nost: avoid pass-by-value in PANB I/O (#465)

## Link to GitHub Issue, if one exists
#452

## Description of change
* Split the MDXF I/O option into three: auto, thread, and backfill.
`auto` is the same as what `thread` used to be before this change (based
on I/O poll rate), but `thread` now always forces thread poll mode.
`auto` is the default (behavior wise, same as before)
* For the poll rate measurement, start polling after 20s mark to avoid
frame drops during game boot
* Minor code clean up

## Testing
manual test in windowed + full screen

src/spice2x/CMakeLists.txt

@@ -526,6 +526,7 @@ set(SOURCE_FILES ${SOURCE_FILES}
         misc/device.cpp
         misc/eamuse.cpp
         misc/extdev.cpp
+        misc/nativetouchhook.cpp
         misc/sciunit.cpp
         misc/sde.cpp
         misc/wintouchemu.cpp

src/spice2x/acio/mdxf/mdxf.cpp

@@ -9,6 +9,15 @@
 #include "util/utils.h"
 #include <mutex>
 
+#define MDFX_DEBUG_VERBOSE 0
+
+#if MDFX_DEBUG_VERBOSE
+#define log_debug(module, format_str, ...) logger::push( \
+    LOG_FORMAT("M", module, format_str, ## __VA_ARGS__), logger::Style::GREY)
+#else
+#define log_debug(module, format_str, ...)
+#endif
+
 // constants
 const size_t STATUS_BUFFER_SIZE = 32;
 const size_t STATUS_BUFFER_NUM_ENTRIES = 16;
@@ -30,6 +39,21 @@ static bool IS_MDXF_ACTIVE = false;
 static const uint8_t BACKFILL_INTERVAL_MS = 4;
 static const uint8_t BACKFILL_PADDING_MS = 2;
 
+// These are used to determine if a thread needs to be spun up to keep pad state ring buffers populated with enough recent polls
+static uint64_t START_TIME = 0;
+static int CALL_COUNT = 0;
+static const int THRESHOLD_REFRESH_RATE = 120;
+
+static std::atomic<bool> IS_REFRESH_RATE_MEASUREMENT_STARTED{false};
+static std::atomic<bool> IS_REFRESH_RATE_DETERMINED{false};
+static std::atomic<bool> IS_THREAD_NEEDED{false};
+static std::atomic<bool> MDXF_THREAD_RUNNING{false};
+
+static std::thread MDXF_THREAD;
+
+static constexpr int THREAD_REFRESH_RATE_HZ = 125;
+static constexpr auto THREAD_PERIOD = std::chrono::milliseconds(1000 / THREAD_REFRESH_RATE_HZ);
+
 // buffers
 #pragma pack(push, 1)
 static struct {
@@ -40,9 +64,20 @@ static struct {
 
 static bool STATUS_BUFFER_FREEZE = false;
 
+// Decides which method to use for populating ring buffer entries for "padding".
+// Overwritten in spicecfg using P4IO Buffer Algorithm option.
+// THREAD_MODE: Spins a thread running at THREAD_REFRESH_RATE_HZ which periodically fills the ring
+//              buffer with auxiliary entries. Falls back on BACKFILL_MODE
+// BACKFILL_MODE: On every update cycle, fill the ring buffer with entries for the last known state
+//                BACKFILL_INTERVAL_MS apart from each other from the time of the last entry to the
+//                current time before adding the entry for the current state.
+// AUTO_MODE: thread mode if <120Hz, backfill if >=120Hz
+acio::MDXFBufferFillMode acio::MDXF_BUFFER_FILL_MODE = acio::MDXFBufferFillMode::AUTO_MODE;
+
 typedef enum {
     ARKMDXP4_POLL = 0,
-    EXTERNAL_POLL = 1
+    INTERNAL_POLL = 1,
+    EXTERNAL_POLL = 2
 } MDXFPollSource;
 
 typedef uint64_t (__cdecl *ARK_GET_TICK_TIME64_T)();
@@ -61,6 +96,106 @@ static uint64_t arkGetTickTime64() {
     return getTickTime64 ? getTickTime64() : timeGetTime();
 }
 
