+    // save config (skipped when caller batches multiple addGame calls
+    // and flushes once at the end via Config::save())
+    if (save) {
+        this->saveConfigFile();
+    }

    // return success
    return true;
 }

+void Config::save() {
+    this->saveConfigFile();
+}
+
 bool Config::updateBinding(const Game &game, const Button &button, int alternative) {

    // get root node
@@ -535,9 +554,11 @@ bool Config::updateBinding(const Game &game, const Button &button, int alternati
                gameButtonNode->SetAttribute("analogtype", (int) button.getAnalogType());
                gameButtonNode->SetAttribute("debounce_up", button.getDebounceUp());
                gameButtonNode->SetAttribute("debounce_down", button.getDebounceDown());
+                gameButtonNode->SetAttribute("bat_threshold", button.getBatThreshold());
                gameButtonNode->SetAttribute("velocity_threshold", button.getVelocityThreshold());
                gameButtonNode->SetAttribute("invert", button.getInvert());
                gameButtonNode->SetAttribute("devid", button.getDeviceIdentifier().c_str());
+                gameButtonNode->SetAttribute("modifiers", button.getModifierMask());
                break;
            }
        }
@@ -551,9 +572,11 @@ bool Config::updateBinding(const Game &game, const Button &button, int alternati
            gameButtonNode->SetAttribute("analogtype", 0);
            gameButtonNode->SetAttribute("debounce_up", 0.0);
            gameButtonNode->SetAttribute("debounce_down", 0.0);
+            gameButtonNode->SetAttribute("bat_threshold", 0);
            gameButtonNode->SetAttribute("velocity_threshold", 0);
            gameButtonNode->SetAttribute("invert", false);
            gameButtonNode->SetAttribute("devid", "");
+            gameButtonNode->SetAttribute("modifiers", 0);
            gameButtonsNode->InsertEndChild(gameButtonNode);
        }
    }
@@ -684,7 +707,7 @@ bool Config::updateBinding(const Game &game, const Analog &analog) {
        gameAnalogNode->SetAttribute("smoothing", analog.getSmoothing());
        gameAnalogNode->SetAttribute("multiplier", analog.getMultiplier());
        gameAnalogNode->SetAttribute("relative", analog.isRelativeMode());
-        gameAnalogNode->SetAttribute("delay", analog.getDelayBufferDepth());
+        gameAnalogNode->SetAttribute("delay_ms", analog.getDelayMs());
        gameAnalogNode->SetAttribute("devid", analog.getDeviceIdentifier().c_str());
    } else {
        gameAnalogNode = this->configFile.NewElement("analog");
@@ -696,7 +719,7 @@ bool Config::updateBinding(const Game &game, const Analog &analog) {
        gameAnalogNode->SetAttribute("smoothing", analog.getSmoothing());
        gameAnalogNode->SetAttribute("multiplier", analog.getMultiplier());
        gameAnalogNode->SetAttribute("relative", analog.isRelativeMode());
-        gameAnalogNode->SetAttribute("delay", analog.getDelayBufferDepth());
+        gameAnalogNode->SetAttribute("delay_ms", analog.getDelayMs());
        gameAnalogNode->SetAttribute("devid", analog.getDeviceIdentifier().c_str());
        gameAnalogsNode->InsertEndChild(gameAnalogNode);
    }
@@ -946,14 +969,18 @@ std::vector<Button> Config::getButtons(const std::string &gameName) {
                auto analogType = (int) BAT_NONE;
                double debounce_up = 0.0;
                double debounce_down = 0.0;
+                int bat_threshold = 0;
                int velocity_threshold = 0;
                bool invert = false;
+                unsigned int modifier_mask = 0;
                gameButtonNode->QueryIntAttribute("vkey", &vKey);
                gameButtonNode->QueryIntAttribute("analogtype", &analogType);
                gameButtonNode->QueryDoubleAttribute("debounce_up", &debounce_up);
                gameButtonNode->QueryDoubleAttribute("debounce_down", &debounce_down);
+                gameButtonNode->QueryIntAttribute("bat_threshold", &bat_threshold);
                gameButtonNode->QueryIntAttribute("velocity_threshold", &velocity_threshold);
                gameButtonNode->QueryBoolAttribute("invert", &invert);
+                gameButtonNode->QueryUnsignedAttribute("modifiers", &modifier_mask);
                const char *devid = gameButtonNode->Attribute("devid");

                // find alternative
@@ -966,8 +993,10 @@ std::vector<Button> Config::getButtons(const std::string &gameName) {
                        alt.setAnalogType((ButtonAnalogType) analogType);
                        alt.setDebounceUp(debounce_up);
                        alt.setDebounceDown(debounce_down);
+                        alt.setBatThreshold(bat_threshold);
                        alt.setVelocityThreshold(velocity_threshold);
                        alt.setInvert(invert);
+                        alt.setModifierMask(static_cast<uint8_t>(modifier_mask));
                        if (devid) {
                            alt.setDeviceIdentifier(std::string(devid));
                        }
@@ -984,8 +1013,10 @@ std::vector<Button> Config::getButtons(const std::string &gameName) {
                    button.setAnalogType((ButtonAnalogType) analogType);
                    button.setDebounceUp(debounce_up);
                    button.setDebounceDown(debounce_down);
+                    button.setBatThreshold(bat_threshold);
                    button.setVelocityThreshold(velocity_threshold);
                    button.setInvert(invert);
+                    button.setModifierMask(static_cast<uint8_t>(modifier_mask));
                    if (devid) {
                        button.setDeviceIdentifier(devid);
                    }
@@ -1130,7 +1161,7 @@ std::vector<Analog> Config::getAnalogs(const std::string &gameName) {
                bool smoothing = false;
                int multiplier = 1;
                bool relative_mode = false;
-                int delay_buffer_depth = 0;
+                uint32_t delay = 0;
                gameAnalogNode->QueryIntAttribute("index", &index);
                gameAnalogNode->QueryFloatAttribute("sensivity", &sensitivity);
                gameAnalogNode->QueryFloatAttribute("deadzone", &deadzone);
@@ -1139,7 +1170,7 @@ std::vector<Analog> Config::getAnalogs(const std::string &gameName) {
                gameAnalogNode->QueryBoolAttribute("smoothing", &smoothing);
                gameAnalogNode->QueryIntAttribute("multiplier", &multiplier);
                gameAnalogNode->QueryBoolAttribute("relative", &relative_mode);
-                gameAnalogNode->QueryIntAttribute("delay", &delay_buffer_depth);
+                gameAnalogNode->QueryUnsignedAttribute("delay_ms", &delay);
                const char *devid = gameAnalogNode->Attribute("devid");

                // create analog and add to list
@@ -1152,7 +1183,7 @@ std::vector<Analog> Config::getAnalogs(const std::string &gameName) {
                analog.setSmoothing(smoothing);
                analog.setMultiplier(multiplier);
                analog.setRelativeMode(relative_mode);
-                analog.setDelayBufferDepth(delay_buffer_depth);
+                analog.setDelayMs(delay);
                if (devid) {
                    analog.setDeviceIdentifier(devid);
                }
@@ -1312,17 +1343,32 @@ bool Config::firstFillConfigFile() {
 }

 void Config::saveConfigFile() {
-    // create a .tmp file and write to it...
+    // write the new config to a .tmp file first...
    const auto xml_result = this->configFile.SaveFile(this->configLocationTemp.c_str(), false);
    if (xml_result != tinyxml2::XMLError::XML_SUCCESS) {
-        log_info("cfg", "failed to write file: {}", this->configLocationTemp.string());
+        log_info("cfg", "failed to write file: {}", this->configLocationTemp);
        return;
    }
-    // copy the .tmp file to the main file...
-    if (CopyFileW(this->configLocationTemp.c_str(), this->configLocation.c_str(), false) == 0) {
-        log_warning("cfg", "CopyFileA failed: 0x{:08x}", GetLastError());
+
+    // ...flush the .tmp file to disk so a crash/power loss can't leave it half-written...
+    HANDLE tmp_handle = CreateFileW(
+        this->configLocationTemp.c_str(),
+        GENERIC_WRITE, FILE_SHARE_READ, nullptr,
+        OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, nullptr);
+    if (tmp_handle != INVALID_HANDLE_VALUE) {
+        FlushFileBuffers(tmp_handle);
+        CloseHandle(tmp_handle);
+    }
+
+    // ...then atomically replace the real config with the .tmp file.
+    // Unlike CopyFile (which truncates and rewrites the destination in place),
+    // an NTFS rename is atomic: the existing config is never left half-overwritten,
+    // so an interrupted save can't corrupt it. MoveFileEx also removes the .tmp on success.
+    if (MoveFileExW(
+            this->configLocationTemp.c_str(),
+            this->configLocation.c_str(),
+            MOVEFILE_REPLACE_EXISTING | MOVEFILE_WRITE_THROUGH) == 0) {
+        log_warning("cfg", "MoveFileExW failed: 0x{:08x}", GetLastError());
        return;
    }
-    // delete the .tmp file (not critical if this fails)
-    DeleteFileW(this->configLocationTemp.c_str());
-}
+}
@@ -22,7 +22,12 @@ public:
    bool getStatus();
    bool createConfigFile();

-    bool addGame(Game &game);
+    bool addGame(Game &game, bool save = true);
+
+    // flush pending in-memory changes to disk; intended to be paired with
+    // addGame(game, false) in bulk-init paths so we write the XML once instead
+    // of once per game
+    void save();

    bool updateBinding(const Game &game, const Button &button, int alternative);
    bool updateBinding(const Game &game, const Analog &analog);
@@ -1,11 +1,16 @@
-#include "configurator.h"
+#include "configurator.h"
+
+#include <d3d9.h>

 #include "overlay/overlay.h"
+#include "util/libutils.h"
+#include "util/logging.h"

 namespace cfg {

    // globals
    bool CONFIGURATOR_STANDALONE = false;
+    bool CONFIGURATOR_FORCE_SOFTWARE_RENDER = false;
    ConfigType CONFIGURATOR_TYPE = ConfigType::Config;

    Configurator::Configurator() {
@@ -16,13 +21,98 @@ namespace cfg {
        CONFIGURATOR_STANDALONE = false;
    }

+    // Attempt to bring up a D3D9 device backing the configurator's window so the
+    // overlay can use the hardware-accelerated imgui_impl_dx9 path instead of
+    // the CPU rasterizer. Returns true if both Direct3DCreate9 and CreateDevice
+    // succeed; the caller falls back to overlay::create_software() otherwise so
+    // that environments without D3D9 (Wine without dxvk, headless test boxes,
+    // GPUs whose drivers reject HAL) still get a working configurator.
+    static bool try_init_d3d9(ConfiguratorWindow &wnd) {
+        if (cfg::CONFIGURATOR_FORCE_SOFTWARE_RENDER) {
+            return false;
+        }
+
+        if (wnd.hWnd == nullptr) {
+            return false;
+        }
+
+        // load d3d9.dll dynamically rather than linking against it statically.
+        // a static import would force the system d3d9.dll to load at process
+        // startup for the main spice executable too (this file is shared with
+        // spice.exe), which loads system32\d3d9.dll before the modules directory
+        // is added to the DLL search path - preventing a user-supplied DXVK
+        // d3d9.dll in modules from ever loading for the game.
+        typedef IDirect3D9 *(WINAPI *Direct3DCreate9_t)(UINT);
+        HMODULE d3d9_module = libutils::try_library("d3d9.dll");
+        if (d3d9_module == nullptr) {
+            log_warning("configurator", "could not load d3d9.dll, falling back to software renderer");
+            return false;
+        }
+        auto Direct3DCreate9_fn =
+                reinterpret_cast<Direct3DCreate9_t>(libutils::try_proc(d3d9_module, "Direct3DCreate9"));
+        if (Direct3DCreate9_fn == nullptr) {
+            log_warning("configurator", "could not find Direct3DCreate9, falling back to software renderer");
+            return false;
+        }
+
+        IDirect3D9 *d3d = Direct3DCreate9_fn(D3D_SDK_VERSION);
+        if (d3d == nullptr) {
+            log_warning("configurator", "Direct3DCreate9 returned NULL, falling back to software renderer");
+            return false;
+        }
+
+        D3DPRESENT_PARAMETERS pp {};
+        pp.Windowed = TRUE;
+        pp.SwapEffect = D3DSWAPEFFECT_DISCARD;
+        // D3DFMT_UNKNOWN -> driver picks the current desktop format
+        pp.BackBufferFormat = D3DFMT_UNKNOWN;
+        pp.BackBufferWidth = static_cast<UINT>(wnd.client_width);
+        pp.BackBufferHeight = static_cast<UINT>(wnd.client_height);
+        pp.hDeviceWindow = wnd.hWnd;
+        pp.EnableAutoDepthStencil = FALSE;
+        pp.PresentationInterval = D3DPRESENT_INTERVAL_ONE;
+
+        IDirect3DDevice9 *device = nullptr;
+        // SOFTWARE_VERTEXPROCESSING is the most compatible behavior flag; the UI
+        // is tiny so we don't need hardware T&L. FPU_PRESERVE keeps our floating
+        // point environment intact in case other spice code relies on it.
+        const DWORD behavior_flags = D3DCREATE_SOFTWARE_VERTEXPROCESSING | D3DCREATE_FPU_PRESERVE;
+        HRESULT hr = d3d->CreateDevice(
+                D3DADAPTER_DEFAULT,
+                D3DDEVTYPE_HAL,
+                wnd.hWnd,
+                behavior_flags,
+                &pp,
+                &device);
+        if (FAILED(hr) || device == nullptr) {
+            log_warning("configurator",
+                "D3D9 CreateDevice failed (hr={:#x}), falling back to software renderer",
+                static_cast<unsigned int>(hr));
+            d3d->Release();
+            return false;
+        }
+
+        wnd.d3d = d3d;
+        wnd.device = device;
+        wnd.pp = pp;
+        wnd.use_d3d9 = true;
+        return true;
+    }
+
    void Configurator::run() {

-        // create instance
+        // bring up the overlay against either a real D3D9 device or the software
+        // rasterizer. The choice is one-shot - the in-game overlay always uses
+        // the same renderer the configurator picked here.
        overlay::ENABLED = true;
-        overlay::create_software(this->wnd.hWnd);
+        if (try_init_d3d9(this->wnd)) {
+            log_info("configurator", "using D3D9 hardware-accelerated renderer");
+            overlay::create_d3d9(this->wnd.hWnd, this->wnd.d3d, this->wnd.device);
+        } else {
+            log_info("configurator", "using software renderer");
+            overlay::create_software(this->wnd.hWnd);
+        }
        overlay::OVERLAY->set_active(true);
-        overlay::OVERLAY->hotkeys_enable = false;
        ImGui::GetIO().MouseDrawCursor = false;

        // run window
@@ -11,6 +11,7 @@ namespace cfg {
    // globals
    extern bool CONFIGURATOR_STANDALONE;
    extern ConfigType CONFIGURATOR_TYPE;
+    extern bool CONFIGURATOR_FORCE_SOFTWARE_RENDER;

    class Configurator {
    private:
@@ -1,11 +1,18 @@
-#include "configurator_wnd.h"
+#include "configurator_wnd.h"
+
+#include <algorithm>
+#include <cstring>

 #include <windows.h>
+#include <windowsx.h>

 #include "build/defs.h"
+#include "external/imgui/imgui.h"
 #include "launcher/shutdown.h"
 #include "overlay/overlay.h"
 #include "util/logging.h"
+#include "util/precise_timer.h"
+#include "util/utils.h"
 #include "cfg/configurator.h"

 #include "icon.h"
@@ -14,8 +21,177 @@ static const char *CLASS_NAME = "ConfiguratorWindow";
 static std::string WINDOW_TITLE;
 static int WINDOW_SIZE_X = 800;
 static int WINDOW_SIZE_Y = 600;
+static const int WINDOW_MIN_SIZE_X = 540;
+static const int WINDOW_MIN_SIZE_Y = 300;
 static HICON WINDOW_ICON = LoadIcon(GetModuleHandle(nullptr), MAKEINTRESOURCE(MAINICON));

+static const UINT_PTR RENDER_TIMER_ID = 1;
+
+// Map a virtual key code to the matching ImGuiKey. Mirrors the table used by
+// the in-game ImGui spice backend so navigation keys behave identically when
+// the configurator runs standalone and drives ImGui from Win32 messages.
+static ImGuiKey vk_to_imgui_key(WPARAM vkey) {
+    switch (vkey) {
+        case VK_TAB: return ImGuiKey_Tab;
+        case VK_LEFT: return ImGuiKey_LeftArrow;
+        case VK_RIGHT: return ImGuiKey_RightArrow;
+        case VK_UP: return ImGuiKey_UpArrow;
+        case VK_DOWN: return ImGuiKey_DownArrow;
+        case VK_PRIOR: return ImGuiKey_PageUp;
+        case VK_NEXT: return ImGuiKey_PageDown;
+        case VK_HOME: return ImGuiKey_Home;
+        case VK_END: return ImGuiKey_End;
+        case VK_INSERT: return ImGuiKey_Insert;
+        case VK_DELETE: return ImGuiKey_Delete;
+        case VK_BACK: return ImGuiKey_Backspace;
+        case VK_SPACE: return ImGuiKey_Space;
+        case VK_RETURN: return ImGuiKey_Enter;
+        case VK_ESCAPE: return ImGuiKey_Escape;
+        case VK_LSHIFT: return ImGuiKey_LeftShift;
+        case VK_RSHIFT: return ImGuiKey_RightShift;
+        case VK_SHIFT: return ImGuiKey_LeftShift;
+        case VK_LCONTROL: return ImGuiKey_LeftCtrl;
+        case VK_RCONTROL: return ImGuiKey_RightCtrl;
+        case VK_CONTROL: return ImGuiKey_LeftCtrl;
+        case 'A': return ImGuiKey_A;
+        case 'C': return ImGuiKey_C;
+        case 'V': return ImGuiKey_V;
+        case 'X': return ImGuiKey_X;
+        case 'Y': return ImGuiKey_Y;
+        case 'Z': return ImGuiKey_Z;
+        default: return ImGuiKey_None;
+    }
+}
+
+static ImGuiKey vk_to_imgui_mod_key(WPARAM vkey) {
+    switch (vkey) {
+        case VK_SHIFT:
+        case VK_LSHIFT:
+        case VK_RSHIFT:
+            return ImGuiMod_Shift;
+        case VK_CONTROL:
+        case VK_LCONTROL:
+        case VK_RCONTROL:
+            return ImGuiMod_Ctrl;
+        case VK_MENU:
+        case VK_LMENU:
+        case VK_RMENU:
+            return ImGuiMod_Alt;
+        default:
+            return ImGuiMod_None;
+    }
+}
+
+static cfg::ConfiguratorWindow *get_state(HWND hWnd) {
+    return reinterpret_cast<cfg::ConfiguratorWindow *>(
+        GetWindowLongPtrW(hWnd, GWLP_USERDATA));
+}
+
+// computes the top-left position that centers a window of the given size on
+// the primary monitor's work area (the desktop minus the taskbar); falls back
+// to (0, 0) if the monitor info can't be queried.
+static POINT center_on_primary_monitor(int width, int height) {
+    POINT pos = { 0, 0 };
+    HMONITOR mon = MonitorFromPoint(pos, MONITOR_DEFAULTTOPRIMARY);
+    MONITORINFO mi {};
+    mi.cbSize = sizeof(mi);
+    if (GetMonitorInfo(mon, &mi)) {
+        const int work_w = mi.rcWork.right - mi.rcWork.left;
+        const int work_h = mi.rcWork.bottom - mi.rcWork.top;
+        pos.x = mi.rcWork.left + (work_w - width) / 2;
+        pos.y = mi.rcWork.top + (work_h - height) / 2;
+    }
+    return pos;
+}
+
+// Returns the refresh rate (Hz) of the monitor the window currently lives on,
+// clamped to a sane range. Falls back to 60 if detection fails or the value
+// looks invalid (DEVMODE may report 0 or 1 to mean "use hardware default").
+static UINT detect_monitor_refresh_hz(HWND hWnd) {
+    UINT hz = 0;
+    HMONITOR mon = MonitorFromWindow(hWnd, MONITOR_DEFAULTTONEAREST);
+    if (mon) {
+        MONITORINFOEXW mi {};
+        mi.cbSize = sizeof(mi);
+        if (GetMonitorInfoW(mon, reinterpret_cast<LPMONITORINFO>(&mi))) {
+            DEVMODEW dm {};
+            dm.dmSize = sizeof(dm);
+            if (EnumDisplaySettingsW(mi.szDevice, ENUM_CURRENT_SETTINGS, &dm)) {
+                hz = dm.dmDisplayFrequency;
+            }
+        }
+    }
+    if (hz <= 1) {
+        HDC hdc = GetDC(hWnd);
+        if (hdc) {
+            int v = GetDeviceCaps(hdc, VREFRESH);
+            if (v > 1) {
+                hz = static_cast<UINT>(v);
+            }
+            ReleaseDC(hWnd, hdc);
+        }
+    }
+    if (hz < 30 || hz > 240) {
+        hz = 60;
+    }
+    return hz;
+}
+
+// Software-path WM_PAINT: blit the overlay's pixel buffer to the window using
+// SetDIBitsToDevice. Avoids creating/destroying a GDI HBITMAP every frame.
+static void paint_software(HWND hWnd) {
+    if (!overlay::OVERLAY) {
+        PAINTSTRUCT ps {};
+        BeginPaint(hWnd, &ps);
+        EndPaint(hWnd, &ps);
+        return;
+    }
+
+    int width = 0;
+    int height = 0;
+    uint32_t *pixel_data = overlay::OVERLAY->sw_get_pixel_data(&width, &height);
+
+    PAINTSTRUCT paint {};
+    HDC hdc = BeginPaint(hWnd, &paint);
+
+    if (pixel_data && width > 0 && height > 0) {
+        BITMAPINFO bmi {};
+        bmi.bmiHeader.biSize = sizeof(BITMAPINFOHEADER);
+        bmi.bmiHeader.biWidth = width;
+        // negative height -> top-down DIB, matching how ImGui packs pixels
+        bmi.bmiHeader.biHeight = -height;
+        bmi.bmiHeader.biPlanes = 1;
+        bmi.bmiHeader.biBitCount = 32;
+        bmi.bmiHeader.biCompression = BI_RGB;
+
+        SetDIBitsToDevice(
+            hdc,
+            0, 0,
+            width, height,
+            0, 0,
+            0, height,
+            pixel_data,
+            &bmi,
+            DIB_RGB_COLORS);
+    }
+
+    EndPaint(hWnd, &paint);
+}
+
+void cfg::ConfiguratorWindow::start_timer() {
+    if (!this->timer_running && this->hWnd) {
+        SetTimer(this->hWnd, RENDER_TIMER_ID, this->timer_interval_ms, nullptr);
+        this->timer_running = true;
+    }
+}
+
+void cfg::ConfiguratorWindow::stop_timer() {
+    if (this->timer_running && this->hWnd) {
+        KillTimer(this->hWnd, RENDER_TIMER_ID);
+        this->timer_running = false;
+    }
+}
+
 cfg::ConfiguratorWindow::ConfiguratorWindow() {

    // register the window class
@@ -27,12 +203,22 @@ cfg::ConfiguratorWindow::ConfiguratorWindow() {
    wc.hIcon = WINDOW_ICON;
    RegisterClass(&wc);

+    // raise SetTimer resolution so high refresh rates (120/144Hz) are actually
+    // achievable. Without this, USER timers quantize to ~15.6ms and cap us
+    // near ~64 FPS. Uses the shared helper so it respects -notimerhacks
+    // (Use Legacy Timers) and the Win11 timer-throttling opt-out. The helper
+    // intentionally never calls timeEndPeriod; the kernel releases the request
+    // when spicecfg exits.
+    timeutils::set_timer_resolution();
+
    // determine window title
    if (cfg::CONFIGURATOR_TYPE == cfg::ConfigType::Config) {
        WINDOW_TITLE = "spice2x config (" + to_string(VERSION_STRING_CFG) + ")";
        WINDOW_SIZE_X = 800;
        WINDOW_SIZE_Y = 600;
    }
+    this->client_width = WINDOW_SIZE_X;
+    this->client_height = WINDOW_SIZE_Y;

    // open window
    this->hWnd = CreateWindowEx(
@@ -48,27 +234,58 @@ cfg::ConfiguratorWindow::ConfiguratorWindow() {

    if (this->hWnd) {
        overlay::USE_WM_CHAR_FOR_IMGUI_CHAR_INPUT = true;
+        // force dark title bar
+        set_window_dark_titlebar(this->hWnd, true);
    }
 }

 cfg::ConfiguratorWindow::~ConfiguratorWindow() {

+    this->stop_timer();
+
    // close window
    DestroyWindow(this->hWnd);

    // unregister class
    UnregisterClass(CLASS_NAME, GetModuleHandle(NULL));
+
+    // release D3D9 resources if any
+    if (this->device) {
+        this->device->Release();
+        this->device = nullptr;
+    }
+    if (this->d3d) {
+        this->d3d->Release();
+        this->d3d = nullptr;
+    }
 }

 void cfg::ConfiguratorWindow::run() {

+    const POINT pos = center_on_primary_monitor(WINDOW_SIZE_X, WINDOW_SIZE_Y);
+    SetWindowPos(this->hWnd, HWND_TOP, pos.x, pos.y, WINDOW_SIZE_X, WINDOW_SIZE_Y, 0);
+
    // show window
-    SetWindowPos(this->hWnd, HWND_TOP, 0, 0, WINDOW_SIZE_X, WINDOW_SIZE_Y, 0);
    ShowWindow(this->hWnd, SW_SHOWNORMAL);
    UpdateWindow(this->hWnd);

-    // draw overlay in 60 FPS
-    SetTimer(this->hWnd, 1, 1000 / 60, nullptr);
+    // SW_SHOWNORMAL usually activates a top-level window, but not always (e.g.,
+    // when the launching process doesn't have foreground rights to transfer).
+    // Force foreground+focus explicitly so WM_MOUSEWHEEL is routed to us from
+    // the very first frame; the Win32 default routes wheel events to the
+    // keyboard-focused window.
+    SetForegroundWindow(this->hWnd);
+    SetFocus(this->hWnd);
+
+    // match the render timer to the monitor refresh rate so scrolling stays smooth
+    // on 60/120/144 Hz panels. Idle CPU is bounded by overlay::sw_pixels_dirty
+    // (software path) and overlay::d3d9_frame_dirty (D3D9 path); the timer is also
+    // paused entirely when the window is minimized (see WM_SIZE handler).
+    const UINT hz = detect_monitor_refresh_hz(this->hWnd);
+    this->timer_interval_ms = std::max<UINT>(1, 1000 / hz);
+    log_info("configurator", "render timer {} ms ({} Hz)",
+        this->timer_interval_ms, hz);
+    this->start_timer();

    // window loop
    BOOL ret;
@@ -94,6 +311,22 @@ LRESULT CALLBACK cfg::ConfiguratorWindow::window_proc(HWND hWnd, UINT uMsg, WPAR
            }
            break;
        }
+        case WM_GETMINMAXINFO: {
+
+            // enforce a minimum window size so the UI can't be shrunk to a
+            // point where the tabs/controls become unusable. The minimum is
+            // expressed in client-area pixels and converted to a full window
+            // size that accounts for the current frame/title-bar style.
+            RECT rc = { 0, 0, WINDOW_MIN_SIZE_X, WINDOW_MIN_SIZE_Y };
+            DWORD style = static_cast<DWORD>(GetWindowLongPtrW(hWnd, GWL_STYLE));
+            DWORD ex_style = static_cast<DWORD>(GetWindowLongPtrW(hWnd, GWL_EXSTYLE));
+            if (AdjustWindowRectEx(&rc, style, FALSE, ex_style)) {
+                auto *mmi = reinterpret_cast<MINMAXINFO *>(lParam);
+                mmi->ptMinTrackSize.x = rc.right - rc.left;
+                mmi->ptMinTrackSize.y = rc.bottom - rc.top;
+            }
+            break;
+        }
        case WM_CREATE: {

            // set user data of window to class pointer
@@ -102,6 +335,72 @@ LRESULT CALLBACK cfg::ConfiguratorWindow::window_proc(HWND hWnd, UINT uMsg, WPAR

            break;
        }
+        case WM_SIZE: {
+
+            // pause/resume the render timer based on visibility
+            auto *state = get_state(hWnd);
+            if (state) {
+                if (wParam == SIZE_MINIMIZED) {
+                    state->window_minimized = true;
+                    state->stop_timer();
+                } else {
+                    if (state->window_minimized) {
+                        state->window_minimized = false;
+                        // force a full repaint on the next frame
+                        state->has_valid_draw_hash = false;
+                        state->start_timer();
+                    }
+                    const int new_w = LOWORD(lParam);
+                    const int new_h = HIWORD(lParam);
+                    if (new_w > 0 && new_h > 0
+                            && (new_w != state->client_width || new_h != state->client_height)) {
+                        state->client_width = new_w;
+                        state->client_height = new_h;
+
+                        // reset the D3D9 device with the new back-buffer size so the
+                        // hardware-accelerated path matches the window dimensions.
+                        if (state->use_d3d9 && state->device) {
+                            if (overlay::OVERLAY) {
+                                overlay::OVERLAY->reset_invalidate();
+                            }
+                            state->pp.BackBufferWidth = static_cast<UINT>(new_w);
+                            state->pp.BackBufferHeight = static_cast<UINT>(new_h);
+                            HRESULT hr = state->device->Reset(&state->pp);
+                            if (FAILED(hr)) {
+                                log_warning("configurator", "D3D9 device Reset failed, hr={:#x}",
+                                    static_cast<unsigned int>(hr));
+                            }
+                            if (overlay::OVERLAY) {
+                                overlay::OVERLAY->reset_recreate();
+                            }
+                        }
+
+                        // force a full repaint after resize
+                        state->has_valid_draw_hash = false;
+                    }
+                }
+            }
+            break;
+        }
+        case WM_DISPLAYCHANGE: {
+
+            // monitor refresh rate may have changed (or the window moved to a
+            // different monitor); re-detect and re-arm the render timer.
+            auto *state = get_state(hWnd);
+            if (state && state->timer_running) {
+                const UINT hz = detect_monitor_refresh_hz(hWnd);
+                const UINT new_interval = std::max<UINT>(1, 1000 / hz);
+                if (new_interval != state->timer_interval_ms) {
+                    state->stop_timer();
+                    state->timer_interval_ms = new_interval;
+                    state->start_timer();
+                    log_info("configurator",
+                        "WM_DISPLAYCHANGE: render timer {} ms ({} Hz)",
+                        new_interval, hz);
+                }
+            }
+            break;
+        }
        case WM_CLOSE:
        case WM_DESTROY: {

@@ -110,62 +409,192 @@ LRESULT CALLBACK cfg::ConfiguratorWindow::window_proc(HWND hWnd, UINT uMsg, WPAR
            break;
        }
        case WM_TIMER: {
+            if (wParam != RENDER_TIMER_ID) {
+                break;
+            }

-            // update overlay
+            auto *state = get_state(hWnd);
+
+            // skip rendering when the window is minimized or hidden - the timer is
+            // already paused on minimize, but defensively skip here as well.
+            if (state && (state->window_minimized || !IsWindowVisible(hWnd))) {
+                break;
+            }
+
+            const bool use_d3d9 = state && state->use_d3d9 && state->device;
+
+            // build the imgui frame (input is already buffered via WM_* messages)
            if (overlay::OVERLAY) {
                overlay::OVERLAY->update();
                overlay::OVERLAY->set_active(true);
                overlay::OVERLAY->new_frame();
-                overlay::OVERLAY->render();
            }

-            // repaint window
-            InvalidateRect(hWnd, nullptr, TRUE);
+            if (use_d3d9) {
+                const HRESULT cl = state->device->TestCooperativeLevel();
+                if (cl == D3DERR_DEVICELOST) {
+                    break;
+                }
+                if (cl == D3DERR_DEVICENOTRESET) {
+                    if (overlay::OVERLAY) {
+                        overlay::OVERLAY->reset_invalidate();
+                    }
+                    const HRESULT reset_hr = state->device->Reset(&state->pp);
+                    if (SUCCEEDED(reset_hr) && overlay::OVERLAY) {
+                        overlay::OVERLAY->reset_recreate();
+                    }
+                    break;
+                }
+                if (FAILED(cl)) {
+                    break;
+                }
+
+                if (overlay::OVERLAY) {
+                    overlay::OVERLAY->render();
+                }
+
+                // skip Clear/Present when ImGui draw data is unchanged; the last
+                // presented frame stays visible (same fast path as sw_pixels_dirty).
+                const bool dirty = overlay::OVERLAY && overlay::OVERLAY->d3d9_frame_dirty;
+                const bool force_present = state && !state->has_valid_draw_hash;
+                if (dirty || force_present) {
+                    if (state) {
+                        state->has_valid_draw_hash = true;
+                    }
+                    state->device->Clear(0, nullptr, D3DCLEAR_TARGET,
+                        D3DCOLOR_RGBA(20, 18, 18, 255), 1.0f, 0);
+                    if (SUCCEEDED(state->device->BeginScene())) {
+                        if (overlay::OVERLAY) {
+                            overlay::OVERLAY->d3d9_render_draw(force_present);
+                        }
+                        state->device->EndScene();
+                    }
+                    const HRESULT hr = state->device->Present(nullptr, nullptr, nullptr, nullptr);
+                    if (hr == D3DERR_DEVICELOST) {
+                        break;
+                    }
+                }
+            } else {
+                if (overlay::OVERLAY) {
+                    overlay::OVERLAY->render();
+                }
+                // software path: the overlay's renderer already skipped the costly
+                // paint_imgui call when the draw data was unchanged. Only invalidate
+                // the window when those pixels actually changed, or on the first
+                // frame / after a resize when we lost the previously cached state.
+                const bool dirty = overlay::OVERLAY && overlay::OVERLAY->sw_pixels_dirty;
+                const bool force_repaint = state && !state->has_valid_draw_hash;
+                if (dirty || force_repaint) {
+                    if (state) {
+                        state->has_valid_draw_hash = true;
+                    }
+                    // pass FALSE for bErase - WM_ERASEBKGND already short-circuits, so
+                    // skipping the erase region avoids one extra Win32 round trip.
+                    InvalidateRect(hWnd, nullptr, FALSE);
+                }
+            }
+
            break;
        }
        case WM_ERASEBKGND: {
            return 1;
        }
        case WM_PAINT: {
-
-            // render overlay
-            if (overlay::OVERLAY) {
-
-                // get pixel data
-                int width, height;
-                uint32_t *pixel_data = overlay::OVERLAY->sw_get_pixel_data(&width, &height);
-                if (width > 0 && height > 0) {
-
-                    // create bitmap
-                    HBITMAP bitmap = CreateBitmap(width, height, 1, 8 * sizeof(uint32_t), pixel_data);
-
-                    // prepare paint
-                    PAINTSTRUCT paint{};
-                    HDC hdc = BeginPaint(hWnd, &paint);
-                    HDC hdcMem = CreateCompatibleDC(hdc);
-                    SetBkMode(hdc, TRANSPARENT);
-
-                    // draw bitmap
-                    SelectObject(hdcMem, bitmap);
-                    BitBlt(hdc, paint.rcPaint.left, paint.rcPaint.top,
-                            paint.rcPaint.right - paint.rcPaint.left,
-                            paint.rcPaint.bottom - paint.rcPaint.top,
-                            hdcMem, paint.rcPaint.left, paint.rcPaint.top, SRCCOPY);
-
-                    // delete bitmap
-                    DeleteObject(bitmap);
-
-                    // clean up
-                    DeleteDC(hdcMem);
-                    EndPaint(hWnd, &paint);
-                } else {
-                    return DefWindowProc(hWnd, uMsg, wParam, lParam);
-                }
+            paint_software(hWnd);
+            break;
+        }
+        case WM_ACTIVATE: {
+            const WORD activation = LOWORD(wParam);
+            if (activation == WA_INACTIVE) {
+                // dropping any held mouse buttons so we don't get stuck in a
+                // "button still pressed" state when we come back.
+                auto &io = ImGui::GetIO();
+                io.AddMouseButtonEvent(ImGuiMouseButton_Left, false);
+                io.AddMouseButtonEvent(ImGuiMouseButton_Right, false);
+                io.AddMouseButtonEvent(ImGuiMouseButton_Middle, false);
            } else {
-                return DefWindowProc(hWnd, uMsg, wParam, lParam);
+                // WA_ACTIVE or WA_CLICKACTIVE: make sure the keyboard focus is
+                // actually on this window so WM_MOUSEWHEEL gets routed to us
+                // again. Without this, scrolling silently stops working after
+                // the user alt-tabs back to spicecfg.
+                SetFocus(hWnd);
            }
            break;
        }
+        case WM_KILLFOCUS: {
+            // mirror the WA_INACTIVE mouse-button drop for the keyboard -
+            // without this, a key held during alt-tab stays "down" in ImGui
+            // state until the user presses and releases it again.
+            ImGui::GetIO().ClearInputKeys();
+            break;
+        }
+        // input messages routed straight into ImGui IO. This replaces the previous
+        // per-frame 256-VK rawinput scan in ImGui_ImplSpice_NewFrame for the
+        // standalone configurator (see CONFIGURATOR_STANDALONE branches there).
+        case WM_KEYDOWN:
+        case WM_SYSKEYDOWN:
+        case WM_KEYUP:
+        case WM_SYSKEYUP: {
+            if (wParam == VK_F4) {
+                return DefWindowProc(hWnd, uMsg, wParam, lParam);
+            }
+            const bool down = (uMsg == WM_KEYDOWN) || (uMsg == WM_SYSKEYDOWN);
+            auto &io = ImGui::GetIO();
+            const ImGuiKey key = vk_to_imgui_key(wParam);
+            if (key != ImGuiKey_None) {
+                io.AddKeyEvent(key, down);
+            }
+            const ImGuiKey mod = vk_to_imgui_mod_key(wParam);
+            if (mod != ImGuiMod_None) {
+                io.AddKeyEvent(mod, down);
+            }
+            break;
+        }
+        case WM_MOUSEMOVE: {
+            auto &io = ImGui::GetIO();
+            io.AddMousePosEvent(static_cast<float>(GET_X_LPARAM(lParam)),
+                                static_cast<float>(GET_Y_LPARAM(lParam)));
+            break;
+        }
+        case WM_LBUTTONDOWN:
+        case WM_LBUTTONUP:
+        case WM_RBUTTONDOWN:
+        case WM_RBUTTONUP:
+        case WM_MBUTTONDOWN:
+        case WM_MBUTTONUP: {
+            int button = 0;
+            bool down = false;
+            switch (uMsg) {
+                case WM_LBUTTONDOWN: button = ImGuiMouseButton_Left; down = true; break;
+                case WM_LBUTTONUP: button = ImGuiMouseButton_Left; down = false; break;
+                case WM_RBUTTONDOWN: button = ImGuiMouseButton_Right; down = true; break;
+                case WM_RBUTTONUP: button = ImGuiMouseButton_Right; down = false; break;
+                case WM_MBUTTONDOWN: button = ImGuiMouseButton_Middle; down = true; break;
+                case WM_MBUTTONUP: button = ImGuiMouseButton_Middle; down = false; break;
+            }
+            auto &io = ImGui::GetIO();
+            io.AddMouseButtonEvent(button, down);
+            if (down) {
+                SetCapture(hWnd);
+            } else {
+                ReleaseCapture();
+            }
+            break;
+        }
+        case WM_MOUSEWHEEL: {
+            const float delta = static_cast<float>(GET_WHEEL_DELTA_WPARAM(wParam))
+                / static_cast<float>(WHEEL_DELTA);
+            ImGui::GetIO().AddMouseWheelEvent(0.0f, delta);
+            break;
+        }
+#if !SPICE_XP
+        case WM_MOUSEHWHEEL: {
+            const float delta = static_cast<float>(GET_WHEEL_DELTA_WPARAM(wParam))
+                / static_cast<float>(WHEEL_DELTA);
+            ImGui::GetIO().AddMouseWheelEvent(delta, 0.0f);
+            break;
+        }
+#endif
        default:
            return DefWindowProc(hWnd, uMsg, wParam, lParam);
    }
@@ -1,6 +1,9 @@
-#pragma once
+#pragma once
+
+#include <cstdint>

 #include <windows.h>
+#include <d3d9.h>

 namespace cfg {

@@ -9,11 +12,38 @@ namespace cfg {

        HWND hWnd;

+        // optional D3D9 device backing the configurator window; nullptr when running
+        // the software-rendered path. Owned by ConfiguratorWindow when set.
+        IDirect3D9 *d3d = nullptr;
+        IDirect3DDevice9 *device = nullptr;
+        D3DPRESENT_PARAMETERS pp {};
+        bool use_d3d9 = false;
+
+        // throttling / pause state. timer_interval_ms is the default until run()
+        // refines it to the actual monitor refresh rate (see detect_monitor_refresh_hz).
+        UINT timer_interval_ms = 1000 / 60;
+        bool timer_running = false;
+        bool window_minimized = false;
+
+        // tracks whether we've issued at least one InvalidateRect since window
+        // creation/resize. The overlay's per-frame "pixels changed" flag suppresses
+        // idle blits; this flag forces the very first blit on startup or after
+        // a resize so the window doesn't show garbage until the user moves the mouse.
+        bool has_valid_draw_hash = false;
+
+        // dimensions of the configurator client area (kept in sync with WM_SIZE)
+        int client_width = 0;
+        int client_height = 0;
+
        ConfiguratorWindow();
        ~ConfiguratorWindow();

        void run();

+        // start/stop the render timer based on visibility state
+        void start_timer();
+        void stop_timer();
+
        static LRESULT CALLBACK window_proc(HWND hwnd, UINT uMsg, WPARAM wParam, LPARAM lParam);
    };
 }
@@ -24,7 +24,9 @@ namespace overlay::windows {
        el->SetAttribute("invert", entry.invert);
        el->SetAttribute("debounce_up", entry.debounce_up);
        el->SetAttribute("debounce_down", entry.debounce_down);
+        el->SetAttribute("bat_threshold", entry.bat_threshold);
        el->SetAttribute("velocity_threshold", entry.velocity_threshold);
+        el->SetAttribute("modifiers", entry.modifier_mask);
        parent->InsertEndChild(el);
    }

@@ -49,6 +51,14 @@ namespace overlay::windows {
        el->QueryIntAttribute("velocity_threshold", &vel);
        entry.velocity_threshold = (unsigned short)vel;

+        int bat = 0;
+        el->QueryIntAttribute("bat_threshold", &bat);
+        entry.bat_threshold = bat;
+
+        unsigned int modifier_mask = 0;
+        el->QueryUnsignedAttribute("modifiers", &modifier_mask);
+        entry.modifier_mask = static_cast<uint8_t>(modifier_mask & UINT8_C(0x0F));
+
        return entry;
    }

@@ -65,7 +75,7 @@ namespace overlay::windows {
        el->SetAttribute("smoothing", analog.smoothing);
        el->SetAttribute("multiplier", analog.multiplier);
        el->SetAttribute("relative", analog.relative_mode);
-        el->SetAttribute("delay", analog.delay_buffer_depth);
+        el->SetAttribute("delay_ms", analog.delay_ms);
        parent->InsertEndChild(el);
    }

@@ -88,7 +98,7 @@ namespace overlay::windows {
        el->QueryBoolAttribute("smoothing", &a.smoothing);
        el->QueryIntAttribute("multiplier", &a.multiplier);
        el->QueryBoolAttribute("relative", &a.relative_mode);
-        el->QueryIntAttribute("delay", &a.delay_buffer_depth);
+        el->QueryUnsignedAttribute("delay_ms", &a.delay_ms);

        return a;
    }
@@ -155,6 +165,28 @@ namespace overlay::windows {
                }
            }

