Update game status report form

fix(gpu): drop colorWriteEnable dynamic state on depth-only pipelines

Pipelines were created with VK_DYNAMIC_STATE_COLOR_WRITE_ENABLE_EXT
unconditionally on non-Apple platforms, but the draw path only calls
vkCmdSetColorWriteEnableEXT when there is at least one color attachment.
A depth-only draw therefore reached vkCmdDrawIndexed with the dynamic
state declared and never set, which violates
VUID-vkCmdDrawIndexed-None-07749.

With validation enabled the error is fatal, so any title that issues a
depth-only pass aborts. The launcher enables validation by default
(vulkan_validation_enabled = true), so this is the default configuration:
PAC-MAN WORLD Re-PAC crashes from the launcher while the same build runs
normally with --vulkan-validation false.

Gate the declaration on the same color_count the draw path uses. Both
sides read state.color_count from the same draw, so the two conditions
are now identical by construction.

.github/ISSUE_TEMPLATE/kytyps5-game-emulation.yaml

@@ -1,6 +1,6 @@
-name: Game Emulation Bug Report
-description: Report KytyPS5 game boot, crash, freeze, rendering, audio, input, or gameplay behavior problems.
-title: "[GAME BUG]: "
+name: Game Emulation Status Report
+description: Submit a KytyPS5 game compatibility result, including boot, rendering, audio, input, or gameplay behavior.
+title: "[GAME STATUS]: "
 body:
   - type: markdown
     attributes:
@@ -10,6 +10,7 @@ body:
         Before submitting, please:
         - Search existing issues first.
         - Re-test with a clean game dump.
+        - Use the latest KytyPS5 build available to you.
         - Attach the complete log file instead of edited snippets.
 
   - type: markdown
@@ -44,6 +45,28 @@ body:
     validations:
       required: true
 
+  - type: dropdown
+    id: compatibility-status
+    attributes:
+      label: Compatibility status
+      description: Select the furthest point the game consistently reaches on this build.
+      options:
+        - "Doesn't boot"
+        - Logo
+        - Main menu
+        - In game
+    validations:
+      required: true
+
+  - type: markdown
+    attributes:
+      value: |
+        **Status guide**
+        - **Doesn't boot:** The game does not show its first logo or startup screen.
+        - **Logo:** The game shows a logo or startup screen but does not reach the main menu.
+        - **Main menu:** The game reaches its menus but does not enter gameplay.
+        - **In game:** The game reaches controllable gameplay, even if bugs prevent completion.
+
   - type: input
     id: regression-builds
     attributes:
@@ -54,22 +77,22 @@ body:
   - type: markdown
     attributes:
       value: |
-        ## Problem details
+        ## Test result
 
   - type: textarea
     id: what-happened
     attributes:
-      label: What happened?
-      description: Describe the failure or incorrect behavior as directly as possible.
-      placeholder: "Example: The game reaches the main menu, then freezes after selecting Continue."
+      label: Result details
+      description: Describe what worked, what failed, and any visible or audible problems.
+      placeholder: "Example: The game reaches the main menu, then freezes after selecting Continue. Audio plays correctly until the freeze."
     validations:
       required: true
 
   - type: textarea
     id: steps-to-reproduce
     attributes:
-      label: Steps to reproduce
-      description: List the shortest path that triggers the problem from a fresh KytyPS5 start.
+      label: Steps to reproduce the result
+      description: List the shortest path to the reported status or failure from a fresh KytyPS5 start.
       placeholder: |
         1. Open KytyPS5.
         2. Boot the game.
@@ -81,7 +104,7 @@ body:
     id: expected-behavior
     attributes:
       label: Expected behavior
-      description: Explain what should happen at the same point in the game.
+      description: Explain what should happen at the same point in the game. If there is no known problem, write "No known issue".
       placeholder: "Example: The save should load and gameplay should continue normally."
     validations:
       required: true

.github/workflows/build.yml

@@ -373,7 +373,13 @@ jobs:
 
   release:
     name: Release KytyPS5
-    if: github.event_name == 'push' && github.repository == 'KytyPS5/KytyPS5'
+    if: >-
+      always() &&
+      github.event_name == 'push' &&
+      github.repository == 'KytyPS5/KytyPS5' &&
+      (needs.windows.result == 'success' ||
+       needs.macos.result == 'success' ||
+       needs.linux.result == 'success')
     needs: [metadata, windows, macos, linux]
     runs-on: ubuntu-latest
     permissions:
@@ -382,34 +388,62 @@ jobs:
       RELEASE_NAME: ${{ needs.metadata.outputs.release_tag }}
 
     steps:
-      - name: Download builds
+      - name: Download Windows build
+        if: needs.windows.result == 'success'
         uses: actions/download-artifact@v4
         with:
-          path: artifacts
+          name: KytyPS5-Windows-x64
+          path: artifacts/KytyPS5-Windows-x64
 