+// Used to keep the ring buffer populated with steady updates. 60Hz interval is too slow
+static void mdxf_thread_start() {
+    bool expected = false;
+    if (!MDXF_THREAD_RUNNING.compare_exchange_strong(expected, true)) {
+        return;
+    }
+
+    log_info("mdxf", "starting poll thread");
+
+    MDXF_THREAD = std::thread([] {
+        SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_BELOW_NORMAL);
+
+        while (MDXF_THREAD_RUNNING.load(std::memory_order_acquire)) {
+            mdxf_poll(false);
+            std::this_thread::sleep_for(THREAD_PERIOD);
+        }
+    });
+}
+
+static void mdxf_thread_stop() {
+    if (!MDXF_THREAD_RUNNING.exchange(false)) {
+        return;
+    }
+
+    if (MDXF_THREAD.joinable()) {
+        MDXF_THREAD.join();
+    }
+
+}
+
+// Snaps measured refresh rate to best fit
+static int snap_refresh_rate(int measured_hz) {
+    static constexpr std::array<int, 6> rates = {
+        60, 120, 144, 165, 180, 240
+    };
+
+    int best = rates[0];
+    int best_err = std::fabs(measured_hz - best);
+
+    for (int r : rates) {
+        int err = std::fabs(measured_hz - r);
+        if (err < best_err) {
+            best = r;
+            best_err = err;
+        }
+    }
+    return best;
+}
+
+// Increments the number of times the update function was called,
+// then calculates the current refresh rate of the game
+// (20 seconds after the game starts, until 25 seconds)
+static void count_calls_from_game() {
+    if (IS_REFRESH_RATE_DETERMINED) {
+        return;
+    }
+
+    const uint64_t current_time = arkGetTickTime64();
+
+    if (!IS_REFRESH_RATE_MEASUREMENT_STARTED) {
+        if (START_TIME == 0) {
+            START_TIME = current_time;
+        }
+
+        // boot screen takes about 10 seconds, so let's wait for double that
+        if ((current_time - START_TIME) < 20000) {
+            // too early, do nothing
+            return;
+        } else {
+            // 20s has passed for the first time, start measuring on next call
+            IS_REFRESH_RATE_MEASUREMENT_STARTED = true;
+            START_TIME = current_time;
+            log_debug("mdxf", "measurement begin");
+            return;
+        }
+    }
+
+    const uint64_t elapsed_time = current_time - START_TIME;
+    CALL_COUNT++;
+    if (elapsed_time >= 5000) {
+        double measured_hz = static_cast<double>(CALL_COUNT) * 1000.0 / static_cast<double>(elapsed_time);
+        
+        // Account for the main loop calling this twice per iteration
+        measured_hz *= 0.5;
+        const int snapped_hz = snap_refresh_rate(static_cast<int>(measured_hz));
+
+        IS_REFRESH_RATE_DETERMINED = true;
+        IS_THREAD_NEEDED = (snapped_hz < THRESHOLD_REFRESH_RATE);
+        log_info(
+            "mdxf",
+            "detected: {} Hz, best fit: {} Hz",
+            static_cast<int>(measured_hz),
+            snapped_hz);
+
+        if (IS_THREAD_NEEDED) {
+            mdxf_thread_start();
+        }
+    }
+}
+
 /*
  * Implementations
  */
@@ -92,22 +227,20 @@ static uint64_t __cdecl ac_io_mdxf_get_control_status_buffer(int node, void *out
         }
         
         std::lock_guard<std::mutex> lock(*mutex);
+        
+        const uint8_t start_index = (head_in == 0xFF) ? *head : head_in;
 
-        if (head_in != 0xFF) {
-            *head = head_in;
-        }
-
-        // Compute ring index: walk backwards from head as index increases
+        // Compute ring index: walk backwards from start_index as index increases
         // Assumes ring buffer size is a power of two
         const size_t mask = size - 1;
         const size_t offset = static_cast<size_t>(index) & mask;
-        const size_t i = (static_cast<size_t>(*head) - offset + size) & mask;
+        const size_t i = (static_cast<size_t>(start_index) - offset + size) & mask;
 
         // Copy the chosen entry
         memcpy(out, buffer[i], STATUS_BUFFER_SIZE);
 
-        // Return the head value actually used
-        return static_cast<uint64_t>(*head);
+        // Return the start value actually used
+        return static_cast<uint64_t>(start_index);
     }
 