+            // modifier buttons
+            auto *mod_el = tmpl_el->FirstChildElement("modifier_buttons");
+            if (mod_el) {
+                auto *btn_el = mod_el->FirstChildElement("button");
+                while (btn_el) {
+                    const char *btn_name = btn_el->Attribute("name");
+                    std::string btn_name_str = btn_name ? btn_name : "";
+
+                    TemplateButtonBinding *binding = nullptr;
+                    for (auto &b : tmpl.modifier_buttons) {
+                        if (b.name == btn_name_str) { binding = &b; break; }
+                    }
+                    if (!binding) {
+                        tmpl.modifier_buttons.push_back(
+                            {btn_name_str, read_button_entry(btn_el), {}});
+                    } else {
+                        binding->alternatives.push_back(read_button_entry(btn_el));
+                    }
+                    btn_el = btn_el->NextSiblingElement("button");
+                }
+            }
+
            // keypad buttons
            auto *kp_el = tmpl_el->FirstChildElement("keypad_buttons");
            if (kp_el) {
@@ -239,6 +271,7 @@ namespace overlay::windows {
    bool save_user_template(
        const ControllerTemplate &tmpl,
        const bool save_buttons,
+        const bool save_modifiers,
        const bool save_keypads,
        const bool save_analogs,
        const bool save_lights) {
@@ -295,6 +328,24 @@ namespace overlay::windows {
        }
        tmpl_el->InsertEndChild(buttons_el);

+        // modifier buttons - skip unbound entries
+        if (!tmpl.modifier_buttons.empty()) {
+            auto *mod_el = doc.NewElement("modifier_buttons");
+            if (save_modifiers) {
+                for (auto &btn : tmpl.modifier_buttons) {
+                    if (!btn.primary.is_unbound()) {
+                        write_button_entry(doc, mod_el, btn.name, btn.primary);
+                    }
+                    for (auto &alt : btn.alternatives) {
+                        if (!alt.is_unbound()) {
+                            write_button_entry(doc, mod_el, btn.name, alt);
+                        }
+                    }
+                }
+            }
+            tmpl_el->InsertEndChild(mod_el);
+        }
+
        // keypad buttons — skip unbound entries
        if (!tmpl.keypad_buttons.empty()) {
            auto *kp_el = doc.NewElement("keypad_buttons");
@@ -354,13 +405,12 @@ namespace overlay::windows {
            return false;
        }

-        std::error_code ec;
-        std::filesystem::rename(path_tmp, path, ec);
-        if (ec) {
-            log_warning("templates", "failed to rename templates file: {}", ec.message());
+        if (MoveFileExW(path_tmp.c_str(), path.c_str(), MOVEFILE_REPLACE_EXISTING | MOVEFILE_WRITE_THROUGH) == 0) {
+            log_warning("templates", "failed to rename templates file: 0x{:08x}", GetLastError());
            return false;
        }

+        log_info("templates", "templates file saved successfully");
        return true;
    }

@@ -392,8 +442,7 @@ namespace overlay::windows {
            auto path_tmp = path;
            path_tmp.replace_extension(L"tmp");
            if (doc.SaveFile(path_tmp.c_str()) == tinyxml2::XML_SUCCESS) {
-                std::error_code ec;
-                std::filesystem::rename(path_tmp, path, ec);
+                MoveFileExW(path_tmp.c_str(), path.c_str(), MOVEFILE_REPLACE_EXISTING | MOVEFILE_WRITE_THROUGH);
            }
        }

@@ -429,8 +478,7 @@ namespace overlay::windows {
            auto path_tmp = path;
            path_tmp.replace_extension(L"tmp");
            if (doc.SaveFile(path_tmp.c_str()) == tinyxml2::XML_SUCCESS) {
-                std::error_code ec;
-                std::filesystem::rename(path_tmp, path, ec);
+                MoveFileExW(path_tmp.c_str(), path.c_str(), MOVEFILE_REPLACE_EXISTING | MOVEFILE_WRITE_THROUGH);
            }
        }

@@ -85,6 +85,15 @@ std::string Light::getDisplayString(rawinput::RawInputManager* manager) {

            return "Invalid SMX Dedicab Light (" + index_string + ")";
        }
+        case rawinput::XINPUT_GAMEPAD: {
+            if (index < static_cast<size_t>(xinput::XInputOutputEnum::COUNT)) {
+                return fmt::format("{} ({})",
+                    xinput::get_output_string(static_cast<xinput::XInputOutputEnum>(index)),
+                    device->desc);
+            }
+
+            return "Invalid XInput Gamepad Light (" + index_string + ")";
+        }
        case rawinput::DESTROYED:
            return "Unplugged device (" + index_string + ")";
        default:
@@ -20,10 +20,10 @@
        </dependentAssembly>
    </dependency>
    <trustInfo xmlns="urn:schemas-microsoft-com:asm.v2">
-            <security>
-                <requestedPrivileges>
-                    <requestedExecutionLevel level="requireAdministrator" uiAccess="false"/>
-                </requestedPrivileges>
-            </security>
+        <security>
+            <requestedPrivileges>
+                <requestedExecutionLevel level="asInvoker" uiAccess="false"/>
+            </requestedPrivileges>
+        </security>
    </trustInfo>
 </assembly>
@@ -20,6 +20,7 @@ enum class OptionPickerType {
    CpuAffinity,
    FilePath,
    DirectoryPath,
+    Monitor,
 };

 struct OptionDefinition {
@@ -45,6 +46,8 @@ struct OptionDefinition {

    // for OptionPickerType::FilePath
    std::string file_extension = "";
+
+    std::string quick_setting_category = "";
 };

 class Option {
@@ -21,7 +21,7 @@ namespace cfg {
        if (SCREEN_RESIZE_CFG_PATH_OVERRIDE.has_value()) {
            this->config_path = SCREEN_RESIZE_CFG_PATH_OVERRIDE.value();
            if (fileutils::file_exists(this->config_path)) {
-                log_info("ScreenResize", "loading config from: {}", this->config_path.string());
+                log_info("ScreenResize", "loading config from: {}", this->config_path);
                file_exists = true;
            }
        } else {
@@ -35,7 +35,7 @@ namespace cfg {

    ScreenResize::~ScreenResize() {
    }
-    
+
    void ScreenResize::config_load() {
        std::string config = fileutils::text_read(this->config_path);
        if (config.empty()) {
@@ -78,7 +78,7 @@ namespace cfg {
            eamuse_get_game(),
            use_game_setting,
            root);
-        
+
        load_bool_value(doc, root + "enable_screen_resize", this->enable_screen_resize);
        if (this->enable_screen_resize) {
            log_misc("ScreenResize", "enabled by config file");
@@ -60,6 +60,7 @@ namespace cfg {
        //     window = rectangle including the frame
        //     client = just the content area without frames.
        bool window_always_on_top = false;
+        bool window_disable_round_corners = false;
        bool client_keep_aspect_ratio = true;
        bool enable_window_resize = false;
        int window_decoration = 0; // enum type WindowDecorationMode
@@ -0,0 +1,72 @@
+# fails the build if a PE binary statically imports a forbidden DLL.
+#
+# some DLLs must never end up in spice's static import table, for two reasons:
+#
+#  1. user-overridable DLLs (e.g. DXVK's d3d9.dll): users drop their own copy
+#     into the modules directory to replace the system one. a static import
+#     forces the loader to load the SYSTEM copy at process startup - before the
+#     modules directory is added to the DLL search path and before the game DLL
+#     loads - so the user-supplied override never takes effect (see issue #779).
+#
+#  2. DLLs that break games when present (e.g. Media Foundation: mf/mfplat/
+#     mfreadwrite): a static import loads them eagerly and breaks Unity games.
+#
+# in both cases the DLL must instead be loaded dynamically (libutils::try_library
+# / GetProcAddress / delay load) so it is only pulled in when actually needed.
+#
+# invoked via `cmake -P` from a POST_BUILD step. required -D variables:
+#   OBJDUMP     - path to objdump (CMAKE_OBJDUMP)
+#   TARGET_FILE - path to the PE binary to inspect
+#   FORBIDDEN   - semicolon-separated list of lowercase DLL names to reject
+
+if(NOT OBJDUMP OR NOT EXISTS "${OBJDUMP}")
+    message(WARNING
+        "check_no_static_dll_imports: objdump not found, skipping import check for ${TARGET_FILE}")
+    return()
+endif()
+
+execute_process(
+    COMMAND "${OBJDUMP}" -p "${TARGET_FILE}"
+    OUTPUT_VARIABLE dump_output
+    RESULT_VARIABLE dump_result
+    ERROR_VARIABLE dump_error)
+
+if(NOT dump_result EQUAL 0)
+    message(WARNING
+        "check_no_static_dll_imports: objdump failed for ${TARGET_FILE}: ${dump_error}")
+    return()
+endif()
+
+# both GNU objdump and llvm-objdump print one "DLL Name: <name>" line per
+# statically imported DLL in their PE private-header dump.
+string(REGEX MATCHALL "DLL Name:[ \t]*[^\n\r]+" dll_lines "${dump_output}")
+
+set(violations "")
+foreach(line IN LISTS dll_lines)
+    string(REGEX REPLACE "DLL Name:[ \t]*" "" dll_name "${line}")
+    string(STRIP "${dll_name}" dll_name)
+    string(TOLOWER "${dll_name}" dll_name_lower)
+    if(dll_name_lower IN_LIST FORBIDDEN)
+        list(APPEND violations "${dll_name}")
+    endif()
+endforeach()
+
+if(violations)
+    list(REMOVE_DUPLICATES violations)
+    string(REPLACE ";" ", " violations_str "${violations}")
+    message(FATAL_ERROR
+        "static DLL import check FAILED for ${TARGET_FILE}\n"
+        "  forbidden static imports found: ${violations_str}\n"
+        "\n"
+        "  these DLLs must never be statically imported by spice:\n"
+        "    * user-overridable DLLs (e.g. DXVK d3d9.dll) - a static import loads the\n"
+        "      system copy at startup and preempts the modules override (issue #779).\n"
+        "    * Media Foundation DLLs (mf/mfplat/mfreadwrite) - a static import breaks\n"
+        "      Unity games.\n"
+        "\n"
+        "  fix: load the DLL dynamically instead - replace the direct API call with a\n"
+        "  libutils::try_library() + libutils::try_proc() lookup (or a delay load), then\n"
+        "  call through the resolved function pointer.")
+endif()
+
+message(STATUS "static DLL import check passed for ${TARGET_FILE}")
@@ -7,6 +7,7 @@ RUN pacman --noconfirm -Syu git \
    ninja \
    cmake \
    unzip \
+    nasm \
    mingw-w64-crt \
    mingw-w64-winpthreads \
    mingw-w64-gcc \
@@ -15,3 +16,11 @@ RUN pacman --noconfirm -Syu git \
    ccache
 RUN useradd user -m && echo "ALL ALL=(ALL) NOPASSWD: ALL" >> /etc/sudoers
 RUN su user -c "cd /tmp/ && git clone https://aur.archlinux.org/yay-bin.git --depth=1 && cd yay-bin && makepkg --noconfirm -si && yay --noconfirm -S mingw-w64-cmake"
+
+RUN mkdir -p /opt/llvm-mingw-xp \
+    && curl -fsSL "https://github.com/mon/llvm-mingw-xp/releases/download/llvm-mingw-xp-22.1.0/llvm-mingw-llvm-mingw-xp-22.1.0-msvcrt-ubuntu-22.04-x86_64.tar.xz" \
+    | tar -xJ --strip-components=1 -C /opt/llvm-mingw-xp
+ENV PATH="$PATH:/opt/llvm-mingw-xp/bin"
+
+RUN curl -fsSL "https://github.com/mon/windows-dll-compat-checker/releases/download/v1.3/windows_dll_compat_checker-linux-x86_64.tar.xz" \
+    | tar -xJ -C /usr/local/bin
@@ -1,95 +1,95 @@
-Copyright (c) 2017, keshikan (http://www.keshikan.net),
-with Reserved Font Name "DSEG".
-
-
-This Font Software is licensed under the SIL Open Font License, Version 1.1.
-This license is copied below, and is also available with a FAQ at:
-http://scripts.sil.org/OFL
-
-
-----------------------------------------------------------
-SIL OPEN FONT LICENSE Version 1.1 - 26 February 2007
-----------------------------------------------------------
-
-PREAMBLE
-The goals of the Open Font License (OFL) are to stimulate worldwide
-development of collaborative font projects, to support the font creation
-efforts of academic and linguistic communities, and to provide a free and
-open framework in which fonts may be shared and improved in partnership
-with others.
-
-The OFL allows the licensed fonts to be used, studied, modified and
-redistributed freely as long as they are not sold by themselves. The
-fonts, including any derivative works, can be bundled, embedded, 
-redistributed and/or sold with any software provided that any reserved
-names are not used by derivative works. The fonts and derivatives,
-however, cannot be released under any other type of license. The
-requirement for fonts to remain under this license does not apply
-to any document created using the fonts or their derivatives.
-
-DEFINITIONS
-"Font Software" refers to the set of files released by the Copyright
-Holder(s) under this license and clearly marked as such. This may
-include source files, build scripts and documentation.
-
-"Reserved Font Name" refers to any names specified as such after the
-copyright statement(s).
-
-"Original Version" refers to the collection of Font Software components as
-distributed by the Copyright Holder(s).
-
-"Modified Version" refers to any derivative made by adding to, deleting,
-or substituting -- in part or in whole -- any of the components of the
-Original Version, by changing formats or by porting the Font Software to a
-new environment.
-
-"Author" refers to any designer, engineer, programmer, technical
-writer or other person who contributed to the Font Software.
-
-PERMISSION & CONDITIONS
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of the Font Software, to use, study, copy, merge, embed, modify,
-redistribute, and sell modified and unmodified copies of the Font
-Software, subject to the following conditions:
-
-1) Neither the Font Software nor any of its individual components,
-in Original or Modified Versions, may be sold by itself.
-
-2) Original or Modified Versions of the Font Software may be bundled,
-redistributed and/or sold with any software, provided that each copy
-contains the above copyright notice and this license. These can be
-included either as stand-alone text files, human-readable headers or
-in the appropriate machine-readable metadata fields within text or
-binary files as long as those fields can be easily viewed by the user.
-
-3) No Modified Version of the Font Software may use the Reserved Font
-Name(s) unless explicit written permission is granted by the corresponding
-Copyright Holder. This restriction only applies to the primary font name as
-presented to the users.
-
-4) The name(s) of the Copyright Holder(s) or the Author(s) of the Font
-Software shall not be used to promote, endorse or advertise any
-Modified Version, except to acknowledge the contribution(s) of the
-Copyright Holder(s) and the Author(s) or with their explicit written
-permission.
-
-5) The Font Software, modified or unmodified, in part or in whole,
-must be distributed entirely under this license, and must not be
-distributed under any other license. The requirement for fonts to
-remain under this license does not apply to any document created
-using the Font Software.
-
-TERMINATION
-This license becomes null and void if any of the above conditions are
-not met.
-
-DISCLAIMER
-THE FONT SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT
-OF COPYRIGHT, PATENT, TRADEMARK, OR OTHER RIGHT. IN NO EVENT SHALL THE
-COPYRIGHT HOLDER BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
-INCLUDING ANY GENERAL, SPECIAL, INDIRECT, INCIDENTAL, OR CONSEQUENTIAL
-DAMAGES, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
-FROM, OUT OF THE USE OR INABILITY TO USE THE FONT SOFTWARE OR FROM
-OTHER DEALINGS IN THE FONT SOFTWARE.
+Copyright (c) 2017, keshikan (http://www.keshikan.net),
+with Reserved Font Name "DSEG".
+
+
+This Font Software is licensed under the SIL Open Font License, Version 1.1.
+This license is copied below, and is also available with a FAQ at:
+http://scripts.sil.org/OFL
+
+
+-----------------------------------------------------------
+SIL OPEN FONT LICENSE Version 1.1 - 26 February 2007
+-----------------------------------------------------------
+
+PREAMBLE
+The goals of the Open Font License (OFL) are to stimulate worldwide
+development of collaborative font projects, to support the font creation
+efforts of academic and linguistic communities, and to provide a free and
+open framework in which fonts may be shared and improved in partnership
+with others.
+
+The OFL allows the licensed fonts to be used, studied, modified and
+redistributed freely as long as they are not sold by themselves. The
+fonts, including any derivative works, can be bundled, embedded, 
+redistributed and/or sold with any software provided that any reserved
+names are not used by derivative works. The fonts and derivatives,
+however, cannot be released under any other type of license. The
+requirement for fonts to remain under this license does not apply
+to any document created using the fonts or their derivatives.
+
+DEFINITIONS
+"Font Software" refers to the set of files released by the Copyright
+Holder(s) under this license and clearly marked as such. This may
+include source files, build scripts and documentation.
+
+"Reserved Font Name" refers to any names specified as such after the
+copyright statement(s).
+
+"Original Version" refers to the collection of Font Software components as
+distributed by the Copyright Holder(s).
+
+"Modified Version" refers to any derivative made by adding to, deleting,
+or substituting -- in part or in whole -- any of the components of the
+Original Version, by changing formats or by porting the Font Software to a
+new environment.
+
+"Author" refers to any designer, engineer, programmer, technical
+writer or other person who contributed to the Font Software.
+
+PERMISSION & CONDITIONS
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of the Font Software, to use, study, copy, merge, embed, modify,
+redistribute, and sell modified and unmodified copies of the Font
+Software, subject to the following conditions:
+
+1) Neither the Font Software nor any of its individual components,
+in Original or Modified Versions, may be sold by itself.
+
+2) Original or Modified Versions of the Font Software may be bundled,
+redistributed and/or sold with any software, provided that each copy
+contains the above copyright notice and this license. These can be
+included either as stand-alone text files, human-readable headers or
+in the appropriate machine-readable metadata fields within text or
+binary files as long as those fields can be easily viewed by the user.
+
+3) No Modified Version of the Font Software may use the Reserved Font
+Name(s) unless explicit written permission is granted by the corresponding
+Copyright Holder. This restriction only applies to the primary font name as
+presented to the users.
+
+4) The name(s) of the Copyright Holder(s) or the Author(s) of the Font
+Software shall not be used to promote, endorse or advertise any
+Modified Version, except to acknowledge the contribution(s) of the
+Copyright Holder(s) and the Author(s) or with their explicit written
+permission.
+
+5) The Font Software, modified or unmodified, in part or in whole,
+must be distributed entirely under this license, and must not be
+distributed under any other license. The requirement for fonts to
+remain under this license does not apply to any document created
+using the Font Software.
+
+TERMINATION
+This license becomes null and void if any of the above conditions are
+not met.
+
+DISCLAIMER
+THE FONT SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT
+OF COPYRIGHT, PATENT, TRADEMARK, OR OTHER RIGHT. IN NO EVENT SHALL THE
+COPYRIGHT HOLDER BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+INCLUDING ANY GENERAL, SPECIAL, INDIRECT, INCIDENTAL, OR CONSEQUENTIAL
+DAMAGES, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+FROM, OUT OF THE USE OR INABILITY TO USE THE FONT SOFTWARE OR FROM
+OTHER DEALINGS IN THE FONT SOFTWARE.
@@ -0,0 +1,552 @@
+// This code comes from:
+// https://github.com/dhbaird/easywsclient
+//
+// To get the latest version:
+// wget https://raw.github.com/dhbaird/easywsclient/master/easywsclient.hpp
+// wget https://raw.github.com/dhbaird/easywsclient/master/easywsclient.cpp
+
+#include "easywsclient.hpp"
+
+#ifdef _WIN32
+    #ifndef WIN32_LEAN_AND_MEAN
+        #define WIN32_LEAN_AND_MEAN
+    #endif
+    #include <winsock2.h>
+    #include <ws2tcpip.h>
+    #pragma comment( lib, "ws2_32" )
+    #include <stdio.h>
+    #include <stdlib.h>
+    #include <string.h>
+    #include <vector>
+    #include <string>
+    typedef SOCKET socket_t;
+    #ifndef _SSIZE_T_DEFINED
+        typedef int ssize_t;
+        #define _SSIZE_T_DEFINED
+    #endif
+    #ifndef _SOCKET_T_DEFINED
+        typedef SOCKET socket_t;
+        #define _SOCKET_T_DEFINED
+    #endif
+    #if defined(_MSC_VER) && !defined(snprintf)
+        #define snprintf _snprintf_s
+    #endif
+    #include <stdint.h>
+    #define socketerrno WSAGetLastError()
+    #define SOCKET_EAGAIN_EINPROGRESS WSAEINPROGRESS
+    #define SOCKET_EWOULDBLOCK WSAEWOULDBLOCK
+#else
+    #include <fcntl.h>
+    #include <netdb.h>
+    #include <netinet/tcp.h>
+    #include <stdint.h>
+    #include <stdio.h>
+    #include <stdlib.h>
+    #include <string.h>
+    #include <sys/socket.h>
+    #include <sys/time.h>
+    #include <sys/types.h>
+    #include <unistd.h>
+    #include <string>
+    #include <vector>
+    #include <errno.h>
+    typedef int socket_t;
+    #ifndef INVALID_SOCKET
+        #define INVALID_SOCKET (-1)
+    #endif
+    #ifndef SOCKET_ERROR
+        #define SOCKET_ERROR (-1)
+    #endif
+    #define closesocket(s) ::close(s)
+    #include <errno.h>
+    #define socketerrno errno
+    #define SOCKET_EAGAIN_EINPROGRESS EAGAIN
+    #define SOCKET_EWOULDBLOCK EWOULDBLOCK
+#endif
+
+#include <vector>
+#include <string>
+
+#include <stdarg.h>
+
+#include "util/logging.h"
+
+// When false, easywsclient's diagnostics are suppressed. The host application
+// opts in by setting this flag (see obs_websocket.cpp).
+bool EASYWSCLIENT_LOGGING_ENABLED = false;
+
+// Route easywsclient's diagnostic output through the project logger instead of
+// writing to stderr, without editing the upstream source lines below: these two
+// macros redirect fprintf(stderr, ...) / fputs(..., stderr) to log_misc.
+// Only emitted when EASYWSCLIENT_LOGGING_ENABLED is set.
+static inline void easywsclient_logf(const char *fmt, ...) {
+    if (!EASYWSCLIENT_LOGGING_ENABLED) {
+        return;
+    }
+    char buf[1024];
+    va_list args;
+    va_start(args, fmt);
+    vsnprintf(buf, sizeof(buf), fmt, args);
+    va_end(args);
+    // trim trailing newline(s); the logger appends its own
+    size_t len = strlen(buf);
+    while (len > 0 && (buf[len - 1] == '\n' || buf[len - 1] == '\r')) {
+        buf[--len] = '\0';
+    }
+    log_misc("easywsclient", "{}", buf);
+}
+#define fprintf(stream, ...) easywsclient_logf(__VA_ARGS__)
+#define fputs(str, stream) easywsclient_logf("%s", str)
+
+using namespace easywsclient;
+
+namespace { // private module-only namespace
+
+socket_t hostname_connect(const std::string& hostname, int port) {
+    struct addrinfo hints;
+    struct addrinfo *result;
+    struct addrinfo *p;
+    int ret;
+    socket_t sockfd = INVALID_SOCKET;
+    char sport[16];
+    memset(&hints, 0, sizeof(hints));
+    hints.ai_family = AF_UNSPEC;
+    hints.ai_socktype = SOCK_STREAM;
+    snprintf(sport, 16, "%d", port);
+    if ((ret = getaddrinfo(hostname.c_str(), sport, &hints, &result)) != 0)
+    {
+      fprintf(stderr, "getaddrinfo: %s\n", gai_strerror(ret));
+      return 1;
+    }
+    for(p = result; p != NULL; p = p->ai_next)
+    {
+        sockfd = socket(p->ai_family, p->ai_socktype, p->ai_protocol);
+        if (sockfd == INVALID_SOCKET) { continue; }
+        if (connect(sockfd, p->ai_addr, p->ai_addrlen) != SOCKET_ERROR) {
+            break;
+        }
+        closesocket(sockfd);
+        sockfd = INVALID_SOCKET;
+    }
+    freeaddrinfo(result);
+    return sockfd;
+}
+
+
+class _DummyWebSocket : public easywsclient::WebSocket
+{
+  public:
+    void poll(int timeout) { }
+    void send(const std::string& message) { }
+    void sendBinary(const std::string& message) { }
+    void sendBinary(const std::vector<uint8_t>& message) { }
+    void sendPing() { }
+    void close() { }
+    readyStateValues getReadyState() const { return CLOSED; }
+    void _dispatch(Callback_Imp & callable) { }
+    void _dispatchBinary(BytesCallback_Imp& callable) { }
+};
+
+
+class _RealWebSocket : public easywsclient::WebSocket
+{
+  public:
+    // http://tools.ietf.org/html/rfc6455#section-5.2  Base Framing Protocol
+    //
+    //  0                   1                   2                   3
+    //  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+    // +-+-+-+-+-------+-+-------------+-------------------------------+
+    // |F|R|R|R| opcode|M| Payload len |    Extended payload length    |
+    // |I|S|S|S|  (4)  |A|     (7)     |             (16/64)           |
+    // |N|V|V|V|       |S|             |   (if payload len==126/127)   |
+    // | |1|2|3|       |K|             |                               |
+    // +-+-+-+-+-------+-+-------------+ - - - - - - - - - - - - - - - +
+    // |     Extended payload length continued, if payload len == 127  |
+    // + - - - - - - - - - - - - - - - +-------------------------------+
+    // |                               |Masking-key, if MASK set to 1  |
+    // +-------------------------------+-------------------------------+
+    // | Masking-key (continued)       |          Payload Data         |
+    // +-------------------------------- - - - - - - - - - - - - - - - +
+    // :                     Payload Data continued ...                :
+    // + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - +
+    // |                     Payload Data continued ...                |
+    // +---------------------------------------------------------------+
+
+    struct wsheader_type {
+        unsigned header_size;
+        bool fin;
+        bool mask;
+        enum opcode_type {
+            CONTINUATION = 0x0,
+            TEXT_FRAME = 0x1,
+            BINARY_FRAME = 0x2,
+            CLOSE = 8,
+            PING = 9,
+            PONG = 0xa,
+        } opcode;
+        int N0;
+        uint64_t N;
+        uint8_t masking_key[4];
+    };
+
+    std::vector<uint8_t> rxbuf;
+    std::vector<uint8_t> txbuf;
+    std::vector<uint8_t> receivedData;
+
+    socket_t sockfd;
+    readyStateValues readyState;
+    bool useMask;
+    bool isRxBad;
+
+    _RealWebSocket(socket_t sockfd, bool useMask) : sockfd(sockfd), readyState(OPEN), useMask(useMask), isRxBad(false) {
+    }
+
+    readyStateValues getReadyState() const {
+      return readyState;
+    }
+
+    void poll(int timeout) { // timeout in milliseconds
+        if (readyState == CLOSED) {
+            if (timeout > 0) {
+                timeval tv = { timeout/1000, (timeout%1000) * 1000 };
+                select(0, NULL, NULL, NULL, &tv);
+            }
+            return;
+        }
+        if (timeout != 0) {
+            fd_set rfds;
+            fd_set wfds;
+            timeval tv = { timeout/1000, (timeout%1000) * 1000 };
+            FD_ZERO(&rfds);
+            FD_ZERO(&wfds);
+            FD_SET(sockfd, &rfds);
+            if (txbuf.size()) { FD_SET(sockfd, &wfds); }
+            select(sockfd + 1, &rfds, &wfds, NULL, timeout > 0 ? &tv : NULL);
+        }
+        while (true) {
+            // FD_ISSET(0, &rfds) will be true
+            int N = rxbuf.size();
+            ssize_t ret;
+            rxbuf.resize(N + 1500);
+            ret = recv(sockfd, (char*)&rxbuf[0] + N, 1500, 0);
+            if (false) { }
+            else if (ret < 0 && (socketerrno == SOCKET_EWOULDBLOCK || socketerrno == SOCKET_EAGAIN_EINPROGRESS)) {
+                rxbuf.resize(N);
+                break;
+            }
+            else if (ret <= 0) {
+                rxbuf.resize(N);
+                closesocket(sockfd);
+                readyState = CLOSED;
+                fputs(ret < 0 ? "Connection error!\n" : "Connection closed!\n", stderr);
+                break;
+            }
+            else {
+                rxbuf.resize(N + ret);
+            }
+        }
+        while (txbuf.size()) {
+            int ret = ::send(sockfd, (char*)&txbuf[0], txbuf.size(), 0);
+            if (false) { } // ??
+            else if (ret < 0 && (socketerrno == SOCKET_EWOULDBLOCK || socketerrno == SOCKET_EAGAIN_EINPROGRESS)) {
+                break;
+            }
+            else if (ret <= 0) {
+                closesocket(sockfd);
+                readyState = CLOSED;
+                fputs(ret < 0 ? "Connection error!\n" : "Connection closed!\n", stderr);
+                break;
+            }
+            else {
+                txbuf.erase(txbuf.begin(), txbuf.begin() + ret);
+            }
+        }
+        if (!txbuf.size() && readyState == CLOSING) {
+            closesocket(sockfd);
+            readyState = CLOSED;
+        }
+    }
+
+    virtual void _dispatch(Callback_Imp & callable) {
+        struct CallbackAdapter : public BytesCallback_Imp
+            // Adapt void(const std::string<uint8_t>&) to void(const std::string&)
+        {
+            Callback_Imp& callable;
+            CallbackAdapter(Callback_Imp& callable) : callable(callable) { }
+            void operator()(const std::vector<uint8_t>& message) {
+                std::string stringMessage(message.begin(), message.end());
+                callable(stringMessage);
+            }
+        };
+        CallbackAdapter bytesCallback(callable);
+        _dispatchBinary(bytesCallback);
+    }
+
+    virtual void _dispatchBinary(BytesCallback_Imp& callable) {
+        // TODO: consider acquiring a lock on rxbuf...
+        while (true) {
+            wsheader_type ws;
+            if (rxbuf.size() < 2) { return; /* Need at least 2 */ }
+            const uint8_t * data = (uint8_t *) &rxbuf[0]; // peek, but don't consume
+            ws.fin = (data[0] & 0x80) == 0x80;
+            ws.opcode = (wsheader_type::opcode_type) (data[0] & 0x0f);
+            ws.mask = (data[1] & 0x80) == 0x80;
+            ws.N0 = (data[1] & 0x7f);
+            ws.header_size = 2 + (ws.N0 == 126? 2 : 0) + (ws.N0 == 127? 8 : 0) + (ws.mask? 4 : 0);
+            if (rxbuf.size() < ws.header_size) { return; /* Need: ws.header_size - rxbuf.size() */ }
+            int i = 0;
+            if (ws.N0 < 126) {
+                ws.N = ws.N0;
+                i = 2;
+            }
+            else if (ws.N0 == 126) {
+                ws.N = 0;
+                ws.N |= ((uint64_t) data[2]) << 8;
+                ws.N |= ((uint64_t) data[3]) << 0;
+                i = 4;
+            }
+            else if (ws.N0 == 127) {
+                ws.N = 0;
+                ws.N |= ((uint64_t) data[2]) << 56;
+                ws.N |= ((uint64_t) data[3]) << 48;
+                ws.N |= ((uint64_t) data[4]) << 40;
+                ws.N |= ((uint64_t) data[5]) << 32;
+                ws.N |= ((uint64_t) data[6]) << 24;
+                ws.N |= ((uint64_t) data[7]) << 16;
+                ws.N |= ((uint64_t) data[8]) << 8;
+                ws.N |= ((uint64_t) data[9]) << 0;
+                i = 10;
+            }
+            if (ws.mask) {
+                ws.masking_key[0] = ((uint8_t) data[i+0]) << 0;
+                ws.masking_key[1] = ((uint8_t) data[i+1]) << 0;
+                ws.masking_key[2] = ((uint8_t) data[i+2]) << 0;
+                ws.masking_key[3] = ((uint8_t) data[i+3]) << 0;
+            }
+            else {
+                ws.masking_key[0] = 0;
+                ws.masking_key[1] = 0;
+                ws.masking_key[2] = 0;
+                ws.masking_key[3] = 0;
+            }
+
+            // Note: The checks above should hopefully ensure this addition
+            //       cannot overflow:
+            if (rxbuf.size() < ws.header_size+ws.N) { return; /* Need: ws.header_size+ws.N - rxbuf.size() */ }
+
+            // We got a whole message, now do something with it:
+            if (false) { }
+            else if (
+                   ws.opcode == wsheader_type::TEXT_FRAME
+                || ws.opcode == wsheader_type::BINARY_FRAME
+                || ws.opcode == wsheader_type::CONTINUATION
+            ) {
+                if (ws.mask) { for (size_t i = 0; i != ws.N; ++i) { rxbuf[i+ws.header_size] ^= ws.masking_key[i&0x3]; } }
+                receivedData.insert(receivedData.end(), rxbuf.begin()+ws.header_size, rxbuf.begin()+ws.header_size+(size_t)ws.N);// just feed
+                if (ws.fin) {
+                    callable((const std::vector<uint8_t>) receivedData);
+                    receivedData.erase(receivedData.begin(), receivedData.end());
+                    std::vector<uint8_t> ().swap(receivedData);// free memory
+                }
+            }
+            else if (ws.opcode == wsheader_type::PING) {
+                if (ws.mask) { for (size_t i = 0; i != ws.N; ++i) { rxbuf[i+ws.header_size] ^= ws.masking_key[i&0x3]; } }
+                std::string data(rxbuf.begin()+ws.header_size, rxbuf.begin()+ws.header_size+(size_t)ws.N);
+                sendData(wsheader_type::PONG, data.size(), data.begin(), data.end());
+            }
+            else if (ws.opcode == wsheader_type::PONG) { }
+            else if (ws.opcode == wsheader_type::CLOSE) { close(); }
+            else { fprintf(stderr, "ERROR: Got unexpected WebSocket message.\n"); close(); }
+
+            rxbuf.erase(rxbuf.begin(), rxbuf.begin() + ws.header_size+(size_t)ws.N);
+        }
+    }
+
+    void sendPing() {
+        std::string empty;
+        sendData(wsheader_type::PING, empty.size(), empty.begin(), empty.end());
+    }
+
+    void send(const std::string& message) {
+        sendData(wsheader_type::TEXT_FRAME, message.size(), message.begin(), message.end());
+    }
+
+    void sendBinary(const std::string& message) {
+        sendData(wsheader_type::BINARY_FRAME, message.size(), message.begin(), message.end());
+    }
+
+    void sendBinary(const std::vector<uint8_t>& message) {
+        sendData(wsheader_type::BINARY_FRAME, message.size(), message.begin(), message.end());
+    }
+
+    template<class Iterator>
+    void sendData(wsheader_type::opcode_type type, uint64_t message_size, Iterator message_begin, Iterator message_end) {
+        // TODO:
+        // Masking key should (must) be derived from a high quality random
+        // number generator, to mitigate attacks on non-WebSocket friendly
+        // middleware:
+        const uint8_t masking_key[4] = { 0x12, 0x34, 0x56, 0x78 };
+        // TODO: consider acquiring a lock on txbuf...
+        if (readyState == CLOSING || readyState == CLOSED) { return; }
+        std::vector<uint8_t> header;
+        header.assign(2 + (message_size >= 126 ? 2 : 0) + (message_size >= 65536 ? 6 : 0) + (useMask ? 4 : 0), 0);
+        header[0] = 0x80 | type;
+        if (false) { }
+        else if (message_size < 126) {
+            header[1] = (message_size & 0xff) | (useMask ? 0x80 : 0);
+            if (useMask) {
+                header[2] = masking_key[0];
+                header[3] = masking_key[1];
+                header[4] = masking_key[2];
+                header[5] = masking_key[3];
+            }
+        }
+        else if (message_size < 65536) {
+            header[1] = 126 | (useMask ? 0x80 : 0);
+            header[2] = (message_size >> 8) & 0xff;
+            header[3] = (message_size >> 0) & 0xff;
+            if (useMask) {
+                header[4] = masking_key[0];
+                header[5] = masking_key[1];
+                header[6] = masking_key[2];
+                header[7] = masking_key[3];
+            }
+        }
+        else { // TODO: run coverage testing here
+            header[1] = 127 | (useMask ? 0x80 : 0);
+            header[2] = (message_size >> 56) & 0xff;
+            header[3] = (message_size >> 48) & 0xff;
+            header[4] = (message_size >> 40) & 0xff;
+            header[5] = (message_size >> 32) & 0xff;
+            header[6] = (message_size >> 24) & 0xff;
+            header[7] = (message_size >> 16) & 0xff;
+            header[8] = (message_size >>  8) & 0xff;
+            header[9] = (message_size >>  0) & 0xff;
+            if (useMask) {
+                header[10] = masking_key[0];
+                header[11] = masking_key[1];
+                header[12] = masking_key[2];
+                header[13] = masking_key[3];
+            }
+        }
+        // N.B. - txbuf will keep growing until it can be transmitted over the socket:
+        txbuf.insert(txbuf.end(), header.begin(), header.end());
+        txbuf.insert(txbuf.end(), message_begin, message_end);
+        if (useMask) {
+            size_t message_offset = txbuf.size() - message_size;
+            for (size_t i = 0; i != message_size; ++i) {
+                txbuf[message_offset + i] ^= masking_key[i&0x3];
+            }
+        }
+    }
+
+    void close() {
+        if(readyState == CLOSING || readyState == CLOSED) { return; }
+        readyState = CLOSING;
+        uint8_t closeFrame[6] = {0x88, 0x80, 0x00, 0x00, 0x00, 0x00}; // last 4 bytes are a masking key
+        std::vector<uint8_t> header(closeFrame, closeFrame+6);
+        txbuf.insert(txbuf.end(), header.begin(), header.end());
+    }
+
+};
+
+
+easywsclient::WebSocket::pointer from_url(const std::string& url, bool useMask, const std::string& origin) {
+    char host[512];
+    int port;
+    char path[512];
+    if (url.size() >= 512) {
+      fprintf(stderr, "ERROR: url size limit exceeded: %s\n", url.c_str());
+      return NULL;
+    }
+    if (origin.size() >= 200) {
+      fprintf(stderr, "ERROR: origin size limit exceeded: %s\n", origin.c_str());
+      return NULL;
+    }
+    if (false) { }
+    else if (sscanf(url.c_str(), "ws://%[^:/]:%d/%s", host, &port, path) == 3) {
+    }
+    else if (sscanf(url.c_str(), "ws://%[^:/]/%s", host, path) == 2) {
+        port = 80;
+    }
+    else if (sscanf(url.c_str(), "ws://%[^:/]:%d", host, &port) == 2) {
+        path[0] = '\0';
+    }
+    else if (sscanf(url.c_str(), "ws://%[^:/]", host) == 1) {
+        port = 80;
+        path[0] = '\0';
+    }
+    else {
+        fprintf(stderr, "ERROR: Could not parse WebSocket url: %s\n", url.c_str());
+        return NULL;
+    }
+    //fprintf(stderr, "easywsclient: connecting: host=%s port=%d path=/%s\n", host, port, path);
+    socket_t sockfd = hostname_connect(host, port);
+    if (sockfd == INVALID_SOCKET) {
+        fprintf(stderr, "Unable to connect to %s:%d\n", host, port);
+        return NULL;
+    }
+    {
+        // XXX: this should be done non-blocking,
+        char line[1024];
+        int status;
+        int i;
+        snprintf(line, 1024, "GET /%s HTTP/1.1\r\n", path); ::send(sockfd, line, strlen(line), 0);
+        if (port == 80) {
+            snprintf(line, 1024, "Host: %s\r\n", host); ::send(sockfd, line, strlen(line), 0);
+        }
+        else {
+            snprintf(line, 1024, "Host: %s:%d\r\n", host, port); ::send(sockfd, line, strlen(line), 0);
+        }
+        snprintf(line, 1024, "Upgrade: websocket\r\n"); ::send(sockfd, line, strlen(line), 0);
+        snprintf(line, 1024, "Connection: Upgrade\r\n"); ::send(sockfd, line, strlen(line), 0);
+        if (!origin.empty()) {
+            snprintf(line, 1024, "Origin: %s\r\n", origin.c_str()); ::send(sockfd, line, strlen(line), 0);
+        }
+        snprintf(line, 1024, "Sec-WebSocket-Key: x3JJHMbDL1EzLkh9GBhXDw==\r\n"); ::send(sockfd, line, strlen(line), 0);
+        snprintf(line, 1024, "Sec-WebSocket-Version: 13\r\n"); ::send(sockfd, line, strlen(line), 0);
+        snprintf(line, 1024, "\r\n"); ::send(sockfd, line, strlen(line), 0);
+        for (i = 0; i < 2 || (i < 1023 && line[i-2] != '\r' && line[i-1] != '\n'); ++i) { if (recv(sockfd, line+i, 1, 0) == 0) { return NULL; } }
+        line[i] = 0;
+        if (i == 1023) { fprintf(stderr, "ERROR: Got invalid status line connecting to: %s\n", url.c_str()); return NULL; }
+        if (sscanf(line, "HTTP/1.1 %d", &status) != 1 || status != 101) { fprintf(stderr, "ERROR: Got bad status connecting to %s: %s", url.c_str(), line); return NULL; }
+        // TODO: verify response headers,
+        while (true) {
+            for (i = 0; i < 2 || (i < 1023 && line[i-2] != '\r' && line[i-1] != '\n'); ++i) { if (recv(sockfd, line+i, 1, 0) == 0) { return NULL; } }
+            if (line[0] == '\r' && line[1] == '\n') { break; }
+        }
+    }
+    int flag = 1;
+    setsockopt(sockfd, IPPROTO_TCP, TCP_NODELAY, (char*) &flag, sizeof(flag)); // Disable Nagle's algorithm
+#ifdef _WIN32
+    u_long on = 1;
+    ioctlsocket(sockfd, FIONBIO, &on);
+#else
+    fcntl(sockfd, F_SETFL, O_NONBLOCK);
+#endif
+    //fprintf(stderr, "Connected to: %s\n", url.c_str());
+    return easywsclient::WebSocket::pointer(new _RealWebSocket(sockfd, useMask));
+}
+
+} // end of module-only namespace
+
+
+
+namespace easywsclient {
+
+WebSocket::pointer WebSocket::create_dummy() {
+    static pointer dummy = pointer(new _DummyWebSocket);
+    return dummy;
+}
+
+
+WebSocket::pointer WebSocket::from_url(const std::string& url, const std::string& origin) {
+    return ::from_url(url, true, origin);
+}
+
+WebSocket::pointer WebSocket::from_url_no_mask(const std::string& url, const std::string& origin) {
+    return ::from_url(url, false, origin);
+}
+
+
+} // namespace easywsclient
@@ -0,0 +1,73 @@
+#ifndef EASYWSCLIENT_HPP_20120819_MIOFVASDTNUASZDQPLFD
+#define EASYWSCLIENT_HPP_20120819_MIOFVASDTNUASZDQPLFD
+
+// This code comes from:
+// https://github.com/dhbaird/easywsclient
+//
+// To get the latest version:
+// wget https://raw.github.com/dhbaird/easywsclient/master/easywsclient.hpp
+// wget https://raw.github.com/dhbaird/easywsclient/master/easywsclient.cpp
+
+#include <string>
+#include <vector>
+#include <cstdint>
+
+namespace easywsclient {
+
+struct Callback_Imp { virtual void operator()(const std::string& message) = 0; };
+struct BytesCallback_Imp { virtual void operator()(const std::vector<uint8_t>& message) = 0; };
+
+class WebSocket {
+  public:
+    typedef WebSocket * pointer;
+    typedef enum readyStateValues { CLOSING, CLOSED, CONNECTING, OPEN } readyStateValues;
+
+    // Factories:
+    static pointer create_dummy();
+    static pointer from_url(const std::string& url, const std::string& origin = std::string());
+    static pointer from_url_no_mask(const std::string& url, const std::string& origin = std::string());
+
+    // Interfaces:
+    virtual ~WebSocket() { }
+    virtual void poll(int timeout = 0) = 0; // timeout in milliseconds
+    virtual void send(const std::string& message) = 0;
+    virtual void sendBinary(const std::string& message) = 0;
+    virtual void sendBinary(const std::vector<uint8_t>& message) = 0;
+    virtual void sendPing() = 0;
+    virtual void close() = 0;
+    virtual readyStateValues getReadyState() const = 0;
+
+    template<class Callable>
+    void dispatch(Callable callable)
+        // For callbacks that accept a string argument.
+    { // N.B. this is compatible with both C++11 lambdas, functors and C function pointers
+        struct _Callback : public Callback_Imp {
+            Callable& callable;
+            _Callback(Callable& callable) : callable(callable) { }
+            void operator()(const std::string& message) { callable(message); }
+        };
+        _Callback callback(callable);
+        _dispatch(callback);
+    }
+
+    template<class Callable>
+    void dispatchBinary(Callable callable)
+        // For callbacks that accept a std::vector<uint8_t> argument.
+    { // N.B. this is compatible with both C++11 lambdas, functors and C function pointers
+        struct _Callback : public BytesCallback_Imp {
+            Callable& callable;
+            _Callback(Callable& callable) : callable(callable) { }
+            void operator()(const std::vector<uint8_t>& message) { callable(message); }
+        };
+        _Callback callback(callable);
+        _dispatchBinary(callback);
+    }
+
+  protected:
+    virtual void _dispatch(Callback_Imp& callable) = 0;
+    virtual void _dispatchBinary(BytesCallback_Imp& callable) = 0;
+};
+
+} // namespace easywsclient
+
+#endif /* EASYWSCLIENT_HPP_20120819_MIOFVASDTNUASZDQPLFD */
@@ -0,0 +1,122 @@
+// fpng.h - unlicense (see end of fpng.cpp)
+#pragma once
+
+#include <stdlib.h>
+#include <stdint.h>
+#include <vector>
+
+#ifndef FPNG_TRAIN_HUFFMAN_TABLES
+	// Set to 1 when using the -t (training) option in fpng_test to generate new opaque/alpha Huffman tables for the single pass encoder.
+	#define FPNG_TRAIN_HUFFMAN_TABLES (0)
+#endif
+
+namespace fpng
+{
+	// ---- Library initialization - call once to identify if the processor supports SSE.
+	// Otherwise you'll only get scalar fallbacks.
+	void fpng_init();
+
+	// ---- Useful Utilities
+
+	// Returns true if the CPU supports SSE 4.1, and SSE support wasn't disabled by setting FPNG_NO_SSE=1.
+	// fpng_init() must have been called first, or it'll assert and return false.
+	bool fpng_cpu_supports_sse41();
+
+	// Fast CRC-32 SSE4.1+pclmul or a scalar fallback (slice by 4)
+	const uint32_t FPNG_CRC32_INIT = 0;
+	uint32_t fpng_crc32(const void* pData, size_t size, uint32_t prev_crc32 = FPNG_CRC32_INIT);
+
+	// Fast Adler32 SSE4.1 Adler-32 with a scalar fallback.
+	const uint32_t FPNG_ADLER32_INIT = 1;
+	uint32_t fpng_adler32(const void* pData, size_t size, uint32_t adler = FPNG_ADLER32_INIT);
+
+	// ---- Compression
+	enum
+	{
+		// Enables computing custom Huffman tables for each file, instead of using the custom global tables. 
+		// Results in roughly 6% smaller files on average, but compression is around 40% slower.
+		FPNG_ENCODE_SLOWER = 1, 
+		
+		// Only use raw Deflate blocks (no compression at all). Intended for testing.
+		FPNG_FORCE_UNCOMPRESSED = 2,
+	};
+
+	// Fast PNG encoding. The resulting file can be decoded either using a standard PNG decoder or by the fpng_decode_memory() function below.
+	// pImage: pointer to RGB or RGBA image pixels, R first in memory, B/A last.
+	// w/h - image dimensions. Image's row pitch in bytes must is w*num_chans.
+	// num_chans must be 3 or 4. 
+	bool fpng_encode_image_to_memory(const void* pImage, uint32_t w, uint32_t h, uint32_t num_chans, std::vector<uint8_t>& out_buf, uint32_t flags = 0);
+
+#ifndef FPNG_NO_STDIO
+	// Fast PNG encoding to the specified file.
+	bool fpng_encode_image_to_file(const char* pFilename, const void* pImage, uint32_t w, uint32_t h, uint32_t num_chans, uint32_t flags = 0);
+#endif
+
+	// ---- Decompression
+		
+	enum
+	{
+		FPNG_DECODE_SUCCESS = 0,				// file is a valid PNG file and written by FPNG and the decode succeeded
+		
+		FPNG_DECODE_NOT_FPNG,					// file is a valid PNG file, but it wasn't written by FPNG so you should try decoding it with a general purpose PNG decoder
+
+		FPNG_DECODE_INVALID_ARG,				// invalid function parameter
+
+		FPNG_DECODE_FAILED_NOT_PNG,				// file cannot be a PNG file
+		FPNG_DECODE_FAILED_HEADER_CRC32,		// a chunk CRC32 check failed, file is likely corrupted or not PNG
+		FPNG_DECODE_FAILED_INVALID_DIMENSIONS,  // invalid image dimensions in IHDR chunk (0 or too large)
+		FPNG_DECODE_FAILED_DIMENSIONS_TOO_LARGE, // decoding the file fully into memory would likely require too much memory (only on 32bpp builds)
+		FPNG_DECODE_FAILED_CHUNK_PARSING,		// failed while parsing the chunk headers, or file is corrupted
+		FPNG_DECODE_FAILED_INVALID_IDAT,		// IDAT data length is too small and cannot be valid, file is either corrupted or it's a bug
+
+		// fpng_decode_file() specific errors
+		FPNG_DECODE_FILE_OPEN_FAILED,
+		FPNG_DECODE_FILE_TOO_LARGE,
+		FPNG_DECODE_FILE_READ_FAILED,
+		FPNG_DECODE_FILE_SEEK_FAILED
+	};
+
+	// Fast PNG decoding of files ONLY created by fpng_encode_image_to_memory() or fpng_encode_image_to_file().
+	// If fpng_get_info() or fpng_decode_memory() returns FPNG_DECODE_NOT_FPNG, you should decode the PNG by falling back to a general purpose decoder.
+	//
+	// fpng_get_info() parses the PNG header and iterates through all chunks to determine if it's a file written by FPNG, but does not decompress the actual image data so it's relatively fast.
+	// 
+	// pImage, image_size: Pointer to PNG image data and its size
+	// width, height: output image's dimensions
+	// channels_in_file: will be 3 or 4
+	// 
+	// Returns FPNG_DECODE_SUCCESS on success, otherwise one of the failure codes above.
+	// If FPNG_DECODE_NOT_FPNG is returned, you must decompress the file with a general purpose PNG decoder.
+	// If another error occurs, the file is likely corrupted or invalid, but you can still try to decompress the file with another decoder (which will likely fail).
+	int fpng_get_info(const void* pImage, uint32_t image_size, uint32_t& width, uint32_t& height, uint32_t& channels_in_file);
+
+	// fpng_decode_memory() decompresses 24/32bpp PNG files ONLY encoded by this module.
+	// If the image was written by FPNG, it will decompress the image data, otherwise it will return FPNG_DECODE_NOT_FPNG in which case you should fall back to a general purpose PNG decoder (lodepng, stb_image, libpng, etc.)
+	//
+	// pImage, image_size: Pointer to PNG image data and its size
+	// out: Output 24/32bpp image buffer
+	// width, height: output image's dimensions
+	// channels_in_file: will be 3 or 4
+	// desired_channels: must be 3 or 4 
+	// 
+	// If the image is 24bpp and 32bpp is requested, the alpha values will be set to 0xFF. 
+	// If the image is 32bpp and 24bpp is requested, the alpha values will be discarded.
+	// 
+	// Returns FPNG_DECODE_SUCCESS on success, otherwise one of the failure codes above.
+	// If FPNG_DECODE_NOT_FPNG is returned, you must decompress the file with a general purpose PNG decoder.
+	// If another error occurs, the file is likely corrupted or invalid, but you can still try to decompress the file with another decoder (which will likely fail).
+	int fpng_decode_memory(const void* pImage, uint32_t image_size, std::vector<uint8_t>& out, uint32_t& width, uint32_t& height, uint32_t& channels_in_file, uint32_t desired_channels);
+
+#ifndef FPNG_NO_STDIO
+	int fpng_decode_file(const char* pFilename, std::vector<uint8_t>& out, uint32_t& width, uint32_t& height, uint32_t& channels_in_file, uint32_t desired_channels);
+#endif
+
+	// ---- Internal API used for Huffman table training purposes
+
+#if FPNG_TRAIN_HUFFMAN_TABLES
+	const uint32_t HUFF_COUNTS_SIZE = 288;
+	extern uint64_t g_huff_counts[HUFF_COUNTS_SIZE];
+	bool create_dynamic_block_prefix(uint64_t* pFreq, uint32_t num_chans, std::vector<uint8_t>& prefix, uint64_t& bit_buf, int& bit_buf_size, uint32_t *pCodes, uint8_t *pCodesizes);
+#endif
+
+} // namespace fpng
@@ -20,6 +20,7 @@ Usage:
 #include <memory>
 #include <string>
 #include <map>
+#include <cstdint>