-      - name: Package builds
+      - name: Download macOS build
+        if: needs.macos.result == 'success'
+        uses: actions/download-artifact@v4
+        with:
+          name: KytyPS5-macOS-x86_64
+          path: artifacts/KytyPS5-macOS-x86_64
+
+      - name: Download Linux build
+        if: needs.linux.result == 'success'
+        uses: actions/download-artifact@v4
+        with:
+          name: KytyPS5-Linux-x86_64
+          path: artifacts/KytyPS5-Linux-x86_64
+
+      - name: Package Windows build
+        if: needs.windows.result == 'success'
         shell: bash
         run: |
           windows_dir="artifacts/KytyPS5-Windows-x64"
-          macos_dir="artifacts/KytyPS5-macOS-x86_64"
-          linux_dir="artifacts/KytyPS5-Linux-x86_64"
           test -d "$windows_dir"
-          test -d "$macos_dir"
-          test -d "$linux_dir"
-          chmod a+x \
-            "$macos_dir/launcher" \
-            "$macos_dir/kyty_emulator" \
-            "$macos_dir/libMoltenVK.dylib" \
-            "$linux_dir/launcher" \
-            "$linux_dir/kyty_emulator"
           (
             cd "$windows_dir"
             zip -r "$GITHUB_WORKSPACE/$RELEASE_NAME-Windows-x64.zip" .
           )
+
+      - name: Package macOS build
+        if: needs.macos.result == 'success'
+        shell: bash
+        run: |
+          macos_dir="artifacts/KytyPS5-macOS-x86_64"
+          test -d "$macos_dir"
+          chmod a+x \
+            "$macos_dir/launcher" \
+            "$macos_dir/kyty_emulator" \
+            "$macos_dir/libMoltenVK.dylib"
           (
             cd "$macos_dir"
             zip -r "$GITHUB_WORKSPACE/$RELEASE_NAME-macOS-x86_64.zip" .
           )
+
+      - name: Package Linux build
+        if: needs.linux.result == 'success'
+        shell: bash
+        run: |
+          linux_dir="artifacts/KytyPS5-Linux-x86_64"
+          test -d "$linux_dir"
+          chmod a+x \
+            "$linux_dir/launcher" \
+            "$linux_dir/kyty_emulator"
           tar -C "$linux_dir" -czf \
             "$GITHUB_WORKSPACE/$RELEASE_NAME-Linux-x86_64.tar.gz" .
 
@@ -418,11 +452,16 @@ jobs:
         env:
           GH_TOKEN: ${{ github.token }}
         run: |
-          assets=(
-            "$RELEASE_NAME-Windows-x64.zip"
-            "$RELEASE_NAME-macOS-x86_64.zip"
-            "$RELEASE_NAME-Linux-x86_64.tar.gz"
-          )
+          assets=()
+          for artifact in \
+            "$RELEASE_NAME-Windows-x64.zip" \
+            "$RELEASE_NAME-macOS-x86_64.zip" \
+            "$RELEASE_NAME-Linux-x86_64.tar.gz"; do
+            if [[ -f "$artifact" ]]; then
+              assets+=("$artifact")
+            fi
+          done
+          test "${#assets[@]}" -gt 0
           if gh release view "$RELEASE_NAME" --repo "$GITHUB_REPOSITORY" > /dev/null 2>&1; then
             gh release upload "$RELEASE_NAME" "${assets[@]}" \
               --clobber \

3rdparty/CMakeLists.txt

@@ -2,6 +2,32 @@ set(BUILD_SHARED_LIBS OFF CACHE BOOL "" FORCE)
 
 add_subdirectory(cpuinfo)
 
+# The Windows guest red-zone patcher needs a decoder/encoder and a small
+# trampoline assembler. Pin the same revisions used by shadPS4's implementation.
+include(FetchContent)
+
+set(XBYAK_BUILD_TESTS OFF CACHE BOOL "" FORCE)
+FetchContent_Declare(
+	xbyak
+	GIT_REPOSITORY https://github.com/herumi/xbyak.git
+	GIT_TAG 44a72f369268f7d552650891b296693e91db86bb
+	GIT_SHALLOW FALSE
+)
+
+set(ZYDIS_BUILD_EXAMPLES OFF CACHE BOOL "" FORCE)
+set(ZYDIS_BUILD_TOOLS OFF CACHE BOOL "" FORCE)
+set(ZYDIS_BUILD_TESTS OFF CACHE BOOL "" FORCE)
+set(ZYDIS_BUILD_DOXYGEN OFF CACHE BOOL "" FORCE)
+FetchContent_Declare(
+	zydis
+	GIT_REPOSITORY https://github.com/zyantific/zydis.git
+	GIT_TAG 120e0e705f8e3b507dc49377ac2879979f0d545c
+	GIT_SUBMODULES_RECURSE TRUE
+	GIT_SHALLOW FALSE
+)
+
+FetchContent_MakeAvailable(xbyak zydis)
+
 set(SDL_SHARED OFF CACHE BOOL "" FORCE)
 add_subdirectory(SDL2)
 

