overlay: remove keyboard / gamepad tab navigation (#755)

Tired of constantly fighting unintentional keyboard navigation behavior.

Also, remove `ImGuiConfigFlags_IsTouchScreen` which literally does
nothing.

.github/ISSUE_TEMPLATE/bug_report.md

@@ -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.

src/spice2x/CMakeLists.txt

@@ -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")
 
@@ -409,6 +416,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
         games/nost/io.cpp
         games/nost/poke.cpp
         games/gitadora/gitadora.cpp
+        games/gitadora/asio.cpp
         games/gitadora/io.cpp
         games/gitadora/handle.cpp
         games/gitadora/j32d.cpp
@@ -486,6 +494,8 @@ 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
@@ -494,8 +504,13 @@ set(SOURCE_FILES ${SOURCE_FILES}
         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
@@ -517,12 +532,19 @@ set(SOURCE_FILES ${SOURCE_FILES}
         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
@@ -556,6 +578,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
 
         # overlay
         overlay/overlay.cpp
+        overlay/notifications.cpp
         overlay/window.cpp
         overlay/imgui/extensions.cpp
         overlay/imgui/impl_spice.cpp
@@ -607,6 +630,9 @@ set(SOURCE_FILES ${SOURCE_FILES}
         reader/structuredmessage.cpp
         reader/crypt.cpp
 
+        # sdk
+        sdk/sdk.cpp
+
         # stubs
         stubs/stubs.cpp
 
@@ -633,6 +659,7 @@ 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
@@ -644,37 +671,47 @@ set(SOURCE_FILES ${SOURCE_FILES}
 
 source_group(TREE ${CMAKE_CURRENT_SOURCE_DIR} PREFIX "Source Files" FILES ${SOURCE_FILES})
 
+# 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)
+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
@@ -707,6 +744,10 @@ 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")
@@ -726,6 +767,9 @@ 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")
@@ -738,7 +782,7 @@ 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
+        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)
 target_link_libraries(spicetools_cfg PUBLIC winscard)
 set_target_properties(spicetools_cfg PROPERTIES PREFIX "")
@@ -756,7 +800,7 @@ 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
+        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)
 set_target_properties(spicetools_cfg_linux PROPERTIES PREFIX "")
 set_target_properties(spicetools_cfg_linux PROPERTIES OUTPUT_NAME "spicecfg_linux")
@@ -860,6 +904,36 @@ 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
 ####################
 
@@ -869,14 +943,14 @@ 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"

src/spice2x/README.md

@@ -283,6 +283,16 @@ 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.

src/spice2x/acio/mdxf/mdxf.cpp

@@ -1,14 +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
 
@@ -109,9 +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);
         }
     });
 }
@@ -512,4 +515,4 @@ acio::MDXFModule::~MDXFModule() {
     if (IS_THREAD_NEEDED) {
         mdxf_thread_stop();
     }
-}
+}

src/spice2x/acioemu/device.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <cstdint>
 #include <string>
 
 #include "util/circular_buffer.h"

src/spice2x/acioemu/icca.cpp

@@ -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);
             }
         });
     }

src/spice2x/api/controller.cpp

@@ -11,6 +11,8 @@
 #include "util/logging.h"
 #include "util/utils.h"
 
+#include "overlay/notifications.h"
+
 #include "module.h"
 #include "modules/analogs.h"
 #include "modules/buttons.h"
@@ -197,6 +199,9 @@ 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 +282,9 @@ 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();

src/spice2x/api/modules/analogs.cpp

@@ -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;
             }
         }

src/spice2x/api/modules/buttons.cpp

@@ -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;
             }
         }

src/spice2x/api/modules/capture.cpp

@@ -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);
@@ -41,6 +93,7 @@ namespace api::modules {
      * reduce: uint for dividing image size
      */
     void Capture::get_jpg(Request &req, Response &res) {
+        CAPTURE_BUFFER.clear();
         CAPTURE_BUFFER.reserve(1024 * 128);
 
         // settings
@@ -71,24 +124,15 @@ namespace api::modules {
         bool success = graphics_capture_receive_jpeg(screen, [] (uint8_t byte) {
             CAPTURE_BUFFER.push_back(byte);
         }, true, quality, true, divide, &timestamp, &width, &height);
-        if (!success) {
+
+        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);
     }
 }

src/spice2x/api/modules/keypads.cpp

@@ -7,6 +7,7 @@
 #include "avs/game.h"
 #include "external/rapidjson/document.h"
 #include "misc/eamuse.h"
+#include "util/precise_timer.h"
 
 using namespace std::placeholders;
 using namespace rapidjson;
@@ -58,6 +59,7 @@ namespace api::modules {
         // get params
         auto keypad = req.params[0].GetUint();
         auto input = std::string(req.params[1].GetString());
+        timeutils::PreciseSleepTimer timer;
 
         // process all chars
         for (auto c : input) {
@@ -91,11 +93,11 @@ namespace api::modules {
 
             // set
             eamuse_set_keypad_overrides(keypad, state);
-            Sleep(sleep_time);
+            timer.sleep(sleep_time);
 
             // unset
             eamuse_set_keypad_overrides(keypad, 0);
-            Sleep(sleep_time);
+            timer.sleep(sleep_time);
         }
     }
 

src/spice2x/api/serial.cpp

@@ -5,6 +5,7 @@
 #include <utility>
 
 #include "util/logging.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"
 
 
@@ -16,6 +17,7 @@ namespace api {
         controller->init_state(this->state);
         this->thread = new std::thread([this] () {
             log_warning("api::serial", "listening on {} (baud: {})", this->port, this->baud);
+            timeutils::PreciseSleepTimer timer;
 
             // read buffer
             uint8_t read_buffer[16*1024];
@@ -160,7 +162,7 @@ namespace api {
 
                 // slow down on reconnect
                 if (this->running) {
-                    Sleep(retry_time);
+                    timer.sleep(retry_time);
                 }
             }
         });

src/spice2x/api/websocket.cpp

@@ -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());

src/spice2x/avs/ea3.cpp

@@ -390,8 +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

src/spice2x/build/controller_presets.xml

@@ -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"/>
-            <analog name="Turntable P2" devid="Player 2" index="4" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
+            <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"/>
-            <analog name="Turntable P2" devid="arcin (2p)" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="true" smoothing="false" multiplier="1"/>
+            <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"/>
-            <analog name="Key 2" devid="Analogs key 01-28" index="5" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 3" devid="Analogs key 01-28" index="6" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 4" devid="Analogs key 01-28" index="7" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 5" devid="Analogs key 01-28" index="8" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 6" devid="Analogs key 01-28" index="9" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 7" devid="Analogs key 01-28" index="10" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 8" devid="Analogs key 01-28" index="11" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 9" devid="Analogs key 01-28" index="12" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 10" devid="Analogs key 01-28" index="13" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 11" devid="Analogs key 01-28" index="14" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 12" devid="Analogs key 01-28" index="15" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 13" devid="Analogs key 01-28" index="16" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 14" devid="Analogs key 01-28" index="17" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 15" devid="Analogs key 01-28" index="18" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 16" devid="Analogs key 01-28" index="19" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 17" devid="Analogs key 01-28" index="20" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 18" devid="Analogs key 01-28" index="21" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 19" devid="Analogs key 01-28" index="22" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 20" devid="Analogs key 01-28" index="23" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 21" devid="Analogs key 01-28" index="24" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 22" devid="Analogs key 01-28" index="25" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 23" devid="Analogs key 01-28" index="26" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 24" devid="Analogs key 01-28" index="27" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 25" devid="Analogs key 01-28" index="28" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 26" devid="Analogs key 01-28" index="29" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 27" devid="Analogs key 01-28" index="30" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Key 28" devid="Analogs key 01-28" index="31" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
+            <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"/>
-            <analog name="Guitar P1 Wail Y" devid="Gitaller P1" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
-            <analog name="Guitar P1 Wail Z" devid="Gitaller P1" index="5" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
+            <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"/>
-            <analog name="VOL-R" devid="Buttons (MI_01)" index="0" sensivity="1" deadzone="0" deadzone_mirror="false" invert="false" smoothing="false" multiplier="1"/>
+            <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"/>

src/spice2x/build/manifest.manifest

@@ -23,7 +23,7 @@
     <trustInfo xmlns="urn:schemas-microsoft-com:asm.v2">
         <security>
             <requestedPrivileges>
-                <requestedExecutionLevel level="requireAdministrator" uiAccess="false"/>
+                <requestedExecutionLevel level="asInvoker" uiAccess="false"/>
             </requestedPrivileges>
         </security>
     </trustInfo>

src/spice2x/build_all.sh

@@ -68,8 +68,8 @@ 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"
 
@@ -91,14 +91,20 @@ fi
 
 # is the XP-compatible toolchain installed?
 XP_MUST_BUILD=0
-BUILD_XP=0
-if [ -f "$TOOLCHAIN_WINXP_32" ] && [ -f "$TOOLCHAIN_WINXP_64" ]; then
-	BUILD_XP=1;
+# 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 "$TOOLCHAIN_WINXP_32" ]; then
+	BUILD_XP_32=1;
 elif ((XP_MUST_BUILD > 0))
 then
-	echo "WinXP toolchain not available, aborting"
+	echo "WinXP 32bit toolchain not available, aborting"
 	exit 1
 fi
+if ((BUILD_XP_64_ENABLE > 0)) && [ -f "$TOOLCHAIN_WINXP_64" ]; then
+	BUILD_XP_64=1;
+fi
 
 # determine number of cores
 CORES=$(nproc)
@@ -113,12 +119,17 @@ 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 > 0))
+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 toolchain not available, skipping WinXP builds"
+	echo "WinXP 64bit builds disabled, skipping WinXP 64bit builds"
 fi
 echo "Distribution Name: $DIST_NAME"
 echo "Build Type: $BUILD_TYPE"
@@ -160,7 +171,7 @@ time (
 	cmake -G "Ninja" -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_64} -DCMAKE_BUILD_TYPE=${BUILD_TYPE} "$OLDPWD" && ninja ${TARGETS_64}
 	popd > /dev/null
 
-	if ((BUILD_XP > 0))
+	if ((BUILD_XP_32 > 0))
 	then
 		# 32 bit Windows XP
 		echo ""
@@ -172,9 +183,15 @@ time (
 		fi
 		mkdir -p ${BUILDDIR_WINXP_32}
 		pushd ${BUILDDIR_WINXP_32} > /dev/null
-		CXXFLAGS="$CXXFLAGS -DSPICE_XP=1" cmake -G "Ninja" -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_WINXP_32} -DCMAKE_BUILD_TYPE=${BUILD_TYPE} "$OLDPWD" && ninja ${TARGETS_XP32}
+		cmake -G "Ninja" -DCMAKE_TOOLCHAIN_FILE=${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)..."
@@ -185,11 +202,11 @@ time (
 		fi
 		mkdir -p ${BUILDDIR_WINXP_64}
 		pushd ${BUILDDIR_WINXP_64} > /dev/null
-		CXXFLAGS="$CXXFLAGS -DSPICE_XP=1" cmake -G "Ninja" -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_WINXP_64} -DCMAKE_BUILD_TYPE=${BUILD_TYPE} "$OLDPWD" && ninja ${TARGETS_XP64}
+		cmake -G "Ninja" -DCMAKE_TOOLCHAIN_FILE=${TOOLCHAIN_WINXP_64} -DCMAKE_BUILD_TYPE=${BUILD_TYPE} -DSPICE_XP=ON "$OLDPWD" && ninja ${TARGETS_XP64}
 		popd > /dev/null
 	else
 		echo ""
-		echo "Skipping WinXP builds, toolchain not specified"
+		echo "Skipping WinXP 64bit builds"
 	fi
 
 	echo ""
@@ -197,7 +214,7 @@ time (
 	echo "==========================="
 )
 
-if ((BUILD_XP > 0))
+if ((BUILD_XP_32 > 0)) || ((BUILD_XP_64 > 0))
 then
 	echo ""
 	echo "Checking XP compatibility..."
@@ -205,11 +222,17 @@ then
 	if ! command -v windows_dll_compat_checker &> /dev/null; then
 		echo "WARNING: windows_dll_compat_checker not found, skipping XP compatibility check"
 	else
-		windows_dll_compat_checker -s PREMADE/winxp_x86_64.ini \
-			${BUILDDIR_WINXP_64}/spicetools/64/spice64.exe
-		windows_dll_compat_checker -s PREMADE/winxp_x86_64_32bit_dlls.ini \
-			${BUILDDIR_WINXP_32}/spicetools/spicecfg.exe \
-			${BUILDDIR_WINXP_32}/spicetools/32/spice.exe
+		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/winxp_x86_64_32bit_dlls.ini \
+				${BUILDDIR_WINXP_32}/spicetools/spicecfg.exe \
+				${BUILDDIR_WINXP_32}/spicetools/32/spice.exe
+		fi
 	fi
 fi
 
@@ -269,7 +292,9 @@ rm -rf ${OUTDIR_EXTRAS}
 mkdir -p ${OUTDIR_EXTRAS}
 mkdir -p ${OUTDIR_EXTRAS}/largeaddressaware
 mkdir -p ${OUTDIR_EXTRAS}/linux
-if ((BUILD_XP > 0))
+mkdir -p ${OUTDIR_EXTRAS}/sdk/samples/32
+mkdir -p ${OUTDIR_EXTRAS}/sdk/samples/64
+if ((BUILD_XP_32 > 0)) || ((BUILD_XP_64 > 0))
 then
 mkdir -p ${OUTDIR_EXTRAS}/winxp
 fi
@@ -281,16 +306,8 @@ 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
@@ -298,20 +315,22 @@ 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
-	if ((BUILD_XP > 0))
+	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
@@ -326,6 +345,9 @@ 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

src/spice2x/build_docker.bat

@@ -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

src/spice2x/cfg/analog.cpp

@@ -237,7 +237,8 @@ float Analog::applyDeadzone(float raw_value) {
         deadzone = -deadzone;
     }
 
-    if (getType() != GameAPI::Analogs::AnalogType::LinearPositive) {
+    // 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;
@@ -290,4 +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;
+}

src/spice2x/cfg/analog.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <cstdint>
 #include <array>
 #include <string>
 #include <cmath>
@@ -34,6 +35,11 @@ struct AnalogMovingAverage {
     float cosine;
 };
 
+struct AnalogDelayEntry {
+    double time_in_ms;
+    float value;
+};
+
 class Analog {
 private:
     std::string name;
@@ -63,6 +69,15 @@ private:
     float divisor_previous_value = 0.5f;
     unsigned short divisor_region = 0;
 
+    // relative input mode
+    float rel_mode_absolute_value = 0.5f;
+    float rel_mode_last_read_time_s = 0.f;
+    bool relative_mode = false;
+
+    // delay
+    uint32_t delay_ms = 0;
+    std::queue<AnalogDelayEntry> delayed_inputs;
+
     float calculateAngularDifference(float old_rads, float new_rads);
     float normalizeAngle(float rads);
     float normalizeAnalogValue(float value);
@@ -83,6 +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) {
@@ -98,7 +115,9 @@ public:
         smoothing = false;
         deadzone_mirror = false;
         setMultiplier(1);
+        setRelativeMode(false);
         setLastState(0.5f);
+        setDelayMs(0);
     }
 
     inline void clearBindings() {
@@ -195,6 +214,19 @@ public:
         this->last_state = last_state;
     }
 
+    inline bool isRelativeMode() const {
+        return this->relative_mode;
+    }
+
+    inline void setRelativeMode(bool relative_mode) {
+        if (relative_mode) {
+            this->smoothing = false;
+        }
+        this->relative_mode = relative_mode;
+        this->rel_mode_absolute_value = 0.5f;
+        this->rel_mode_last_read_time_s = 0.f;
+    }
+
     inline GameAPI::Analogs::AnalogType getType() const {
         return this->type;
     }
@@ -202,4 +234,12 @@ public:
     inline void setType(GameAPI::Analogs::AnalogType type) {
         this->type = type;
     }
+
+    inline uint32_t getDelayMs() const {
+        return this->delay_ms;
+    }
+
+    inline void setDelayMs(uint32_t delay_ms) {
+        this->delay_ms = delay_ms;
+    }
 };

src/spice2x/cfg/api.cpp

@@ -679,38 +679,45 @@ float GameAPI::Analogs::getState(rawinput::RawInputManager *manager, rawinput::D
             }
 
             // deadzone
-            // do not apply deadzone to circular analogs since it doesn't make sense
-            if (analog.isDeadzoneSet() && analog.getType() != AnalogType::Circular) {
+            // 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.getType() == AnalogType::Circular) {
-                // integer multiplier
-                value = analog.applyMultiplier(value);
 
-                // smoothing/sensitivity
-                if (analog.getSmoothing() || analog.isSensitivitySet()) {
-                    float rads = value * (float) M_TAU;
+                if (analog.isRelativeMode()) {
+                    value = analog.getRelativeModeValue(value);
 
-                    // smoothing
-                    if (analog.getSmoothing()) {
+                } else {
+                    // integer multiplier
+                    value = analog.applyMultiplier(value);
 
-                        // preserve direction
-                        if (rads >= M_TAU) {
-                            rads -= 0.0001f;
+                    // smoothing/sensitivity
+                    if (analog.getSmoothing() || analog.isSensitivitySet()) {
+                        float rads = value * (float) M_TAU;
+
+                        // 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);
+                        }
 
-                    // sensitivity
-                    if (analog.isSensitivitySet()) {
-                        rads = analog.applyAngularSensitivity(rads);
+                        // apply to value
+                        value = rads * (float) M_1_TAU;
                     }
-
-                    // apply to value
-                    value = rads * (float) M_1_TAU;
                 }
             } else {
                 // sensitivity
@@ -760,6 +767,12 @@ float GameAPI::Analogs::getState(rawinput::RawInputManager *manager, rawinput::D
                     value = std::clamp(value, 0.f, 1.f);
                 }
             }
+
+            // delay
+            if (analog.getDelayMs() > 0) {
+                value = analog.getDelayedValue(value);
+            }
+
             break;
         }
         case rawinput::MIDI: {

src/spice2x/cfg/config.cpp

@@ -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");
 
@@ -252,6 +252,8 @@ bool Config::addGame(Game &game) {
                 bool invert = false;
                 bool smoothing = false;
                 int multiplier = 1;
+                bool relative_mode = false;
+                uint32_t delay = 0;
                 tinyxml2::XMLError err1 = gameAnalogNode->QueryIntAttribute("index", &index);
                 gameAnalogNode->QueryFloatAttribute("sensivity", &sensitivity);
                 gameAnalogNode->QueryFloatAttribute("deadzone", &deadzone);
@@ -259,6 +261,8 @@ bool Config::addGame(Game &game) {
                 gameAnalogNode->QueryBoolAttribute("invert", &invert);
                 gameAnalogNode->QueryBoolAttribute("smoothing", &smoothing);
                 gameAnalogNode->QueryIntAttribute("multiplier", &multiplier);
+                gameAnalogNode->QueryBoolAttribute("relative", &relative_mode);
+                gameAnalogNode->QueryUnsignedAttribute("delay_ms", &delay);
                 const char *devid = gameAnalogNode->Attribute("devid");
 
                 if (err1 != tinyxml2::XMLError::XML_SUCCESS || !devid) {
@@ -273,6 +277,8 @@ bool Config::addGame(Game &game) {
                     gameAnalogNode->SetAttribute("invert", it.getInvert());
                     gameAnalogNode->SetAttribute("smoothing", it.getSmoothing());
                     gameAnalogNode->SetAttribute("multiplier", it.getMultiplier());
+                    gameAnalogNode->SetAttribute("relative", it.isRelativeMode());
+                    gameAnalogNode->SetAttribute("delay_ms", it.getDelayMs());
                     gameAnalogsNode->InsertEndChild(gameAnalogNode);
                 } else {
                     it.setIndex(static_cast<unsigned short int>(index));
@@ -283,6 +289,8 @@ bool Config::addGame(Game &game) {
                     it.setInvert(invert);
                     it.setSmoothing(smoothing);
                     it.setMultiplier(multiplier);
+                    it.setRelativeMode(relative_mode);
+                    it.setDelayMs(delay);
                 }
             } else {
                 gameAnalogNode = this->configFile.NewElement("analog");
@@ -294,6 +302,8 @@ bool Config::addGame(Game &game) {
                 gameAnalogNode->SetAttribute("invert", it.getInvert());
                 gameAnalogNode->SetAttribute("smoothing", it.getSmoothing());
                 gameAnalogNode->SetAttribute("multiplier", it.getMultiplier());
+                gameAnalogNode->SetAttribute("relative", it.isRelativeMode());
+                gameAnalogNode->SetAttribute("delay_ms", it.getDelayMs());
                 gameAnalogNode->SetAttribute("devid", it.getDeviceIdentifier().c_str());
                 gameAnalogsNode->InsertEndChild(gameAnalogNode);
             }
@@ -447,6 +457,8 @@ bool Config::addGame(Game &game) {
             gameAnalogNode->SetAttribute("invert", it.getInvert());
             gameAnalogNode->SetAttribute("smoothing", it.getSmoothing());
             gameAnalogNode->SetAttribute("multiplier", it.getMultiplier());
+            gameAnalogNode->SetAttribute("relative", it.isRelativeMode());
+            gameAnalogNode->SetAttribute("delay_ms", it.getDelayMs());
             gameAnalogsNode->InsertEndChild(gameAnalogNode);
         }
 
@@ -474,13 +486,20 @@ 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
@@ -679,6 +698,8 @@ bool Config::updateBinding(const Game &game, const Analog &analog) {
         gameAnalogNode->SetAttribute("invert", analog.getInvert());
         gameAnalogNode->SetAttribute("smoothing", analog.getSmoothing());
         gameAnalogNode->SetAttribute("multiplier", analog.getMultiplier());
+        gameAnalogNode->SetAttribute("relative", analog.isRelativeMode());
+        gameAnalogNode->SetAttribute("delay_ms", analog.getDelayMs());
         gameAnalogNode->SetAttribute("devid", analog.getDeviceIdentifier().c_str());
     } else {
         gameAnalogNode = this->configFile.NewElement("analog");
@@ -689,6 +710,8 @@ bool Config::updateBinding(const Game &game, const Analog &analog) {
         gameAnalogNode->SetAttribute("invert", analog.getInvert());
         gameAnalogNode->SetAttribute("smoothing", analog.getSmoothing());
         gameAnalogNode->SetAttribute("multiplier", analog.getMultiplier());
+        gameAnalogNode->SetAttribute("relative", analog.isRelativeMode());
+        gameAnalogNode->SetAttribute("delay_ms", analog.getDelayMs());
         gameAnalogNode->SetAttribute("devid", analog.getDeviceIdentifier().c_str());
         gameAnalogsNode->InsertEndChild(gameAnalogNode);
     }
@@ -1125,6 +1148,8 @@ std::vector<Analog> Config::getAnalogs(const std::string &gameName) {
                 bool invert = false;
                 bool smoothing = false;
                 int multiplier = 1;
+                bool relative_mode = false;
+                uint32_t delay = 0;
                 gameAnalogNode->QueryIntAttribute("index", &index);
                 gameAnalogNode->QueryFloatAttribute("sensivity", &sensitivity);
                 gameAnalogNode->QueryFloatAttribute("deadzone", &deadzone);
@@ -1132,6 +1157,8 @@ std::vector<Analog> Config::getAnalogs(const std::string &gameName) {
                 gameAnalogNode->QueryBoolAttribute("invert", &invert);
                 gameAnalogNode->QueryBoolAttribute("smoothing", &smoothing);
                 gameAnalogNode->QueryIntAttribute("multiplier", &multiplier);
+                gameAnalogNode->QueryBoolAttribute("relative", &relative_mode);
+                gameAnalogNode->QueryUnsignedAttribute("delay_ms", &delay);
                 const char *devid = gameAnalogNode->Attribute("devid");
 
                 // create analog and add to list
@@ -1143,6 +1170,8 @@ std::vector<Analog> Config::getAnalogs(const std::string &gameName) {
                 analog.setInvert(invert);
                 analog.setSmoothing(smoothing);
                 analog.setMultiplier(multiplier);
+                analog.setRelativeMode(relative_mode);
+                analog.setDelayMs(delay);
                 if (devid) {
                     analog.setDeviceIdentifier(devid);
                 }

src/spice2x/cfg/config.h

@@ -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);

src/spice2x/cfg/configurator.cpp

@@ -1,11 +1,15 @@
-#include "configurator.h"
+#include "configurator.h"
+
+#include <d3d9.h>
 
 #include "overlay/overlay.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 +20,79 @@ 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;
+        }
+
+        IDirect3D9 *d3d = Direct3DCreate9(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

src/spice2x/cfg/configurator.h

@@ -11,6 +11,7 @@ namespace cfg {
     // globals
     extern bool CONFIGURATOR_STANDALONE;
     extern ConfigType CONFIGURATOR_TYPE;
+    extern bool CONFIGURATOR_FORCE_SOFTWARE_RENDER;
 
     class Configurator {
     private:

src/spice2x/cfg/configurator_wnd.cpp

@@ -1,11 +1,17 @@
-#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 "cfg/configurator.h"
 
 #include "icon.h"
@@ -16,6 +22,156 @@ static int WINDOW_SIZE_X = 800;
 static int WINDOW_SIZE_Y = 600;
 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));
+}
+
+// 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 +183,22 @@ 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(
@@ -53,11 +219,23 @@ cfg::ConfiguratorWindow::ConfiguratorWindow() {
 
 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() {
@@ -67,8 +245,23 @@ void cfg::ConfiguratorWindow::run() {
     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;
@@ -102,6 +295,72 @@ 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 +369,192 @@ 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);
     }

src/spice2x/cfg/configurator_wnd.h

@@ -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);
     };
 }

src/spice2x/cfg/controller_presets.cpp

@@ -69,6 +69,8 @@ namespace overlay::windows {
         el->SetAttribute("invert", analog.invert);
         el->SetAttribute("smoothing", analog.smoothing);
         el->SetAttribute("multiplier", analog.multiplier);
+        el->SetAttribute("relative", analog.relative_mode);
+        el->SetAttribute("delay_ms", analog.delay_ms);
         parent->InsertEndChild(el);
     }
 
@@ -90,6 +92,8 @@ namespace overlay::windows {
         el->QueryBoolAttribute("invert", &a.invert);
         el->QueryBoolAttribute("smoothing", &a.smoothing);
         el->QueryIntAttribute("multiplier", &a.multiplier);
+        el->QueryBoolAttribute("relative", &a.relative_mode);
+        el->QueryUnsignedAttribute("delay_ms", &a.delay_ms);
 
         return a;
     }

src/spice2x/cfg/manifest.manifest

@@ -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>

src/spice2x/cfg/option.h

@@ -46,6 +46,8 @@ struct OptionDefinition {
 
     // for OptionPickerType::FilePath
     std::string file_extension = "";
+
+    std::string quick_setting_category = "";
 };
 
 class Option {

src/spice2x/external/headsocket.h

@@ -20,6 +20,7 @@ Usage:
 #include <memory>
 #include <string>
 #include <map>
+#include <cstdint>
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 

src/spice2x/external/imgui/CMakeLists.txt

@@ -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})

src/spice2x/external/imgui/backends/imgui_impl_dx11.cpp

@@ -0,0 +1,729 @@
+// 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'.
+
+// 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)
+//  2026-01-19: 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).
+//  2025-01-06: 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_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;
+}
+
+// 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);
+}
+
+// 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 = 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 = 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.SamplerLinear = bd->pTexSamplerLinear;
+    render_state.SamplerNearest = bd->pTexSamplerNearest;
+    render_state.VertexConstantBuffer = bd->pVertexConstantBuffer;
+    platform_io.Renderer_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()
+                // (ImDrawCallback_ResetRenderState is a special callback value used by the user to request the renderer to reset render state.)
+                if (pcmd->UserCallback == ImDrawCallback_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 = 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.
+
+    ImGuiPlatformIO& platform_io = ImGui::GetPlatformIO();
+    platform_io.Renderer_TextureMaxWidth = platform_io.Renderer_TextureMaxHeight = D3D11_REQ_TEXTURE2D_U_OR_V_DIMENSION;
+
+    // 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();
+
+    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_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);
+    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!");
+}
+
+//-----------------------------------------------------------------------------
+
+#endif // #ifndef IMGUI_DISABLE

src/spice2x/external/imgui/backends/imgui_impl_dx11.h

@@ -0,0 +1,52 @@
+// 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'.
+
+// 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;
+    ID3D11SamplerState*     SamplerLinear;
+    ID3D11SamplerState*     SamplerNearest;
+    ID3D11Buffer*           VertexConstantBuffer;
+};
+
+#endif // #ifndef IMGUI_DISABLE

src/spice2x/external/robin_hood.h

@@ -39,6 +39,7 @@
 #define ROBIN_HOOD_VERSION_PATCH 2  // for backwards-compatible bug fixes
 
 #include <algorithm>
+#include <cstdint>
 #include <cstdlib>
 #include <cstring>
 #include <functional>

src/spice2x/games/ccj/trackball.cpp

@@ -6,6 +6,7 @@
 #include "util/logging.h"
 #include "rawinput/rawinput.h"
 #include "games/io.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"
 #include "io.h"
 
@@ -112,7 +113,7 @@ namespace games::ccj {
                 static std::chrono::time_point<std::chrono::steady_clock> lastModified = std::chrono::steady_clock::now();
                 static std::chrono::milliseconds debounceDuration(100);
                 auto currentTime = std::chrono::steady_clock::now();
-                bool pressed = get_async_secondary_mouse() & 0x8000;
+                bool pressed = get_async_secondary_mouse();
                 bool focused = GetForegroundWindow() == hWnd;
 
                 if (focused && MOUSE_TRACKBALL_USE_TOGGLE && pressed && (currentTime - lastModified > debounceDuration)) {
@@ -202,10 +203,19 @@ namespace games::ccj {
     }
 
     static BOOL WINAPI RegisterRawInputDevices_hook(PCRAWINPUTDEVICE pRawInputDevices, UINT uiNumDevices, UINT cbSize) {
-        if (uiNumDevices == 2 && pRawInputDevices[1].usUsage == HID_USAGE_GENERIC_GAMEPAD)
-            uiNumDevices = 1;
 
-        return RegisterRawInputDevices_orig(pRawInputDevices, uiNumDevices, cbSize);
+        // if the caller is spice itself, then pass through.
+        if (pRawInputDevices &&
+            (uiNumDevices > 0) &&
+            (pRawInputDevices[0].hwndTarget == RI_MGR->input_hwnd)) {
+
+            return RegisterRawInputDevices_orig(pRawInputDevices, uiNumDevices, cbSize);
+        }
+
+        // otherwise, it must be the game; prevent the game from registering for raw input
+        // and hijacking WM_INPUT messages; we need that for rawinput to work.
+        // even if we drop this, trackball emulation and mouse-as-touch input still work
+        return TRUE;
     }
 
 
@@ -243,6 +253,7 @@ namespace games::ccj {
         log_info("trackball", "thread start, use mouse: {}, toggle: {}", MOUSE_TRACKBALL, MOUSE_TRACKBALL_USE_TOGGLE);
 
         tbThread = new std::thread([&] {
+            timeutils::PreciseSleepTimer timer;
             while (tbThreadRunning) {
                 if (hWnd && wndProc) {
                     wndProc(hWnd, WM_INPUT, RIM_INPUT, (LPARAM)fakeHandle);
@@ -251,7 +262,7 @@ namespace games::ccj {
                 if (!tbThreadRunning)
                     break;
 
-                std::this_thread::sleep_for(10ms);
+                timer.sleep(10);
             }
         });
     }

src/spice2x/games/ddr/p3io/p3io.cpp

@@ -3,6 +3,7 @@
 #include "cfg/api.h"
 #include "rawinput/rawinput.h"
 #include "util/logging.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"
 
 #include "../ddr.h"
@@ -459,7 +460,8 @@ int games::ddr::DDRP3IOHandle::device_io(
     if (nOutBufferSize >= 4) {
 
         // cool down
-        Sleep(1);
+        static thread_local timeutils::PreciseSleepTimer timer;
+        timer.sleep(1);
 
         // get controls as single variable (4 bytes)
         auto &controls = *(uint32_t*) lpOutBuffer;

src/spice2x/games/ddr/p4io/p4io.cpp

@@ -2,6 +2,7 @@
 
 #include "rawinput/rawinput.h"
 #include "util/logging.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"
 
 #include "../io.h"
@@ -255,7 +256,8 @@ namespace games::ddr {
                 case P4IO_IOCTL_GET_INPUTS: {
                     
                     // Prevents this function from being called at its normal 2000-2500 kHz cadence and overloading the CPU, instead reduces it to 250 Hz
-                    Sleep(4);
+                    static thread_local timeutils::PreciseSleepTimer timer;
+                    timer.sleep(4);
                     
                     memset(lpOutBuffer, 0, 16);
                     auto controls = (uint32_t*) lpOutBuffer;

src/spice2x/games/drs/drs.cpp

@@ -7,6 +7,7 @@
 #include "games/game.h"
 #include "util/detour.h"
 #include "util/logging.h"
+#include "util/precise_timer.h"
 #include "util/memutils.h"
 #include "rgb_cam.h"
 
@@ -221,10 +222,11 @@ namespace games::drs {
     void start_touch() {
         std::thread t([] {
             log_info("drs", "starting touch input thread");
+            timeutils::PreciseSleepTimer timer;
 
             // main loop
             while (TRUE) {
-                std::this_thread::sleep_for(std::chrono::milliseconds(1));
+                timer.sleep(1);
 
                 TOUCH_EVENTS.clear();
                 touch_get_events(TOUCH_EVENTS);

src/spice2x/games/gitadora/asio.cpp

@@ -0,0 +1,171 @@
+#include "asio.h"
+
+#include <windows.h>
+#include <cstring>
+
+#include "avs/game.h"
+#include "gitadora.h"
+#include "util/detour.h"
+#include "util/logging.h"
+
+namespace games::gitadora {
+
+    // Redirects the game's hard-coded "XONAR" ASIO driver lookup to the
+    // driver name in ASIO_DRIVER by intercepting registry calls to
+    // HKLM\SOFTWARE\ASIO. Sentinel HKEY values mark the redirected handles
+    // so we can recognise them on subsequent reg* calls.
+
+    static const HKEY PARENT_ASIO_REG_HANDLE = reinterpret_cast<HKEY>(0x4001);
+    static const HKEY DEVICE_ASIO_REG_HANDLE = reinterpret_cast<HKEY>(0x4002);
+    static const char *FAKE_ASIO_DEVICE_NAME = "XONAR";
+
+    static decltype(RegCloseKey) *RegCloseKey_orig = nullptr;
+    static decltype(RegEnumKeyA) *RegEnumKeyA_orig = nullptr;
+    static decltype(RegOpenKeyA) *RegOpenKeyA_orig = nullptr;
+    static decltype(RegOpenKeyExA) *RegOpenKeyExA_orig = nullptr;
+    static decltype(RegQueryValueExA) *RegQueryValueExA_orig = nullptr;
+
+    static HKEY real_asio_reg_handle = nullptr;
+    static HKEY real_asio_device_reg_handle = nullptr;
+
+    static LONG WINAPI RegOpenKeyExA_hook(HKEY hKey, LPCSTR lpSubKey, DWORD ulOptions, REGSAM samDesired,
+            PHKEY phkResult)
+    {
+        if (ASIO_DRIVER.has_value() &&
+            lpSubKey != nullptr &&
+            phkResult != nullptr &&
+            hKey == PARENT_ASIO_REG_HANDLE &&
+            _stricmp(lpSubKey, FAKE_ASIO_DEVICE_NAME) == 0) {
+
+            *phkResult = DEVICE_ASIO_REG_HANDLE;
+
+            log_info("gitadora::asio", "replacing '{}' with '{}'", lpSubKey, ASIO_DRIVER.value());
+            const auto result = RegOpenKeyExA_orig(
+                    real_asio_reg_handle,
+                    ASIO_DRIVER.value().c_str(),
+                    ulOptions,
+                    samDesired,
+                    &real_asio_device_reg_handle);
+
+            if (result != ERROR_SUCCESS) {
+                log_warning(
+                    "gitadora::asio",
+                    "failed to open registry subkey '{}', error=0x{:x}",
+                    ASIO_DRIVER.value(), result);
+                log_warning(
+                    "gitadora::asio",
+                    "due to improper ASIO setting, audio init will fail");
+            }
+
+            return result;
+        }
+
+        return RegOpenKeyExA_orig(hKey, lpSubKey, ulOptions, samDesired, phkResult);
+    }
+
+    static LONG WINAPI RegOpenKeyA_hook(HKEY hKey, LPCSTR lpSubKey, PHKEY phkResult) {
+        if (ASIO_DRIVER.has_value() &&
+            lpSubKey != nullptr &&
+            phkResult != nullptr &&
+            hKey == HKEY_LOCAL_MACHINE &&
+            _stricmp(lpSubKey, "software\\asio") == 0)
+        {
+            *phkResult = PARENT_ASIO_REG_HANDLE;
+            return RegOpenKeyA_orig(hKey, lpSubKey, &real_asio_reg_handle);
+        }
+
+        return RegOpenKeyA_orig(hKey, lpSubKey, phkResult);
+    }
+
+    static LONG WINAPI RegEnumKeyA_hook(HKEY hKey, DWORD dwIndex, LPSTR lpName, DWORD cchName) {
+        if (hKey == PARENT_ASIO_REG_HANDLE && ASIO_DRIVER.has_value()) {
+            if (dwIndex == 0) {
+                // forward to real handle just to verify the key exists; we
+                // overwrite the name with our fake driver string regardless
+                auto ret = RegEnumKeyA_orig(real_asio_reg_handle, dwIndex, lpName, cchName);
+                if (ret == ERROR_SUCCESS && lpName != nullptr && cchName > 0) {
+                    log_info("gitadora::asio", "stubbing '{}' with '{}'", lpName, FAKE_ASIO_DEVICE_NAME);
+                    strncpy(lpName, FAKE_ASIO_DEVICE_NAME, cchName);
+                    lpName[cchName - 1] = '\0';
+                }
+                return ret;
+            } else {
+                return ERROR_NO_MORE_ITEMS;
+            }
+        }
+
+        return RegEnumKeyA_orig(hKey, dwIndex, lpName, cchName);
+    }
+
+    static LONG WINAPI RegQueryValueExA_hook(HKEY hKey, LPCSTR lpValueName, LPDWORD lpReserved, LPDWORD lpType,
+            LPBYTE lpData, LPDWORD lpcbData)
+    {
+        HKEY target = hKey;
+
+        if (ASIO_DRIVER.has_value() &&
+            lpValueName != nullptr &&
+            lpData != nullptr &&
+            lpcbData != nullptr &&
+            hKey == DEVICE_ASIO_REG_HANDLE) {
+            if (_stricmp(lpValueName, "Description") == 0) {
+                // engine may verify the driver name after open; ensure it still
+                // sees something containing "XONAR" so the substring check passes
+                const size_t len = strlen(FAKE_ASIO_DEVICE_NAME) + 1;
+                if (*lpcbData < len) {
+                    *lpcbData = static_cast<DWORD>(len);
+                    return ERROR_MORE_DATA;
+                }
+                memcpy(lpData, FAKE_ASIO_DEVICE_NAME, len);
+                *lpcbData = static_cast<DWORD>(len);
+                if (lpType != nullptr) {
+                    *lpType = REG_SZ;
+                }
+                return ERROR_SUCCESS;
+            }
+
+            // for everything else (CLSID etc.) defer to the real driver subkey
+            target = real_asio_device_reg_handle;
+        }
+
+        return RegQueryValueExA_orig(target, lpValueName, lpReserved, lpType, lpData, lpcbData);
+    }
+
+    static LONG WINAPI RegCloseKey_hook(HKEY hKey) {
+        if (hKey == PARENT_ASIO_REG_HANDLE) {
+            if (real_asio_reg_handle != nullptr) {
+                RegCloseKey_orig(real_asio_reg_handle);
+                real_asio_reg_handle = nullptr;
+            }
+            return ERROR_SUCCESS;
+        }
+
+        if (hKey == DEVICE_ASIO_REG_HANDLE) {
+            if (real_asio_device_reg_handle != nullptr) {
+                RegCloseKey_orig(real_asio_device_reg_handle);
+                real_asio_device_reg_handle = nullptr;
+            }
+            return ERROR_SUCCESS;
+        }
+
+        return RegCloseKey_orig(hKey);
+    }
+
+    void asio_hook_init() {
+        if (!ASIO_DRIVER.has_value()) {
+            return;
+        }
+
+        log_info("gitadora::asio", "installing ASIO driver redirect: XONAR -> {}", ASIO_DRIVER.value());
+
+        RegCloseKey_orig = detour::iat_try(
+                "RegCloseKey", RegCloseKey_hook, avs::game::DLL_INSTANCE);
+        RegEnumKeyA_orig = detour::iat_try(
+                "RegEnumKeyA", RegEnumKeyA_hook, avs::game::DLL_INSTANCE);
+        RegOpenKeyA_orig = detour::iat_try(
+                "RegOpenKeyA", RegOpenKeyA_hook, avs::game::DLL_INSTANCE);
+        RegOpenKeyExA_orig = detour::iat_try(
+                "RegOpenKeyExA", RegOpenKeyExA_hook, avs::game::DLL_INSTANCE);
+        RegQueryValueExA_orig = detour::iat_try(
+                "RegQueryValueExA", RegQueryValueExA_hook, avs::game::DLL_INSTANCE);
+    }
+}

src/spice2x/games/gitadora/asio.h

@@ -0,0 +1,12 @@
+#pragma once
+
+namespace games::gitadora {
+
+    // installs IAT registry hooks in gfdm.dll that redirect the game's
+    // ASIO driver lookup (hard-coded "XONAR" substring) to a user-chosen
+    // driver name read from games::gitadora::ASIO_DRIVER.
+    //
+    // safe to call unconditionally; if ASIO_DRIVER is unset the hooks
+    // forward every call straight through to advapi32.
+    void asio_hook_init();
+}

src/spice2x/games/gitadora/gitadora.cpp

@@ -1,18 +1,27 @@
 #include "gitadora.h"
+#include "asio.h"
 #include "handle.h"
 #include "bi2x_hook.h"
 #include <unordered_map>
+
+#include <ks.h>
+#include <ksmedia.h>
+
 #include "cfg/configurator.h"
+#include "hooks/audio/audio.h"
 #include "hooks/audio/mme.h"
 #include "hooks/graphics/graphics.h"
 #include "misc/wintouchemu.h"
 #include "overlay/overlay.h"
 #include "util/cpuutils.h"
+#include "util/deferlog.h"
 #include "util/detour.h"
 #include "util/libutils.h"
 #include "util/logging.h"
 #include "util/sigscan.h"
 #include "util/socd_cleaner.h"
+#include "util/sysutils.h"
+#include "util/utils.h"
 #include "hooks/setupapihook.h"
 
 namespace games::gitadora {
@@ -20,11 +29,13 @@ namespace games::gitadora {
     // settings
     bool TWOCHANNEL = false;
     std::optional<unsigned int> CAB_TYPE = std::nullopt;
-    bool ARENA_SINGLE_WINDOW = false;
     bool P1_LEFTY = false;
     bool P2_LEFTY = false;
     std::optional<std::string> SUBSCREEN_OVERLAY_SIZE;
     std::optional<socd::SocdAlgorithm> PICK_ALGO = socd::SocdAlgorithm::PreferRecent;
+    std::optional<uint8_t> ARENA_WINDOW_COUNT = std::nullopt;
+    std::optional<std::string> ASIO_DRIVER = std::nullopt;
+    bool ALLOW_REALTEK_AUDIO = false;
 