 ///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////

@@ -21,5 +21,11 @@ set(IMGUI_SOURCES
        misc/cpp/imgui_stdlib.cpp
 )

+# spice2x: DX11 backend is only built for non-XP toolchains
+if(NOT SPICE_XP)
+    list(APPEND IMGUI_HEADERS backends/imgui_impl_dx11.h)
+    list(APPEND IMGUI_SOURCES backends/imgui_impl_dx11.cpp)
+endif()
+
 add_library(imgui STATIC ${IMGUI_HEADERS} ${IMGUI_SOURCES})
 target_include_directories(imgui PRIVATE ${PROJECT_SOURCE_DIR})
@@ -1,6 +1,6 @@
 The MIT License (MIT)

-Copyright (c) 2014-2023 Omar Cornut
+Copyright (c) 2014-2026 Omar Cornut

 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal
@@ -0,0 +1,894 @@
+// dear imgui: Renderer Backend for DirectX11
+// This needs to be used along with a Platform Backend (e.g. Win32)
+
+// Implemented features:
+//  [X] Renderer: User texture binding. Use 'ID3D11ShaderResourceView*' as texture identifier. Read the FAQ about ImTextureID/ImTextureRef!
+//  [X] Renderer: Large meshes support (64k+ vertices) even with 16-bit indices (ImGuiBackendFlags_RendererHasVtxOffset).
+//  [X] Renderer: Texture updates support for dynamic font atlas (ImGuiBackendFlags_RendererHasTextures).
+//  [X] Renderer: Expose selected render state for draw callbacks to use. Access in '(ImGui_ImplXXXX_RenderState*)GetPlatformIO().Renderer_RenderState'.
+//  [X] Renderer: Multi-viewport support (multiple windows). Enable with 'io.ConfigFlags |= ImGuiConfigFlags_ViewportsEnable'.
+
+// You can use unmodified imgui_impl_* files in your project. See examples/ folder for examples of using this.
+// Prefer including the entire imgui/ repository into your project (either as a copy or as a submodule), and only build the backends you need.
+// Learn about Dear ImGui:
+// - FAQ                  https://dearimgui.com/faq
+// - Getting Started      https://dearimgui.com/getting-started
+// - Documentation        https://dearimgui.com/docs (same as your local docs/ folder).
+// - Introduction, links and more at the top of imgui.cpp
+
+// CHANGELOG
+// (minor and older changes stripped away, please see git history for details)
+//  2026-XX-XX: Platform: Added support for multiple windows via the ImGuiPlatformIO interface.
+//  2026-04-23: DirectX11: Added support for standard draw callbacks (in platform_io): DrawCallback_ResetRenderState, DrawCallback_SetSamplerLinear, DrawCallback_SetSamplerNearest. Obsoleting samplers from ImGui_ImplDX11_RenderState. (#9378)
+//  2026-01-19: DirectX11: Added 'SamplerNearest' in ImGui_ImplDX11_RenderState. Renamed 'SamplerDefault' to 'SamplerLinear'.
+//  2025-09-18: Call platform_io.ClearRendererHandlers() on shutdown.
+//  2025-06-11: DirectX11: Added support for ImGuiBackendFlags_RendererHasTextures, for dynamic font atlas.
+//  2025-05-07: DirectX11: Honor draw_data->FramebufferScale to allow for custom backends and experiment using it (consistently with other renderer backends, even though in normal condition it is not set under Windows).
+//  2025-02-24: [Docking] Added undocumented ImGui_ImplDX11_SetSwapChainDescs() to configure swap chain creation for secondary viewports.
+//  2025-01-06: DirectX11: Expose VertexConstantBuffer in ImGui_ImplDX11_RenderState. Reset projection matrix in ImDrawCallback_ResetRenderState handler.
+//  2024-10-07: DirectX11: Changed default texture sampler to Clamp instead of Repeat/Wrap.
+//  2024-10-07: DirectX11: Expose selected render state in ImGui_ImplDX11_RenderState, which you can access in 'void* platform_io.Renderer_RenderState' during draw callbacks.
+//  2022-10-11: Using 'nullptr' instead of 'NULL' as per our switch to C++11.
+//  2021-06-29: Reorganized backend to pull data from a single structure to facilitate usage with multiple-contexts (all g_XXXX access changed to bd->XXXX).
+//  2021-05-19: DirectX11: Replaced direct access to ImDrawCmd::TextureId with a call to ImDrawCmd::GetTexID(). (will become a requirement)
+//  2021-02-18: DirectX11: Change blending equation to preserve alpha in output buffer.
+//  2019-08-01: DirectX11: Fixed code querying the Geometry Shader state (would generally error with Debug layer enabled).
+//  2019-07-21: DirectX11: Backup, clear and restore Geometry Shader is any is bound when calling ImGui_ImplDX11_RenderDrawData. Clearing Hull/Domain/Compute shaders without backup/restore.
+//  2019-05-29: DirectX11: Added support for large mesh (64K+ vertices), enable ImGuiBackendFlags_RendererHasVtxOffset flag.
+//  2019-04-30: DirectX11: Added support for special ImDrawCallback_ResetRenderState callback to reset render state.
+//  2018-12-03: Misc: Added #pragma comment statement to automatically link with d3dcompiler.lib when using D3DCompile().
+//  2018-11-30: Misc: Setting up io.BackendRendererName so it can be displayed in the About Window.
+//  2018-08-01: DirectX11: Querying for IDXGIFactory instead of IDXGIFactory1 to increase compatibility.
+//  2018-07-13: DirectX11: Fixed unreleased resources in Init and Shutdown functions.
+//  2018-06-08: Misc: Extracted imgui_impl_dx11.cpp/.h away from the old combined DX11+Win32 example.
+//  2018-06-08: DirectX11: Use draw_data->DisplayPos and draw_data->DisplaySize to setup projection matrix and clipping rectangle.
+//  2018-02-16: Misc: Obsoleted the io.RenderDrawListsFn callback and exposed ImGui_ImplDX11_RenderDrawData() in the .h file so you can call it yourself.
+//  2018-02-06: Misc: Removed call to ImGui::Shutdown() which is not available from 1.60 WIP, user needs to call CreateContext/DestroyContext themselves.
+//  2016-05-07: DirectX11: Disabling depth-write.
+
+#include "imgui.h"
+#ifndef IMGUI_DISABLE
+#include "imgui_impl_dx11.h"
+
+// DirectX
+#include <stdio.h>
+#include <d3d11.h>
+#include <d3dcompiler.h>
+// #ifdef _MSC_VER
+// #pragma comment(lib, "d3dcompiler") // Automatically link with d3dcompiler.lib as we are using D3DCompile() below.
+// #endif
+
+// spice2x: resolve D3DCompile at runtime rather than static-linking d3dcompiler,
+// which would hardwire an import on d3dcompiler_47.dll (absent on stock Win7).
+static pD3DCompile ImGui_ImplDX11_GetD3DCompile()
+{
+    static pD3DCompile fn = []() -> pD3DCompile {
+        static const char *dlls[] = {
+            "d3dcompiler_47.dll",
+            "d3dcompiler_46.dll",
+            "d3dcompiler_43.dll" };
+
+        for (const char *dll : dlls) {
+            HMODULE mod = GetModuleHandleA(dll);
+            if (!mod) {
+                mod = LoadLibraryA(dll);
+            }
+            if (mod) {
+                if (auto p = (pD3DCompile) GetProcAddress(mod, "D3DCompile")) {
+                    return p;
+                }
+            }
+        }
+        return nullptr;
+    }();
+    return fn;
+}
+
+// Clang/GCC warnings with -Weverything
+#if defined(__clang__)
+#pragma clang diagnostic ignored "-Wold-style-cast"         // warning: use of old-style cast                            // yes, they are more terse.
+#pragma clang diagnostic ignored "-Wsign-conversion"        // warning: implicit conversion changes signedness
+#endif
+
+// DirectX11 data
+struct ImGui_ImplDX11_Texture
+{
+    ID3D11Texture2D*            pTexture;
+    ID3D11ShaderResourceView*   pTextureView;
+};
+
+struct ImGui_ImplDX11_Data
+{
+    ID3D11Device*               pd3dDevice;
+    ID3D11DeviceContext*        pd3dDeviceContext;
+    IDXGIFactory*               pFactory;
+    ID3D11Buffer*               pVB;
+    ID3D11Buffer*               pIB;
+    ID3D11VertexShader*         pVertexShader;
+    ID3D11InputLayout*          pInputLayout;
+    ID3D11Buffer*               pVertexConstantBuffer;
+    ID3D11PixelShader*          pPixelShader;
+    ID3D11SamplerState*         pTexSamplerLinear;
+    ID3D11SamplerState*         pTexSamplerNearest;
+    ID3D11RasterizerState*      pRasterizerState;
+    ID3D11BlendState*           pBlendState;
+    ID3D11DepthStencilState*    pDepthStencilState;
+    int                         VertexBufferSize;
+    int                         IndexBufferSize;
+    ImGui_ImplDX11_RenderState* RenderState;            // == (ImGui_ImplDX11_RenderState*)ImGui::GetPlatformIO().Renderer_RenderState during rendering.
+    ImVector<DXGI_SWAP_CHAIN_DESC> SwapChainDescsForViewports;
+
+    ImGui_ImplDX11_Data()       { memset((void*)this, 0, sizeof(*this)); VertexBufferSize = 5000; IndexBufferSize = 10000; }
+};
+
+struct VERTEX_CONSTANT_BUFFER_DX11
+{
+    float   mvp[4][4];
+};
+
+// Backend data stored in io.BackendRendererUserData to allow support for multiple Dear ImGui contexts
+// It is STRONGLY preferred that you use docking branch with multi-viewports (== single Dear ImGui context + multiple windows) instead of multiple Dear ImGui contexts.
+static ImGui_ImplDX11_Data* ImGui_ImplDX11_GetBackendData()
+{
+    return ImGui::GetCurrentContext() ? (ImGui_ImplDX11_Data*)ImGui::GetIO().BackendRendererUserData : nullptr;
+}
+
+// Forward Declarations
+static void ImGui_ImplDX11_InitMultiViewportSupport();
+static void ImGui_ImplDX11_ShutdownMultiViewportSupport();
+
+// Functions
+static void ImGui_ImplDX11_SetupRenderState(const ImDrawData* draw_data, ID3D11DeviceContext* device_ctx)
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+
+    // Setup viewport
+    D3D11_VIEWPORT vp = {};
+    vp.Width = draw_data->DisplaySize.x * draw_data->FramebufferScale.x;
+    vp.Height = draw_data->DisplaySize.y * draw_data->FramebufferScale.y;
+    vp.MinDepth = 0.0f;
+    vp.MaxDepth = 1.0f;
+    vp.TopLeftX = vp.TopLeftY = 0;
+    device_ctx->RSSetViewports(1, &vp);
+
+    // Setup orthographic projection matrix into our constant buffer
+    // Our visible imgui space lies from draw_data->DisplayPos (top left) to draw_data->DisplayPos+data_data->DisplaySize (bottom right). DisplayPos is (0,0) for single viewport apps.
+    D3D11_MAPPED_SUBRESOURCE mapped_resource;
+    if (device_ctx->Map(bd->pVertexConstantBuffer, 0, D3D11_MAP_WRITE_DISCARD, 0, &mapped_resource) == S_OK)
+    {
+        VERTEX_CONSTANT_BUFFER_DX11* constant_buffer = (VERTEX_CONSTANT_BUFFER_DX11*)mapped_resource.pData;
+        float L = draw_data->DisplayPos.x;
+        float R = draw_data->DisplayPos.x + draw_data->DisplaySize.x;
+        float T = draw_data->DisplayPos.y;
+        float B = draw_data->DisplayPos.y + draw_data->DisplaySize.y;
+        float mvp[4][4] =
+        {
+            { 2.0f/(R-L),   0.0f,           0.0f,       0.0f },
+            { 0.0f,         2.0f/(T-B),     0.0f,       0.0f },
+            { 0.0f,         0.0f,           0.5f,       0.0f },
+            { (R+L)/(L-R),  (T+B)/(B-T),    0.5f,       1.0f },
+        };
+        memcpy(&constant_buffer->mvp, mvp, sizeof(mvp));
+        device_ctx->Unmap(bd->pVertexConstantBuffer, 0);
+    }
+
+    // Setup shader and vertex buffers
+    unsigned int stride = sizeof(ImDrawVert);
+    unsigned int offset = 0;
+    device_ctx->IASetInputLayout(bd->pInputLayout);
+    device_ctx->IASetVertexBuffers(0, 1, &bd->pVB, &stride, &offset);
+    device_ctx->IASetIndexBuffer(bd->pIB, sizeof(ImDrawIdx) == 2 ? DXGI_FORMAT_R16_UINT : DXGI_FORMAT_R32_UINT, 0);
+    device_ctx->IASetPrimitiveTopology(D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST);
+    device_ctx->VSSetShader(bd->pVertexShader, nullptr, 0);
+    device_ctx->VSSetConstantBuffers(0, 1, &bd->pVertexConstantBuffer);
+    device_ctx->PSSetShader(bd->pPixelShader, nullptr, 0);
+    device_ctx->PSSetSamplers(0, 1, &bd->pTexSamplerLinear);
+    device_ctx->GSSetShader(nullptr, nullptr, 0);
+    device_ctx->HSSetShader(nullptr, nullptr, 0); // In theory we should backup and restore this as well.. very infrequently used..
+    device_ctx->DSSetShader(nullptr, nullptr, 0); // In theory we should backup and restore this as well.. very infrequently used..
+    device_ctx->CSSetShader(nullptr, nullptr, 0); // In theory we should backup and restore this as well.. very infrequently used..
+
+    // Setup render state
+    const float blend_factor[4] = { 0.f, 0.f, 0.f, 0.f };
+    device_ctx->OMSetBlendState(bd->pBlendState, blend_factor, 0xffffffff);
+    device_ctx->OMSetDepthStencilState(bd->pDepthStencilState, 0);
+    device_ctx->RSSetState(bd->pRasterizerState);
+}
+
+// Draw callbacks
+static void ImGui_ImplDX11_DrawCallback_ResetRenderState(const ImDrawList*, const ImDrawCmd*)       {} // Intentionally empty. Used as an identifier for rendering loop to call its code. Simpler to implement this way.
+static void ImGui_ImplDX11_DrawCallback_SetSamplerLinear(const ImDrawList*, const ImDrawCmd*)       { ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData(); bd->RenderState->DeviceContext->PSSetSamplers(0, 1, &bd->pTexSamplerLinear); }
+static void ImGui_ImplDX11_DrawCallback_SetSamplerNearest(const ImDrawList*, const ImDrawCmd*)      { ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData(); bd->RenderState->DeviceContext->PSSetSamplers(0, 1, &bd->pTexSamplerNearest); }
+
+// Render function
+void ImGui_ImplDX11_RenderDrawData(ImDrawData* draw_data)
+{
+    // Avoid rendering when minimized
+    if (draw_data->DisplaySize.x <= 0.0f || draw_data->DisplaySize.y <= 0.0f)
+        return;
+
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    ID3D11DeviceContext* device = bd->pd3dDeviceContext;
+
+    // Catch up with texture updates. Most of the times, the list will have 1 element with an OK status, aka nothing to do.
+    // (This almost always points to ImGui::GetPlatformIO().Textures[] but is part of ImDrawData to allow overriding or disabling texture updates).
+    if (draw_data->Textures != nullptr)
+        for (ImTextureData* tex : *draw_data->Textures)
+            if (tex->Status != ImTextureStatus_OK)
+                ImGui_ImplDX11_UpdateTexture(tex);
+
+    // Create and grow vertex/index buffers if needed
+    if (!bd->pVB || bd->VertexBufferSize < draw_data->TotalVtxCount)
+    {
+        if (bd->pVB) { bd->pVB->Release(); bd->pVB = nullptr; }
+        bd->VertexBufferSize = draw_data->TotalVtxCount + 5000;
+        D3D11_BUFFER_DESC desc = {};
+        desc.Usage = D3D11_USAGE_DYNAMIC;
+        desc.ByteWidth = (UINT)bd->VertexBufferSize * sizeof(ImDrawVert);
+        desc.BindFlags = D3D11_BIND_VERTEX_BUFFER;
+        desc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
+        desc.MiscFlags = 0;
+        if (bd->pd3dDevice->CreateBuffer(&desc, nullptr, &bd->pVB) < 0)
+            return;
+    }
+    if (!bd->pIB || bd->IndexBufferSize < draw_data->TotalIdxCount)
+    {
+        if (bd->pIB) { bd->pIB->Release(); bd->pIB = nullptr; }
+        bd->IndexBufferSize = draw_data->TotalIdxCount + 10000;
+        D3D11_BUFFER_DESC desc = {};
+        desc.Usage = D3D11_USAGE_DYNAMIC;
+        desc.ByteWidth = (UINT)bd->IndexBufferSize * sizeof(ImDrawIdx);
+        desc.BindFlags = D3D11_BIND_INDEX_BUFFER;
+        desc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
+        if (bd->pd3dDevice->CreateBuffer(&desc, nullptr, &bd->pIB) < 0)
+            return;
+    }
+
+    // Upload vertex/index data into a single contiguous GPU buffer
+    D3D11_MAPPED_SUBRESOURCE vtx_resource, idx_resource;
+    if (device->Map(bd->pVB, 0, D3D11_MAP_WRITE_DISCARD, 0, &vtx_resource) != S_OK)
+        return;
+    if (device->Map(bd->pIB, 0, D3D11_MAP_WRITE_DISCARD, 0, &idx_resource) != S_OK)
+        return;
+    ImDrawVert* vtx_dst = (ImDrawVert*)vtx_resource.pData;
+    ImDrawIdx* idx_dst = (ImDrawIdx*)idx_resource.pData;
+    for (const ImDrawList* draw_list : draw_data->CmdLists)
+    {
+        memcpy(vtx_dst, draw_list->VtxBuffer.Data, draw_list->VtxBuffer.Size * sizeof(ImDrawVert));
+        memcpy(idx_dst, draw_list->IdxBuffer.Data, draw_list->IdxBuffer.Size * sizeof(ImDrawIdx));
+        vtx_dst += draw_list->VtxBuffer.Size;
+        idx_dst += draw_list->IdxBuffer.Size;
+    }
+    device->Unmap(bd->pVB, 0);
+    device->Unmap(bd->pIB, 0);
+
+    // Backup DX state that will be modified to restore it afterwards (unfortunately this is very ugly looking and verbose. Close your eyes!)
+    struct BACKUP_DX11_STATE
+    {
+        UINT                        ScissorRectsCount, ViewportsCount;
+        D3D11_RECT                  ScissorRects[D3D11_VIEWPORT_AND_SCISSORRECT_OBJECT_COUNT_PER_PIPELINE];
+        D3D11_VIEWPORT              Viewports[D3D11_VIEWPORT_AND_SCISSORRECT_OBJECT_COUNT_PER_PIPELINE];
+        ID3D11RasterizerState*      RS;
+        ID3D11BlendState*           BlendState;
+        FLOAT                       BlendFactor[4];
+        UINT                        SampleMask;
+        UINT                        StencilRef;
+        ID3D11DepthStencilState*    DepthStencilState;
+        ID3D11ShaderResourceView*   PSShaderResource;
+        ID3D11SamplerState*         PSSampler;
+        ID3D11PixelShader*          PS;
+        ID3D11VertexShader*         VS;
+        ID3D11GeometryShader*       GS;
+        UINT                        PSInstancesCount, VSInstancesCount, GSInstancesCount;
+        ID3D11ClassInstance         *PSInstances[256], *VSInstances[256], *GSInstances[256];   // 256 is max according to PSSetShader documentation
+        D3D11_PRIMITIVE_TOPOLOGY    PrimitiveTopology;
+        ID3D11Buffer*               IndexBuffer, *VertexBuffer, *VSConstantBuffer;
+        UINT                        IndexBufferOffset, VertexBufferStride, VertexBufferOffset;
+        DXGI_FORMAT                 IndexBufferFormat;
+        ID3D11InputLayout*          InputLayout;
+    };
+    BACKUP_DX11_STATE old = {};
+    old.ScissorRectsCount = old.ViewportsCount = D3D11_VIEWPORT_AND_SCISSORRECT_OBJECT_COUNT_PER_PIPELINE;
+    device->RSGetScissorRects(&old.ScissorRectsCount, old.ScissorRects);
+    device->RSGetViewports(&old.ViewportsCount, old.Viewports);
+    device->RSGetState(&old.RS);
+    device->OMGetBlendState(&old.BlendState, old.BlendFactor, &old.SampleMask);
+    device->OMGetDepthStencilState(&old.DepthStencilState, &old.StencilRef);
+    device->PSGetShaderResources(0, 1, &old.PSShaderResource);
+    device->PSGetSamplers(0, 1, &old.PSSampler);
+    old.PSInstancesCount = old.VSInstancesCount = old.GSInstancesCount = 256;
+    device->PSGetShader(&old.PS, old.PSInstances, &old.PSInstancesCount);
+    device->VSGetShader(&old.VS, old.VSInstances, &old.VSInstancesCount);
+    device->VSGetConstantBuffers(0, 1, &old.VSConstantBuffer);
+    device->GSGetShader(&old.GS, old.GSInstances, &old.GSInstancesCount);
+
+    device->IAGetPrimitiveTopology(&old.PrimitiveTopology);
+    device->IAGetIndexBuffer(&old.IndexBuffer, &old.IndexBufferFormat, &old.IndexBufferOffset);
+    device->IAGetVertexBuffers(0, 1, &old.VertexBuffer, &old.VertexBufferStride, &old.VertexBufferOffset);
+    device->IAGetInputLayout(&old.InputLayout);
+
+    // Setup desired DX state
+    ImGui_ImplDX11_SetupRenderState(draw_data, device);
+
+    // Setup render state structure (for callbacks and custom texture bindings)
+    ImGuiPlatformIO& platform_io = ImGui::GetPlatformIO();
+    ImGui_ImplDX11_RenderState render_state;
+    render_state.Device = bd->pd3dDevice;
+    render_state.DeviceContext = bd->pd3dDeviceContext;
+    render_state.VertexConstantBuffer = bd->pVertexConstantBuffer;
+    platform_io.Renderer_RenderState = bd->RenderState = &render_state;
+
+    // Render command lists
+    // (Because we merged all buffers into a single one, we maintain our own offset into them)
+    int global_idx_offset = 0;
+    int global_vtx_offset = 0;
+    ImVec2 clip_off = draw_data->DisplayPos;
+    ImVec2 clip_scale = draw_data->FramebufferScale;
+    for (const ImDrawList* draw_list : draw_data->CmdLists)
+    {
+        for (int cmd_i = 0; cmd_i < draw_list->CmdBuffer.Size; cmd_i++)
+        {
+            const ImDrawCmd* pcmd = &draw_list->CmdBuffer[cmd_i];
+            if (pcmd->UserCallback != nullptr)
+            {
+                // User callback, registered via ImDrawList::AddCallback()
+                if (pcmd->UserCallback == ImGui_ImplDX11_DrawCallback_ResetRenderState)
+                    ImGui_ImplDX11_SetupRenderState(draw_data, device);
+                else
+                    pcmd->UserCallback(draw_list, pcmd);
+            }
+            else
+            {
+                // Project scissor/clipping rectangles into framebuffer space
+                ImVec2 clip_min((pcmd->ClipRect.x - clip_off.x) * clip_scale.x, (pcmd->ClipRect.y - clip_off.y) * clip_scale.y);
+                ImVec2 clip_max((pcmd->ClipRect.z - clip_off.x) * clip_scale.x, (pcmd->ClipRect.w - clip_off.y) * clip_scale.y);
+                if (clip_max.x <= clip_min.x || clip_max.y <= clip_min.y)
+                    continue;
+
+                // Apply scissor/clipping rectangle
+                const D3D11_RECT r = { (LONG)clip_min.x, (LONG)clip_min.y, (LONG)clip_max.x, (LONG)clip_max.y };
+                device->RSSetScissorRects(1, &r);
+
+                // Bind texture, Draw
+                ID3D11ShaderResourceView* texture_srv = (ID3D11ShaderResourceView*)pcmd->GetTexID();
+                device->PSSetShaderResources(0, 1, &texture_srv);
+                device->DrawIndexed(pcmd->ElemCount, pcmd->IdxOffset + global_idx_offset, pcmd->VtxOffset + global_vtx_offset);
+            }
+        }
+        global_idx_offset += draw_list->IdxBuffer.Size;
+        global_vtx_offset += draw_list->VtxBuffer.Size;
+    }
+    platform_io.Renderer_RenderState = bd->RenderState = nullptr;
+
+    // Restore modified DX state
+    device->RSSetScissorRects(old.ScissorRectsCount, old.ScissorRects);
+    device->RSSetViewports(old.ViewportsCount, old.Viewports);
+    device->RSSetState(old.RS); if (old.RS) old.RS->Release();
+    device->OMSetBlendState(old.BlendState, old.BlendFactor, old.SampleMask); if (old.BlendState) old.BlendState->Release();
+    device->OMSetDepthStencilState(old.DepthStencilState, old.StencilRef); if (old.DepthStencilState) old.DepthStencilState->Release();
+    device->PSSetShaderResources(0, 1, &old.PSShaderResource); if (old.PSShaderResource) old.PSShaderResource->Release();
+    device->PSSetSamplers(0, 1, &old.PSSampler); if (old.PSSampler) old.PSSampler->Release();
+    device->PSSetShader(old.PS, old.PSInstances, old.PSInstancesCount); if (old.PS) old.PS->Release();
+    for (UINT i = 0; i < old.PSInstancesCount; i++) if (old.PSInstances[i]) old.PSInstances[i]->Release();
+    device->VSSetShader(old.VS, old.VSInstances, old.VSInstancesCount); if (old.VS) old.VS->Release();
+    device->VSSetConstantBuffers(0, 1, &old.VSConstantBuffer); if (old.VSConstantBuffer) old.VSConstantBuffer->Release();
+    device->GSSetShader(old.GS, old.GSInstances, old.GSInstancesCount); if (old.GS) old.GS->Release();
+    for (UINT i = 0; i < old.VSInstancesCount; i++) if (old.VSInstances[i]) old.VSInstances[i]->Release();
+    device->IASetPrimitiveTopology(old.PrimitiveTopology);
+    device->IASetIndexBuffer(old.IndexBuffer, old.IndexBufferFormat, old.IndexBufferOffset); if (old.IndexBuffer) old.IndexBuffer->Release();
+    device->IASetVertexBuffers(0, 1, &old.VertexBuffer, &old.VertexBufferStride, &old.VertexBufferOffset); if (old.VertexBuffer) old.VertexBuffer->Release();
+    device->IASetInputLayout(old.InputLayout); if (old.InputLayout) old.InputLayout->Release();
+}
+
+static void ImGui_ImplDX11_DestroyTexture(ImTextureData* tex)
+{
+    if (ImGui_ImplDX11_Texture* backend_tex = (ImGui_ImplDX11_Texture*)tex->BackendUserData)
+    {
+        IM_ASSERT(backend_tex->pTextureView == (ID3D11ShaderResourceView*)(intptr_t)tex->TexID);
+        backend_tex->pTextureView->Release();
+        backend_tex->pTexture->Release();
+        IM_DELETE(backend_tex);
+
+        // Clear identifiers and mark as destroyed (in order to allow e.g. calling InvalidateDeviceObjects while running)
+        tex->SetTexID(ImTextureID_Invalid);
+        tex->BackendUserData = nullptr;
+    }
+    tex->SetStatus(ImTextureStatus_Destroyed);
+}
+
+void ImGui_ImplDX11_UpdateTexture(ImTextureData* tex)
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    if (tex->Status == ImTextureStatus_WantCreate)
+    {
+        // Create and upload new texture to graphics system
+        //IMGUI_DEBUG_LOG("UpdateTexture #%03d: WantCreate %dx%d\n", tex->UniqueID, tex->Width, tex->Height);
+        IM_ASSERT(tex->TexID == ImTextureID_Invalid && tex->BackendUserData == nullptr);
+        IM_ASSERT(tex->Format == ImTextureFormat_RGBA32);
+        unsigned int* pixels = (unsigned int*)tex->GetPixels();
+        ImGui_ImplDX11_Texture* backend_tex = IM_NEW(ImGui_ImplDX11_Texture)();
+
+        // Create texture
+        D3D11_TEXTURE2D_DESC desc;
+        ZeroMemory(&desc, sizeof(desc));
+        desc.Width = (UINT)tex->Width;
+        desc.Height = (UINT)tex->Height;
+        desc.MipLevels = 1;
+        desc.ArraySize = 1;
+#ifdef IMGUI_USE_BGRA_PACKED_COLOR
+        desc.Format = DXGI_FORMAT_B8G8R8A8_UNORM;
+#else
+        desc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
+#endif
+        desc.SampleDesc.Count = 1;
+        desc.Usage = D3D11_USAGE_DEFAULT;
+        desc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
+        desc.CPUAccessFlags = 0;
+        D3D11_SUBRESOURCE_DATA subResource;
+        subResource.pSysMem = pixels;
+        subResource.SysMemPitch = desc.Width * 4;
+        subResource.SysMemSlicePitch = 0;
+        bd->pd3dDevice->CreateTexture2D(&desc, &subResource, &backend_tex->pTexture);
+        IM_ASSERT(backend_tex->pTexture != nullptr && "Backend failed to create texture!");
+
+        // Create texture view
+        D3D11_SHADER_RESOURCE_VIEW_DESC srvDesc;
+        ZeroMemory(&srvDesc, sizeof(srvDesc));
+#ifdef IMGUI_USE_BGRA_PACKED_COLOR
+        srvDesc.Format = DXGI_FORMAT_B8G8R8A8_UNORM;
+#else
+        srvDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
+#endif
+        srvDesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2D;
+        srvDesc.Texture2D.MipLevels = desc.MipLevels;
+        srvDesc.Texture2D.MostDetailedMip = 0;
+        bd->pd3dDevice->CreateShaderResourceView(backend_tex->pTexture, &srvDesc, &backend_tex->pTextureView);
+        IM_ASSERT(backend_tex->pTextureView != nullptr && "Backend failed to create texture!");
+
+        // Store identifiers
+        tex->SetTexID((ImTextureID)(intptr_t)backend_tex->pTextureView);
+        tex->SetStatus(ImTextureStatus_OK);
+        tex->BackendUserData = backend_tex;
+    }
+    else if (tex->Status == ImTextureStatus_WantUpdates)
+    {
+        // Update selected blocks. We only ever write to textures regions which have never been used before!
+        // This backend choose to use tex->Updates[] but you can use tex->UpdateRect to upload a single region.
+        ImGui_ImplDX11_Texture* backend_tex = (ImGui_ImplDX11_Texture*)tex->BackendUserData;
+        IM_ASSERT(backend_tex->pTextureView == (ID3D11ShaderResourceView*)(intptr_t)tex->TexID);
+        for (ImTextureRect& r : tex->Updates)
+        {
+            D3D11_BOX box = { (UINT)r.x, (UINT)r.y, (UINT)0, (UINT)(r.x + r.w), (UINT)(r.y + r .h), (UINT)1 };
+            bd->pd3dDeviceContext->UpdateSubresource(backend_tex->pTexture, 0, &box, tex->GetPixelsAt(r.x, r.y), (UINT)tex->GetPitch(), 0);
+        }
+        tex->SetStatus(ImTextureStatus_OK);
+    }
+    if (tex->Status == ImTextureStatus_WantDestroy && tex->UnusedFrames > 0)
+        ImGui_ImplDX11_DestroyTexture(tex);
+}
+
+bool    ImGui_ImplDX11_CreateDeviceObjects()
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    if (!bd->pd3dDevice)
+        return false;
+    ImGui_ImplDX11_InvalidateDeviceObjects();
+
+    // By using D3DCompile() from <d3dcompiler.h> / d3dcompiler.lib, we introduce a dependency to a given version of d3dcompiler_XX.dll (see D3DCOMPILER_DLL_A)
+    // If you would like to use this DX11 sample code but remove this dependency you can:
+    //  1) compile once, save the compiled shader blobs into a file or source code and pass them to CreateVertexShader()/CreatePixelShader() [preferred solution]
+    //  2) use code to detect any version of the DLL and grab a pointer to D3DCompile from the DLL.
+    // See https://github.com/ocornut/imgui/pull/638 for sources and details.
+
+    // spice2x: dynamically resolved; see ImGui_ImplDX11_GetD3DCompile.
+    pD3DCompile D3DCompile_fn = ImGui_ImplDX11_GetD3DCompile();
+    if (!D3DCompile_fn) {
+        return false;
+    }
+
+    // Create the vertex shader
+    {
+        static const char* vertexShader =
+            "cbuffer vertexBuffer : register(b0) \
+            {\
+              float4x4 ProjectionMatrix; \
+            };\
+            struct VS_INPUT\
+            {\
+              float2 pos : POSITION;\
+              float4 col : COLOR0;\
+              float2 uv  : TEXCOORD0;\
+            };\
+            \
+            struct PS_INPUT\
+            {\
+              float4 pos : SV_POSITION;\
+              float4 col : COLOR0;\
+              float2 uv  : TEXCOORD0;\
+            };\
+            \
+            PS_INPUT main(VS_INPUT input)\
+            {\
+              PS_INPUT output;\
+              output.pos = mul( ProjectionMatrix, float4(input.pos.xy, 0.f, 1.f));\
+              output.col = input.col;\
+              output.uv  = input.uv;\
+              return output;\
+            }";
+
+        ID3DBlob* vertexShaderBlob;
+        if (FAILED(D3DCompile_fn(vertexShader, strlen(vertexShader), nullptr, nullptr, nullptr, "main", "vs_4_0", 0, 0, &vertexShaderBlob, nullptr)))
+            return false; // NB: Pass ID3DBlob* pErrorBlob to D3DCompile() to get error showing in (const char*)pErrorBlob->GetBufferPointer(). Make sure to Release() the blob!
+        if (bd->pd3dDevice->CreateVertexShader(vertexShaderBlob->GetBufferPointer(), vertexShaderBlob->GetBufferSize(), nullptr, &bd->pVertexShader) != S_OK)
+        {
+            vertexShaderBlob->Release();
+            return false;
+        }
+
+        // Create the input layout
+        D3D11_INPUT_ELEMENT_DESC local_layout[] =
+        {
+            { "POSITION", 0, DXGI_FORMAT_R32G32_FLOAT,   0, (UINT)offsetof(ImDrawVert, pos), D3D11_INPUT_PER_VERTEX_DATA, 0 },
+            { "TEXCOORD", 0, DXGI_FORMAT_R32G32_FLOAT,   0, (UINT)offsetof(ImDrawVert, uv),  D3D11_INPUT_PER_VERTEX_DATA, 0 },
+#ifdef IMGUI_USE_BGRA_PACKED_COLOR
+            { "COLOR",    0, DXGI_FORMAT_B8G8R8A8_UNORM, 0, (UINT)offsetof(ImDrawVert, col), D3D11_INPUT_PER_VERTEX_DATA, 0 },
+#else
+            { "COLOR",    0, DXGI_FORMAT_R8G8B8A8_UNORM, 0, (UINT)offsetof(ImDrawVert, col), D3D11_INPUT_PER_VERTEX_DATA, 0 },
+#endif
+        };
+        if (bd->pd3dDevice->CreateInputLayout(local_layout, 3, vertexShaderBlob->GetBufferPointer(), vertexShaderBlob->GetBufferSize(), &bd->pInputLayout) != S_OK)
+        {
+            vertexShaderBlob->Release();
+            return false;
+        }
+        vertexShaderBlob->Release();
+
+        // Create the constant buffer
+        {
+            D3D11_BUFFER_DESC desc = {};
+            desc.ByteWidth = sizeof(VERTEX_CONSTANT_BUFFER_DX11);
+            desc.Usage = D3D11_USAGE_DYNAMIC;
+            desc.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
+            desc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
+            desc.MiscFlags = 0;
+            bd->pd3dDevice->CreateBuffer(&desc, nullptr, &bd->pVertexConstantBuffer);
+        }
+    }
+
+    // Create the pixel shader
+    {
+        static const char* pixelShader =
+            "struct PS_INPUT\
+            {\
+            float4 pos : SV_POSITION;\
+            float4 col : COLOR0;\
+            float2 uv  : TEXCOORD0;\
+            };\
+            sampler sampler0;\
+            Texture2D texture0;\
+            \
+            float4 main(PS_INPUT input) : SV_Target\
+            {\
+            float4 out_col = input.col * texture0.Sample(sampler0, input.uv); \
+            return out_col; \
+            }";
+
+        ID3DBlob* pixelShaderBlob;
+        if (FAILED(D3DCompile_fn(pixelShader, strlen(pixelShader), nullptr, nullptr, nullptr, "main", "ps_4_0", 0, 0, &pixelShaderBlob, nullptr)))
+            return false; // NB: Pass ID3DBlob* pErrorBlob to D3DCompile() to get error showing in (const char*)pErrorBlob->GetBufferPointer(). Make sure to Release() the blob!
+        if (bd->pd3dDevice->CreatePixelShader(pixelShaderBlob->GetBufferPointer(), pixelShaderBlob->GetBufferSize(), nullptr, &bd->pPixelShader) != S_OK)
+        {
+            pixelShaderBlob->Release();
+            return false;
+        }
+        pixelShaderBlob->Release();
+    }
+
+    // Create the blending setup
+    {
+        D3D11_BLEND_DESC desc;
+        ZeroMemory(&desc, sizeof(desc));
+        desc.AlphaToCoverageEnable = false;
+        desc.RenderTarget[0].BlendEnable = true;
+        desc.RenderTarget[0].SrcBlend = D3D11_BLEND_SRC_ALPHA;
+        desc.RenderTarget[0].DestBlend = D3D11_BLEND_INV_SRC_ALPHA;
+        desc.RenderTarget[0].BlendOp = D3D11_BLEND_OP_ADD;
+        desc.RenderTarget[0].SrcBlendAlpha = D3D11_BLEND_ONE;
+        desc.RenderTarget[0].DestBlendAlpha = D3D11_BLEND_INV_SRC_ALPHA;
+        desc.RenderTarget[0].BlendOpAlpha = D3D11_BLEND_OP_ADD;
+        desc.RenderTarget[0].RenderTargetWriteMask = D3D11_COLOR_WRITE_ENABLE_ALL;
+        bd->pd3dDevice->CreateBlendState(&desc, &bd->pBlendState);
+    }
+
+    // Create the rasterizer state
+    {
+        D3D11_RASTERIZER_DESC desc;
+        ZeroMemory(&desc, sizeof(desc));
+        desc.FillMode = D3D11_FILL_SOLID;
+        desc.CullMode = D3D11_CULL_NONE;
+        desc.ScissorEnable = true;
+        desc.DepthClipEnable = true;
+        bd->pd3dDevice->CreateRasterizerState(&desc, &bd->pRasterizerState);
+    }
+
+    // Create depth-stencil State
+    {
+        D3D11_DEPTH_STENCIL_DESC desc;
+        ZeroMemory(&desc, sizeof(desc));
+        desc.DepthEnable = false;
+        desc.DepthWriteMask = D3D11_DEPTH_WRITE_MASK_ALL;
+        desc.DepthFunc = D3D11_COMPARISON_ALWAYS;
+        desc.StencilEnable = false;
+        desc.FrontFace.StencilFailOp = desc.FrontFace.StencilDepthFailOp = desc.FrontFace.StencilPassOp = D3D11_STENCIL_OP_KEEP;
+        desc.FrontFace.StencilFunc = D3D11_COMPARISON_ALWAYS;
+        desc.BackFace = desc.FrontFace;
+        bd->pd3dDevice->CreateDepthStencilState(&desc, &bd->pDepthStencilState);
+    }
+
+    // Create texture sampler
+    // (Bilinear sampling is required by default. Set 'io.Fonts->Flags |= ImFontAtlasFlags_NoBakedLines' or 'style.AntiAliasedLinesUseTex = false' to allow point/nearest sampling)
+    {
+        D3D11_SAMPLER_DESC desc;
+        ZeroMemory(&desc, sizeof(desc));
+        desc.Filter = D3D11_FILTER_MIN_MAG_MIP_LINEAR;
+        desc.AddressU = D3D11_TEXTURE_ADDRESS_CLAMP;
+        desc.AddressV = D3D11_TEXTURE_ADDRESS_CLAMP;
+        desc.AddressW = D3D11_TEXTURE_ADDRESS_CLAMP;
+        desc.MipLODBias = 0.f;
+        desc.ComparisonFunc = D3D11_COMPARISON_ALWAYS;
+        desc.MinLOD = 0.f;
+        desc.MaxLOD = 0.f;
+        bd->pd3dDevice->CreateSamplerState(&desc, &bd->pTexSamplerLinear);
+        desc.Filter = D3D11_FILTER_MIN_MAG_MIP_POINT;
+        bd->pd3dDevice->CreateSamplerState(&desc, &bd->pTexSamplerNearest);
+    }
+
+    return true;
+}
+
+void    ImGui_ImplDX11_InvalidateDeviceObjects()
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    if (!bd->pd3dDevice)
+        return;
+
+    // Destroy all textures
+    for (ImTextureData* tex : ImGui::GetPlatformIO().Textures)
+        if (tex->RefCount == 1)
+            ImGui_ImplDX11_DestroyTexture(tex);
+
+    if (bd->pTexSamplerLinear)      { bd->pTexSamplerLinear->Release(); bd->pTexSamplerLinear = nullptr; }
+    if (bd->pTexSamplerNearest)     { bd->pTexSamplerNearest->Release(); bd->pTexSamplerNearest = nullptr; }
+    if (bd->pIB)                    { bd->pIB->Release(); bd->pIB = nullptr; }
+    if (bd->pVB)                    { bd->pVB->Release(); bd->pVB = nullptr; }
+    if (bd->pBlendState)            { bd->pBlendState->Release(); bd->pBlendState = nullptr; }
+    if (bd->pDepthStencilState)     { bd->pDepthStencilState->Release(); bd->pDepthStencilState = nullptr; }
+    if (bd->pRasterizerState)       { bd->pRasterizerState->Release(); bd->pRasterizerState = nullptr; }
+    if (bd->pPixelShader)           { bd->pPixelShader->Release(); bd->pPixelShader = nullptr; }
+    if (bd->pVertexConstantBuffer)  { bd->pVertexConstantBuffer->Release(); bd->pVertexConstantBuffer = nullptr; }
+    if (bd->pInputLayout)           { bd->pInputLayout->Release(); bd->pInputLayout = nullptr; }
+    if (bd->pVertexShader)          { bd->pVertexShader->Release(); bd->pVertexShader = nullptr; }
+}
+
+bool    ImGui_ImplDX11_Init(ID3D11Device* device, ID3D11DeviceContext* device_context)
+{
+    ImGuiIO& io = ImGui::GetIO();
+    IMGUI_CHECKVERSION();
+    IM_ASSERT(io.BackendRendererUserData == nullptr && "Already initialized a renderer backend!");
+
+    // Setup backend capabilities flags
+    ImGui_ImplDX11_Data* bd = IM_NEW(ImGui_ImplDX11_Data)();
+    io.BackendRendererUserData = (void*)bd;
+    io.BackendRendererName = "imgui_impl_dx11";
+    io.BackendFlags |= ImGuiBackendFlags_RendererHasVtxOffset;  // We can honor the ImDrawCmd::VtxOffset field, allowing for large meshes.
+    io.BackendFlags |= ImGuiBackendFlags_RendererHasTextures;   // We can honor ImGuiPlatformIO::Textures[] requests during render.
+    io.BackendFlags |= ImGuiBackendFlags_RendererHasViewports;  // We can create multi-viewports on the Renderer side (optional)
+
+    ImGuiPlatformIO& platform_io = ImGui::GetPlatformIO();
+    platform_io.Renderer_TextureMaxWidth = platform_io.Renderer_TextureMaxHeight = D3D11_REQ_TEXTURE2D_U_OR_V_DIMENSION;
+    platform_io.DrawCallback_ResetRenderState = ImGui_ImplDX11_DrawCallback_ResetRenderState;
+    platform_io.DrawCallback_SetSamplerLinear = ImGui_ImplDX11_DrawCallback_SetSamplerLinear;
+    platform_io.DrawCallback_SetSamplerNearest = ImGui_ImplDX11_DrawCallback_SetSamplerNearest;
+
+    // Get factory from device
+    IDXGIDevice* pDXGIDevice = nullptr;
+    IDXGIAdapter* pDXGIAdapter = nullptr;
+    IDXGIFactory* pFactory = nullptr;
+
+    if (device->QueryInterface(IID_PPV_ARGS(&pDXGIDevice)) == S_OK)
+        if (pDXGIDevice->GetParent(IID_PPV_ARGS(&pDXGIAdapter)) == S_OK)
+            if (pDXGIAdapter->GetParent(IID_PPV_ARGS(&pFactory)) == S_OK)
+            {
+                bd->pd3dDevice = device;
+                bd->pd3dDeviceContext = device_context;
+                bd->pFactory = pFactory;
+            }
+    if (pDXGIDevice) pDXGIDevice->Release();
+    if (pDXGIAdapter) pDXGIAdapter->Release();
+    bd->pd3dDevice->AddRef();
+    bd->pd3dDeviceContext->AddRef();
+
+    ImGui_ImplDX11_InitMultiViewportSupport();
+
+    return true;
+}
+
+void ImGui_ImplDX11_Shutdown()
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    IM_ASSERT(bd != nullptr && "No renderer backend to shutdown, or already shutdown?");
+    ImGuiIO& io = ImGui::GetIO();
+    ImGuiPlatformIO& platform_io = ImGui::GetPlatformIO();
+
+    ImGui_ImplDX11_ShutdownMultiViewportSupport();
+    ImGui_ImplDX11_InvalidateDeviceObjects();
+    if (bd->pFactory)             { bd->pFactory->Release(); }
+    if (bd->pd3dDevice)           { bd->pd3dDevice->Release(); }
+    if (bd->pd3dDeviceContext)    { bd->pd3dDeviceContext->Release(); }
+
+    io.BackendRendererName = nullptr;
+    io.BackendRendererUserData = nullptr;
+    io.BackendFlags &= ~(ImGuiBackendFlags_RendererHasVtxOffset | ImGuiBackendFlags_RendererHasTextures | ImGuiBackendFlags_RendererHasViewports);
+    platform_io.ClearRendererHandlers();
+    IM_DELETE(bd);
+}
+
+void ImGui_ImplDX11_NewFrame()
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    IM_ASSERT(bd != nullptr && "Context or backend not initialized! Did you call ImGui_ImplDX11_Init()?");
+
+    if (!bd->pVertexShader)
+        if (!ImGui_ImplDX11_CreateDeviceObjects())
+            IM_ASSERT(0 && "ImGui_ImplDX11_CreateDeviceObjects() failed!");
+}
+
+//--------------------------------------------------------------------------------------------------------
+// MULTI-VIEWPORT / PLATFORM INTERFACE SUPPORT
+// This is an _advanced_ and _optional_ feature, allowing the backend to create and handle multiple viewports simultaneously.
+// If you are new to dear imgui or creating a new binding for dear imgui, it is recommended that you completely ignore this section first..
+//--------------------------------------------------------------------------------------------------------
+
+// Helper structure we store in the void* RendererUserData field of each ImGuiViewport to easily retrieve our backend data.
+struct ImGui_ImplDX11_ViewportData
+{
+    IDXGISwapChain*                 SwapChain;
+    ID3D11RenderTargetView*         RTView;
+
+    ImGui_ImplDX11_ViewportData()   { SwapChain = nullptr; RTView = nullptr; }
+    ~ImGui_ImplDX11_ViewportData()  { IM_ASSERT(SwapChain == nullptr && RTView == nullptr); }
+};
+
+// Multi-Viewports: configure templates used when creating swapchains for secondary viewports. Will try them in order.
+// This is intentionally not declared in the .h file yet, so you will need to copy this declaration:
+void ImGui_ImplDX11_SetSwapChainDescs(const DXGI_SWAP_CHAIN_DESC* desc_templates, int desc_templates_count);
+void ImGui_ImplDX11_SetSwapChainDescs(const DXGI_SWAP_CHAIN_DESC* desc_templates, int desc_templates_count)
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    bd->SwapChainDescsForViewports.resize(desc_templates_count);
+    memcpy(bd->SwapChainDescsForViewports.Data, desc_templates, sizeof(DXGI_SWAP_CHAIN_DESC));
+}
+
+static void ImGui_ImplDX11_CreateWindow(ImGuiViewport* viewport)
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    ImGui_ImplDX11_ViewportData* vd = IM_NEW(ImGui_ImplDX11_ViewportData)();
+    viewport->RendererUserData = vd;
+
+    // PlatformHandleRaw should always be a HWND, whereas PlatformHandle might be a higher-level handle (e.g. GLFWWindow*, SDL's WindowID).
+    // Some backends will leave PlatformHandleRaw == 0, in which case we assume PlatformHandle will contain the HWND.
+    HWND hwnd = viewport->PlatformHandleRaw ? (HWND)viewport->PlatformHandleRaw : (HWND)viewport->PlatformHandle;
+    IM_ASSERT(hwnd != 0);
+    IM_ASSERT(vd->SwapChain == nullptr && vd->RTView == nullptr);
+
+    // Create swap chain
+    HRESULT hr = DXGI_ERROR_UNSUPPORTED;
+    for (const DXGI_SWAP_CHAIN_DESC& sd_template : bd->SwapChainDescsForViewports)
+    {
+        IM_ASSERT(sd_template.BufferDesc.Width == 0 && sd_template.BufferDesc.Height == 0 && sd_template.OutputWindow == nullptr);
+        DXGI_SWAP_CHAIN_DESC sd = sd_template;
+        sd.BufferDesc.Width = (UINT)viewport->Size.x;
+        sd.BufferDesc.Height = (UINT)viewport->Size.y;
+        sd.OutputWindow = hwnd;
+        hr = bd->pFactory->CreateSwapChain(bd->pd3dDevice, &sd, &vd->SwapChain);
+        if (SUCCEEDED(hr))
+            break;
+    }
+    IM_ASSERT(SUCCEEDED(hr));
+    bd->pFactory->MakeWindowAssociation(hwnd, DXGI_MWA_NO_ALT_ENTER | DXGI_MWA_NO_WINDOW_CHANGES); // Disable e.g. Alt+Enter
+
+    // Create the render target
+    if (vd->SwapChain != nullptr)
+    {
+        ID3D11Texture2D* pBackBuffer;
+        vd->SwapChain->GetBuffer(0, IID_PPV_ARGS(&pBackBuffer));
+        bd->pd3dDevice->CreateRenderTargetView(pBackBuffer, nullptr, &vd->RTView);
+        pBackBuffer->Release();
+    }
+}
+
+static void ImGui_ImplDX11_DestroyWindow(ImGuiViewport* viewport)
+{
+    // The main viewport (owned by the application) will always have RendererUserData == nullptr since we didn't create the data for it.
+    if (ImGui_ImplDX11_ViewportData* vd = (ImGui_ImplDX11_ViewportData*)viewport->RendererUserData)
+    {
+        if (vd->SwapChain)
+            vd->SwapChain->Release();
+        vd->SwapChain = nullptr;
+        if (vd->RTView)
+            vd->RTView->Release();
+        vd->RTView = nullptr;
+        IM_DELETE(vd);
+    }
+    viewport->RendererUserData = nullptr;
+}
+
+static void ImGui_ImplDX11_SetWindowSize(ImGuiViewport* viewport, ImVec2 size)
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    ImGui_ImplDX11_ViewportData* vd = (ImGui_ImplDX11_ViewportData*)viewport->RendererUserData;
+    if (vd->RTView)
+    {
+        vd->RTView->Release();
+        vd->RTView = nullptr;
+    }
+    if (vd->SwapChain)
+    {
+        ID3D11Texture2D* pBackBuffer = nullptr;
+        vd->SwapChain->ResizeBuffers(0, (UINT)size.x, (UINT)size.y, DXGI_FORMAT_UNKNOWN, 0);
+        vd->SwapChain->GetBuffer(0, IID_PPV_ARGS(&pBackBuffer));
+        if (pBackBuffer == nullptr) { fprintf(stderr, "ImGui_ImplDX11_SetWindowSize() failed creating buffers.\n"); return; }
+        bd->pd3dDevice->CreateRenderTargetView(pBackBuffer, nullptr, &vd->RTView);
+        pBackBuffer->Release();
+    }
+}
+
+static void ImGui_ImplDX11_RenderWindow(ImGuiViewport* viewport, void*)
+{
+    ImGui_ImplDX11_Data* bd = ImGui_ImplDX11_GetBackendData();
+    ImGui_ImplDX11_ViewportData* vd = (ImGui_ImplDX11_ViewportData*)viewport->RendererUserData;
+    ImVec4 clear_color = ImVec4(0.0f, 0.0f, 0.0f, 1.0f);
+    bd->pd3dDeviceContext->OMSetRenderTargets(1, &vd->RTView, nullptr);
+    if (!(viewport->Flags & ImGuiViewportFlags_NoRendererClear))
+        bd->pd3dDeviceContext->ClearRenderTargetView(vd->RTView, (float*)&clear_color);
+    ImGui_ImplDX11_RenderDrawData(viewport->DrawData);
+}
+
+static void ImGui_ImplDX11_SwapBuffers(ImGuiViewport* viewport, void*)
+{
+    ImGui_ImplDX11_ViewportData* vd = (ImGui_ImplDX11_ViewportData*)viewport->RendererUserData;
+    if (vd->SwapChain)
+        vd->SwapChain->Present(0, 0); // Present without vsync
+}
+
+static void ImGui_ImplDX11_InitMultiViewportSupport()
+{
+    ImGuiPlatformIO& platform_io = ImGui::GetPlatformIO();
+    platform_io.Renderer_CreateWindow = ImGui_ImplDX11_CreateWindow;
+    platform_io.Renderer_DestroyWindow = ImGui_ImplDX11_DestroyWindow;
+    platform_io.Renderer_SetWindowSize = ImGui_ImplDX11_SetWindowSize;
+    platform_io.Renderer_RenderWindow = ImGui_ImplDX11_RenderWindow;
+    platform_io.Renderer_SwapBuffers = ImGui_ImplDX11_SwapBuffers;
+
+    // Default swapchain format
+    DXGI_SWAP_CHAIN_DESC sd;
+    ZeroMemory(&sd, sizeof(sd));
+    sd.BufferDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
+    sd.SampleDesc.Count = 1;
+    sd.SampleDesc.Quality = 0;
+    sd.BufferUsage = DXGI_USAGE_RENDER_TARGET_OUTPUT;
+    sd.BufferCount = 1;
+    sd.Windowed = TRUE;
+    sd.SwapEffect = DXGI_SWAP_EFFECT_DISCARD;
+    sd.Flags = 0;
+    ImGui_ImplDX11_SetSwapChainDescs(&sd, 1);
+}
+
+static void ImGui_ImplDX11_ShutdownMultiViewportSupport()
+{
+    ImGui::DestroyPlatformWindows();
+}
+
+//-----------------------------------------------------------------------------
+
+#endif // #ifndef IMGUI_DISABLE
@@ -0,0 +1,53 @@
+// dear imgui: Renderer Backend for DirectX11
+// This needs to be used along with a Platform Backend (e.g. Win32)
+
+// Implemented features:
+//  [X] Renderer: User texture binding. Use 'ID3D11ShaderResourceView*' as texture identifier. Read the FAQ about ImTextureID/ImTextureRef!
+//  [X] Renderer: Large meshes support (64k+ vertices) even with 16-bit indices (ImGuiBackendFlags_RendererHasVtxOffset).
+//  [X] Renderer: Texture updates support for dynamic font atlas (ImGuiBackendFlags_RendererHasTextures).
+//  [X] Renderer: Expose selected render state for draw callbacks to use. Access in '(ImGui_ImplXXXX_RenderState*)GetPlatformIO().Renderer_RenderState'.
+//  [X] Renderer: Multi-viewport support (multiple windows). Enable with 'io.ConfigFlags |= ImGuiConfigFlags_ViewportsEnable'.
+
+// You can use unmodified imgui_impl_* files in your project. See examples/ folder for examples of using this.
+// Prefer including the entire imgui/ repository into your project (either as a copy or as a submodule), and only build the backends you need.
+// Learn about Dear ImGui:
+// - FAQ                  https://dearimgui.com/faq
+// - Getting Started      https://dearimgui.com/getting-started
+// - Documentation        https://dearimgui.com/docs (same as your local docs/ folder).
+// - Introduction, links and more at the top of imgui.cpp
+
+#pragma once
+#include "imgui.h"      // IMGUI_IMPL_API
+#ifndef IMGUI_DISABLE
+
+struct ID3D11Device;
+struct ID3D11DeviceContext;
+struct ID3D11SamplerState;
+struct ID3D11Buffer;
+
+// Follow "Getting Started" link and check examples/ folder to learn about using backends!
+IMGUI_IMPL_API bool     ImGui_ImplDX11_Init(ID3D11Device* device, ID3D11DeviceContext* device_context);
+IMGUI_IMPL_API void     ImGui_ImplDX11_Shutdown();
+IMGUI_IMPL_API void     ImGui_ImplDX11_NewFrame();
+IMGUI_IMPL_API void     ImGui_ImplDX11_RenderDrawData(ImDrawData* draw_data);
+
+// Use if you want to reset your rendering device without losing Dear ImGui state.
+IMGUI_IMPL_API bool     ImGui_ImplDX11_CreateDeviceObjects();
+IMGUI_IMPL_API void     ImGui_ImplDX11_InvalidateDeviceObjects();
+
+// (Advanced) Use e.g. if you need to precisely control the timing of texture updates (e.g. for staged rendering), by setting ImDrawData::Textures = nullptr to handle this manually.
+IMGUI_IMPL_API void     ImGui_ImplDX11_UpdateTexture(ImTextureData* tex);
+
+// [BETA] Selected render state data shared with callbacks.
+// This is temporarily stored in GetPlatformIO().Renderer_RenderState during the ImGui_ImplDX11_RenderDrawData() call.
+// (Please open an issue if you feel you need access to more data)
+struct ImGui_ImplDX11_RenderState
+{
+    ID3D11Device*           Device;
+    ID3D11DeviceContext*    DeviceContext;
+    ID3D11Buffer*           VertexConstantBuffer;
+    //ID3D11SamplerState*   SamplerLinear;          // Use ImDrawList::AddCallback(ImGui::GetPlatform().DrawCallback_SetSamplerLinear)
+    //ID3D11SamplerState*   SamplerNearest;         // Use ImDrawList::AddCallback(ImGui::GetPlatform().DrawCallback_SetSamplerNearest)
+};
+
+#endif // #ifndef IMGUI_DISABLE
@@ -1,25 +1,38 @@
-// dear imgui: Renderer for DirectX9
-// This needs to be used along with a Platform Binding (e.g. Win32)
+// dear imgui: Renderer Backend for DirectX9
+// This needs to be used along with a Platform Backend (e.g. Win32)