CMakeLists.txt

@@ -292,7 +292,7 @@ if (KYTY_CLANG_CL)
 	set_source_files_properties(${kyty_emulator_src} PROPERTIES COMPILE_FLAGS "-Wno-pragma-pack -Wno-deprecated-declarations -D_TIMESPEC_DEFINED")
 endif()
 
-set(kyty_emulator_link_libraries common Vulkan::Headers spirv-tools-opt spirv-tools SDL2-static imgui xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version)
+set(kyty_emulator_link_libraries common Vulkan::Headers spirv-tools-opt spirv-tools SDL2-static imgui xxhash FFmpeg::ffmpeg fmt::fmt nlohmann_json::nlohmann_json LibAtrac9 kyty_git_version Zydis::Zydis xbyak::xbyak)
 
 # Linux system libraries required by the static FFmpeg archive.
 if(LINUX)
@@ -433,13 +433,6 @@ add_executable(resource_tracking_tests EXCLUDE_FROM_ALL
 target_link_libraries(resource_tracking_tests fmt::fmt)
 target_include_directories(resource_tracking_tests PRIVATE ${inc_headers})
 
-add_executable(resource_mutex_tests EXCLUDE_FROM_ALL
-	"${KYTY_TESTS_DIR}/ResourceMutexTests.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/host_gpu/renderer/cache/resourceMutex.cpp"
-)
-target_link_libraries(resource_mutex_tests common)
-target_include_directories(resource_mutex_tests PRIVATE ${inc_headers})
-
 add_executable(audio_out2_port_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/AudioOut2PortTests.cpp"
 	"${KYTY_SOURCE_DIR}/libs/libAudio2.cpp"
@@ -515,13 +508,14 @@ if(BUILD_TESTING)
 	add_test(NAME shader_vertex_metadata COMMAND $<TARGET_FILE:shader_vertex_metadata_tests>)
 	add_test(NAME shader_stage_runtime COMMAND $<TARGET_FILE:shader_stage_runtime_tests>)
 	add_test(NAME resource_tracking COMMAND $<TARGET_FILE:resource_tracking_tests>)
-	add_test(NAME resource_mutex COMMAND $<TARGET_FILE:resource_mutex_tests>)
 	add_test(NAME event_queue_lifetime COMMAND $<TARGET_FILE:event_queue_lifetime_tests>)
 	add_test(NAME sync_on_address COMMAND $<TARGET_FILE:sync_on_address_tests>)
 	add_test(NAME audio_out2_port COMMAND $<TARGET_FILE:audio_out2_port_tests>)
 	add_test(NAME shader_recompiler_compute COMMAND $<TARGET_FILE:shader_recompiler_compute_tests>)
 	add_test(NAME virtual_memory_allocation
 		COMMAND $<TARGET_FILE:virtual_memory_allocation_tests>)
+	add_test(NAME guest_red_zone_patcher
+		COMMAND $<TARGET_FILE:virtual_memory_allocation_tests> --red-zone-patcher-only)
 	add_test(NAME command_scheduler_timeline
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --scheduler-only)
 	add_test(NAME stream_buffer_ring
@@ -565,7 +559,6 @@ if(BUILD_TESTING)
 		shader_vertex_metadata_tests
 		shader_stage_runtime_tests
 		resource_tracking_tests
-		resource_mutex_tests
 		event_queue_lifetime_tests
 		sync_on_address_tests
 		shader_recompiler_compute_tests

src/common/CMakeLists.txt

@@ -24,6 +24,7 @@ target_link_libraries(common
 		Tracy::TracyClient
 	PRIVATE
 		cpuinfo
+		kyty_git_version
 		spdlog::spdlog
 )
 

src/common/assert.cpp

@@ -3,6 +3,7 @@
 #include "common/debug.h"
 #include "common/logging/log.h"
 #include "common/subsystems.h"
+#include "kytyGitVersion.h"
 
 #include <cstdio>
 #include <cstdlib>
@@ -23,7 +24,7 @@ static std::string BuildFatalReport(const char* title, std::string_view text, co
 	DebugStack stack;
 	DebugStack::Trace(&stack);
 