     /*
      * Prevent GitaDora from creating folders on F drive
@@ -221,13 +232,6 @@ namespace games::gitadora {
             }
 #endif
 
-            // arena model launches a tiny window yet backbuffer at 4k, resulting in unusable overlay
-            // force scaling to make things usable
-            if (!overlay::UI_SCALE_PERCENT.has_value() && is_arena_model() &&
-                GRAPHICS_WINDOWED && !cfg::CONFIGURATOR_STANDALONE) {
-                overlay::UI_SCALE_PERCENT = 250;
-            }
-
             // for guitar wail SOCD cleaning
             socd::ALGORITHM = socd::SocdAlgorithm::PreferRecent;
 
@@ -237,12 +241,334 @@ namespace games::gitadora {
             } else {
                 log_info("gitadora", "guitar pick SOCD algorithm: legacy");
             }
+
+#if SPICE64 && !SPICE_XP
+
+            if (is_arena_model()) {
+                // in full screen, if single-adapter option is checked, it's functionally
+                // the same as forcing a single monitor
+                if (!GRAPHICS_WINDOWED && GRAPHICS_FORCE_SINGLE_ADAPTER) {
+                    ARENA_WINDOW_COUNT = 1;
+                }
+
+                // figure out default settings if user didn't provide one
+                if (!ARENA_WINDOW_COUNT.has_value()) {
+                    if (!GRAPHICS_WINDOWED && sysutils::enumerate_monitors().size() < 4) {
+                        log_info("gitadora", "arena model: <4 monitors, defaulting to single window mode");
+                        ARENA_WINDOW_COUNT = 1;
+                    } else {
+                        ARENA_WINDOW_COUNT = 4;
+                    }
+                }
+
+                const int count = ARENA_WINDOW_COUNT.value();
+                switch (count) {
+                    case 1:
+                        log_info("gitadora", "arena model: single-window mode");
+                        GRAPHICS_FORCE_SINGLE_ADAPTER = true;
+                        GRAPHICS_PREVENT_SECONDARY_WINDOWS = true;
+                        break;
+                    case 2:
+                        if (!GRAPHICS_WINDOWED) {
+                            log_fatal(
+                                "gitadora",
+                                "arena model: 2-window mode is not supported in fullscreen, choose 1 or 4");
+                        }
+                        log_info("gitadora", "arena model: two-window mode");
+                        GRAPHICS_GITADORA_HIDE_SIDE_WINDOWS = true;
+                        break;
+                    case 4:
+                        log_info("gitadora", "arena model: four-window mode");
+                        break;
+                    default:
+                        log_fatal(
+                            "gitadora",
+                            "arena model: unsupported window count: {}", count);
+                }
+            }
+
+#endif
+
         }
     }
 
+#if SPICE64 && !SPICE_XP
+
+    static decltype(GetDisplayConfigBufferSizes) *GetDisplayConfigBufferSizes_orig = nullptr;
+    static decltype(QueryDisplayConfig) *QueryDisplayConfig_orig = nullptr;
+    static decltype(DisplayConfigGetDeviceInfo) *DisplayConfigGetDeviceInfo_orig = nullptr;
+
+    // cached primary real monitor: its path + source/target mode entries.
+    // modeInfoIdx values on the path are renumbered to 0 / 1 so the cache is self-contained.
+    static DISPLAYCONFIG_PATH_INFO real_primary_path = {};
+    static DISPLAYCONFIG_MODE_INFO real_primary_modes[2] = {}; // [0]=source, [1]=target
+
+    // fake monitors appended after the real ones. the game classifies monitors
+    // by outputTechnology + connectorInstance:
+    //   HDMI               -> main 4k monitor (real primary)
+    //   DP connInstance 0  -> left
+    //   DP connInstance 1  -> right
+    //   DP connInstance 2  -> small (sub/touch)
+    // ids are negated on the fake monitor headers so they can be distinguished
+    // from real ones in DisplayConfigGetDeviceInfo.
+    struct FakeMonitor {
+        LONG id;
+        int width;
+        int height;
+        int offset_x;
+        int offset_y;
+        UINT32 connector_instance;
+    };
+
+    // ORDERING MATTERS: the d3d9 wrapper (FAKE_SUBSCREEN_ADAPTER) hands out
+    // adapters as "\\.\DISPLAY_SPICE_FAKE_{N}" for N=1,2,3. The game maps each
+    // adapter to a swap chain role via its DisplayConfig connector instance,
+    // so the entries here must be listed in the same order the wrapper enumerates
+    // them: id=1 -> adapter 1 -> left, id=2 -> adapter 2 -> right, id=3 -> adapter 3 -> small.
+    static constexpr FakeMonitor FAKE_MONITORS[] = {
+        { 1, 1080, 1920, -100000, -100000, 0 }, // left  (DP connector instance 0)
+        { 2, 1080, 1920, -200000, -200000, 1 }, // right (DP connector instance 1)
+        { 3,  800, 1280, -300000, -300000, 2 }, // small (DP connector instance 2, touch)
+    };
+    static constexpr UINT32 FAKE_MONITOR_COUNT = static_cast<UINT32>(std::size(FAKE_MONITORS));
+
+    // call QueryDisplayConfig once, keep only the primary monitor's path and its
+    // two referenced modes (source + target). modeInfoIdx values are rewritten to
+    // 0 and 1 so the cache is self-consistent.
+    static void cache_primary_monitor_info() {
+        UINT32 path_count = 0;
+        UINT32 mode_count = 0;
+        if (GetDisplayConfigBufferSizes_orig(
+                QDC_ONLY_ACTIVE_PATHS, &path_count, &mode_count) != ERROR_SUCCESS) {
+            log_fatal("gitadora", "cache_primary_monitor_info: GetDisplayConfigBufferSizes failed");
+        }
+
+        std::vector<DISPLAYCONFIG_PATH_INFO> all_paths(path_count);
+        std::vector<DISPLAYCONFIG_MODE_INFO> all_modes(mode_count);
+        if (QueryDisplayConfig_orig(
+                QDC_ONLY_ACTIVE_PATHS,
+                &path_count, all_paths.data(),
+                &mode_count, all_modes.data(),
+                nullptr) != ERROR_SUCCESS) {
+            log_fatal("gitadora", "cache_primary_monitor_info: QueryDisplayConfig failed");
+        }
+        all_paths.resize(path_count);
+        all_modes.resize(mode_count);
+
+        // pick the primary monitor: source mode at (0, 0)
+        auto primary = std::find_if(all_paths.begin(), all_paths.end(),
+            [&](const auto &p) {
+                const auto idx = p.sourceInfo.modeInfoIdx;
+                return idx < all_modes.size() &&
+                       all_modes[idx].infoType == DISPLAYCONFIG_MODE_INFO_TYPE_SOURCE &&
+                       all_modes[idx].sourceMode.position.x == 0 &&
+                       all_modes[idx].sourceMode.position.y == 0;
+            });
+        if (primary == all_paths.end()) {
+            log_fatal("gitadora", "cache_primary_monitor_info: no primary monitor found");
+        }
+
+        real_primary_modes[0] = all_modes[primary->sourceInfo.modeInfoIdx];
+        real_primary_modes[1] = all_modes[primary->targetInfo.modeInfoIdx];
+        real_primary_path = *primary;
+        real_primary_path.sourceInfo.modeInfoIdx = 0;
+        real_primary_path.targetInfo.modeInfoIdx = 1;
+
+        log_info("gitadora", "cache_primary_monitor_info: cached primary monitor");
+    }
+
+    static
+    LONG
+    WINAPI
+    GetDisplayConfigBufferSizes_hook(
+        UINT32 Flags,
+        UINT32 *pNumPathArrayElements,
+        UINT32 *pNumModeInfoArrayElements)
+    {
+        // populate cached primary real monitor on the first call
+        static std::once_flag populate_once;
+        std::call_once(populate_once, cache_primary_monitor_info);
+
+        // always report exactly 1 real + N fake monitors
+        *pNumPathArrayElements = 1 + FAKE_MONITOR_COUNT;
+        *pNumModeInfoArrayElements = 2 + FAKE_MONITOR_COUNT * 2;
+
+        log_info(
+            "gitadora",
+            "GetDisplayConfigBufferSizes: 1 real path + {} fake monitor(s)",
+            FAKE_MONITOR_COUNT);
+
+        return ERROR_SUCCESS;
+    }
+
+    // write fake monitor i into the caller's path/mode arrays. layout (single-monitor
+    // assumption): index 0 in both arrays holds the cached primary real monitor, so
+    // fake i occupies path slot (1 + i) and mode slots (2 + i*2) / (2 + i*2 + 1).
+    static void insert_fake_monitor(
+        DISPLAYCONFIG_PATH_INFO *paths,
+        DISPLAYCONFIG_MODE_INFO *modes,
+        UINT32 i)
+    {
+        const FakeMonitor &m = FAKE_MONITORS[i];
+        const UINT32 src_idx = 2 + i * 2;
+        const UINT32 tgt_idx = src_idx + 1;
+        const LUID adapter_id { .LowPart = static_cast<DWORD>(-m.id), .HighPart = -m.id };
+        const UINT32 uid = static_cast<UINT32>(-m.id);
+
+        paths[1 + i] = {
+            .sourceInfo = {
+                .adapterId = adapter_id,
+                .id = uid,
+                .modeInfoIdx = src_idx,
+                .statusFlags = DISPLAYCONFIG_SOURCE_IN_USE,
+            },
+            .targetInfo = {
+                .adapterId = adapter_id,
+                .id = uid,
+                .modeInfoIdx = tgt_idx,
+                .outputTechnology = DISPLAYCONFIG_OUTPUT_TECHNOLOGY_DISPLAYPORT_EXTERNAL,
+                .rotation = DISPLAYCONFIG_ROTATION_IDENTITY,
+                .scaling = DISPLAYCONFIG_SCALING_IDENTITY,
+                .refreshRate = { .Numerator = 60000, .Denominator = 1000 },
+                .scanLineOrdering = DISPLAYCONFIG_SCANLINE_ORDERING_PROGRESSIVE,
+                .targetAvailable = TRUE,
+                .statusFlags = DISPLAYCONFIG_TARGET_IN_USE,
+            },
+            .flags = DISPLAYCONFIG_PATH_ACTIVE,
+        };
+
+        modes[src_idx] = {
+            .infoType = DISPLAYCONFIG_MODE_INFO_TYPE_SOURCE,
+            .id = uid,
+            .adapterId = adapter_id,
+            .sourceMode = {
+                .width = static_cast<UINT32>(m.width),
+                .height = static_cast<UINT32>(m.height),
+                .pixelFormat = DISPLAYCONFIG_PIXELFORMAT_32BPP,
+                .position = { .x = m.offset_x, .y = m.offset_y },
+            },
+        };
+
+        modes[tgt_idx] = {
+            .infoType = DISPLAYCONFIG_MODE_INFO_TYPE_TARGET,
+            .id = uid,
+            .adapterId = adapter_id,
+            .targetMode = {},
+        };
+
+        log_misc(
+            "gitadora",
+            "inserted fake monitor: id={}, width={}, height={}, offset_x={}, offset_y={}",
+            m.id, m.width, m.height, m.offset_x, m.offset_y);
+    }
+
+    static
+    LONG
+    WINAPI
+    QueryDisplayConfig_hook(
+        UINT32 flags,
+        UINT32* numPathArrayElements,
+        DISPLAYCONFIG_PATH_INFO* pathArray,
+        UINT32* numModeInfoArrayElements,
+        DISPLAYCONFIG_MODE_INFO* modeInfoArray,
+        DISPLAYCONFIG_TOPOLOGY_ID* currentTopologyId)
+    {
+        // copy cached primary real monitor into caller buffers at index 0
+        pathArray[0] = real_primary_path;
+        modeInfoArray[0] = real_primary_modes[0];
+        modeInfoArray[1] = real_primary_modes[1];
+        *numPathArrayElements = 1 + FAKE_MONITOR_COUNT;
+        *numModeInfoArrayElements = 2 + FAKE_MONITOR_COUNT * 2;
+        if (currentTopologyId != nullptr) {
+            *currentTopologyId = DISPLAYCONFIG_TOPOLOGY_EXTEND;
+        }
+
+        log_misc("gitadora", "QueryDisplayConfig returning fake monitor paths and modes");
+
+        // append fake monitors after the real one
+        for (UINT32 i = 0; i < FAKE_MONITOR_COUNT; i++) {
+            insert_fake_monitor(pathArray, modeInfoArray, i);
+        }
+
+        return ERROR_SUCCESS;
+    }
+
+    static
+    LONG
+    WINAPI
+    DisplayConfigGetDeviceInfo_hook(DISPLAYCONFIG_DEVICE_INFO_HEADER* requestPacket)
+    {
+        if (requestPacket == nullptr) {
+            return DisplayConfigGetDeviceInfo_orig(requestPacket);
+        }
+
+        // handle fake monitors (negative id) directly without calling orig
+        const auto id = static_cast<int>(requestPacket->id);
+        if (id < 0) {
+            if (requestPacket->type == DISPLAYCONFIG_DEVICE_INFO_GET_SOURCE_NAME) {
+                const auto sourceName = reinterpret_cast<DISPLAYCONFIG_SOURCE_DEVICE_NAME*>(requestPacket);
+                // name must match WrappedIDirect3D9::GetAdapterIdentifier
+                const std::string adapter_name = fmt::format("\\\\.\\DISPLAY_SPICE_FAKE_{}", -id);
+                wcscpy(sourceName->viewGdiDeviceName, s2ws(adapter_name).c_str());
+                log_misc("gitadora",
+                    "DisplayConfigGetDeviceInfo: fake source id={} name={}", id, adapter_name);
+                return ERROR_SUCCESS;
+            }
+            if (requestPacket->type == DISPLAYCONFIG_DEVICE_INFO_GET_TARGET_NAME) {
+                const auto targetName = reinterpret_cast<DISPLAYCONFIG_TARGET_DEVICE_NAME*>(requestPacket);
+                const LONG fake_id = -id;
+                UINT32 conn_inst = 0xff;
+                for (const auto& f : FAKE_MONITORS) {
+                    if (f.id == fake_id) {
+                        conn_inst = f.connector_instance;
+                        break;
+                    }
+                }
+                targetName->outputTechnology = DISPLAYCONFIG_OUTPUT_TECHNOLOGY_DISPLAYPORT_EXTERNAL;
+                targetName->connectorInstance = conn_inst;
+                wcscpy(targetName->monitorFriendlyDeviceName, L"Spice Fake Monitor");
+                wcscpy(targetName->monitorDevicePath, L"\\\\?\\SpiceFakeMonitor");
+                log_misc("gitadora",
+                    "DisplayConfigGetDeviceInfo: fake target id={} -> DP connInst {}",
+                    id, targetName->connectorInstance);
+                return ERROR_SUCCESS;
+            }
+        }
+
+        const auto ret = DisplayConfigGetDeviceInfo_orig(requestPacket);
+        if (ret != ERROR_SUCCESS ||
+            requestPacket->type != DISPLAYCONFIG_DEVICE_INFO_GET_TARGET_NAME) {
+            return ret;
+        }
+
+        // override the cached primary real monitor target info to look like HDMI/0
+        const auto targetName = reinterpret_cast<DISPLAYCONFIG_TARGET_DEVICE_NAME*>(requestPacket);
+        const auto& target = real_primary_path.targetInfo;
+        if (target.id == targetName->header.id &&
+            target.adapterId.HighPart == targetName->header.adapterId.HighPart &&
+            target.adapterId.LowPart == targetName->header.adapterId.LowPart)
+        {
+            targetName->outputTechnology = DISPLAYCONFIG_OUTPUT_TECHNOLOGY_HDMI;
+            targetName->connectorInstance = 0;
+            log_info("gitadora",
+                "overriding primary monitor (id={}) to pretend to be HDMI",
+                targetName->header.id);
+        }
+        return ret;
+    }
+
+#endif
+
     void GitaDoraGame::attach() {
         Game::attach();
 
+        // arena model launches a tiny window yet backbuffer at 4k, resulting in unusable overlay
+        // force scaling to make things usable
+        if (!overlay::UI_SCALE_PERCENT.has_value() && is_arena_model()) {
+            log_info("gitadora", "forcing UI scale to 250% for arena model");
+            overlay::UI_SCALE_PERCENT = 250;
+        }
+
         // modules
         HMODULE sharepj_module = libutils::try_module("libshare-pj.dll");
         HMODULE bmsd2_module = libutils::try_module("libbmsd2.dll");
@@ -269,6 +595,12 @@ namespace games::gitadora {
 #ifdef SPICE64
         // gitadora arena model 
         auto aio = libutils::try_library("libaio.dll");
+
+        // before we start patching and hooking things, detect invalid configuration
+        if (aio != nullptr && !is_arena_model()) {
+            log_fatal("gitadora", "arena model i/o (libaio.dll) detected, but <spec> is not an arena model - bad prop XML files?");
+        }
+
         if (aio != nullptr) {
             SETUPAPI_SETTINGS settings{};
             settings.class_guid[0] = 0x86E0D1E0;
@@ -294,39 +626,76 @@ namespace games::gitadora {
             detour::iat_try("GetDriveTypeA", GetDriveTypeA_hook, avs::game::DLL_INSTANCE);
             detour::iat_try("CreateDirectoryA", CreateDirectoryA_hook, avs::game::DLL_INSTANCE);
 
+            // ASIO driver redirect (XONAR -> user-configured driver)
+            asio_hook_init();
+
             // volume change prevention
             hooks::audio::mme::init(avs::game::DLL_INSTANCE);
 
-            // touch hook
-            if (ARENA_SINGLE_WINDOW) {
+            // fake Realtek audio injection
+            // if ASIO init succeeds, game tries to look for audio device with `Realtek` in friendly name
+            // if ASIO init fails, game opens default audio device
+            // therefore, it's safe to enable this hook by default regardless of ASIO preference
+            // (unless the user explicitly disables it, of course)
+            if (ALLOW_REALTEK_AUDIO) {
+                log_info(
+                    "gitadora",
+                    "fake Realtek audio injection disabled "
+                    "(user's real Realtek audio may be used after successful ASIO init)");
+            } else {
+                log_info(
+                    "gitadora",
+                    "fake Realtek audio injection enabled "
+                    "(create a fake Realtek audio device to prevent crashes after successful ASIO init)");
+                hooks::audio::INJECT_FAKE_REALTEK_AUDIO = true;
+            }
+
+            // monitor/touch hooks (windowed or full screen)
+            if (GRAPHICS_PREVENT_SECONDARY_WINDOWS) {
+                // enable touch hook for subscreen overlay
                 wintouchemu::FORCE = true;
                 wintouchemu::INJECT_MOUSE_AS_WM_TOUCH = true;
                 wintouchemu::hook("GITADORA", avs::game::DLL_INSTANCE);
-            }
-            return;
-        }
 
-        HMODULE gdme_module = libutils::try_module("libgdme.dll");
+#if !SPICE_XP
+
+                if (!GRAPHICS_WINDOWED) {
+                    // monitor hook: always pretend to have 1 primary real monitor + 3 fake monitors
+                    // (LEFT / RIGHT / SMALL) so the game accepts the arena cab topology
+                    GetDisplayConfigBufferSizes_orig =
+                        detour::iat_try("GetDisplayConfigBufferSizes",
+                            GetDisplayConfigBufferSizes_hook, avs::game::DLL_INSTANCE);
+                    QueryDisplayConfig_orig =
+                        detour::iat_try("QueryDisplayConfig",
+                            QueryDisplayConfig_hook, avs::game::DLL_INSTANCE);
+                    DisplayConfigGetDeviceInfo_orig =
+                        detour::iat_try("DisplayConfigGetDeviceInfo",
+                            DisplayConfigGetDeviceInfo_hook, avs::game::DLL_INSTANCE);
+                }
+#endif
+
+            }
+        }
 
         // window patch
-        if (GRAPHICS_WINDOWED && !replace_pattern(
-                gdme_module,
-                "754185ED753D8B4118BF0000CB02",
-                "9090????9090??????????????12", 0, 0))
-        {
-            log_warning("gitadora", "windowed mode failed");
+        if (!is_arena_model()) {
+            HMODULE gdme_module = libutils::try_module("libgdme.dll");
+            if (GRAPHICS_WINDOWED && !replace_pattern(
+                    gdme_module,
+                    "754185ED753D8B4118BF0000CB02",
+                    "9090????9090??????????????12", 0, 0)) {
+                log_warning("gitadora", "windowed mode failed");
+            }
         }
 
-        HMODULE bmsd_engine_module = libutils::try_module("libbmsd-engine.dll");
-        HMODULE bmsd_module = libutils::try_module("libbmsd.dll");
-
         // two channel mod
-        if (TWOCHANNEL) {
-            bmsd2_boot_orig = detour::iat_try("bmsd2_boot", bmsd2_boot_hook, bmsd_module);
+        if (TWOCHANNEL && !is_arena_model()) {
+            HMODULE bmsd_engine_module = libutils::try_module("libbmsd-engine.dll");
+            HMODULE bmsd_module = libutils::try_module("libbmsd.dll");
 
+            bmsd2_boot_orig = detour::iat_try("bmsd2_boot", bmsd2_boot_hook, bmsd_module);
             if (!(replace_pattern(bmsd_engine_module, "33000000488D", "03??????????", 0, 0) ||
-                    replace_pattern(bmsd_engine_module, "330000000F10", "03??????????", 0, 0)))
-            {
+                    replace_pattern(bmsd_engine_module, "330000000F10", "03??????????", 0, 0))) {
                 log_warning("gitadora", "two channel mode failed");
             }
         }
@@ -334,4 +703,18 @@ namespace games::gitadora {
 #endif
 
     }
+
+    void fix_audio_channel_mask(WAVEFORMATEX *format) {
+        if (!format || format->wFormatTag != WAVE_FORMAT_EXTENSIBLE) {
+            return;
+        }
+
+        auto ext = reinterpret_cast<WAVEFORMATEXTENSIBLE *>(format);
+
+        // fix the legacy 7.1 channel mask to the modern surround layout
+        // makes it more compatible with modern audio cards
+        if (ext->dwChannelMask == KSAUDIO_SPEAKER_7POINT1) {
+            ext->dwChannelMask = KSAUDIO_SPEAKER_7POINT1_SURROUND;
+        }
+    }
 }

src/spice2x/games/gitadora/gitadora.h

@@ -2,6 +2,9 @@
 
 #include <optional>
 
+#include <windows.h>
+#include <mmreg.h>
+
 #include "avs/game.h"
 #include "games/game.h"
 #include "util/socd_cleaner.h"
@@ -11,11 +14,13 @@ namespace games::gitadora {
     // settings
     extern bool TWOCHANNEL;
     extern std::optional<unsigned int> CAB_TYPE;
-    extern bool ARENA_SINGLE_WINDOW;
     extern bool P1_LEFTY;
     extern bool P2_LEFTY;
     extern std::optional<std::string> SUBSCREEN_OVERLAY_SIZE;
     extern std::optional<socd::SocdAlgorithm> PICK_ALGO;
+    extern std::optional<uint8_t> ARENA_WINDOW_COUNT;
+    extern std::optional<std::string> ASIO_DRIVER;
+    extern bool ALLOW_REALTEK_AUDIO;
 
     class GitaDoraGame : public games::Game {
     public:
@@ -24,6 +29,8 @@ namespace games::gitadora {
         virtual void attach() override;
     };
 
+    void fix_audio_channel_mask(WAVEFORMATEX *format);
+
     static inline bool is_drum() {
         return (
             avs::game::is_model({ "J32", "K32", "L32" }) ||

src/spice2x/games/iidx/bi2a.cpp

@@ -2,6 +2,7 @@
 
 #include "misc/eamuse.h"
 #include "rawinput/rawinput.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"
 
 #include "iidx.h"
@@ -82,7 +83,8 @@ bool games::iidx::IIDXFMSerialHandle::FMSerialDevice::parse_msg(
             }
 
             // sleep - otherwise the IO thread will go too hard on the CPU
-            Sleep(1);
+            static thread_local timeutils::PreciseSleepTimer timer;
+            timer.sleep(1);
 
             // generate message
             auto msg = this->create_msg(msg_in, 0x2E);

src/spice2x/games/iidx/bi2x.cpp

@@ -2,6 +2,7 @@
 
 #include "misc/eamuse.h"
 #include "rawinput/rawinput.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"
 
 #include "iidx.h"
@@ -82,7 +83,8 @@ bool games::iidx::BI2XSerialHandle::BI2XDevice::parse_msg(
             }
 
             // sleep - otherwise the IO thread will go too hard on the CPU
-            Sleep(1);
+            static thread_local timeutils::PreciseSleepTimer timer;
+            timer.sleep(1);
 
             // generate message
             auto msg = this->create_msg(msg_in, 0x2E);

src/spice2x/games/iidx/iidx.cpp

@@ -543,11 +543,7 @@ namespace games::iidx {
         // if the user specified a value (other than auto), use it as the environment var
         // probably "wasapi" or "asio", but it's not explicitly checked here for forward compat
         if (SOUND_OUTPUT_DEVICE.has_value() && SOUND_OUTPUT_DEVICE.value() != "auto") {
-            log_info(
-                "iidx",
-                "using user-supplied \"{}\" for SOUND_OUTPUT_DEVICE",
-                SOUND_OUTPUT_DEVICE.value());
-            SetEnvironmentVariable("SOUND_OUTPUT_DEVICE", SOUND_OUTPUT_DEVICE.value().c_str());
+            // the environemnt variable was already set in launcher.cpp
             SOUND_OUTPUT_DEVICE_IN_EFFECT = SOUND_OUTPUT_DEVICE;
             return;
         }
@@ -783,42 +779,7 @@ namespace games::iidx {
         }
 
         // return higher 8 bit
-        uint8_t result = (uint8_t) (ret_value >> 2);
-
-        // delay
-        if ((player == 0 && TT_DELAY_P1 > 0) ||
-            (player == 1 && TT_DELAY_P2 > 0)) {
-
-            static std::queue<std::pair<double, uint8_t>> delay_queue[2];
-            auto &queue = delay_queue[player];
-
-            const auto max_delta_ms =
-                static_cast<double>((player == 0) ? TT_DELAY_P1 : TT_DELAY_P2);
-
-            // always push a new value
-            const auto now = get_performance_milliseconds();
-            queue.push(std::make_pair(now, result));
-            
-            // drain the queue down to reasonable length to prevent unconstrained growth
-            // this would accommodate 1 second at ~1000Hz
-            // (in reality all three iidx I/O emulation runs well under 500Hz)
-            while (queue.size() > 1024) {
-                queue.pop();
-            }
-
-            // pop until we find one that falls just under the time threshold
-            while (!queue.empty()) {
-                const auto delta_t = now - queue.front().first;
-                if (delta_t <= max_delta_ms) {
-                    break;
-                }
-                queue.pop();
-            }
-
-            result = queue.front().second;
-        }
-
-        return result;
+        return (uint8_t)(ret_value >> 2);
     }
 
     unsigned char get_slider(uint8_t slider) {
@@ -928,38 +889,10 @@ namespace games::iidx {
             }
         }
 
-        // patch iidx32+ for asio compatibility
-        // only do this if NOT wasapi (as opposed to checking if it's asio)
-        // the patch is only really needed for (some) non-XONAR devices but since people sometimes disguise
-        // other devices as a XONAR, don't check for the exact string (common ASIO workaround for INF)
-        if (avs::game::is_ext(2024090100, INT_MAX) &&
-            !(SOUND_OUTPUT_DEVICE_IN_EFFECT.has_value() &&
-              SOUND_OUTPUT_DEVICE_IN_EFFECT.value() == "wasapi")) {
-
-            // in iidx32 final:
-            // ff 50 08      call   QWORD PTR [rax+0x8]     ; ASIO instance AddRef
-            // 48 8b 4b 08   mov    rcx,QWORD PTR [rbx+0x8]
-            // 48 8b 01      mov    rax,QWORD PTR [rcx]
-            // ff 50 08      call   QWORD PTR [rax+0x8]     ; ASIO instance AddRef
-
-            intptr_t result = replace_pattern(
-                avs::game::DLL_INSTANCE,
-                "FF50????????????????FF50??4533C94533C0418D51",
-                "????????????????????909090??????????????????",
-                0, 0);
-
-            if (result == 0) {
-                log_warning(
-                    "iidx",
-                    "Failed to apply ASIO compatibility fix for iidx32+. "
-                    "Unless patches are applied, your ASIO device may hang or fail to work");
-            } else {
-                log_info(
-                    "iidx",
-                    "Successfully applied ASIO compatibility fix for iidx32+ using signature matching @ 0x{:x}.",
-                    result);
-            }
-        }
+        // note: the iidx32+ ASIO refcount bug (a duplicate AddRef on the ASIO instance with
+        // no matching Release, which leaks the driver and can hang non-XONAR devices) is now
+        // handled transparently by the WrappedAsio proxy (see hooks/audio/asio_proxy.cpp),
+        // so no game-DLL signature patch is needed here anymore
 
 #endif
 

src/spice2x/games/iidx/iidx.h

@@ -31,8 +31,6 @@ namespace games::iidx {
     extern bool NATIVE_TOUCH;
     extern std::optional<std::string> SOUND_OUTPUT_DEVICE;
     extern std::optional<std::string> ASIO_DRIVER;
-    extern uint32_t TT_DELAY_P1;
-    extern uint32_t TT_DELAY_P2;
     extern uint8_t DIGITAL_TT_SENS;
     extern std::optional<std::string> SUBSCREEN_OVERLAY_SIZE;
     extern std::optional<std::string> SCREEN_MODE;

src/spice2x/games/iidx/local_camera.cpp

@@ -3,6 +3,7 @@
 #if SPICE64 && !SPICE_XP
 
 #include "util/logging.h"
+#include "util/precise_timer.h"
 #include "util/utils.h"
 #include "mf_wrappers.h"
 
@@ -399,6 +400,7 @@ namespace games::iidx {
     void IIDXLocalCamera::CreateThread() {
         // Create thread
         m_drawThread = new std::thread([this]() {
+            timeutils::PreciseSleepTimer timer;
             SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_BELOW_NORMAL);
 
             double accumulator = 0.0;
@@ -412,7 +414,7 @@ namespace games::iidx {
                     accumulator -= 1.0;
                     floorFrameTimeMicroSec += 1;
                 }
-                std::this_thread::sleep_for(std::chrono::microseconds(floorFrameTimeMicroSec));
+                timer.sleep(std::chrono::microseconds(floorFrameTimeMicroSec));
             }
         });
     }

src/spice2x/games/iidx/poke.cpp

@@ -15,6 +15,7 @@
 #include "touch/touch.h"
 #include "util/libutils.h"
 #include "util/logging.h"
+#include "util/precise_timer.h"
 
 #define POKE_NATIVE_TOUCH 0
 
@@ -200,6 +201,7 @@ namespace games::iidx::poke {
         // create new thread
         THREAD_RUNNING = true;
         THREAD = new std::thread([] {
+            timeutils::PreciseSleepTimer timer;
 
             // log
             log_info("poke", "enabled");
@@ -319,7 +321,7 @@ namespace games::iidx::poke {
                 }
 
                 // slow down
-                Sleep(50);
+                timer.sleep(50);
             }
 
             return nullptr;

src/spice2x/games/io.cpp

@@ -444,6 +444,8 @@ namespace games {
         // overlay button definitions
         names.emplace_back("Screenshot");
         vkey_defaults.push_back(VK_SNAPSHOT);
+        names.emplace_back("Toggle All Windows");
+        vkey_defaults.push_back(VK_OEM_3); // backtick `
         names.emplace_back("Toggle Main Menu");
         vkey_defaults.push_back(VK_ESCAPE);
         names.emplace_back("Toggle Sub Screen");
@@ -468,7 +470,7 @@ namespace games {
         vkey_defaults.push_back(VK_F10);
         names.emplace_back("Toggle Screen Resize");
         vkey_defaults.push_back(VK_F11);
-        names.emplace_back("Toggle Overlay");
+        names.emplace_back("Toggle FPS");
         vkey_defaults.push_back(VK_F12);
         names.emplace_back("Toggle Camera Control");
         vkey_defaults.push_back(0xFF);

src/spice2x/games/io.h

@@ -8,6 +8,7 @@ namespace games {
     namespace OverlayButtons {
         enum {
             Screenshot,
+            ToggleAllWindows,
             ToggleMainMenu,
             ToggleSubScreen,
             InsertCoin,
@@ -20,7 +21,7 @@ namespace games {
             ToggleControl,
             TogglePatchManager,
             ToggleScreenResize,
-            ToggleOverlay,
+            ToggleFps,
             ToggleCameraControl,
             TriggerPinMacroP1,
             TriggerPinMacroP2,

src/spice2x/games/jb/jb.cpp

@@ -19,7 +19,7 @@
 namespace games::jb {
 
     // touch stuff
-    bool TOUCH_LEGACY_BOX = false;
+    JubeatTouchAlgorithm TOUCH_ALGORITHM = Improved;
     static bool TOUCH_ENABLE = false;
     static bool TOUCH_ATTACHED = false;
     static bool IS_PORTRAIT = true;
@@ -78,7 +78,7 @@ namespace games::jb {
         // note that the IO code in device.cpp will correctly compensate for orientation, depending on the model.
         TOUCH_POINTS.clear();
         touch_get_points(TOUCH_POINTS);
-        if (TOUCH_LEGACY_BOX) {
+        if (TOUCH_ALGORITHM == Legacy) {
             auto offset = IS_PORTRAIT ? 580 : 0;
             for (auto &tp : TOUCH_POINTS) {
 