 // Implemented features:
-//  [X] Renderer: User texture binding. Use 'LPDIRECT3DTEXTURE9' as ImTextureID. Read the FAQ about ImTextureID in imgui.cpp.
-//  [X] Renderer: Support for large meshes (64k+ vertices) with 16-bits indices.
+//  [X] Renderer: User texture binding. Use 'LPDIRECT3DTEXTURE9' as texture identifier. Read the FAQ about ImTextureID/ImTextureRef!
+//  [X] Renderer: Large meshes support (64k+ vertices) even with 16-bit indices (ImGuiBackendFlags_RendererHasVtxOffset).
+//  [X] Renderer: Texture updates support for dynamic font atlas (ImGuiBackendFlags_RendererHasTextures).
+//  [X] Renderer: IMGUI_USE_BGRA_PACKED_COLOR support, as this is the optimal color encoding for DirectX9.
+//  [X] Renderer: Multi-viewport support (multiple windows). Enable with 'io.ConfigFlags |= ImGuiConfigFlags_ViewportsEnable'.

-// You can copy and use unmodified imgui_impl_* files in your project. See main.cpp for an example of using this.
-// If you are new to dear imgui, read examples/README.txt and read the documentation at the top of imgui.cpp.
-// https://github.com/ocornut/imgui
+// You can use unmodified imgui_impl_* files in your project. See examples/ folder for examples of using this.
+// Prefer including the entire imgui/ repository into your project (either as a copy or as a submodule), and only build the backends you need.
+// Learn about Dear ImGui:
+// - FAQ                  https://dearimgui.com/faq
+// - Getting Started      https://dearimgui.com/getting-started
+// - Documentation        https://dearimgui.com/docs (same as your local docs/ folder).
+// - Introduction, links and more at the top of imgui.cpp

 #pragma once
-
-#include "imgui.h"
+#include "imgui.h"      // IMGUI_IMPL_API
+#ifndef IMGUI_DISABLE

 struct IDirect3DDevice9;

-IMGUI_IMPL_API bool     ImGui_ImplDX9_Init(IDirect3DDevice9 *device);
+// Follow "Getting Started" link and check examples/ folder to learn about using backends!
+IMGUI_IMPL_API bool     ImGui_ImplDX9_Init(IDirect3DDevice9* device);
 IMGUI_IMPL_API void     ImGui_ImplDX9_Shutdown();
 IMGUI_IMPL_API void     ImGui_ImplDX9_NewFrame();
-IMGUI_IMPL_API void     ImGui_ImplDX9_RenderDrawData(ImDrawData *draw_data);
+IMGUI_IMPL_API void     ImGui_ImplDX9_RenderDrawData(ImDrawData* draw_data);

-// Use if you want to reset your rendering device without losing ImGui state.
+// Use if you want to reset your rendering device without losing Dear ImGui state.
 IMGUI_IMPL_API bool     ImGui_ImplDX9_CreateDeviceObjects();
 IMGUI_IMPL_API void     ImGui_ImplDX9_InvalidateDeviceObjects();
+
+// (Advanced) Use e.g. if you need to precisely control the timing of texture updates (e.g. for staged rendering), by setting ImDrawData::Textures = nullptr to handle this manually.
+IMGUI_IMPL_API void     ImGui_ImplDX9_UpdateTexture(ImTextureData* tex);
+
+#endif // #ifndef IMGUI_DISABLE
@@ -15,7 +15,8 @@
 #pragma once

 //---- Define assertion handler. Defaults to calling assert().
-// If your macro uses multiple statements, make sure is enclosed in a 'do { .. } while (0)' block so it can be used as a single statement.
+// - If your macro uses multiple statements, make sure is enclosed in a 'do { .. } while (0)' block so it can be used as a single statement.
+// - Compiling with NDEBUG will usually strip out assert() to nothing, which is NOT recommended because we use asserts to notify of programmer mistakes.
 //#define IM_ASSERT(_EXPR)  MyAssert(_EXPR)
 #define IM_ASSERT(_EXPR)  ((void)(_EXPR))     // Disable asserts

@@ -29,7 +30,6 @@

 //---- Don't define obsolete functions/enums/behaviors. Consider enabling from time to time after updating to clean your code of obsolete function/names.
 //#define IMGUI_DISABLE_OBSOLETE_FUNCTIONS
-//#define IMGUI_DISABLE_OBSOLETE_KEYIO                      // 1.87+ disable legacy io.KeyMap[]+io.KeysDown[] in favor io.AddKeyEvent(). This is automatically done by IMGUI_DISABLE_OBSOLETE_FUNCTIONS.

 //---- Disable all of Dear ImGui or don't implement standard windows/tools.
 // It is very strongly recommended to NOT disable the demo windows and debug tool during development. They are extremely useful in day to day work. Please read comments in imgui_demo.cpp.
@@ -49,6 +49,7 @@
 //#define IMGUI_DISABLE_FILE_FUNCTIONS                      // Don't implement ImFileOpen/ImFileClose/ImFileRead/ImFileWrite and ImFileHandle at all (replace them with dummies)
 //#define IMGUI_DISABLE_DEFAULT_FILE_FUNCTIONS              // Don't implement ImFileOpen/ImFileClose/ImFileRead/ImFileWrite and ImFileHandle so you can implement them yourself if you don't want to link with fopen/fclose/fread/fwrite. This will also disable the LogToTTY() function.
 #define IMGUI_DISABLE_DEFAULT_ALLOCATORS                  // Don't implement default allocators calling malloc()/free() to avoid linking with them. You will need to call ImGui::SetAllocatorFunctions().
+//#define IMGUI_DISABLE_DEFAULT_FONT                        // Disable default embedded fonts (ProggyClean/ProggyForever), remove ~9 KB + ~14 KB from output binary. AddFontDefaultXXX() functions will assert.
 //#define IMGUI_DISABLE_SSE                                 // Disable use of SSE intrinsics even if available

 //---- Enable Test Engine / Automation features.
@@ -59,12 +60,15 @@
 //#define IMGUI_INCLUDE_IMGUI_USER_H
 //#define IMGUI_USER_H_FILENAME         "my_folder/my_imgui_user.h"

-//---- Pack colors to BGRA8 instead of RGBA8 (to avoid converting from one to another)
+//---- Pack vertex colors as BGRA8 instead of RGBA8 (to avoid converting from one to another). Need dedicated backend support.
 #define IMGUI_USE_BGRA_PACKED_COLOR

 //---- Performance optimization for DX9
 #define IMGUI_OVERRIDE_DRAWVERT_STRUCT_LAYOUT struct ImDrawVert { ImVec2 pos; float z; ImU32 col; ImVec2 uv; }

+//---- Use legacy CRC32-adler tables (used before 1.91.6), in order to preserve old .ini data that you cannot afford to invalidate.
+//#define IMGUI_USE_LEGACY_CRC32_ADLER
+
 //---- Use 32-bit for ImWchar (default is 16-bit) to support Unicode planes 1-16. (e.g. point beyond 0xFFFF like emoticons, dingbats, symbols, shapes, ancient languages, etc...)
 //#define IMGUI_USE_WCHAR32

@@ -83,13 +87,13 @@

 //---- Use FreeType to build and rasterize the font atlas (instead of stb_truetype which is embedded by default in Dear ImGui)
 // Requires FreeType headers to be available in the include path. Requires program to be compiled with 'misc/freetype/imgui_freetype.cpp' (in this repository) + the FreeType library (not provided).
+// Note that imgui_freetype.cpp may be used _without_ this define, if you manually call ImFontAtlas::SetFontLoader(). The define is simply a convenience.
 // On Windows you may use vcpkg with 'vcpkg install freetype --triplet=x64-windows' + 'vcpkg integrate install'.
 //#define IMGUI_ENABLE_FREETYPE

 //---- Use FreeType + plutosvg or lunasvg to render OpenType SVG fonts (SVGinOT)
 // Only works in combination with IMGUI_ENABLE_FREETYPE.
-// - lunasvg is currently easier to acquire/install, as e.g. it is part of vcpkg.
-// - plutosvg will support more fonts and may load them faster. It currently requires to be built manually but it is fairly easy. See misc/freetype/README for instructions.
+// - plutosvg is currently easier to install, as e.g. it is part of vcpkg. It will support more fonts and may load them faster. See misc/freetype/README for instructions.
 // - Both require headers to be available in the include path + program to be linked with the library code (not provided).
 // - (note: lunasvg implementation is based on Freetype's rsvg-port.c which is licensed under CeCILL-C Free Software License Agreement)
 //#define IMGUI_ENABLE_FREETYPE_PLUTOSVG
@@ -130,6 +134,10 @@
 //#define IM_DEBUG_BREAK  IM_ASSERT(0)
 //#define IM_DEBUG_BREAK  __debugbreak()

+//---- Debug Tools: Enable highlight ID conflicts _before_ hovering items. When io.ConfigDebugHighlightIdConflicts is set.
+// (THIS WILL SLOW DOWN DEAR IMGUI. Only use occasionally and disable after use)
+//#define IMGUI_DEBUG_HIGHLIGHT_ALL_ID_CONFLICTS
+
 //---- Debug Tools: Enable slower asserts
 //#define IMGUI_DEBUG_PARANOID

@@ -878,7 +878,15 @@ inline void ImGui::FileBrowser::UpdateFileRecords()

 inline void ImGui::FileBrowser::SetPwdUncatched(const std::filesystem::path &pwd)
 {
-    pwd_ = absolute(pwd);
+    // spice2x
+    // avoid MinGW std::filesystem::absolute() duplicating UNC server/share
+    // prefixes, which makes directory_iterator() throw during startup
+    // spice2x
+    const auto &nativePwd = pwd.native();
+    const bool hasUncPrefix = nativePwd.size() >= 2 &&
+        (nativePwd[0] == '\\' || nativePwd[0] == '/') &&
+        (nativePwd[1] == '\\' || nativePwd[1] == '/');
+    pwd_ = hasUncPrefix ? pwd : absolute(pwd);
    UpdateFileRecords();
    selectedFilenames_.clear();
    (*inputNameBuf_)[0] = '\0';
@@ -315,7 +315,7 @@ static int stbrp__skyline_find_min_y(stbrp_context *c, stbrp_node *first, int x0
      if (node->y > min_y) {
         // raise min_y higher.
         // we've accounted for all waste up to min_y,
-         // but we'll now add more waste for everything we've visted
+         // but we'll now add more waste for everything we've visited
         waste_area += visited_width * (node->y - min_y);
         min_y = node->y;
         // the first time through, visited_width might be reduced
@@ -470,7 +470,7 @@ static stbrp__findresult stbrp__skyline_pack_rectangle(stbrp_context *context, i

   // insert the new node into the right starting point, and
   // let 'cur' point to the remaining nodes needing to be
-   // stiched back in
+   // stitched back in

   cur = *res.prev_link;
   if (cur->x < res.x) {
@@ -5,7 +5,8 @@
 // - Fix in stb_textedit_find_charpos to handle last line (see https://github.com/ocornut/imgui/issues/6000 + #6783)
 // - Added name to struct or it may be forward declared in our code.
 // - Added UTF-8 support (see https://github.com/nothings/stb/issues/188 + https://github.com/ocornut/imgui/pull/7925)
-// Grep for [DEAR IMGUI] to find the changes.
+// - Changed STB_TEXTEDIT_INSERTCHARS() to return inserted count (instead of 0/1 bool), allowing partial insertion.
+// Grep for [DEAR IMGUI] to find some changes.
 // - Also renamed macros used or defined outside of IMSTB_TEXTEDIT_IMPLEMENTATION block from STB_TEXTEDIT_* to IMSTB_TEXTEDIT_*

 // stb_textedit.h - v1.14  - public domain - Sean Barrett
@@ -39,6 +40,7 @@
 //
 // VERSION HISTORY
 //
+//   !!!! (2025-10-23) changed STB_TEXTEDIT_INSERTCHARS() to return inserted count (instead of 0/1 bool), allowing partial insertion.
 //   1.14 (2021-07-11) page up/down, various fixes
 //   1.13 (2019-02-07) fix bug in undo size management
 //   1.12 (2018-01-29) user can change STB_TEXTEDIT_KEYTYPE, fix redo to avoid crash
@@ -141,12 +143,14 @@
 //                                        with previous char)
 //    STB_TEXTEDIT_KEYTOTEXT(k)         maps a keyboard input to an insertable character
 //                                        (return type is int, -1 means not valid to insert)
+//                                        (not supported if you want to use UTF-8, see below)
 //    STB_TEXTEDIT_GETCHAR(obj,i)       returns the i'th character of obj, 0-based
 //    STB_TEXTEDIT_NEWLINE              the character returned by _GETCHAR() we recognize
 //                                        as manually wordwrapping for end-of-line positioning
 //
 //    STB_TEXTEDIT_DELETECHARS(obj,i,n)      delete n characters starting at i
-//    STB_TEXTEDIT_INSERTCHARS(obj,i,c*,n)   insert n characters at i (pointed to by STB_TEXTEDIT_CHARTYPE*)
+//    STB_TEXTEDIT_INSERTCHARS(obj,i,c*,n)   try to insert n characters at i (pointed to by STB_TEXTEDIT_CHARTYPE*)
+//                                           returns number of characters actually inserted. [DEAR IMGUI]
 //
 //    STB_TEXTEDIT_K_SHIFT       a power of two that is or'd in to a keyboard input to represent the shift key
 //
@@ -178,6 +182,13 @@
 //    STB_TEXTEDIT_K_TEXTSTART2          secondary keyboard input to move cursor to start of text
 //    STB_TEXTEDIT_K_TEXTEND2            secondary keyboard input to move cursor to end of text
 //
+// To support UTF-8:
+//
+//    STB_TEXTEDIT_GETPREVCHARINDEX      returns index of previous character
+//    STB_TEXTEDIT_GETNEXTCHARINDEX      returns index of next character
+//    Do NOT define STB_TEXTEDIT_KEYTOTEXT.
+//    Instead, call stb_textedit_text() directly for text contents.
+//
 // Keyboard input must be encoded as a single integer value; e.g. a character code
 // and some bitflags that represent shift states. to simplify the interface, SHIFT must
 // be a bitflag, so we can test the shifted state of cursor movements to allow selection,
@@ -250,8 +261,10 @@
 //          if the STB_TEXTEDIT_KEYTOTEXT function is defined, selected keys are
 //          transformed into text and stb_textedit_text() is automatically called.
 //
-//      text: [DEAR IMGUI] added 2024-09
-//          call this to text inputs sent to the textfield.
+//      text: (added 2025)
+//          call this to directly send text input the textfield, which is required
+//          for UTF-8 support, because stb_textedit_key() + STB_TEXTEDIT_KEYTOTEXT()
+//          cannot infer text length.
 //
 //
 //   When rendering, you can read the cursor position and selection state from
@@ -400,6 +413,16 @@ typedef struct
 #define IMSTB_TEXTEDIT_memmove memmove
 #endif

+// [DEAR IMGUI]
+// Functions must be implemented for UTF8 support
+// Code in this file that uses those functions is modified for [DEAR IMGUI] and deviates from the original stb_textedit.
+// There is not necessarily a '[DEAR IMGUI]' at the usage sites.
+#ifndef IMSTB_TEXTEDIT_GETPREVCHARINDEX
+#define IMSTB_TEXTEDIT_GETPREVCHARINDEX(OBJ, IDX) ((IDX) - 1)
+#endif
+#ifndef IMSTB_TEXTEDIT_GETNEXTCHARINDEX
+#define IMSTB_TEXTEDIT_GETNEXTCHARINDEX(OBJ, IDX) ((IDX) + 1)
+#endif

 /////////////////////////////////////////////////////////////////////////////
 //
@@ -407,7 +430,7 @@ typedef struct
 //

 // traverse the layout to locate the nearest character to a display position
-static int stb_text_locate_coord(IMSTB_TEXTEDIT_STRING *str, float x, float y)
+static int stb_text_locate_coord(IMSTB_TEXTEDIT_STRING *str, float x, float y, int* out_side_on_line)
 {
   StbTexteditRow r;
   int n = STB_TEXTEDIT_STRINGLEN(str);
@@ -417,6 +440,7 @@ static int stb_text_locate_coord(IMSTB_TEXTEDIT_STRING *str, float x, float y)
   r.x0 = r.x1 = 0;
   r.ymin = r.ymax = 0;
   r.num_chars = 0;
+   *out_side_on_line = 0;

   // search rows to find one that straddles 'y'
   while (i < n) {
@@ -436,7 +460,10 @@ static int stb_text_locate_coord(IMSTB_TEXTEDIT_STRING *str, float x, float y)

   // below all text, return 'after' last character
   if (i >= n)
+   {
+      *out_side_on_line = 1;
      return n;
+   }

   // check if it's before the beginning of the line
   if (x < r.x0)
@@ -449,6 +476,7 @@ static int stb_text_locate_coord(IMSTB_TEXTEDIT_STRING *str, float x, float y)
      for (k=0; k < r.num_chars; k = IMSTB_TEXTEDIT_GETNEXTCHARINDEX(str, i + k) - i) {
         float w = STB_TEXTEDIT_GETWIDTH(str, i, k);
         if (x < prev_x+w) {
+            *out_side_on_line = (k == 0) ? 0 : 1;
            if (x < prev_x+w/2)
               return k+i;
            else
@@ -460,6 +488,7 @@ static int stb_text_locate_coord(IMSTB_TEXTEDIT_STRING *str, float x, float y)
   }

   // if the last character is a newline, return that. otherwise return 'after' the last character
+   *out_side_on_line = 1;
   if (STB_TEXTEDIT_GETCHAR(str, i+r.num_chars-1) == STB_TEXTEDIT_NEWLINE)
      return i+r.num_chars-1;
   else
@@ -471,6 +500,7 @@ static void stb_textedit_click(IMSTB_TEXTEDIT_STRING *str, STB_TexteditState *st
 {
   // In single-line mode, just always make y = 0. This lets the drag keep working if the mouse
   // goes off the top or bottom of the text
+   int side_on_line;
   if( state->single_line )
   {
      StbTexteditRow r;
@@ -478,16 +508,18 @@ static void stb_textedit_click(IMSTB_TEXTEDIT_STRING *str, STB_TexteditState *st
      y = r.ymin;
   }

-   state->cursor = stb_text_locate_coord(str, x, y);
+   state->cursor = stb_text_locate_coord(str, x, y, &side_on_line);
   state->select_start = state->cursor;
   state->select_end = state->cursor;
   state->has_preferred_x = 0;
+   str->LastMoveDirectionLR = (ImS8)(side_on_line ? ImGuiDir_Right : ImGuiDir_Left);
 }

 // API drag: on mouse drag, move the cursor and selection endpoint to the clicked location
 static void stb_textedit_drag(IMSTB_TEXTEDIT_STRING *str, STB_TexteditState *state, float x, float y)
 {
   int p = 0;
+   int side_on_line;

   // In single-line mode, just always make y = 0. This lets the drag keep working if the mouse
   // goes off the top or bottom of the text
@@ -501,8 +533,9 @@ static void stb_textedit_drag(IMSTB_TEXTEDIT_STRING *str, STB_TexteditState *sta
   if (state->select_start == state->select_end)
      state->select_start = state->cursor;

-   p = stb_text_locate_coord(str, x, y);
+   p = stb_text_locate_coord(str, x, y, &side_on_line);
   state->cursor = state->select_end = p;
+   str->LastMoveDirectionLR = (ImS8)(side_on_line ? ImGuiDir_Right : ImGuiDir_Left);
 }

 /////////////////////////////////////////////////////////////////////////////
@@ -552,6 +585,8 @@ static void stb_textedit_find_charpos(StbFindState *find, IMSTB_TEXTEDIT_STRING
      STB_TEXTEDIT_LAYOUTROW(&r, str, i);
      if (n < i + r.num_chars)
         break;
+      if (str->LastMoveDirectionLR == ImGuiDir_Right && str->Stb->cursor > 0 && str->Stb->cursor == i + r.num_chars && STB_TEXTEDIT_GETCHAR(str, i + r.num_chars - 1) != STB_TEXTEDIT_NEWLINE) // [DEAR IMGUI] Wrapping point handling
+         break;
      if (i + r.num_chars == z && z > 0 && STB_TEXTEDIT_GETCHAR(str, z - 1) != STB_TEXTEDIT_NEWLINE)  // [DEAR IMGUI] special handling for last line
         break;   // [DEAR IMGUI]
      prev_start = i;
@@ -648,15 +683,33 @@ static void stb_textedit_move_to_last(IMSTB_TEXTEDIT_STRING *str, STB_TexteditSt
   }
 }