-	std::string report = "--- Stack Trace ---\n";
+	std::string report = "--- Build ---\n" KYTY_BUILD_LABEL "\n--- Stack Trace ---\n";
 	for (int i = PRINT_STACK_FROM; i < stack.depth; i++) {
 		report += fmt::format("[{}] {:016x}\n", i - PRINT_STACK_FROM,
 		                      static_cast<uint64_t>(stack.GetAddr(i)));

src/common/emulatorConfig.cpp

@@ -101,6 +101,12 @@ bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+bool RedZoneProtectionEnabled() {
+	return g_config->red_zone_protection_enabled;
+}
+#endif
+
 const Keymap& GetKeymap() {
 	return g_config->keymap;
 }

src/common/emulatorConfig.h

@@ -45,7 +45,10 @@ struct ConfigOptions {
 	bool                   spirv_debug_printf_enabled  = false;
 	bool                   renderdoc_enabled           = false;
 	bool                   readback_linear_images      = false;
-	Keymap                 keymap;
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	bool red_zone_protection_enabled = false;
+#endif
+	Keymap keymap;
 };
 
 void Load(const ConfigOptions& cfg);
@@ -76,6 +79,9 @@ bool SpirvDebugPrintfEnabled();
 
 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+bool RedZoneProtectionEnabled();
+#endif
 
 const Keymap& GetKeymap();
 

src/common/threads.cpp

@@ -28,7 +28,8 @@
 #ifdef KYTY_WIN_CS
 #include <windows.h> // IWYU pragma: keep
 // IWYU pragma: no_include <winbase.h>
-constexpr DWORD KYTY_CS_SPIN_COUNT = 4000;
+constexpr DWORD    KYTY_CS_SPIN_COUNT          = 4000;
+constexpr uint64_t KYTY_SLEEP_SPIN_LIMIT_100NS = 500; // 50 us
 
 #ifndef CREATE_WAITABLE_TIMER_HIGH_RESOLUTION
 #define CREATE_WAITABLE_TIMER_HIGH_RESOLUTION 0x00000002
@@ -39,7 +40,10 @@ static void SleepHighResolution100ns(uint64_t units_100ns) {
 		return;
 	}
 
-	if (units_100ns <= 10000) {
+	// Keep spinning only where a kernel transition is
+	// likely to cost more than the requested delay; ordinary millisecond sleeps use the
+	// per-thread high-resolution waitable timer below.
+	if (units_100ns <= KYTY_SLEEP_SPIN_LIMIT_100NS) {
 		LARGE_INTEGER frequency {};
 		LARGE_INTEGER start {};
 		if (QueryPerformanceFrequency(&frequency) != 0 && QueryPerformanceCounter(&start) != 0 &&

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -18,7 +18,9 @@
 #include <algorithm>
 #include <array>
 #include <atomic>
+#include <bit>
 #include <cstdio>
+#include <cstring>
 #include <vector>
 
 #define KYTY_HW_CTX_PARSER_ARGS                                                                    \
@@ -895,23 +897,44 @@ KYTY_HW_CTX_PARSER(HwCtxSetModeControl) {
 	return 1;
 }
 
-static void HwCtxIgnorePolyOffsetRegister(uint32_t cmd_offset, uint32_t value) {
-	static std::atomic<uint32_t> log_count = 0;
+static void HwCtxSetPolyOffsetRegister(CommandProcessor& cp, uint32_t cmd_offset, uint32_t value) {
+	HW::PolyOffset offset = cp.GetCtx().GetPolyOffset();
 
-	auto count = log_count.fetch_add(1);
-	if (count < 8) {
-		LOGF_COLOR(Log::Color::Red,
-		           "\t temporary: ignoring polygon offset context register 0x%03" PRIx32
-		           " = 0x%08" PRIx32 " (depth bias not implemented)\n",
-		           cmd_offset, value);
+	switch (cmd_offset) {
+		case Pm4::PA_SU_POLY_OFFSET_DB_FMT_CNTL: {
+			const auto neg_bits =
+			    KYTY_PM4_GET(value, PA_SU_POLY_OFFSET_DB_FMT_CNTL, NEG_NUM_DB_BITS);
+			offset.neg_num_db_bits = std::bit_cast<int8_t>(static_cast<uint8_t>(neg_bits));
+			offset.db_is_float_fmt =
+			    KYTY_PM4_GET(value, PA_SU_POLY_OFFSET_DB_FMT_CNTL, DB_IS_FLOAT_FMT) != 0;
+			break;
+		}
+		case Pm4::PA_SU_POLY_OFFSET_CLAMP: offset.clamp = std::bit_cast<float>(value); break;
+		case Pm4::PA_SU_POLY_OFFSET_FRONT_SCALE:
+			offset.front_scale = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_FRONT_OFFSET:
+			offset.front_offset = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_BACK_SCALE:
+			offset.back_scale = std::bit_cast<float>(value);
+			break;
+		case Pm4::PA_SU_POLY_OFFSET_BACK_OFFSET:
+			offset.back_offset = std::bit_cast<float>(value);
+			break;
+		default:
+			EXIT("unknown polygon offset context register 0x%03" PRIx32 " = 0x%08" PRIx32 "\n",
+			     cmd_offset, value);
 	}
+
+	cp.GetCtx().SetPolyOffset(offset);
 }
 