@@ -94,6 +94,17 @@ namespace games::jb {
             }
         } else {
             for (auto &tp : TOUCH_POINTS) {
+                // check window out of bounds
+                if (IS_PORTRAIT) {
+                    if (tp.x > 768 || tp.y > 1360) {
+                        continue;
+                    }
+                } else {
+                    if (tp.x > 1360 || tp.y > 768) {
+                        continue;
+                    }
+                }
+
                 int x_relative = tp.x;
                 int y_relative = tp.y;
 
@@ -110,19 +121,66 @@ namespace games::jb {
                     y_relative -= 8;
                 }
 
-                if (x_relative < 0 || y_relative < 0) {
+                int x_index = -1;
+                int x_hitbox = 0;
+                int y_index = -1;
+                int y_hitbox = 0;
+
+                // x_hitbox and y_hitbox is relative to top-left pixel of each button
+                if (x_relative >= 0) {
+                    x_index = x_relative / JB_BUTTON_HITBOX;
+                    x_hitbox = x_relative % JB_BUTTON_HITBOX;
+                }
+                if (y_relative >= 0) {
+                    y_index = y_relative / JB_BUTTON_HITBOX;
+                    y_hitbox = y_relative % JB_BUTTON_HITBOX;
+                }
+
+                // log_info("jb", "raw={}, idx={}, hitbox={}", tp.x, x_index, x_hitbox);
+
+                if (TOUCH_ALGORITHM == Improved) {
+                    if (IS_PORTRAIT) {
+                        // extend to left border
+                        if (x_relative < 0) {
+                            x_index = 0;
+                        }
+                        // right and bottom borders are covered by the hit box
+                    } else {
+                        // extend to top border
+                        if (y_relative < 0) {
+                            y_index = 0;
+                        }
+                        // extend to left border
+                        if (x_relative < 0) {
+                            x_index = 0;
+                        }
+                        // bottom border is covered by the hit box
+                        // rightmost edge - ignore them entirely
+                        if (x_index == 3 && JB_BUTTON_SIZE < x_hitbox) {
+                            continue;
+                        }
+                    }
+                }
+
+                if (x_index < 0 || y_index < 0 || x_index > 3 || y_index > 3) {
                     continue;
                 }
 
-                // x_hitbox and y_hitbox is relative to top-left pixel of each button
-                int x_index = x_relative / JB_BUTTON_HITBOX;
-                int x_hitbox = x_relative % JB_BUTTON_HITBOX;
-                int y_index = y_relative / JB_BUTTON_HITBOX;
-                int y_hitbox = y_relative % JB_BUTTON_HITBOX;
-
                 // check if the gap was touched
-                if (x_hitbox > JB_BUTTON_SIZE || y_hitbox > JB_BUTTON_SIZE) {
-                    continue;
+                if (TOUCH_ALGORITHM == AcAccurate) {
+                    // in ac-accurate mode, touching the gap is ignored
+                    if (x_hitbox > JB_BUTTON_SIZE || y_hitbox > JB_BUTTON_SIZE) {
+                        continue;
+                    }
+
+                } else if (TOUCH_ALGORITHM == Improved) {
+                    // in improved mode, touching the gap triggers the closest button
+                    if (x_index <= 2 && (JB_BUTTON_SIZE + JB_BUTTON_GAP / 2) < x_hitbox) {
+                        x_index++;
+                    }
+                    if (y_index <= 2 && (JB_BUTTON_SIZE + JB_BUTTON_GAP / 2) < y_hitbox) {
+                        y_index++;
+                    }
                 }
 
                 // set the corresponding state
@@ -193,6 +251,21 @@ namespace games::jb {
         // enable touch
         TOUCH_ENABLE = true;
 
+        switch (TOUCH_ALGORITHM) {
+            case Legacy:
+                log_info("jubeat", "using 'legacy' touch targets");
+                break;
+            case Improved:
+                log_info("jubeat", "using 'improved' touch targets");
+                break;
+            case AcAccurate:
+                log_info("jubeat", "using 'ac accurate' touch targets");
+                break;
+            default:
+                log_fatal("jubeat", "unknown touch algo detected in touch_update, this is a bug");
+                break;
+        }
+
         // enable debug logging of gftools
         HMODULE gftools = libutils::try_module("gftools.dll");
         detour::inline_hook((void *) GFDbgSetReportFunc, libutils::try_proc(

src/spice2x/games/jb/jb.h

@@ -4,8 +4,14 @@
 
 namespace games::jb {
 
+    enum JubeatTouchAlgorithm {
+        Legacy,
+        Improved,
+        AcAccurate
+    };
+
     // touch stuff
-    extern bool TOUCH_LEGACY_BOX;
+    extern JubeatTouchAlgorithm TOUCH_ALGORITHM;
     extern bool TOUCH_STATE[16];
     void touch_update();
 

src/spice2x/games/nost/poke.cpp

@@ -9,6 +9,7 @@
 #include "misc/eamuse.h"
 #include "touch/touch.h"
 #include "util/logging.h"
+#include "util/precise_timer.h"
 
 namespace games::nost::poke {
 
@@ -63,6 +64,7 @@ namespace games::nost::poke {
         // create new thread
         THREAD_RUNNING = true;
         THREAD = new std::thread([] {
+            timeutils::PreciseSleepTimer timer;
             const DWORD touch_id = (DWORD)(0xFFFFFFFE);
 
             const int swipe_anim_total_frames = 6;
@@ -178,7 +180,7 @@ namespace games::nost::poke {
                 }
 
                 // slow down
-                Sleep(30);
+                timer.sleep(30);
             }
 
             return nullptr;

src/spice2x/games/onpara/touchpanel.cpp

@@ -4,6 +4,7 @@
 #include "touchpanel.h"
 
 #include "util/logging.h"
+#include "util/precise_timer.h"
 #include "touch/touch.h"
 
 using namespace std::chrono_literals;
@@ -44,6 +45,9 @@ bool games::onpara::TouchPanelHandle::open(LPCWSTR lpFileName) {
 }
 
 int games::onpara::TouchPanelHandle::read(LPVOID lpBuffer, DWORD nNumberOfBytesToRead) {
+
+    static thread_local timeutils::PreciseSleepTimer timer;
+
     DWORD i;
     auto buffer = reinterpret_cast<uint8_t *>(lpBuffer);
 
@@ -67,7 +71,7 @@ int games::onpara::TouchPanelHandle::read(LPVOID lpBuffer, DWORD nNumberOfBytesT
         enqueue_packet(report);
 
         // prevent cpu bullying
-        std::this_thread::sleep_for(1ms);
+        timer.sleep(1);
     }
 
     // copy from output queue
@@ -133,4 +137,4 @@ bool games::onpara::TouchPanelHandle::close() {
     log_info("touchpanel", "Closed COM1 (Touch Panel)");
 
     return true;
-}
+}

src/spice2x/games/otoca/otoca.cpp

@@ -4,6 +4,7 @@
 #include "util/detour.h"
 #include "util/libutils.h"
 #include "util/logging.h"
+#include "util/sigscan.h"
 
 #define OTOCA_DEBUG_VERBOSE 0
 #if OTOCA_DEBUG_VERBOSE
@@ -116,6 +117,73 @@ namespace games::otoca {
         return ret;
     }
 
+    static void patch_holo_print_hang() {
+
+        // arkkep.dll's print method sets the printer busy (0x66 / Printing_Busy),
+        // then bails out with -1 WITHOUT printing when its holo parameter is
+        // nonzero (the card path passes 0 and works). The completion callback that
+        // clears busy never fires, so the page's poll loop spins forever -> hang.
+        // The image band is already built, so flip the je that skips the bail-out
+        // into a jmp (74 -> EB) to route holo through the print path like card.
+        //
+        //   je   issue_print     ; <- patched to jmp
+        //   or   eax, -1         ; holo: return without printing
+        //   ret  10h
+        auto arkkep = libutils::try_module("arkkep.dll");
+        if (arkkep == nullptr) {
+            log_warning("otoca", "arkkep.dll not loaded; skipping holo print hang fix");
+            return;
+        }
+
+        auto result = replace_pattern(
+            arkkep,
+            "C7465866000000895DC0C74710080700007416",
+            "C7465866000000895DC0C7471008070000EB16",
+            0, 0);
+        if (result) {
+            log_info("otoca", "patched hologram print hang in arkkep.dll");
+        } else {
+            log_warning("otoca", "could not patch hologram print hang (incompatible arkkep.dll?)");
+        }
+    }
+
+    static void patch_holo_print_search() {
+
+        // before printing, arkkep walks the page's printer list looking for one
+        // whose capability flag (byte at entry+120h) matches the job type
+        // (0 = card, 1 = holo). Our emulated printer is only card-capable, so a
+        // holo job finds no match and aborts without printing ("not found holo
+        // printer"). The emulated printer serves both media, so force the compare
+        // to always match by turning `cmp cl,[edi+120h]` into `cmp cl,cl` + NOPs,
+        // which always sets ZF. Two such compares exist, so patch both.
+        //
+        //   cmp  cl, [edi+120h]  ; <- patched to `cmp cl,cl` (3A 8F.. -> 3A C9..)
+        //   je   printer_found
+        auto arkkep = libutils::try_module("arkkep.dll");
+        if (arkkep == nullptr) {
+            log_warning("otoca", "arkkep.dll not loaded; skipping holo print search fix");
+            return;
+        }
+
+        int patched = 0;
+        while (patched < 2) {
+            auto result = replace_pattern(
+                arkkep,
+                "3A8F20010000",
+                "3AC990909090",
+                0, 0);
+            if (!result) {
+                break;
+            }
+            patched++;
+        }
+        if (patched > 0) {
+            log_info("otoca", "patched hologram printer search in arkkep.dll ({} site(s))", patched);
+        } else {
+            log_warning("otoca", "could not patch hologram printer search (incompatible arkkep.dll?)");
+        }
+    }
+
     void OtocaGame::attach() {
         Game::attach();
 
@@ -129,6 +197,12 @@ namespace games::otoca {
         p4io_hook();
         games::shared::printer_attach();
 
+        // fix hologram print hard hang in arkkep.dll
+        patch_holo_print_hang();
+
+        // make hologram jobs find the emulated printer (else they abort unprinted)
+        patch_holo_print_search();
+
         if (BYPASS_CAMERA) {
             libutils::try_library("libcamera.dll");
             const auto libcamera = "libcamera.dll";

src/spice2x/games/popn/popn.cpp

@@ -261,7 +261,7 @@ namespace games::popn {
             } else if (requestPacket->type == DISPLAYCONFIG_DEVICE_INFO_GET_SOURCE_NAME) {
                 const auto source = reinterpret_cast<DISPLAYCONFIG_SOURCE_DEVICE_NAME*>(requestPacket);
                 // value must match WrappedIDirect3D9::GetAdapterIdentifier
-                wcscpy(source->viewGdiDeviceName, L"\\\\.\\DISPLAY_SPICE_FAKE");
+                wcscpy(source->viewGdiDeviceName, L"\\\\.\\DISPLAY_SPICE_FAKE_1");
             } else {
                 log_fatal(
                     "popn",

src/spice2x/games/rb/rb.cpp

@@ -4,29 +4,34 @@
 #include "hooks/devicehook.h"
 #include "hooks/audio/backends/dsound/dsound_backend.h"
 #include "util/detour.h"
+#include "util/precise_timer.h"
+#include "util/logging.h"
+#include "cfg/configurator.h"
 
 #include "touch.h"
 
+namespace games::rb {
+    uint16_t TOUCH_POLL_RATE = 0;
+}
+
 static decltype(SleepEx) *SleepEx_orig;
 
-static DWORD WINAPI SleepEx_hook(DWORD dwMilliseconds, BOOL bAltertable) {
+static DWORD WINAPI SleepEx_hook(DWORD dwMilliseconds, BOOL bAlertable) {
 
-    /*
-     * Increase touch poll from ~110 FPS to ~500 FPS
-     */
+    // increase touch poll rate
     if (dwMilliseconds == 8) {
-        static bool initialized = false;
-        if (!initialized) {
-            initialized = true;
+        dwMilliseconds = 1000 / games::rb::TOUCH_POLL_RATE;
+    }
 
-            // if we only sleep for 1ms we also don't need the high priority RB sets
-            SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_BELOW_NORMAL);
-        }
-        dwMilliseconds = 1;
+    // most calls from rb are actually non-alertable
+    if (!bAlertable) {
+        static thread_local timeutils::PreciseSleepTimer timer;
+        timer.sleep(dwMilliseconds);
+        return 0;
     }
 
     // call original
-    return SleepEx_orig(dwMilliseconds, bAltertable);
+    return SleepEx_orig(dwMilliseconds, bAlertable);
 }
 
 games::rb::RBGame::RBGame() : Game("Reflec Beat") {
@@ -35,6 +40,10 @@ games::rb::RBGame::RBGame() : Game("Reflec Beat") {
 void games::rb::RBGame::attach() {
     Game::attach();
 
+    if (1000 < games::rb::TOUCH_POLL_RATE) {
+        log_fatal("rb", "-rbtouchsize is set too high; cannot exceed 1000");
+    }
+
     // init stuff
     devicehook_init();
 
@@ -47,7 +56,12 @@ void games::rb::RBGame::attach() {
     }
 
     // hooks
-    SleepEx_orig = detour::iat_try("SleepEx", SleepEx_hook, avs::game::DLL_INSTANCE);
+    if (0 < TOUCH_POLL_RATE) {
+        log_info("rb", "force increasing touch poll rate by hooking SleepEx ({}Hz)", TOUCH_POLL_RATE);
+        SleepEx_orig = detour::iat_try("SleepEx", SleepEx_hook, avs::game::DLL_INSTANCE);
+    } else {
+        log_info("rb", "not changing touch poll rate (120Hz by default)");
+    }
 }
 
 void games::rb::RBGame::detach() {

src/spice2x/games/rb/rb.h

@@ -7,6 +7,8 @@
 namespace games::rb {
 
     extern uint16_t TOUCH_SCALING;
+    extern uint8_t TOUCH_SIZE;
+    extern uint16_t TOUCH_POLL_RATE;
 
     class RBGame : public games::Game {
     public:

src/spice2x/games/rb/touch.cpp

@@ -13,6 +13,7 @@ static std::string WINDOW_TITLE = "REFLEC BEAT";
 
 namespace games::rb {
     uint16_t TOUCH_SCALING = 1000;
+    uint8_t TOUCH_SIZE = 1;
 }
 
 games::rb::ReflecBeatTouchDeviceHandle::ReflecBeatTouchDeviceHandle(bool log_fps) {
@@ -160,16 +161,20 @@ int games::rb::ReflecBeatTouchDeviceHandle::read(LPVOID lpBuffer, DWORD nNumberO
         const auto point_x = (int) ((x - window_width / 2.0) * 48.0 / 54.0 + window_width / 2.0);
         const auto point_y = (int) (y - window_height / 76);
 
-        // insert 9 times with offset to double the precision
+        // center
         grid_insert(data, point_x, point_y);
-        grid_insert(data, point_x - offset_x, point_y);
-        grid_insert(data, point_x - offset_x, point_y - offset_y);
-        grid_insert(data, point_x - offset_x, point_y + offset_y);
-        grid_insert(data, point_x + offset_x, point_y);
-        grid_insert(data, point_x + offset_x, point_y + offset_y);
-        grid_insert(data, point_x + offset_x, point_y - offset_y);
-        grid_insert(data, point_x, point_y - offset_y);
-        grid_insert(data, point_x, point_y + offset_y);
+
+        // surrounding points
+        if (games::rb::TOUCH_SIZE == 3) {
+            grid_insert(data, point_x - offset_x, point_y);            // west
+            grid_insert(data, point_x - offset_x, point_y - offset_y); // northwest
+            grid_insert(data, point_x - offset_x, point_y + offset_y); // southwest
+            grid_insert(data, point_x + offset_x, point_y);            // east
+            grid_insert(data, point_x + offset_x, point_y + offset_y); // southeast
+            grid_insert(data, point_x + offset_x, point_y - offset_y); // northeast
+            grid_insert(data, point_x, point_y - offset_y);            // north
+            grid_insert(data, point_x, point_y + offset_y);            // south
+        }
     }
 
     // copy data to buffer

src/spice2x/games/shared/printer.cpp

@@ -7,6 +7,7 @@
 #include "hooks/sleephook.h"
 #include "hooks/libraryhook.h"
 #include "launcher/launcher.h"
+#include "overlay/notifications.h"
 #include "util/detour.h"
 #include "util/fileutils.h"
 #include "util/libutils.h"
@@ -397,8 +398,14 @@ namespace games::shared {
                 // logging
                 if (success) {
                     log_info("printer", "printer emulation has written an image to {}", image_path);
+                    overlay::notifications::add(
+                            overlay::notifications::Severity::Success,
+                            fmt::format("Printer: saved {}", fileutils::basename(image_path)));
                 } else {
                     log_warning("printer", "printer emulation failed to write image to {}", image_path);
+                    overlay::notifications::add(
+                            overlay::notifications::Severity::Error,
+                            fmt::format("Printer: failed to write {}", fileutils::basename(image_path)));
                 }
             }
         }

src/spice2x/hooks/audio/asio_driver_scan.cpp

@@ -0,0 +1,74 @@
+#include "asio_driver_scan.h"
+
+#include <algorithm>
+
+#include <windows.h>
+
+#include "util/utils.h"
+
+namespace hooks::audio {
+
+    static constexpr char ASIO_REG_PATH[] = "software\\asio";
+    static constexpr char ASIO_REG_DESC[] = "description";
+
+    // enumerate a single registry view, appending to entries while merging
+    // duplicates discovered in another view. Drivers are matched by name (not
+    // CLSID): the game's ASIO loader selects drivers by name, and some vendors
+    // register the same CLSID under different 32-bit/64-bit names (e.g. "XONAR
+    // SOUND CARD" vs "XONAR SOUND CARD(64)"), which are distinct user choices.
+    static void scan_view(
+            REGSAM wow64_flag,
+            bool is_64bit,
+            std::vector<AsioDriverScanEntry> &entries) {
+
+        HKEY hkEnum = nullptr;
+        if (RegOpenKeyExA(HKEY_LOCAL_MACHINE, ASIO_REG_PATH, 0,
+                KEY_READ | wow64_flag, &hkEnum) != ERROR_SUCCESS) {
+            return;
+        }
+
+        char key_name[256];
+        for (DWORD index = 0;
+             RegEnumKeyA(hkEnum, index, key_name, sizeof(key_name)) == ERROR_SUCCESS;
+             index++) {
+
+            // read description (display name), fall back to the key name
+            char desc[256] = { 0 };
+            DWORD size = sizeof(desc);
+            std::string name = key_name;
+            if (RegGetValueA(hkEnum,
+                             key_name,
+                             ASIO_REG_DESC,
+                             RRF_RT_REG_SZ | wow64_flag,
+                             nullptr,
+                             desc,
+                             &size) == ERROR_SUCCESS && desc[0]) {
+                name = desc;
+            }
+
+            // merge with an existing entry from the other view (match by name)
+            const std::string name_lower = strtolower(name);
+            auto it = std::find_if(entries.begin(), entries.end(), [&](const auto &e) {
+                return strtolower(e.name) == name_lower;
+            });
+            if (it == entries.end()) {
+                entries.push_back({ name });
+                it = entries.end() - 1;
+            }
+            it->found_32bit |= !is_64bit;
+            it->found_64bit |= is_64bit;
+        }
+
+        RegCloseKey(hkEnum);
+    }
+
+    std::vector<AsioDriverScanEntry> scan_asio_drivers() {
+        std::vector<AsioDriverScanEntry> entries;
+
+        // 64-bit view first so it wins ordering when present in both
+        scan_view(KEY_WOW64_64KEY, true, entries);
+        scan_view(KEY_WOW64_32KEY, false, entries);
+
+        return entries;
+    }
+}

src/spice2x/hooks/audio/asio_driver_scan.h

@@ -0,0 +1,15 @@
+#pragma once
+
+#include <string>
+#include <vector>
+
+namespace hooks::audio {
+
+    struct AsioDriverScanEntry {
+        std::string name;
+        bool found_32bit = false;
+        bool found_64bit = false;
+    };
+
+    std::vector<AsioDriverScanEntry> scan_asio_drivers();
+}

src/spice2x/hooks/audio/asio_proxy.h

@@ -0,0 +1,216 @@
+#pragma once
+
+#include <atomic>
+#include <memory>
+#include <string>
+#include <vector>
+
+#include <windows.h>
+
+#include "external/asio/asio.h"
+#include "external/asio/iasiodrv.h"
+
+namespace hooks::audio::asio {
+
+    // returns true if a CoCreateInstance call is instantiating a registered ASIO driver.
+    // ASIO hosts pass the driver CLSID as both class id and interface id; we also validate
+    // it against the system's registered ASIO drivers to avoid false positives
+    bool is_asio_creation(REFCLSID rclsid, REFIID riid);
+
+    // wrap a real ASIO driver instance, taking ownership of the supplied reference, and
+    // return a proxy that forwards every call to it
+    IUnknown *wrap(REFCLSID clsid, void *real);
+}
+
+// transparent proxy around a real ASIO driver; a single place to intercept ASIO traffic
+struct WrappedAsio final : IAsio {
+    WrappedAsio(IAsio *real, REFCLSID clsid, std::string name)
+        : pReal(real), clsid(clsid), driver_name(std::move(name)) {
+    }
+
+    WrappedAsio(const WrappedAsio &) = delete;
+    WrappedAsio &operator=(const WrappedAsio &) = delete;
+
+    virtual ~WrappedAsio();
+
+    // selects which source channel pair of a multichannel ASIO output reaches the device's
+    // 2.0 front pair. when not None, the proxy presents the game's expected multichannel
+    // layout to the host so it proceeds to create_buffers, then opens only a two-channel
+    // stream on the real device and routes the selected pair onto it (see create_buffers).
+    // Front is the plain "force two channel" case (forward the device's own front pair);
+    // the others copy a different pair onto 0/1. assumes a standard 7.1 layout (0-indexed).
+    // set once at boot, before any wrapper exists, so it needs no synchronization
+    enum class StereoDownmix {
+        None,   // feature disabled - full multichannel passthrough
+        Front,  // channels 0/1 - the device front pair is forwarded as-is (no copy)
+        Center, // channel 2 duplicated to both 0 and 1
+        Rear,   // channels 4/5 -> 0/1
+        Side,   // channels 6/7 -> 0/1
+    };
+    static StereoDownmix STEREO_DOWNMIX;
+
+    // true when a stereo extraction is configured, i.e. the real device should open a 2.0
+    // stream and only the selected pair should reach it. the former standalone
+    // FORCE_TWO_CHANNELS flag is now just the Front case of this
+    static bool force_two_channels() {
+        return STEREO_DOWNMIX != StereoDownmix::None;
+    }
+
+    // some games hardcode a multichannel ASIO output and bail before create_buffers if
+    // get_channels reports fewer, so we report at least this many output channels when a
+    // stereo extraction is active
+    static constexpr long FORCED_OUTPUT_CHANNELS = 8;
+
+    // maps an option string ("front", "center", "rear", "side") to a StereoDownmix value,
+    // returning None for anything unrecognized
+    static StereoDownmix name_to_stereo_downmix(const char *name);
+
+#pragma region IUnknown
+    HRESULT STDMETHODCALLTYPE QueryInterface(REFIID riid, void **ppv) override;
+    ULONG STDMETHODCALLTYPE AddRef() override;
+    ULONG STDMETHODCALLTYPE Release() override;
+#pragma endregion
+
+#pragma region IAsio
+    AsioBool __thiscall init(void *sys_handle) override;
+    void __thiscall get_driver_name(char *name) override;
+    long __thiscall get_driver_version() override;
+    void __thiscall get_error_message(char *string) override;
+    AsioError __thiscall start() override;
+    AsioError __thiscall stop() override;
+    AsioError __thiscall get_channels(long *num_input_channels, long *num_output_channels) override;
+    AsioError __thiscall get_latencies(long *input_latency, long *output_latency) override;
+    AsioError __thiscall get_buffer_size(
+        long *min_size,
+        long *max_size,
+        long *preferred_size,
+        long *granularity) override;
+    AsioError __thiscall can_sample_rate(AsioSampleRate sample_rate) override;
+    AsioError __thiscall get_sample_rate(AsioSampleRate *sample_rate) override;
+    AsioError __thiscall set_sample_rate(AsioSampleRate sample_rate) override;
+    AsioError __thiscall get_clock_sources(ASIOClockSource *clocks, long *num_sources) override;
+    AsioError __thiscall set_clock_source(long reference) override;
+    AsioError __thiscall get_sample_position(ASIOSamples *s_pos, ASIOTimeStamp *t_stamp) override;
+    AsioError __thiscall get_channel_info(AsioChannelInfo *info) override;
+    AsioError __thiscall create_buffers(
+        AsioBufferInfo *buffer_infos,
+        long num_channels,
+        long buffer_size,
+        AsioCallbacks *callbacks) override;
+    AsioError __thiscall dispose_buffers() override;
+    AsioError __thiscall control_panel() override;
+    AsioError __thiscall future(long selector, void *opt) override;
+    AsioError __thiscall output_ready() override;
+#pragma endregion
+
+private:
+    // create_buffers implementation used when a stereo extraction is active: forwards only
+    // the channels the real device has and hands the game throwaway buffers for the rest
+    AsioError create_buffers_front_pair(
+        AsioBufferInfo *buffer_infos,
+        long num_channels,
+        long buffer_size,
+        AsioCallbacks *callbacks);
+
+    // if any post-processing effect (volume boost or stereo downmix) is active, saves the
+    // game's callbacks and returns a proxy callback set (our buffer-switch trampolines) to
+    // hand the real driver instead, so we can rework its output buffers after the game
+    // fills them. otherwise returns the game's callbacks unchanged. called at create_buffers
+    // time, before the stream starts
+    AsioCallbacks *install_proxy_callbacks(AsioCallbacks *game_callbacks);
+
+    // records a device output channel whose buffers we scale by the volume boost. queries
+    // the real driver for the channel's sample format. called at create_buffers time
+    void record_volume_output_channel(const AsioBufferInfo &info);
+
+    // the real device's output sample format, queried from its first output channel. all
+    // output channels of a device share one format, so this characterizes them all. returns
+    // ASIOSTLastEntry if the device has no output channels or the query fails
+    AsioSampleType device_output_sample_type();
+
+    // locates the destination pair (device channels 0/1) and the configured source channels
+    // in the game's buffer set so the realtime path can copy the selected pair onto 0/1.
+    // a no-op unless STEREO_DOWNMIX selects a non-front pair. called at create_buffers time
+    void record_downmix_channels(AsioBufferInfo *buffer_infos, long num_channels, long buffer_size);
+
+    // publishes the captured post-process state to the realtime thread once the buffers
+    // exist, making our trampolines start reworking output. called at the end of either
+    // create_buffers path
+    void publish_post_process(long buffer_size);
+
+    // detaches this instance from the realtime trampolines so they stop touching its
+    // buffers. called from dispose_buffers and the destructor
+    void detach_post_process();
+
+    // multiplies every recorded output channel's buffer for the given double-buffer index
+    // by the volume boost. runs on the driver's realtime thread from our buffer switch
+    void apply_output_volume(long double_buffer_index);
+
+    // copies the configured source channel pair onto device channels 0/1 for the given
+    // double-buffer index. runs on the driver's realtime thread from our buffer switch
+    void apply_downmix(long double_buffer_index);
+
+    // realtime-thread trampolines for the buffer-switch callbacks, handed to the real
+    // driver in place of the game's; ASIO callbacks carry no user data, so they reach the
+    // active wrapper through active_instance, call the game's original, then rework output.
+    // the other two callbacks (sample_rate_did_change, asio_message) are forwarded as the
+    // game's own pointers, so they need no trampoline
+    static void __cdecl proxy_buffer_switch(long double_buffer_index, AsioBool direct_process);
+    static AsioTime * __cdecl proxy_buffer_switch_time_info(
+        AsioTime *params, long double_buffer_index, AsioBool direct_process);
+
+    // the single wrapper whose proxy callbacks are installed (ASIO is single-instance with
+    // one running stream); read by the static trampolines to reach the right wrapper
+    static std::atomic<WrappedAsio *> active_instance;
+
+    IAsio *const pReal;
+    const CLSID clsid;
+
+    // registry name of the driver (not get_driver_name), used in our logs as a single
+    // unambiguous name; constant for our lifetime
+    std::string driver_name;
+
+    // our own reference count; we hold one reference on pReal and release it when this
+    // drops to zero
+    std::atomic<ULONG> ref_count {1};
+
+    // throwaway double buffers handed to the channels we discard when a stereo extraction
+    // is active (see create_buffers). owned for the lifetime of the buffer set and freed
+    // in dispose_buffers; only read by the game from its own bufferSwitch, never by us
+    std::vector<std::unique_ptr<uint8_t[]>> dummy_buffers;
+
+    // one device output channel scaled by the volume boost in our buffer switch
+    struct VolumeOutputChannel {
+        void *buffers[2];
+        AsioSampleType type;
+    };
+
+    // the game's original callbacks (captured when we install our proxy set) and the proxy
+    // set we hand the real driver; the realtime trampolines reach the game's buffer_switch
+    // through game_callbacks regardless of which effect is active
+    AsioCallbacks game_callbacks {};
+    AsioCallbacks proxy_callbacks {};
+
+    // volume boost state, captured at create_buffers time and published to the realtime
+    // thread via active_instance once fully built; untouched while the stream runs.
+    // volume_active gates whether the realtime path scales any buffers
+    bool volume_active = false;
+    float volume_gain = 1.0f;
+    long volume_buffer_size = 0;
+    std::vector<VolumeOutputChannel> volume_channels;
+
+    // one device channel (0 or 1) fed by a source channel during stereo downmix; both
+    // buffer pointers are indexed by the ASIO double-buffer index, the same as the channels
+    struct DownmixCopy {
+        void *dst[2];
+        void *src[2];
+    };
+
+    // stereo downmix state, captured at create_buffers time and published alongside the
+    // volume state; untouched while the stream runs. downmix_active gates whether the
+    // realtime path copies the selected source pair onto device channels 0/1. copies[0]
+    // feeds device channel 0, copies[1] feeds device channel 1
+    bool downmix_active = false;
+    DownmixCopy downmix_copies[2] {};
+    size_t downmix_bytes = 0;
+};

src/spice2x/hooks/audio/audio.cpp

@@ -14,6 +14,7 @@
 
 #include "audio_private.h"
 #include "acm.h"
+#include "asio_proxy.h"
 
 #ifdef _MSC_VER
 DEFINE_GUID(CLSID_MMDeviceEnumerator,
@@ -39,9 +40,15 @@ namespace hooks::audio {
     // public globals
     bool ENABLED = true;
     bool VOLUME_HOOK_ENABLED = true;
+    std::optional<DownmixAlgorithm> DOWNMIX_ALGORITHM = std::nullopt;
+    float VOLUME_BOOST = 1.0f;
+    std::optional<uint32_t> RESAMPLE_RATE = std::nullopt;
+    std::optional<uint32_t> EXCLUSIVE_BUFFER_MS = std::nullopt;
+    bool WASAPI_COMPATIBILITY_MODE = false;
     bool USE_DUMMY = false;
     WAVEFORMATEXTENSIBLE FORMAT {};
     std::optional<Backend> BACKEND = std::nullopt;
+    bool INJECT_FAKE_REALTEK_AUDIO = false;
     std::optional<size_t> ASIO_DRIVER_ID = std::nullopt;
     std::string ASIO_DRIVER_NAME = "";
     bool ASIO_FORCE_UNLOAD_ON_STOP = false;
@@ -91,6 +98,10 @@ static HRESULT STDAPICALLTYPE CoCreateInstance_hook(
         // wrap object
         auto mmde = reinterpret_cast<IMMDeviceEnumerator **>(ppv);
         *mmde = new WrappedIMMDeviceEnumerator(*mmde);
+    } else if (ppv != nullptr && *ppv != nullptr && hooks::audio::asio::is_asio_creation(rclsid, riid)) {
+
+        // wrap every ASIO driver so calls pass through to the real driver
+        *ppv = hooks::audio::asio::wrap(rclsid, *ppv);
     }
 
     // return original result

src/spice2x/hooks/audio/audio.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <cstdint>
 #include <optional>
 #include <string>
 
@@ -16,11 +17,40 @@ namespace hooks::audio {
         WaveOut,
     };
 
+    // surround-to-stereo downmix algorithm
+    enum class DownmixAlgorithm {
+        FrontOnly, // keep only the front channels
+        RearOnly,  // keep only the rear/back channels
+        SideOnly,  // keep only the side channels
+        AC4,       // AC-4 stereo downmix coefficients (ETSI TS 103 190-1)
+        Normalize, // all channels equally loud, LFE dropped
+    };
+
     extern bool ENABLED;
     extern bool VOLUME_HOOK_ENABLED;
+    extern std::optional<DownmixAlgorithm> DOWNMIX_ALGORITHM;
+    extern float VOLUME_BOOST;
+
+    // target sample rate the hooked output is resampled to, if set
+    extern std::optional<uint32_t> RESAMPLE_RATE;
+
+    // minimum WASAPI exclusive buffer duration (milliseconds), if set. enlarges the device buffer
+    // to avoid underrun crackle on endpoints that cannot service a tiny buffer in time.
+    extern std::optional<uint32_t> EXCLUSIVE_BUFFER_MS;
+
+    // when true, WASAPI compatibility mode is active: exclusive-mode streams are redirected to
+    // shared mode. the Windows audio engine performs any required sample-rate / channel / bit-depth
+    // conversion, so the game's format is passed through unchanged and other applications can play
+    // audio simultaneously.
+    extern bool WASAPI_COMPATIBILITY_MODE;
     extern bool USE_DUMMY;
     extern WAVEFORMATEXTENSIBLE FORMAT;
     extern std::optional<Backend> BACKEND;
+
+    // when true, a synthetic "Realtek" render endpoint is injected into device enumeration that
+    // discards all audio. used by gitadora arena, whose device search crashes when no render
+    // endpoint reports a "Realtek" friendly name.
+    extern bool INJECT_FAKE_REALTEK_AUDIO;
     extern std::optional<size_t> ASIO_DRIVER_ID;
     extern std::string ASIO_DRIVER_NAME;
     extern bool ASIO_FORCE_UNLOAD_ON_STOP;

src/spice2x/hooks/audio/backends/mmdevice/device_collection.cpp

@@ -1,5 +1,6 @@
 #include "device_collection.h"
 #include "device.h"
+#include "null_device.h"
 #include "util/utils.h"
 #include "util/logging.h"
 
@@ -28,17 +29,48 @@ ULONG STDMETHODCALLTYPE WrappedIMMDeviceCollection::Release() {
     return refs;
 }
 
+bool WrappedIMMDeviceCollection::should_inject_fake_realtek() const {
+    return null_render_device_enabled()
+        && (data_flow == eRender || data_flow == eAll);
+}
+
 HRESULT STDMETHODCALLTYPE WrappedIMMDeviceCollection::GetCount(UINT *pcDevices) {
+    // when active, hide all real devices and present only the synthetic one
+    if (should_inject_fake_realtek()) {
+        if (pcDevices == nullptr) {
+            return E_POINTER;
+        }
+        *pcDevices = 1;
+        return S_OK;
+    }
+
     return pReal->GetCount(pcDevices);
 }
 
 HRESULT STDMETHODCALLTYPE WrappedIMMDeviceCollection::Item(UINT nDevice, IMMDevice **ppDevice) {
+    if (ppDevice == nullptr) {
+        return E_POINTER;
+    }
+
+    // when active, the only device in the collection is the synthetic fake Realtek
+    // render device; all real devices are hidden
+    if (should_inject_fake_realtek()) {
+        if (nDevice != 0) {
+            return E_INVALIDARG;
+        }
+        log_info("audio", "WrappedIMMDeviceCollection::Item[{}] -> synthetic fake Realtek render device", nDevice);
+        *ppDevice = new NullMMDevice();
+        return S_OK;
+    }
+
     log_info("audio", "WrappedIMMDeviceCollection::Item[{}]", nDevice);
 
     // call original
     const auto hr = pReal->Item(nDevice, ppDevice);
 
     // wrap interface
-    *ppDevice = new WrappedIMMDevice(*ppDevice);
+    if (SUCCEEDED(hr) && *ppDevice != nullptr) {
+        *ppDevice = new WrappedIMMDevice(*ppDevice);
+    }
     return hr;
 }

src/spice2x/hooks/audio/backends/mmdevice/device_collection.h

@@ -4,7 +4,8 @@
 #include <mmdeviceapi.h>
 
 struct WrappedIMMDeviceCollection : IMMDeviceCollection {
-    explicit WrappedIMMDeviceCollection(IMMDeviceCollection *orig) : pReal(orig) {
+    WrappedIMMDeviceCollection(IMMDeviceCollection *orig, EDataFlow dataFlow)
+        : pReal(orig), data_flow(dataFlow) {
     }
 