     return error_ret;
@@ -162,8 +295,18 @@ static bool __cdecl ac_io_mdxf_update_control_status_buffer_impl(int node, MDXFP
     // Dance Dance Revolution
     if (avs::game::is_model("MDX")) {
         // Marks this module as actively being used, allowing this function to be called from other sources
-        if (!IS_MDXF_ACTIVE && source == ARKMDXP4_POLL) {
-            IS_MDXF_ACTIVE = true;
+        if (source == ARKMDXP4_POLL) {
+            if (!IS_MDXF_ACTIVE) {
+                log_debug("mdxf", "initializing mdxf I/O support");
+                IS_MDXF_ACTIVE = true;
+                if (acio::MDXF_BUFFER_FILL_MODE == acio::MDXFBufferFillMode::THREAD_MODE) {
+                    IS_THREAD_NEEDED = true;
+                    mdxf_thread_start();
+                }
+            }
+            if (acio::MDXF_BUFFER_FILL_MODE == acio::MDXFBufferFillMode::AUTO_MODE) {
+                count_calls_from_game();
+            }
         }
         
         uint8_t (*buffer)[STATUS_BUFFER_SIZE];
@@ -226,6 +369,8 @@ static bool __cdecl ac_io_mdxf_update_control_status_buffer_impl(int node, MDXFP
         }
         
         uint16_t current_state;
+        
+        // Only call getState() if called externally when actual input events happen, otherwise use previous known state
         if (source == EXTERNAL_POLL) {
             // get buttons
             auto &buttons = games::ddr::get_buttons();
@@ -276,8 +421,8 @@ static bool __cdecl ac_io_mdxf_update_control_status_buffer_impl(int node, MDXFP
             start_time = min_time;
         }
         
-        // Only write one entry if called externally
-        if (source == EXTERNAL_POLL) {
+        // Don't backfill entries if called externally or if a separate thread is being used to fill auxiliary entries
+        if (source == EXTERNAL_POLL || IS_THREAD_NEEDED) {
             start_time = stop_time - 1;
         }
         
@@ -323,11 +468,12 @@ static bool __cdecl ac_io_mdxf_update_control_status_buffer(int node) {
 }
 
 // Used for triggering updates of the controller states from outside arkmdxp4.dll main refresh loop (i.e. within rawinput.cpp on controller events)
-void mdxf_poll() {
+void mdxf_poll(bool isExternal) {
     if (IS_MDXF_ACTIVE) {
+        const MDXFPollSource source = isExternal ? EXTERNAL_POLL : INTERNAL_POLL;
         const uint64_t call_time_ms = arkGetTickTime64();
-        ac_io_mdxf_update_control_status_buffer_impl(17, EXTERNAL_POLL, call_time_ms);
-        ac_io_mdxf_update_control_status_buffer_impl(18, EXTERNAL_POLL, call_time_ms);
+        ac_io_mdxf_update_control_status_buffer_impl(17, source, call_time_ms);
+        ac_io_mdxf_update_control_status_buffer_impl(18, source, call_time_ms);
     }
 }
 
@@ -348,4 +494,10 @@ void acio::MDXFModule::attach() {
     ACIO_MODULE_HOOK(ac_io_mdxf_get_control_status_buffer);
     ACIO_MODULE_HOOK(ac_io_mdxf_set_output_level);
     ACIO_MODULE_HOOK(ac_io_mdxf_update_control_status_buffer);
+}
+
+acio::MDXFModule::~MDXFModule() {
+    if (IS_THREAD_NEEDED) {
+        mdxf_thread_stop();
+    }
 }

src/spice2x/acio/mdxf/mdxf.h

@@ -4,10 +4,24 @@
 
 namespace acio {
 
+    enum class MDXFBufferFillMode {
+        // backfill mode, but if <120Hz, enable poll thread
+        AUTO_MODE,
+
+        // forces poll thread
+        THREAD_MODE,
+
+        // forces backfill mode (no poll thread)
+        BACKFILL_MODE
+    };
+
+    extern MDXFBufferFillMode MDXF_BUFFER_FILL_MODE;
+
     class MDXFModule : public ACIOModule {
     public:
         MDXFModule(HMODULE module, HookMode hookMode);
 
         virtual void attach() override;
+        ~MDXFModule() override; 
     };
 }

src/spice2x/acio/mdxf/mdxf_poll.h

@@ -2,4 +2,4 @@
 #pragma once
 
 // Called from rawinput thread whenever inputs have just been updated
-void mdxf_poll();
+void mdxf_poll(bool isExternal);

src/spice2x/acio/panb/panb.cpp

@@ -84,7 +84,7 @@ static bool __cdecl ac_io_panb_start_auto_input() {
     return true;
 }
 