 KYTY_HW_CTX_PARSER(HwCtxSetPolyOffsetRegisters) {
 	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
 
 	for (uint32_t i = 0; i < num_values; i++) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset + i, buffer[i]);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset + i, buffer[i]);
 	}
 
 	return num_values;
@@ -1312,26 +1335,36 @@ static void HwShSetCsRegister(CommandProcessor& cp, uint32_t cmd_offset, uint32_
 			    (value >> Pm4::COMPUTE_PGM_RSRC1_VGPRS_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_VGPRS_MASK;
 			cs_regs.sgprs =
 			    (value >> Pm4::COMPUTE_PGM_RSRC1_SGPRS_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_SGPRS_MASK;
-			cs_regs.bulky =
-			    (value >> Pm4::COMPUTE_PGM_RSRC1_BULKY_SHIFT) & Pm4::COMPUTE_PGM_RSRC1_BULKY_MASK;
-			cs_regs.wave_size = (((value >> Pm4::COMPUTE_PGM_RSRC1_W32_EN_SHIFT) &
-			                      Pm4::COMPUTE_PGM_RSRC1_W32_EN_MASK) != 0u)
-			                        ? 32u
-			                        : 64u;
+			cs_regs.float_mode    = (value >> Pm4::COMPUTE_PGM_RSRC1_FLOAT_MODE_SHIFT) &
+			                        Pm4::COMPUTE_PGM_RSRC1_FLOAT_MODE_MASK;
+			cs_regs.dx10_clamp    = ((value >> Pm4::COMPUTE_PGM_RSRC1_DX10_CLAMP_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_DX10_CLAMP_MASK) != 0u;
+			cs_regs.ieee_mode     = ((value >> Pm4::COMPUTE_PGM_RSRC1_IEEE_MODE_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_IEEE_MODE_MASK) != 0u;
+			cs_regs.fp16_overflow = ((value >> Pm4::COMPUTE_PGM_RSRC1_FP16_OVFL_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_FP16_OVFL_MASK) != 0u;
+			cs_regs.bulky         = ((value >> Pm4::COMPUTE_PGM_RSRC1_BULKY_SHIFT) &
+			                         Pm4::COMPUTE_PGM_RSRC1_BULKY_MASK) != 0u;
+			cs_regs.threadgroup_configuration = ((value >> Pm4::COMPUTE_PGM_RSRC1_WGP_MODE_SHIFT) &
+			                                     Pm4::COMPUTE_PGM_RSRC1_WGP_MODE_MASK) != 0u;
+			cs_regs.wave_size                 = (((value >> Pm4::COMPUTE_PGM_RSRC1_W32_EN_SHIFT) &
+			                                      Pm4::COMPUTE_PGM_RSRC1_W32_EN_MASK) != 0u)
+			                                        ? 32u
+			                                        : 64u;
 			break;
 		case Pm4::COMPUTE_PGM_RSRC2:
-			cs_regs.scratch_en     = (value >> Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_MASK;
+			cs_regs.scratch_en     = ((value >> Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_MASK) != 0u;
 			cs_regs.user_sgpr      = (value >> Pm4::COMPUTE_PGM_RSRC2_USER_SGPR_SHIFT) &
 			                         Pm4::COMPUTE_PGM_RSRC2_USER_SGPR_MASK;
-			cs_regs.tgid_x_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_MASK;
-			cs_regs.tgid_y_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_MASK;
-			cs_regs.tgid_z_en      = (value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_MASK;
-			cs_regs.tg_size_en     = (value >> Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_SHIFT) &
-			                         Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_MASK;
+			cs_regs.tgid_x_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_X_EN_MASK) != 0u;
+			cs_regs.tgid_y_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_Y_EN_MASK) != 0u;
+			cs_regs.tgid_z_en      = ((value >> Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TGID_Z_EN_MASK) != 0u;
+			cs_regs.tg_size_en     = ((value >> Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_SHIFT) &
+			                          Pm4::COMPUTE_PGM_RSRC2_TG_SIZE_EN_MASK) != 0u;
 			cs_regs.tidig_comp_cnt = (value >> Pm4::COMPUTE_PGM_RSRC2_TIDIG_COMP_CNT_SHIFT) &
 			                         Pm4::COMPUTE_PGM_RSRC2_TIDIG_COMP_CNT_MASK;
 			cs_regs.lds_size       = (value >> Pm4::COMPUTE_PGM_RSRC2_LDS_SIZE_SHIFT) &
@@ -2102,7 +2135,7 @@ KYTY_CP_OP_PARSER(CpOpClearState) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0001200);
 	EXIT_NOT_IMPLEMENTED((buffer[0] & ~0xfu) != 0);
 