     WrappedIMMDeviceCollection(const WrappedIMMDeviceCollection &) = delete;
@@ -24,5 +25,9 @@ struct WrappedIMMDeviceCollection : IMMDeviceCollection {
 #pragma endregion
 
 private:
+    // whether the synthetic fake Realtek render device should be appended to this collection
+    bool should_inject_fake_realtek() const;
+
     IMMDeviceCollection *const pReal;
+    const EDataFlow data_flow;
 };

src/spice2x/hooks/audio/backends/mmdevice/device_enumerator.cpp

@@ -45,7 +45,7 @@ HRESULT STDMETHODCALLTYPE WrappedIMMDeviceEnumerator::EnumAudioEndpoints(
 {
     const auto hr = pReal->EnumAudioEndpoints(dataFlow, dwStateMask, ppDevices);
     if (SUCCEEDED(hr) && (ppDevices != nullptr) && (*ppDevices != nullptr)) {
-        *ppDevices = new WrappedIMMDeviceCollection(*ppDevices);
+        *ppDevices = new WrappedIMMDeviceCollection(*ppDevices, dataFlow);
     }
     return hr;
 }

src/spice2x/hooks/audio/backends/mmdevice/null_device.cpp

@@ -0,0 +1,195 @@
+#include "null_device.h"
+
+#include <atomic>
+#include <cstring>
+
+#include <audioclient.h>
+
+#include "hooks/audio/audio.h"
+#include "hooks/audio/audio_private.h"
+#include "hooks/audio/backends/wasapi/dummy_audio_client.h"
+#include "util/logging.h"
+#include "util/utils.h"
+
+#include "null_discard_backend.h"
+
+// friendly name reported by the synthetic device. must contain "Realtek" so the
+// gitadora arena device search matches it.
+static const wchar_t NULL_DEVICE_FRIENDLY_NAME[] = L"Realtek High Definition Audio";
+
+// arbitrary identifier reported by the synthetic device.
+static const wchar_t NULL_DEVICE_ID[] = L"{spice2x-null-render-device}";
+
+// PKEY_Device_FriendlyName, hardcoded to avoid pulling in functiondiscoverykeys_devpkey.h
+static const PROPERTYKEY PKEY_DEVICE_FRIENDLY_NAME_LOCAL = {
+    { 0xa45c254e, 0xdf1c, 0x4efd, { 0x80, 0x20, 0x67, 0xd1, 0x46, 0xa8, 0x50, 0xe0 } },
+    14
+};
+
+bool null_render_device_enabled() {
+    return hooks::audio::INJECT_FAKE_REALTEK_AUDIO;
+}
+
+// duplicate a wide string into CoTaskMem so the caller can free it with
+// CoTaskMemFree / PropVariantClear as the COM API contract requires.
+static LPWSTR co_task_wcsdup(const wchar_t *src) {
+    const size_t bytes = (wcslen(src) + 1) * sizeof(wchar_t);
+    auto *dst = static_cast<LPWSTR>(CoTaskMemAlloc(bytes));
+    if (dst != nullptr) {
+        memcpy(dst, src, bytes);
+    }
+    return dst;
+}
+
+namespace {
+
+    // minimal IPropertyStore that only answers PKEY_Device_FriendlyName.
+    struct NullPropertyStore : IPropertyStore {
+        std::atomic<ULONG> ref_cnt = 1;
+
+        virtual ~NullPropertyStore() = default;
+
+        HRESULT STDMETHODCALLTYPE QueryInterface(REFIID riid, void **ppvObj) override {
+            if (ppvObj == nullptr) {
+                return E_POINTER;
+            }
+            if (riid == __uuidof(IUnknown) || riid == __uuidof(IPropertyStore)) {
+                this->AddRef();
+                *ppvObj = this;
+                return S_OK;
+            }
+            *ppvObj = nullptr;
+            return E_NOINTERFACE;
+        }
+        ULONG STDMETHODCALLTYPE AddRef() override {
+            return ++this->ref_cnt;
+        }
+        ULONG STDMETHODCALLTYPE Release() override {
+            const ULONG refs = --this->ref_cnt;
+            if (refs == 0) {
+                delete this;
+            }
+            return refs;
+        }
+        HRESULT STDMETHODCALLTYPE GetCount(DWORD *cProps) override {
+            if (cProps == nullptr) {
+                return E_POINTER;
+            }
+            *cProps = 1;
+            return S_OK;
+        }
+        HRESULT STDMETHODCALLTYPE GetAt(DWORD iProp, PROPERTYKEY *pkey) override {
+            if (pkey == nullptr) {
+                return E_POINTER;
+            }
+            if (iProp != 0) {
+                return E_INVALIDARG;
+            }
+            *pkey = PKEY_DEVICE_FRIENDLY_NAME_LOCAL;
+            return S_OK;
+        }
+        HRESULT STDMETHODCALLTYPE GetValue(REFPROPERTYKEY key, PROPVARIANT *pv) override {
+            if (pv == nullptr) {
+                return E_POINTER;
+            }
+            PropVariantInit(pv);
+            if (key.fmtid == PKEY_DEVICE_FRIENDLY_NAME_LOCAL.fmtid
+                    && key.pid == PKEY_DEVICE_FRIENDLY_NAME_LOCAL.pid) {
+                pv->pwszVal = co_task_wcsdup(NULL_DEVICE_FRIENDLY_NAME);
+                if (pv->pwszVal == nullptr) {
+                    return E_OUTOFMEMORY;
+                }
+                pv->vt = VT_LPWSTR;
+            }
+            // unknown keys are returned as VT_EMPTY / S_OK
+            return S_OK;
+        }
+        HRESULT STDMETHODCALLTYPE SetValue(REFPROPERTYKEY, REFPROPVARIANT) override {
+            return STG_E_ACCESSDENIED;
+        }
+        HRESULT STDMETHODCALLTYPE Commit() override {
+            return S_OK;
+        }
+    };
+}
+
+#pragma region IUnknown
+HRESULT STDMETHODCALLTYPE NullMMDevice::QueryInterface(REFIID riid, void **ppvObj) {
+    if (ppvObj == nullptr) {
+        return E_POINTER;
+    }
+    if (riid == __uuidof(IUnknown) || riid == __uuidof(IMMDevice)) {
+        this->AddRef();
+        *ppvObj = this;
+        return S_OK;
+    }
+    *ppvObj = nullptr;
+    return E_NOINTERFACE;
+}
+ULONG STDMETHODCALLTYPE NullMMDevice::AddRef() {
+    return ++this->ref_cnt;
+}
+ULONG STDMETHODCALLTYPE NullMMDevice::Release() {
+    const ULONG refs = --this->ref_cnt;
+    if (refs == 0) {
+        delete this;
+    }
+    return refs;
+}
+#pragma endregion
+
+#pragma region IMMDevice
+HRESULT STDMETHODCALLTYPE NullMMDevice::Activate(
+    REFIID iid,
+    DWORD,
+    PROPVARIANT *,
+    void **ppInterface)
+{
+    if (ppInterface == nullptr) {
+        return E_POINTER;
+    }
+    *ppInterface = nullptr;
+
+    log_info("audio::null", "NullMMDevice::Activate {}", guid2s(iid));
+
+    if (iid == IID_IAudioClient) {
+
+        // release any previously persisted client
+        if (hooks::audio::CLIENT != nullptr) {
+            hooks::audio::CLIENT->Release();
+        }
+
+        auto *client = static_cast<IAudioClient *>(new DummyIAudioClient(new NullDiscardBackend()));
+        *ppInterface = client;
+
+        // persist the audio client
+        hooks::audio::CLIENT = client;
+        hooks::audio::CLIENT->AddRef();
+
+        return S_OK;
+    }
+
+    return E_NOINTERFACE;
+}
+HRESULT STDMETHODCALLTYPE NullMMDevice::OpenPropertyStore(DWORD, IPropertyStore **ppProperties) {
+    if (ppProperties == nullptr) {
+        return E_POINTER;
+    }
+    *ppProperties = new NullPropertyStore();
+    return S_OK;
+}
+HRESULT STDMETHODCALLTYPE NullMMDevice::GetId(LPWSTR *ppstrId) {
+    if (ppstrId == nullptr) {
+        return E_POINTER;
+    }
+    *ppstrId = co_task_wcsdup(NULL_DEVICE_ID);
+    return *ppstrId != nullptr ? S_OK : E_OUTOFMEMORY;
+}
+HRESULT STDMETHODCALLTYPE NullMMDevice::GetState(DWORD *pdwState) {
+    if (pdwState == nullptr) {
+        return E_POINTER;
+    }
+    *pdwState = DEVICE_STATE_ACTIVE;
+    return S_OK;
+}
+#pragma endregion

src/spice2x/hooks/audio/backends/mmdevice/null_device.h

@@ -0,0 +1,39 @@
+#pragma once
+
+#include <atomic>
+
+#include <mmdeviceapi.h>
+
+// returns true when a synthetic render endpoint should be injected into device
+// enumeration. games like gitadora arena search the render endpoint list for a
+// device whose friendly name contains "Realtek" and crash with a null pointer
+// dereference when no match exists. presenting a fake match that routes to the
+// null audio backend lets the search succeed while discarding the audio.
+bool null_render_device_enabled();
+
+// fake IMMDevice that reports a "Realtek" friendly name and activates straight
+// into the null audio backend, never touching real hardware.
+struct NullMMDevice : IMMDevice {
+    NullMMDevice() = default;
+
+    NullMMDevice(const NullMMDevice &) = delete;
+    NullMMDevice &operator=(const NullMMDevice &) = delete;
+
+    virtual ~NullMMDevice() = default;
+
+#pragma region IUnknown
+    HRESULT STDMETHODCALLTYPE QueryInterface(REFIID riid, void **ppvObj) override;
+    ULONG STDMETHODCALLTYPE AddRef() override;
+    ULONG STDMETHODCALLTYPE Release() override;
+#pragma endregion
+
+#pragma region IMMDevice
+    HRESULT STDMETHODCALLTYPE Activate(REFIID iid, DWORD dwClsCtx, PROPVARIANT *pActivationParams, void **ppInterface) override;
+    HRESULT STDMETHODCALLTYPE OpenPropertyStore(DWORD stgmAccess, IPropertyStore **ppProperties) override;
+    HRESULT STDMETHODCALLTYPE GetId(LPWSTR *ppstrId) override;
+    HRESULT STDMETHODCALLTYPE GetState(DWORD *pdwState) override;
+#pragma endregion
+
+private:
+    std::atomic<ULONG> ref_cnt = 1;
+};

src/spice2x/hooks/audio/backends/mmdevice/null_discard_backend.cpp

@@ -0,0 +1,141 @@
+#include "null_discard_backend.h"
+
+#include <algorithm>
+#include <chrono>
+
+#include "hooks/audio/util.h"
+#include "util/logging.h"
+#include "util/precise_timer.h"
+
+NullDiscardBackend::~NullDiscardBackend() {
+    this->running = false;
+    if (this->pacing_thread.joinable()) {
+        this->pacing_thread.join();
+    }
+}
+
+const WAVEFORMATEXTENSIBLE &NullDiscardBackend::format() const noexcept {
+    return this->format_;
+}
+
+HRESULT NullDiscardBackend::on_initialize(
+    AUDCLNT_SHAREMODE *,
+    DWORD *,
+    REFERENCE_TIME *hnsBufferDuration,
+    REFERENCE_TIME *,
+    const WAVEFORMATEX *pFormat,
+    LPCGUID)
+{
+    copy_wave_format(&this->format_, pFormat);
+
+    // honor the game's requested buffer duration, falling back to 10 ms
+    constexpr REFERENCE_TIME DEFAULT_REFTIME = 100000; // 10 ms in 100-ns units
+    this->period_reftime = (hnsBufferDuration && *hnsBufferDuration > 0)
+            ? *hnsBufferDuration
+            : DEFAULT_REFTIME;
+
+    this->buffer_frames = std::max<uint32_t>(1, static_cast<uint32_t>(
+            static_cast<double>(this->format_.Format.nSamplesPerSec)
+                    * this->period_reftime / 10000000.0 + 0.5));
+
+    log_info("audio::null", "initializing null render device with {} channels, {} Hz, {}-bit",
+             this->format_.Format.nChannels,
+             this->format_.Format.nSamplesPerSec,
+             this->format_.Format.wBitsPerSample);
+
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_get_buffer_size(uint32_t *buffer_frames) {
+    *buffer_frames = this->buffer_frames;
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_get_stream_latency(REFERENCE_TIME *latency) {
+    *latency = this->period_reftime;
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_get_current_padding(std::optional<uint32_t> &padding_frames) {
+    // discarded immediately, so the buffer always reads as fully drained
+    padding_frames = 0;
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_is_format_supported(
+    AUDCLNT_SHAREMODE *,
+    const WAVEFORMATEX *,
+    WAVEFORMATEX **ppClosestMatch)
+{
+    if (ppClosestMatch) {
+        *ppClosestMatch = nullptr;
+    }
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_get_mix_format(WAVEFORMATEX **) {
+    return E_NOTIMPL;
+}
+
+HRESULT NullDiscardBackend::on_get_device_period(
+    REFERENCE_TIME *default_device_period,
+    REFERENCE_TIME *minimum_device_period)
+{
+    if (default_device_period) {
+        *default_device_period = this->period_reftime;
+    }
+    if (minimum_device_period) {
+        *minimum_device_period = this->period_reftime;
+    }
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_start() {
+    if (!this->running.exchange(true)) {
+        this->pacing_thread = std::thread(&NullDiscardBackend::pace_loop, this);
+    }
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_stop() {
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_set_event_handle(HANDLE *event_handle) {
+    // keep the game's event so pace_loop() can wake it; there is no real device behind it
+    this->relay_handle = *event_handle;
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_get_buffer(uint32_t num_frames_requested, BYTE **ppData) {
+    const size_t buffer_size =
+            static_cast<size_t>(this->format_.Format.nBlockAlign) * num_frames_requested;
+    if (this->scratch.size() < buffer_size) {
+        this->scratch.resize(buffer_size);
+    }
+    *ppData = this->scratch.data();
+    return S_OK;
+}
+
+HRESULT NullDiscardBackend::on_release_buffer(uint32_t, DWORD) {
+    // discard the audio entirely
+    return S_OK;
+}
+
+void NullDiscardBackend::pace_loop() {
+    using namespace std::chrono;
+
+    timeutils::PreciseSleepTimer timer;
+
+    // audio is discarded, so timing precision and drift do not matter; just wake the
+    // game once per buffer period to keep its render thread from blocking on the event.
+    const auto period = duration_cast<steady_clock::duration>(
+            duration<double>(this->period_reftime / 10000000.0));
+
+    while (this->running.load()) {
+        if (this->relay_handle) {
+            SetEvent(this->relay_handle);
+        }
+        timer.sleep(period);
+    }
+}

src/spice2x/hooks/audio/backends/mmdevice/null_discard_backend.h

@@ -0,0 +1,54 @@
+#pragma once
+
+#include <atomic>
+#include <optional>
+#include <thread>
+#include <vector>
+
+#include <audioclient.h>
+
+#include "hooks/audio/implementations/backend.h"
+
+// discards all audio while pacing the game's event handle once per buffer period, so the game
+// keeps running normally with nothing output to any real device. routed through the shared
+// DummyIAudioClient, the same plumbing the asio backend uses.
+struct NullDiscardBackend final : AudioBackend {
+    ~NullDiscardBackend() final;
+
+    const WAVEFORMATEXTENSIBLE &format() const noexcept override;
+
+    HRESULT on_initialize(
+        AUDCLNT_SHAREMODE *,
+        DWORD *,
+        REFERENCE_TIME *hnsBufferDuration,
+        REFERENCE_TIME *,
+        const WAVEFORMATEX *pFormat,
+        LPCGUID) override;
+    HRESULT on_get_buffer_size(uint32_t *buffer_frames) override;
+    HRESULT on_get_stream_latency(REFERENCE_TIME *latency) override;
+    HRESULT on_get_current_padding(std::optional<uint32_t> &padding_frames) override;
+    HRESULT on_is_format_supported(
+        AUDCLNT_SHAREMODE *,
+        const WAVEFORMATEX *,
+        WAVEFORMATEX **ppClosestMatch) override;
+    HRESULT on_get_mix_format(WAVEFORMATEX **) override;
+    HRESULT on_get_device_period(
+        REFERENCE_TIME *default_device_period,
+        REFERENCE_TIME *minimum_device_period) override;
+    HRESULT on_start() override;
+    HRESULT on_stop() override;
+    HRESULT on_set_event_handle(HANDLE *event_handle) override;
+    HRESULT on_get_buffer(uint32_t num_frames_requested, BYTE **ppData) override;
+    HRESULT on_release_buffer(uint32_t, DWORD) override;
+
+private:
+    void pace_loop();
+
+    WAVEFORMATEXTENSIBLE format_ {};
+    uint32_t buffer_frames = 0;
+    REFERENCE_TIME period_reftime = 0;
+    HANDLE relay_handle = nullptr;
+    std::vector<BYTE> scratch;
+    std::thread pacing_thread;
+    std::atomic<bool> running = false;
+};

src/spice2x/hooks/audio/backends/wasapi/audio_client.cpp

@@ -4,6 +4,7 @@
 #include <ksmedia.h>
 
 #include "avs/game.h"
+#include "games/gitadora/gitadora.h"
 #include "hooks/audio/audio.h"
 #include "hooks/audio/util.h"
 #include "hooks/audio/backends/wasapi/util.h"
@@ -41,6 +42,27 @@ static void fix_rec_format(WAVEFORMATEX *pFormat) {
     pFormat->nAvgBytesPerSec = pFormat->nSamplesPerSec * pFormat->nBlockAlign;
 }
 
+// decide whether the given multi-channel format should be downmixed to stereo and which algorithm
+// to use. an explicit user selection (-downmix) takes precedence; otherwise gitadora arena
+// two-channel mode defaults to the AC-4 algorithm.
+static std::optional<hooks::audio::DownmixAlgorithm> resolve_downmix(const WAVEFORMATEX *format) {
+    if (format == nullptr
+            || format->nChannels <= 2
+            || format->wFormatTag != WAVE_FORMAT_EXTENSIBLE) {
+        return std::nullopt;
+    }
+
+    if (hooks::audio::DOWNMIX_ALGORITHM.has_value()) {
+        return hooks::audio::DOWNMIX_ALGORITHM;
+    }
+
+    if (games::gitadora::is_arena_model() && games::gitadora::TWOCHANNEL) {
+        return hooks::audio::DownmixAlgorithm::AC4;
+    }
+
+    return std::nullopt;
+}
+
 IAudioClient *wrap_audio_client(IAudioClient *audio_client) {
     log_misc("audio::wasapi", "wrapping IAudioClient");
 
@@ -145,13 +167,51 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioClient::Initialize(
         fix_rec_format(const_cast<WAVEFORMATEX *>(pFormat));
     }
 
+    // apply the -wasapishared option: redirect an exclusive request to shared mode. once redirected,
+    // spice's own downmix/resample paths below are skipped (gated on redirected_from_exclusive) and
+    // the shared engine handles any format conversion via AUTOCONVERTPCM (PCM / float only).
+    if (hooks::audio::SharedRedirect::wants(ShareMode, pFormat)) {
+        this->shared.apply(&ShareMode, &StreamFlags, &hnsPeriodicity);
+    } else if (hooks::audio::WASAPI_COMPATIBILITY_MODE && ShareMode == AUDCLNT_SHAREMODE_SHARED) {
+        log_warning(
+            "audio::wasapi",
+            "-wasapishared is enabled but the game is already opening a shared-mode stream; "
+            "the option has no effect");
+    }
+
+    WAVEFORMATEXTENSIBLE stereo_storage = {};
+    WAVEFORMATEXTENSIBLE resample_storage = {};
+    const WAVEFORMATEX *device_format = pFormat;
+
+    if (!this->shared.redirected_from_exclusive) {
+
+        // when downmixing, open the real device as stereo while the game keeps writing its native
+        // multi-channel format into the scratch buffer.
+        if (auto algorithm = resolve_downmix(pFormat)) {
+            this->downmix.setup(pFormat, &stereo_storage, *algorithm);
+            device_format = reinterpret_cast<const WAVEFORMATEX *>(&stereo_storage);
+            log_info("audio::wasapi", "downmix enabled: {} channels -> 2 channels ({})",
+                    pFormat->nChannels, hooks::audio::Downmix::algorithm_name(*algorithm));
+        } else if (games::gitadora::is_arena_model()) {
+            games::gitadora::fix_audio_channel_mask(const_cast<WAVEFORMATEX *>(pFormat));
+        }
+
+        // when resampling, open the real device at the target rate while the game keeps writing its
+        // native-rate audio into the scratch buffer. this runs on whatever device_format is now: the
+        // game's native format, or the stereo format produced above when downmix is also active, so
+        // the two stages chain as multi-channel -> stereo -> resampled stereo.
+        if (auto target_rate = hooks::audio::Resampler::resolve(device_format)) {
+            const uint32_t src_rate = device_format->nSamplesPerSec;
+            this->resample.setup(device_format, &resample_storage, *target_rate);
+            device_format = reinterpret_cast<const WAVEFORMATEX *>(&resample_storage);
+            log_info("audio::wasapi", "resample enabled: {} Hz -> {} Hz{}",
+                    src_rate, *target_rate, this->downmix.enabled ? " (after downmix)" : "");
+        }
+    }
+
     // verbose output
     log_info("audio::wasapi", "IAudioClient::Initialize hook hit");
-    log_info("audio::wasapi", "... ShareMode         : {}", share_mode_str(ShareMode));
-    log_info("audio::wasapi", "... StreamFlags       : {}", stream_flags_str(StreamFlags));
-    log_info("audio::wasapi", "... hnsBufferDuration : {}", hnsBufferDuration);
-    log_info("audio::wasapi", "... hnsPeriodicity    : {}", hnsPeriodicity);
-    print_format(pFormat);
+    print_format(ShareMode, StreamFlags, hnsBufferDuration, hnsPeriodicity, device_format);
 
     if (this->backend) {
         SAFE_CALL("AudioBackend", "on_initialize", this->backend->on_initialize(
@@ -163,27 +223,67 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioClient::Initialize(
                 AudioSessionGuid));
 
         log_info("audio::wasapi", "AudioBackend::on_initialize call finished");
-        log_info("audio::wasapi", "... ShareMode         : {}", share_mode_str(ShareMode));
-        log_info("audio::wasapi", "... StreamFlags       : {}", stream_flags_str(StreamFlags));
-        log_info("audio::wasapi", "... hnsBufferDuration : {}", hnsBufferDuration);
-        log_info("audio::wasapi", "... hnsPeriodicity    : {}", hnsPeriodicity);
-        print_format(pFormat);
+        print_format(ShareMode, StreamFlags, hnsBufferDuration, hnsPeriodicity, pFormat);
     }
 
     // check for exclusive mode
     if (ShareMode == AUDCLNT_SHAREMODE_EXCLUSIVE) {
         this->exclusive_mode = true;
-        this->frame_size = pFormat->nChannels * (pFormat->wBitsPerSample / 8);
+        this->frame_size = device_format->nChannels * (device_format->wBitsPerSample / 8);
+
+        // optionally enlarge the exclusive buffer. games request a very small buffer (e.g. 3 ms)
+        // which some endpoints (notably NVIDIA HDMI/DP display audio) cannot service in time,
+        // underrunning mid-period and crackling. a larger buffer gives the device slack. exclusive
+        // mode requires periodicity == buffer_duration, so raise both together; the initialize
+        // paths below handle any required buffer-size realignment.
+        if (hooks::audio::EXCLUSIVE_BUFFER_MS.has_value()) {
+            const REFERENCE_TIME min_duration =
+                    (REFERENCE_TIME) hooks::audio::EXCLUSIVE_BUFFER_MS.value() * 10000;
+            if (hnsBufferDuration < min_duration) {
+                log_info("audio::wasapi",
+                        "raising exclusive buffer from {} hns to {} hns ({} ms)",
+                        hnsBufferDuration, min_duration, hooks::audio::EXCLUSIVE_BUFFER_MS.value());
+                hnsBufferDuration = min_duration;
+                if (hnsPeriodicity != 0) {
+                    hnsPeriodicity = min_duration;
+                }
+            }
+        }
     }
 
-    // call next
-    HRESULT ret = pReal->Initialize(
-            ShareMode,
-            StreamFlags,
-            hnsBufferDuration,
-            hnsPeriodicity,
-            pFormat,
-            AudioSessionGuid);
+    // call next. the resampler owns the device interaction whenever it is active (including when
+    // chained after the downmix), otherwise the downmix does, otherwise the device is opened
+    // directly.
+    HRESULT ret;
+    if (this->resample.enabled) {
+        ret = this->resample.initialize(
+                pReal,
+                ShareMode,
+                StreamFlags,
+                hnsBufferDuration,
+                hnsPeriodicity,
+                device_format,
+                AudioSessionGuid);
+    } else if (this->downmix.enabled) {
+        ret = this->downmix.initialize(
+                pReal,
+                ShareMode,
+                StreamFlags,
+                hnsBufferDuration,
+                hnsPeriodicity,
+                device_format,
+                AudioSessionGuid);
+    } else {
+        ret = initialize_with_alignment_retry(
+                pReal,
+                "audio::wasapi",
+                ShareMode,
+                StreamFlags,
+                hnsBufferDuration,
+                hnsPeriodicity,
+                device_format,
+                AudioSessionGuid);
+    }
 
     // check for failure
     if (FAILED(ret)) {
@@ -192,7 +292,8 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioClient::Initialize(
     }
 
     log_info("audio::wasapi", "IAudioClient::Initialize success, hr={}", FMT_HRESULT(ret));
-    copy_wave_format(&hooks::audio::FORMAT, pFormat);
+    copy_wave_format(&hooks::audio::FORMAT, device_format);
+    copy_wave_format(&this->device_format, device_format);
 
     return ret;
 }
@@ -214,7 +315,21 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioClient::GetBufferSize(UINT32 *pNumBufferF
         }
     }
 
-    CHECK_RESULT(pReal->GetBufferSize(pNumBufferFrames));
+    HRESULT ret = pReal->GetBufferSize(pNumBufferFrames);
+
+    // report the buffer size at the game's native rate; the real device buffer is at the
+    // resampled rate, so translate it back so the game paces its writes correctly.
+    if (SUCCEEDED(ret) && this->resample.enabled && pNumBufferFrames) {
+        *pNumBufferFrames = this->resample.frames_device_to_game(*pNumBufferFrames);
+    }
+
+    // redirected to shared mode: clamp the reported buffer to one device period (see SharedRedirect).
+    if (SUCCEEDED(ret) && this->shared.redirected_from_exclusive && pNumBufferFrames) {
+        *pNumBufferFrames = this->shared.clamp_buffer_size(
+                pReal, this->device_format.Format.nSamplesPerSec, *pNumBufferFrames);
+    }
+
+    CHECK_RESULT(ret);
 }
 HRESULT STDMETHODCALLTYPE WrappedIAudioClient::GetStreamLatency(REFERENCE_TIME *phnsLatency) {
     static std::once_flag printed;
@@ -256,7 +371,15 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioClient::GetCurrentPadding(UINT32 *pNumPad
         }
     }
 
-    CHECK_RESULT(pReal->GetCurrentPadding(pNumPaddingFrames));
+    HRESULT ret = pReal->GetCurrentPadding(pNumPaddingFrames);
+
+    // the device buffer is at the resampled rate; report padding at the game's native rate so the
+    // game's free-space calculation stays paced correctly.
+    if (SUCCEEDED(ret) && this->resample.enabled && pNumPaddingFrames) {
+        *pNumPaddingFrames = this->resample.padding_device_to_game(*pNumPaddingFrames);
+    }
+
+    CHECK_RESULT(ret);
 }
 HRESULT STDMETHODCALLTYPE WrappedIAudioClient::IsFormatSupported(
     AUDCLNT_SHAREMODE ShareMode,
@@ -274,6 +397,54 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioClient::IsFormatSupported(
         fix_rec_format(const_cast<WAVEFORMATEX *>(pFormat));
     }
 
+    // log the format the game is asking about
+    log_info("audio::wasapi", "IAudioClient::IsFormatSupported hook hit");
+    print_format(ShareMode, pFormat);
+
+    // under the exclusive->shared redirect, report the exclusive format as supported so the game
+    // doesn't fall back before reaching Initialize.
+    if (hooks::audio::SharedRedirect::wants(ShareMode, pFormat)) {
+        log_info("audio::wasapi", "... reporting supported (will redirect to shared mode)");
+        if (ppClosestMatch) {
+            *ppClosestMatch = nullptr;
+        }
+        return S_OK;
+    }
+
+    // when downmixing, the real device is opened as stereo, so check whether the equivalent
+    // stereo format is supported instead of the multi-channel one. when resampling is also active
+    // it chains onto that stereo format, so check the resampled stereo format.
+    if (resolve_downmix(pFormat)) {
+        WAVEFORMATEXTENSIBLE stereo_storage = {};
+        hooks::audio::Downmix::make_stereo_format(pFormat, &stereo_storage);
+        const WAVEFORMATEX *check_format = reinterpret_cast<const WAVEFORMATEX *>(&stereo_storage);
+
+        WAVEFORMATEXTENSIBLE resample_storage = {};
+        if (auto target_rate = hooks::audio::Resampler::resolve(check_format)) {
+            hooks::audio::Resampler::make_device_format(check_format, &resample_storage, *target_rate);
+            check_format = reinterpret_cast<const WAVEFORMATEX *>(&resample_storage);
+        }
+
+        log_info("audio::wasapi", "... checking device format instead (after downmix/resample):");
+        print_format(check_format);
+
+        CHECK_RESULT(pReal->IsFormatSupported(ShareMode, check_format, ppClosestMatch));
+    } else if (games::gitadora::is_arena_model()) {
+        games::gitadora::fix_audio_channel_mask(const_cast<WAVEFORMATEX *>(pFormat));
+    } else if (auto target_rate = hooks::audio::Resampler::resolve(pFormat)) {
+
+        // when resampling, the real device is opened at the target rate, so check whether the
+        // equivalent format at that rate is supported instead of the game's native rate.
+        WAVEFORMATEXTENSIBLE resample_storage = {};
+        hooks::audio::Resampler::make_device_format(pFormat, &resample_storage, *target_rate);
+        const auto resample_format = reinterpret_cast<const WAVEFORMATEX *>(&resample_storage);
+
+        log_info("audio::wasapi", "... checking device format instead (after resample):");
+        print_format(resample_format);
+
+        CHECK_RESULT(pReal->IsFormatSupported(ShareMode, resample_format, ppClosestMatch));
+    }
+
     if (this->backend) {
         HRESULT ret = this->backend->on_is_format_supported(&ShareMode, pFormat, ppClosestMatch);
 
@@ -466,5 +637,6 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioClient::InitializeSharedAudioStream(
 
     log_info("audio::wasapi", "IAudioClient3::InitializeSharedAudioStream success, hr={}", FMT_HRESULT(ret));
     copy_wave_format(&hooks::audio::FORMAT, pFormat);
+    copy_wave_format(&this->device_format, pFormat);
     return ret;
 }

src/spice2x/hooks/audio/backends/wasapi/audio_client.h

@@ -8,8 +8,11 @@
 #include "hooks/audio/audio_private.h"
 #include "util/logging.h"
 
-#include "audio_render_client.h"
+#include "downmix.h"
+#include "resample.h"
+#include "shared.h"
 
+#include "audio_render_client.h"
 // {1FBC8530-AF3E-4128-B418-115DE72F76B6}
 static const GUID IID_WrappedIAudioClient = {
     0x1fbc8530, 0xaf3e, 0x4128, { 0xb4, 0x18, 0x11, 0x5d, 0xe7, 0x2f, 0x76, 0xb6 }
@@ -92,5 +95,22 @@ struct WrappedIAudioClient : IAudioClient3 {
     IAudioClient3 *const pReal3;
     AudioBackend *const backend;
     bool exclusive_mode = false;
+
+    // -wasapishared redirect state: when an exclusive request was redirected to shared mode, the
+    // engine converts the native format and the reported buffer size is clamped (see SharedRedirect).
+    hooks::audio::SharedRedirect shared;
     int frame_size = 0;
+
+    // the format the real device was opened with (after any downmix). used to scale the final
+    // output buffer for the volume boost.
+    WAVEFORMATEXTENSIBLE device_format = {};
+
+    // surround -> stereo downmix. the real device is opened as stereo while the game keeps
+    // writing multi-channel audio into a scratch buffer that we downmix in the render client.
+    hooks::audio::Downmix downmix;
+
+    // native-rate -> target-rate sample-rate conversion. the real device is opened at the target
+    // rate while the game keeps writing its native-rate audio into a scratch buffer that we
+    // resample in the render client.
+    hooks::audio::Resampler resample;
 };

src/spice2x/hooks/audio/backends/wasapi/audio_render_client.cpp

@@ -1,10 +1,71 @@
 #include "audio_render_client.h"
 
+#include <algorithm>
+#include <cmath>
+#include <cstdint>
+#include <cstring>
+
 #include "audio_client.h"
+#include "hooks/audio/audio.h"
+#include "util.h"
 #include "wasapi_private.h"
 
 const char CLASS_NAME[] = "WrappedIAudioRenderClient";
 
+// scale every sample of an interleaved device buffer by `gain`, clamped to the format's range.
+// supports 16/24/32-bit PCM and 32-bit float; other formats are left untouched.
+static void apply_gain(BYTE *buffer, UINT32 frames, const WAVEFORMATEXTENSIBLE &fmt, float gain) {
+    const WAVEFORMATEX &f = fmt.Format;
+    const size_t samples = (size_t) frames * f.nChannels;
+
+    bool is_float = is_ieee_float(&f);
+
+    if (is_float && f.wBitsPerSample == 32) {
+        auto p = reinterpret_cast<float *>(buffer);
+        for (size_t i = 0; i < samples; i++) {
+            p[i] = std::clamp(p[i] * gain, -1.0f, 1.0f);
+        }
+        return;
+    }
+
+    switch (f.wBitsPerSample) {
+        case 16: {
+            auto p = reinterpret_cast<int16_t *>(buffer);
+            for (size_t i = 0; i < samples; i++) {
+                p[i] = (int16_t) std::clamp((int) std::lround(p[i] * gain), -32768, 32767);
+            }
+            break;
+        }
+        case 24: {
+            // packed 24-bit little-endian
+            for (size_t i = 0; i < samples; i++) {
+                BYTE *s = buffer + i * 3;
+                int32_t v = s[0] | (s[1] << 8) | (s[2] << 16);
+                if (v & 0x800000) {
+                    v |= ~0xFFFFFF; // sign extend
+                }
+                int64_t scaled = std::clamp<int64_t>(
+                    std::llround((double) v * gain), -8388608, 8388607);
+                s[0] = scaled & 0xFF;
+                s[1] = (scaled >> 8) & 0xFF;
+                s[2] = (scaled >> 16) & 0xFF;
+            }
+            break;
+        }
+        case 32: {
+            auto p = reinterpret_cast<int32_t *>(buffer);
+            for (size_t i = 0; i < samples; i++) {
+                p[i] = (int32_t) std::clamp(
+                    std::llround((double) p[i] * gain),
+                    (long long) INT32_MIN, (long long) INT32_MAX);
+            }
+            break;
+        }
+        default:
+            break;
+    }
+}
+
 HRESULT STDMETHODCALLTYPE WrappedIAudioRenderClient::QueryInterface(REFIID riid, void **ppvObj) {
     if (ppvObj == nullptr) {
         return E_POINTER;
@@ -51,6 +112,25 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioRenderClient::GetBuffer(UINT32 NumFramesR
         return S_OK;
     }
 
+    // downmix + resample chained: the game writes its multi-channel native-rate audio into the
+    // downmix scratch, which is downmixed to stereo and then resampled on release. size the
+    // resampler's (stereo) input scratch now and hand the game the multi-channel downmix scratch.
+    if (this->client->downmix.enabled && this->client->resample.enabled) {
+        BYTE *resample_scratch = nullptr;
+        this->client->resample.get_buffer(NumFramesRequested, &resample_scratch);
+        CHECK_RESULT(this->client->downmix.get_scratch(NumFramesRequested, ppData));
+
+    // surround downmix: reserve the real (stereo) device buffer, but hand the game a
+    // multi-channel scratch buffer that we downmix on release
+    } else if (this->client->downmix.enabled) {
+        CHECK_RESULT(this->client->downmix.get_buffer(pReal, NumFramesRequested, ppData));
+
+    // resample: hand the game a native-rate scratch buffer that we convert on release. the real
+    // device buffer is acquired in ReleaseBuffer once the converted frame count is known.
+    } else if (this->client->resample.enabled) {
+        CHECK_RESULT(this->client->resample.get_buffer(NumFramesRequested, ppData));
+    }
+
     // call original
     HRESULT ret = pReal->GetBuffer(NumFramesRequested, ppData);
 
@@ -75,14 +155,72 @@ HRESULT STDMETHODCALLTYPE WrappedIAudioRenderClient::ReleaseBuffer(UINT32 NumFra
         return S_OK;
     }
 
-    // fix for audio pop effect
-    if (this->buffers_to_mute > 0 && this->client->frame_size > 0) {
-
-        // zero out = mute
-        memset(this->audio_buffer, 0, NumFramesWritten * this->client->frame_size);
-
-        this->buffers_to_mute--;
+    // downmix + resample chained: downmix the game's multi-channel scratch into the resampler's
+    // stereo input scratch, then let the resampler convert and push it to the device. a silent
+    // buffer skips the downmix and feeds silence straight through.
+    if (this->client->downmix.enabled && this->client->resample.enabled) {
+        if ((dwFlags & AUDCLNT_BUFFERFLAGS_SILENT) == 0) {
+            this->client->downmix.downmix_into(
+                    this->client->resample.input_data(), NumFramesWritten);
+        }
+        return this->client->resample.flush(
+                pReal,
+                this->client->pReal,
+                NumFramesWritten,
+                dwFlags,
+                hooks::audio::VOLUME_BOOST);
     }
 