-// [DEAR IMGUI]
-// Functions must be implemented for UTF8 support
-// Code in this file that uses those functions is modified for [DEAR IMGUI] and deviates from the original stb_textedit.
-// There is not necessarily a '[DEAR IMGUI]' at the usage sites.
-#ifndef IMSTB_TEXTEDIT_GETPREVCHARINDEX
-#define IMSTB_TEXTEDIT_GETPREVCHARINDEX(obj, idx) (idx - 1)
+// [DEAR IMGUI] Extracted this function so we can more easily add support for word-wrapping.
+#ifndef STB_TEXTEDIT_MOVELINESTART
+static int stb_textedit_move_line_start(IMSTB_TEXTEDIT_STRING *str, STB_TexteditState *state, int cursor)
+{
+   if (state->single_line)
+      return 0;
+   while (cursor > 0) {
+      int prev = IMSTB_TEXTEDIT_GETPREVCHARINDEX(str, cursor);
+      if (STB_TEXTEDIT_GETCHAR(str, prev) == STB_TEXTEDIT_NEWLINE)
+         break;
+      cursor = prev;
+   }
+   return cursor;
+}
+#define STB_TEXTEDIT_MOVELINESTART stb_textedit_move_line_start
 #endif
-#ifndef IMSTB_TEXTEDIT_GETNEXTCHARINDEX
-#define IMSTB_TEXTEDIT_GETNEXTCHARINDEX(obj, idx) (idx + 1)
+#ifndef STB_TEXTEDIT_MOVELINEEND
+static int stb_textedit_move_line_end(IMSTB_TEXTEDIT_STRING *str, STB_TexteditState *state, int cursor)
+{
+   int n = STB_TEXTEDIT_STRINGLEN(str);
+   if (state->single_line)
+      return n;
+   while (cursor < n && STB_TEXTEDIT_GETCHAR(str, cursor) != STB_TEXTEDIT_NEWLINE)
+      cursor = IMSTB_TEXTEDIT_GETNEXTCHARINDEX(str, cursor);
+   return cursor;
+}
+#define STB_TEXTEDIT_MOVELINEEND stb_textedit_move_line_end
 #endif

 #ifdef STB_TEXTEDIT_IS_SPACE
@@ -668,9 +721,9 @@ static int is_word_boundary( IMSTB_TEXTEDIT_STRING *str, int idx )
 #ifndef STB_TEXTEDIT_MOVEWORDLEFT
 static int stb_textedit_move_to_word_previous( IMSTB_TEXTEDIT_STRING *str, int c )
 {
-   --c; // always move at least one character
-   while( c >= 0 && !is_word_boundary( str, c ) )
-      --c;
+   c = IMSTB_TEXTEDIT_GETPREVCHARINDEX( str, c ); // always move at least one character
+   while (c >= 0 && !is_word_boundary(str, c))
+      c = IMSTB_TEXTEDIT_GETPREVCHARINDEX(str, c);

   if( c < 0 )
      c = 0;
@@ -684,9 +737,9 @@ static int stb_textedit_move_to_word_previous( IMSTB_TEXTEDIT_STRING *str, int c
 static int stb_textedit_move_to_word_next( IMSTB_TEXTEDIT_STRING *str, int c )
 {
   const int len = STB_TEXTEDIT_STRINGLEN(str);
-   ++c; // always move at least one character
+   c = IMSTB_TEXTEDIT_GETNEXTCHARINDEX(str, c); // always move at least one character
   while( c < len && !is_word_boundary( str, c ) )
-      ++c;
+      c = IMSTB_TEXTEDIT_GETNEXTCHARINDEX(str, c);

   if( c > len )
      c = len;
@@ -725,7 +778,8 @@ static int stb_textedit_paste_internal(IMSTB_TEXTEDIT_STRING *str, STB_TexteditS
   stb_textedit_clamp(str, state);
   stb_textedit_delete_selection(str,state);
   // try to insert the characters
-   if (STB_TEXTEDIT_INSERTCHARS(str, state->cursor, text, len)) {
+   len = STB_TEXTEDIT_INSERTCHARS(str, state->cursor, text, len);
+   if (len) {
      stb_text_makeundo_insert(state, state->cursor, len);
      state->cursor += len;
      state->has_preferred_x = 0;
@@ -739,6 +793,7 @@ static int stb_textedit_paste_internal(IMSTB_TEXTEDIT_STRING *str, STB_TexteditS
 #define STB_TEXTEDIT_KEYTYPE int
 #endif

+// API key: process text input
 // [DEAR IMGUI] Added stb_textedit_text(), extracted out and called by stb_textedit_key() for backward compatibility.
 static void stb_textedit_text(IMSTB_TEXTEDIT_STRING* str, STB_TexteditState* state, const IMSTB_TEXTEDIT_CHARTYPE* text, int text_len)
 {
@@ -749,14 +804,15 @@ static void stb_textedit_text(IMSTB_TEXTEDIT_STRING* str, STB_TexteditState* sta
   if (state->insert_mode && !STB_TEXT_HAS_SELECTION(state) && state->cursor < STB_TEXTEDIT_STRINGLEN(str)) {
      stb_text_makeundo_replace(str, state, state->cursor, 1, 1);
      STB_TEXTEDIT_DELETECHARS(str, state->cursor, 1);
-      if (STB_TEXTEDIT_INSERTCHARS(str, state->cursor, text, text_len)) {
+      text_len = STB_TEXTEDIT_INSERTCHARS(str, state->cursor, text, text_len);
+      if (text_len) {
         state->cursor += text_len;
         state->has_preferred_x = 0;
      }
-   }
-   else {
+   } else {
      stb_textedit_delete_selection(str, state); // implicitly clamps
-      if (STB_TEXTEDIT_INSERTCHARS(str, state->cursor, text, text_len)) {
+      text_len = STB_TEXTEDIT_INSERTCHARS(str, state->cursor, text, text_len);
+      if (text_len) {
         stb_text_makeundo_insert(state, state->cursor, text_len);
         state->cursor += text_len;
         state->has_preferred_x = 0;
@@ -771,6 +827,7 @@ retry:
   switch (key) {
      default: {
 #ifdef STB_TEXTEDIT_KEYTOTEXT
+         // This is not suitable for UTF-8 support.
         int c = STB_TEXTEDIT_KEYTOTEXT(key);
         if (c > 0) {
            IMSTB_TEXTEDIT_CHARTYPE ch = (IMSTB_TEXTEDIT_CHARTYPE)c;
@@ -911,15 +968,16 @@ retry:

            // [DEAR IMGUI]
            // going down while being on the last line shouldn't bring us to that line end
-            if (STB_TEXTEDIT_GETCHAR(str, find.first_char + find.length - 1) != STB_TEXTEDIT_NEWLINE)
-               break;
+            //if (STB_TEXTEDIT_GETCHAR(str, find.first_char + find.length - 1) != STB_TEXTEDIT_NEWLINE)
+            //   break;

            // now find character position down a row
            state->cursor = start;
            STB_TEXTEDIT_LAYOUTROW(&row, str, state->cursor);
            x = row.x0;
-            for (i=0; i < row.num_chars; ++i) {
+            for (i=0; i < row.num_chars; ) {
               float dx = STB_TEXTEDIT_GETWIDTH(str, start, i);
+               int next = IMSTB_TEXTEDIT_GETNEXTCHARINDEX(str, state->cursor);
               #ifdef IMSTB_TEXTEDIT_GETWIDTH_NEWLINE
               if (dx == IMSTB_TEXTEDIT_GETWIDTH_NEWLINE)
                  break;
@@ -927,10 +985,13 @@ retry:
               x += dx;
               if (x > goal_x)
                  break;
-               state->cursor = IMSTB_TEXTEDIT_GETNEXTCHARINDEX(str, state->cursor);
+               i += next - state->cursor;
+               state->cursor = next;
            }
            stb_textedit_clamp(str, state);

+            if (state->cursor == find.first_char + find.length)
+               str->LastMoveDirectionLR = ImGuiDir_Left;
            state->has_preferred_x = 1;
            state->preferred_x = goal_x;

@@ -980,8 +1041,9 @@ retry:
            state->cursor = find.prev_first;
            STB_TEXTEDIT_LAYOUTROW(&row, str, state->cursor);
            x = row.x0;
-            for (i=0; i < row.num_chars; ++i) {
+            for (i=0; i < row.num_chars; ) {
               float dx = STB_TEXTEDIT_GETWIDTH(str, find.prev_first, i);
+               int next = IMSTB_TEXTEDIT_GETNEXTCHARINDEX(str, state->cursor);
               #ifdef IMSTB_TEXTEDIT_GETWIDTH_NEWLINE
               if (dx == IMSTB_TEXTEDIT_GETWIDTH_NEWLINE)
                  break;
@@ -989,10 +1051,15 @@ retry:
               x += dx;
               if (x > goal_x)
                  break;
-               state->cursor = IMSTB_TEXTEDIT_GETNEXTCHARINDEX(str, state->cursor);
+               i += next - state->cursor;
+               state->cursor = next;
            }
            stb_textedit_clamp(str, state);

+            if (state->cursor == find.first_char)
+               str->LastMoveDirectionLR = ImGuiDir_Right;
+            else if (state->cursor == find.prev_first)
+               str->LastMoveDirectionLR = ImGuiDir_Left;
            state->has_preferred_x = 1;
            state->preferred_x = goal_x;

@@ -1002,10 +1069,15 @@ retry:
            // go to previous line
            // (we need to scan previous line the hard way. maybe we could expose this as a new API function?)
            prev_scan = find.prev_first > 0 ? find.prev_first - 1 : 0;
-            while (prev_scan > 0 && STB_TEXTEDIT_GETCHAR(str, prev_scan - 1) != STB_TEXTEDIT_NEWLINE)
-               --prev_scan;
+            while (prev_scan > 0)
+            {
+               int prev = IMSTB_TEXTEDIT_GETPREVCHARINDEX(str, prev_scan);
+               if (STB_TEXTEDIT_GETCHAR(str, prev) == STB_TEXTEDIT_NEWLINE)
+                  break;
+               prev_scan = prev;
+            }
            find.first_char = find.prev_first;
-            find.prev_first = prev_scan;
+            find.prev_first = STB_TEXTEDIT_MOVELINESTART(str, state, prev_scan);
         }
         break;
      }
@@ -1079,10 +1151,7 @@ retry:
      case STB_TEXTEDIT_K_LINESTART:
         stb_textedit_clamp(str, state);
         stb_textedit_move_to_first(state);
-         if (state->single_line)
-            state->cursor = 0;
-         else while (state->cursor > 0 && STB_TEXTEDIT_GETCHAR(str, state->cursor-1) != STB_TEXTEDIT_NEWLINE)
-            --state->cursor;
+         state->cursor = STB_TEXTEDIT_MOVELINESTART(str, state, state->cursor);
         state->has_preferred_x = 0;
         break;

@@ -1090,13 +1159,9 @@ retry:
      case STB_TEXTEDIT_K_LINEEND2:
 #endif
      case STB_TEXTEDIT_K_LINEEND: {
-         int n = STB_TEXTEDIT_STRINGLEN(str);
         stb_textedit_clamp(str, state);
-         stb_textedit_move_to_first(state);
-         if (state->single_line)
-             state->cursor = n;
-         else while (state->cursor < n && STB_TEXTEDIT_GETCHAR(str, state->cursor) != STB_TEXTEDIT_NEWLINE)
-             ++state->cursor;
+         stb_textedit_move_to_last(str, state);
+         state->cursor = STB_TEXTEDIT_MOVELINEEND(str, state, state->cursor);
         state->has_preferred_x = 0;
         break;
      }
@@ -1107,10 +1172,7 @@ retry:
      case STB_TEXTEDIT_K_LINESTART | STB_TEXTEDIT_K_SHIFT:
         stb_textedit_clamp(str, state);
         stb_textedit_prep_selection_at_cursor(state);
-         if (state->single_line)
-            state->cursor = 0;
-         else while (state->cursor > 0 && STB_TEXTEDIT_GETCHAR(str, state->cursor-1) != STB_TEXTEDIT_NEWLINE)
-            --state->cursor;
+         state->cursor = STB_TEXTEDIT_MOVELINESTART(str, state, state->cursor);
         state->select_end = state->cursor;
         state->has_preferred_x = 0;
         break;
@@ -1119,13 +1181,9 @@ retry:
      case STB_TEXTEDIT_K_LINEEND2 | STB_TEXTEDIT_K_SHIFT:
 #endif
      case STB_TEXTEDIT_K_LINEEND | STB_TEXTEDIT_K_SHIFT: {
-         int n = STB_TEXTEDIT_STRINGLEN(str);
         stb_textedit_clamp(str, state);
         stb_textedit_prep_selection_at_cursor(state);
-         if (state->single_line)
-             state->cursor = n;
-         else while (state->cursor < n && STB_TEXTEDIT_GETCHAR(str, state->cursor) != STB_TEXTEDIT_NEWLINE)
-            ++state->cursor;
+         state->cursor = STB_TEXTEDIT_MOVELINEEND(str, state, state->cursor);
         state->select_end = state->cursor;
         state->has_preferred_x = 0;
         break;
@@ -1300,7 +1358,7 @@ static void stb_text_undo(IMSTB_TEXTEDIT_STRING *str, STB_TexteditState *state)
   // check type of recorded action:
   if (u.insert_length) {
      // easy case: was a deletion, so we need to insert n characters
-      STB_TEXTEDIT_INSERTCHARS(str, u.where, &s->undo_char[u.char_storage], u.insert_length);
+      u.insert_length = STB_TEXTEDIT_INSERTCHARS(str, u.where, &s->undo_char[u.char_storage], u.insert_length);
      s->undo_char_point -= u.insert_length;
   }

@@ -1351,7 +1409,7 @@ static void stb_text_redo(IMSTB_TEXTEDIT_STRING *str, STB_TexteditState *state)

   if (r.insert_length) {
      // easy case: need to insert n characters
-      STB_TEXTEDIT_INSERTCHARS(str, r.where, &s->undo_char[r.char_storage], r.insert_length);
+      r.insert_length = STB_TEXTEDIT_INSERTCHARS(str, r.where, &s->undo_char[r.char_storage], r.insert_length);
      s->redo_char_point += r.insert_length;
   }

@@ -886,7 +886,7 @@ STBTT_DEF unsigned char *stbtt_GetCodepointBitmap(const stbtt_fontinfo *info, fl
 // xoff/yoff are the offset it pixel space from the glyph origin to the top-left of the bitmap

 STBTT_DEF unsigned char *stbtt_GetCodepointBitmapSubpixel(const stbtt_fontinfo *info, float scale_x, float scale_y, float shift_x, float shift_y, int codepoint, int *width, int *height, int *xoff, int *yoff);
-// the same as stbtt_GetCodepoitnBitmap, but you can specify a subpixel
+// the same as stbtt_GetCodepointBitmap, but you can specify a subpixel
 // shift for the character