-	cp.Reset();
+	cp.ApplyContextStateOperation(ContextStateOperation::Clear);
 
 	return 1;
 }
@@ -2649,14 +2682,21 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 	const bool gl2_writeback = ((gcr_cntl & GcrGl2Writeback) != 0);
 
 	auto trigger_interrupt = [&]() {
+		bool queued = false;
 		switch (interrupt_selector) {
 			case 0x00:
 			case 0x03: break;
 			case 0x01:
 			case 0x02:
-			case 0x04: cp.TriggerEopEventAtEndOfPipe(interrupt_context_id); break;
+			case 0x04:
+				cp.TriggerEopEventAtEndOfPipe(interrupt_context_id);
+				queued = true;
+				break;
 			default: EXIT("unknown release_mem interrupt selector\n");
 		}
+		if (queued) {
+			cp.BufferFlush();
+		}
 	};
 
 	if (data_sel == 0 || interrupt_selector == 4) {
@@ -2664,10 +2704,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			cp.EmitGlobalBarrier();
 		}
 
-		if (gl2_writeback) {
-			cp.SynchronizeGpu();
-		}
-
 		trigger_interrupt();
 
 		return 7;
@@ -2678,10 +2714,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 			cp.EmitGlobalBarrier();
 		}
 
-		if (gl2_writeback) {
-			cp.SynchronizeGpu();
-		}
-
 		trigger_interrupt();
 
 		return 7;
@@ -2695,17 +2727,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 	cache_policy      = 0;
 
 	if (data_sel == 1) {
-		static std::atomic<uint32_t> log_count {0};
-		const auto                   dst_addr = reinterpret_cast<uint64_t>(dst_gpu_addr);
-		if ((dst_addr & 0xffffffffff000000ull) == 0x0000001021000000ull || log_count.load() < 64) {
-			if (log_count.fetch_add(1) < 256) {
-				LOGF(
-				    "\t runtime release_mem32: dst=0x%016" PRIx64 ", value=0x%08" PRIx32
-				    ", action=0x%02" PRIx32 ", gcr=0x%04" PRIx32 ", cache_action=0x%08" PRIx32 "\n",
-				    dst_addr, static_cast<uint32_t>(value), eop_event_type, gcr_cntl, cache_action);
-			}
-		}
-
 		eop_event_type    = 0x2f;
 		event_index       = 6;
 		auto event_source = 2u;
@@ -2720,7 +2741,6 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 
 	if (data_sel == 5) {
 		if (gl2_writeback) {
-			cp.SynchronizeGpu();
 			cache_action = 0;
 		}
 
@@ -2731,7 +2751,9 @@ KYTY_CP_OP_PARSER(CpOpReleaseMem) {
 		cp.WriteAtEndOfPipe32(cache_policy, event_write_dest, eop_event_type, cache_action,
 		                      event_index, event_source, dst_gpu_addr, static_cast<uint32_t>(value),
 		                      interrupt_selector, interrupt_context_id);
-		cp.BufferFlush();
+		if (interrupt_selector == 0x01) {
+			cp.BufferFlush();
+		}
 
 		return 7;
 	}
@@ -3695,22 +3717,22 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_DB_FMT_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_CLAMP] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_FRONT_SCALE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_FRONT_OFFSET] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_BACK_SCALE] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_BACK_OFFSET] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwCtxIgnorePolyOffsetRegister(cmd_offset, value);
+		HwCtxSetPolyOffsetRegister(cp, cmd_offset, value);
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_Z_INFO] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -4233,29 +4255,28 @@ void GraphicsInitJmpTablesShIndirect() {
 	};
 
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_RSRC1_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HW::VsShaderResource1 r1;
-		r1.vgprs                = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, VGPRS);
-		r1.sgprs                = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, SGPRS);
-		r1.priority             = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, PRIORITY);
-		r1.float_mode           = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, FLOAT_MODE);
-		r1.dx10_clamp           = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, DX10_CLAMP) != 0;
-		r1.ieee_mode            = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, IEEE_MODE) != 0;
-		r1.vgpr_component_count = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, VGPR_COMP_CNT);
-		r1.cu_group_enable = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, CU_GROUP_ENABLE) != 0;
+		HW::HsShaderResource1 r1;
+		r1.vgprs      = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, VGPRS);
+		r1.priority   = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, PRIORITY);
+		r1.float_mode = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, FLOAT_MODE);
+		r1.dx10_clamp = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, DX10_CLAMP) != 0;
+		r1.debug_mode = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, DEBUG_MODE) != 0;
+		r1.ieee_mode  = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, IEEE_MODE) != 0;
 		r1.require_forward_progress =
-		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, FWD_PROGRESS) != 0;
-		r1.fp16_overflow = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_VS, FP16_OVFL) != 0;
+		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, FWD_PROGRESS) != 0;
+		r1.threadgroup_configuration = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, WGP_MODE) != 0;
+		r1.ls_vgpr_component_count = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, LS_VGPR_COMP_CNT);
+		r1.fp16_overflow           = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_HS, FP16_OVFL) != 0;
 		cp.GetShCtx().SetHsShaderResource1(r1);
 	};
 