-    CHECK_RESULT(pReal->ReleaseBuffer(NumFramesWritten, dwFlags));
+    // resample: convert the game's native-rate scratch and push as many output frames as the
+    // device has room for, applying the volume boost to the converted output. handles acquiring
+    // and releasing the real device buffer itself.
+    if (this->client->resample.enabled) {
+        return this->client->resample.flush(
+                pReal,
+                this->client->pReal,
+                NumFramesWritten,
+                dwFlags,
+                hooks::audio::VOLUME_BOOST);
+    }
+
+    // resolve the real device buffer for whichever path produced the audio
+    BYTE *device_buffer;
+    if (this->client->downmix.enabled) {
+
+        // downmix the game's multi-channel scratch into the real stereo buffer held since GetBuffer
+        this->client->downmix.write_device_buffer(NumFramesWritten, dwFlags);
+        device_buffer = this->client->downmix.current_buffer();
+    } else {
+        device_buffer = this->audio_buffer;
+
+        // mute the first few buffers to avoid a startup pop
+        if (this->buffers_to_mute > 0 && this->client->frame_size > 0) {
+            memset(this->audio_buffer, 0, NumFramesWritten * this->client->frame_size);
+            this->buffers_to_mute--;
+        }
+    }
+
+    // boost the final output volume just before it reaches the device, layout-agnostic
+    if (hooks::audio::VOLUME_BOOST != 1.0f
+            && device_buffer != nullptr
+            && (dwFlags & AUDCLNT_BUFFERFLAGS_SILENT) == 0) {
+        static std::once_flag boost_printed;
+        std::call_once(boost_printed, []() {
+            log_info("audio::wasapi", "volume boost active: gain={}", hooks::audio::VOLUME_BOOST);
+        });
+        apply_gain(device_buffer, NumFramesWritten, this->client->device_format,
+                hooks::audio::VOLUME_BOOST);
+    }
+
+    HRESULT ret = pReal->ReleaseBuffer(NumFramesWritten, dwFlags);
+
+    if (this->client->downmix.enabled) {
+        this->client->downmix.buffer_released();
+    }
+
+    if (FAILED(ret)) {
+        PRINT_FAILED_RESULT(CLASS_NAME, __func__, ret);
+    }
+
+    return ret;
 }

src/spice2x/hooks/audio/backends/wasapi/downmix.cpp

@@ -0,0 +1,264 @@
+#include "downmix.h"
+
+#include <algorithm>
+#include <cmath>
+#include <cstdint>
+#include <cstring>
+
+#include <audioclient.h>
+#include <ks.h>
+#include <ksmedia.h>
+
+#include "util/logging.h"
+
+#include "util.h"
+
+namespace hooks::audio {
+
+    namespace {
+
+        constexpr float ATT_3DB = 0.70710678f;
+
+        // speakers routed to the left/right output; anything else (center) feeds both sides
+        constexpr DWORD LEFT_SPEAKERS = SPEAKER_FRONT_LEFT | SPEAKER_BACK_LEFT | SPEAKER_SIDE_LEFT
+            | SPEAKER_FRONT_LEFT_OF_CENTER | SPEAKER_TOP_FRONT_LEFT | SPEAKER_TOP_BACK_LEFT;
+        constexpr DWORD RIGHT_SPEAKERS = SPEAKER_FRONT_RIGHT | SPEAKER_BACK_RIGHT | SPEAKER_SIDE_RIGHT
+            | SPEAKER_FRONT_RIGHT_OF_CENTER | SPEAKER_TOP_FRONT_RIGHT | SPEAKER_TOP_BACK_RIGHT;
+
+        // the speaker mask is only present on WAVE_FORMAT_EXTENSIBLE formats
+        DWORD read_channel_mask(const WAVEFORMATEX *fmt) {
+            if (fmt->wFormatTag == WAVE_FORMAT_EXTENSIBLE
+                    && fmt->cbSize >= sizeof(WAVEFORMATEXTENSIBLE) - sizeof(WAVEFORMATEX)) {
+                return reinterpret_cast<const WAVEFORMATEXTENSIBLE *>(fmt)->dwChannelMask;
+            }
+            return 0;
+        }
+
+        // call visit(channel_index, speaker_bit) for each present speaker, in channel order
+        template <typename F>
+        void for_each_speaker(DWORD mask, int channels, F &&visit) {
+            int channel = 0;
+            for (int bit = 0; bit < 18 && channel < channels; bit++) {
+                const DWORD speaker = 1u << bit;
+                if (mask & speaker) {
+                    visit(channel++, speaker);
+                }
+            }
+        }
+    }
+
+    void Downmix::setup(const WAVEFORMATEX *game_format, WAVEFORMATEXTENSIBLE *stereo_out,
+            DownmixAlgorithm algorithm) {
+        this->enabled = true;
+        this->algorithm = algorithm;
+        this->bytes_per_sample = game_format->wBitsPerSample / 8;
+        this->game_frame_size = game_format->nChannels * this->bytes_per_sample;
+
+        this->is_float = is_ieee_float(game_format);
+
+        // supported: 16/24/32-bit integer PCM and 32-bit float; anything else mixes to silence
+        const bool supported = this->is_float
+            ? this->bytes_per_sample == 4
+            : (this->bytes_per_sample >= 2 && this->bytes_per_sample <= 4);
+        if (!supported) {
+            log_fatal(
+                "audio::downmix",
+                "unsupported sample format ({}-bit {}), downmix will output silence",
+                game_format->wBitsPerSample, this->is_float ? "float" : "int");
+        }
+
+        this->left_mix.clear();
+        this->right_mix.clear();
+        this->build_layout_mix(game_format);
+
+        make_stereo_format(game_format, stereo_out);
+    }
+
+    void Downmix::make_stereo_format(const WAVEFORMATEX *game_format,
+            WAVEFORMATEXTENSIBLE *stereo_out) {
+        const int bytes_per_sample = game_format->wBitsPerSample / 8;
+
+        memcpy(stereo_out, game_format, sizeof(WAVEFORMATEXTENSIBLE));
+        stereo_out->Format.nChannels = 2;
+        stereo_out->Format.nBlockAlign = 2 * bytes_per_sample;
+        stereo_out->Format.nAvgBytesPerSec =
+            game_format->nSamplesPerSec * stereo_out->Format.nBlockAlign;
+        stereo_out->dwChannelMask = SPEAKER_FRONT_LEFT | SPEAKER_FRONT_RIGHT;
+    }
+
+    HRESULT Downmix::initialize(IAudioClient *real, AUDCLNT_SHAREMODE share_mode, DWORD stream_flags,
+            REFERENCE_TIME buffer_duration, REFERENCE_TIME periodicity,
+            const WAVEFORMATEX *device_format, LPCGUID session_guid) {
+
+        // the smaller stereo buffer can end up unaligned for the device when the game sized the
+        // duration for its larger multi-channel format; the helper recovers from that.
+        return initialize_with_alignment_retry(real, "audio::downmix", share_mode, stream_flags,
+                buffer_duration, periodicity, device_format, session_guid);
+    }
+
+    void Downmix::add_channel(int channel, DWORD speaker, float gain) {
+        if (speaker & LEFT_SPEAKERS) {
+            this->left_mix.push_back({ channel, gain });
+        } else if (speaker & RIGHT_SPEAKERS) {
+            this->right_mix.push_back({ channel, gain });
+        } else { // center: feed both sides
+            this->left_mix.push_back({ channel, gain });
+            this->right_mix.push_back({ channel, gain });
+        }
+    }
+
+    // AC-4 stereo downmix (ETSI TS 103 190-1): front pair at unity, everything else -3 dB, LFE dropped
+    void Downmix::build_ac4_mix(DWORD mask, int channels) {
+        for_each_speaker(mask, channels, [&](int ch, DWORD speaker) {
+            if (speaker == SPEAKER_LOW_FREQUENCY) {
+                return;
+            }
+            const bool front_pair = speaker & (SPEAKER_FRONT_LEFT | SPEAKER_FRONT_RIGHT);
+            this->add_channel(ch, speaker, front_pair ? 1.0f : ATT_3DB);
+        });
+    }
+
+    // keep only the channels in `keep` (front/rear/side), each at unity gain
+    void Downmix::build_extract_mix(DWORD mask, int channels, DWORD keep) {
+        for_each_speaker(mask, channels, [&](int ch, DWORD speaker) {
+            if (speaker & keep) {
+                this->add_channel(ch, speaker, 1.0f);
+            }
+        });
+    }
+
+    // keep every channel (LFE dropped), then average each side so its gains sum to unity
+    void Downmix::build_normalize_mix(DWORD mask, int channels) {
+        for_each_speaker(mask, channels, [&](int ch, DWORD speaker) {
+            if (speaker != SPEAKER_LOW_FREQUENCY) {
+                this->add_channel(ch, speaker, 1.0f);
+            }
+        });
+
+        for (auto *mix : { &this->left_mix, &this->right_mix }) {
+            if (!mix->empty()) {
+                const float gain = 1.0f / mix->size();
+                for (auto &c : *mix) {
+                    c.gain = gain;
+                }
+            }
+        }
+    }
+
+    // fallback when no speaker mask is present: fold interleaved L/R pairs (even->left, odd->right)
+    void Downmix::build_pairs_mix(int channels, float gain) {
+        for (int ch = 0; ch < channels; ch++) {
+            (((ch & 1) == 0) ? this->left_mix : this->right_mix).push_back({ ch, gain });
+        }
+    }
+
+    void Downmix::build_layout_mix(const WAVEFORMATEX *game_format) {
+        const int channels = game_format->nChannels;
+        const DWORD mask = read_channel_mask(game_format);
+
+        // without a mask the layout is unknown: extract/normalize have nothing to act on, so all
+        // algorithms fall back to folding L/R pairs (AC-4 still attenuates by -3 dB)
+        if (mask == 0) {
+            this->build_pairs_mix(channels,
+                this->algorithm == DownmixAlgorithm::AC4 ? ATT_3DB : 1.0f);
+            return;
+        }
+
+        switch (this->algorithm) {
+            case DownmixAlgorithm::FrontOnly:
+                this->build_extract_mix(mask, channels,
+                    SPEAKER_FRONT_LEFT | SPEAKER_FRONT_RIGHT);
+                break;
+            case DownmixAlgorithm::RearOnly:
+                this->build_extract_mix(mask, channels,
+                    SPEAKER_BACK_LEFT | SPEAKER_BACK_RIGHT | SPEAKER_BACK_CENTER);
+                break;
+            case DownmixAlgorithm::SideOnly:
+                this->build_extract_mix(mask, channels,
+                    SPEAKER_SIDE_LEFT | SPEAKER_SIDE_RIGHT);
+                break;
+            case DownmixAlgorithm::Normalize:
+                this->build_normalize_mix(mask, channels);
+                break;
+            case DownmixAlgorithm::AC4:
+                this->build_ac4_mix(mask, channels);
+                break;
+        }
+    }
+
+    void Downmix::process(BYTE *dst, const BYTE *src, UINT32 frames) const {
+        const int bps = this->bytes_per_sample;
+        const int src_stride = this->game_frame_size;
+        const int dst_stride = 2 * bps;
+
+        if (dst == nullptr || src == nullptr || bps <= 0) {
+            return;
+        }
+
+        // sum each speaker's source channels into the matching stereo output
+        for (UINT32 i = 0; i < frames; i++) {
+            const BYTE *in = src + (size_t) i * src_stride;
+            BYTE *out = dst + (size_t) i * dst_stride;
+
+            float left = 0.0f;
+            float right = 0.0f;
+            for (const auto &c : this->left_mix) {
+                left += read_sample(in + c.channel * bps, bps, this->is_float) * c.gain;
+            }
+            for (const auto &c : this->right_mix) {
+                right += read_sample(in + c.channel * bps, bps, this->is_float) * c.gain;
+            }
+            write_sample(out, bps, this->is_float, left);
+            write_sample(out + bps, bps, this->is_float, right);
+        }
+    }
+
+    HRESULT Downmix::get_buffer(IAudioRenderClient *real, UINT32 frames, BYTE **ppData) {
+        const size_t needed = (size_t) frames * this->game_frame_size;
+        if (this->scratch.size() < needed) {
+            this->scratch.resize(needed);
+        }
+
+        HRESULT ret = real->GetBuffer(frames, &this->device_buffer);
+        if (FAILED(ret)) {
+            this->device_buffer = nullptr;
+            return ret;
+        }
+
+        *ppData = this->scratch.data();
+
+        return S_OK;
+    }
+
+    HRESULT Downmix::get_scratch(UINT32 frames, BYTE **ppData) {
+        const size_t needed = (size_t) frames * this->game_frame_size;
+        if (this->scratch.size() < needed) {
+            this->scratch.resize(needed);
+        }
+
+        *ppData = this->scratch.data();
+
+        return S_OK;
+    }
+
+    void Downmix::downmix_into(BYTE *dst, UINT32 frames) const {
+        this->process(dst, this->scratch.data(), frames);
+    }
+
+    void Downmix::write_device_buffer(UINT32 frames, DWORD flags) {
+        const int bps = this->bytes_per_sample;
+        const int dst_stride = 2 * bps;
+
+        if (this->device_buffer == nullptr || frames == 0 || bps <= 0) {
+            return;
+        }
+
+        // mute the first few buffers to avoid a pop on stream start
+        if (this->buffers_to_mute > 0) {
+            memset(this->device_buffer, 0, (size_t) frames * dst_stride);
+            this->buffers_to_mute--;
+        } else if ((flags & AUDCLNT_BUFFERFLAGS_SILENT) == 0) {
+            this->process(this->device_buffer, this->scratch.data(), frames);
+        }
+    }
+}

src/spice2x/hooks/audio/backends/wasapi/downmix.h

@@ -0,0 +1,150 @@
+#pragma once
+
+#include <optional>
+#include <vector>
+
+#include <windows.h>
+#include <mmreg.h>
+#include <audioclient.h>
+
+#include "hooks/audio/audio.h"
+
+struct IAudioClient;
+struct IAudioRenderClient;
+
+namespace hooks::audio {
+
+    // Generic WASAPI surround-to-stereo downmix. The real device is opened in stereo while the
+    // game keeps writing its native multi-channel audio into a scratch buffer; on release that
+    // buffer is mixed down into the two front channels.
+    //
+    // The mix is derived from the source format's speaker mask according to the selected
+    // DownmixAlgorithm:
+    //   FrontOnly / RearOnly / SideOnly - keep only that group of channels, routed to their side
+    //   AC4       - AC-4 stereo downmix coefficients (ETSI TS 103 190-1 §6.2.17): front left/right
+    //               pass at 0 dB, center and surrounds fold in at -3 dB, LFE dropped
+    //   Normalize - every channel folded in (center to both sides) with each output side averaged
+    //               so its channels are equally loud, LFE dropped
+    struct Downmix {
+
+        // a source channel routed into one output speaker at the given gain
+        struct Contribution {
+            int channel;
+            float gain;
+        };
+
+        // map an option value (front/rear/side/ac4/normalize) to its algorithm.
+        static std::optional<DownmixAlgorithm> name_to_algorithm(const char *value) {
+            if (_stricmp(value, "front") == 0) {
+                return DownmixAlgorithm::FrontOnly;
+            } else if (_stricmp(value, "rear") == 0) {
+                return DownmixAlgorithm::RearOnly;
+            } else if (_stricmp(value, "side") == 0) {
+                return DownmixAlgorithm::SideOnly;
+            } else if (_stricmp(value, "ac4") == 0) {
+                return DownmixAlgorithm::AC4;
+            } else if (_stricmp(value, "normalize") == 0) {
+                return DownmixAlgorithm::Normalize;
+            }
+
+            return std::nullopt;
+        }
+
+        // human-readable name of an algorithm, for logging.
+        static const char *algorithm_name(DownmixAlgorithm algorithm) {
+            switch (algorithm) {
+                case DownmixAlgorithm::FrontOnly: return "front";
+                case DownmixAlgorithm::RearOnly: return "rear";
+                case DownmixAlgorithm::SideOnly: return "side";
+                case DownmixAlgorithm::AC4: return "ac4";
+                case DownmixAlgorithm::Normalize: return "normalize";
+                default: return "unknown";
+            }
+        }
+
+        // whether the downmix is active for the current stream
+        bool enabled = false;
+
+        // algorithm used to fold the multi-channel audio into stereo
+        DownmixAlgorithm algorithm = DownmixAlgorithm::AC4;
+
+        // size in bytes of one frame of the game's multi-channel format
+        int game_frame_size = 0;
+
+        // size in bytes of a single sample (per channel)
+        int bytes_per_sample = 0;
+
+        // whether samples are IEEE floating point rather than integer PCM
+        bool is_float = false;
+
+        // enable the downmix for the given game format and fill stereo_out with the equivalent
+        // stereo format to open the real device with.
+        void setup(const WAVEFORMATEX *game_format, WAVEFORMATEXTENSIBLE *stereo_out,
+                   DownmixAlgorithm algorithm);
+
+        // build the stereo format equivalent to game_format (same sample rate and bit depth).
+        static void make_stereo_format(const WAVEFORMATEX *game_format,
+                                       WAVEFORMATEXTENSIBLE *stereo_out);
+
+        // initialize the real device with the stereo format. downmixing reduces the channel count,
+        // shrinking the buffer's byte size, so the duration the game sized for its multi-channel
+        // format can leave the smaller stereo buffer unaligned. on AUDCLNT_E_BUFFER_SIZE_NOT_ALIGNED
+        // this performs the standard WASAPI realignment and retries.
+        HRESULT initialize(IAudioClient *real, AUDCLNT_SHAREMODE share_mode, DWORD stream_flags,
+                           REFERENCE_TIME buffer_duration, REFERENCE_TIME periodicity,
+                           const WAVEFORMATEX *device_format, LPCGUID session_guid);
+
+        // mix `frames` frames of multi-channel `src` down into stereo `dst`.
+        void process(BYTE *dst, const BYTE *src, UINT32 frames) const;
+
+        // grab the real stereo device buffer and hand the game the scratch buffer to write into.
+        HRESULT get_buffer(IAudioRenderClient *real, UINT32 frames, BYTE **ppData);
+
+        // size the scratch and hand it to the game without acquiring a device buffer. used when a
+        // later stage (the resampler) owns the device interaction.
+        HRESULT get_scratch(UINT32 frames, BYTE **ppData);
+
+        // downmix the scratch the game wrote into the caller's stereo buffer, without touching the
+        // device. used to feed the resampler when the two stages are chained.
+        void downmix_into(BYTE *dst, UINT32 frames) const;
+
+        // mix the scratch buffer into the stereo device buffer held since get_buffer. the caller
+        // owns releasing the device buffer afterwards (see current_buffer / buffer_released).
+        void write_device_buffer(UINT32 frames, DWORD flags);
+
+        // the real device buffer currently held, or null.
+        BYTE *current_buffer() const { return this->device_buffer; }
+
+        // forget the held device buffer once the caller has released it.
+        void buffer_released() { this->device_buffer = nullptr; }
+
+    private:
+
+        // build the mix from the source speaker layout for the selected algorithm
+        void build_layout_mix(const WAVEFORMATEX *game_format);
+
+        // per-algorithm builders, each filling left_mix / right_mix from the speaker mask
+        void build_ac4_mix(DWORD mask, int channels);
+        void build_extract_mix(DWORD mask, int channels, DWORD keep);
+        void build_normalize_mix(DWORD mask, int channels);
+
+        // fallback for streams without a speaker mask: fold interleaved L/R pairs at `gain`
+        void build_pairs_mix(int channels, float gain);
+
+        // append one source channel to the output side(s) matching its speaker, at `gain`
+        void add_channel(int channel, DWORD speaker, float gain);
+
+        // source channels summed into each output speaker
+        std::vector<Contribution> left_mix;
+        std::vector<Contribution> right_mix;
+
+        // buffer the game writes its multi-channel audio into between get/release
+        std::vector<BYTE> scratch;
+
+        // the real stereo device buffer currently held, or null
+        BYTE *device_buffer = nullptr;
+
+        // leading buffers to silence to avoid a pop on stream start
+        int buffers_to_mute = 16;
+    };
+}

src/spice2x/hooks/audio/backends/wasapi/dummy_audio_client.cpp

@@ -73,11 +73,7 @@ HRESULT STDMETHODCALLTYPE DummyIAudioClient::Initialize(
 
     // verbose output
     log_info("audio::wasapi", "IAudioClient::Initialize hook hit");
-    log_info("audio::wasapi", "... ShareMode         : {}", share_mode_str(ShareMode));
-    log_info("audio::wasapi", "... StreamFlags       : {}", stream_flags_str(StreamFlags));
-    log_info("audio::wasapi", "... hnsBufferDuration : {}", hnsBufferDuration);
-    log_info("audio::wasapi", "... hnsPeriodicity    : {}", hnsPeriodicity);
-    print_format(pFormat);
+    print_format(ShareMode, StreamFlags, hnsBufferDuration, hnsPeriodicity, pFormat);
 
     CHECK_RESULT(this->backend->on_initialize(
         &ShareMode,

src/spice2x/hooks/audio/backends/wasapi/resample.cpp

@@ -0,0 +1,437 @@
+#include "resample.h"
+
+#include <algorithm>
+#include <cmath>
+#include <cstdint>
+#include <cstring>
+#include <mutex>
+
+#include <audioclient.h>
+
+#include "util/logging.h"
+
+#include "util.h"
+
+namespace hooks::audio {
+
+    namespace {
+
+        constexpr double PI = 3.14159265358979323846;
+
+        // normalized sinc: sin(pi*x) / (pi*x), with the removable singularity at 0 filled in
+        inline double sinc(double x) {
+            if (x == 0.0) {
+                return 1.0;
+            }
+            const double px = PI * x;
+            return std::sin(px) / px;
+        }
+
+        // Blackman window across the kernel radius; zero at +/- radius
+        inline double blackman(double x, double radius) {
+            const double n = (x + radius) / (2.0 * radius);
+            if (n <= 0.0 || n >= 1.0) {
+                return 0.0;
+            }
+            return 0.42 - 0.5 * std::cos(2.0 * PI * n) + 0.08 * std::cos(4.0 * PI * n);
+        }
+    }
+
+    std::optional<uint32_t> Resampler::resolve(const WAVEFORMATEX *game_format) {
+        if (game_format == nullptr || !RESAMPLE_RATE.has_value()) {
+            return std::nullopt;
+        }
+        if (game_format->nSamplesPerSec == 0
+                || game_format->nSamplesPerSec == RESAMPLE_RATE.value()) {
+            return std::nullopt;
+        }
+        return RESAMPLE_RATE;
+    }
+
+    void Resampler::setup(const WAVEFORMATEX *game_format, WAVEFORMATEXTENSIBLE *device_out,
+            uint32_t target_rate) {
+        this->enabled = true;
+        this->channels = game_format->nChannels;
+        this->bytes_per_sample = game_format->wBitsPerSample / 8;
+        this->game_frame_size = this->channels * this->bytes_per_sample;
+
+        this->is_float = is_ieee_float(game_format);
+
+        const bool supported = this->is_float
+            ? this->bytes_per_sample == 4
+            : (this->bytes_per_sample >= 2 && this->bytes_per_sample <= 4);
+        if (!supported) {
+            log_fatal(
+                "audio::resample",
+                "unsupported sample format ({}-bit {}) for -resample",
+                game_format->wBitsPerSample, this->is_float ? "float" : "int");
+        }
+
+        this->src_rate = game_format->nSamplesPerSec;
+        this->dst_rate = target_rate;
+
+        // anti-alias cutoff: full bandwidth when upsampling, scaled down when decimating
+        this->cutoff = std::min(1.0, (double) this->dst_rate / (double) this->src_rate);
+        this->half_taps = 16;
+
+        // precompute the windowed-sinc kernel now that cutoff is known
+        this->build_kernel();
+
+        // prime the queue with half a window of silence so the first outputs have left history
+        this->in_queue.assign((size_t) this->half_taps * this->channels, 0.0f);
+        this->in_pos = this->half_taps;
+
+        this->make_device_format(game_format, device_out, target_rate);
+    }
+
+    void Resampler::make_device_format(const WAVEFORMATEX *game_format,
+            WAVEFORMATEXTENSIBLE *device_out, uint32_t target_rate) {
+        const size_t src_size = sizeof(WAVEFORMATEX) + game_format->cbSize;
+
+        memset(device_out, 0, sizeof(WAVEFORMATEXTENSIBLE));
+        memcpy(device_out, game_format, std::min(src_size, sizeof(WAVEFORMATEXTENSIBLE)));
+
+        device_out->Format.nSamplesPerSec = target_rate;
+        device_out->Format.nAvgBytesPerSec = target_rate * device_out->Format.nBlockAlign;
+    }
+
+    HRESULT Resampler::initialize(IAudioClient *real, AUDCLNT_SHAREMODE share_mode,
+            DWORD stream_flags, REFERENCE_TIME buffer_duration, REFERENCE_TIME periodicity,
+            const WAVEFORMATEX *device_format, LPCGUID session_guid) {
+
+        // the resampler bypasses the OS mixer and talks to the device directly, so it only makes
+        // sense (and only works) for exclusive streams. shared streams are already resampled by
+        // the Windows audio engine, so refuse loudly rather than silently doing nothing.
+        if (share_mode != AUDCLNT_SHAREMODE_EXCLUSIVE) {
+            log_fatal("audio::resample",
+                    "-resample requires WASAPI exclusive mode, but this stream is shared "
+                    "(Windows already resamples shared streams)");
+        }
+
+        // record the pacing model. event-driven streams fill the whole device buffer each period
+        // (produce_exact); timer-driven streams poll padding and write variable partial chunks, so
+        // they drain the pending output to the device's free space each call (flush_timer).
+        this->event_driven = (stream_flags & AUDCLNT_STREAMFLAGS_EVENTCALLBACK) != 0;
+
+        return initialize_with_alignment_retry(real, "audio::resample", share_mode, stream_flags,
+                buffer_duration, periodicity, device_format, session_guid);
+    }
+
+    UINT32 Resampler::frames_device_to_game(UINT32 device_frames) const {
+        if (this->dst_rate == 0) {
+            return device_frames;
+        }
+        // round down so the game never believes it has more room than the device can hold
+        return (UINT32) (((double) device_frames * this->src_rate) / this->dst_rate);
+    }
+
+    UINT32 Resampler::padding_device_to_game(UINT32 device_padding) const {
+        if (this->dst_rate == 0) {
+            return device_padding;
+        }
+        // round up so the reported free space stays conservative
+        return (UINT32) std::ceil(((double) device_padding * this->src_rate) / this->dst_rate);
+    }
+
+    HRESULT Resampler::get_buffer(UINT32 frames, BYTE **ppData) {
+        const size_t needed = (size_t) frames * this->game_frame_size;
+        if (this->scratch.size() < needed) {
+            this->scratch.resize(needed);
+        }
+
+        *ppData = this->scratch.data();
+
+        return S_OK;
+    }
+
+    void Resampler::enqueue_input(UINT32 frames, bool silent) {
+        const int bps = this->bytes_per_sample;
+        const int ch = this->channels;
+        const size_t base = this->in_queue.size();
+
+        this->in_queue.resize(base + (size_t) frames * ch);
+
+        if (silent || bps <= 0 || ch <= 0) {
+            std::fill(this->in_queue.begin() + base, this->in_queue.end(), 0.0f);
+            return;
+        }
+
+        const BYTE *src = this->scratch.data();
+        for (UINT32 f = 0; f < frames; f++) {
+            for (int c = 0; c < ch; c++) {
+                const size_t s = (size_t) f * ch + c;
+                this->in_queue[base + s] = read_sample(src + s * bps, bps, this->is_float);
+            }
+        }
+    }
+
+    void Resampler::build_kernel() {
+        const int taps = 2 * this->half_taps;
+        const int phases = this->kernel_phases;
+        const double cut = this->cutoff;
+        const double radius = (double) this->half_taps;
+
+        // one extra row at frac == 1.0 so emit_frame can interpolate against row p + 1 safely
+        this->kernel_table.resize((size_t) (phases + 1) * taps);
+
+        for (int p = 0; p <= phases; p++) {
+            const double frac = (double) p / (double) phases;
+            for (int k = 0; k < taps; k++) {
+                // tap k maps to input offset t = k - (half_taps - 1), matching emit_frame
+                const double x = frac - (double) (k - (this->half_taps - 1));
+                this->kernel_table[(size_t) p * taps + k] =
+                        (float) (cut * sinc(cut * x) * blackman(x, radius));
+            }
+        }
+    }
+
+    void Resampler::emit_frame() {
+        const int ch = this->channels;
+        const int radius = this->half_taps;
+        const int taps = 2 * radius;
+        const long avail = (long) (this->in_queue.size() / ch);
+        const long center = (long) std::floor(this->in_pos);
+
+        // pick the two kernel rows bracketing this fractional position and the blend between them
+        const double frac = this->in_pos - (double) center;
+        const double fp = frac * (double) this->kernel_phases;
+        const int p0 = (int) fp;
+        const float blend = (float) (fp - (double) p0);
+        const float *row0 = &this->kernel_table[(size_t) p0 * taps];
+        const float *row1 = &this->kernel_table[(size_t) (p0 + 1) * taps];
+
+        // base input index for tap 0 (t = -(radius - 1))
+        const long base = center - (radius - 1);
+
+        for (int c = 0; c < ch; c++) {
+            double acc = 0.0;
+            for (int k = 0; k < taps; k++) {
+                const long idx = base + k;
+                if (idx < 0 || idx >= avail) {
+                    continue;
+                }
+                const float w = row0[k] + blend * (row1[k] - row0[k]);
+                acc += (double) this->in_queue[(size_t) idx * ch + c] * w;
+            }
+            this->out_float.push_back((float) acc);
+        }
+    }
+
+    void Resampler::drop_consumed() {
+        const int ch = this->channels;
+        const long drop = (long) std::floor(this->in_pos) - this->half_taps;
+        if (drop > 0) {
+            const size_t drop_samples = (size_t) drop * ch;
+            if (drop_samples <= this->in_queue.size()) {
+                this->in_queue.erase(this->in_queue.begin(),
+                        this->in_queue.begin() + drop_samples);
+                this->in_pos -= drop;
+            }
+        }
+    }
+
+    UINT32 Resampler::produce_exact(UINT32 out_frames) {
+        const int ch = this->channels;
+        this->out_float.clear();
+        if (ch <= 0 || out_frames == 0) {
+            return 0;
+        }
+        this->out_float.reserve((size_t) out_frames * ch);
+
+        // resample ratio. drive it from the buffer size actually advertised to the game rather
+        // than the nominal src/dst ratio: GetBufferSize reports floor(dev_buf * src/dst) game
+        // frames, so the game only ever delivers that many input frames per device period.
+        // consuming at the nominal ratio would eat slightly more input than arrives on any device
+        // where dev_buf * src/dst is non-integer (e.g. 144 -> 132.3, floored to 132), slowly
+        // draining the queue until it underruns to permanent silence. using the advertised integer
+        // ratio keeps input and output exactly balanced; the resulting pitch error is below 0.3%
+        // and inaudible, and it collapses to the exact ratio when the division is integer (160 ->
+        // 147 stays 147/160 = 44100/48000).
+        const double step = (double) this->frames_device_to_game(this->device_buffer_frames)
+                / (double) this->device_buffer_frames;
+
+        // input frames the block will touch: from in_pos through the right edge of the sinc kernel
+        // at the final output sample. if the queue is short of this, the kernel tail reads past the
+        // end and distorts every buffer, so buffer one extra block of input before the first output
+        // (emitting silence without consuming) to build a cushion the kernel can always reach into.
+        const long avail = (long) (this->in_queue.size() / ch);
+        const long need = (long) std::ceil(this->in_pos + step * (double) out_frames)
+                + this->half_taps;
+
+        if (this->priming) {
+            if (avail < need + (long) out_frames) {
+                this->out_float.assign((size_t) out_frames * ch, 0.0f);
+                return out_frames;
+            }
+            this->priming = false;
+        }
+
+        for (UINT32 o = 0; o < out_frames; o++) {
+            this->emit_frame();
+            this->in_pos += step;
+        }
+
+        this->drop_consumed();
+        return out_frames;
+    }
+
+    UINT32 Resampler::produce_variable() {
+        const int ch = this->channels;
+        if (ch <= 0) {
+            return 0;
+        }
+
+        // input frames consumed per output frame. timer-driven streams write variable partial
+        // chunks, so produce however many output frames the currently queued input can fully
+        // support and leave the rest for the next call; this keeps input and output balanced at
+        // the exact src/dst ratio over time without depending on the device buffer size.
+        const double step = (double) this->src_rate / (double) this->dst_rate;
+        const long avail = (long) (this->in_queue.size() / ch);
+
+        // emit only while the sinc kernel's right edge stays within the queued input. the kernel
+        // reaches from in_pos out to half_taps frames ahead, so stop once that would read past the
+        // end; the remaining input becomes the next block's lookahead.
+        UINT32 produced = 0;
+        while ((long) std::ceil(this->in_pos) + this->half_taps < avail) {
+            this->emit_frame();
+            this->in_pos += step;
+            produced++;
+        }
+
+        this->drop_consumed();
+        return produced;
+    }
+
+    void Resampler::write_output(BYTE *dst, UINT32 frames, float gain) const {
+        const int bps = this->bytes_per_sample;
+        const int ch = this->channels;
+        const size_t count = (size_t) frames * ch;
+
+        for (size_t i = 0; i < count; i++) {
+            write_sample(dst + i * bps, bps, this->is_float, this->out_float[i] * gain);
+        }
+    }
+
+    HRESULT Resampler::flush(IAudioRenderClient *real, IAudioClient *client, UINT32 frames,
+            DWORD flags, float boost) {
+        if (!this->enabled) {
+            return S_OK;
+        }
+
+        // cache the device buffer size once
+        if (this->device_buffer_frames == 0) {
+            client->GetBufferSize(&this->device_buffer_frames);
+        }
+        if (this->device_buffer_frames == 0) {
+            return S_OK;
+        }
+
+        const bool silent = (flags & AUDCLNT_BUFFERFLAGS_SILENT) != 0;
+        this->enqueue_input(frames, silent);
+
+        // confirm once that conversion actually started producing output
+        static std::once_flag active_printed;
+        std::call_once(active_printed, [this]() {
+            log_info("audio::resample", "resample active: {} Hz -> {} Hz ({} ch, {})",
+                    this->src_rate, this->dst_rate, this->channels,
+                    this->event_driven ? "event-driven" : "timer-driven");
+        });
+
+        // the boost is applied here (inside write_output) rather than in the standard ReleaseBuffer
+        // path, so log it once for parity with that path's "volume boost active" line.
+        if (boost != 1.0f) {
+            static std::once_flag boost_printed;
+            std::call_once(boost_printed, [boost]() {
+                log_info("audio::resample", "volume boost active (resample): gain={}", boost);
+            });
+        }
+
+        return this->event_driven
+                ? this->flush_event(real, boost)
+                : this->flush_timer(real, client, boost);
+    }
+
+    HRESULT Resampler::flush_event(IAudioRenderClient *real, float boost) {
+
+        // event-driven exclusive streams must hand the device a full buffer every period and may
+        // not push partial counts. resample the whole input block into exactly the device buffer
+        // size.
+        const UINT32 produced = this->produce_exact(this->device_buffer_frames);
+        if (produced == 0) {
+            return S_OK;
+        }
+
+        BYTE *dev = nullptr;
+        HRESULT ret = real->GetBuffer(produced, &dev);
+        if (FAILED(ret) || dev == nullptr) {
+            return ret;
+        }
+
+        // mute the first few buffers to avoid a pop on stream start
+        float gain = boost;
+        if (this->buffers_to_mute > 0) {
+            gain = 0.0f;
+            this->buffers_to_mute--;
+        }
+
+        this->write_output(dev, produced, gain);
+
+        return real->ReleaseBuffer(produced, 0);
+    }
+
+    HRESULT Resampler::flush_timer(IAudioRenderClient *real, IAudioClient *client, float boost) {
+
+        // convert everything currently queued into the pending output FIFO (out_float). timer-
+        // driven games write variable partial chunks, so produce only what the queued input can
+        // fully support and keep the remainder for the next call.
+        this->produce_variable();
+
+        const int ch = this->channels;
+        if (ch <= 0) {
+            return S_OK;
+        }
+
+        const UINT32 pending = (UINT32) (this->out_float.size() / ch);
+        if (pending == 0) {
+            return S_OK;
+        }
+
+        // push as many frames as the device currently has free, keeping the rest queued for the
+        // next call. timer-driven games poll padding and write whenever there is room, so matching
+        // the device's free space here avoids overflowing the ring while staying device-paced.
+        UINT32 padding = 0;
+        if (FAILED(client->GetCurrentPadding(&padding))) {
+            return S_OK;
+        }
+        const UINT32 device_free = this->device_buffer_frames > padding
+                ? this->device_buffer_frames - padding
+                : 0;
+        if (device_free == 0) {
+            return S_OK;
+        }
+
+        const UINT32 to_write = std::min(pending, device_free);
+
+        BYTE *dev = nullptr;
+        HRESULT ret = real->GetBuffer(to_write, &dev);
+        if (FAILED(ret) || dev == nullptr) {
+            return ret;
+        }
+
+        // mute the first few buffers to avoid a pop on stream start
+        float gain = boost;
+        if (this->buffers_to_mute > 0) {
+            gain = 0.0f;
+            this->buffers_to_mute--;
+        }
+
+        this->write_output(dev, to_write, gain);
+        ret = real->ReleaseBuffer(to_write, 0);
+
+        // drop the frames just written from the front of the pending FIFO
+        this->out_float.erase(this->out_float.begin(),
+                this->out_float.begin() + (size_t) to_write * ch);
+
+        return ret;
+    }
+}

src/spice2x/hooks/audio/backends/wasapi/resample.h

@@ -0,0 +1,149 @@
+#pragma once
+
+#include <cstdint>
+#include <optional>
+#include <vector>
+
+#include <windows.h>
+#include <mmreg.h>
+#include <audioclient.h>
+
+#include "hooks/audio/audio.h"
+
+struct IAudioClient;
+struct IAudioRenderClient;
+
+namespace hooks::audio {
+
+    // Streaming sample-rate converter for the WASAPI render path. The real device is opened at the
+    // target rate while the game keeps writing its native-rate audio into a scratch buffer; on
+    // release that buffer is converted with a windowed-sinc kernel and pushed to the device.
+    // Channel count and sample format are preserved; only the sample rate changes.
+    //
+    // Frame counts differ between the two rates, so unlike the per-frame downmix this is stateful:
+    // a fractional read position and a window of input history carry across ReleaseBuffer calls,
+    // and the device buffer is only filled up to the space the device currently has free.
+    struct Resampler {
+
+        // whether the resampler is active for the current stream
+        bool enabled = false;
+
+        // whether the stream is event-driven (AUDCLNT_STREAMFLAGS_EVENTCALLBACK). timer-driven
+        // streams instead poll padding and write variable partial chunks, so they drain the
+        // pending output to the device's free space rather than pushing a full buffer per period.
+        bool event_driven = true;
+
+        // decide whether the stream should be resampled and to which rate. returns the target rate
+        // when RESAMPLE_RATE is set and differs from the game's rate, otherwise nullopt.
+        static std::optional<uint32_t> resolve(const WAVEFORMATEX *game_format);
+
+        // enable resampling for game_format and fill device_out with the equivalent format at the
+        // target rate to open the real device with.
+        void setup(const WAVEFORMATEX *game_format, WAVEFORMATEXTENSIBLE *device_out,
+                   uint32_t target_rate);
+
+        // build the device format equivalent to game_format at target_rate (same channels/depth).
+        static void make_device_format(const WAVEFORMATEX *game_format,
+                                       WAVEFORMATEXTENSIBLE *device_out, uint32_t target_rate);
+
+        // initialize the real device at the target rate, performing the standard WASAPI buffer
+        // realignment retry on AUDCLNT_E_BUFFER_SIZE_NOT_ALIGNED.
+        HRESULT initialize(IAudioClient *real, AUDCLNT_SHAREMODE share_mode, DWORD stream_flags,
+                           REFERENCE_TIME buffer_duration, REFERENCE_TIME periodicity,
+                           const WAVEFORMATEX *device_format, LPCGUID session_guid);
+
+        // translate a device-rate frame count to the equivalent game-rate count, so the buffer-size
+        // and padding values reported to the game stay paced at the game's native rate.
+        UINT32 frames_device_to_game(UINT32 device_frames) const;
+        UINT32 padding_device_to_game(UINT32 device_padding) const;
+
+        // hand the game a scratch buffer sized for `frames` of its native format to write into.
+        HRESULT get_buffer(UINT32 frames, BYTE **ppData);
+
+        // pointer to the input scratch (sized by get_buffer). when chained after the downmix, the
+        // downmix writes its stereo output here for the resampler to consume on the next flush.
+        BYTE *input_data() { return this->scratch.data(); }
+
+        // convert the `frames` the game wrote and push output to the real render client. `boost`
+        // is applied to the converted output. event-driven streams fill exactly one device buffer
+        // per period; timer-driven streams push as many converted frames as the device has free.
+        HRESULT flush(IAudioRenderClient *real, IAudioClient *client, UINT32 frames, DWORD flags,
+                      float boost);
+
+    private:
+
+        // append `frames` of the scratch buffer (native format), or silence, to the input queue
+        void enqueue_input(UINT32 frames, bool silent);
+
+        // event-driven path: produce exactly one full device buffer and push it.
+        HRESULT flush_event(IAudioRenderClient *real, float boost);
+
+        // timer-driven path: convert all queued input into the pending output FIFO, then push as
+        // many frames as the device currently has free, keeping the remainder for the next call.
+        HRESULT flush_timer(IAudioRenderClient *real, IAudioClient *client, float boost);
+
+        // produce exactly out_frames output frames using the fixed src/dst ratio. event-driven
+        // exclusive streams must fill the whole device buffer every period; a small input cushion
+        // is buffered first (see priming) so the sinc kernel always has lookahead.
+        UINT32 produce_exact(UINT32 out_frames);
+
+        // convert all input the kernel can fully support into the pending output FIFO (out_float),
+        // appending without clearing. returns the number of frames produced. used by the
+        // timer-driven path where output is drained to the device in device-paced chunks.
+        UINT32 produce_variable();
+
+        // convolve the windowed-sinc kernel at the current in_pos and append the resulting frame
+        // (one sample per channel) to out_float
+        void emit_frame();
+
+        // precompute the windowed-sinc kernel sampled at kernel_phases sub-sample positions, so
+        // emit_frame is a table lookup instead of recomputing sin/cos per tap (which is far too
+        // expensive to run per sample on the audio callback thread and causes underrun crackle).
+        void build_kernel();
+
+        // drop input frames that in_pos has advanced past, keeping a window of history for the
+        // next block's left context
+        void drop_consumed();
+
+        // convert the first `frames` of out_float to the device format, scaled by `gain`
+        void write_output(BYTE *dst, UINT32 frames, float gain) const;
+
+        // sample format of the stream
+        int channels = 0;
+        int bytes_per_sample = 0;
+        bool is_float = false;
+        int game_frame_size = 0;
+
+        uint32_t src_rate = 0;
+        uint32_t dst_rate = 0;
+
+        // sinc low-pass cutoff (1.0 when upsampling, dst/src when downsampling) and window radius
+        double cutoff = 1.0;
+        int half_taps = 16;
+
+        // precomputed kernel: (kernel_phases + 1) rows of 2*half_taps weights, indexed by the
+        // fractional sample position (linearly interpolated between adjacent rows in emit_frame)
+        std::vector<float> kernel_table;
+        int kernel_phases = 1024;
+
+        // interleaved float input queue and the fractional read position within it (in frames)
+        std::vector<float> in_queue;
+        double in_pos = 0.0;
+
+        // emit silence until a full block of input lookahead has accumulated, so the sinc kernel
+        // never reads past the end of the queue (which would distort the tail of every buffer)
+        bool priming = true;
+
+        // interleaved float scratch for produced output
+        std::vector<float> out_float;
+
+        // buffer the game writes its native-rate audio into between get_buffer / flush
+        std::vector<BYTE> scratch;
+
+        // cached device buffer size (frames); a full buffer is produced every period
+        UINT32 device_buffer_frames = 0;
+
+        // leading buffers to silence to avoid a pop on stream start
+        int buffers_to_mute = 16;
+    };
+}