 STBTT_DEF void stbtt_MakeCodepointBitmap(const stbtt_fontinfo *info, unsigned char *output, int out_w, int out_h, int out_stride, float scale_x, float scale_y, int codepoint);
@@ -4516,8 +4516,8 @@ static int stbtt__compute_crossings_x(float x, float y, int nverts, stbtt_vertex
            q2[0] = (float)x2;
            q2[1] = (float)y2;
            if (equal(q0,q1) || equal(q1,q2)) {
-               x0 = (int)verts[i-1].x;
-               y0 = (int)verts[i-1].y;
+               x0 = (int)verts[i-1].x; //-V1048
+               y0 = (int)verts[i-1].y; //-V1048
               x1 = (int)verts[i  ].x;
               y1 = (int)verts[i  ].y;
               if (y > STBTT_min(y0,y1) && y < STBTT_max(y0,y1) && x > STBTT_min(x0,x1)) {
@@ -0,0 +1,143 @@
+# Minimal build for the vendored libjpeg-turbo.
+# Upstream's CMake pulls in Java bindings, tools, tests and packaging and sets
+# its own global flags, none of which survive contact with this project's four
+# cross toolchains, so only the static library is described here.
+
+set(LJT_VERSION "3.0.4")
+set(VERSION ${LJT_VERSION})
+set(JPEG_LIB_VERSION 62)
+set(LIBJPEG_TURBO_VERSION_NUMBER 3000004)
+set(BUILD "spice2x")
+
+set(C_ARITH_CODING_SUPPORTED 1)
+set(D_ARITH_CODING_SUPPORTED 1)
+
+if(CMAKE_SIZEOF_VOID_P EQUAL 8)
+    set(LJT_ARCH "x86_64")
+else()
+    set(LJT_ARCH "i386")
+endif()
+
+# SIMD is the entire reason for taking this dependency, but a missing assembler
+# must not break the build; the C paths are still correct, just slower.
+# ASM_NASM is enabled by the top level CMakeLists.
+if(CMAKE_ASM_NASM_COMPILER)
+    set(WITH_SIMD 1)
+else()
+    message(WARNING "nasm not found, building libjpeg-turbo without SIMD (screen capture will be slower)")
+endif()
+
+if(MSVC)
+    set(HIDDEN "")
+    set(INLINE "__forceinline")
+    set(THREAD_LOCAL "__declspec(thread)")
+    set(HAVE_INTRIN_H 1)
+else()
+    set(HIDDEN "__attribute__((visibility(\"hidden\")))")
+    set(INLINE "inline __attribute__((always_inline))")
+    set(THREAD_LOCAL "__thread")
+    set(HAVE_BUILTIN_CTZL 1)
+endif()
+
+set(SIZE_T ${CMAKE_SIZEOF_VOID_P})
+
+configure_file(jconfig.h.in ${CMAKE_CURRENT_BINARY_DIR}/jconfig.h)
+configure_file(jconfigint.h.in ${CMAKE_CURRENT_BINARY_DIR}/jconfigint.h)
+configure_file(jversion.h.in ${CMAKE_CURRENT_BINARY_DIR}/jversion.h)
+
+# libjpeg-turbo 3.x supports 8, 12 and 16 bit samples by compiling the
+# precision dependent units once per precision; jcinit.c dispatches to all
+# three at runtime, so none of them can be dropped.
+# jstdhuff.c and the *ext.c units are textually included and never compiled.
+set(JPEG16_SOURCES
+        jcapistd.c jccolor.c jcdiffct.c jclossls.c jcmainct.c jcprepct.c
+        jcsample.c jdapistd.c jdcolor.c jddiffct.c jdlossls.c jdmainct.c
+        jdpostct.c jdsample.c jutils.c)
+
+set(JPEG12_SOURCES ${JPEG16_SOURCES}
+        jccoefct.c jcdctmgr.c jdcoefct.c jddctmgr.c jdmerge.c jfdctfst.c
+        jfdctint.c jidctflt.c jidctfst.c jidctint.c jidctred.c jquant1.c
+        jquant2.c)
+
+set(JPEG_SOURCES ${JPEG12_SOURCES}
+        jcapimin.c jchuff.c jcicc.c jcinit.c jclhuff.c jcmarker.c jcmaster.c
+        jcomapi.c jcparam.c jcphuff.c jctrans.c jdapimin.c jdatadst.c
+        jdatasrc.c jdhuff.c jdicc.c jdinput.c jdlhuff.c jdmarker.c jdmaster.c
+        jdphuff.c jdtrans.c jerror.c jfdctflt.c jmemmgr.c jmemnobs.c
+        jpeg_nbits.c
+        jaricom.c jcarith.c jdarith.c)
+
+if(WITH_SIMD)
+    # the *ext-*.asm units are textually included by these, so the list is
+    # explicit rather than globbed
+    if(LJT_ARCH STREQUAL "x86_64")
+        set(LJT_SIMD_ASM
+                x86_64/jsimdcpu.asm x86_64/jfdctflt-sse.asm
+                x86_64/jccolor-sse2.asm x86_64/jcgray-sse2.asm x86_64/jchuff-sse2.asm
+                x86_64/jcphuff-sse2.asm x86_64/jcsample-sse2.asm x86_64/jdcolor-sse2.asm
+                x86_64/jdmerge-sse2.asm x86_64/jdsample-sse2.asm x86_64/jfdctfst-sse2.asm
+                x86_64/jfdctint-sse2.asm x86_64/jidctflt-sse2.asm x86_64/jidctfst-sse2.asm
+                x86_64/jidctint-sse2.asm x86_64/jidctred-sse2.asm x86_64/jquantf-sse2.asm
+                x86_64/jquanti-sse2.asm
+                x86_64/jccolor-avx2.asm x86_64/jcgray-avx2.asm x86_64/jcsample-avx2.asm
+                x86_64/jdcolor-avx2.asm x86_64/jdmerge-avx2.asm x86_64/jdsample-avx2.asm
+                x86_64/jfdctint-avx2.asm x86_64/jidctint-avx2.asm x86_64/jquanti-avx2.asm)
+    else()
+        set(LJT_SIMD_ASM
+                i386/jsimdcpu.asm i386/jfdctflt-3dn.asm
+                i386/jidctflt-3dn.asm i386/jquant-3dn.asm
+                i386/jccolor-mmx.asm i386/jcgray-mmx.asm i386/jcsample-mmx.asm
+                i386/jdcolor-mmx.asm i386/jdmerge-mmx.asm i386/jdsample-mmx.asm
+                i386/jfdctfst-mmx.asm i386/jfdctint-mmx.asm i386/jidctfst-mmx.asm
+                i386/jidctint-mmx.asm i386/jidctred-mmx.asm i386/jquant-mmx.asm
+                i386/jfdctflt-sse.asm i386/jidctflt-sse.asm i386/jquant-sse.asm
+                i386/jccolor-sse2.asm i386/jcgray-sse2.asm i386/jchuff-sse2.asm
+                i386/jcphuff-sse2.asm i386/jcsample-sse2.asm i386/jdcolor-sse2.asm
+                i386/jdmerge-sse2.asm i386/jdsample-sse2.asm i386/jfdctfst-sse2.asm
+                i386/jfdctint-sse2.asm i386/jidctflt-sse2.asm i386/jidctfst-sse2.asm
+                i386/jidctint-sse2.asm i386/jidctred-sse2.asm i386/jquantf-sse2.asm
+                i386/jquanti-sse2.asm
+                i386/jccolor-avx2.asm i386/jcgray-avx2.asm i386/jcsample-avx2.asm
+                i386/jdcolor-avx2.asm i386/jdmerge-avx2.asm i386/jdsample-avx2.asm
+                i386/jfdctint-avx2.asm i386/jidctint-avx2.asm i386/jquanti-avx2.asm)
+    endif()
+
+    list(TRANSFORM LJT_SIMD_ASM PREPEND "${CMAKE_CURRENT_SOURCE_DIR}/simd/")
+    list(APPEND JPEG_SOURCES simd/${LJT_ARCH}/jsimd.c ${LJT_SIMD_ASM})
+
+    # nasm resolves %include against these, and requires the trailing separator
+    set(CMAKE_ASM_NASM_FLAGS
+            "${CMAKE_ASM_NASM_FLAGS} -I${CMAKE_CURRENT_SOURCE_DIR}/simd/nasm/ -I${CMAKE_CURRENT_SOURCE_DIR}/simd/${LJT_ARCH}/")
+
+    if(LJT_ARCH STREQUAL "x86_64")
+        set(CMAKE_ASM_NASM_OBJECT_FORMAT win64)
+        set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -D__x86_64__ -DWIN64")
+    else()
+        set(CMAKE_ASM_NASM_OBJECT_FORMAT win32)
+        set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DWIN32")
+    endif()
+endif()
+
+add_library(jpeg12-static OBJECT ${JPEG12_SOURCES})
+set_property(TARGET jpeg12-static PROPERTY COMPILE_FLAGS "-DBITS_IN_JSAMPLE=12")
+
+add_library(jpeg16-static OBJECT ${JPEG16_SOURCES})
+set_property(TARGET jpeg16-static PROPERTY COMPILE_FLAGS "-DBITS_IN_JSAMPLE=16")
+
+add_library(jpeg-static STATIC ${JPEG_SOURCES}
+        $<TARGET_OBJECTS:jpeg12-static>
+        $<TARGET_OBJECTS:jpeg16-static>)
+
+foreach(ljt_target jpeg-static jpeg12-static jpeg16-static)
+    target_include_directories(${ljt_target} PUBLIC
+            ${CMAKE_CURRENT_SOURCE_DIR}
+            ${CMAKE_CURRENT_BINARY_DIR})
+
+    target_include_directories(${ljt_target} PRIVATE
+            ${CMAKE_CURRENT_SOURCE_DIR}/simd
+            ${CMAKE_CURRENT_SOURCE_DIR}/simd/${LJT_ARCH})
+
+    if(NOT MSVC)
+        target_compile_options(${ljt_target} PRIVATE -Wno-unused-parameter -Wno-sign-compare)
+    endif()
+endforeach()
@@ -0,0 +1,135 @@
+libjpeg-turbo Licenses
+======================
+
+libjpeg-turbo is covered by two compatible BSD-style open source licenses:
+
+- The IJG (Independent JPEG Group) License, which is listed in
+  [README.ijg](README.ijg)
+
+  This license applies to the libjpeg API library and associated programs,
+  including any code inherited from libjpeg and any modifications to that
+  code.  Note that the libjpeg-turbo SIMD source code bears the
+  [zlib License](https://opensource.org/licenses/Zlib), but in the context of
+  the overall libjpeg API library, the terms of the zlib License are subsumed
+  by the terms of the IJG License.
+
+- The Modified (3-clause) BSD License, which is listed below
+
+  This license applies to the TurboJPEG API library and associated programs, as
+  well as the build system.  Note that the TurboJPEG API library wraps the
+  libjpeg API library, so in the context of the overall TurboJPEG API library,
+  both the terms of the IJG License and the terms of the Modified (3-clause)
+  BSD License apply.
+
+
+Complying with the libjpeg-turbo Licenses
+=========================================
+
+This section provides a roll-up of the libjpeg-turbo licensing terms, to the
+best of our understanding.  This is not a license in and of itself.  It is
+intended solely for clarification.
+
+1.  If you are distributing a modified version of the libjpeg-turbo source,
+    then:
+
+    1.  You cannot alter or remove any existing copyright or license notices
+        from the source.
+
+        **Origin**
+        - Clause 1 of the IJG License
+        - Clause 1 of the Modified BSD License
+        - Clauses 1 and 3 of the zlib License
+
+    2.  You must add your own copyright notice to the header of each source
+        file you modified, so others can tell that you modified that file.  (If
+        there is not an existing copyright header in that file, then you can
+        simply add a notice stating that you modified the file.)
+
+        **Origin**
+        - Clause 1 of the IJG License
+        - Clause 2 of the zlib License
+
+    3.  You must include the IJG README file, and you must not alter any of the
+        copyright or license text in that file.
+
+        **Origin**
+        - Clause 1 of the IJG License
+
+2.  If you are distributing only libjpeg-turbo binaries without the source, or
+    if you are distributing an application that statically links with
+    libjpeg-turbo, then:
+
+    1.  Your product documentation must include a message stating:
+
+        This software is based in part on the work of the Independent JPEG
+        Group.
+
+        **Origin**
+        - Clause 2 of the IJG license
+
+    2.  If your binary distribution includes or uses the TurboJPEG API, then
+        your product documentation must include the text of the Modified BSD
+        License (see below.)
+
+        **Origin**
+        - Clause 2 of the Modified BSD License
+
+3.  You cannot use the name of the IJG or The libjpeg-turbo Project or the
+    contributors thereof in advertising, publicity, etc.
+
+    **Origin**
+    - IJG License
+    - Clause 3 of the Modified BSD License
+
+4.  The IJG and The libjpeg-turbo Project do not warrant libjpeg-turbo to be
+    free of defects, nor do we accept any liability for undesirable
+    consequences resulting from your use of the software.
+
+    **Origin**
+    - IJG License
+    - Modified BSD License
+    - zlib License
+
+
+The Modified (3-clause) BSD License
+===================================
+
+Copyright (C)2009-2023 D. R. Commander.  All Rights Reserved.<br>
+Copyright (C)2015 Viktor Szathmáry.  All Rights Reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+- Redistributions of source code must retain the above copyright notice,
+  this list of conditions and the following disclaimer.
+- Redistributions in binary form must reproduce the above copyright notice,
+  this list of conditions and the following disclaimer in the documentation
+  and/or other materials provided with the distribution.
+- Neither the name of the libjpeg-turbo Project nor the names of its
+  contributors may be used to endorse or promote products derived from this
+  software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS",
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT HOLDERS OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGE.
+
+
+Why Two Licenses?
+=================
+
+The zlib License could have been used instead of the Modified (3-clause) BSD
+License, and since the IJG License effectively subsumes the distribution
+conditions of the zlib License, this would have effectively placed
+libjpeg-turbo binary distributions under the IJG License.  However, the IJG
+License specifically refers to the Independent JPEG Group and does not extend
+attribution and endorsement protections to other entities.  Thus, it was
+desirable to choose a license that granted us the same protections for new code
+that were granted to the IJG for code derived from their software.
@@ -0,0 +1,258 @@
+libjpeg-turbo note:  This file has been modified by The libjpeg-turbo Project
+to include only information relevant to libjpeg-turbo, to wordsmith certain
+sections, and to remove impolitic language that existed in the libjpeg v8
+README.  It is included only for reference.  Please see README.md for
+information specific to libjpeg-turbo.
+
+
+The Independent JPEG Group's JPEG software
+==========================================
+
+This distribution contains a release of the Independent JPEG Group's free JPEG
+software.  You are welcome to redistribute this software and to use it for any
+purpose, subject to the conditions under LEGAL ISSUES, below.
+
+This software is the work of Tom Lane, Guido Vollbeding, Philip Gladstone,
+Bill Allombert, Jim Boucher, Lee Crocker, Bob Friesenhahn, Ben Jackson,
+Julian Minguillon, Luis Ortiz, George Phillips, Davide Rossi, Ge' Weijers,
+and other members of the Independent JPEG Group.
+
+IJG is not affiliated with the ISO/IEC JTC1/SC29/WG1 standards committee
+(also known as JPEG, together with ITU-T SG16).
+
+
+DOCUMENTATION ROADMAP
+=====================
+
+This file contains the following sections:
+
+OVERVIEW            General description of JPEG and the IJG software.
+LEGAL ISSUES        Copyright, lack of warranty, terms of distribution.
+REFERENCES          Where to learn more about JPEG.
+ARCHIVE LOCATIONS   Where to find newer versions of this software.
+FILE FORMAT WARS    Software *not* to get.
+TO DO               Plans for future IJG releases.
+
+Other documentation files in the distribution are:
+
+User documentation:
+  usage.txt         Usage instructions for cjpeg, djpeg, jpegtran,
+                    rdjpgcom, and wrjpgcom.
+  *.1               Unix-style man pages for programs (same info as usage.txt).
+  wizard.txt        Advanced usage instructions for JPEG wizards only.
+  change.log        Version-to-version change highlights.
+Programmer and internal documentation:
+  libjpeg.txt       How to use the JPEG library in your own programs.
+  example.c         Sample code for calling the JPEG library.
+  structure.txt     Overview of the JPEG library's internal structure.
+  coderules.txt     Coding style rules --- please read if you contribute code.
+
+Please read at least usage.txt.  Some information can also be found in the JPEG
+FAQ (Frequently Asked Questions) article.  See ARCHIVE LOCATIONS below to find
+out where to obtain the FAQ article.
+
+If you want to understand how the JPEG code works, we suggest reading one or
+more of the REFERENCES, then looking at the documentation files (in roughly
+the order listed) before diving into the code.
+
+
+OVERVIEW
+========
+
+This package contains C software to implement JPEG image encoding, decoding,
+and transcoding.  JPEG (pronounced "jay-peg") is a standardized compression
+method for full-color and grayscale images.  JPEG's strong suit is compressing
+photographic images or other types of images that have smooth color and
+brightness transitions between neighboring pixels.  Images with sharp lines or
+other abrupt features may not compress well with JPEG, and a higher JPEG
+quality may have to be used to avoid visible compression artifacts with such
+images.
+
+JPEG is normally lossy, meaning that the output pixels are not necessarily
+identical to the input pixels.  However, on photographic content and other
+"smooth" images, very good compression ratios can be obtained with no visible
+compression artifacts, and extremely high compression ratios are possible if
+you are willing to sacrifice image quality (by reducing the "quality" setting
+in the compressor.)
+
+This software implements JPEG baseline, extended-sequential, progressive, and
+lossless compression processes.  Provision is made for supporting all variants
+of these processes, although some uncommon parameter settings aren't
+implemented yet.  We have made no provision for supporting the hierarchical
+processes defined in the standard.
+
+We provide a set of library routines for reading and writing JPEG image files,
+plus two sample applications "cjpeg" and "djpeg", which use the library to
+perform conversion between JPEG and some other popular image file formats.
+The library is intended to be reused in other applications.
+
+In order to support file conversion and viewing software, we have included
+considerable functionality beyond the bare JPEG coding/decoding capability;
+for example, the color quantization modules are not strictly part of JPEG
+decoding, but they are essential for output to colormapped file formats.  These
+extra functions can be compiled out of the library if not required for a
+particular application.
+
+We have also included "jpegtran", a utility for lossless transcoding between
+different JPEG processes, and "rdjpgcom" and "wrjpgcom", two simple
+applications for inserting and extracting textual comments in JFIF files.
+
+The emphasis in designing this software has been on achieving portability and
+flexibility, while also making it fast enough to be useful.  In particular,
+the software is not intended to be read as a tutorial on JPEG.  (See the
+REFERENCES section for introductory material.)  Rather, it is intended to
+be reliable, portable, industrial-strength code.  We do not claim to have
+achieved that goal in every aspect of the software, but we strive for it.
+
+We welcome the use of this software as a component of commercial products.
+No royalty is required, but we do ask for an acknowledgement in product
+documentation, as described under LEGAL ISSUES.
+
+
+LEGAL ISSUES
+============
+
+In plain English:
+
+1. We don't promise that this software works.  (But if you find any bugs,
+   please let us know!)
+2. You can use this software for whatever you want.  You don't have to pay us.
+3. You may not pretend that you wrote this software.  If you use it in a
+   program, you must acknowledge somewhere in your documentation that
+   you've used the IJG code.
+
+In legalese:
+
+The authors make NO WARRANTY or representation, either express or implied,
+with respect to this software, its quality, accuracy, merchantability, or
+fitness for a particular purpose.  This software is provided "AS IS", and you,
+its user, assume the entire risk as to its quality and accuracy.
+
+This software is copyright (C) 1991-2020, Thomas G. Lane, Guido Vollbeding.
+All Rights Reserved except as specified below.
+
+Permission is hereby granted to use, copy, modify, and distribute this
+software (or portions thereof) for any purpose, without fee, subject to these
+conditions:
+(1) If any part of the source code for this software is distributed, then this
+README file must be included, with this copyright and no-warranty notice
+unaltered; and any additions, deletions, or changes to the original files
+must be clearly indicated in accompanying documentation.
+(2) If only executable code is distributed, then the accompanying
+documentation must state that "this software is based in part on the work of
+the Independent JPEG Group".
+(3) Permission for use of this software is granted only if the user accepts
+full responsibility for any undesirable consequences; the authors accept
+NO LIABILITY for damages of any kind.
+
+These conditions apply to any software derived from or based on the IJG code,
+not just to the unmodified library.  If you use our work, you ought to
+acknowledge us.
+
+Permission is NOT granted for the use of any IJG author's name or company name
+in advertising or publicity relating to this software or products derived from
+it.  This software may be referred to only as "the Independent JPEG Group's
+software".
+
+We specifically permit and encourage the use of this software as the basis of
+commercial products, provided that all warranty or liability claims are
+assumed by the product vendor.
+
+
+REFERENCES
+==========
+
+We recommend reading one or more of these references before trying to
+understand the innards of the JPEG software.
+
+The best short technical introduction to the JPEG compression algorithm is
+        Wallace, Gregory K.  "The JPEG Still Picture Compression Standard",
+        Communications of the ACM, April 1991 (vol. 34 no. 4), pp. 30-44.
+(Adjacent articles in that issue discuss MPEG motion picture compression,
+applications of JPEG, and related topics.)  If you don't have the CACM issue
+handy, a PDF file containing a revised version of Wallace's article is
+available at http://www.ijg.org/files/Wallace.JPEG.pdf.  The file (actually
+a preprint for an article that appeared in IEEE Trans. Consumer Electronics)
+omits the sample images that appeared in CACM, but it includes corrections
+and some added material.  Note: the Wallace article is copyright ACM and IEEE,
+and it may not be used for commercial purposes.
+
+A somewhat less technical, more leisurely introduction to JPEG can be found in
+"The Data Compression Book" by Mark Nelson and Jean-loup Gailly, published by
+M&T Books (New York), 2nd ed. 1996, ISBN 1-55851-434-1.  This book provides
+good explanations and example C code for a multitude of compression methods
+including JPEG.  It is an excellent source if you are comfortable reading C
+code but don't know much about data compression in general.  The book's JPEG
+sample code is far from industrial-strength, but when you are ready to look
+at a full implementation, you've got one here...
+
+The best currently available description of JPEG is the textbook "JPEG Still
+Image Data Compression Standard" by William B. Pennebaker and Joan L.
+Mitchell, published by Van Nostrand Reinhold, 1993, ISBN 0-442-01272-1.
+Price US$59.95, 638 pp.  The book includes the complete text of the ISO JPEG
+standards (DIS 10918-1 and draft DIS 10918-2).
+
+The original JPEG standard is divided into two parts, Part 1 being the actual
+specification, while Part 2 covers compliance testing methods.  Part 1 is
+titled "Digital Compression and Coding of Continuous-tone Still Images,
+Part 1: Requirements and guidelines" and has document numbers ISO/IEC IS
+10918-1, ITU-T T.81.  Part 2 is titled "Digital Compression and Coding of
+Continuous-tone Still Images, Part 2: Compliance testing" and has document
+numbers ISO/IEC IS 10918-2, ITU-T T.83.
+
+The JPEG standard does not specify all details of an interchangeable file
+format.  For the omitted details, we follow the "JFIF" conventions, revision
+1.02.  JFIF version 1 has been adopted as ISO/IEC 10918-5 (05/2013) and
+Recommendation ITU-T T.871 (05/2011): Information technology - Digital
+compression and coding of continuous-tone still images: JPEG File Interchange
+Format (JFIF).  It is available as a free download in PDF file format from
+https://www.iso.org/standard/54989.html and http://www.itu.int/rec/T-REC-T.871.
+A PDF file of the older JFIF 1.02 specification is available at
+http://www.w3.org/Graphics/JPEG/jfif3.pdf.
+
+The TIFF 6.0 file format specification can be obtained from
+http://mirrors.ctan.org/graphics/tiff/TIFF6.ps.gz.  The JPEG incorporation
+scheme found in the TIFF 6.0 spec of 3-June-92 has a number of serious
+problems.  IJG does not recommend use of the TIFF 6.0 design (TIFF Compression
+tag 6).  Instead, we recommend the JPEG design proposed by TIFF Technical Note
+#2 (Compression tag 7).  Copies of this Note can be obtained from
+http://www.ijg.org/files/.  It is expected that the next revision
+of the TIFF spec will replace the 6.0 JPEG design with the Note's design.
+Although IJG's own code does not support TIFF/JPEG, the free libtiff library
+uses our library to implement TIFF/JPEG per the Note.
+
+
+ARCHIVE LOCATIONS
+=================
+
+The "official" archive site for this software is www.ijg.org.
+The most recent released version can always be found there in
+directory "files".
+
+The JPEG FAQ (Frequently Asked Questions) article is a source of some
+general information about JPEG.  It is available at
+http://www.faqs.org/faqs/jpeg-faq.
+
+
+FILE FORMAT COMPATIBILITY
+=========================
+
+This software implements ITU T.81 | ISO/IEC 10918 with some extensions from
+ITU T.871 | ISO/IEC 10918-5 (JPEG File Interchange Format-- see REFERENCES).
+Informally, the term "JPEG image" or "JPEG file" most often refers to JFIF or
+a subset thereof, but there are other formats containing the name "JPEG" that
+are incompatible with the original JPEG standard or with JFIF (for instance,
+JPEG 2000 and JPEG XR).  This software therefore does not support these
+formats.  Indeed, one of the original reasons for developing this free software
+was to help force convergence on a common, interoperable format standard for
+JPEG files.
+
+JFIF is a minimal or "low end" representation.  TIFF/JPEG (TIFF revision 6.0 as
+modified by TIFF Technical Note #2) can be used for "high end" applications
+that need to record a lot of additional data about an image.
+
+
+TO DO
+=====
+
+Please send bug reports, offers of help, etc. to jpeg-info@jpegclub.org.
@@ -0,0 +1,157 @@
+/*
+ * jaricom.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Developed 1997-2009 by Guido Vollbeding.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2015, 2018, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains probability estimation tables for common use in
+ * arithmetic entropy encoding and decoding routines.
+ *
+ * This data represents Table D.2 in
+ * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994 and Table 24 in
+ * Recommendation ITU-T T.82 (1993) | ISO/IEC 11544:1993.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+/* The following #define specifies the packing of the four components
+ * into the compact JLONG representation.
+ * Note that this formula must match the actual arithmetic encoder
+ * and decoder implementation.  The implementation has to be changed
+ * if this formula is changed.
+ * The current organization is leaned on Markus Kuhn's JBIG
+ * implementation (jbig_tab.c).
+ */
+
+#define V(i, a, b, c, d) \
+  (((JLONG)a << 16) | ((JLONG)c << 8) | ((JLONG)d << 7) | b)
+
+const JLONG jpeg_aritab[113 + 1] = {
+/*
+ * Index, Qe_Value, Next_Index_LPS, Next_Index_MPS, Switch_MPS
+ */
+  V(   0, 0x5a1d,   1,   1, 1 ),
+  V(   1, 0x2586,  14,   2, 0 ),
+  V(   2, 0x1114,  16,   3, 0 ),
+  V(   3, 0x080b,  18,   4, 0 ),
+  V(   4, 0x03d8,  20,   5, 0 ),
+  V(   5, 0x01da,  23,   6, 0 ),
+  V(   6, 0x00e5,  25,   7, 0 ),
+  V(   7, 0x006f,  28,   8, 0 ),
+  V(   8, 0x0036,  30,   9, 0 ),
+  V(   9, 0x001a,  33,  10, 0 ),
+  V(  10, 0x000d,  35,  11, 0 ),
+  V(  11, 0x0006,   9,  12, 0 ),
+  V(  12, 0x0003,  10,  13, 0 ),
+  V(  13, 0x0001,  12,  13, 0 ),
+  V(  14, 0x5a7f,  15,  15, 1 ),
+  V(  15, 0x3f25,  36,  16, 0 ),
+  V(  16, 0x2cf2,  38,  17, 0 ),
+  V(  17, 0x207c,  39,  18, 0 ),
+  V(  18, 0x17b9,  40,  19, 0 ),
+  V(  19, 0x1182,  42,  20, 0 ),
+  V(  20, 0x0cef,  43,  21, 0 ),
+  V(  21, 0x09a1,  45,  22, 0 ),
+  V(  22, 0x072f,  46,  23, 0 ),
+  V(  23, 0x055c,  48,  24, 0 ),
+  V(  24, 0x0406,  49,  25, 0 ),
+  V(  25, 0x0303,  51,  26, 0 ),
+  V(  26, 0x0240,  52,  27, 0 ),
+  V(  27, 0x01b1,  54,  28, 0 ),
+  V(  28, 0x0144,  56,  29, 0 ),
+  V(  29, 0x00f5,  57,  30, 0 ),
+  V(  30, 0x00b7,  59,  31, 0 ),
+  V(  31, 0x008a,  60,  32, 0 ),
+  V(  32, 0x0068,  62,  33, 0 ),
+  V(  33, 0x004e,  63,  34, 0 ),
+  V(  34, 0x003b,  32,  35, 0 ),
+  V(  35, 0x002c,  33,   9, 0 ),
+  V(  36, 0x5ae1,  37,  37, 1 ),
+  V(  37, 0x484c,  64,  38, 0 ),
+  V(  38, 0x3a0d,  65,  39, 0 ),
+  V(  39, 0x2ef1,  67,  40, 0 ),
+  V(  40, 0x261f,  68,  41, 0 ),
+  V(  41, 0x1f33,  69,  42, 0 ),
+  V(  42, 0x19a8,  70,  43, 0 ),
+  V(  43, 0x1518,  72,  44, 0 ),
+  V(  44, 0x1177,  73,  45, 0 ),
+  V(  45, 0x0e74,  74,  46, 0 ),
+  V(  46, 0x0bfb,  75,  47, 0 ),
+  V(  47, 0x09f8,  77,  48, 0 ),
+  V(  48, 0x0861,  78,  49, 0 ),
+  V(  49, 0x0706,  79,  50, 0 ),
+  V(  50, 0x05cd,  48,  51, 0 ),
+  V(  51, 0x04de,  50,  52, 0 ),
+  V(  52, 0x040f,  50,  53, 0 ),
+  V(  53, 0x0363,  51,  54, 0 ),
+  V(  54, 0x02d4,  52,  55, 0 ),
+  V(  55, 0x025c,  53,  56, 0 ),
+  V(  56, 0x01f8,  54,  57, 0 ),
+  V(  57, 0x01a4,  55,  58, 0 ),
+  V(  58, 0x0160,  56,  59, 0 ),
+  V(  59, 0x0125,  57,  60, 0 ),
+  V(  60, 0x00f6,  58,  61, 0 ),
+  V(  61, 0x00cb,  59,  62, 0 ),
+  V(  62, 0x00ab,  61,  63, 0 ),
+  V(  63, 0x008f,  61,  32, 0 ),
+  V(  64, 0x5b12,  65,  65, 1 ),
+  V(  65, 0x4d04,  80,  66, 0 ),
+  V(  66, 0x412c,  81,  67, 0 ),
+  V(  67, 0x37d8,  82,  68, 0 ),
+  V(  68, 0x2fe8,  83,  69, 0 ),
+  V(  69, 0x293c,  84,  70, 0 ),
+  V(  70, 0x2379,  86,  71, 0 ),
+  V(  71, 0x1edf,  87,  72, 0 ),
+  V(  72, 0x1aa9,  87,  73, 0 ),
+  V(  73, 0x174e,  72,  74, 0 ),
+  V(  74, 0x1424,  72,  75, 0 ),
+  V(  75, 0x119c,  74,  76, 0 ),
+  V(  76, 0x0f6b,  74,  77, 0 ),
+  V(  77, 0x0d51,  75,  78, 0 ),
+  V(  78, 0x0bb6,  77,  79, 0 ),
+  V(  79, 0x0a40,  77,  48, 0 ),
+  V(  80, 0x5832,  80,  81, 1 ),
+  V(  81, 0x4d1c,  88,  82, 0 ),
+  V(  82, 0x438e,  89,  83, 0 ),
+  V(  83, 0x3bdd,  90,  84, 0 ),
+  V(  84, 0x34ee,  91,  85, 0 ),
+  V(  85, 0x2eae,  92,  86, 0 ),
+  V(  86, 0x299a,  93,  87, 0 ),
+  V(  87, 0x2516,  86,  71, 0 ),
+  V(  88, 0x5570,  88,  89, 1 ),
+  V(  89, 0x4ca9,  95,  90, 0 ),
+  V(  90, 0x44d9,  96,  91, 0 ),
+  V(  91, 0x3e22,  97,  92, 0 ),
+  V(  92, 0x3824,  99,  93, 0 ),
+  V(  93, 0x32b4,  99,  94, 0 ),
+  V(  94, 0x2e17,  93,  86, 0 ),
+  V(  95, 0x56a8,  95,  96, 1 ),
+  V(  96, 0x4f46, 101,  97, 0 ),
+  V(  97, 0x47e5, 102,  98, 0 ),
+  V(  98, 0x41cf, 103,  99, 0 ),
+  V(  99, 0x3c3d, 104, 100, 0 ),
+  V( 100, 0x375e,  99,  93, 0 ),
+  V( 101, 0x5231, 105, 102, 0 ),
+  V( 102, 0x4c0f, 106, 103, 0 ),
+  V( 103, 0x4639, 107, 104, 0 ),
+  V( 104, 0x415e, 103,  99, 0 ),
+  V( 105, 0x5627, 105, 106, 1 ),
+  V( 106, 0x50e7, 108, 107, 0 ),
+  V( 107, 0x4b85, 109, 103, 0 ),
+  V( 108, 0x5597, 110, 109, 0 ),
+  V( 109, 0x504f, 111, 107, 0 ),
+  V( 110, 0x5a10, 110, 111, 1 ),
+  V( 111, 0x5522, 112, 109, 0 ),
+  V( 112, 0x59eb, 112, 111, 1 ),
+/*
+ * This last entry is used for fixed probability estimate of 0.5
+ * as recommended in Section 10.3 Table 5 of ITU-T Rec. T.851.
+ */
+  V( 113, 0x5a1d, 113, 113, 0 )
+};
@@ -0,0 +1,318 @@
+/*
+ * jcapimin.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1994-1998, Thomas G. Lane.
+ * Modified 2003-2010 by Guido Vollbeding.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2022, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains application interface code for the compression half
+ * of the JPEG library.  These are the "minimum" API routines that may be
+ * needed in either the normal full-compression case or the transcoding-only
+ * case.
+ *
+ * Most of the routines intended to be called directly by an application
+ * are in this file or in jcapistd.c.  But also see jcparam.c for
+ * parameter-setup helper routines, jcomapi.c for routines shared by
+ * compression and decompression, and jctrans.c for the transcoding case.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jcmaster.h"
+
+
+/*
+ * Initialization of a JPEG compression object.
+ * The error manager must already be set up (in case memory manager fails).
+ */
+
+GLOBAL(void)
+jpeg_CreateCompress(j_compress_ptr cinfo, int version, size_t structsize)
+{
+  int i;
+
+  /* Guard against version mismatches between library and caller. */
+  cinfo->mem = NULL;            /* so jpeg_destroy knows mem mgr not called */
+  if (version != JPEG_LIB_VERSION)
+    ERREXIT2(cinfo, JERR_BAD_LIB_VERSION, JPEG_LIB_VERSION, version);
+  if (structsize != sizeof(struct jpeg_compress_struct))
+    ERREXIT2(cinfo, JERR_BAD_STRUCT_SIZE,
+             (int)sizeof(struct jpeg_compress_struct), (int)structsize);
+
+  /* For debugging purposes, we zero the whole master structure.
+   * But the application has already set the err pointer, and may have set
+   * client_data, so we have to save and restore those fields.
+   * Note: if application hasn't set client_data, tools like Purify may
+   * complain here.
+   */
+  {
+    struct jpeg_error_mgr *err = cinfo->err;
+    void *client_data = cinfo->client_data; /* ignore Purify complaint here */
+    memset(cinfo, 0, sizeof(struct jpeg_compress_struct));
+    cinfo->err = err;
+    cinfo->client_data = client_data;
+  }
+  cinfo->is_decompressor = FALSE;
+
+  /* Initialize a memory manager instance for this object */
+  jinit_memory_mgr((j_common_ptr)cinfo);
+
+  /* Zero out pointers to permanent structures. */
+  cinfo->progress = NULL;
+  cinfo->dest = NULL;
+
+  cinfo->comp_info = NULL;
+
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    cinfo->quant_tbl_ptrs[i] = NULL;
+#if JPEG_LIB_VERSION >= 70
+    cinfo->q_scale_factor[i] = 100;
+#endif
+  }
+
+  for (i = 0; i < NUM_HUFF_TBLS; i++) {
+    cinfo->dc_huff_tbl_ptrs[i] = NULL;
+    cinfo->ac_huff_tbl_ptrs[i] = NULL;
+  }
+
+#if JPEG_LIB_VERSION >= 80
+  /* Must do it here for emit_dqt in case jpeg_write_tables is used */
+  cinfo->block_size = DCTSIZE;
+  cinfo->natural_order = jpeg_natural_order;
+  cinfo->lim_Se = DCTSIZE2 - 1;
+#endif
+
+  cinfo->script_space = NULL;
+
+  cinfo->input_gamma = 1.0;     /* in case application forgets */
+
+  cinfo->data_precision = BITS_IN_JSAMPLE;
+
+  /* OK, I'm ready */
+  cinfo->global_state = CSTATE_START;
+
+  /* The master struct is used to store extension parameters, so we allocate it
+   * here.
+   */
+  cinfo->master = (struct jpeg_comp_master *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT,
+                                  sizeof(my_comp_master));
+  memset(cinfo->master, 0, sizeof(my_comp_master));
+}
+
+
+/*
+ * Destruction of a JPEG compression object
+ */
+
+GLOBAL(void)
+jpeg_destroy_compress(j_compress_ptr cinfo)
+{
+  jpeg_destroy((j_common_ptr)cinfo); /* use common routine */
+}
+
+
+/*
+ * Abort processing of a JPEG compression operation,
+ * but don't destroy the object itself.
+ */
+
+GLOBAL(void)
+jpeg_abort_compress(j_compress_ptr cinfo)
+{
+  jpeg_abort((j_common_ptr)cinfo); /* use common routine */
+}
+
+
+/*
+ * Forcibly suppress or un-suppress all quantization and Huffman tables.
+ * Marks all currently defined tables as already written (if suppress)
+ * or not written (if !suppress).  This will control whether they get emitted
+ * by a subsequent jpeg_start_compress call.
+ *
+ * This routine is exported for use by applications that want to produce
+ * abbreviated JPEG datastreams.  It logically belongs in jcparam.c, but
+ * since it is called by jpeg_start_compress, we put it here --- otherwise
+ * jcparam.o would be linked whether the application used it or not.
+ */
+
+GLOBAL(void)
+jpeg_suppress_tables(j_compress_ptr cinfo, boolean suppress)
+{
+  int i;
+  JQUANT_TBL *qtbl;
+  JHUFF_TBL *htbl;
+
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    if ((qtbl = cinfo->quant_tbl_ptrs[i]) != NULL)
+      qtbl->sent_table = suppress;
+  }
+
+  for (i = 0; i < NUM_HUFF_TBLS; i++) {
+    if ((htbl = cinfo->dc_huff_tbl_ptrs[i]) != NULL)
+      htbl->sent_table = suppress;
+    if ((htbl = cinfo->ac_huff_tbl_ptrs[i]) != NULL)
+      htbl->sent_table = suppress;
+  }
+}
+
+
+/*
+ * Finish JPEG compression.
+ *
+ * If a multipass operating mode was selected, this may do a great deal of
+ * work including most of the actual output.
+ */
+
+GLOBAL(void)
+jpeg_finish_compress(j_compress_ptr cinfo)
+{
+  JDIMENSION iMCU_row;
+
+  if (cinfo->global_state == CSTATE_SCANNING ||
+      cinfo->global_state == CSTATE_RAW_OK) {
+    /* Terminate first pass */
+    if (cinfo->next_scanline < cinfo->image_height)
+      ERREXIT(cinfo, JERR_TOO_LITTLE_DATA);
+    (*cinfo->master->finish_pass) (cinfo);
+  } else if (cinfo->global_state != CSTATE_WRCOEFS)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  /* Perform any remaining passes */
+  while (!cinfo->master->is_last_pass) {
+    (*cinfo->master->prepare_for_pass) (cinfo);
+    for (iMCU_row = 0; iMCU_row < cinfo->total_iMCU_rows; iMCU_row++) {
+      if (cinfo->progress != NULL) {
+        cinfo->progress->pass_counter = (long)iMCU_row;
+        cinfo->progress->pass_limit = (long)cinfo->total_iMCU_rows;
+        (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo);
+      }
+      /* We bypass the main controller and invoke coef controller directly;
+       * all work is being done from the coefficient buffer.
+       */
+      if (cinfo->data_precision == 16) {
+#ifdef C_LOSSLESS_SUPPORTED
+        if (!(*cinfo->coef->compress_data_16) (cinfo, (J16SAMPIMAGE)NULL))
+          ERREXIT(cinfo, JERR_CANT_SUSPEND);
+#else
+        ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+#endif
+      } else if (cinfo->data_precision == 12) {
+        if (!(*cinfo->coef->compress_data_12) (cinfo, (J12SAMPIMAGE)NULL))
+          ERREXIT(cinfo, JERR_CANT_SUSPEND);
+      } else {
+        if (!(*cinfo->coef->compress_data) (cinfo, (JSAMPIMAGE)NULL))
+          ERREXIT(cinfo, JERR_CANT_SUSPEND);
+      }
+    }
+    (*cinfo->master->finish_pass) (cinfo);
+  }
+  /* Write EOI, do final cleanup */
+  (*cinfo->marker->write_file_trailer) (cinfo);
+  (*cinfo->dest->term_destination) (cinfo);
+  /* We can use jpeg_abort to release memory and reset global_state */
+  jpeg_abort((j_common_ptr)cinfo);
+}
+
+
+/*
+ * Write a special marker.
+ * This is only recommended for writing COM or APPn markers.
+ * Must be called after jpeg_start_compress() and before
+ * first call to jpeg_write_scanlines() or jpeg_write_raw_data().
+ */
+
+GLOBAL(void)
+jpeg_write_marker(j_compress_ptr cinfo, int marker, const JOCTET *dataptr,
+                  unsigned int datalen)
+{
+  void (*write_marker_byte) (j_compress_ptr info, int val);
+
+  if (cinfo->next_scanline != 0 ||
+      (cinfo->global_state != CSTATE_SCANNING &&
+       cinfo->global_state != CSTATE_RAW_OK &&
+       cinfo->global_state != CSTATE_WRCOEFS))
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  (*cinfo->marker->write_marker_header) (cinfo, marker, datalen);
+  write_marker_byte = cinfo->marker->write_marker_byte; /* copy for speed */
+  while (datalen--) {
+    (*write_marker_byte) (cinfo, *dataptr);
+    dataptr++;
+  }
+}
+
+/* Same, but piecemeal. */
+
+GLOBAL(void)
+jpeg_write_m_header(j_compress_ptr cinfo, int marker, unsigned int datalen)
+{
+  if (cinfo->next_scanline != 0 ||
+      (cinfo->global_state != CSTATE_SCANNING &&
+       cinfo->global_state != CSTATE_RAW_OK &&
+       cinfo->global_state != CSTATE_WRCOEFS))
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  (*cinfo->marker->write_marker_header) (cinfo, marker, datalen);
+}
+
+GLOBAL(void)
+jpeg_write_m_byte(j_compress_ptr cinfo, int val)
+{
+  (*cinfo->marker->write_marker_byte) (cinfo, val);
+}
+
+
+/*
+ * Alternate compression function: just write an abbreviated table file.
+ * Before calling this, all parameters and a data destination must be set up.
+ *
+ * To produce a pair of files containing abbreviated tables and abbreviated
+ * image data, one would proceed as follows:
+ *
+ *              initialize JPEG object
+ *              set JPEG parameters
+ *              set destination to table file
+ *              jpeg_write_tables(cinfo);
+ *              set destination to image file
+ *              jpeg_start_compress(cinfo, FALSE);
+ *              write data...
+ *              jpeg_finish_compress(cinfo);
+ *
+ * jpeg_write_tables has the side effect of marking all tables written
+ * (same as jpeg_suppress_tables(..., TRUE)).  Thus a subsequent start_compress
+ * will not re-emit the tables unless it is passed write_all_tables=TRUE.
+ */
+
+GLOBAL(void)
+jpeg_write_tables(j_compress_ptr cinfo)
+{
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  /* (Re)initialize error mgr and destination modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr)cinfo);
+  (*cinfo->dest->init_destination) (cinfo);
+  /* Initialize the marker writer ... bit of a crock to do it here. */
+  jinit_marker_writer(cinfo);
+  /* Write them tables! */
+  (*cinfo->marker->write_tables_only) (cinfo);
+  /* And clean up. */
+  (*cinfo->dest->term_destination) (cinfo);
+  /*
+   * In library releases up through v6a, we called jpeg_abort() here to free
+   * any working memory allocated by the destination manager and marker
+   * writer.  Some applications had a problem with that: they allocated space
+   * of their own from the library memory manager, and didn't want it to go
+   * away during write_tables.  So now we do nothing.  This will cause a
+   * memory leak if an app calls write_tables repeatedly without doing a full
+   * compression cycle or otherwise resetting the JPEG object.  However, that
+   * seems less bad than unexpectedly freeing memory in the normal case.
+   * An app that prefers the old behavior can call jpeg_abort for itself after
+   * each call to jpeg_write_tables().
+   */
+}
@@ -0,0 +1,187 @@
+/*
+ * jcapistd.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2022, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains application interface code for the compression half
+ * of the JPEG library.  These are the "standard" API routines that are
+ * used in the normal full-compression case.  They are not used by a
+ * transcoding-only application.  Note that if an application links in
+ * jpeg_start_compress, it will end up linking in the entire compressor.
+ * We thus must separate this file from jcapimin.c to avoid linking the
+ * whole compression library into a transcoder.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jsamplecomp.h"
+
+
+#if BITS_IN_JSAMPLE == 8
+
+/*
+ * Compression initialization.
+ * Before calling this, all parameters and a data destination must be set up.
+ *
+ * We require a write_all_tables parameter as a failsafe check when writing
+ * multiple datastreams from the same compression object.  Since prior runs
+ * will have left all the tables marked sent_table=TRUE, a subsequent run
+ * would emit an abbreviated stream (no tables) by default.  This may be what
+ * is wanted, but for safety's sake it should not be the default behavior:
+ * programmers should have to make a deliberate choice to emit abbreviated
+ * images.  Therefore the documentation and examples should encourage people
+ * to pass write_all_tables=TRUE; then it will take active thought to do the
+ * wrong thing.
+ */
+
+GLOBAL(void)
+jpeg_start_compress(j_compress_ptr cinfo, boolean write_all_tables)
+{
+  if (cinfo->global_state != CSTATE_START)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+
+  if (write_all_tables)
+    jpeg_suppress_tables(cinfo, FALSE); /* mark all tables to be written */
+
+  /* (Re)initialize error mgr and destination modules */
+  (*cinfo->err->reset_error_mgr) ((j_common_ptr)cinfo);
+  (*cinfo->dest->init_destination) (cinfo);
+  /* Perform master selection of active modules */
+  jinit_compress_master(cinfo);
+  /* Set up for the first pass */
+  (*cinfo->master->prepare_for_pass) (cinfo);
+  /* Ready for application to drive first pass through _jpeg_write_scanlines
+   * or _jpeg_write_raw_data.
+   */
+  cinfo->next_scanline = 0;
+  cinfo->global_state = (cinfo->raw_data_in ? CSTATE_RAW_OK : CSTATE_SCANNING);
+}
+
+#endif
+
+
+/*
+ * Write some scanlines of data to the JPEG compressor.
+ *
+ * The return value will be the number of lines actually written.
+ * This should be less than the supplied num_lines only in case that
+ * the data destination module has requested suspension of the compressor,
+ * or if more than image_height scanlines are passed in.
+ *
+ * Note: we warn about excess calls to _jpeg_write_scanlines() since
+ * this likely signals an application programmer error.  However,
+ * excess scanlines passed in the last valid call are *silently* ignored,
+ * so that the application need not adjust num_lines for end-of-image
+ * when using a multiple-scanline buffer.
+ */
+
+GLOBAL(JDIMENSION)
+_jpeg_write_scanlines(j_compress_ptr cinfo, _JSAMPARRAY scanlines,
+                      JDIMENSION num_lines)
+{
+#if BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED)
+  JDIMENSION row_ctr, rows_left;
+
+  if (cinfo->data_precision != BITS_IN_JSAMPLE)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  if (cinfo->global_state != CSTATE_SCANNING)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  if (cinfo->next_scanline >= cinfo->image_height)
+    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
+
+  /* Call progress monitor hook if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->pass_counter = (long)cinfo->next_scanline;
+    cinfo->progress->pass_limit = (long)cinfo->image_height;
+    (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo);
+  }
+
+  /* Give master control module another chance if this is first call to
+   * _jpeg_write_scanlines.  This lets output of the frame/scan headers be
+   * delayed so that application can write COM, etc, markers between
+   * jpeg_start_compress and _jpeg_write_scanlines.
+   */
+  if (cinfo->master->call_pass_startup)
+    (*cinfo->master->pass_startup) (cinfo);
+
+  /* Ignore any extra scanlines at bottom of image. */
+  rows_left = cinfo->image_height - cinfo->next_scanline;
+  if (num_lines > rows_left)
+    num_lines = rows_left;
+
+  row_ctr = 0;
+  (*cinfo->main->_process_data) (cinfo, scanlines, &row_ctr, num_lines);
+  cinfo->next_scanline += row_ctr;
+  return row_ctr;
+#else
+  ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+  return 0;
+#endif
+}
+
+
+#if BITS_IN_JSAMPLE != 16
+
+/*
+ * Alternate entry point to write raw data.
+ * Processes exactly one iMCU row per call, unless suspended.
+ */
+
+GLOBAL(JDIMENSION)
+_jpeg_write_raw_data(j_compress_ptr cinfo, _JSAMPIMAGE data,
+                     JDIMENSION num_lines)
+{
+  JDIMENSION lines_per_iMCU_row;
+
+  if (cinfo->data_precision != BITS_IN_JSAMPLE)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  if (cinfo->master->lossless)
+    ERREXIT(cinfo, JERR_NOTIMPL);
+
+  if (cinfo->global_state != CSTATE_RAW_OK)
+    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
+  if (cinfo->next_scanline >= cinfo->image_height) {
+    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
+    return 0;
+  }
+
+  /* Call progress monitor hook if present */
+  if (cinfo->progress != NULL) {
+    cinfo->progress->pass_counter = (long)cinfo->next_scanline;
+    cinfo->progress->pass_limit = (long)cinfo->image_height;
+    (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo);
+  }
+
+  /* Give master control module another chance if this is first call to
+   * _jpeg_write_raw_data.  This lets output of the frame/scan headers be
+   * delayed so that application can write COM, etc, markers between
+   * jpeg_start_compress and _jpeg_write_raw_data.
+   */
+  if (cinfo->master->call_pass_startup)
+    (*cinfo->master->pass_startup) (cinfo);
+
+  /* Verify that at least one iMCU row has been passed. */
+  lines_per_iMCU_row = cinfo->max_v_samp_factor * DCTSIZE;
+  if (num_lines < lines_per_iMCU_row)
+    ERREXIT(cinfo, JERR_BUFFER_SIZE);
+
+  /* Directly compress the row. */
+  if (!(*cinfo->coef->_compress_data) (cinfo, data)) {
+    /* If compressor did not consume the whole row, suspend processing. */
+    return 0;
+  }
+
+  /* OK, we processed one iMCU row. */
+  cinfo->next_scanline += lines_per_iMCU_row;
+  return lines_per_iMCU_row;
+}
+
+#endif /* BITS_IN_JSAMPLE != 16 */
@@ -0,0 +1,932 @@
+/*
+ * jcarith.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Developed 1997-2009 by Guido Vollbeding.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2015, 2018, 2021-2022, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains portable arithmetic entropy encoding routines for JPEG
+ * (implementing Recommendation ITU-T T.81 | ISO/IEC 10918-1).
+ *
+ * Both sequential and progressive modes are supported in this single module.
+ *
+ * Suspension is not currently supported in this module.
+ *
+ * NOTE: All referenced figures are from
+ * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+
+
+/* Expanded entropy encoder object for arithmetic encoding. */
+
+typedef struct {
+  struct jpeg_entropy_encoder pub; /* public fields */
+
+  JLONG c; /* C register, base of coding interval, layout as in sec. D.1.3 */
+  JLONG a;               /* A register, normalized size of coding interval */
+  JLONG sc;        /* counter for stacked 0xFF values which might overflow */
+  JLONG zc;          /* counter for pending 0x00 output values which might *
+                          * be discarded at the end ("Pacman" termination) */
+  int ct;  /* bit shift counter, determines when next byte will be written */
+  int buffer;                /* buffer for most recent output byte != 0xFF */
+
+  int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */
+  int dc_context[MAX_COMPS_IN_SCAN]; /* context index for DC conditioning */
+
+  unsigned int restarts_to_go;  /* MCUs left in this restart interval */
+  int next_restart_num;         /* next restart number to write (0-7) */
+
+  /* Pointers to statistics areas (these workspaces have image lifespan) */
+  unsigned char *dc_stats[NUM_ARITH_TBLS];
+  unsigned char *ac_stats[NUM_ARITH_TBLS];
+
+  /* Statistics bin for coding with fixed probability 0.5 */
+  unsigned char fixed_bin[4];
+} arith_entropy_encoder;
+
+typedef arith_entropy_encoder *arith_entropy_ptr;
+
+/* The following two definitions specify the allocation chunk size
+ * for the statistics area.
+ * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least
+ * 49 statistics bins for DC, and 245 statistics bins for AC coding.
+ *
+ * We use a compact representation with 1 byte per statistics bin,
+ * thus the numbers directly represent byte sizes.
+ * This 1 byte per statistics bin contains the meaning of the MPS
+ * (more probable symbol) in the highest bit (mask 0x80), and the
+ * index into the probability estimation state machine table
+ * in the lower bits (mask 0x7F).
+ */
+
+#define DC_STAT_BINS  64
+#define AC_STAT_BINS  256
+
+/* NOTE: Uncomment the following #define if you want to use the
+ * given formula for calculating the AC conditioning parameter Kx
+ * for spectral selection progressive coding in section G.1.3.2
+ * of the spec (Kx = Kmin + SRL (8 + Se - Kmin) 4).
+ * Although the spec and P&M authors claim that this "has proven
+ * to give good results for 8 bit precision samples", I'm not
+ * convinced yet that this is really beneficial.
+ * Early tests gave only very marginal compression enhancements
+ * (a few - around 5 or so - bytes even for very large files),
+ * which would turn out rather negative if we'd suppress the
+ * DAC (Define Arithmetic Conditioning) marker segments for
+ * the default parameters in the future.
+ * Note that currently the marker writing module emits 12-byte
+ * DAC segments for a full-component scan in a color image.
+ * This is not worth worrying about IMHO. However, since the
+ * spec defines the default values to be used if the tables
+ * are omitted (unlike Huffman tables, which are required
+ * anyway), one might optimize this behaviour in the future,
+ * and then it would be disadvantageous to use custom tables if
+ * they don't provide sufficient gain to exceed the DAC size.
+ *
+ * On the other hand, I'd consider it as a reasonable result
+ * that the conditioning has no significant influence on the
+ * compression performance. This means that the basic
+ * statistical model is already rather stable.
+ *
+ * Thus, at the moment, we use the default conditioning values
+ * anyway, and do not use the custom formula.
+ *
+#define CALCULATE_SPECTRAL_CONDITIONING
+ */
+