 	g_hw_sh_indirect_func[Pm4::SPI_SHADER_PGM_RSRC2_HS] = [](KYTY_HW_SH_INDIRECT_ARGS) {
-		HW::VsShaderResource2 r2;
-		r2.scratch_en        = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, SCRATCH_EN) != 0;
-		r2.user_sgpr         = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, USER_SGPR) +
-		                       (KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, USER_SGPR_MSB) << 5u);
-		r2.offchip_lds       = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, OC_LDS_EN) != 0;
-		r2.streamout_enabled = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, SO_EN) != 0;
-		r2.shared_vgprs      = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_VS, SHARED_VGPR_CNT);
+		HW::HsShaderResource2 r2;
+		r2.scratch_en   = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, SCRATCH_EN) != 0;
+		r2.user_sgpr    = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, USER_SGPR) +
+		                  (KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, USER_SGPR_MSB) << 5u);
+		r2.lds_size     = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, LDS_SIZE);
+		r2.shared_vgprs = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_HS, SHARED_VGPR_CNT);
 		cp.GetShCtx().SetHsShaderResource2(r2);
 	};
 
@@ -4343,7 +4364,7 @@ void GraphicsInitJmpTablesShIndirect() {
 		r1.cu_group_enable = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, CU_GROUP_ENABLE) != 0;
 		r1.require_forward_progress =
 		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, FWD_PROGRESS) != 0;
-		r1.lds_configuration       = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, WGP_MODE) != 0;
+		r1.threadgroup_configuration = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, WGP_MODE) != 0;
 		r1.gs_vgpr_component_count = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, GS_VGPR_COMP_CNT);
 		r1.fp16_overflow           = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC1_GS, FP16_OVFL) != 0;
 		cp.GetShCtx().SetGsShaderResource1(r1);
@@ -4403,7 +4424,7 @@ void GraphicsInitJmpTablesShIndirect() {
 		r2.wave_cnt_en    = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, WAVE_CNT_EN);
 		r2.extra_lds_size = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, EXTRA_LDS_SIZE);
 		r2.raster_ordered_shading =
-		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, LOAD_INTRAWAVE_COLLISION);
+		    KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, RASTER_ORDERED_SHADING);
 		r2.shared_vgprs = KYTY_PM4_GET(value, SPI_SHADER_PGM_RSRC2_PS, SHARED_VGPR_CNT);
 		cp.GetShCtx().SetPsShaderResource2(r2);
 	};

src/graphics/guest_gpu/gpu_defs.h

@@ -60,24 +60,34 @@ enum class GsOutputPrimitiveType : uint32_t {
 };
 
 enum class ChannelLayout : uint32_t {
-	kInvalid     = 0,
-	k8           = 1,
-	k16          = 2,
-	k8_8         = 3,
-	k32          = 4,
-	k16_16       = 5,
-	k11_11_10    = 6,
-	k10_10_10_2  = 9,
-	k8_8_8_8     = 10,
-	k32_32       = 11,
-	k16_16_16_16 = 12,
-	k32_32_32_32 = 14,
-	k5_6_5       = 16,
-	k5_5_5_1     = 17,
-	k4_4_4_4     = 19,
-	kBc1         = 35,
-	kBc2         = 36,
-	kBc3         = 37,
+	kInvalid         = 0,
+	k8               = 1,
+	k16              = 2,
+	k8_8             = 3,
+	k32              = 4,
+	k16_16           = 5,
+	k11_11_10        = 6,
+	k10_11_11        = 7,
+	k2_10_10_10      = 8,
+	k10_10_10_2      = 9,
+	k8_8_8_8         = 10,
+	k32_32           = 11,
+	k16_16_16_16     = 12,
+	k32_32_32        = 13,
+	k32_32_32_32     = 14,
+	k5_6_5           = 16,
+	k5_5_5_1         = 17,
+	k1_5_5_5         = 18,
+	k4_4_4_4         = 19,
+	k10_10_10_2Float = 31,
+	k9_9_9_5         = 34,
+	kBc1             = 35,
+	kBc2             = 36,
+	kBc3             = 37,
+	kBc4             = 38,
+	kBc5             = 39,
+	kBc6             = 40,
+	kBc7             = 41,
 };
 
 enum class ChannelType : uint32_t {

src/graphics/guest_gpu/graphicsRun.cpp

@@ -124,7 +124,6 @@ private:
 		uint64_t       flip_request_id               = 0;
 	};
 