src/spice2x/hooks/audio/backends/wasapi/shared.cpp

@@ -0,0 +1,78 @@
+#include "shared.h"
+
+#include <audioclient.h>
+
+#include "hooks/audio/audio.h"
+#include "util/logging.h"
+
+#include "defs.h"
+
+namespace hooks::audio {
+
+    // whether the engine's PCM converter can handle this format. PCM / float only; non-PCM
+    // bitstream (AC-3 / DTS passthrough) must be left alone.
+    static bool is_pcm_or_float(const WAVEFORMATEX *format) {
+        if (format == nullptr) {
+            return false;
+        }
+
+        switch (format->wFormatTag) {
+            case WAVE_FORMAT_PCM:
+            case WAVE_FORMAT_IEEE_FLOAT:
+                return true;
+            case WAVE_FORMAT_EXTENSIBLE: {
+
+                // SubFormat is only valid when the extra-bytes block is large enough
+                if (format->cbSize < sizeof(WAVEFORMATEXTENSIBLE) - sizeof(WAVEFORMATEX)) {
+                    return false;
+                }
+                const auto *ext = reinterpret_cast<const WAVEFORMATEXTENSIBLE *>(format);
+                return ext->SubFormat == GUID_KSDATAFORMAT_SUBTYPE_PCM
+                        || ext->SubFormat == GUID_KSDATAFORMAT_SUBTYPE_IEEE_FLOAT;
+            }
+            default:
+                return false;
+        }
+    }
+
+    bool SharedRedirect::wants(AUDCLNT_SHAREMODE share_mode, const WAVEFORMATEX *format) {
+
+        // only redirect PCM / float exclusive streams: the engine converter (AUTOCONVERTPCM) can
+        // handle those, but non-PCM bitstream (AC-3 / DTS passthrough) would fail in shared mode,
+        // so leave it in exclusive untouched.
+        return hooks::audio::WASAPI_COMPATIBILITY_MODE
+                && share_mode == AUDCLNT_SHAREMODE_EXCLUSIVE
+                && is_pcm_or_float(format);
+    }
+
+    void SharedRedirect::apply(AUDCLNT_SHAREMODE *share_mode, DWORD *stream_flags,
+            REFERENCE_TIME *periodicity) {
+
+        // shared mode requires periodicity == 0; AUTOCONVERTPCM lets the engine accept the game's
+        // native format (else shared Initialize returns AUDCLNT_E_UNSUPPORTED_FORMAT).
+        log_info("audio::wasapi", "redirecting exclusive WASAPI to shared mode");
+        *share_mode = AUDCLNT_SHAREMODE_SHARED;
+        *periodicity = 0;
+        *stream_flags |= AUDCLNT_STREAMFLAGS_AUTOCONVERTPCM | AUDCLNT_STREAMFLAGS_SRC_DEFAULT_QUALITY;
+        this->redirected_from_exclusive = true;
+    }
+
+    UINT32 SharedRedirect::clamp_buffer_size(IAudioClient *real, uint32_t sample_rate,
+            UINT32 device_frames) const {
+        if (!this->redirected_from_exclusive || real == nullptr || sample_rate == 0 || device_frames == 0) {
+            return device_frames;
+        }
+
+        // GetDevicePeriod returns REFERENCE_TIME units (100 ns), 10^7 per second, so
+        // period_frames = period * sample_rate / 10^7.
+        REFERENCE_TIME period = 0;
+        if (SUCCEEDED(real->GetDevicePeriod(&period, nullptr)) && period > 0) {
+            const UINT32 period_frames = (UINT32) ((period * sample_rate) / 10000000);
+            if (period_frames > 0 && period_frames < device_frames) {
+                return period_frames;
+            }
+        }
+
+        return device_frames;
+    }
+}

src/spice2x/hooks/audio/backends/wasapi/shared.h

@@ -0,0 +1,31 @@
+#pragma once
+
+#include <cstdint>
+
+#include <windows.h>
+#include <audioclient.h>
+
+namespace hooks::audio {
+
+    // The -wasapishared option redirects an exclusive WASAPI stream to shared mode. lets other apps
+    // play sound and works on devices that can't open the exclusive format, at the cost of some
+    // latency. Only PCM / float is converted; bitstream (AC-3 / DTS) is left alone.
+    struct SharedRedirect {
+
+        // true once apply() has redirected an exclusive request. gates the buffer clamp; stays false
+        // for a natively-shared stream (it paces itself, so must not be clamped).
+        bool redirected_from_exclusive = false;
+
+        // whether an exclusive-mode request should be redirected, given the -wasapishared option.
+        // only PCM / float is eligible; bitstream (AC-3 / DTS) is left in exclusive mode.
+        static bool wants(AUDCLNT_SHAREMODE share_mode, const WAVEFORMATEX *format);
+
+        // redirect an exclusive request to shared mode. caller must have checked wants() first.
+        void apply(AUDCLNT_SHAREMODE *share_mode, DWORD *stream_flags, REFERENCE_TIME *periodicity);
+
+        // clamp a reported buffer size to one device period. some games write the whole buffer per
+        // event, which overflows shared-mode buffering (AUDCLNT_E_BUFFER_TOO_LARGE -> stutter); one
+        // period always fits. a no-op unless an exclusive request was redirected.
+        UINT32 clamp_buffer_size(IAudioClient *real, uint32_t sample_rate, UINT32 device_frames) const;
+    };
+}

src/spice2x/hooks/audio/backends/wasapi/util.cpp

@@ -2,7 +2,9 @@
 
 #include <audioclient.h>
 
+#include "hooks/audio/util.h"
 #include "util/flags_helper.h"
+#include "util/logging.h"
 
 #include "defs.h"
 
@@ -18,3 +20,46 @@ std::string stream_flags_str(DWORD flags) {
     FLAG(flags, AUDCLNT_STREAMFLAGS_SRC_DEFAULT_QUALITY);
     FLAGS_END(flags);
 }
+
+void print_format(AUDCLNT_SHAREMODE share_mode, DWORD stream_flags, REFERENCE_TIME buffer_duration,
+        REFERENCE_TIME periodicity, const WAVEFORMATEX *device_format) {
+    log_info("audio::wasapi", "... ShareMode         : {}", share_mode_str(share_mode));
+    log_info("audio::wasapi", "... StreamFlags       : {}", stream_flags_str(stream_flags));
+    log_info("audio::wasapi", "... hnsBufferDuration : {} ({:.3f} ms)",
+            buffer_duration, buffer_duration / 10000.0);
+    log_info("audio::wasapi", "... hnsPeriodicity    : {} ({:.3f} ms)",
+            periodicity, periodicity / 10000.0);
+    print_format(device_format);
+}
+
+void print_format(AUDCLNT_SHAREMODE share_mode, const WAVEFORMATEX *device_format) {
+    log_info("audio::wasapi", "... ShareMode         : {}", share_mode_str(share_mode));
+    print_format(device_format);
+}
+
+HRESULT initialize_with_alignment_retry(IAudioClient *client, const char *log_group,
+        AUDCLNT_SHAREMODE share_mode, DWORD stream_flags, REFERENCE_TIME buffer_duration,
+        REFERENCE_TIME periodicity, const WAVEFORMATEX *device_format, LPCGUID session_guid) {
+
+    HRESULT ret = client->Initialize(share_mode, stream_flags, buffer_duration, periodicity,
+            device_format, session_guid);
+
+    // the requested buffer size can end up unaligned for the device; recover by asking for the next
+    // aligned buffer size and re-initializing with a matching duration.
+    if (ret == AUDCLNT_E_BUFFER_SIZE_NOT_ALIGNED) {
+        UINT32 aligned_frames = 0;
+        if (SUCCEEDED(client->GetBufferSize(&aligned_frames)) && aligned_frames > 0) {
+            REFERENCE_TIME aligned_duration = (REFERENCE_TIME)
+                    (10000.0 * 1000 / device_format->nSamplesPerSec * aligned_frames + 0.5);
+
+            log_info(log_group, "buffer not aligned, retrying with {} frames ({} hns)",
+                    aligned_frames, aligned_duration);
+
+            ret = client->Initialize(share_mode, stream_flags, aligned_duration,
+                    periodicity != 0 ? aligned_duration : 0, device_format, session_guid);
+        }
+    }
+
+    return ret;
+}
+

src/spice2x/hooks/audio/backends/wasapi/util.h

@@ -1,8 +1,100 @@
 #pragma once
 
+#include <algorithm>
+#include <cmath>
+#include <cstdint>
+#include <cstring>
 #include <string>
 
 #include <windows.h>
 #include <mmreg.h>
+#include <audioclient.h>
 
 std::string stream_flags_str(DWORD flags);
+
+// log the stream parameters (share mode, flags, buffer duration, periodicity) followed by the wave
+// format, matching the block printed at the top of IAudioClient::Initialize.
+void print_format(AUDCLNT_SHAREMODE share_mode, DWORD stream_flags, REFERENCE_TIME buffer_duration,
+        REFERENCE_TIME periodicity, const WAVEFORMATEX *device_format);
+
+// log the share mode followed by the wave format, for paths that only have a share mode (e.g.
+// IAudioClient::IsFormatSupported).
+void print_format(AUDCLNT_SHAREMODE share_mode, const WAVEFORMATEX *device_format);
+
+// detect IEEE float samples: WAVE_FORMAT_IEEE_FLOAT, or WAVE_FORMAT_EXTENSIBLE whose SubFormat is
+// KSDATAFORMAT_SUBTYPE_IEEE_FLOAT (Data1 == 3; _PCM has Data1 == 1)
+inline bool is_ieee_float(const WAVEFORMATEX *fmt) {
+    return fmt->wFormatTag == WAVE_FORMAT_IEEE_FLOAT
+        || (fmt->wFormatTag == WAVE_FORMAT_EXTENSIBLE
+            && reinterpret_cast<const WAVEFORMATEXTENSIBLE *>(fmt)->SubFormat.Data1 == 0x00000003);
+}
+
+// read one sample at `p` as a normalized float in [-1, 1]
+inline float read_sample(const BYTE *p, int bytes, bool is_float) {
+    if (is_float) {
+        float v;
+        memcpy(&v, p, sizeof(float));
+        return v;
+    }
+    switch (bytes) {
+        case 2: {
+            int16_t v;
+            memcpy(&v, p, sizeof(v));
+            return v * (1.0f / 32768.0f);
+        }
+        case 3: {
+            int32_t v = p[0] | (p[1] << 8) | (p[2] << 16);
+            if (v & 0x800000) {
+                v |= ~0xFFFFFF; // sign extend
+            }
+            return v * (1.0f / 8388608.0f);
+        }
+        case 4: {
+            int32_t v;
+            memcpy(&v, p, sizeof(v));
+            return (float) (v * (1.0 / 2147483648.0));
+        }
+        default:
+            return 0.0f;
+    }
+}
+
+// write the normalized float `value` to the sample at `p`, clamping to the format's range
+inline void write_sample(BYTE *p, int bytes, bool is_float, float value) {
+    if (is_float) {
+        float v = std::clamp(value, -1.0f, 1.0f);
+        memcpy(p, &v, sizeof(v));
+        return;
+    }
+    switch (bytes) {
+        case 2: {
+            int16_t v = (int16_t) std::clamp(
+                (int) std::lround(value * 32768.0f), -32768, 32767);
+            memcpy(p, &v, sizeof(v));
+            break;
+        }
+        case 3: {
+            int32_t v = (int32_t) std::clamp<int64_t>(
+                std::llround((double) value * 8388608.0), -8388608, 8388607);
+            p[0] = v & 0xFF;
+            p[1] = (v >> 8) & 0xFF;
+            p[2] = (v >> 16) & 0xFF;
+            break;
+        }
+        case 4: {
+            int32_t v = (int32_t) std::clamp<int64_t>(
+                std::llround((double) value * 2147483648.0), INT32_MIN, INT32_MAX);
+            memcpy(p, &v, sizeof(v));
+            break;
+        }
+        default:
+            break;
+    }
+}
+
+// initialize the real audio client, recovering from AUDCLNT_E_BUFFER_SIZE_NOT_ALIGNED by asking the
+// device for the next aligned buffer size and re-initializing with a matching duration. log_group
+// names the subsystem in the retry log line.
+HRESULT initialize_with_alignment_retry(IAudioClient *client, const char *log_group,
+        AUDCLNT_SHAREMODE share_mode, DWORD stream_flags, REFERENCE_TIME buffer_duration,
+        REFERENCE_TIME periodicity, const WAVEFORMATEX *device_format, LPCGUID session_guid);

src/spice2x/hooks/audio/implementations/wave_out.cpp

@@ -3,6 +3,7 @@
 #include "hooks/audio/audio.h"
 #include "hooks/audio/backends/wasapi/audio_client.h"
 #include "hooks/audio/backends/wasapi/defs.h"
+#include "util/precise_timer.h"
 
 static REFERENCE_TIME WASAPI_TARGET_REFTIME = TARGET_REFTIME;
 
@@ -185,6 +186,7 @@ HRESULT WaveOutBackend::on_get_buffer(uint32_t num_frames_requested, BYTE **ppDa
 }
 HRESULT WaveOutBackend::on_release_buffer(uint32_t num_frames_written, DWORD dwFlags) {
     bool written = false;
+    timeutils::PreciseSleepTimer timer;
 
     // reset the dispatcher event
     ResetEvent(this->dispatcher_event);
@@ -209,7 +211,7 @@ HRESULT WaveOutBackend::on_release_buffer(uint32_t num_frames_written, DWORD dwF
 
         // avoid pegging the CPU
         if (!written) {
-            Sleep(1);
+            timer.sleep(1);
         }
     }
 

src/spice2x/hooks/audio/util.cpp

@@ -48,33 +48,33 @@ void copy_wave_format(WAVEFORMATEXTENSIBLE *destination, const WAVEFORMATEX *sou
 }
 
 void print_format(const WAVEFORMATEX *pFormat) {
-    log_info("audio", "Wave Format:");
+    log_info("audio::wasapi", "Wave Format:");
 
     // format specific
     if (pFormat->wFormatTag == WAVE_FORMAT_EXTENSIBLE) {
         auto format = reinterpret_cast<const WAVEFORMATEXTENSIBLE *>(pFormat);
-        log_info("audio", "... SubFormat           : {}", guid2s(format->SubFormat));
+        log_info("audio::wasapi", "... SubFormat           : {}", guid2s(format->SubFormat));
     } else {
-        log_info("audio", "... wFormatTag          : {}", pFormat->wFormatTag);
+        log_info("audio::wasapi", "... wFormatTag          : {}", pFormat->wFormatTag);
     }
 
     // generic
-    log_info("audio", "... nChannels           : {}", pFormat->nChannels);
-    log_info("audio", "... nSamplesPerSec      : {}", pFormat->nSamplesPerSec);
-    log_info("audio", "... nAvgBytesPerSec     : {}", pFormat->nAvgBytesPerSec);
-    log_info("audio", "... nBlockAlign         : {}", pFormat->nBlockAlign);
-    log_info("audio", "... wBitsPerSample      : {}", pFormat->wBitsPerSample);
+    log_info("audio::wasapi", "... nChannels           : {}", pFormat->nChannels);
+    log_info("audio::wasapi", "... nSamplesPerSec      : {}", pFormat->nSamplesPerSec);
+    log_info("audio::wasapi", "... nAvgBytesPerSec     : {}", pFormat->nAvgBytesPerSec);
+    log_info("audio::wasapi", "... nBlockAlign         : {}", pFormat->nBlockAlign);
+    log_info("audio::wasapi", "... wBitsPerSample      : {}", pFormat->wBitsPerSample);
 
     // format specific
     if (pFormat->wFormatTag == WAVE_FORMAT_EXTENSIBLE) {
         auto format = reinterpret_cast<const WAVEFORMATEXTENSIBLE *>(pFormat);
 
         if (pFormat->wBitsPerSample == 0) {
-            log_info("audio", "... wSamplesPerBlock    : {}", format->Samples.wSamplesPerBlock);
+            log_info("audio::wasapi", "... wSamplesPerBlock    : {}", format->Samples.wSamplesPerBlock);
         } else {
-            log_info("audio", "... wValidBitsPerSample : {}", format->Samples.wValidBitsPerSample);
+            log_info("audio::wasapi", "... wValidBitsPerSample : {}", format->Samples.wValidBitsPerSample);
         }
 
-        log_info("audio", "... dwChannelMask       : {}", channel_mask_str(format->dwChannelMask));
+        log_info("audio::wasapi", "... dwChannelMask       : {}", channel_mask_str(format->dwChannelMask));
     }
 }

src/spice2x/hooks/graphics/backends/d3d11/d3d11_backend.cpp

@@ -0,0 +1,304 @@
+// dx11 / dxgi hook entrypoint. trampolines d3d11.dll / dxgi.dll exports
+// the moment those DLLs appear (LDR notification + poll-thread fallback),
+// then drives proactive vtable capture so we don't lose the race against
+// the execexe loader. per-vtable hook implementations live in the sibling
+// files (d3d11_swapchain / d3d11_factory / d3d11_vtable_capture /
+// d3d11_screenshot).
+//
+// note: never LoadLibrary d3d11/dxgi -- execexe pre-loads them itself and
+// fails (error 0xa) if they're already in the loader's module list.
+//
+// 64-bit only.
+
+#include "d3d11_backend.h"
+
+#ifndef SPICE_D3D11
+
+void graphics_d3d11_init() {}
+void graphics_d3d11_shutdown() {}
+
+#else
+
+#include <atomic>
+#include <thread>
+#include <chrono>
+#include <cwchar>
+#include <mutex>
+
+#include <windows.h>
+#include <d3d11.h>
+#include <dxgi.h>
+#include <dxgi1_2.h>
+
+#include "d3d11_internal.h"
+#include "util/nt_loader.h"
+
+namespace {
+
+using D3D11CreateDeviceAndSwapChain_t = HRESULT(WINAPI *)(
+    IDXGIAdapter *, D3D_DRIVER_TYPE, HMODULE, UINT,
+    const D3D_FEATURE_LEVEL *, UINT, UINT,
+    const DXGI_SWAP_CHAIN_DESC *, IDXGISwapChain **,
+    ID3D11Device **, D3D_FEATURE_LEVEL *, ID3D11DeviceContext **);
+using CreateDXGIFactory_t = HRESULT(WINAPI *)(REFIID, void **);
+using CreateDXGIFactory1_t = HRESULT(WINAPI *)(REFIID, void **);
+using CreateDXGIFactory2_t = HRESULT(WINAPI *)(UINT, REFIID, void **);
+
+D3D11CreateDeviceAndSwapChain_t D3D11CreateDeviceAndSwapChain_orig = nullptr;
+CreateDXGIFactory_t CreateDXGIFactory_orig = nullptr;
+CreateDXGIFactory1_t CreateDXGIFactory1_orig = nullptr;
+CreateDXGIFactory2_t CreateDXGIFactory2_orig = nullptr;
+
+std::atomic<bool> g_d3d11_exports_hooked { false };
+std::atomic<bool> g_dxgi_exports_hooked { false };
+
+// ----------------------------------------------------------------------
+// top-level export hooks
+
+HRESULT WINAPI D3D11CreateDeviceAndSwapChain_hook(
+        IDXGIAdapter *pAdapter, D3D_DRIVER_TYPE DriverType, HMODULE Software, UINT Flags,
+        const D3D_FEATURE_LEVEL *pFeatureLevels, UINT FeatureLevels, UINT SDKVersion,
+        const DXGI_SWAP_CHAIN_DESC *pSwapChainDesc, IDXGISwapChain **ppSwapChain,
+        ID3D11Device **ppDevice, D3D_FEATURE_LEVEL *pFeatureLevel,
+        ID3D11DeviceContext **ppImmediateContext)
+{
+    HRESULT res = D3D11CreateDeviceAndSwapChain_orig(
+        pAdapter, DriverType, Software, Flags,
+        pFeatureLevels, FeatureLevels, SDKVersion,
+        pSwapChainDesc, ppSwapChain, ppDevice, pFeatureLevel, ppImmediateContext);
+    if (SUCCEEDED(res) && ppSwapChain && *ppSwapChain) {
+        if (pSwapChainDesc) {
+            d3d11_hooks::note_main_hwnd(pSwapChainDesc->OutputWindow);
+        }
+        d3d11_hooks::install_swapchain_hooks(*ppSwapChain);
+    }
+    return res;
+}
+
+#define DEFINE_FACTORY_HOOK(NAME, SIG_PARAMS, ORIG_ARGS)                       \
+    HRESULT WINAPI NAME##_hook SIG_PARAMS {                                    \
+        HRESULT res = NAME##_orig ORIG_ARGS;                                   \
+        if (SUCCEEDED(res) && ppFactory && *ppFactory) {                       \
+            d3d11_hooks::install_factory_hooks(                                \
+                reinterpret_cast<IUnknown *>(*ppFactory));                     \
+        }                                                                      \
+        return res;                                                            \
+    }
+
+DEFINE_FACTORY_HOOK(CreateDXGIFactory,
+    (REFIID riid, void **ppFactory),
+    (riid, ppFactory))
+DEFINE_FACTORY_HOOK(CreateDXGIFactory1,
+    (REFIID riid, void **ppFactory),
+    (riid, ppFactory))
+DEFINE_FACTORY_HOOK(CreateDXGIFactory2,
+    (UINT Flags, REFIID riid, void **ppFactory),
+    (Flags, riid, ppFactory))
+
+#undef DEFINE_FACTORY_HOOK
+
+// ----------------------------------------------------------------------
+// export trampoline plumbing
+
+// serializes trampoline_export() so the LDR notification callback and the
+// poll thread don't race each other into MinHook against the same target.
+std::mutex g_export_mutex;
+
+bool trampoline_export(const char *dll, const char *name, void *hook, void **orig) {
+    std::lock_guard<std::mutex> lock(g_export_mutex);
+    if (*orig) {
+        return true;
+    }
+    HMODULE mod = GetModuleHandleA(dll);
+    if (!mod) {
+        return false;
+    }
+    void *addr = reinterpret_cast<void *>(GetProcAddress(mod, name));
+    if (!addr) {
+        return false;
+    }
+    *orig = addr; // trampoline_try reads *orig before overwriting it.
+    if (!detour::trampoline_try(addr, hook, orig)) {
+        *orig = nullptr;
+        return false;
+    }
+    log_info("graphics::d3d11", "trampolined {}!{}", dll, name);
+    return true;
+}
+
+void try_install_d3d11_exports() {
+    if (g_d3d11_exports_hooked) {
+        return;
+    }
+    if (trampoline_export("d3d11.dll", "D3D11CreateDeviceAndSwapChain",
+            (void *) D3D11CreateDeviceAndSwapChain_hook,
+            (void **) &D3D11CreateDeviceAndSwapChain_orig)) {
+        g_d3d11_exports_hooked = true;
+    }
+}
+
+void try_install_dxgi_exports() {
+    if (g_dxgi_exports_hooked) {
+        return;
+    }
+    struct entry { const char *name; void *hook; void **orig; };
+    const entry entries[] = {
+        { "CreateDXGIFactory", (void *) CreateDXGIFactory_hook,
+          (void **) &CreateDXGIFactory_orig },
+        { "CreateDXGIFactory1", (void *) CreateDXGIFactory1_hook,
+          (void **) &CreateDXGIFactory1_orig },
+        { "CreateDXGIFactory2", (void *) CreateDXGIFactory2_hook,
+          (void **) &CreateDXGIFactory2_orig },
+    };
+    bool any = false;
+    for (auto &e : entries) {
+        any |= trampoline_export("dxgi.dll", e.name, e.hook, e.orig);
+    }
+    if (any) {
+        g_dxgi_exports_hooked = true;
+    }
+}
+
+void try_capture_if_ready() {
+    if (g_d3d11_exports_hooked && g_dxgi_exports_hooked) {
+        d3d11_hooks::try_capture_vtables();
+    }
+}
+
+// ----------------------------------------------------------------------
+// LDR notification + polling fallback
+
+bool dll_name_ends_with(PCUNICODE_STRING name, const wchar_t *suffix) {
+    if (!name || !name->Buffer) {
+        return false;
+    }
+    const size_t n = name->Length / sizeof(WCHAR);
+    const size_t s = wcslen(suffix);
+    return n >= s && _wcsnicmp(name->Buffer + n - s, suffix, s) == 0;
+}
+
+VOID CALLBACK ldr_dll_notification(
+        ULONG reason, PCLDR_DLL_NOTIFICATION_DATA data, PVOID /*context*/)
+{
+    if (reason != LDR_DLL_NOTIFICATION_REASON_LOADED || !data) {
+        return;
+    }
+    if (dll_name_ends_with(data->Loaded.BaseDllName, L"d3d11.dll")) {
+        try_install_d3d11_exports();
+    } else if (dll_name_ends_with(data->Loaded.BaseDllName, L"dxgi.dll")) {
+        try_install_dxgi_exports();
+    }
+}
+
+// execexe maps d3d11/dxgi via a path that bypasses LdrLoadDll, so the
+// notification above never fires for those DLLs and we have to poll.
+std::atomic<bool> g_stop { false };
+std::thread g_poll_thread;
+std::mutex g_init_mutex;
+PVOID g_ldr_cookie = nullptr;
+
+void poll_thread() {
+    using namespace std::chrono_literals;
+    for (int32_t i = 0; i < 120 && !g_stop.load(); ++i) {
+        try_install_d3d11_exports();
+        try_install_dxgi_exports();
+        if (g_d3d11_exports_hooked && g_dxgi_exports_hooked) {
+            d3d11_hooks::try_capture_vtables();
+            return;
+        }
+        // sliced so shutdown doesn't have to wait a full second.
+        for (int32_t s = 0; s < 10 && !g_stop.load(); ++s) {
+            std::this_thread::sleep_for(100ms);
+        }
+    }
+}
+
+// the overlay's imgui dx11 backend needs D3DCompile (d3dcompiler_XX.dll) to
+// build its shaders. _43 ships with the DX June 2010 redist on stock Win7;
+// _46/_47 come with newer Windows.
+bool d3dcompiler_available() {
+    static const wchar_t *names[] = {
+        L"d3dcompiler_47.dll",
+        L"d3dcompiler_46.dll",
+        L"d3dcompiler_43.dll",
+    };
+    for (auto name : names) {
+        HMODULE mod = GetModuleHandleW(name);
+        if (!mod) {
+            mod = LoadLibraryW(name);
+        }
+        if (mod && GetProcAddress(mod, "D3DCompile")) {
+            return true;
+        }
+    }
+    return false;
+}
+
+} // namespace
+
+void graphics_d3d11_init() {
+    // dx11 titles always run under execexe. skipping on pure-dx9 games keeps
+    // their startup path completely untouched (no exports patched, no poll
+    // thread, no LDR callback).
+    if (!GetModuleHandleW(L"execexe.dll")) {
+        return;
+    }
+
+    // no d3dcompiler -> overlay can't build shaders; skip dx11 overlay
+    if (!d3dcompiler_available()) {
+        log_warning(
+            "graphics::d3d11",
+            "d3dcompiler not found; dx11 overlay disabled");
+        return;
+    }
+
+    std::lock_guard<std::mutex> lock(g_init_mutex);
+    if (g_poll_thread.joinable()) {
+        return; // already initialized
+    }
+
+    log_info("graphics::d3d11", "initializing");
+
+    // trampoline now if either DLL is already in the PEB.
+    try_install_d3d11_exports();
+    try_install_dxgi_exports();
+    try_capture_if_ready();
+
+    // catches standard LdrLoadDll loads.
+    auto reg = reinterpret_cast<decltype(&LdrRegisterDllNotification)>(
+        GetProcAddress(GetModuleHandleW(L"ntdll.dll"), "LdrRegisterDllNotification"));
+    if (reg) {
+        NTSTATUS st = reg(0, ldr_dll_notification, nullptr, &g_ldr_cookie);
+        if (NT_SUCCESS(st)) {
+            log_info("graphics::d3d11", "registered LDR DLL notification");
+        } else {
+            g_ldr_cookie = nullptr;
+            log_warning("graphics::d3d11",
+                "LdrRegisterDllNotification failed: {:#x}", (unsigned long)st);
+        }
+    }
+
+    // catches the execexe loader path that bypasses LdrLoadDll.
+    g_poll_thread = std::thread(poll_thread);
+}
+
+void graphics_d3d11_shutdown() {
+    std::lock_guard<std::mutex> lock(g_init_mutex);
+
+    // unregister first so the callback can't fire mid-teardown.
+    if (g_ldr_cookie) {
+        auto unreg = reinterpret_cast<decltype(&LdrUnregisterDllNotification)>(
+            GetProcAddress(GetModuleHandleW(L"ntdll.dll"), "LdrUnregisterDllNotification"));
+        if (unreg) {
+            unreg(g_ldr_cookie);
+        }
+        g_ldr_cookie = nullptr;
+    }
+
+    g_stop.store(true);
+    if (g_poll_thread.joinable()) {
+        g_poll_thread.join();
+    }
+}
+
+#endif // SPICE_D3D11

src/spice2x/hooks/graphics/backends/d3d11/d3d11_backend.h

@@ -0,0 +1,24 @@
+#pragma once
+
+#include "overlay/overlay.h"
+
+void graphics_d3d11_init();
+void graphics_d3d11_shutdown();
+
+#ifdef SPICE_D3D11
+
+struct ID3D11Device;
+struct ID3D11DeviceContext;
+struct ID3D11RenderTargetView;
+struct IDXGISwapChain;
+
+namespace overlay::d3d11 {
+
+    void render(ID3D11Device *device,
+                ID3D11DeviceContext *context,
+                IDXGISwapChain *swapchain,
+                ID3D11RenderTargetView **rtv);
+
+}
+
+#endif

src/spice2x/hooks/graphics/backends/d3d11/d3d11_factory.cpp

@@ -0,0 +1,102 @@
+// dx11 factory vtable hooks. patches CreateSwapChain / CreateSwapChainForHwnd
+// so we can install_swapchain_hooks against every newly-created swapchain.
+
+#include "d3d11_backend.h"
+
+#ifdef SPICE_D3D11
+
+#include <mutex>
+
+#include <windows.h>
+#include <d3d11.h>
+#include <dxgi.h>
+#include <dxgi1_2.h>
+
+#include "d3d11_internal.h"
+
+namespace {
+
+using CreateSwapChain_t = HRESULT(STDMETHODCALLTYPE *)(
+    IDXGIFactory *, IUnknown *, DXGI_SWAP_CHAIN_DESC *, IDXGISwapChain **);
+using CreateSwapChainForHwnd_t = HRESULT(STDMETHODCALLTYPE *)(
+    IDXGIFactory2 *, IUnknown *, HWND,
+    const DXGI_SWAP_CHAIN_DESC1 *,
+    const DXGI_SWAP_CHAIN_FULLSCREEN_DESC *,
+    IDXGIOutput *, IDXGISwapChain1 **);
+
+CreateSwapChain_t CreateSwapChain_orig = nullptr;
+CreateSwapChainForHwnd_t CreateSwapChainForHwnd_orig = nullptr;
+
+bool g_factory_hooked = false;
+bool g_factory2_hooked = false;
+std::mutex g_hook_mutex;
+
+HRESULT STDMETHODCALLTYPE CreateSwapChain_hook(
+        IDXGIFactory *factory, IUnknown *pDevice,
+        DXGI_SWAP_CHAIN_DESC *pDesc, IDXGISwapChain **ppSwapChain)
+{
+    HRESULT res = CreateSwapChain_orig(factory, pDevice, pDesc, ppSwapChain);
+    if (SUCCEEDED(res) && ppSwapChain && *ppSwapChain) {
+        if (pDesc) {
+            d3d11_hooks::note_main_hwnd(pDesc->OutputWindow);
+        }
+        d3d11_hooks::install_swapchain_hooks(*ppSwapChain);
+    }
+    return res;
+}
+
+HRESULT STDMETHODCALLTYPE CreateSwapChainForHwnd_hook(
+        IDXGIFactory2 *factory, IUnknown *pDevice, HWND hWnd,
+        const DXGI_SWAP_CHAIN_DESC1 *pDesc,
+        const DXGI_SWAP_CHAIN_FULLSCREEN_DESC *pFullscreenDesc,
+        IDXGIOutput *pRestrictToOutput, IDXGISwapChain1 **ppSwapChain)
+{
+    HRESULT res = CreateSwapChainForHwnd_orig(
+        factory, pDevice, hWnd, pDesc, pFullscreenDesc, pRestrictToOutput, ppSwapChain);
+    if (SUCCEEDED(res) && ppSwapChain && *ppSwapChain) {
+        d3d11_hooks::note_main_hwnd(hWnd);
+        d3d11_hooks::install_swapchain_hooks(*ppSwapChain);
+    }
+    return res;
+}
+
+// QI-and-hook helper: dedupes the IDXGIFactory / IDXGIFactory2 install paths.
+template<typename Iface>
+void install_on(IUnknown *factory, bool &flag,
+                size_t vtbl_index, void *hook, void **orig, const char *name)
+{
+    if (flag) {
+        return;
+    }
+    Iface *f = nullptr;
+    if (FAILED(factory->QueryInterface(IID_PPV_ARGS(&f))) || !f) {
+        return;
+    }
+    if (d3d11_hooks::hook_vtbl(f, vtbl_index, hook, orig, name)) {
+        flag = true;
+    }
+    f->Release();
+}
+
+} // namespace
+
+namespace d3d11_hooks {
+
+void install_factory_hooks(IUnknown *factory) {
+    if (!factory) {
+        return;
+    }
+    std::lock_guard<std::mutex> lock(g_hook_mutex);
+
+    install_on<IDXGIFactory>(factory, g_factory_hooked, 10,
+        (void *) CreateSwapChain_hook, (void **) &CreateSwapChain_orig,
+        "IDXGIFactory::CreateSwapChain");
+
+    install_on<IDXGIFactory2>(factory, g_factory2_hooked, 15,
+        (void *) CreateSwapChainForHwnd_hook, (void **) &CreateSwapChainForHwnd_orig,
+        "IDXGIFactory2::CreateSwapChainForHwnd");
+}
+
+}
+
+#endif // SPICE_D3D11

src/spice2x/hooks/graphics/backends/d3d11/d3d11_internal.h

@@ -0,0 +1,59 @@
+#pragma once
+
+// internal glue for the dx11 backend. all symbols gated on SPICE_D3D11.
+
+#include "overlay/overlay.h"
+
+#ifdef SPICE_D3D11
+
+#include <memory>
+
+#include "util/detour.h"
+#include "util/logging.h"
+
+struct HWND__; typedef HWND__ *HWND;
+struct IUnknown;
+struct IDXGISwapChain;
+
+namespace d3d11_hooks {
+
+    void install_swapchain_hooks(IDXGISwapChain *swapchain);
+    void install_factory_hooks(IUnknown *factory);
+    void try_capture_vtables();
+
+    // first non-null swapchain HWND wins; later ones (sub-screens, IME
+    // helpers) are ignored. the dummy capture window is exempted via
+    // ignore_hwnd.
+    void note_main_hwnd(HWND hwnd);
+    HWND main_hwnd();
+    void ignore_hwnd(HWND hwnd);
+
+    // capture backbuffer to PNG if a screenshot was requested.
+    void try_screenshot(IDXGISwapChain *swapchain);
+
+    // trampoline a virtual method by vtable index. on failure *orig is null.
+    inline bool hook_vtbl(void *iface, size_t index,
+                          void *hook, void **orig, const char *name)
+    {
+        void **vtbl = *reinterpret_cast<void ***>(iface);
+        void *target = vtbl[index];
+        // trampoline_try reads *orig before overwriting it.
+        *orig = target;
+        if (!detour::trampoline_try(target, hook, orig)) {
+            *orig = nullptr;
+            log_warning("graphics::d3d11", "failed to hook {}", name);
+            return false;
+        }
+        log_info("graphics::d3d11", "hooked {}", name);
+        return true;
+    }
+
+    // minimal COM RAII used by capture / screenshot paths.
+    struct com_release {
+        void operator()(IUnknown *p) const { if (p) p->Release(); }
+    };
+    template<typename T> using com_ptr = std::unique_ptr<T, com_release>;
+
+}
+
+#endif