+/* IRIGHT_SHIFT is like RIGHT_SHIFT, but works on int rather than JLONG.
+ * We assume that int right shift is unsigned if JLONG right shift is,
+ * which should be safe.
+ */
+
+#ifdef RIGHT_SHIFT_IS_UNSIGNED
+#define ISHIFT_TEMPS    int ishift_temp;
+#define IRIGHT_SHIFT(x, shft) \
+  ((ishift_temp = (x)) < 0 ? \
+   (ishift_temp >> (shft)) | ((~0) << (16 - (shft))) : \
+   (ishift_temp >> (shft)))
+#else
+#define ISHIFT_TEMPS
+#define IRIGHT_SHIFT(x, shft)   ((x) >> (shft))
+#endif
+
+
+LOCAL(void)
+emit_byte(int val, j_compress_ptr cinfo)
+/* Write next output byte; we do not support suspension in this module. */
+{
+  struct jpeg_destination_mgr *dest = cinfo->dest;
+
+  *dest->next_output_byte++ = (JOCTET)val;
+  if (--dest->free_in_buffer == 0)
+    if (!(*dest->empty_output_buffer) (cinfo))
+      ERREXIT(cinfo, JERR_CANT_SUSPEND);
+}
+
+
+/*
+ * Finish up at the end of an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+finish_pass(j_compress_ptr cinfo)
+{
+  arith_entropy_ptr e = (arith_entropy_ptr)cinfo->entropy;
+  JLONG temp;
+
+  /* Section D.1.8: Termination of encoding */
+
+  /* Find the e->c in the coding interval with the largest
+   * number of trailing zero bits */
+  if ((temp = (e->a - 1 + e->c) & 0xFFFF0000UL) < e->c)
+    e->c = temp + 0x8000L;
+  else
+    e->c = temp;
+  /* Send remaining bytes to output */
+  e->c <<= e->ct;
+  if (e->c & 0xF8000000UL) {
+    /* One final overflow has to be handled */
+    if (e->buffer >= 0) {
+      if (e->zc)
+        do emit_byte(0x00, cinfo);
+        while (--e->zc);
+      emit_byte(e->buffer + 1, cinfo);
+      if (e->buffer + 1 == 0xFF)
+        emit_byte(0x00, cinfo);
+    }
+    e->zc += e->sc;  /* carry-over converts stacked 0xFF bytes to 0x00 */
+    e->sc = 0;
+  } else {
+    if (e->buffer == 0)
+      ++e->zc;
+    else if (e->buffer >= 0) {
+      if (e->zc)
+        do emit_byte(0x00, cinfo);
+        while (--e->zc);
+      emit_byte(e->buffer, cinfo);
+    }
+    if (e->sc) {
+      if (e->zc)
+        do emit_byte(0x00, cinfo);
+        while (--e->zc);
+      do {
+        emit_byte(0xFF, cinfo);
+        emit_byte(0x00, cinfo);
+      } while (--e->sc);
+    }
+  }
+  /* Output final bytes only if they are not 0x00 */
+  if (e->c & 0x7FFF800L) {
+    if (e->zc)  /* output final pending zero bytes */
+      do emit_byte(0x00, cinfo);
+      while (--e->zc);
+    emit_byte((e->c >> 19) & 0xFF, cinfo);
+    if (((e->c >> 19) & 0xFF) == 0xFF)
+      emit_byte(0x00, cinfo);
+    if (e->c & 0x7F800L) {
+      emit_byte((e->c >> 11) & 0xFF, cinfo);
+      if (((e->c >> 11) & 0xFF) == 0xFF)
+        emit_byte(0x00, cinfo);
+    }
+  }
+}
+
+
+/*
+ * The core arithmetic encoding routine (common in JPEG and JBIG).
+ * This needs to go as fast as possible.
+ * Machine-dependent optimization facilities
+ * are not utilized in this portable implementation.
+ * However, this code should be fairly efficient and
+ * may be a good base for further optimizations anyway.
+ *
+ * Parameter 'val' to be encoded may be 0 or 1 (binary decision).
+ *
+ * Note: I've added full "Pacman" termination support to the
+ * byte output routines, which is equivalent to the optional
+ * Discard_final_zeros procedure (Figure D.15) in the spec.
+ * Thus, we always produce the shortest possible output
+ * stream compliant to the spec (no trailing zero bytes,
+ * except for FF stuffing).
+ *
+ * I've also introduced a new scheme for accessing
+ * the probability estimation state machine table,
+ * derived from Markus Kuhn's JBIG implementation.
+ */
+
+LOCAL(void)
+arith_encode(j_compress_ptr cinfo, unsigned char *st, int val)
+{
+  register arith_entropy_ptr e = (arith_entropy_ptr)cinfo->entropy;
+  register unsigned char nl, nm;
+  register JLONG qe, temp;
+  register int sv;
+
+  /* Fetch values from our compact representation of Table D.2:
+   * Qe values and probability estimation state machine
+   */
+  sv = *st;
+  qe = jpeg_aritab[sv & 0x7F];  /* => Qe_Value */
+  nl = qe & 0xFF;  qe >>= 8;    /* Next_Index_LPS + Switch_MPS */
+  nm = qe & 0xFF;  qe >>= 8;    /* Next_Index_MPS */
+
+  /* Encode & estimation procedures per sections D.1.4 & D.1.5 */
+  e->a -= qe;
+  if (val != (sv >> 7)) {
+    /* Encode the less probable symbol */
+    if (e->a >= qe) {
+      /* If the interval size (qe) for the less probable symbol (LPS)
+       * is larger than the interval size for the MPS, then exchange
+       * the two symbols for coding efficiency, otherwise code the LPS
+       * as usual: */
+      e->c += e->a;
+      e->a = qe;
+    }
+    *st = (sv & 0x80) ^ nl;     /* Estimate_after_LPS */
+  } else {
+    /* Encode the more probable symbol */
+    if (e->a >= 0x8000L)
+      return;  /* A >= 0x8000 -> ready, no renormalization required */
+    if (e->a < qe) {
+      /* If the interval size (qe) for the less probable symbol (LPS)
+       * is larger than the interval size for the MPS, then exchange
+       * the two symbols for coding efficiency: */
+      e->c += e->a;
+      e->a = qe;
+    }
+    *st = (sv & 0x80) ^ nm;     /* Estimate_after_MPS */
+  }
+
+  /* Renormalization & data output per section D.1.6 */
+  do {
+    e->a <<= 1;
+    e->c <<= 1;
+    if (--e->ct == 0) {
+      /* Another byte is ready for output */
+      temp = e->c >> 19;
+      if (temp > 0xFF) {
+        /* Handle overflow over all stacked 0xFF bytes */
+        if (e->buffer >= 0) {
+          if (e->zc)
+            do emit_byte(0x00, cinfo);
+            while (--e->zc);
+          emit_byte(e->buffer + 1, cinfo);
+          if (e->buffer + 1 == 0xFF)
+            emit_byte(0x00, cinfo);
+        }
+        e->zc += e->sc;  /* carry-over converts stacked 0xFF bytes to 0x00 */
+        e->sc = 0;
+        /* Note: The 3 spacer bits in the C register guarantee
+         * that the new buffer byte can't be 0xFF here
+         * (see page 160 in the P&M JPEG book). */
+        e->buffer = temp & 0xFF;  /* new output byte, might overflow later */
+      } else if (temp == 0xFF) {
+        ++e->sc;  /* stack 0xFF byte (which might overflow later) */
+      } else {
+        /* Output all stacked 0xFF bytes, they will not overflow any more */
+        if (e->buffer == 0)
+          ++e->zc;
+        else if (e->buffer >= 0) {
+          if (e->zc)
+            do emit_byte(0x00, cinfo);
+            while (--e->zc);
+          emit_byte(e->buffer, cinfo);
+        }
+        if (e->sc) {
+          if (e->zc)
+            do emit_byte(0x00, cinfo);
+            while (--e->zc);
+          do {
+            emit_byte(0xFF, cinfo);
+            emit_byte(0x00, cinfo);
+          } while (--e->sc);
+        }
+        e->buffer = temp & 0xFF;  /* new output byte (can still overflow) */
+      }
+      e->c &= 0x7FFFFL;
+      e->ct += 8;
+    }
+  } while (e->a < 0x8000L);
+}
+
+
+/*
+ * Emit a restart marker & resynchronize predictions.
+ */
+
+LOCAL(void)
+emit_restart(j_compress_ptr cinfo, int restart_num)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy;
+  int ci;
+  jpeg_component_info *compptr;
+
+  finish_pass(cinfo);
+
+  emit_byte(0xFF, cinfo);
+  emit_byte(JPEG_RST0 + restart_num, cinfo);
+
+  /* Re-initialize statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->progressive_mode == 0 || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+      memset(entropy->dc_stats[compptr->dc_tbl_no], 0, DC_STAT_BINS);
+      /* Reset DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->progressive_mode == 0 || cinfo->Se) {
+      memset(entropy->ac_stats[compptr->ac_tbl_no], 0, AC_STAT_BINS);
+    }
+  }
+
+  /* Reset arithmetic encoding variables */
+  entropy->c = 0;
+  entropy->a = 0x10000L;
+  entropy->sc = 0;
+  entropy->zc = 0;
+  entropy->ct = 11;
+  entropy->buffer = -1;  /* empty */
+}
+
+
+/*
+ * MCU encoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_first(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl;
+  int v, v2, m;
+  ISHIFT_TEMPS
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    tbl = cinfo->cur_comp_info[ci]->dc_tbl_no;
+
+    /* Compute the DC value after the required point transform by Al.
+     * This is simply an arithmetic right shift.
+     */
+    m = IRIGHT_SHIFT((int)((*block)[0]), cinfo->Al);
+
+    /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.4: Encode_DC_DIFF */
+    if ((v = m - entropy->last_dc_val[ci]) == 0) {
+      arith_encode(cinfo, st, 0);
+      entropy->dc_context[ci] = 0;      /* zero diff category */
+    } else {
+      entropy->last_dc_val[ci] = m;
+      arith_encode(cinfo, st, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+        arith_encode(cinfo, st + 1, 0); /* Table F.4: SS = S0 + 1 */
+        st += 2;                        /* Table F.4: SP = S0 + 2 */
+        entropy->dc_context[ci] = 4;    /* small positive diff category */
+      } else {
+        v = -v;
+        arith_encode(cinfo, st + 1, 1); /* Table F.4: SS = S0 + 1 */
+        st += 3;                        /* Table F.4: SN = S0 + 3 */
+        entropy->dc_context[ci] = 8;    /* small negative diff category */
+      }
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+        arith_encode(cinfo, st, 1);
+        m = 1;
+        v2 = v;
+        st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */
+        while (v2 >>= 1) {
+          arith_encode(cinfo, st, 1);
+          m <<= 1;
+          st += 1;
+        }
+      }
+      arith_encode(cinfo, st, 0);
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int)((1L << cinfo->arith_dc_L[tbl]) >> 1))
+        entropy->dc_context[ci] = 0;    /* zero diff category */
+      else if (m > (int)((1L << cinfo->arith_dc_U[tbl]) >> 1))
+        entropy->dc_context[ci] += 8;   /* large diff category */
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+        arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_first(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke;
+  int v, v2, m;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */
+
+  /* Establish EOB (end-of-block) index */
+  for (ke = cinfo->Se; ke > 0; ke--)
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if ((v = (*block)[jpeg_natural_order[ke]]) >= 0) {
+      if (v >>= cinfo->Al) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Al) break;
+    }
+
+  /* Figure F.5: Encode_AC_Coefficients */
+  for (k = cinfo->Ss; k <= ke; k++) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 0);         /* EOB decision */
+    for (;;) {
+      if ((v = (*block)[jpeg_natural_order[k]]) >= 0) {
+        if (v >>= cinfo->Al) {
+          arith_encode(cinfo, st + 1, 1);
+          arith_encode(cinfo, entropy->fixed_bin, 0);
+          break;
+        }
+      } else {
+        v = -v;
+        if (v >>= cinfo->Al) {
+          arith_encode(cinfo, st + 1, 1);
+          arith_encode(cinfo, entropy->fixed_bin, 1);
+          break;
+        }
+      }
+      arith_encode(cinfo, st + 1, 0);  st += 3;  k++;
+    }
+    st += 2;
+    /* Figure F.8: Encoding the magnitude category of v */
+    m = 0;
+    if (v -= 1) {
+      arith_encode(cinfo, st, 1);
+      m = 1;
+      v2 = v;
+      if (v2 >>= 1) {
+        arith_encode(cinfo, st, 1);
+        m <<= 1;
+        st = entropy->ac_stats[tbl] +
+             (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+        while (v2 >>= 1) {
+          arith_encode(cinfo, st, 1);
+          m <<= 1;
+          st += 1;
+        }
+      }
+    }
+    arith_encode(cinfo, st, 0);
+    /* Figure F.9: Encoding the magnitude bit pattern of v */
+    st += 14;
+    while (m >>= 1)
+      arith_encode(cinfo, st, (m & v) ? 1 : 0);
+  }
+  /* Encode EOB decision only if k <= cinfo->Se */
+  if (k <= cinfo->Se) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for DC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+encode_mcu_DC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy;
+  unsigned char *st;
+  int Al, blkn;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  st = entropy->fixed_bin;      /* use fixed probability estimation */
+  Al = cinfo->Al;
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    /* We simply emit the Al'th bit of the DC coefficient value. */
+    arith_encode(cinfo, st, (MCU_data[blkn][0][0] >> Al) & 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * MCU encoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+encode_mcu_AC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy;
+  JBLOCKROW block;
+  unsigned char *st;
+  int tbl, k, ke, kex;
+  int v;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data block */
+  block = MCU_data[0];
+  tbl = cinfo->cur_comp_info[0]->ac_tbl_no;
+
+  /* Section G.1.3.3: Encoding of AC coefficients */
+
+  /* Establish EOB (end-of-block) index */
+  for (ke = cinfo->Se; ke > 0; ke--)
+    /* We must apply the point transform by Al.  For AC coefficients this
+     * is an integer division with rounding towards 0.  To do this portably
+     * in C, we shift after obtaining the absolute value.
+     */
+    if ((v = (*block)[jpeg_natural_order[ke]]) >= 0) {
+      if (v >>= cinfo->Al) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Al) break;
+    }
+
+  /* Establish EOBx (previous stage end-of-block) index */
+  for (kex = ke; kex > 0; kex--)
+    if ((v = (*block)[jpeg_natural_order[kex]]) >= 0) {
+      if (v >>= cinfo->Ah) break;
+    } else {
+      v = -v;
+      if (v >>= cinfo->Ah) break;
+    }
+
+  /* Figure G.10: Encode_AC_Coefficients_SA */
+  for (k = cinfo->Ss; k <= ke; k++) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    if (k > kex)
+      arith_encode(cinfo, st, 0);       /* EOB decision */
+    for (;;) {
+      if ((v = (*block)[jpeg_natural_order[k]]) >= 0) {
+        if (v >>= cinfo->Al) {
+          if (v >> 1)                   /* previously nonzero coef */
+            arith_encode(cinfo, st + 2, (v & 1));
+          else {                        /* newly nonzero coef */
+            arith_encode(cinfo, st + 1, 1);
+            arith_encode(cinfo, entropy->fixed_bin, 0);
+          }
+          break;
+        }
+      } else {
+        v = -v;
+        if (v >>= cinfo->Al) {
+          if (v >> 1)                   /* previously nonzero coef */
+            arith_encode(cinfo, st + 2, (v & 1));
+          else {                        /* newly nonzero coef */
+            arith_encode(cinfo, st + 1, 1);
+            arith_encode(cinfo, entropy->fixed_bin, 1);
+          }
+          break;
+        }
+      }
+      arith_encode(cinfo, st + 1, 0);  st += 3;  k++;
+    }
+  }
+  /* Encode EOB decision only if k <= cinfo->Se */
+  if (k <= cinfo->Se) {
+    st = entropy->ac_stats[tbl] + 3 * (k - 1);
+    arith_encode(cinfo, st, 1);
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Encode and output one MCU's worth of arithmetic-compressed coefficients.
+ */
+
+METHODDEF(boolean)
+encode_mcu(j_compress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy;
+  jpeg_component_info *compptr;
+  JBLOCKROW block;
+  unsigned char *st;
+  int blkn, ci, tbl, k, ke;
+  int v, v2, m;
+
+  /* Emit restart marker if needed */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0) {
+      emit_restart(cinfo, entropy->next_restart_num);
+      entropy->restarts_to_go = cinfo->restart_interval;
+      entropy->next_restart_num++;
+      entropy->next_restart_num &= 7;
+    }
+    entropy->restarts_to_go--;
+  }
+
+  /* Encode the MCU data blocks */
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+    ci = cinfo->MCU_membership[blkn];
+    compptr = cinfo->cur_comp_info[ci];
+
+    /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */
+
+    tbl = compptr->dc_tbl_no;
+
+    /* Table F.4: Point to statistics bin S0 for DC coefficient coding */
+    st = entropy->dc_stats[tbl] + entropy->dc_context[ci];
+
+    /* Figure F.4: Encode_DC_DIFF */
+    if ((v = (*block)[0] - entropy->last_dc_val[ci]) == 0) {
+      arith_encode(cinfo, st, 0);
+      entropy->dc_context[ci] = 0;      /* zero diff category */
+    } else {
+      entropy->last_dc_val[ci] = (*block)[0];
+      arith_encode(cinfo, st, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+        arith_encode(cinfo, st + 1, 0); /* Table F.4: SS = S0 + 1 */
+        st += 2;                        /* Table F.4: SP = S0 + 2 */
+        entropy->dc_context[ci] = 4;    /* small positive diff category */
+      } else {
+        v = -v;
+        arith_encode(cinfo, st + 1, 1); /* Table F.4: SS = S0 + 1 */
+        st += 3;                        /* Table F.4: SN = S0 + 3 */
+        entropy->dc_context[ci] = 8;    /* small negative diff category */
+      }
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+        arith_encode(cinfo, st, 1);
+        m = 1;
+        v2 = v;
+        st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */
+        while (v2 >>= 1) {
+          arith_encode(cinfo, st, 1);
+          m <<= 1;
+          st += 1;
+        }
+      }
+      arith_encode(cinfo, st, 0);
+      /* Section F.1.4.4.1.2: Establish dc_context conditioning category */
+      if (m < (int)((1L << cinfo->arith_dc_L[tbl]) >> 1))
+        entropy->dc_context[ci] = 0;    /* zero diff category */
+      else if (m > (int)((1L << cinfo->arith_dc_U[tbl]) >> 1))
+        entropy->dc_context[ci] += 8;   /* large diff category */
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+        arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+
+    /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */
+
+    tbl = compptr->ac_tbl_no;
+
+    /* Establish EOB (end-of-block) index */
+    for (ke = DCTSIZE2 - 1; ke > 0; ke--)
+      if ((*block)[jpeg_natural_order[ke]]) break;
+
+    /* Figure F.5: Encode_AC_Coefficients */
+    for (k = 1; k <= ke; k++) {
+      st = entropy->ac_stats[tbl] + 3 * (k - 1);
+      arith_encode(cinfo, st, 0);       /* EOB decision */
+      while ((v = (*block)[jpeg_natural_order[k]]) == 0) {
+        arith_encode(cinfo, st + 1, 0);  st += 3;  k++;
+      }
+      arith_encode(cinfo, st + 1, 1);
+      /* Figure F.6: Encoding nonzero value v */
+      /* Figure F.7: Encoding the sign of v */
+      if (v > 0) {
+        arith_encode(cinfo, entropy->fixed_bin, 0);
+      } else {
+        v = -v;
+        arith_encode(cinfo, entropy->fixed_bin, 1);
+      }
+      st += 2;
+      /* Figure F.8: Encoding the magnitude category of v */
+      m = 0;
+      if (v -= 1) {
+        arith_encode(cinfo, st, 1);
+        m = 1;
+        v2 = v;
+        if (v2 >>= 1) {
+          arith_encode(cinfo, st, 1);
+          m <<= 1;
+          st = entropy->ac_stats[tbl] +
+               (k <= cinfo->arith_ac_K[tbl] ? 189 : 217);
+          while (v2 >>= 1) {
+            arith_encode(cinfo, st, 1);
+            m <<= 1;
+            st += 1;
+          }
+        }
+      }
+      arith_encode(cinfo, st, 0);
+      /* Figure F.9: Encoding the magnitude bit pattern of v */
+      st += 14;
+      while (m >>= 1)
+        arith_encode(cinfo, st, (m & v) ? 1 : 0);
+    }
+    /* Encode EOB decision only if k <= DCTSIZE2 - 1 */
+    if (k <= DCTSIZE2 - 1) {
+      st = entropy->ac_stats[tbl] + 3 * (k - 1);
+      arith_encode(cinfo, st, 1);
+    }
+  }
+
+  return TRUE;
+}
+
+
+/*
+ * Initialize for an arithmetic-compressed scan.
+ */
+
+METHODDEF(void)
+start_pass(j_compress_ptr cinfo, boolean gather_statistics)
+{
+  arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy;
+  int ci, tbl;
+  jpeg_component_info *compptr;
+
+  if (gather_statistics)
+    /* Make sure to avoid that in the master control logic!
+     * We are fully adaptive here and need no extra
+     * statistics gathering pass!
+     */
+    ERREXIT(cinfo, JERR_NOTIMPL);
+
+  /* We assume jcmaster.c already validated the progressive scan parameters. */
+
+  /* Select execution routines */
+  if (cinfo->progressive_mode) {
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+        entropy->pub.encode_mcu = encode_mcu_DC_first;
+      else
+        entropy->pub.encode_mcu = encode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+        entropy->pub.encode_mcu = encode_mcu_DC_refine;
+      else
+        entropy->pub.encode_mcu = encode_mcu_AC_refine;
+    }
+  } else
+    entropy->pub.encode_mcu = encode_mcu;
+
+  /* Allocate & initialize requested statistics areas */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    /* DC needs no table for refinement scan */
+    if (cinfo->progressive_mode == 0 || (cinfo->Ss == 0 && cinfo->Ah == 0)) {
+      tbl = compptr->dc_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+        ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->dc_stats[tbl] == NULL)
+        entropy->dc_stats[tbl] = (unsigned char *)(*cinfo->mem->alloc_small)
+          ((j_common_ptr)cinfo, JPOOL_IMAGE, DC_STAT_BINS);
+      memset(entropy->dc_stats[tbl], 0, DC_STAT_BINS);
+      /* Initialize DC predictions to 0 */
+      entropy->last_dc_val[ci] = 0;
+      entropy->dc_context[ci] = 0;
+    }
+    /* AC needs no table when not present */
+    if (cinfo->progressive_mode == 0 || cinfo->Se) {
+      tbl = compptr->ac_tbl_no;
+      if (tbl < 0 || tbl >= NUM_ARITH_TBLS)
+        ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl);
+      if (entropy->ac_stats[tbl] == NULL)
+        entropy->ac_stats[tbl] = (unsigned char *)(*cinfo->mem->alloc_small)
+          ((j_common_ptr)cinfo, JPOOL_IMAGE, AC_STAT_BINS);
+      memset(entropy->ac_stats[tbl], 0, AC_STAT_BINS);
+#ifdef CALCULATE_SPECTRAL_CONDITIONING
+      if (cinfo->progressive_mode)
+        /* Section G.1.3.2: Set appropriate arithmetic conditioning value Kx */
+        cinfo->arith_ac_K[tbl] = cinfo->Ss +
+                                 ((8 + cinfo->Se - cinfo->Ss) >> 4);
+#endif
+    }
+  }
+
+  /* Initialize arithmetic encoding variables */
+  entropy->c = 0;
+  entropy->a = 0x10000L;
+  entropy->sc = 0;
+  entropy->zc = 0;
+  entropy->ct = 11;
+  entropy->buffer = -1;  /* empty */
+
+  /* Initialize restart stuff */
+  entropy->restarts_to_go = cinfo->restart_interval;
+  entropy->next_restart_num = 0;
+}
+
+
+/*
+ * Module initialization routine for arithmetic entropy encoding.
+ */
+
+GLOBAL(void)
+jinit_arith_encoder(j_compress_ptr cinfo)
+{
+  arith_entropy_ptr entropy;
+  int i;
+
+  entropy = (arith_entropy_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                sizeof(arith_entropy_encoder));
+  cinfo->entropy = (struct jpeg_entropy_encoder *)entropy;
+  entropy->pub.start_pass = start_pass;
+  entropy->pub.finish_pass = finish_pass;
+
+  /* Mark tables unallocated */
+  for (i = 0; i < NUM_ARITH_TBLS; i++) {
+    entropy->dc_stats[i] = NULL;
+    entropy->ac_stats[i] = NULL;
+  }
+
+  /* Initialize index for fixed probability estimation */
+  entropy->fixed_bin[0] = 113;
+}
@@ -0,0 +1,454 @@
+/*
+ * jccoefct.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1994-1997, Thomas G. Lane.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2022, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains the coefficient buffer controller for compression.
+ * This controller is the top level of the lossy JPEG compressor proper.
+ * The coefficient buffer lies between forward-DCT and entropy encoding steps.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jsamplecomp.h"
+
+
+/* We use a full-image coefficient buffer when doing Huffman optimization,
+ * and also for writing multiple-scan JPEG files.  In all cases, the DCT
+ * step is run during the first pass, and subsequent passes need only read
+ * the buffered coefficients.
+ */
+#ifdef ENTROPY_OPT_SUPPORTED
+#define FULL_COEF_BUFFER_SUPPORTED
+#else
+#ifdef C_MULTISCAN_FILES_SUPPORTED
+#define FULL_COEF_BUFFER_SUPPORTED
+#endif
+#endif
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_c_coef_controller pub; /* public fields */
+
+  JDIMENSION iMCU_row_num;      /* iMCU row # within image */
+  JDIMENSION mcu_ctr;           /* counts MCUs processed in current row */
+  int MCU_vert_offset;          /* counts MCU rows within iMCU row */
+  int MCU_rows_per_iMCU_row;    /* number of such rows needed */
+
+  /* For single-pass compression, it's sufficient to buffer just one MCU
+   * (although this may prove a bit slow in practice).  We allocate a
+   * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each
+   * MCU constructed and sent.  In multi-pass modes, this array points to the
+   * current MCU's blocks within the virtual arrays.
+   */
+  JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU];
+
+  /* In multi-pass modes, we need a virtual block array for each component. */
+  jvirt_barray_ptr whole_image[MAX_COMPONENTS];
+} my_coef_controller;
+
+typedef my_coef_controller *my_coef_ptr;
+
+
+/* Forward declarations */
+METHODDEF(boolean) compress_data(j_compress_ptr cinfo, _JSAMPIMAGE input_buf);
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+METHODDEF(boolean) compress_first_pass(j_compress_ptr cinfo,
+                                       _JSAMPIMAGE input_buf);
+METHODDEF(boolean) compress_output(j_compress_ptr cinfo,
+                                   _JSAMPIMAGE input_buf);
+#endif
+
+
+LOCAL(void)
+start_iMCU_row(j_compress_ptr cinfo)
+/* Reset within-iMCU-row counters for a new row */
+{
+  my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
+
+  /* In an interleaved scan, an MCU row is the same as an iMCU row.
+   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
+   * But at the bottom of the image, process only what's left.
+   */
+  if (cinfo->comps_in_scan > 1) {
+    coef->MCU_rows_per_iMCU_row = 1;
+  } else {
+    if (coef->iMCU_row_num < (cinfo->total_iMCU_rows - 1))
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
+    else
+      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
+  }
+
+  coef->mcu_ctr = 0;
+  coef->MCU_vert_offset = 0;
+}
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_coef(j_compress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
+
+  coef->iMCU_row_num = 0;
+  start_iMCU_row(cinfo);
+
+  switch (pass_mode) {
+  case JBUF_PASS_THRU:
+    if (coef->whole_image[0] != NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub._compress_data = compress_data;
+    break;
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+  case JBUF_SAVE_AND_PASS:
+    if (coef->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub._compress_data = compress_first_pass;
+    break;
+  case JBUF_CRANK_DEST:
+    if (coef->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    coef->pub._compress_data = compress_output;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    break;
+  }
+}
+
+
+/*
+ * Process some data in the single-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the image.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf contains a plane for each component in image,
+ * which we index according to the component's SOF position.
+ */
+
+METHODDEF(boolean)
+compress_data(j_compress_ptr cinfo, _JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
+  JDIMENSION MCU_col_num;       /* index of current MCU within row */
+  JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  int blkn, bi, ci, yindex, yoffset, blockcnt;
+  JDIMENSION ypos, xpos;
+  jpeg_component_info *compptr;
+
+  /* Loop to write as much as one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->mcu_ctr; MCU_col_num <= last_MCU_col;
+         MCU_col_num++) {
+      /* Determine where data comes from in input_buf and do the DCT thing.
+       * Each call on forward_DCT processes a horizontal row of DCT blocks
+       * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks
+       * sequentially.  Dummy blocks at the right or bottom edge are filled in
+       * specially.  The data in them does not matter for image reconstruction,
+       * so we fill them with values that will encode to the smallest amount of
+       * data, viz: all zeroes in the AC entries, DC entries equal to previous
+       * block's DC value.  (Thanks to Thomas Kinsman for this idea.)
+       */
+      blkn = 0;
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+        compptr = cinfo->cur_comp_info[ci];
+        blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width :
+                                                  compptr->last_col_width;
+        xpos = MCU_col_num * compptr->MCU_sample_width;
+        ypos = yoffset * DCTSIZE; /* ypos == (yoffset+yindex) * DCTSIZE */
+        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+          if (coef->iMCU_row_num < last_iMCU_row ||
+              yoffset + yindex < compptr->last_row_height) {
+            (*cinfo->fdct->_forward_DCT) (cinfo, compptr,
+                                          input_buf[compptr->component_index],
+                                          coef->MCU_buffer[blkn],
+                                          ypos, xpos, (JDIMENSION)blockcnt);
+            if (blockcnt < compptr->MCU_width) {
+              /* Create some dummy blocks at the right edge of the image. */
+              jzero_far((void *)coef->MCU_buffer[blkn + blockcnt],
+                        (compptr->MCU_width - blockcnt) * sizeof(JBLOCK));
+              for (bi = blockcnt; bi < compptr->MCU_width; bi++) {
+                coef->MCU_buffer[blkn + bi][0][0] =
+                  coef->MCU_buffer[blkn + bi - 1][0][0];
+              }
+            }
+          } else {
+            /* Create a row of dummy blocks at the bottom of the image. */
+            jzero_far((void *)coef->MCU_buffer[blkn],
+                      compptr->MCU_width * sizeof(JBLOCK));
+            for (bi = 0; bi < compptr->MCU_width; bi++) {
+              coef->MCU_buffer[blkn + bi][0][0] =
+                coef->MCU_buffer[blkn - 1][0][0];
+            }
+          }
+          blkn += compptr->MCU_width;
+          ypos += DCTSIZE;
+        }
+      }
+      /* Try to write the MCU.  In event of a suspension failure, we will
+       * re-DCT the MCU on restart (a bit inefficient, could be fixed...)
+       */
+      if (!(*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
+        /* Suspension forced; update state counters and exit */
+        coef->MCU_vert_offset = yoffset;
+        coef->mcu_ctr = MCU_col_num;
+        return FALSE;
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->mcu_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  coef->iMCU_row_num++;
+  start_iMCU_row(cinfo);
+  return TRUE;
+}
+
+
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+
+/*
+ * Process some data in the first pass of a multi-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the image.
+ * This amount of data is read from the source buffer, DCT'd and quantized,
+ * and saved into the virtual arrays.  We also generate suitable dummy blocks
+ * as needed at the right and lower edges.  (The dummy blocks are constructed
+ * in the virtual arrays, which have been padded appropriately.)  This makes
+ * it possible for subsequent passes not to worry about real vs. dummy blocks.
+ *
+ * We must also emit the data to the entropy encoder.  This is conveniently
+ * done by calling compress_output() after we've loaded the current strip
+ * of the virtual arrays.
+ *
+ * NB: input_buf contains a plane for each component in image.  All
+ * components are DCT'd and loaded into the virtual arrays in this pass.
+ * However, it may be that only a subset of the components are emitted to
+ * the entropy encoder during this first pass; be careful about looking
+ * at the scan-dependent variables (MCU dimensions, etc).
+ */
+
+METHODDEF(boolean)
+compress_first_pass(j_compress_ptr cinfo, _JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  JDIMENSION blocks_across, MCUs_across, MCUindex;
+  int bi, ci, h_samp_factor, block_row, block_rows, ndummy;
+  JCOEF lastDC;
+  jpeg_component_info *compptr;
+  JBLOCKARRAY buffer;
+  JBLOCKROW thisblockrow, lastblockrow;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Align the virtual buffer for this component. */
+    buffer = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr)cinfo, coef->whole_image[ci],
+       coef->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION)compptr->v_samp_factor, TRUE);
+    /* Count non-dummy DCT block rows in this iMCU row. */
+    if (coef->iMCU_row_num < last_iMCU_row)
+      block_rows = compptr->v_samp_factor;
+    else {
+      /* NB: can't use last_row_height here, since may not be set! */
+      block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor);
+      if (block_rows == 0) block_rows = compptr->v_samp_factor;
+    }
+    blocks_across = compptr->width_in_blocks;
+    h_samp_factor = compptr->h_samp_factor;
+    /* Count number of dummy blocks to be added at the right margin. */
+    ndummy = (int)(blocks_across % h_samp_factor);
+    if (ndummy > 0)
+      ndummy = h_samp_factor - ndummy;
+    /* Perform DCT for all non-dummy blocks in this iMCU row.  Each call
+     * on forward_DCT processes a complete horizontal row of DCT blocks.
+     */
+    for (block_row = 0; block_row < block_rows; block_row++) {
+      thisblockrow = buffer[block_row];
+      (*cinfo->fdct->_forward_DCT) (cinfo, compptr,
+                                    input_buf[ci], thisblockrow,
+                                    (JDIMENSION)(block_row * DCTSIZE),
+                                    (JDIMENSION)0, blocks_across);
+      if (ndummy > 0) {
+        /* Create dummy blocks at the right edge of the image. */
+        thisblockrow += blocks_across; /* => first dummy block */
+        jzero_far((void *)thisblockrow, ndummy * sizeof(JBLOCK));
+        lastDC = thisblockrow[-1][0];
+        for (bi = 0; bi < ndummy; bi++) {
+          thisblockrow[bi][0] = lastDC;
+        }
+      }
+    }
+    /* If at end of image, create dummy block rows as needed.
+     * The tricky part here is that within each MCU, we want the DC values
+     * of the dummy blocks to match the last real block's DC value.
+     * This squeezes a few more bytes out of the resulting file...
+     */
+    if (coef->iMCU_row_num == last_iMCU_row) {
+      blocks_across += ndummy;  /* include lower right corner */
+      MCUs_across = blocks_across / h_samp_factor;
+      for (block_row = block_rows; block_row < compptr->v_samp_factor;
+           block_row++) {
+        thisblockrow = buffer[block_row];
+        lastblockrow = buffer[block_row - 1];
+        jzero_far((void *)thisblockrow,
+                  (size_t)(blocks_across * sizeof(JBLOCK)));
+        for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) {
+          lastDC = lastblockrow[h_samp_factor - 1][0];
+          for (bi = 0; bi < h_samp_factor; bi++) {
+            thisblockrow[bi][0] = lastDC;
+          }
+          thisblockrow += h_samp_factor; /* advance to next MCU in row */
+          lastblockrow += h_samp_factor;
+        }
+      }
+    }
+  }
+  /* NB: compress_output will increment iMCU_row_num if successful.
+   * A suspension return will result in redoing all the work above next time.
+   */
+
+  /* Emit data to the entropy encoder, sharing code with subsequent passes */
+  return compress_output(cinfo, input_buf);
+}
+
+
+/*
+ * Process some data in subsequent passes of a multi-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor block rows for each component in the scan.
+ * The data is obtained from the virtual arrays and fed to the entropy coder.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf is ignored; it is likely to be a NULL pointer.
+ */
+
+METHODDEF(boolean)
+compress_output(j_compress_ptr cinfo, _JSAMPIMAGE input_buf)
+{
+  my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
+  JDIMENSION MCU_col_num;       /* index of current MCU within row */
+  int blkn, ci, xindex, yindex, yoffset;
+  JDIMENSION start_col;
+  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
+  JBLOCKROW buffer_ptr;
+  jpeg_component_info *compptr;
+
+  /* Align the virtual buffers for the components used in this scan.
+   * NB: during first pass, this is safe only because the buffers will
+   * already be aligned properly, so jmemmgr.c won't need to do any I/O.
+   */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    buffer[ci] = (*cinfo->mem->access_virt_barray)
+      ((j_common_ptr)cinfo, coef->whole_image[compptr->component_index],
+       coef->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION)compptr->v_samp_factor, FALSE);
+  }
+
+  /* Loop to process one whole iMCU row */
+  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
+       yoffset++) {
+    for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row;
+         MCU_col_num++) {
+      /* Construct list of pointers to DCT blocks belonging to this MCU */
+      blkn = 0;                 /* index of current DCT block within MCU */
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+        compptr = cinfo->cur_comp_info[ci];
+        start_col = MCU_col_num * compptr->MCU_width;
+        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
+          buffer_ptr = buffer[ci][yindex + yoffset] + start_col;
+          for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
+            coef->MCU_buffer[blkn++] = buffer_ptr++;
+          }
+        }
+      }
+      /* Try to write the MCU. */
+      if (!(*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
+        /* Suspension forced; update state counters and exit */
+        coef->MCU_vert_offset = yoffset;
+        coef->mcu_ctr = MCU_col_num;
+        return FALSE;
+      }
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    coef->mcu_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  coef->iMCU_row_num++;
+  start_iMCU_row(cinfo);
+  return TRUE;
+}
+
+#endif /* FULL_COEF_BUFFER_SUPPORTED */
+
+
+/*
+ * Initialize coefficient buffer controller.
+ */
+
+GLOBAL(void)
+_jinit_c_coef_controller(j_compress_ptr cinfo, boolean need_full_buffer)
+{
+  my_coef_ptr coef;
+
+  if (cinfo->data_precision != BITS_IN_JSAMPLE)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  coef = (my_coef_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                sizeof(my_coef_controller));
+  cinfo->coef = (struct jpeg_c_coef_controller *)coef;
+  coef->pub.start_pass = start_pass_coef;
+
+  /* Create the coefficient buffer. */
+  if (need_full_buffer) {
+#ifdef FULL_COEF_BUFFER_SUPPORTED
+    /* Allocate a full-image virtual array for each component, */
+    /* padded to a multiple of samp_factor DCT blocks in each direction. */
+    int ci;
+    jpeg_component_info *compptr;
+
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+         ci++, compptr++) {
+      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
+        ((j_common_ptr)cinfo, JPOOL_IMAGE, FALSE,
+         (JDIMENSION)jround_up((long)compptr->width_in_blocks,
+                               (long)compptr->h_samp_factor),
+         (JDIMENSION)jround_up((long)compptr->height_in_blocks,
+                               (long)compptr->v_samp_factor),
+         (JDIMENSION)compptr->v_samp_factor);
+    }
+#else
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+#endif
+  } else {
+    /* We only need a single-MCU buffer. */
+    JBLOCKROW buffer;
+    int i;
+
+    buffer = (JBLOCKROW)
+      (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                  C_MAX_BLOCKS_IN_MCU * sizeof(JBLOCK));
+    for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) {
+      coef->MCU_buffer[i] = buffer + i;
+    }
+    coef->whole_image[0] = NULL; /* flag for no virtual arrays */
+  }
+}
@@ -0,0 +1,152 @@
+/*
+ * jccolext.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2009-2012, 2015, 2022, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains input colorspace conversion routines.
+ */
+
+
+/* This file is included by jccolor.c */
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ *
+ * Note that we change from the application's interleaved-pixel format
+ * to our internal noninterleaved, one-plane-per-component format.
+ * The input buffer is therefore three times as wide as the output buffer.
+ *
+ * A starting row offset is provided only for the output buffer.  The caller
+ * can easily adjust the passed input_buf value to accommodate any row
+ * offset required on that side.
+ */
+
+INLINE
+LOCAL(void)
+rgb_ycc_convert_internal(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+                         _JSAMPIMAGE output_buf, JDIMENSION output_row,
+                         int num_rows)
+{
+#if BITS_IN_JSAMPLE != 16
+  my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert;
+  register int r, g, b;
+  register JLONG *ctab = cconvert->rgb_ycc_tab;
+  register _JSAMPROW inptr;
+  register _JSAMPROW outptr0, outptr1, outptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = RANGE_LIMIT(inptr[RGB_RED]);
+      g = RANGE_LIMIT(inptr[RGB_GREEN]);
+      b = RANGE_LIMIT(inptr[RGB_BLUE]);
+      inptr += RGB_PIXELSIZE;
+      /* If the inputs are 0.._MAXJSAMPLE, the outputs of these equations
+       * must be too; we do not need an explicit range-limiting operation.
+       * Hence the value being shifted is never negative, and we don't
+       * need the general RIGHT_SHIFT macro.
+       */
+      /* Y */
+      outptr0[col] = (_JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] +
+                                 ctab[b + B_Y_OFF]) >> SCALEBITS);
+      /* Cb */
+      outptr1[col] = (_JSAMPLE)((ctab[r + R_CB_OFF] + ctab[g + G_CB_OFF] +
+                                 ctab[b + B_CB_OFF]) >> SCALEBITS);
+      /* Cr */
+      outptr2[col] = (_JSAMPLE)((ctab[r + R_CR_OFF] + ctab[g + G_CR_OFF] +
+                                 ctab[b + B_CR_OFF]) >> SCALEBITS);
+    }
+  }
+#else
+  ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+#endif
+}
+
+
+/**************** Cases other than RGB -> YCbCr **************/
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles RGB->grayscale conversion, which is the same
+ * as the RGB->Y portion of RGB->YCbCr.
+ * We assume rgb_ycc_start has been called (we only use the Y tables).
+ */
+
+INLINE
+LOCAL(void)
+rgb_gray_convert_internal(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+                          _JSAMPIMAGE output_buf, JDIMENSION output_row,
+                          int num_rows)
+{
+#if BITS_IN_JSAMPLE != 16
+  my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert;
+  register int r, g, b;
+  register JLONG *ctab = cconvert->rgb_ycc_tab;
+  register _JSAMPROW inptr;
+  register _JSAMPROW outptr;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr = output_buf[0][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = RANGE_LIMIT(inptr[RGB_RED]);
+      g = RANGE_LIMIT(inptr[RGB_GREEN]);
+      b = RANGE_LIMIT(inptr[RGB_BLUE]);
+      inptr += RGB_PIXELSIZE;
+      /* Y */
+      outptr[col] = (_JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] +
+                                ctab[b + B_Y_OFF]) >> SCALEBITS);
+    }
+  }
+#else
+  ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+#endif
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles extended RGB->plain RGB conversion
+ */
+
+INLINE
+LOCAL(void)
+rgb_rgb_convert_internal(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+                         _JSAMPIMAGE output_buf, JDIMENSION output_row,
+                         int num_rows)
+{
+  register _JSAMPROW inptr;
+  register _JSAMPROW outptr0, outptr1, outptr2;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      outptr0[col] = inptr[RGB_RED];
+      outptr1[col] = inptr[RGB_GREEN];
+      outptr2[col] = inptr[RGB_BLUE];
+      inptr += RGB_PIXELSIZE;
+    }
+  }
+}
@@ -0,0 +1,732 @@
+/*
+ * jccolor.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1991-1996, Thomas G. Lane.
+ * libjpeg-turbo Modifications:
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright (C) 2009-2012, 2015, 2022, D. R. Commander.
+ * Copyright (C) 2014, MIPS Technologies, Inc., California.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains input colorspace conversion routines.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jsimd.h"
+#include "jsamplecomp.h"
+
+
+#if BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED)
+
+/* Private subobject */
+
+typedef struct {
+  struct jpeg_color_converter pub; /* public fields */
+
+#if BITS_IN_JSAMPLE != 16
+  /* Private state for RGB->YCC conversion */
+  JLONG *rgb_ycc_tab;           /* => table for RGB to YCbCr conversion */
+#endif
+} my_color_converter;
+
+typedef my_color_converter *my_cconvert_ptr;
+
+
+/**************** RGB -> YCbCr conversion: most common case **************/
+
+/*
+ * YCbCr is defined per CCIR 601-1, except that Cb and Cr are
+ * normalized to the range 0.._MAXJSAMPLE rather than -0.5 .. 0.5.
+ * The conversion equations to be implemented are therefore
+ *      Y  =  0.29900 * R + 0.58700 * G + 0.11400 * B
+ *      Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B  + _CENTERJSAMPLE
+ *      Cr =  0.50000 * R - 0.41869 * G - 0.08131 * B  + _CENTERJSAMPLE
+ * (These numbers are derived from TIFF 6.0 section 21, dated 3-June-92.)
+ * Note: older versions of the IJG code used a zero offset of _MAXJSAMPLE/2,
+ * rather than _CENTERJSAMPLE, for Cb and Cr.  This gave equal positive and
+ * negative swings for Cb/Cr, but meant that grayscale values (Cb=Cr=0)
+ * were not represented exactly.  Now we sacrifice exact representation of
+ * maximum red and maximum blue in order to get exact grayscales.
+ *
+ * To avoid floating-point arithmetic, we represent the fractional constants
+ * as integers scaled up by 2^16 (about 4 digits precision); we have to divide
+ * the products by 2^16, with appropriate rounding, to get the correct answer.
+ *
+ * For even more speed, we avoid doing any multiplications in the inner loop
+ * by precalculating the constants times R,G,B for all possible values.
+ * For 8-bit samples this is very reasonable (only 256 entries per table);
+ * for 12-bit samples it is still acceptable.  It's not very reasonable for
+ * 16-bit samples, but if you want lossless storage you shouldn't be changing
+ * colorspace anyway.
+ * The _CENTERJSAMPLE offsets and the rounding fudge-factor of 0.5 are included
+ * in the tables to save adding them separately in the inner loop.
+ */
+
+#define SCALEBITS       16      /* speediest right-shift on some machines */
+#define CBCR_OFFSET     ((JLONG)_CENTERJSAMPLE << SCALEBITS)
+#define ONE_HALF        ((JLONG)1 << (SCALEBITS - 1))
+#define FIX(x)          ((JLONG)((x) * (1L << SCALEBITS) + 0.5))
+
+/* We allocate one big table and divide it up into eight parts, instead of
+ * doing eight alloc_small requests.  This lets us use a single table base
+ * address, which can be held in a register in the inner loops on many
+ * machines (more than can hold all eight addresses, anyway).
+ */
+
+#define R_Y_OFF         0                       /* offset to R => Y section */
+#define G_Y_OFF         (1 * (_MAXJSAMPLE + 1)) /* offset to G => Y section */
+#define B_Y_OFF         (2 * (_MAXJSAMPLE + 1)) /* etc. */
+#define R_CB_OFF        (3 * (_MAXJSAMPLE + 1))
+#define G_CB_OFF        (4 * (_MAXJSAMPLE + 1))
+#define B_CB_OFF        (5 * (_MAXJSAMPLE + 1))
+#define R_CR_OFF        B_CB_OFF                /* B=>Cb, R=>Cr are the same */
+#define G_CR_OFF        (6 * (_MAXJSAMPLE + 1))
+#define B_CR_OFF        (7 * (_MAXJSAMPLE + 1))
+#define TABLE_SIZE      (8 * (_MAXJSAMPLE + 1))
+
+/* 12-bit samples use a 16-bit data type, so it is possible to pass
+ * out-of-range sample values (< 0 or > 4095) to jpeg_write_scanlines().
+ * Thus, we mask the incoming 12-bit samples to guard against overrunning
+ * or underrunning the conversion tables.
+ */
+
+#if BITS_IN_JSAMPLE == 12
+#define RANGE_LIMIT(value)  ((value) & 0xFFF)
+#else
+#define RANGE_LIMIT(value)  (value)
+#endif
+
+
+/* Include inline routines for colorspace extensions */
+
+#include "jccolext.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+
+#define RGB_RED  EXT_RGB_RED
+#define RGB_GREEN  EXT_RGB_GREEN
+#define RGB_BLUE  EXT_RGB_BLUE
+#define RGB_PIXELSIZE  EXT_RGB_PIXELSIZE
+#define rgb_ycc_convert_internal  extrgb_ycc_convert_internal
+#define rgb_gray_convert_internal  extrgb_gray_convert_internal
+#define rgb_rgb_convert_internal  extrgb_rgb_convert_internal
+#include "jccolext.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+#undef rgb_rgb_convert_internal
+
+#define RGB_RED  EXT_RGBX_RED
+#define RGB_GREEN  EXT_RGBX_GREEN
+#define RGB_BLUE  EXT_RGBX_BLUE
+#define RGB_PIXELSIZE  EXT_RGBX_PIXELSIZE
+#define rgb_ycc_convert_internal  extrgbx_ycc_convert_internal
+#define rgb_gray_convert_internal  extrgbx_gray_convert_internal
+#define rgb_rgb_convert_internal  extrgbx_rgb_convert_internal
+#include "jccolext.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+#undef rgb_rgb_convert_internal
+
+#define RGB_RED  EXT_BGR_RED
+#define RGB_GREEN  EXT_BGR_GREEN
+#define RGB_BLUE  EXT_BGR_BLUE
+#define RGB_PIXELSIZE  EXT_BGR_PIXELSIZE
+#define rgb_ycc_convert_internal  extbgr_ycc_convert_internal
+#define rgb_gray_convert_internal  extbgr_gray_convert_internal
+#define rgb_rgb_convert_internal  extbgr_rgb_convert_internal
+#include "jccolext.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+#undef rgb_rgb_convert_internal
+
+#define RGB_RED  EXT_BGRX_RED
+#define RGB_GREEN  EXT_BGRX_GREEN
+#define RGB_BLUE  EXT_BGRX_BLUE
+#define RGB_PIXELSIZE  EXT_BGRX_PIXELSIZE
+#define rgb_ycc_convert_internal  extbgrx_ycc_convert_internal
+#define rgb_gray_convert_internal  extbgrx_gray_convert_internal
+#define rgb_rgb_convert_internal  extbgrx_rgb_convert_internal
+#include "jccolext.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+#undef rgb_rgb_convert_internal
+
+#define RGB_RED  EXT_XBGR_RED
+#define RGB_GREEN  EXT_XBGR_GREEN
+#define RGB_BLUE  EXT_XBGR_BLUE
+#define RGB_PIXELSIZE  EXT_XBGR_PIXELSIZE
+#define rgb_ycc_convert_internal  extxbgr_ycc_convert_internal
+#define rgb_gray_convert_internal  extxbgr_gray_convert_internal
+#define rgb_rgb_convert_internal  extxbgr_rgb_convert_internal
+#include "jccolext.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+#undef rgb_rgb_convert_internal
+
+#define RGB_RED  EXT_XRGB_RED
+#define RGB_GREEN  EXT_XRGB_GREEN
+#define RGB_BLUE  EXT_XRGB_BLUE
+#define RGB_PIXELSIZE  EXT_XRGB_PIXELSIZE
+#define rgb_ycc_convert_internal  extxrgb_ycc_convert_internal
+#define rgb_gray_convert_internal  extxrgb_gray_convert_internal
+#define rgb_rgb_convert_internal  extxrgb_rgb_convert_internal
+#include "jccolext.c"
+#undef RGB_RED
+#undef RGB_GREEN
+#undef RGB_BLUE
+#undef RGB_PIXELSIZE
+#undef rgb_ycc_convert_internal
+#undef rgb_gray_convert_internal
+#undef rgb_rgb_convert_internal
+
+
+/*
+ * Initialize for RGB->YCC colorspace conversion.
+ */
+
+METHODDEF(void)
+rgb_ycc_start(j_compress_ptr cinfo)
+{
+#if BITS_IN_JSAMPLE != 16
+  my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert;
+  JLONG *rgb_ycc_tab;
+  JLONG i;
+
+  /* Allocate and fill in the conversion tables. */
+  cconvert->rgb_ycc_tab = rgb_ycc_tab = (JLONG *)
+    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                (TABLE_SIZE * sizeof(JLONG)));
+
+  for (i = 0; i <= _MAXJSAMPLE; i++) {
+    rgb_ycc_tab[i + R_Y_OFF] = FIX(0.29900) * i;
+    rgb_ycc_tab[i + G_Y_OFF] = FIX(0.58700) * i;
+    rgb_ycc_tab[i + B_Y_OFF] = FIX(0.11400) * i   + ONE_HALF;
+    rgb_ycc_tab[i + R_CB_OFF] = (-FIX(0.16874)) * i;
+    rgb_ycc_tab[i + G_CB_OFF] = (-FIX(0.33126)) * i;
+    /* We use a rounding fudge-factor of 0.5-epsilon for Cb and Cr.
+     * This ensures that the maximum output will round to _MAXJSAMPLE
+     * not _MAXJSAMPLE+1, and thus that we don't have to range-limit.
+     */
+    rgb_ycc_tab[i + B_CB_OFF] = FIX(0.50000) * i  + CBCR_OFFSET + ONE_HALF - 1;
+/*  B=>Cb and R=>Cr tables are the same
+    rgb_ycc_tab[i + R_CR_OFF] = FIX(0.50000) * i  + CBCR_OFFSET + ONE_HALF - 1;
+*/
+    rgb_ycc_tab[i + G_CR_OFF] = (-FIX(0.41869)) * i;
+    rgb_ycc_tab[i + B_CR_OFF] = (-FIX(0.08131)) * i;
+  }
+#else
+  ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+#endif