-static uint8_t panb_get_button_velocity(Button button, Button button_soft, Button button_medium, Button button_hard) {
+static uint8_t panb_get_button_velocity(Button& button, Button& button_soft, Button& button_medium, Button& button_hard) {
     const auto velocity = Buttons::getVelocity(RI_MGR, button);
     const auto velocity_soft = Buttons::getVelocity(RI_MGR, button_soft);
     const auto velocity_medium = Buttons::getVelocity(RI_MGR, button_medium);

src/spice2x/acio2emu/packet.cpp

@@ -95,12 +95,19 @@ namespace acio2emu {
             break;
 
         default:
-            log_fatal("acio2emu", "cannot set step: unknown value: {}", s);
+            log_fatal(
+                "acio2emu",
+                "cannot set step: unknown value: {}",
+                static_cast<uint32_t>(s));
             break;
         }
 
         if (!valid) {
-            log_fatal("acio2emu", "illegal transition detected: {} -> {}", step_, s);
+            log_fatal(
+                "acio2emu",
+                "illegal transition detected: {} -> {}",
+                static_cast<uint32_t>(step_),
+                static_cast<uint32_t>(s));
         }
 #endif
         step_ = s;

src/spice2x/games/iidx/iidx.cpp

@@ -20,6 +20,7 @@
 #include "hooks/sleephook.h"
 #include "launcher/options.h"
 #include "touch/touch.h"
+#include "misc/nativetouchhook.h"
 #include "misc/wintouchemu.h"
 #include "misc/eamuse.h"
 #include "util/detour.h"
@@ -332,7 +333,9 @@ namespace games::iidx {
                 // need to hook `avs2-core.dll` so AVS win32fs operations go through rom hook
                 devicehook_init(avs::core::DLL_INSTANCE);
 
-                if (!NATIVE_TOUCH) {
+                if (NATIVE_TOUCH) {
+                    nativetouchhook::hook(avs::game::DLL_INSTANCE);
+                } else {
                     wintouchemu::FORCE = true;
                     wintouchemu::INJECT_MOUSE_AS_WM_TOUCH = true;
                     wintouchemu::hook_title_ends("beatmania IIDX", "main", avs::game::DLL_INSTANCE);

src/spice2x/launcher/launcher.cpp

@@ -10,6 +10,7 @@
 #include "api/modules/capture.h"
 #include "acio/acio.h"
 #include "acio/icca/icca.h"
+#include "acio/mdxf/mdxf.h"
 #include "api/controller.h"
 #include "avs/automap.h"
 #include "avs/core.h"
@@ -1149,6 +1150,13 @@ int main_implementation(int argc, char *argv[]) {
             rawinput::RAWINPUT_REQUIRE_FOCUS = false;
         }
     }
+    if (options[launcher::Options::DDRP4IOBufferMode].is_active()) {
+        if (options[launcher::Options::DDRP4IOBufferMode].value_text() == "thread") {
+            acio::MDXF_BUFFER_FILL_MODE = acio::MDXFBufferFillMode::THREAD_MODE;
+        } else if (options[launcher::Options::DDRP4IOBufferMode].value_text() == "backfill") {
+            acio::MDXF_BUFFER_FILL_MODE = acio::MDXFBufferFillMode::BACKFILL_MODE;
+        } 
+    }
 
     if (options[launcher::Options::MidiAlgoVer].is_active()) {
         if (options[launcher::Options::MidiAlgoVer].value_text() == "legacy") {

src/spice2x/launcher/options.cpp

@@ -2291,6 +2291,26 @@ static const std::vector<OptionDefinition> OPTION_DEFINITIONS = {
         .setting_name = "20",
         .category = "I/O Options",
     },
+    {
+        // DDRP4IOBufferMode
+        .title = "DDR P4IO Buffer Algorithm",
+        .name = "ddrp4iobuffer",
+        .desc =
+            "Remember to restart after changing this value.\n\n"
+            "Sets the algorithm used to populate entries to the buffer of controller polls read by the game.\n\n"
+            "auto (default): thread mode if <120Hz, backfill if >=120Hz\n\n"
+            "thread: starts a thread to periodically insert polls into the buffer\n\n"
+            "backfill: fills the buffer on each frame with last known state info at short regular intervals up to the current time, then writes the current state.\n\n"
+            "Only has an effect when emulating P4IO (arkmdxp4.dll)",
+        .type = OptionType::Enum,
+        .game_name = "Dance Dance Revolution",
+        .category = "Game Options (Advanced)",
+        .elements = {
+            {"auto", ""},
+            {"thread", ""},
+            {"backfill", ""},
+        },
+    },
     {
         // InputRequiresFocus
         .title = "Input Requires Focus",