src/spice2x/hooks/graphics/backends/d3d11/d3d11_screenshot.cpp

@@ -0,0 +1,165 @@
+// dx11 screenshot capture. mirrors the d3d9 backend: copy the current
+// backbuffer into a staging texture, force alpha=255, write PNG via
+// stb_image_write, push to clipboard and notify.
+
+#include "d3d11_backend.h"
+
+#ifdef SPICE_D3D11
+
+#include <vector>
+
+#include <windows.h>
+#include <d3d11.h>
+#include <dxgi.h>
+
+#include "d3d11_internal.h"
+
+#include "external/stb_image_write.h"
+#include "hooks/graphics/graphics.h"
+#include "misc/clipboard.h"
+#include "overlay/notifications.h"
+#include "util/fileutils.h"
+
+using d3d11_hooks::com_ptr;
+
+namespace {
+
+// copy the swapchain backbuffer into a CPU-readable staging texture and
+// flatten it into an RGBA8 buffer (BGRA backbuffers are swizzled,
+// alpha is forced to 255).
+bool copy_backbuffer_to_rgba(IDXGISwapChain *swapchain,
+                             ID3D11Device *device,
+                             ID3D11DeviceContext *context,
+                             std::vector<uint8_t> &out,
+                             uint32_t &out_w, uint32_t &out_h)
+{
+    ID3D11Texture2D *raw_bb = nullptr;
+    if (FAILED(swapchain->GetBuffer(0, IID_PPV_ARGS(&raw_bb))) || !raw_bb) {
+        return false;
+    }
+    com_ptr<ID3D11Texture2D> backbuffer(raw_bb);
+
+    D3D11_TEXTURE2D_DESC desc {};
+    backbuffer->GetDesc(&desc);
+
+    // MSAA backbuffers can't be CopyResource'd into a non-MS staging target.
+    com_ptr<ID3D11Texture2D> resolved;
+    ID3D11Texture2D *source = backbuffer.get();
+    if (desc.SampleDesc.Count > 1) {
+        D3D11_TEXTURE2D_DESC rd = desc;
+        rd.SampleDesc.Count = 1;
+        rd.SampleDesc.Quality = 0;
+        rd.Usage = D3D11_USAGE_DEFAULT;
+        rd.BindFlags = D3D11_BIND_RENDER_TARGET;
+        rd.CPUAccessFlags = 0;
+        rd.MiscFlags = 0;
+        ID3D11Texture2D *r = nullptr;
+        if (FAILED(device->CreateTexture2D(&rd, nullptr, &r)) || !r) {
+            return false;
+        }
+        resolved.reset(r);
+        context->ResolveSubresource(resolved.get(), 0, backbuffer.get(), 0, desc.Format);
+        source = resolved.get();
+    }
+
+    D3D11_TEXTURE2D_DESC sd {};
+    sd.Width = desc.Width;
+    sd.Height = desc.Height;
+    sd.MipLevels = 1;
+    sd.ArraySize = 1;
+    sd.Format = desc.Format;
+    sd.SampleDesc.Count = 1;
+    sd.Usage = D3D11_USAGE_STAGING;
+    sd.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
+
+    ID3D11Texture2D *raw_staging = nullptr;
+    if (FAILED(device->CreateTexture2D(&sd, nullptr, &raw_staging)) || !raw_staging) {
+        return false;
+    }
+    com_ptr<ID3D11Texture2D> staging(raw_staging);
+    context->CopyResource(staging.get(), source);
+
+    D3D11_MAPPED_SUBRESOURCE mapped {};
+    if (FAILED(context->Map(staging.get(), 0, D3D11_MAP_READ, 0, &mapped))) {
+        return false;
+    }
+
+    // backbuffers from GetDesc are always fully-typed (never _TYPELESS).
+    const bool is_bgra = desc.Format == DXGI_FORMAT_B8G8R8A8_UNORM
+                      || desc.Format == DXGI_FORMAT_B8G8R8A8_UNORM_SRGB;
+
+    out.resize(static_cast<size_t>(desc.Width) * desc.Height * 4);
+    const uint8_t *src_base = reinterpret_cast<const uint8_t *>(mapped.pData);
+    for (uint32_t y = 0; y < desc.Height; ++y) {
+        const uint8_t *row = src_base + static_cast<size_t>(y) * mapped.RowPitch;
+        uint8_t *dst = out.data() + static_cast<size_t>(y) * desc.Width * 4;
+        for (uint32_t x = 0; x < desc.Width; ++x) {
+            dst[x * 4 + 0] = row[x * 4 + (is_bgra ? 2 : 0)];
+            dst[x * 4 + 1] = row[x * 4 + 1];
+            dst[x * 4 + 2] = row[x * 4 + (is_bgra ? 0 : 2)];
+            dst[x * 4 + 3] = 255;
+        }
+    }
+
+    context->Unmap(staging.get(), 0);
+
+    out_w = desc.Width;
+    out_h = desc.Height;
+    return true;
+}
+
+} // namespace
+
+namespace d3d11_hooks {
+
+void try_screenshot(IDXGISwapChain *swapchain) {
+    if (!swapchain || !graphics_screenshot_consume()) {
+        return;
+    }
+
+    auto file_path = graphics_screenshot_genpath();
+    if (file_path.empty()) {
+        return;
+    }
+
+    ID3D11Device *raw_device = nullptr;
+    if (FAILED(swapchain->GetDevice(IID_PPV_ARGS(&raw_device))) || !raw_device) {
+        return;
+    }
+    com_ptr<ID3D11Device> device(raw_device);
+    ID3D11DeviceContext *raw_ctx = nullptr;
+    device->GetImmediateContext(&raw_ctx);
+    if (!raw_ctx) {
+        return;
+    }
+    com_ptr<ID3D11DeviceContext> context(raw_ctx);
+
+    std::vector<uint8_t> pixels;
+    uint32_t w = 0, h = 0;
+    if (!copy_backbuffer_to_rgba(swapchain, device.get(), context.get(), pixels, w, h)) {
+        log_warning("graphics::d3d11", "screenshot: failed to capture backbuffer");
+        overlay::notifications::add(
+            overlay::notifications::Severity::Error,
+            "Screenshot failed to capture");
+        return;
+    }
+
+    log_info("graphics::d3d11", "saving screenshot to {}", file_path);
+    if (stbi_write_png(file_path.c_str(), (int) w, (int) h, 4,
+                       pixels.data(), (int) w * 4))
+    {
+        clipboard::copy_image(file_path);
+        overlay::notifications::add(
+            overlay::notifications::Severity::Success,
+            fmt::format("Screenshot saved: {}", fileutils::basename(file_path)));
+    } else {
+        log_warning("graphics::d3d11", "screenshot: stbi_write_png failed");
+        overlay::notifications::add(
+            overlay::notifications::Severity::Error,
+            "Screenshot failed to save");
+    }
+}
+
+}
+
+#endif // SPICE_D3D11

src/spice2x/hooks/graphics/backends/d3d11/d3d11_swapchain.cpp

@@ -0,0 +1,289 @@
+// dx11 swapchain vtable hooks + per-frame overlay pump.
+//
+// dxgi shares vtables across swapchain instances, so we only need to patch
+// Present / Present1 / ResizeBuffers once on the first instance we see.
+// each frame we lazily attach the overlay to whichever swapchain is
+// presenting, then drive its imgui update / new_frame / render cycle.
+
+#include "d3d11_backend.h"
+
+#ifdef SPICE_D3D11
+
+#include <atomic>
+#include <mutex>
+
+#include <windows.h>
+#include <d3d11.h>
+#include <dxgi.h>
+#include <dxgi1_2.h>
+
+#include "d3d11_internal.h"
+
+#include "external/imgui/imgui.h"
+#include "external/imgui/backends/imgui_impl_dx11.h"
+#include "overlay/imgui/impl_spice.h"
+
+#include "games/io.h"
+#include "hooks/graphics/graphics.h"
+#include "launcher/launcher.h"
+#include "misc/eamuse.h"
+
+// --------------------------------------------------------------------------
+// overlay render bridge
+
+namespace overlay::d3d11 {
+
+    // sRGB backbuffers need a UNORM view: ImGui vertex colors are already
+    // sRGB-encoded, so an extra linear->sRGB conversion would wash the
+    // overlay out white.
+    static DXGI_FORMAT to_unorm_view(DXGI_FORMAT fmt) {
+        switch (fmt) {
+            case DXGI_FORMAT_R8G8B8A8_UNORM_SRGB: return DXGI_FORMAT_R8G8B8A8_UNORM;
+            case DXGI_FORMAT_B8G8R8A8_UNORM_SRGB: return DXGI_FORMAT_B8G8R8A8_UNORM;
+            default:                              return fmt;
+        }
+    }
+
+    static void ensure_rtv(ID3D11Device *device,
+                           IDXGISwapChain *swapchain,
+                           ID3D11RenderTargetView **rtv)
+    {
+        if (*rtv || !device || !swapchain) {
+            return;
+        }
+        ID3D11Texture2D *backbuffer = nullptr;
+        if (FAILED(swapchain->GetBuffer(0, IID_PPV_ARGS(&backbuffer))) || !backbuffer) {
+            return;
+        }
+        D3D11_TEXTURE2D_DESC td {};
+        backbuffer->GetDesc(&td);
+        const DXGI_FORMAT view_fmt = to_unorm_view(td.Format);
+        if (view_fmt != td.Format) {
+            D3D11_RENDER_TARGET_VIEW_DESC rtvd {};
+            rtvd.Format = view_fmt;
+            rtvd.ViewDimension = D3D11_RTV_DIMENSION_TEXTURE2D;
+            device->CreateRenderTargetView(backbuffer, &rtvd, rtv);
+        } else {
+            device->CreateRenderTargetView(backbuffer, nullptr, rtv);
+        }
+        backbuffer->Release();
+    }
+
+    // bind the backbuffer (lazily creating the RTV) and draw the imgui
+    // frame on top. reset_invalidate releases *rtv on ResizeBuffers.
+    void render(ID3D11Device *device,
+                ID3D11DeviceContext *context,
+                IDXGISwapChain *swapchain,
+                ID3D11RenderTargetView **rtv)
+    {
+        ensure_rtv(device, swapchain, rtv);
+        if (!*rtv || !context) {
+            return;
+        }
+        // present happens immediately after, so no need to save the previous
+        // RT binding (flip-model resets it anyway).
+        context->OMSetRenderTargets(1, rtv, nullptr);
+        ImGui_ImplDX11_RenderDrawData(ImGui::GetDrawData());
+    }
+
+}
+
+// --------------------------------------------------------------------------
+// file-local state + per-frame helpers
+
+namespace {
+
+using Present_t = HRESULT(STDMETHODCALLTYPE *)(
+        IDXGISwapChain *, UINT, UINT);
+using ResizeBuffers_t = HRESULT(STDMETHODCALLTYPE *)(
+        IDXGISwapChain *, UINT, UINT, UINT, DXGI_FORMAT, UINT);
+using Present1_t = HRESULT(STDMETHODCALLTYPE *)(
+        IDXGISwapChain1 *, UINT, UINT, const DXGI_PRESENT_PARAMETERS *);
+
+Present_t Present_orig = nullptr;
+ResizeBuffers_t ResizeBuffers_orig = nullptr;
+Present1_t Present1_orig = nullptr;
+
+bool g_swapchain_hooked = false;
+bool g_swapchain1_hooked = false;
+
+void try_create_overlay(IDXGISwapChain *swapchain) {
+    if (!swapchain || overlay::OVERLAY) {
+        return;
+    }
+
+    DXGI_SWAP_CHAIN_DESC desc {};
+    if (FAILED(swapchain->GetDesc(&desc)) || !desc.OutputWindow) {
+        return;
+    }
+
+    // only attach to the main game window; ignore sub-screens / IME helpers.
+    HWND main = d3d11_hooks::main_hwnd();
+    if (main && desc.OutputWindow != main) {
+        return;
+    }
+
+    ID3D11Device *device = nullptr;
+    if (FAILED(swapchain->GetDevice(IID_PPV_ARGS(&device))) || !device) {
+        return;
+    }
+    ID3D11DeviceContext *context = nullptr;
+    device->GetImmediateContext(&context);
+
+    if (context) {
+        overlay::create_d3d11(desc.OutputWindow, device, context, swapchain);
+        RECT cr {};
+        ::GetClientRect(desc.OutputWindow, &cr);
+        log_info("graphics::d3d11",
+            "attached overlay to swapchain hwnd=0x{:x} backbuffer={}x{} client={}x{}",
+            (uintptr_t) desc.OutputWindow,
+            desc.BufferDesc.Width, desc.BufferDesc.Height,
+            cr.right - cr.left, cr.bottom - cr.top);
+        context->Release();
+    }
+    device->Release();
+}
+
+// rising-edge screenshot hotkey poll (mirrors d3d9 backend behaviour).
+void poll_screenshot_hotkey() {
+    static bool s_down = false;
+    auto buttons = games::get_buttons_overlay(eamuse_get_game());
+    const bool pressed = buttons
+        && (!overlay::OVERLAY || overlay::OVERLAY->hotkeys_triggered())
+        && GameAPI::Buttons::getState(RI_MGR,
+                buttons->at(games::OverlayButtons::Screenshot));
+    if (pressed && !s_down) {
+        graphics_screenshot_trigger();
+    }
+    s_down = pressed;
+}
+
+void pump_overlay(IDXGISwapChain *swapchain) {
+    if (!overlay::OVERLAY || !overlay::OVERLAY->uses_swapchain(swapchain)) {
+        return;
+    }
+
+    poll_screenshot_hotkey();
+
+    // size imgui to the backbuffer (not window client). dxgi may upscale
+    // a small backbuffer into a larger client rect; without this override
+    // imgui would draw past the RTV and the mouse mapping would be off.
+    DXGI_SWAP_CHAIN_DESC desc {};
+    if (SUCCEEDED(swapchain->GetDesc(&desc))) {
+        ImGui_ImplSpice_SetDisplaySizeOverride(
+            (float) desc.BufferDesc.Width,
+            (float) desc.BufferDesc.Height);
+    }
+
+    overlay::OVERLAY->update();
+    overlay::OVERLAY->new_frame();
+    overlay::OVERLAY->render();
+
+    // after overlay render so toasts/menus end up in the saved image.
+    d3d11_hooks::try_screenshot(swapchain);
+}
+
+// ----------------------------------------------------------------------
+// swapchain method hooks
+
+HRESULT STDMETHODCALLTYPE Present_hook(
+        IDXGISwapChain *swapchain, UINT SyncInterval, UINT Flags)
+{
+    try_create_overlay(swapchain);
+    pump_overlay(swapchain);
+    return Present_orig(swapchain, SyncInterval, Flags);
+}
+
+HRESULT STDMETHODCALLTYPE Present1_hook(
+        IDXGISwapChain1 *swapchain, UINT SyncInterval, UINT Flags,
+        const DXGI_PRESENT_PARAMETERS *pParams)
+{
+    try_create_overlay(swapchain);
+    pump_overlay(swapchain);
+    return Present1_orig(swapchain, SyncInterval, Flags, pParams);
+}
+
+HRESULT STDMETHODCALLTYPE ResizeBuffers_hook(
+        IDXGISwapChain *swapchain, UINT BufferCount, UINT Width, UINT Height,
+        DXGI_FORMAT NewFormat, UINT SwapChainFlags)
+{
+    const bool ours = overlay::OVERLAY && overlay::OVERLAY->uses_swapchain(swapchain);
+    if (ours) {
+        log_info("graphics::d3d11", "ResizeBuffers {}x{} fmt={}",
+            Width, Height, (int32_t) NewFormat);
+        overlay::OVERLAY->reset_invalidate();
+    }
+    HRESULT res = ResizeBuffers_orig(
+        swapchain, BufferCount, Width, Height, NewFormat, SwapChainFlags);
+    if (ours && SUCCEEDED(res)) {
+        overlay::OVERLAY->reset_recreate();
+    }
+    return res;
+}
+
+} // namespace
+
+// --------------------------------------------------------------------------
+// d3d11_hooks public surface: main-window tracking + vtable install.
+
+namespace d3d11_hooks {
+
+namespace {
+    std::atomic<HWND> g_main_hwnd { nullptr };
+    std::atomic<HWND> g_ignored_hwnd { nullptr };
+}
+
+void note_main_hwnd(HWND hwnd) {
+    if (!hwnd || hwnd == g_ignored_hwnd.load()) {
+        return;
+    }
+    HWND expected = nullptr;
+    if (g_main_hwnd.compare_exchange_strong(expected, hwnd)) {
+        log_info("graphics::d3d11", "main hwnd recorded: 0x{:x}",
+            (uintptr_t) hwnd);
+    }
+}
+
+HWND main_hwnd() {
+    return g_main_hwnd.load();
+}
+
+void ignore_hwnd(HWND hwnd) {
+    g_ignored_hwnd.store(hwnd);
+}
+
+// patch IDXGISwapChain::Present + ResizeBuffers and (if implemented)
+// IDXGISwapChain1::Present1. idempotent; flag is set only after success
+// so failed attempts can be retried on the next swapchain.
+void install_swapchain_hooks(IDXGISwapChain *swapchain) {
+    if (!swapchain) {
+        return;
+    }
+    static std::mutex s_hook_mutex;
+    std::lock_guard<std::mutex> lock(s_hook_mutex);
+
+    if (!g_swapchain_hooked) {
+        const bool a = hook_vtbl(swapchain, 8,  (void *) Present_hook,
+                  (void **) &Present_orig,       "IDXGISwapChain::Present");
+        const bool b = hook_vtbl(swapchain, 13, (void *) ResizeBuffers_hook,
+                  (void **) &ResizeBuffers_orig, "IDXGISwapChain::ResizeBuffers");
+        if (a && b) {
+            g_swapchain_hooked = true;
+        }
+    }
+
+    if (!g_swapchain1_hooked) {
+        IDXGISwapChain1 *sc1 = nullptr;
+        if (SUCCEEDED(swapchain->QueryInterface(IID_PPV_ARGS(&sc1))) && sc1) {
+            if (hook_vtbl(sc1, 22, (void *) Present1_hook,
+                          (void **) &Present1_orig, "IDXGISwapChain1::Present1")) {
+                g_swapchain1_hooked = true;
+            }
+            sc1->Release();
+        }
+    }
+}
+
+}
+
+#endif // SPICE_D3D11

src/spice2x/hooks/graphics/backends/d3d11/d3d11_vtable_capture.cpp

@@ -0,0 +1,175 @@
+// proactive vtable capture for the dx11 backend.
+//
+// titles under the execexe loader routinely race past our export-level
+// trampolines, so the game's first real swapchain never goes through us.
+// we sidestep that by creating a throwaway device + swapchain ourselves
+// the moment d3d11.dll + dxgi.dll appear, which patches the shared
+// IDXGISwapChain[1] / IDXGIFactory[2] vtables ahead of the game.
+
+#include "d3d11_backend.h"
+
+#ifdef SPICE_D3D11
+
+#include <atomic>
+#include <memory>
+
+#include <windows.h>
+#include <d3d11.h>
+#include <dxgi.h>
+#include <dxgi1_2.h>
+
+#include "d3d11_internal.h"
+
+using d3d11_hooks::com_ptr;
+
+namespace {
+
+using D3D11CreateDevice_t = HRESULT(WINAPI *)(
+    IDXGIAdapter *, D3D_DRIVER_TYPE, HMODULE, UINT,
+    const D3D_FEATURE_LEVEL *, UINT, UINT,
+    ID3D11Device **, D3D_FEATURE_LEVEL *, ID3D11DeviceContext **);
+using CreateDXGIFactory1_t = HRESULT(WINAPI *)(REFIID, void **);
+using CreateDXGIFactory2_t = HRESULT(WINAPI *)(UINT, REFIID, void **);
+
+std::atomic<bool> g_vtables_captured { false };
+
+template<typename Fn>
+Fn resolve(HMODULE mod, const char *name) {
+    return reinterpret_cast<Fn>(GetProcAddress(mod, name));
+}
+
+com_ptr<IDXGIFactory2> create_factory2(CreateDXGIFactory2_t f2,
+                                       CreateDXGIFactory1_t f1)
+{
+    IDXGIFactory2 *raw = nullptr;
+    if (f2 && SUCCEEDED(f2(0, IID_PPV_ARGS(&raw))) && raw) {
+        return com_ptr<IDXGIFactory2>(raw);
+    }
+    IDXGIFactory1 *factory1 = nullptr;
+    if (f1 && SUCCEEDED(f1(IID_PPV_ARGS(&factory1))) && factory1) {
+        factory1->QueryInterface(IID_PPV_ARGS(&raw));
+        factory1->Release();
+    }
+    return com_ptr<IDXGIFactory2>(raw);
+}
+
+bool create_dummy_device(D3D11CreateDevice_t create,
+                         com_ptr<ID3D11Device> &device,
+                         com_ptr<ID3D11DeviceContext> &context)
+{
+    static constexpr D3D_FEATURE_LEVEL levels[] = {
+        D3D_FEATURE_LEVEL_11_1, D3D_FEATURE_LEVEL_11_0,
+        D3D_FEATURE_LEVEL_10_1, D3D_FEATURE_LEVEL_10_0,
+    };
+    // hardware first, then WARP so headless / unusual configs still work.
+    for (auto type : { D3D_DRIVER_TYPE_HARDWARE, D3D_DRIVER_TYPE_WARP }) {
+        ID3D11Device *d = nullptr;
+        ID3D11DeviceContext *c = nullptr;
+        D3D_FEATURE_LEVEL got;
+        if (SUCCEEDED(create(nullptr, type, nullptr, 0,
+                             levels, ARRAYSIZE(levels), D3D11_SDK_VERSION,
+                             &d, &got, &c)) && d) {
+            device.reset(d);
+            context.reset(c);
+            return true;
+        }
+    }
+    return false;
+}
+
+} // namespace
+
+namespace d3d11_hooks {
+
+// create a throwaway device + swapchain to patch the shared vtables before
+// the game's loader races past our export trampolines. safe to call
+// repeatedly; runs at most once.
+void try_capture_vtables() {
+    if (g_vtables_captured.load()) {
+        return;
+    }
+
+    HMODULE d3d11 = GetModuleHandleW(L"d3d11.dll");
+    HMODULE dxgi = GetModuleHandleW(L"dxgi.dll");
+    if (!d3d11 || !dxgi) {
+        return;
+    }
+
+    auto create_device = resolve<D3D11CreateDevice_t>(d3d11, "D3D11CreateDevice");
+    auto f2 = resolve<CreateDXGIFactory2_t>(dxgi, "CreateDXGIFactory2");
+    auto f1 = resolve<CreateDXGIFactory1_t>(dxgi, "CreateDXGIFactory1");
+    if (!create_device || (!f1 && !f2)) {
+        return;
+    }
+
+    // serialize concurrent calls (poll thread + LDR notification). only
+    // flip g_vtables_captured after success so failed attempts remain
+    // retriable on the next tick.
+    static std::atomic<bool> in_progress { false };
+    if (in_progress.exchange(true)) {
+        return;
+    }
+    struct scope_clear {
+        std::atomic<bool> &flag;
+        ~scope_clear() { flag.store(false); }
+    } clear { in_progress };
+
+    // hidden message-only window; STATIC is always registered by user32.
+    HWND dummy_hwnd = CreateWindowExW(
+        0, L"STATIC", L"", 0, 0, 0, 1, 1,
+        HWND_MESSAGE, nullptr, GetModuleHandleW(nullptr), nullptr);
+    if (!dummy_hwnd) {
+        log_warning("graphics::d3d11",
+            "vtable capture: CreateWindowExW failed (gle={})", (unsigned long)GetLastError());
+        return;
+    }
+    auto destroy_hwnd = std::unique_ptr<HWND__, decltype(&DestroyWindow)>(
+        dummy_hwnd, &DestroyWindow);
+
+    // if the game's CreateDXGIFactory_hook already raced us, our
+    // CreateSwapChainForHwnd call below would trip the hook and try to
+    // record dummy_hwnd as the main window. block that.
+    ignore_hwnd(dummy_hwnd);
+
+    auto factory2 = create_factory2(f2, f1);
+    if (!factory2) {
+        log_warning("graphics::d3d11", "vtable capture: CreateDXGIFactory* failed");
+        return;
+    }
+
+    com_ptr<ID3D11Device>        device;
+    com_ptr<ID3D11DeviceContext> context;
+    if (!create_dummy_device(create_device, device, context)) {
+        log_warning("graphics::d3d11", "vtable capture: D3D11CreateDevice failed");
+        return;
+    }
+
+    DXGI_SWAP_CHAIN_DESC1 desc {};
+    desc.Width = 1;
+    desc.Height = 1;
+    desc.Format = DXGI_FORMAT_B8G8R8A8_UNORM;
+    desc.SampleDesc.Count = 1;
+    desc.BufferUsage = DXGI_USAGE_RENDER_TARGET_OUTPUT;
+    desc.BufferCount = 2;
+    desc.SwapEffect = DXGI_SWAP_EFFECT_FLIP_DISCARD;
+
+    IDXGISwapChain1 *raw_sc = nullptr;
+    HRESULT hr = factory2->CreateSwapChainForHwnd(
+        device.get(), dummy_hwnd, &desc, nullptr, nullptr, &raw_sc);
+    if (FAILED(hr) || !raw_sc) {
+        log_warning("graphics::d3d11",
+            "vtable capture: CreateSwapChainForHwnd failed (hr={:#x})", (unsigned long)hr);
+        return;
+    }
+    com_ptr<IDXGISwapChain1> swapchain(raw_sc);
+
+    install_swapchain_hooks(swapchain.get());
+    install_factory_hooks(factory2.get());
+
+    g_vtables_captured.store(true);
+    log_info("graphics::d3d11", "vtable capture complete (via dummy swapchain)");
+}
+
+}
+
+#endif // SPICE_D3D11

src/spice2x/hooks/graphics/backends/d3d9/d3d9_backend.cpp

@@ -28,8 +28,10 @@
 #include "misc/eamuse.h"
 #include "misc/wintouchemu.h"
 #include "overlay/overlay.h"
+#include "overlay/notifications.h"
 #include "util/detour.h"
 #include "util/deferlog.h"
+#include "util/fileutils.h"
 #include "util/flags_helper.h"
 #include "util/libutils.h"
 #include "util/logging.h"
@@ -465,15 +467,21 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::RegisterSoftwareDevice(void *pIniti
 UINT STDMETHODCALLTYPE WrappedIDirect3D9::GetAdapterCount() {
     UINT result = pReal->GetAdapterCount();
 
-    if (!FAKE_SUBSCREEN_ADAPTER && games::popn::is_pikapika_model() && result == 1) {
-        FAKE_SUBSCREEN_ADAPTER = true;
-        log_info(
-            "graphics::d3d9",
-            "GetAdapterCount returned 1, popn needs 2 adapters - enabliing fake submonitor mode");
-    }
+    if (!FAKE_SUBSCREEN_ADAPTER) {
+        if (games::popn::is_pikapika_model() && result == 1) {
+            FAKE_SUBSCREEN_ADAPTER = true;
+            log_info(
+                "graphics::d3d9",
+                "GetAdapterCount returned {}, popn needs 2 adapters - enabling fake submonitor mode", result);
+            return 2;
 
-    if (FAKE_SUBSCREEN_ADAPTER) {
-        return 2;
+        } else if (games::gitadora::is_arena_model() && !GRAPHICS_WINDOWED && result < 4) {
+            FAKE_SUBSCREEN_ADAPTER = true;
+            log_info(
+                "graphics::d3d9",
+                "GetAdapterCount returned {}, gfdm needs 4 adapters - enabling fake submonitor mode", result);
+            return 4;
+        }
     }
     
     return result;
@@ -485,12 +493,14 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::GetAdapterIdentifier(
         D3DADAPTER_IDENTIFIER9 *pIdentifier)
 {
 
-    if (FAKE_SUBSCREEN_ADAPTER && Adapter == 1 && pIdentifier) {
+    if (FAKE_SUBSCREEN_ADAPTER && Adapter > 0 && pIdentifier) {
         *pIdentifier = {};
-        strcpy(pIdentifier->DeviceName, "\\\\.\\DISPLAY_SPICE_FAKE"); 
+        const std::string adapter_name = fmt::format("\\\\.\\DISPLAY_SPICE_FAKE_{}", Adapter);
+        strcpy(pIdentifier->DeviceName, adapter_name.c_str()); 
         log_misc(
             "graphics::d3d9",
-            "GetAdapterIdentifier called for fake subscreen adapter 1: {}",
+            "GetAdapterIdentifier called for fake subscreen adapter {}: {}",
+            Adapter,
             pIdentifier->DeviceName);
         return S_OK;
     }
@@ -499,8 +509,8 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::GetAdapterIdentifier(
 }
 
 UINT STDMETHODCALLTYPE WrappedIDirect3D9::GetAdapterModeCount(UINT Adapter, D3DFORMAT Format) {
-    if (FAKE_SUBSCREEN_ADAPTER && Adapter == 1) {
-        log_misc("graphics::d3d9", "GetAdapterModeCount called for fake subscreen adapter");
+    if (FAKE_SUBSCREEN_ADAPTER && Adapter > 0) {
+        log_misc("graphics::d3d9", "GetAdapterModeCount called for fake subscreen adapter {}", Adapter);
         return 1;
     }
 
@@ -513,8 +523,8 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::EnumAdapterModes(
         UINT Mode,
         D3DDISPLAYMODE *pMode)
 {
-    if (FAKE_SUBSCREEN_ADAPTER && Adapter == 1) {
-        log_misc("graphics::d3d9", "EnumAdapterModes called for fake subscreen adapter");
+    if (FAKE_SUBSCREEN_ADAPTER && Adapter > 0) {
+        log_misc("graphics::d3d9", "EnumAdapterModes called for fake subscreen adapter {}", Adapter);
         if (Mode == 0 && pMode) {
             pMode->Width = 1280;
             pMode->Height = 800;
@@ -622,8 +632,8 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::CheckDeviceFormat(
         D3DRESOURCETYPE RType,
         D3DFORMAT CheckFormat)
 {
-    if (FAKE_SUBSCREEN_ADAPTER && Adapter == 1) {
-        log_misc("graphics::d3d9", "CheckDeviceFormat called for fake subscreen adapter");
+    if (FAKE_SUBSCREEN_ADAPTER && Adapter > 0) {
+        log_misc("graphics::d3d9", "CheckDeviceFormat called for fake subscreen adapter {}", Adapter);
         return D3D_OK;
     }
 
@@ -708,7 +718,7 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::GetDeviceCaps(UINT Adapter, D3DDEVT
         }
         // in windowed mode, LDJ will always launch two windows, no special handling needed here
     } else if (avs::game::is_model("KFC")) {
-        if (GRAPHICS_WINDOWED && GRAPHICS_PREVENT_SECONDARY_WINDOW) {
+        if (GRAPHICS_WINDOWED && GRAPHICS_PREVENT_SECONDARY_WINDOWS) {
             // user wants windowed mode but does not want subscreen at all
             pCaps->NumberOfAdaptersInGroup = 1;
         } else {
@@ -1143,10 +1153,15 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3D9::CreateDeviceEx(
 
         *ppReturnedDeviceInterface = new WrappedIDirect3DDevice9(hFocusWindow, *ppReturnedDeviceInterface);
 
-        // initialize sub screen if IIDX/SDVX requested a multi-head context
-        if (avs::game::is_model({"LDJ", "KFC", "M39"}) &&
+        // initialize sub screen if the game requested a multi-head context
+        if (avs::game::is_model({"LDJ", "KFC", "M39", "M32"}) &&
             (orig_behavior_flags & D3DCREATE_ADAPTERGROUP_DEVICE)) {
-            graphics_d3d9_ldj_init_sub_screen(*ppReturnedDeviceInterface, &pPresentationParameters[1]);
+
+            UINT i = 1;
+            if (games::gitadora::is_arena_model()) {
+                i = 2;
+            }
+            graphics_d3d9_ldj_init_sub_screen(*ppReturnedDeviceInterface, &pPresentationParameters[i]);
         }
     }
 
@@ -1196,7 +1211,13 @@ static void graphics_d3d9_ldj_init_sub_screen(IDirect3DDevice9Ex *device, D3DPRE
             log_info("graphics::d3d9", "created additional swap chain for windowed mode");
         }
     } else {
-        hr = device->GetSwapChain(1, &SUB_SWAP_CHAIN);
+
+        int swapchain = 1;
+        if (games::gitadora::is_arena_model()) {
+            swapchain = 2;
+        }
+
+        hr = device->GetSwapChain(swapchain, &SUB_SWAP_CHAIN);
         if (FAILED(hr)) {
             log_warning("graphics::d3d9", "failed to acquire fullscreen sub swap chain, hr={}", FMT_HRESULT(hr));
         } else {
@@ -1233,7 +1254,7 @@ static void graphics_d3d9_ldj_on_present(IDirect3DDevice9 *wrapped_device) {
     // iidx/sdvx
     int swapchain = 1;
     if (games::gitadora::is_arena_model()) {
-        swapchain = 3;
+        swapchain = 2;
     }
 
     if (!ATTEMPTED_SUB_SWAP_CHAIN_ACQUIRE && SUB_SWAP_CHAIN == nullptr) {
@@ -1415,11 +1436,18 @@ static void save_screenshot(const std::string &file_path, UINT height, IDirect3D
 
         if (FAILED(hr)) {
             log_warning("graphics::d3d9", "Failed to save screenshot");
+            overlay::notifications::add(
+                overlay::notifications::Severity::Error,
+                "Screenshot failed to save");
             return;
         }
 
         // save to clipboard
         clipboard::copy_image(file_path);
+
+        overlay::notifications::add(
+            overlay::notifications::Severity::Success,
+            fmt::format("Screenshot saved: {}", fileutils::basename(file_path)));
     } else {
         log_warning("graphics::d3d9", "Direct3D save helper function not available");
     }
@@ -1430,6 +1458,15 @@ void graphics_d3d9_on_present(
         IDirect3DDevice9 *device,
         IDirect3DDevice9 *wrapped_device) {
 
+    // image resize / orientation swap. run here (the present path) rather than from `EndScene`,
+    // which may fire several times per frame on multi-pass / render-to-texture games. this is the
+    // single point guaranteed to be after the game's last `EndScene` and before the real `Present`,
+    // so the back buffer is fully drawn and the expensive StretchRect work happens exactly once per
+    // frame. it must run before the overlay is rendered so the overlay isn't scaled with the image.
+    if (cfg::SCREENRESIZE->enable_screen_resize || GRAPHICS_FS_ORIENTATION_SWAP) {
+        SurfaceHook(device);
+    }
+
     // Do overlay init as many d3d9 hooks create a dummy instance to get vtable offsets and never
     // call `Present`. This avoids race conditions on `IDirect3D9::CreateDevice` like with
     // `dx9osd.dll` for pfreepanic.
@@ -1452,8 +1489,7 @@ void graphics_d3d9_on_present(
     // for IIDX TDJ / SDVX UFC, handle subscreen
     const bool is_vm = games::sdvx::is_valkyrie_model();
     const bool is_tdj = avs::game::is_model("LDJ") && games::iidx::TDJ_MODE;
-    const bool is_gfdm_arena =
-        games::gitadora::is_arena_model() && games::gitadora::ARENA_SINGLE_WINDOW;
+    const bool is_gfdm_arena = games::gitadora::is_arena_model() && GRAPHICS_PREVENT_SECONDARY_WINDOWS;
     const bool is_pika = games::popn::is_pikapika_model();
     if (is_vm || is_tdj || is_gfdm_arena || is_pika) {
         graphics_d3d9_ldj_on_present(wrapped_device);
@@ -1499,6 +1535,9 @@ void graphics_d3d9_on_present(
             log_warning("graphics::d3d9",
                     "failed to get back buffer, hr={}",
                     FMT_HRESULT(hr));
+            if (capture) {
+                graphics_capture_skip(capture_screen);
+            }
             return;
         }
 
@@ -1509,6 +1548,9 @@ void graphics_d3d9_on_present(
                     "failed to acquire back buffer descriptor, hr={}",
                     FMT_HRESULT(hr));
             buffer->Release();
+            if (capture) {
+                graphics_capture_skip(capture_screen);
+            }
             return;
         }
 
@@ -1522,6 +1564,9 @@ void graphics_d3d9_on_present(
                     "failed to acquire temporary surface, hr={}",
                     FMT_HRESULT(hr));
             buffer->Release();
+            if (capture) {
+                graphics_capture_skip(capture_screen);
+            }
             return;
         }
 
@@ -1532,6 +1577,9 @@ void graphics_d3d9_on_present(
                     FMT_HRESULT(hr));
             temp_surface->Release();
             buffer->Release();
+            if (capture) {
+                graphics_capture_skip(capture_screen);
+            }
             return;
         }
 

src/spice2x/hooks/graphics/backends/d3d9/d3d9_device.cpp

@@ -129,13 +129,17 @@ ULONG STDMETHODCALLTYPE WrappedIDirect3DDevice9::Release() {
             this->main_swapchain->Release();
             this->main_swapchain = nullptr;
         }
-        if (this->sub_swapchain) {
-            this->sub_swapchain->Release();
-            this->sub_swapchain = nullptr;
+        for (auto &sc : this->sub_swapchain) {
+            if (sc) {
+                sc->Release();
+                sc = nullptr;
+            }
         }
-        if (this->fake_sub_swapchain) {
-            this->fake_sub_swapchain->Release();
-            this->fake_sub_swapchain = nullptr;
+        for (auto &sc : this->fake_sub_swapchain) {
+            if (sc) {
+                sc->Release();
+                sc = nullptr;
+            }
         }
 
         if (overlay::ENABLED) {
@@ -249,28 +253,69 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3DDevice9::CreateAdditionalSwapChain(
 {
     WRAP_VERBOSE;
 