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ */
+
+METHODDEF(void)
+rgb_ycc_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+                _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)
+{
+  switch (cinfo->in_color_space) {
+  case JCS_EXT_RGB:
+    extrgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                num_rows);
+    break;
+  case JCS_EXT_RGBX:
+  case JCS_EXT_RGBA:
+    extrgbx_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  case JCS_EXT_BGR:
+    extbgr_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                num_rows);
+    break;
+  case JCS_EXT_BGRX:
+  case JCS_EXT_BGRA:
+    extbgrx_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  case JCS_EXT_XBGR:
+  case JCS_EXT_ABGR:
+    extxbgr_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  case JCS_EXT_XRGB:
+  case JCS_EXT_ARGB:
+    extxrgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  default:
+    rgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row,
+                             num_rows);
+    break;
+  }
+}
+
+
+/**************** Cases other than RGB -> YCbCr **************/
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ */
+
+METHODDEF(void)
+rgb_gray_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+                 _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)
+{
+  switch (cinfo->in_color_space) {
+  case JCS_EXT_RGB:
+    extrgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  case JCS_EXT_RGBX:
+  case JCS_EXT_RGBA:
+    extrgbx_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                  num_rows);
+    break;
+  case JCS_EXT_BGR:
+    extbgr_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  case JCS_EXT_BGRX:
+  case JCS_EXT_BGRA:
+    extbgrx_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                  num_rows);
+    break;
+  case JCS_EXT_XBGR:
+  case JCS_EXT_ABGR:
+    extxbgr_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                  num_rows);
+    break;
+  case JCS_EXT_XRGB:
+  case JCS_EXT_ARGB:
+    extxrgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                  num_rows);
+    break;
+  default:
+    rgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row,
+                              num_rows);
+    break;
+  }
+}
+
+
+/*
+ * Extended RGB to plain RGB conversion
+ */
+
+METHODDEF(void)
+rgb_rgb_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+                _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)
+{
+  switch (cinfo->in_color_space) {
+  case JCS_EXT_RGB:
+    extrgb_rgb_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                num_rows);
+    break;
+  case JCS_EXT_RGBX:
+  case JCS_EXT_RGBA:
+    extrgbx_rgb_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  case JCS_EXT_BGR:
+    extbgr_rgb_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                num_rows);
+    break;
+  case JCS_EXT_BGRX:
+  case JCS_EXT_BGRA:
+    extbgrx_rgb_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  case JCS_EXT_XBGR:
+  case JCS_EXT_ABGR:
+    extxbgr_rgb_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  case JCS_EXT_XRGB:
+  case JCS_EXT_ARGB:
+    extxrgb_rgb_convert_internal(cinfo, input_buf, output_buf, output_row,
+                                 num_rows);
+    break;
+  default:
+    rgb_rgb_convert_internal(cinfo, input_buf, output_buf, output_row,
+                             num_rows);
+    break;
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles Adobe-style CMYK->YCCK conversion,
+ * where we convert R=1-C, G=1-M, and B=1-Y to YCbCr using the same
+ * conversion as above, while passing K (black) unchanged.
+ * We assume rgb_ycc_start has been called.
+ */
+
+METHODDEF(void)
+cmyk_ycck_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+                  _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)
+{
+#if BITS_IN_JSAMPLE != 16
+  my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert;
+  register int r, g, b;
+  register JLONG *ctab = cconvert->rgb_ycc_tab;
+  register _JSAMPROW inptr;
+  register _JSAMPROW outptr0, outptr1, outptr2, outptr3;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr0 = output_buf[0][output_row];
+    outptr1 = output_buf[1][output_row];
+    outptr2 = output_buf[2][output_row];
+    outptr3 = output_buf[3][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      r = _MAXJSAMPLE - RANGE_LIMIT(inptr[0]);
+      g = _MAXJSAMPLE - RANGE_LIMIT(inptr[1]);
+      b = _MAXJSAMPLE - RANGE_LIMIT(inptr[2]);
+      /* K passes through as-is */
+      outptr3[col] = inptr[3];
+      inptr += 4;
+      /* If the inputs are 0.._MAXJSAMPLE, the outputs of these equations
+       * must be too; we do not need an explicit range-limiting operation.
+       * Hence the value being shifted is never negative, and we don't
+       * need the general RIGHT_SHIFT macro.
+       */
+      /* Y */
+      outptr0[col] = (_JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] +
+                                 ctab[b + B_Y_OFF]) >> SCALEBITS);
+      /* Cb */
+      outptr1[col] = (_JSAMPLE)((ctab[r + R_CB_OFF] + ctab[g + G_CB_OFF] +
+                                 ctab[b + B_CB_OFF]) >> SCALEBITS);
+      /* Cr */
+      outptr2[col] = (_JSAMPLE)((ctab[r + R_CR_OFF] + ctab[g + G_CR_OFF] +
+                                 ctab[b + B_CR_OFF]) >> SCALEBITS);
+    }
+  }
+#else
+  ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+#endif
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles grayscale output with no conversion.
+ * The source can be either plain grayscale or YCbCr (since Y == gray).
+ */
+
+METHODDEF(void)
+grayscale_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+                  _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)
+{
+  register _JSAMPROW inptr;
+  register _JSAMPROW outptr;
+  register JDIMENSION col;
+  JDIMENSION num_cols = cinfo->image_width;
+  int instride = cinfo->input_components;
+
+  while (--num_rows >= 0) {
+    inptr = *input_buf++;
+    outptr = output_buf[0][output_row];
+    output_row++;
+    for (col = 0; col < num_cols; col++) {
+      outptr[col] = inptr[0];
+      inptr += instride;
+    }
+  }
+}
+
+
+/*
+ * Convert some rows of samples to the JPEG colorspace.
+ * This version handles multi-component colorspaces without conversion.
+ * We assume input_components == num_components.
+ */
+
+METHODDEF(void)
+null_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf,
+             _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows)
+{
+  register _JSAMPROW inptr;
+  register _JSAMPROW outptr, outptr0, outptr1, outptr2, outptr3;
+  register JDIMENSION col;
+  register int ci;
+  int nc = cinfo->num_components;
+  JDIMENSION num_cols = cinfo->image_width;
+
+  if (nc == 3) {
+    while (--num_rows >= 0) {
+      inptr = *input_buf++;
+      outptr0 = output_buf[0][output_row];
+      outptr1 = output_buf[1][output_row];
+      outptr2 = output_buf[2][output_row];
+      output_row++;
+      for (col = 0; col < num_cols; col++) {
+        outptr0[col] = *inptr++;
+        outptr1[col] = *inptr++;
+        outptr2[col] = *inptr++;
+      }
+    }
+  } else if (nc == 4) {
+    while (--num_rows >= 0) {
+      inptr = *input_buf++;
+      outptr0 = output_buf[0][output_row];
+      outptr1 = output_buf[1][output_row];
+      outptr2 = output_buf[2][output_row];
+      outptr3 = output_buf[3][output_row];
+      output_row++;
+      for (col = 0; col < num_cols; col++) {
+        outptr0[col] = *inptr++;
+        outptr1[col] = *inptr++;
+        outptr2[col] = *inptr++;
+        outptr3[col] = *inptr++;
+      }
+    }
+  } else {
+    while (--num_rows >= 0) {
+      /* It seems fastest to make a separate pass for each component. */
+      for (ci = 0; ci < nc; ci++) {
+        inptr = *input_buf;
+        outptr = output_buf[ci][output_row];
+        for (col = 0; col < num_cols; col++) {
+          outptr[col] = inptr[ci];
+          inptr += nc;
+        }
+      }
+      input_buf++;
+      output_row++;
+    }
+  }
+}
+
+
+/*
+ * Empty method for start_pass.
+ */
+
+METHODDEF(void)
+null_method(j_compress_ptr cinfo)
+{
+  /* no work needed */
+}
+
+
+/*
+ * Module initialization routine for input colorspace conversion.
+ */
+
+GLOBAL(void)
+_jinit_color_converter(j_compress_ptr cinfo)
+{
+  my_cconvert_ptr cconvert;
+
+  if (cinfo->data_precision != BITS_IN_JSAMPLE)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  cconvert = (my_cconvert_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                sizeof(my_color_converter));
+  cinfo->cconvert = (struct jpeg_color_converter *)cconvert;
+  /* set start_pass to null method until we find out differently */
+  cconvert->pub.start_pass = null_method;
+
+  /* Make sure input_components agrees with in_color_space */
+  switch (cinfo->in_color_space) {
+  case JCS_GRAYSCALE:
+    if (cinfo->input_components != 1)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  case JCS_RGB:
+  case JCS_EXT_RGB:
+  case JCS_EXT_RGBX:
+  case JCS_EXT_BGR:
+  case JCS_EXT_BGRX:
+  case JCS_EXT_XBGR:
+  case JCS_EXT_XRGB:
+  case JCS_EXT_RGBA:
+  case JCS_EXT_BGRA:
+  case JCS_EXT_ABGR:
+  case JCS_EXT_ARGB:
+    if (cinfo->input_components != rgb_pixelsize[cinfo->in_color_space])
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  case JCS_YCbCr:
+    if (cinfo->input_components != 3)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  case JCS_CMYK:
+  case JCS_YCCK:
+    if (cinfo->input_components != 4)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+
+  default:                      /* JCS_UNKNOWN can be anything */
+    if (cinfo->input_components < 1)
+      ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE);
+    break;
+  }
+
+  /* Check num_components, set conversion method based on requested space.
+   * NOTE: We do not allow any lossy color conversion algorithms in lossless
+   * mode.
+   */
+  switch (cinfo->jpeg_color_space) {
+  case JCS_GRAYSCALE:
+    if (cinfo->master->lossless &&
+        cinfo->in_color_space != cinfo->jpeg_color_space)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    if (cinfo->num_components != 1)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space == JCS_GRAYSCALE)
+      cconvert->pub._color_convert = grayscale_convert;
+    else if (IsExtRGB(cinfo->in_color_space)) {
+#ifdef WITH_SIMD
+      if (jsimd_can_rgb_gray())
+        cconvert->pub._color_convert = jsimd_rgb_gray_convert;
+      else
+#endif
+      {
+        cconvert->pub.start_pass = rgb_ycc_start;
+        cconvert->pub._color_convert = rgb_gray_convert;
+      }
+    } else if (cinfo->in_color_space == JCS_YCbCr)
+      cconvert->pub._color_convert = grayscale_convert;
+    else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  case JCS_RGB:
+    if (cinfo->master->lossless && !IsExtRGB(cinfo->in_color_space))
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    if (cinfo->num_components != 3)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (rgb_red[cinfo->in_color_space] == 0 &&
+        rgb_green[cinfo->in_color_space] == 1 &&
+        rgb_blue[cinfo->in_color_space] == 2 &&
+        rgb_pixelsize[cinfo->in_color_space] == 3) {
+#if defined(WITH_SIMD) && defined(__mips__)
+      if (jsimd_c_can_null_convert())
+        cconvert->pub._color_convert = jsimd_c_null_convert;
+      else
+#endif
+        cconvert->pub._color_convert = null_convert;
+    } else if (IsExtRGB(cinfo->in_color_space))
+      cconvert->pub._color_convert = rgb_rgb_convert;
+    else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  case JCS_YCbCr:
+    if (cinfo->master->lossless &&
+        cinfo->in_color_space != cinfo->jpeg_color_space)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    if (cinfo->num_components != 3)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (IsExtRGB(cinfo->in_color_space)) {
+#ifdef WITH_SIMD
+      if (jsimd_can_rgb_ycc())
+        cconvert->pub._color_convert = jsimd_rgb_ycc_convert;
+      else
+#endif
+      {
+        cconvert->pub.start_pass = rgb_ycc_start;
+        cconvert->pub._color_convert = rgb_ycc_convert;
+      }
+    } else if (cinfo->in_color_space == JCS_YCbCr) {
+#if defined(WITH_SIMD) && defined(__mips__)
+      if (jsimd_c_can_null_convert())
+        cconvert->pub._color_convert = jsimd_c_null_convert;
+      else
+#endif
+        cconvert->pub._color_convert = null_convert;
+    } else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  case JCS_CMYK:
+    if (cinfo->master->lossless &&
+        cinfo->in_color_space != cinfo->jpeg_color_space)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    if (cinfo->num_components != 4)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space == JCS_CMYK) {
+#if defined(WITH_SIMD) && defined(__mips__)
+      if (jsimd_c_can_null_convert())
+        cconvert->pub._color_convert = jsimd_c_null_convert;
+      else
+#endif
+        cconvert->pub._color_convert = null_convert;
+    } else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  case JCS_YCCK:
+    if (cinfo->master->lossless &&
+        cinfo->in_color_space != cinfo->jpeg_color_space)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    if (cinfo->num_components != 4)
+      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
+    if (cinfo->in_color_space == JCS_CMYK) {
+      cconvert->pub.start_pass = rgb_ycc_start;
+      cconvert->pub._color_convert = cmyk_ycck_convert;
+    } else if (cinfo->in_color_space == JCS_YCCK) {
+#if defined(WITH_SIMD) && defined(__mips__)
+      if (jsimd_c_can_null_convert())
+        cconvert->pub._color_convert = jsimd_c_null_convert;
+      else
+#endif
+        cconvert->pub._color_convert = null_convert;
+    } else
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+    break;
+
+  default:                      /* allow null conversion of JCS_UNKNOWN */
+    if (cinfo->jpeg_color_space != cinfo->in_color_space ||
+        cinfo->num_components != cinfo->input_components)
+      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
+#if defined(WITH_SIMD) && defined(__mips__)
+    if (jsimd_c_can_null_convert())
+      cconvert->pub._color_convert = jsimd_c_null_convert;
+    else
+#endif
+      cconvert->pub._color_convert = null_convert;
+    break;
+  }
+}
+
+#endif /* BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) */
@@ -0,0 +1,748 @@
+/*
+ * jcdctmgr.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 1999-2006, MIYASAKA Masaru.
+ * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright (C) 2011, 2014-2015, 2022, 2024, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains the forward-DCT management logic.
+ * This code selects a particular DCT implementation to be used,
+ * and it performs related housekeeping chores including coefficient
+ * quantization.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"               /* Private declarations for DCT subsystem */
+#include "jsimddct.h"
+
+
+/* Private subobject for this module */
+
+typedef void (*forward_DCT_method_ptr) (DCTELEM *data);
+typedef void (*float_DCT_method_ptr) (FAST_FLOAT *data);
+
+typedef void (*convsamp_method_ptr) (_JSAMPARRAY sample_data,
+                                     JDIMENSION start_col,
+                                     DCTELEM *workspace);
+typedef void (*float_convsamp_method_ptr) (_JSAMPARRAY sample_data,
+                                           JDIMENSION start_col,
+                                           FAST_FLOAT *workspace);
+
+typedef void (*quantize_method_ptr) (JCOEFPTR coef_block, DCTELEM *divisors,
+                                     DCTELEM *workspace);
+typedef void (*float_quantize_method_ptr) (JCOEFPTR coef_block,
+                                           FAST_FLOAT *divisors,
+                                           FAST_FLOAT *workspace);
+
+METHODDEF(void) quantize(JCOEFPTR, DCTELEM *, DCTELEM *);
+
+typedef struct {
+  struct jpeg_forward_dct pub;  /* public fields */
+
+  /* Pointer to the DCT routine actually in use */
+  forward_DCT_method_ptr dct;
+  convsamp_method_ptr convsamp;
+  quantize_method_ptr quantize;
+
+  /* The actual post-DCT divisors --- not identical to the quant table
+   * entries, because of scaling (especially for an unnormalized DCT).
+   * Each table is given in normal array order.
+   */
+  DCTELEM *divisors[NUM_QUANT_TBLS];
+
+  /* work area for FDCT subroutine */
+  DCTELEM *workspace;
+
+#ifdef DCT_FLOAT_SUPPORTED
+  /* Same as above for the floating-point case. */
+  float_DCT_method_ptr float_dct;
+  float_convsamp_method_ptr float_convsamp;
+  float_quantize_method_ptr float_quantize;
+  FAST_FLOAT *float_divisors[NUM_QUANT_TBLS];
+  FAST_FLOAT *float_workspace;
+#endif
+} my_fdct_controller;
+
+typedef my_fdct_controller *my_fdct_ptr;
+
+
+#if BITS_IN_JSAMPLE == 8
+
+/*
+ * Find the highest bit in an integer through binary search.
+ */
+
+LOCAL(int)
+flss(UINT16 val)
+{
+  int bit;
+
+  bit = 16;
+
+  if (!val)
+    return 0;
+
+  if (!(val & 0xff00)) {
+    bit -= 8;
+    val <<= 8;
+  }
+  if (!(val & 0xf000)) {
+    bit -= 4;
+    val <<= 4;
+  }
+  if (!(val & 0xc000)) {
+    bit -= 2;
+    val <<= 2;
+  }
+  if (!(val & 0x8000)) {
+    bit -= 1;
+    val <<= 1;
+  }
+
+  return bit;
+}
+
+
+/*
+ * Compute values to do a division using reciprocal.
+ *
+ * This implementation is based on an algorithm described in
+ *   "Optimizing subroutines in assembly language:
+ *   An optimization guide for x86 platforms" (https://agner.org/optimize).
+ * More information about the basic algorithm can be found in
+ * the paper "Integer Division Using Reciprocals" by Robert Alverson.
+ *
+ * The basic idea is to replace x/d by x * d^-1. In order to store
+ * d^-1 with enough precision we shift it left a few places. It turns
+ * out that this algoright gives just enough precision, and also fits
+ * into DCTELEM:
+ *
+ *   b = (the number of significant bits in divisor) - 1
+ *   r = (word size) + b
+ *   f = 2^r / divisor
+ *
+ * f will not be an integer for most cases, so we need to compensate
+ * for the rounding error introduced:
+ *
+ *   no fractional part:
+ *
+ *       result = input >> r
+ *
+ *   fractional part of f < 0.5:
+ *
+ *       round f down to nearest integer
+ *       result = ((input + 1) * f) >> r
+ *
+ *   fractional part of f > 0.5:
+ *
+ *       round f up to nearest integer
+ *       result = (input * f) >> r
+ *
+ * This is the original algorithm that gives truncated results. But we
+ * want properly rounded results, so we replace "input" with
+ * "input + divisor/2".
+ *
+ * In order to allow SIMD implementations we also tweak the values to
+ * allow the same calculation to be made at all times:
+ *
+ *   dctbl[0] = f rounded to nearest integer
+ *   dctbl[1] = divisor / 2 (+ 1 if fractional part of f < 0.5)
+ *   dctbl[2] = 1 << ((word size) * 2 - r)
+ *   dctbl[3] = r - (word size)
+ *
+ * dctbl[2] is for stupid instruction sets where the shift operation
+ * isn't member wise (e.g. MMX).
+ *
+ * The reason dctbl[2] and dctbl[3] reduce the shift with (word size)
+ * is that most SIMD implementations have a "multiply and store top
+ * half" operation.
+ *
+ * Lastly, we store each of the values in their own table instead
+ * of in a consecutive manner, yet again in order to allow SIMD
+ * routines.
+ */
+
+LOCAL(int)
+compute_reciprocal(UINT16 divisor, DCTELEM *dtbl)
+{
+  UDCTELEM2 fq, fr;
+  UDCTELEM c;
+  int b, r;
+
+  if (divisor == 1) {
+    /* divisor == 1 means unquantized, so these reciprocal/correction/shift
+     * values will cause the C quantization algorithm to act like the
+     * identity function.  Since only the C quantization algorithm is used in
+     * these cases, the scale value is irrelevant.
+     */
+    dtbl[DCTSIZE2 * 0] = (DCTELEM)1;                        /* reciprocal */
+    dtbl[DCTSIZE2 * 1] = (DCTELEM)0;                        /* correction */
+    dtbl[DCTSIZE2 * 2] = (DCTELEM)1;                        /* scale */
+    dtbl[DCTSIZE2 * 3] = -(DCTELEM)(sizeof(DCTELEM) * 8);   /* shift */
+    return 0;
+  }
+
+  b = flss(divisor) - 1;
+  r  = sizeof(DCTELEM) * 8 + b;
+
+  fq = ((UDCTELEM2)1 << r) / divisor;
+  fr = ((UDCTELEM2)1 << r) % divisor;
+
+  c = divisor / 2;                      /* for rounding */
+
+  if (fr == 0) {                        /* divisor is power of two */
+    /* fq will be one bit too large to fit in DCTELEM, so adjust */
+    fq >>= 1;
+    r--;
+  } else if (fr <= (divisor / 2U)) {    /* fractional part is < 0.5 */
+    c++;
+  } else {                              /* fractional part is > 0.5 */
+    fq++;
+  }
+
+  dtbl[DCTSIZE2 * 0] = (DCTELEM)fq;     /* reciprocal */
+  dtbl[DCTSIZE2 * 1] = (DCTELEM)c;      /* correction + roundfactor */
+#ifdef WITH_SIMD
+  dtbl[DCTSIZE2 * 2] = (DCTELEM)(1 << (sizeof(DCTELEM) * 8 * 2 - r)); /* scale */
+#else
+  dtbl[DCTSIZE2 * 2] = 1;
+#endif
+  dtbl[DCTSIZE2 * 3] = (DCTELEM)r - sizeof(DCTELEM) * 8; /* shift */
+
+  if (r <= 16) return 0;
+  else return 1;
+}
+
+#endif
+
+
+/*
+ * Initialize for a processing pass.
+ * Verify that all referenced Q-tables are present, and set up
+ * the divisor table for each one.
+ * In the current implementation, DCT of all components is done during
+ * the first pass, even if only some components will be output in the
+ * first scan.  Hence all components should be examined here.
+ */
+
+METHODDEF(void)
+start_pass_fdctmgr(j_compress_ptr cinfo)
+{
+  my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct;
+  int ci, qtblno, i;
+  jpeg_component_info *compptr;
+  JQUANT_TBL *qtbl;
+  DCTELEM *dtbl;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    qtblno = compptr->quant_tbl_no;
+    /* Make sure specified quantization table is present */
+    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
+        cinfo->quant_tbl_ptrs[qtblno] == NULL)
+      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
+    qtbl = cinfo->quant_tbl_ptrs[qtblno];
+    /* Compute divisors for this quant table */
+    /* We may do this more than once for same table, but it's not a big deal */
+    switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+    case JDCT_ISLOW:
+      /* For LL&M IDCT method, divisors are equal to raw quantization
+       * coefficients multiplied by 8 (to counteract scaling).
+       */
+      if (fdct->divisors[qtblno] == NULL) {
+        fdct->divisors[qtblno] = (DCTELEM *)
+          (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                      (DCTSIZE2 * 4) * sizeof(DCTELEM));
+      }
+      dtbl = fdct->divisors[qtblno];
+      for (i = 0; i < DCTSIZE2; i++) {
+#if BITS_IN_JSAMPLE == 8
+#ifdef WITH_SIMD
+        if (!compute_reciprocal(qtbl->quantval[i] << 3, &dtbl[i]) &&
+            fdct->quantize == jsimd_quantize)
+          fdct->quantize = quantize;
+#else
+        compute_reciprocal(qtbl->quantval[i] << 3, &dtbl[i]);
+#endif
+#else
+        dtbl[i] = ((DCTELEM)qtbl->quantval[i]) << 3;
+#endif
+      }
+      break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+    case JDCT_IFAST:
+      {
+        /* For AA&N IDCT method, divisors are equal to quantization
+         * coefficients scaled by scalefactor[row]*scalefactor[col], where
+         *   scalefactor[0] = 1
+         *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+         * We apply a further scale factor of 8.
+         */
+#define CONST_BITS  14
+        static const INT16 aanscales[DCTSIZE2] = {
+          /* precomputed values scaled up by 14 bits */
+          16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+          22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
+          21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
+          19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
+          16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+          12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
+           8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
+           4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
+        };
+        SHIFT_TEMPS
+
+        if (fdct->divisors[qtblno] == NULL) {
+          fdct->divisors[qtblno] = (DCTELEM *)
+            (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                        (DCTSIZE2 * 4) * sizeof(DCTELEM));
+        }
+        dtbl = fdct->divisors[qtblno];
+        for (i = 0; i < DCTSIZE2; i++) {
+#if BITS_IN_JSAMPLE == 8
+#ifdef WITH_SIMD
+          if (!compute_reciprocal(
+                DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i],
+                                      (JLONG)aanscales[i]),
+                        CONST_BITS - 3), &dtbl[i]) &&
+              fdct->quantize == jsimd_quantize)
+            fdct->quantize = quantize;
+#else
+          compute_reciprocal(
+            DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i],
+                                  (JLONG)aanscales[i]),
+                    CONST_BITS-3), &dtbl[i]);
+#endif
+#else
+          dtbl[i] = (DCTELEM)
+            DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i],
+                                  (JLONG)aanscales[i]),
+                    CONST_BITS - 3);
+#endif
+        }
+      }
+      break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+    case JDCT_FLOAT:
+      {
+        /* For float AA&N IDCT method, divisors are equal to quantization
+         * coefficients scaled by scalefactor[row]*scalefactor[col], where
+         *   scalefactor[0] = 1
+         *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+         * We apply a further scale factor of 8.
+         * What's actually stored is 1/divisor so that the inner loop can
+         * use a multiplication rather than a division.
+         */
+        FAST_FLOAT *fdtbl;
+        int row, col;
+        static const double aanscalefactor[DCTSIZE] = {
+          1.0, 1.387039845, 1.306562965, 1.175875602,
+          1.0, 0.785694958, 0.541196100, 0.275899379
+        };
+
+        if (fdct->float_divisors[qtblno] == NULL) {
+          fdct->float_divisors[qtblno] = (FAST_FLOAT *)
+            (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                        DCTSIZE2 * sizeof(FAST_FLOAT));
+        }
+        fdtbl = fdct->float_divisors[qtblno];
+        i = 0;
+        for (row = 0; row < DCTSIZE; row++) {
+          for (col = 0; col < DCTSIZE; col++) {
+            fdtbl[i] = (FAST_FLOAT)
+              (1.0 / (((double)qtbl->quantval[i] *
+                       aanscalefactor[row] * aanscalefactor[col] * 8.0)));
+            i++;
+          }
+        }
+      }
+      break;
+#endif
+    default:
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+      break;
+    }
+  }
+}
+
+
+/*
+ * Load data into workspace, applying unsigned->signed conversion.
+ */
+
+METHODDEF(void)
+convsamp(_JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace)
+{
+  register DCTELEM *workspaceptr;
+  register _JSAMPROW elemptr;
+  register int elemr;
+
+  workspaceptr = workspace;
+  for (elemr = 0; elemr < DCTSIZE; elemr++) {
+    elemptr = sample_data[elemr] + start_col;
+
+#if DCTSIZE == 8                /* unroll the inner loop */
+    *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+    *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+#else
+    {
+      register int elemc;
+      for (elemc = DCTSIZE; elemc > 0; elemc--)
+        *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE;
+    }
+#endif
+  }
+}
+
+
+/*
+ * Quantize/descale the coefficients, and store into coef_blocks[].
+ */
+
+METHODDEF(void)
+quantize(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace)
+{
+  int i;
+  DCTELEM temp;
+  JCOEFPTR output_ptr = coef_block;
+
+#if BITS_IN_JSAMPLE == 8
+
+  UDCTELEM recip, corr;
+  int shift;
+  UDCTELEM2 product;
+
+  for (i = 0; i < DCTSIZE2; i++) {
+    temp = workspace[i];
+    recip = divisors[i + DCTSIZE2 * 0];
+    corr =  divisors[i + DCTSIZE2 * 1];
+    shift = divisors[i + DCTSIZE2 * 3];
+
+    if (temp < 0) {
+      temp = -temp;
+      product = (UDCTELEM2)(temp + corr) * recip;
+      product >>= shift + sizeof(DCTELEM) * 8;
+      temp = (DCTELEM)product;
+      temp = -temp;
+    } else {
+      product = (UDCTELEM2)(temp + corr) * recip;
+      product >>= shift + sizeof(DCTELEM) * 8;
+      temp = (DCTELEM)product;
+    }
+    output_ptr[i] = (JCOEF)temp;
+  }
+
+#else
+
+  register DCTELEM qval;
+
+  for (i = 0; i < DCTSIZE2; i++) {
+    qval = divisors[i];
+    temp = workspace[i];
+    /* Divide the coefficient value by qval, ensuring proper rounding.
+     * Since C does not specify the direction of rounding for negative
+     * quotients, we have to force the dividend positive for portability.
+     *
+     * In most files, at least half of the output values will be zero
+     * (at default quantization settings, more like three-quarters...)
+     * so we should ensure that this case is fast.  On many machines,
+     * a comparison is enough cheaper than a divide to make a special test
+     * a win.  Since both inputs will be nonnegative, we need only test
+     * for a < b to discover whether a/b is 0.
+     * If your machine's division is fast enough, define FAST_DIVIDE.
+     */
+#ifdef FAST_DIVIDE
+#define DIVIDE_BY(a, b)  a /= b
+#else
+#define DIVIDE_BY(a, b)  if (a >= b) a /= b;  else a = 0
+#endif
+    if (temp < 0) {
+      temp = -temp;
+      temp += qval >> 1;        /* for rounding */
+      DIVIDE_BY(temp, qval);
+      temp = -temp;
+    } else {
+      temp += qval >> 1;        /* for rounding */
+      DIVIDE_BY(temp, qval);
+    }
+    output_ptr[i] = (JCOEF)temp;
+  }
+
+#endif
+
+}
+
+
+/*
+ * Perform forward DCT on one or more blocks of a component.
+ *
+ * The input samples are taken from the sample_data[] array starting at
+ * position start_row/start_col, and moving to the right for any additional
+ * blocks. The quantized coefficients are returned in coef_blocks[].
+ */
+
+METHODDEF(void)
+forward_DCT(j_compress_ptr cinfo, jpeg_component_info *compptr,
+            _JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+            JDIMENSION start_row, JDIMENSION start_col, JDIMENSION num_blocks)
+/* This version is used for integer DCT implementations. */
+{
+  /* This routine is heavily used, so it's worth coding it tightly. */
+  my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct;
+  DCTELEM *divisors = fdct->divisors[compptr->quant_tbl_no];
+  DCTELEM *workspace;
+  JDIMENSION bi;
+
+  /* Make sure the compiler doesn't look up these every pass */
+  forward_DCT_method_ptr do_dct = fdct->dct;
+  convsamp_method_ptr do_convsamp = fdct->convsamp;
+  quantize_method_ptr do_quantize = fdct->quantize;
+  workspace = fdct->workspace;
+
+  sample_data += start_row;     /* fold in the vertical offset once */
+
+  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
+    /* Load data into workspace, applying unsigned->signed conversion */
+    (*do_convsamp) (sample_data, start_col, workspace);
+
+    /* Perform the DCT */
+    (*do_dct) (workspace);
+
+    /* Quantize/descale the coefficients, and store into coef_blocks[] */
+    (*do_quantize) (coef_blocks[bi], divisors, workspace);
+  }
+}
+
+
+#ifdef DCT_FLOAT_SUPPORTED
+
+METHODDEF(void)
+convsamp_float(_JSAMPARRAY sample_data, JDIMENSION start_col,
+               FAST_FLOAT *workspace)
+{
+  register FAST_FLOAT *workspaceptr;
+  register _JSAMPROW elemptr;
+  register int elemr;
+
+  workspaceptr = workspace;
+  for (elemr = 0; elemr < DCTSIZE; elemr++) {
+    elemptr = sample_data[elemr] + start_col;
+#if DCTSIZE == 8                /* unroll the inner loop */
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+#else
+    {
+      register int elemc;
+      for (elemc = DCTSIZE; elemc > 0; elemc--)
+        *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE);
+    }
+#endif
+  }
+}
+
+
+METHODDEF(void)
+quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors,
+               FAST_FLOAT *workspace)
+{
+  register FAST_FLOAT temp;
+  register int i;
+  register JCOEFPTR output_ptr = coef_block;
+
+  for (i = 0; i < DCTSIZE2; i++) {
+    /* Apply the quantization and scaling factor */
+    temp = workspace[i] * divisors[i];
+
+    /* Round to nearest integer.
+     * Since C does not specify the direction of rounding for negative
+     * quotients, we have to force the dividend positive for portability.
+     * The maximum coefficient size is +-16K (for 12-bit data), so this
+     * code should work for either 16-bit or 32-bit ints.
+     */
+    output_ptr[i] = (JCOEF)((int)(temp + (FAST_FLOAT)16384.5) - 16384);
+  }
+}
+
+
+METHODDEF(void)
+forward_DCT_float(j_compress_ptr cinfo, jpeg_component_info *compptr,
+                  _JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
+                  JDIMENSION start_row, JDIMENSION start_col,
+                  JDIMENSION num_blocks)
+/* This version is used for floating-point DCT implementations. */
+{
+  /* This routine is heavily used, so it's worth coding it tightly. */
+  my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct;
+  FAST_FLOAT *divisors = fdct->float_divisors[compptr->quant_tbl_no];
+  FAST_FLOAT *workspace;
+  JDIMENSION bi;
+
+
+  /* Make sure the compiler doesn't look up these every pass */
+  float_DCT_method_ptr do_dct = fdct->float_dct;
+  float_convsamp_method_ptr do_convsamp = fdct->float_convsamp;
+  float_quantize_method_ptr do_quantize = fdct->float_quantize;
+  workspace = fdct->float_workspace;
+
+  sample_data += start_row;     /* fold in the vertical offset once */
+
+  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
+    /* Load data into workspace, applying unsigned->signed conversion */
+    (*do_convsamp) (sample_data, start_col, workspace);
+
+    /* Perform the DCT */
+    (*do_dct) (workspace);
+
+    /* Quantize/descale the coefficients, and store into coef_blocks[] */
+    (*do_quantize) (coef_blocks[bi], divisors, workspace);
+  }
+}
+
+#endif /* DCT_FLOAT_SUPPORTED */
+
+
+/*
+ * Initialize FDCT manager.
+ */
+
+GLOBAL(void)
+_jinit_forward_dct(j_compress_ptr cinfo)
+{
+  my_fdct_ptr fdct;
+  int i;
+
+  if (cinfo->data_precision != BITS_IN_JSAMPLE)
+    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
+
+  fdct = (my_fdct_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                sizeof(my_fdct_controller));
+  cinfo->fdct = (struct jpeg_forward_dct *)fdct;
+  fdct->pub.start_pass = start_pass_fdctmgr;
+
+  /* First determine the DCT... */
+  switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+  case JDCT_ISLOW:
+    fdct->pub._forward_DCT = forward_DCT;
+#ifdef WITH_SIMD
+    if (jsimd_can_fdct_islow())
+      fdct->dct = jsimd_fdct_islow;
+    else
+#endif
+      fdct->dct = _jpeg_fdct_islow;
+    break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+  case JDCT_IFAST:
+    fdct->pub._forward_DCT = forward_DCT;
+#ifdef WITH_SIMD
+    if (jsimd_can_fdct_ifast())
+      fdct->dct = jsimd_fdct_ifast;
+    else
+#endif
+      fdct->dct = _jpeg_fdct_ifast;
+    break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+  case JDCT_FLOAT:
+    fdct->pub._forward_DCT = forward_DCT_float;
+#ifdef WITH_SIMD
+    if (jsimd_can_fdct_float())
+      fdct->float_dct = jsimd_fdct_float;
+    else
+#endif
+      fdct->float_dct = jpeg_fdct_float;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+    break;
+  }
+
+  /* ...then the supporting stages. */
+  switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+  case JDCT_ISLOW:
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+  case JDCT_IFAST:
+#endif
+#if defined(DCT_ISLOW_SUPPORTED) || defined(DCT_IFAST_SUPPORTED)
+#ifdef WITH_SIMD
+    if (jsimd_can_convsamp())
+      fdct->convsamp = jsimd_convsamp;
+    else
+#endif
+      fdct->convsamp = convsamp;
+#ifdef WITH_SIMD
+    if (jsimd_can_quantize())
+      fdct->quantize = jsimd_quantize;
+    else
+#endif
+      fdct->quantize = quantize;
+    break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+  case JDCT_FLOAT:
+#ifdef WITH_SIMD
+    if (jsimd_can_convsamp_float())
+      fdct->float_convsamp = jsimd_convsamp_float;
+    else
+#endif
+      fdct->float_convsamp = convsamp_float;
+#ifdef WITH_SIMD
+    if (jsimd_can_quantize_float())
+      fdct->float_quantize = jsimd_quantize_float;
+    else
+#endif
+      fdct->float_quantize = quantize_float;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_NOT_COMPILED);
+    break;
+  }
+
+  /* Allocate workspace memory */
+#ifdef DCT_FLOAT_SUPPORTED
+  if (cinfo->dct_method == JDCT_FLOAT)
+    fdct->float_workspace = (FAST_FLOAT *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                  sizeof(FAST_FLOAT) * DCTSIZE2);
+  else
+#endif
+    fdct->workspace = (DCTELEM *)
+      (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                  sizeof(DCTELEM) * DCTSIZE2);
+
+  /* Mark divisor tables unallocated */
+  for (i = 0; i < NUM_QUANT_TBLS; i++) {
+    fdct->divisors[i] = NULL;
+#ifdef DCT_FLOAT_SUPPORTED
+    fdct->float_divisors[i] = NULL;
+#endif
+  }
+}
@@ -0,0 +1,411 @@
+/*
+ * jcdiffct.c
+ *
+ * This file was part of the Independent JPEG Group's software:
+ * Copyright (C) 1994-1997, Thomas G. Lane.
+ * Lossless JPEG Modifications:
+ * Copyright (C) 1999, Ken Murchison.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2022, D. R. Commander.
+ * For conditions of distribution and use, see the accompanying README.ijg
+ * file.
+ *
+ * This file contains the difference buffer controller for compression.
+ * This controller is the top level of the lossless JPEG compressor proper.
+ * The difference buffer lies between the prediction/differencing and entropy
+ * encoding steps.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jlossls.h"            /* Private declarations for lossless codec */
+
+
+#ifdef C_LOSSLESS_SUPPORTED
+
+/* We use a full-image sample buffer when doing Huffman optimization,
+ * and also for writing multiple-scan JPEG files.  In all cases, the
+ * full-image buffer is filled during the first pass, and the scaling,
+ * prediction and differencing steps are run during subsequent passes.
+ */
+#ifdef ENTROPY_OPT_SUPPORTED
+#define FULL_SAMP_BUFFER_SUPPORTED
+#else
+#ifdef C_MULTISCAN_FILES_SUPPORTED
+#define FULL_SAMP_BUFFER_SUPPORTED
+#endif
+#endif
+
+
+/* Private buffer controller object */
+
+typedef struct {
+  struct jpeg_c_coef_controller pub; /* public fields */
+
+  JDIMENSION iMCU_row_num;      /* iMCU row # within image */
+  JDIMENSION mcu_ctr;           /* counts MCUs processed in current row */
+  int MCU_vert_offset;          /* counts MCU rows within iMCU row */
+  int MCU_rows_per_iMCU_row;    /* number of such rows needed */
+
+  _JSAMPROW cur_row[MAX_COMPONENTS];    /* row of point-transformed samples */
+  _JSAMPROW prev_row[MAX_COMPONENTS];   /* previous row of Pt'd samples */
+  JDIFFARRAY diff_buf[MAX_COMPONENTS];  /* iMCU row of differences */
+
+  /* In multi-pass modes, we need a virtual sample array for each component. */
+  jvirt_sarray_ptr whole_image[MAX_COMPONENTS];
+} my_diff_controller;
+
+typedef my_diff_controller *my_diff_ptr;
+
+
+/* Forward declarations */
+METHODDEF(boolean) compress_data(j_compress_ptr cinfo, _JSAMPIMAGE input_buf);
+#ifdef FULL_SAMP_BUFFER_SUPPORTED
+METHODDEF(boolean) compress_first_pass(j_compress_ptr cinfo,
+                                       _JSAMPIMAGE input_buf);
+METHODDEF(boolean) compress_output(j_compress_ptr cinfo,
+                                   _JSAMPIMAGE input_buf);
+#endif
+
+
+LOCAL(void)
+start_iMCU_row(j_compress_ptr cinfo)
+/* Reset within-iMCU-row counters for a new row */
+{
+  my_diff_ptr diff = (my_diff_ptr)cinfo->coef;
+
+  /* In an interleaved scan, an MCU row is the same as an iMCU row.
+   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
+   * But at the bottom of the image, process only what's left.
+   */
+  if (cinfo->comps_in_scan > 1) {
+    diff->MCU_rows_per_iMCU_row = 1;
+  } else {
+    if (diff->iMCU_row_num < (cinfo->total_iMCU_rows-1))
+      diff->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
+    else
+      diff->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
+  }
+
+  diff->mcu_ctr = 0;
+  diff->MCU_vert_offset = 0;
+}
+
+
+/*
+ * Initialize for a processing pass.
+ */
+
+METHODDEF(void)
+start_pass_diff(j_compress_ptr cinfo, J_BUF_MODE pass_mode)
+{
+  my_diff_ptr diff = (my_diff_ptr)cinfo->coef;
+
+  /* Because it is hitching a ride on the jpeg_forward_dct struct,
+   * start_pass_lossless() will be called at the start of the initial pass.
+   * This ensures that it will be called at the start of the Huffman
+   * optimization and output passes as well.
+   */
+  if (pass_mode == JBUF_CRANK_DEST)
+    (*cinfo->fdct->start_pass) (cinfo);
+
+  diff->iMCU_row_num = 0;
+  start_iMCU_row(cinfo);
+
+  switch (pass_mode) {
+  case JBUF_PASS_THRU:
+    if (diff->whole_image[0] != NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    diff->pub._compress_data = compress_data;
+    break;
+#ifdef FULL_SAMP_BUFFER_SUPPORTED
+  case JBUF_SAVE_AND_PASS:
+    if (diff->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    diff->pub._compress_data = compress_first_pass;
+    break;
+  case JBUF_CRANK_DEST:
+    if (diff->whole_image[0] == NULL)
+      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    diff->pub._compress_data = compress_output;
+    break;
+#endif
+  default:
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+    break;
+  }
+}
+
+
+#define SWAP_ROWS(rowa, rowb) { \
+  _JSAMPROW temp = rowa; \
+  rowa = rowb;  rowb = temp; \
+}
+
+/*
+ * Process some data in the single-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor rows for each component in the image.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf contains a plane for each component in image,
+ * which we index according to the component's SOF position.
+ */
+
+METHODDEF(boolean)
+compress_data(j_compress_ptr cinfo, _JSAMPIMAGE input_buf)
+{
+  my_diff_ptr diff = (my_diff_ptr)cinfo->coef;
+  lossless_comp_ptr losslessc = (lossless_comp_ptr)cinfo->fdct;
+  JDIMENSION MCU_col_num;       /* index of current MCU within row */
+  JDIMENSION MCU_count;         /* number of MCUs encoded */
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  int ci, compi, yoffset, samp_row, samp_rows, samps_across;
+  jpeg_component_info *compptr;
+
+  /* Loop to write as much as one whole iMCU row */
+  for (yoffset = diff->MCU_vert_offset; yoffset < diff->MCU_rows_per_iMCU_row;
+       yoffset++) {
+
+    MCU_col_num = diff->mcu_ctr;
+
+    /* Scale and predict each scanline of the MCU row separately.
+     *
+     * Note: We only do this if we are at the start of an MCU row, ie,
+     * we don't want to reprocess a row suspended by the output.
+     */
+    if (MCU_col_num == 0) {
+      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+        compptr = cinfo->cur_comp_info[ci];
+        compi = compptr->component_index;
+        if (diff->iMCU_row_num < last_iMCU_row)
+          samp_rows = compptr->v_samp_factor;
+        else {
+          /* NB: can't use last_row_height here, since may not be set! */
+          samp_rows =
+            (int)(compptr->height_in_blocks % compptr->v_samp_factor);
+          if (samp_rows == 0) samp_rows = compptr->v_samp_factor;
+          else {
+            /* Fill dummy difference rows at the bottom edge with zeros, which
+             * will encode to the smallest amount of data.
+             */
+            for (samp_row = samp_rows; samp_row < compptr->v_samp_factor;
+                 samp_row++)
+              memset(diff->diff_buf[compi][samp_row], 0,
+                     jround_up((long)compptr->width_in_blocks,
+                               (long)compptr->h_samp_factor) * sizeof(JDIFF));
+          }
+        }
+        samps_across = compptr->width_in_blocks;
+
+        for (samp_row = 0; samp_row < samp_rows; samp_row++) {
+          (*losslessc->scaler_scale) (cinfo,
+                                      input_buf[compi][samp_row],
+                                      diff->cur_row[compi],
+                                      samps_across);
+          (*losslessc->predict_difference[compi])
+            (cinfo, compi, diff->cur_row[compi], diff->prev_row[compi],
+             diff->diff_buf[compi][samp_row], samps_across);
+          SWAP_ROWS(diff->cur_row[compi], diff->prev_row[compi]);
+        }
+      }
+    }
+    /* Try to write the MCU row (or remaining portion of suspended MCU row). */
+    MCU_count =
+      (*cinfo->entropy->encode_mcus) (cinfo,
+                                      diff->diff_buf, yoffset, MCU_col_num,
+                                      cinfo->MCUs_per_row - MCU_col_num);
+    if (MCU_count != cinfo->MCUs_per_row - MCU_col_num) {
+      /* Suspension forced; update state counters and exit */
+      diff->MCU_vert_offset = yoffset;
+      diff->mcu_ctr += MCU_col_num;
+      return FALSE;
+    }
+    /* Completed an MCU row, but perhaps not an iMCU row */
+    diff->mcu_ctr = 0;
+  }
+  /* Completed the iMCU row, advance counters for next one */
+  diff->iMCU_row_num++;
+  start_iMCU_row(cinfo);
+  return TRUE;
+}
+
+
+#ifdef FULL_SAMP_BUFFER_SUPPORTED
+
+/*
+ * Process some data in the first pass of a multi-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor rows for each component in the image.
+ * This amount of data is read from the source buffer and saved into the
+ * virtual arrays.
+ *
+ * We must also emit the data to the compressor.  This is conveniently
+ * done by calling compress_output() after we've loaded the current strip
+ * of the virtual arrays.
+ *
+ * NB: input_buf contains a plane for each component in image.  All components
+ * are loaded into the virtual arrays in this pass.  However, it may be that
+ * only a subset of the components are emitted to the compressor during
+ * this first pass; be careful about looking at the scan-dependent variables
+ * (MCU dimensions, etc).
+ */
+
+METHODDEF(boolean)
+compress_first_pass(j_compress_ptr cinfo, _JSAMPIMAGE input_buf)
+{
+  my_diff_ptr diff = (my_diff_ptr)cinfo->coef;
+  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
+  JDIMENSION samps_across;
+  int ci, samp_row, samp_rows;
+  _JSAMPARRAY buffer;
+  jpeg_component_info *compptr;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Align the virtual buffer for this component. */
+    buffer = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray)
+      ((j_common_ptr)cinfo, diff->whole_image[ci],
+       diff->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION)compptr->v_samp_factor, TRUE);
+
+    /* Count non-dummy sample rows in this iMCU row. */
+    if (diff->iMCU_row_num < last_iMCU_row)
+      samp_rows = compptr->v_samp_factor;
+    else {
+      /* NB: can't use last_row_height here, since may not be set! */
+      samp_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor);
+      if (samp_rows == 0) samp_rows = compptr->v_samp_factor;
+    }
+    samps_across = compptr->width_in_blocks;
+
+    /* Perform point transform scaling and prediction/differencing for all
+     * non-dummy rows in this iMCU row.  Each call on these functions
+     * processes a complete row of samples.
+     */
+    for (samp_row = 0; samp_row < samp_rows; samp_row++) {
+      memcpy(buffer[samp_row], input_buf[ci][samp_row],
+             samps_across * sizeof(_JSAMPLE));
+    }
+  }
+  /* NB: compress_output will increment iMCU_row_num if successful.
+   * A suspension return will result in redoing all the work above next time.
+   */
+
+  /* Emit data to the compressor, sharing code with subsequent passes */
+  return compress_output(cinfo, input_buf);
+}
+
+
+/*
+ * Process some data in subsequent passes of a multi-pass case.
+ * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
+ * per call, ie, v_samp_factor rows for each component in the scan.
+ * The data is obtained from the virtual arrays and fed to the compressor.
+ * Returns TRUE if the iMCU row is completed, FALSE if suspended.
+ *
+ * NB: input_buf is ignored; it is likely to be a NULL pointer.
+ */
+
+METHODDEF(boolean)
+compress_output(j_compress_ptr cinfo, _JSAMPIMAGE input_buf)
+{
+  my_diff_ptr diff = (my_diff_ptr)cinfo->coef;
+  int ci, compi;
+  _JSAMPARRAY buffer[MAX_COMPS_IN_SCAN];
+  jpeg_component_info *compptr;
+
+  /* Align the virtual buffers for the components used in this scan.
+   * NB: during first pass, this is safe only because the buffers will
+   * already be aligned properly, so jmemmgr.c won't need to do any I/O.
+   */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+    compptr = cinfo->cur_comp_info[ci];
+    compi = compptr->component_index;
+    buffer[compi] = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray)
+      ((j_common_ptr)cinfo, diff->whole_image[compi],
+       diff->iMCU_row_num * compptr->v_samp_factor,
+       (JDIMENSION)compptr->v_samp_factor, FALSE);
+  }
+
+  return compress_data(cinfo, buffer);
+}
+
+#endif /* FULL_SAMP_BUFFER_SUPPORTED */
+
+
+/*
+ * Initialize difference buffer controller.
+ */
+
+GLOBAL(void)
+_jinit_c_diff_controller(j_compress_ptr cinfo, boolean need_full_buffer)
+{
+  my_diff_ptr diff;
+  int ci, row;
+  jpeg_component_info *compptr;
+
+  diff = (my_diff_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
+                                sizeof(my_diff_controller));
+  cinfo->coef = (struct jpeg_c_coef_controller *)diff;
+  diff->pub.start_pass = start_pass_diff;
+
+  /* Create the prediction row buffers. */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    diff->cur_row[ci] = *(_JSAMPARRAY)(*cinfo->mem->alloc_sarray)
+      ((j_common_ptr)cinfo, JPOOL_IMAGE,
+       (JDIMENSION)jround_up((long)compptr->width_in_blocks,
+                             (long)compptr->h_samp_factor),
+       (JDIMENSION)1);
+    diff->prev_row[ci] = *(_JSAMPARRAY)(*cinfo->mem->alloc_sarray)
+      ((j_common_ptr)cinfo, JPOOL_IMAGE,
+       (JDIMENSION)jround_up((long)compptr->width_in_blocks,
+                             (long)compptr->h_samp_factor),
+       (JDIMENSION)1);
+  }
+
+  /* Create the difference buffer. */
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    diff->diff_buf[ci] =
+      ALLOC_DARRAY(JPOOL_IMAGE,
+                   (JDIMENSION)jround_up((long)compptr->width_in_blocks,
+                                         (long)compptr->h_samp_factor),
+                   (JDIMENSION)compptr->v_samp_factor);
+    /* Prefill difference rows with zeros.  We do this because only actual
+     * data is placed in the buffers during prediction/differencing, leaving
+     * any dummy differences at the right edge as zeros, which will encode
+     * to the smallest amount of data.
+     */
+    for (row = 0; row < compptr->v_samp_factor; row++)
+      memset(diff->diff_buf[ci][row], 0,
+             jround_up((long)compptr->width_in_blocks,
+                       (long)compptr->h_samp_factor) * sizeof(JDIFF));
+  }
+
+  /* Create the sample buffer. */
+  if (need_full_buffer) {
+#ifdef FULL_SAMP_BUFFER_SUPPORTED
+    /* Allocate a full-image virtual array for each component, */
+    /* padded to a multiple of samp_factor differences in each direction. */
+    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+         ci++, compptr++) {
+      diff->whole_image[ci] = (*cinfo->mem->request_virt_sarray)
+        ((j_common_ptr)cinfo, JPOOL_IMAGE, FALSE,
+         (JDIMENSION)jround_up((long)compptr->width_in_blocks,
+                               (long)compptr->h_samp_factor),
+         (JDIMENSION)jround_up((long)compptr->height_in_blocks,
+                               (long)compptr->v_samp_factor),
+         (JDIMENSION)compptr->v_samp_factor);
+    }
+#else
+    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
+#endif
+  } else
+    diff->whole_image[0] = NULL; /* flag for no virtual arrays */
+}
+
+#endif /* C_LOSSLESS_SUPPORTED */
--- a/Show More
+++ b/Show More