src/spice2x/launcher/options.h

@@ -241,6 +241,7 @@ namespace launcher {
             IIDXRecDisable,
             MidiAlgoVer,
             MidiNoteSustain,
+            DDRP4IOBufferMode,
             InputRequiresFocus,
             NostalgiaPoke,
             ForceBackBufferCount,

src/spice2x/misc/nativetouchhook.cpp

@@ -0,0 +1,91 @@
+// enable touch functions - set version to windows 7
+// mingw otherwise doesn't load touch stuff
+#define _WIN32_WINNT 0x0601
+
+#include "wintouchemu.h"
+
+#include "util/detour.h"
+#include "util/logging.h"
+
+#define TOUCH_SIMULATE_FAT_FINGERS 0
+#define TOUCH_DEBUG_VERBOSE 0
+
+#if TOUCH_DEBUG_VERBOSE
+#define log_debug(module, format_str, ...) logger::push( \
+    LOG_FORMAT("M", module, format_str, ## __VA_ARGS__), logger::Style::GREY)
+#else
+#define log_debug(module, format_str, ...)
+#endif
+
+namespace nativetouchhook {
+
+    static decltype(GetTouchInputInfo) *GetTouchInputInfo_orig = nullptr;
+
+    static void strip_contact_size(PTOUCHINPUT point) {
+
+#if TOUCH_SIMULATE_FAT_FINGERS
+        point->dwMask |= 0x004;
+        point->cxContact = 80 * 100;
+        point->cyContact = 60 * 100;
+#endif
+
+        // most monitors do not set TOUCHEVENTFMASK_CONTACTAREA, but for
+        // monitors that do set it, IIDX can get very confused (SDVX is not
+        // affected)
+        //
+        // while the test menu and the touch "glow" seem to work properly,
+        // interacting with subscreen menu items or entering PIN becomes
+        // very unpredictable
+        //
+        // to fix this, simply remove the contact area width and height
+        // 
+        // note: test menu > I/O > touch test  gives 5 numbers:
+        //       n: x, y, w, h
+        //       where 
+        //       n is the nth touch input since boot
+        //       x, y are coordinates (center of finger)
+        //       w, h are contact width and height
+        //
+        // when TOUCHEVENTFMASK_CONTACTAREA is not set, w/h will
+        // automatically be seen as 1x1, which works perfectly fine
+
+        log_debug(
+            "touch::native",
+            "[{}, {}] dwMask = 0x{:x}, cxContact = {}, cyContact = {}",
+            point->x / 100,
+            point->y / 100,
+            point->dwMask,
+            point->cxContact,
+            point->cyContact);
+
+        point->dwMask &= ~(0x004ul); // clear TOUCHEVENTFMASK_CONTACTAREA 
+        point->cxContact = 0;
+        point->cyContact = 0;
+    }
+
+    static BOOL WINAPI GetTouchInputInfoHook(
+        HTOUCHINPUT hTouchInput, UINT cInputs, PTOUCHINPUT pInputs, int cbSize) {
+
+        // call the original fist
+        const auto result = GetTouchInputInfo_orig(hTouchInput, cInputs, pInputs, cbSize);
+        if (result == 0) {
+            return result;
+        }
+
+        for (size_t i = 0; i < cInputs; i++) {
+            PTOUCHINPUT point = &pInputs[i];
+            strip_contact_size(point);
+        }
+        
+        return result;
+    }
+
+    void hook(HMODULE module) {
+        GetTouchInputInfo_orig = detour::iat_try("GetTouchInputInfo", GetTouchInputInfoHook, module);
+        if (GetTouchInputInfo_orig != nullptr) {
+            log_misc("touch::native", "GetTouchInputInfo hooked");
+        }
+    }
+
+}
+ 

src/spice2x/misc/nativetouchhook.h

@@ -0,0 +1,6 @@
+#include <windows.h>
+
+namespace nativetouchhook {
+    void hook(HMODULE module);
+}
+ 

src/spice2x/rawinput/rawinput.cpp

@@ -1850,7 +1850,7 @@ LRESULT CALLBACK rawinput::RawInputManager::input_wnd_proc(
             }
             
             // update controller state ring buffers (DDR/MDXF)
-            mdxf_poll();
+            mdxf_poll(true);
 
             // call the default window handler for cleanup
             DefWindowProc(hWnd, msg, wparam, lParam);
@@ -2835,4 +2835,4 @@ void rawinput::RawInputManager::remove_callback_midi(void * data, const std::fun
             [data, callback](MidiCallback const &cb) {
                 return cb.data == data && cb.f.target<void>() == callback.target<void>();
             }), this->callback_midi.end());
-}
+}