-    HRESULT hr = pReal->CreateAdditionalSwapChain(pPresentationParameters, ppSwapChain);
+    int index = 0;
     bool create_swap_chain = false;
+    bool create_fake_swap_chain = false;
     if (avs::game::is_model({"LDJ", "KFC", "M39"})) {
         create_swap_chain = true;
+
     } else if (games::gitadora::is_arena_model() &&
-        games::gitadora::ARENA_SINGLE_WINDOW &&
-        pPresentationParameters->BackBufferWidth == 800) {
-        create_swap_chain = true;
+        (GRAPHICS_PREVENT_SECONDARY_WINDOWS || GRAPHICS_GITADORA_HIDE_SIDE_WINDOWS)) {
+
+        if (pPresentationParameters->BackBufferWidth == 800) {
+            // SMALL (subscreen)
+            create_swap_chain = true;
+            index = 0;
+        } else if (pPresentationParameters->BackBufferWidth == 1080) {
+            // LEFT/RIGHT
+            create_swap_chain = true;
+            index = 1;
+            if (sub_swapchain[index] || fake_sub_swapchain[index]) {
+                index = 2;
+            }
+            create_fake_swap_chain = GRAPHICS_GITADORA_HIDE_SIDE_WINDOWS &&
+                !GRAPHICS_PREVENT_SECONDARY_WINDOWS;
+        } else {
+            log_warning("graphics::d3d9", "unknown swap chain detected in CreateAdditionalSwapChain");
+        }
     }
 
+    if (create_fake_swap_chain) {
+        if (!fake_sub_swapchain[index]) {
+            log_info(
+                "graphics::d3d9",
+                "CreateAdditionalSwapChain called for hidden GITADORA side swap chain {}, "
+                "using fake swap chain",
+                index);
+
+            fake_sub_swapchain[index] =
+                new FakeIDirect3DSwapChain9(this, pPresentationParameters, false);
+        }
+        fake_sub_swapchain[index]->AddRef();
+        *ppSwapChain = static_cast<IDirect3DSwapChain9 *>(fake_sub_swapchain[index]);
+
+        return D3D_OK;
+    }
+
+    HRESULT hr = pReal->CreateAdditionalSwapChain(pPresentationParameters, ppSwapChain);
+
     if (create_swap_chain) {
-        if (SUCCEEDED(hr) && !sub_swapchain) {
-            sub_swapchain = new WrappedIDirect3DSwapChain9(this, *ppSwapChain);
-            sub_swapchain->should_run_hooks = false;
-        } else if (FAILED(hr) && !fake_sub_swapchain) {
+        log_misc(
+            "graphics::d3d9",
+            "CreateAdditionalSwapChain called for swap chain {}, creating swap chain",
+            index);
+
+        if (SUCCEEDED(hr) && !sub_swapchain[index]) {
+            sub_swapchain[index] = new WrappedIDirect3DSwapChain9(this, *ppSwapChain);
+            sub_swapchain[index]->should_run_hooks = false;
+        } else if (FAILED(hr) && !fake_sub_swapchain[index]) {
             log_warning("graphics::d3d9",
                     "failed to create sub swap chain, hr={}, using fake swap chain",
                     FMT_HRESULT(hr));
 
-            fake_sub_swapchain = new FakeIDirect3DSwapChain9(this, pPresentationParameters, false);
-            fake_sub_swapchain->AddRef();
-            *ppSwapChain = static_cast<IDirect3DSwapChain9 *>(fake_sub_swapchain);
+            fake_sub_swapchain[index] = new FakeIDirect3DSwapChain9(this, pPresentationParameters, false);
+            fake_sub_swapchain[index]->AddRef();
+            *ppSwapChain = static_cast<IDirect3DSwapChain9 *>(fake_sub_swapchain[index]);
 
             return D3D_OK;
         }
@@ -308,22 +353,20 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3DDevice9::GetSwapChain(
     if (iSwapChain == 1 && avs::game::is_model({"LDJ", "KFC", "M39"})) {
         swap = true;
     }
-    if (games::gitadora::is_arena_model() &&
-        games::gitadora::ARENA_SINGLE_WINDOW &&
-        iSwapChain == 3) {
+    if (games::gitadora::is_arena_model() && iSwapChain == 2) {
         swap = true;
     }
 
     if (swap) {
-        if (sub_swapchain) {
-            sub_swapchain->AddRef();
-            *ppSwapChain = static_cast<IDirect3DSwapChain9 *>(sub_swapchain);
+        if (sub_swapchain[0]) {
+            sub_swapchain[0]->AddRef();
+            *ppSwapChain = static_cast<IDirect3DSwapChain9 *>(sub_swapchain[0]);
 
             graphics_screens_register(iSwapChain);
             return D3D_OK;
-        } else if (fake_sub_swapchain) {
-            fake_sub_swapchain->AddRef();
-            *ppSwapChain = static_cast<IDirect3DSwapChain9 *>(fake_sub_swapchain);
+        } else if (fake_sub_swapchain[0]) {
+            fake_sub_swapchain[0]->AddRef();
+            *ppSwapChain = static_cast<IDirect3DSwapChain9 *>(fake_sub_swapchain[0]);
 
             graphics_screens_register(iSwapChain);
             return D3D_OK;
@@ -344,7 +387,7 @@ UINT STDMETHODCALLTYPE WrappedIDirect3DDevice9::GetNumberOfSwapChains() {
 
     UINT n = pReal->GetNumberOfSwapChains();
 
-    if (sub_swapchain && avs::game::is_model({"LDJ", "KFC", "M39"})) {
+    if (sub_swapchain[0] && avs::game::is_model({"LDJ", "KFC", "M39"})) {
         n += 1;
     }
 
@@ -717,18 +760,15 @@ HRESULT STDMETHODCALLTYPE WrappedIDirect3DDevice9::BeginScene() {
 
 static IDirect3DSurface9 *topSurface = nullptr;
 static IDirect3DSurface9 *backbuffer = nullptr;
-static LPDIRECT3DSWAPCHAIN9 mSwapChain = nullptr;
-static IDirect3DTexture9* tex;
+static IDirect3DTexture9 *tex = nullptr;
 
 static UINT topSurface_width = 0;
 static UINT topSurface_height = 0;
 static float topSurface_aspect_ratio = 1.f;
+static UINT backbuffer_width = 0;
+static UINT backbuffer_height = 0;
 
 void SurfaceHook(IDirect3DDevice9 *pReal) {
-    D3DPRESENT_PARAMETERS param {};
-
-    pReal->GetSwapChain(0, &mSwapChain);
-    mSwapChain->GetPresentParameters(&param);
 
     // phase 0 - create a surface that has the same aspect ratio as the original back buffer
     //           but larger in size. this is needed to ensure that when the user zooms out, we have
@@ -736,8 +776,20 @@ void SurfaceHook(IDirect3DDevice9 *pReal) {
     //
     //           (only done once)
     if (!topSurface) {
-        topSurface_width = param.BackBufferWidth * 3;
-        topSurface_height = param.BackBufferHeight * 3;
+        // GetSwapChain/GetBackBuffer AddRef their out parameters; the swap chain is only
+        // needed here (during one-time init), so release it after use to avoid a per-frame
+        // ref leak. The backbuffer ref is intentionally retained for the lifetime of the
+        // cached resources.
+        LPDIRECT3DSWAPCHAIN9 swapChain = nullptr;
+        D3DPRESENT_PARAMETERS param {};
+
+        pReal->GetSwapChain(0, &swapChain);
+        swapChain->GetPresentParameters(&param);
+
+        backbuffer_width = param.BackBufferWidth;
+        backbuffer_height = param.BackBufferHeight;
+        topSurface_width = backbuffer_width * 3;
+        topSurface_height = backbuffer_height * 3;
         topSurface_aspect_ratio = (float)topSurface_width / (float)topSurface_height;
         if (pReal->CreateTexture(topSurface_width, topSurface_height, 1,
             D3DUSAGE_RENDERTARGET, D3DFMT_A8R8G8B8,
@@ -750,16 +802,18 @@ void SurfaceHook(IDirect3DDevice9 *pReal) {
             param.BackBufferWidth, param.BackBufferHeight, param.BackBufferCount);
 
         tex->GetSurfaceLevel(0, &topSurface);
-        if (mSwapChain->GetBackBuffer(0, D3DBACKBUFFER_TYPE_MONO, &backbuffer) != D3D_OK) {
+        if (swapChain->GetBackBuffer(0, D3DBACKBUFFER_TYPE_MONO, &backbuffer) != D3D_OK) {
             log_warning("graphics::d3d9", "SurfaceHook - GetBackBuffer failed");
         }
+
+        swapChain->Release();
     }
 
     // pre-calculate dimensions used for phase 1
-    const int rectLeft = param.BackBufferWidth;
-    const int rectTop = param.BackBufferHeight;
-    const int w = param.BackBufferWidth;
-    const int h = param.BackBufferHeight;
+    const int rectLeft = backbuffer_width;
+    const int rectTop = backbuffer_height;
+    const int w = backbuffer_width;
+    const int h = backbuffer_height;
 
     // this code used to clear the surface using ColorFill on every call, but
     // this turned out to be very expensive, leading to major frame drops in
@@ -769,9 +823,7 @@ void SurfaceHook(IDirect3DDevice9 *pReal) {
         cfg::SCREENRESIZE->need_surface_clean = false;
     }
 
-    D3DLOCKED_RECT rect;
     HRESULT hr;
-    topSurface->LockRect(&rect, NULL, D3DLOCK_DONOTWAIT);
 
     // phase 1 - copy the original back buffer onto the new surface.
     //           we draw it 1:1 in the center of the surface.
@@ -834,7 +886,6 @@ void SurfaceHook(IDirect3DDevice9 *pReal) {
                 originRect2.left, originRect2.top, originRect2.right, originRect2.bottom);
         }
     }
-    topSurface->UnlockRect();
 
     // phase 2 - viewport calculation - do the actual zoom / offset math
     //           figure out what region of the surface to copy back to the back buffer.
@@ -909,7 +960,6 @@ void SurfaceHook(IDirect3DDevice9 *pReal) {
     //     targetRect.left, targetRect.top, targetRect.right, targetRect.bottom);
 
     // phase 3 - draw the surface to back buffer
-    backbuffer->LockRect(&rect, NULL, D3DLOCK_DONOTWAIT);
     bool use_linear_filter = true;
     if (cfg::SCREENRESIZE->enable_screen_resize) {
         use_linear_filter = cfg::SCREENRESIZE->enable_linear_filter;
@@ -918,7 +968,6 @@ void SurfaceHook(IDirect3DDevice9 *pReal) {
         topSurface, &targetRect,
         backbuffer, nullptr,
         use_linear_filter ? D3DTEXF_LINEAR : D3DTEXF_NONE);
-    backbuffer->UnlockRect();
     if (hr != D3D_OK) {
         log_warning(
             "graphics::d3d9",
@@ -930,10 +979,6 @@ void SurfaceHook(IDirect3DDevice9 *pReal) {
 HRESULT STDMETHODCALLTYPE WrappedIDirect3DDevice9::EndScene() {
     WRAP_DEBUG;
 
-    if (cfg::SCREENRESIZE->enable_screen_resize || GRAPHICS_FS_ORIENTATION_SWAP) {
-        SurfaceHook(pReal);
-    }
-
     static std::once_flag printed;
     std::call_once(printed, []() {
         log_misc("graphics::d3d9", "WrappedIDirect3DDevice9::EndScene");

src/spice2x/hooks/graphics/backends/d3d9/d3d9_device.h

@@ -35,6 +35,10 @@ static const GUID IID_WrappedIDirect3DDevice9 = {
     0x6dec0d40, 0x1339, 0x4bda, { 0xa5, 0xf2, 0x22, 0x31, 0xd4, 0x1, 0xf, 0xd1 }
 };
 
+// applies the image resize / orientation swap by copying the back buffer through an
+// oversized intermediate surface and back. expects the real (unwrapped) device.
+void SurfaceHook(IDirect3DDevice9 *pReal);
+
 struct WrappedIDirect3DDevice9 : IDirect3DDevice9Ex {
     explicit WrappedIDirect3DDevice9(HWND hFocusWindow, IDirect3DDevice9 *orig)
         : hFocusWindow(hFocusWindow), pReal(orig), is_d3d9ex(false) {
@@ -204,7 +208,8 @@ struct WrappedIDirect3DDevice9 : IDirect3DDevice9Ex {
     std::atomic_ulong refs = 1;
 
     WrappedIDirect3DSwapChain9 *main_swapchain = nullptr;
-    WrappedIDirect3DSwapChain9 *sub_swapchain = nullptr;
-    FakeIDirect3DSwapChain9 *fake_sub_swapchain = nullptr;
+    WrappedIDirect3DSwapChain9 *sub_swapchain[3] = { nullptr, nullptr, nullptr };
+    FakeIDirect3DSwapChain9 *fake_sub_swapchain[3] = { nullptr, nullptr, nullptr };
+
     IDirect3DVertexShader9 *vertex_shader = nullptr;
 };

src/spice2x/hooks/graphics/graphics.cpp

@@ -5,6 +5,7 @@
 
 #include "graphics.h"
 
+#include <chrono>
 #include <set>
 #include <vector>
 #include <mutex>
@@ -18,6 +19,7 @@
 #include "games/iidx/iidx.h"
 #include "games/popn/popn.h"
 #include "hooks/graphics/backends/d3d9/d3d9_backend.h"
+#include "hooks/graphics/backends/d3d11/d3d11_backend.h"
 #include "launcher/shutdown.h"
 #include "overlay/overlay.h"
 #include "touch/touch.h"
@@ -39,6 +41,8 @@ struct CaptureData {
 HWND TDJ_SUBSCREEN_WINDOW = nullptr;
 HWND SDVX_SUBSCREEN_WINDOW = nullptr;
 HWND GFDM_SUBSCREEN_WINDOW = nullptr;
+static HWND GFDM_LEFT_WINDOW = nullptr;
+static HWND GFDM_RIGHT_WINDOW = nullptr;
 HWND POPN_SUBSCREEN_WINDOW = nullptr;
 bool FAKE_SUBSCREEN_ADAPTER = false;
 
@@ -58,9 +62,23 @@ static std::mutex GRAPHICS_CAPTURE_SCREENS_M {};
 static CaptureData GRAPHICS_CAPTURE_BUFFER[GRAPHICS_CAPTURE_SCREEN_NO] {};
 static std::mutex GRAPHICS_CAPTURE_BUFFER_M[GRAPHICS_CAPTURE_SCREEN_NO] {};
 static std::condition_variable GRAPHICS_CAPTURE_CV[GRAPHICS_CAPTURE_SCREEN_NO] {};
+static bool GRAPHICS_CAPTURE_SKIP_SIGNAL[GRAPHICS_CAPTURE_SCREEN_NO] {};
+static constexpr std::chrono::milliseconds GRAPHICS_CAPTURE_RECEIVE_TIMEOUT {2000};
+
+static void graphics_capture_cancel_pending(int screen) {
+    std::lock_guard<std::mutex> lock(GRAPHICS_CAPTURE_SCREENS_M);
+
+    for (auto it = GRAPHICS_CAPTURE_SCREENS.rbegin(); it != GRAPHICS_CAPTURE_SCREENS.rend(); ++it) {
+        if (*it == screen) {
+            GRAPHICS_CAPTURE_SCREENS.erase(std::next(it).base());
+            return;
+        }
+    }
+}
 
 static std::optional<graphics_orientation> target_orientation_on_boot;
 static UINT target_refresh_rate_on_boot = 0;
+static std::optional<std::pair<uint32_t, uint32_t>> target_resolution_on_boot;
 static bool monitor_settings_changed = false;
 static bool monitor_layout_needs_reset = false;
 
@@ -75,7 +93,8 @@ UINT GRAPHICS_FORCE_REFRESH = 0;
 std::optional<uint32_t> GRAPHICS_FORCE_REFRESH_SUB;
 std::optional<int> GRAPHICS_FORCE_VSYNC_BUFFER;
 bool GRAPHICS_FORCE_SINGLE_ADAPTER = false;
-bool GRAPHICS_PREVENT_SECONDARY_WINDOW = false;
+bool GRAPHICS_PREVENT_SECONDARY_WINDOWS = false;
+bool GRAPHICS_GITADORA_HIDE_SIDE_WINDOWS = false;
 graphics_dx9on12_state GRAPHICS_9_ON_12_STATE = DX9ON12_AUTO;
 bool GRAPHICS_9_ON_12_REQUESTED_BY_GAME = false;
 bool SUBSCREEN_FORCE_REDRAW = false;
@@ -120,6 +139,128 @@ static void reset_window_hook(HWND hWnd) {
     }
 }
 
+static std::string gitadora_canonical_window_name(const std::string &window_name) {
+    if (window_name == "GITADORA") {
+        return "GITADORA";
+    }
+    if (window_name.ends_with("LEFT")) {
+        return "LEFT";
+    }
+    if (window_name.ends_with("RIGHT")) {
+        return "RIGHT";
+    }
+    if (window_name.ends_with("SMALL")) {
+        return "SMALL";
+    }
+    return "";
+}
+
+static const char *gitadora_window_name_for_hwnd(HWND hWnd) {
+    if (hWnd == nullptr) {
+        return nullptr;
+    }
+    if (GRAPHICS_HOOKED_WINDOW.has_value() && hWnd == GRAPHICS_HOOKED_WINDOW.value()) {
+        return "GITADORA";
+    }
+    if (hWnd == GFDM_LEFT_WINDOW) {
+        return "LEFT";
+    }
+    if (hWnd == GFDM_RIGHT_WINDOW) {
+        return "RIGHT";
+    }
+    if (hWnd == GFDM_SUBSCREEN_WINDOW) {
+        return "SMALL";
+    }
+    return nullptr;
+}
+
+static bool is_gfdm_known_window(HWND hWnd) {
+    return gitadora_window_name_for_hwnd(hWnd) != nullptr;
+}
+
+static bool gitadora_should_block_game_window_placement(HWND hWnd) {
+    if (!GRAPHICS_WINDOWED || !games::gitadora::is_arena_model()) {
+        return false;
+    }
+
+    const auto window_name = gitadora_window_name_for_hwnd(hWnd);
+    return window_name != nullptr && graphics_gitadora_has_window_monitor(window_name);
+}
+
+static void gitadora_remember_window(HWND hWnd, const std::string &window_name) {
+    if (window_name == "LEFT") {
+        GFDM_LEFT_WINDOW = hWnd;
+    } else if (window_name == "RIGHT") {
+        GFDM_RIGHT_WINDOW = hWnd;
+    } else if (window_name == "SMALL") {
+        GFDM_SUBSCREEN_WINDOW = hWnd;
+    }
+}
+
+static bool gitadora_should_allow_small_resize() {
+    return GRAPHICS_WINDOWED &&
+        games::gitadora::is_arena_model() &&
+        !graphics_gitadora_is_borderless_windowed();
+}
+
+static void gitadora_apply_small_resize_style(DWORD &style) {
+    if (!gitadora_should_allow_small_resize()) {
+        return;
+    }
+
+    style |= WS_SIZEBOX;
+    style |= WS_MAXIMIZEBOX;
+    style |= WS_SYSMENU;
+}
+
+static void gitadora_force_window_style(HWND hWnd) {
+    if (!GRAPHICS_WINDOWED || !games::gitadora::is_arena_model() || hWnd == nullptr) {
+        return;
+    }
+
+    DWORD style = GetWindowLongA(hWnd, GWL_STYLE);
+    DWORD style_ex = GetWindowLongA(hWnd, GWL_EXSTYLE);
+    const DWORD style_orig = style;
+    const DWORD style_ex_orig = style_ex;
+    graphics_gitadora_apply_window_style(style, style_ex);
+    if (hWnd == GFDM_SUBSCREEN_WINDOW) {
+        gitadora_apply_small_resize_style(style);
+    }
+
+    if (style == style_orig && style_ex == style_ex_orig) {
+        return;
+    }
+
+    if (SetWindowLongA_orig != nullptr) {
+        SetWindowLongA_orig(hWnd, GWL_STYLE, static_cast<LONG>(style));
+        SetWindowLongA_orig(hWnd, GWL_EXSTYLE, static_cast<LONG>(style_ex));
+    } else {
+        SetWindowLongA(hWnd, GWL_STYLE, static_cast<LONG>(style));
+        SetWindowLongA(hWnd, GWL_EXSTYLE, static_cast<LONG>(style_ex));
+    }
+
+    const UINT flags =
+        SWP_NOMOVE |
+        SWP_NOSIZE |
+        SWP_NOZORDER |
+        SWP_NOACTIVATE |
+        SWP_FRAMECHANGED;
+    if (SetWindowPos_orig != nullptr) {
+        SetWindowPos_orig(hWnd, nullptr, 0, 0, 0, 0, flags);
+    } else {
+        SetWindowPos(hWnd, nullptr, 0, 0, 0, 0, flags);
+    }
+
+    log_misc(
+        "graphics",
+        "GITADORA window style override: hwnd={}, style 0x{:x}->0x{:x}, ex 0x{:x}->0x{:x}",
+        fmt::ptr(hWnd),
+        style_orig,
+        style,
+        style_ex_orig,
+        style_ex);
+}
+
 // window procedure
 static LRESULT CALLBACK WindowProc(HWND hWnd, UINT uMsg, WPARAM wParam, LPARAM lParam) {
 
@@ -163,6 +304,7 @@ static LRESULT CALLBACK WindowProc(HWND hWnd, UINT uMsg, WPARAM wParam, LPARAM l
     if (wintouchemu::INJECT_MOUSE_AS_WM_TOUCH) {
         // drop mouse inputs since only wintouches should be used
         switch (uMsg) {
+            case WM_MOUSEMOVE:
             case WM_LBUTTONDOWN:
             case WM_LBUTTONUP:
             case WM_MBUTTONDOWN:
@@ -280,6 +422,18 @@ static LRESULT CALLBACK WsubWindowProc(HWND hWnd, UINT uMsg, WPARAM wParam, LPAR
         log_misc("graphics", "ignore WM_CLOSE for subscreen window");
         return false;
     }
+
+    if (hWnd == GFDM_SUBSCREEN_WINDOW && (uMsg == WM_MOVE || uMsg == WM_SIZE)) {
+        update_spicetouch_window_dimensions(hWnd);
+        if (SPICETOUCH_TOUCH_HWND) {
+            SetWindowPos(
+                SPICETOUCH_TOUCH_HWND, HWND_TOP,
+                SPICETOUCH_TOUCH_X, SPICETOUCH_TOUCH_Y,
+                SPICETOUCH_TOUCH_WIDTH, SPICETOUCH_TOUCH_HEIGHT,
+                SWP_NOZORDER | SWP_NOREDRAW | SWP_NOREPOSITION | SWP_NOACTIVATE);
+        }
+    }
+
     return CallWindowProcA(WSUB_WNDPROC_ORIG, hWnd, uMsg, wParam, lParam);
 }
 
@@ -343,28 +497,55 @@ static HWND WINAPI CreateWindowExA_hook(DWORD dwExStyle, LPCSTR lpClassName, LPC
             fmt::ptr(lpParam));
 
     // gfdm
+    std::string effective_window_name = window_name;
     if (avs::game::is_model({"J32", "J33", "K32", "K33", "L32", "L33", "M32"})) {
         // set window name
         if (!lpWindowName) {
             lpWindowName = "GITADORA";
+            effective_window_name = "GITADORA";
         }
     }
 
     bool is_tdj_sub_window = avs::game::is_model("LDJ") && window_name.ends_with(" sub");
     bool is_sdvx_sub_window = avs::game::is_model("KFC") && window_name.ends_with(" Sub Screen");
     bool is_popn_sub_window = avs::game::is_model("M39") && window_name.ends_with("Sub Screen");
-    bool is_gfdm_sub_window = games::gitadora::is_arena_model() && window_name.ends_with("SMALL");
+    const std::string gfdm_window_name = games::gitadora::is_arena_model()
+        ? gitadora_canonical_window_name(effective_window_name)
+        : "";
+    const bool is_gfdm_window = !gfdm_window_name.empty();
+    const bool is_gfdm_sub_window = gfdm_window_name == "SMALL";
+    const bool allow_gfdm_small_resize =
+        is_gfdm_sub_window && gitadora_should_allow_small_resize();
 
-    // hide maximize button (prevent misaligned touches)
-    if ((is_tdj_sub_window && GRAPHICS_IIDX_WSUB) || is_sdvx_sub_window || is_gfdm_sub_window || is_popn_sub_window) {
-        dwStyle &= ~(WS_MAXIMIZEBOX);
+    // update style / ex-style
+    if (is_tdj_sub_window || is_sdvx_sub_window || is_gfdm_sub_window || is_popn_sub_window) {
+        // hide maximize button (prevent misaligned touches)
+        if (!allow_gfdm_small_resize) {
+            dwStyle &= ~(WS_MAXIMIZEBOX);
+        }
+
+        // mouse clicks become misaligned when resized
+        if (!allow_gfdm_small_resize) {
+            dwStyle &= ~(WS_SIZEBOX);
+        }
+
+        // borderless
+        if (GRAPHICS_WINDOWED && GRAPHICS_WSUB_BORDERLESS) {
+            dwStyle &= ~(WS_OVERLAPPEDWINDOW);
+        }
+
+        // don't show the sub window on task bar / alt-tab targets
+        if (!GRAPHICS_WINDOWED) {
+            dwExStyle &= ~(WS_EX_APPWINDOW);
+            dwExStyle |= WS_EX_TOOLWINDOW;
+        }
     }
-    // mouse clicks become misaligned when resized
-    if (is_gfdm_sub_window) {
-        dwStyle &= ~(WS_SIZEBOX);
+
+    if (allow_gfdm_small_resize) {
+        gitadora_apply_small_resize_style(dwStyle);
     }
-    if ((is_tdj_sub_window || is_sdvx_sub_window || is_popn_sub_window) && GRAPHICS_WSUB_BORDERLESS) {
-        dwStyle &= ~(WS_OVERLAPPEDWINDOW);
+    if (is_gfdm_window) {
+        graphics_gitadora_apply_window_style(dwStyle, dwExStyle);
     }
 
     if (is_sdvx_sub_window) {
@@ -385,6 +566,16 @@ static HWND WINAPI CreateWindowExA_hook(DWORD dwExStyle, LPCSTR lpClassName, LPC
         }
     }
 
+    if (is_gfdm_window) {
+        graphics_gitadora_apply_window_monitor(
+            gfdm_window_name,
+            x,
+            y,
+            nWidth,
+            nHeight,
+            true);
+    }
+
     if (GRAPHICS_WINDOWED) {
         graphics_window_check_bounds_before_creation(x, y, nWidth, nHeight);
     }
@@ -411,14 +602,20 @@ static HWND WINAPI CreateWindowExA_hook(DWORD dwExStyle, LPCSTR lpClassName, LPC
     }
 
     // only hook touch window if multiple windows are allowed
-    if (is_gfdm_sub_window && !games::gitadora::ARENA_SINGLE_WINDOW) {
-        GFDM_SUBSCREEN_WINDOW = result;
+    if (gfdm_window_name == "LEFT" || gfdm_window_name == "RIGHT") {
+        gitadora_remember_window(result, gfdm_window_name);
+    }
+    if (is_gfdm_sub_window && GRAPHICS_WINDOWED && !GRAPHICS_PREVENT_SECONDARY_WINDOWS) {
+        gitadora_remember_window(result, gfdm_window_name);
         graphics_hook_subscreen_window(GFDM_SUBSCREEN_WINDOW);
     }
+    if (is_gfdm_window && GRAPHICS_WINDOWED && !GRAPHICS_PREVENT_SECONDARY_WINDOWS) {
+        gitadora_force_window_style(result);
+    }
 
     if (is_popn_sub_window) {
         POPN_SUBSCREEN_WINDOW = result;
-        if (!GRAPHICS_PREVENT_SECONDARY_WINDOW) {
+        if (!GRAPHICS_PREVENT_SECONDARY_WINDOWS) {
             graphics_hook_subscreen_window(POPN_SUBSCREEN_WINDOW);
         }
     }
@@ -603,6 +800,12 @@ static BOOL WINAPI MoveWindow_hook(HWND hWnd, int X, int Y, int nWidth, int nHei
         }
     }
 
+    // Monitor overrides are applied at creation time. Suppress later game
+    // placement calls instead of resizing again during scene transitions.
+    if (gitadora_should_block_game_window_placement(hWnd)) {
+        return TRUE;
+    }
+
     // call original
     return MoveWindow_orig(hWnd, X, Y, nWidth, nHeight, bRepaint);
 }
@@ -656,8 +859,17 @@ static LONG WINAPI SetWindowLongA_hook(HWND hWnd, int nIndex, LONG dwNewLong) {
         dwNewLong |= WS_OVERLAPPEDWINDOW;
     }
 
-    // call original
-    return SetWindowLongA_orig(hWnd, nIndex, dwNewLong);
+    const bool force_gfdm_style =
+        GRAPHICS_WINDOWED &&
+        games::gitadora::is_arena_model() &&
+        is_gfdm_known_window(hWnd) &&
+        (nIndex == GWL_STYLE || nIndex == GWL_EXSTYLE);
+
+    const auto result = SetWindowLongA_orig(hWnd, nIndex, dwNewLong);
+    if (force_gfdm_style) {
+        gitadora_force_window_style(hWnd);
+    }
+    return result;
 }
 
 static LONG WINAPI SetWindowLongW_hook(HWND hWnd, int nIndex, LONG dwNewLong) {
@@ -667,8 +879,17 @@ static LONG WINAPI SetWindowLongW_hook(HWND hWnd, int nIndex, LONG dwNewLong) {
         dwNewLong |= WS_OVERLAPPEDWINDOW;
     }
 
-    // call original
-    return SetWindowLongW_orig(hWnd, nIndex, dwNewLong);
+    const bool force_gfdm_style =
+        GRAPHICS_WINDOWED &&
+        games::gitadora::is_arena_model() &&
+        is_gfdm_known_window(hWnd) &&
+        (nIndex == GWL_STYLE || nIndex == GWL_EXSTYLE);
+
+    const auto result = SetWindowLongW_orig(hWnd, nIndex, dwNewLong);
+    if (force_gfdm_style) {
+        gitadora_force_window_style(hWnd);
+    }
+    return result;
 }
 
 static BOOL WINAPI SetWindowPos_hook(HWND hWnd, HWND hWndInsertAfter,
@@ -679,6 +900,13 @@ static BOOL WINAPI SetWindowPos_hook(HWND hWnd, HWND hWndInsertAfter,
         return TRUE;
     }
 
+    // Monitor overrides are applied at creation time. Suppress later game
+    // placement calls instead of resizing again during scene transitions.
+    if (gitadora_should_block_game_window_placement(hWnd) &&
+        (uFlags & (SWP_NOMOVE | SWP_NOSIZE)) != (SWP_NOMOVE | SWP_NOSIZE)) {
+        return TRUE;
+    }
+
     // prevent gitadora arena model from shifting windows around if the user has preferences
     if (GRAPHICS_WINDOWED && games::gitadora::is_arena_model() &&
         GRAPHICS_HOOKED_WINDOW.has_value() && hWnd == GRAPHICS_HOOKED_WINDOW.value() &&
@@ -692,14 +920,21 @@ static BOOL WINAPI SetWindowPos_hook(HWND hWnd, HWND hWndInsertAfter,
 
 static BOOL WINAPI ShowWindow_hook(HWND hWnd, int nCmdShow) {
     if (games::gitadora::is_arena_model() &&
-        games::gitadora::ARENA_SINGLE_WINDOW &&
+        GRAPHICS_PREVENT_SECONDARY_WINDOWS &&
         hWnd != GRAPHICS_HOOKED_WINDOW) {
         log_info("graphics", "ShowWindow_hook - hiding sub window {}", fmt::ptr(hWnd));
         return true;
     }
 
+    if (games::gitadora::is_arena_model() &&
+        GRAPHICS_GITADORA_HIDE_SIDE_WINDOWS &&
+        (hWnd == GFDM_LEFT_WINDOW || hWnd == GFDM_RIGHT_WINDOW)) {
+        log_info("graphics", "ShowWindow_hook - hiding GITADORA side window {}", fmt::ptr(hWnd));
+        return true;
+    }
+
     if (games::popn::is_pikapika_model() &&
-        GRAPHICS_PREVENT_SECONDARY_WINDOW &&
+        GRAPHICS_PREVENT_SECONDARY_WINDOWS &&
         hWnd == POPN_SUBSCREEN_WINDOW) {
         log_info("graphics", "ShowWindow_hook - hiding sub window {}", fmt::ptr(hWnd));
         return true;
@@ -831,6 +1066,7 @@ void graphics_init() {
 
     // init backends
     graphics_d3d9_init();
+    graphics_d3d11_init();
 
     // general hooks
     ChangeDisplaySettingsA_orig = detour::iat_try("ChangeDisplaySettingsA", ChangeDisplaySettingsA_hook);
@@ -987,18 +1223,20 @@ void graphics_capture_trigger(int screen) {
 }
 
 bool graphics_capture_consume(int *screen) {
-    auto flag = !GRAPHICS_CAPTURE_SCREENS.empty();
-    if (flag) {
-        std::lock_guard<std::mutex> lock(GRAPHICS_CAPTURE_SCREENS_M);
+    std::lock_guard<std::mutex> lock(GRAPHICS_CAPTURE_SCREENS_M);
 
-        *screen = GRAPHICS_CAPTURE_SCREENS.back();
-        GRAPHICS_CAPTURE_SCREENS.pop_back();
+    if (GRAPHICS_CAPTURE_SCREENS.empty()) {
+        return false;
     }
-    return flag;
+
+    *screen = GRAPHICS_CAPTURE_SCREENS.back();
+    GRAPHICS_CAPTURE_SCREENS.pop_back();
+    return true;
 }
 
 void graphics_capture_enqueue(int screen, uint8_t *data, size_t width, size_t height) {
     GRAPHICS_CAPTURE_BUFFER_M[screen].lock();
+    GRAPHICS_CAPTURE_SKIP_SIGNAL[screen] = false;
     auto &capture = GRAPHICS_CAPTURE_BUFFER[screen];
     capture.data.reset(data);
     capture.width = width;
@@ -1009,6 +1247,14 @@ void graphics_capture_enqueue(int screen, uint8_t *data, size_t width, size_t he
 }
 
 void graphics_capture_skip(int screen) {
+    if (screen < 0 || screen >= static_cast<int>(GRAPHICS_CAPTURE_SCREEN_NO)) {
+        return;
+    }
+
+    {
+        std::lock_guard<std::mutex> lock(GRAPHICS_CAPTURE_BUFFER_M[screen]);
+        GRAPHICS_CAPTURE_SKIP_SIGNAL[screen] = true;
+    }
     GRAPHICS_CAPTURE_CV[screen].notify_one();
 }
 
@@ -1016,11 +1262,32 @@ bool graphics_capture_receive_jpeg(int screen, TooJpeg::WRITE_ONE_BYTE receiver,
         bool rgb, int quality, bool downsample, int divide, uint64_t *timestamp,
         int *width, int *height) {
 
-    // wait for capture event
+    if (screen < 0 || screen >= static_cast<int>(GRAPHICS_CAPTURE_SCREEN_NO)) {
+        return false;
+    }
+
+    // wait for capture event (with timeout)
     std::unique_lock<std::mutex> lock(GRAPHICS_CAPTURE_BUFFER_M[screen]);
-    GRAPHICS_CAPTURE_CV[screen].wait(lock, [screen] {
-        return GRAPHICS_CAPTURE_BUFFER[screen].data != nullptr;
-    });
+    const bool ready = GRAPHICS_CAPTURE_CV[screen].wait_for(
+            lock,
+            GRAPHICS_CAPTURE_RECEIVE_TIMEOUT,
+            [screen] {
+                return GRAPHICS_CAPTURE_BUFFER[screen].data != nullptr
+                    || GRAPHICS_CAPTURE_SKIP_SIGNAL[screen];
+            });
+
+    if (!ready) {
+        lock.unlock();
+        graphics_capture_cancel_pending(screen);
+        return false;
+    }
+
+    if (GRAPHICS_CAPTURE_SKIP_SIGNAL[screen]) {
+        GRAPHICS_CAPTURE_SKIP_SIGNAL[screen] = false;
+        lock.unlock();
+        return false;
+    }
+
     auto &capture = GRAPHICS_CAPTURE_BUFFER[screen];
     auto capture_data = capture.data;
     auto capture_width = capture.width;
@@ -1278,8 +1545,13 @@ void change_primary_monitor(const std::string &monitor_name) {
     Sleep(2000);
 }
 
-void update_monitor(bool is_boot, std::optional<graphics_orientation> target_orientation, UINT target_refresh_rate) {
-    // note: all of this is only being done for the primary motnior
+void update_monitor(
+    bool is_boot,
+    std::optional<graphics_orientation> target_orientation,
+    UINT target_refresh_rate,
+    std::optional<std::pair<uint32_t, uint32_t>> target_resolution) {
+
+    // note: all of this is only being done for the primary monitor
 
     // get current settings
     DEVMODEA dm = {};
@@ -1357,6 +1629,22 @@ void update_monitor(bool is_boot, std::optional<graphics_orientation> target_ori
         needs_update = true;
     }
 
+    if (target_resolution.has_value()) {
+        if (dm.dmPelsWidth != target_resolution.value().first ||
+            dm.dmPelsHeight != target_resolution.value().second) {
+
+            log_misc(
+                "graphics",
+                "current resolution {}, {} => desired resolution {}, {}",
+                dm.dmPelsWidth, dm.dmPelsHeight,
+                target_resolution.value().first, target_resolution.value().second);
+
+            dm.dmPelsWidth = target_resolution.value().first;
+            dm.dmPelsHeight = target_resolution.value().second;
+            needs_update = true;
+        }
+    }
+
     if (!needs_update) {
         // nothing to do
         log_misc("graphics", "display settings are already up to date, no changes needed");
@@ -1393,17 +1681,30 @@ void update_monitor(bool is_boot, std::optional<graphics_orientation> target_ori
     }
 }
 
-void update_monitor_on_boot(std::optional<graphics_orientation> target_orientation, UINT target_refresh_rate) {
+void update_monitor_on_boot(
+    std::optional<graphics_orientation> target_orientation,
+    UINT target_refresh_rate,
+    std::optional<std::pair<uint32_t, uint32_t>> target_resolution) {
+
     target_orientation_on_boot = target_orientation;
     target_refresh_rate_on_boot = target_refresh_rate;
+    target_resolution_on_boot = target_resolution;
     log_misc("graphics", "applying monitor updates at boot...");
-    update_monitor(true, target_orientation, target_refresh_rate);
+    update_monitor(
+        true,
+        target_orientation,
+        target_refresh_rate,
+        target_resolution);
 }
 
 void update_monitor_at_runtime() {
-    if (monitor_settings_changed) {
+    if (!GRAPHICS_WINDOWED && monitor_settings_changed) {
         log_misc("graphics", "applying monitor updates at runtime as window regained focus...");
-        update_monitor(false, target_orientation_on_boot, target_refresh_rate_on_boot);   
+        update_monitor(
+            false,
+            target_orientation_on_boot,
+            target_refresh_rate_on_boot,
+            target_resolution_on_boot);
     }
 }