-	void              WaitLocked();
 	void              Enqueue(Submission submission);
 	void              WaitForIdle();
 	void              ProcessCommands();
@@ -276,10 +275,8 @@ void GpuState::SubmitFlipPreparation(uint64_t request_id) {
 
 void GpuState::Done() {
 	GpuMutexLock lock(m_submission_mutex);
-	if (IsGpuThread()) {
-		m_gfx_cp->BufferWait();
-	} else {
-		WaitLocked();
+	if (!IsGpuThread()) {
+		WaitForIdle();
 	}
 	m_graphics_done = true;
 	m_done_num++;
@@ -289,11 +286,6 @@ int GpuState::GetFrameNum() {
 	return m_done_num;
 }
 
-void GpuState::WaitLocked() {
-	WaitForIdle();
-	SendCommandSync([this] { m_gfx_cp->BufferWait(); });
-}
-
 CommandProcessor& GpuState::GetProcessor(uint32_t queue_id) {
 	EXIT_IF(queue_id >= QueueCount);
 	if (queue_id == 0) {
@@ -307,13 +299,11 @@ CommandProcessor& GpuState::GetProcessor(uint32_t queue_id) {
 }
 
 void CommandProcessor::Reset() {
-	BufferWait();
-
 	m_sh_ctx.Reset();
 	m_ucfg.Reset();
 	m_ctx.Reset();
 	m_saved_ctx.Reset();
-	m_context_state_pushed = false;
+	m_context_state_pushed             = false;
 	m_index_type_and_size              = 0;
 	m_index_buffer_size                = 0;
 	m_user_data_marker                 = HW::UserSgprType::Unknown;
@@ -385,7 +375,6 @@ void CommandProcessor::WaitDeDiff(uint32_t diff) {
 }
 
 void CommandProcessor::IncrementDe() {
-	BufferWait();
 	m_de_count++;
 }
 
@@ -1176,9 +1165,9 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_group_y,
                                       uint32_t thread_group_z, uint32_t mode) {
 	uint32_t frame_num = 0;
-	uint32_t local_x   = 1;
-	uint32_t local_y   = 1;
-	uint32_t local_z   = 1;
+	// uint32_t local_x   = 1;
+	// uint32_t local_y   = 1;
+	// uint32_t local_z   = 1;
 
 	{
 		CheckBuffer();
@@ -1202,9 +1191,9 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 		}
 
 		const auto& cs = m_sh_ctx.GetCs().cs_regs;
-		local_x        = std::max(cs.num_thread_x, 1u);
-		local_y        = std::max(cs.num_thread_y, 1u);
-		local_z        = std::max(cs.num_thread_z, 1u);
+		// local_x        = std::max(cs.num_thread_x, 1u);
+		// local_y        = std::max(cs.num_thread_y, 1u);
+		// local_z        = std::max(cs.num_thread_z, 1u);
 		if (cs.wave_size == 64u) {
 			static std::atomic_bool logged_wave64_shader {false};
 			if (!logged_wave64_shader.exchange(true, std::memory_order_relaxed)) {
@@ -1220,27 +1209,27 @@ void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_g
 		                                              thread_group_y, thread_group_z, mode);
 	}
 
-	constexpr uint32_t DispatchInitiatorUseThreadDimensions = 1u << 5u;
+	/*constexpr uint32_t DispatchInitiatorUseThreadDimensions = 1u << 5u;
 	auto               group_count = [](uint32_t threads, uint32_t group_size) {
-		return (threads == 0
-		            ? 0u
-		            : (threads + std::max(group_size, 1u) - 1u) / std::max(group_size, 1u));
+	    return (threads == 0
+	                ? 0u
+	                : (threads + std::max(group_size, 1u) - 1u) / std::max(group_size, 1u));
 	};
 
 	auto groups_x = thread_group_x;
 	auto groups_y = thread_group_y;
 	auto groups_z = thread_group_z;
 	if ((mode & DispatchInitiatorUseThreadDimensions) != 0) {
-		groups_x = group_count(thread_group_x, local_x);
-		groups_y = group_count(thread_group_y, local_y);
-		groups_z = group_count(thread_group_z, local_z);
+	    groups_x = group_count(thread_group_x, local_x);
+	    groups_y = group_count(thread_group_y, local_y);
+	    groups_z = group_count(thread_group_z, local_z);
 	}
 
 	const uint64_t invocations =
 	    static_cast<uint64_t>(groups_x) * groups_y * groups_z * local_x * local_y * local_z;
 	if (invocations != 0) {
-		BufferFlushAndWait();
-	}
+	    BufferFlushAndWait();
+	}*/
 }
 
 void CommandProcessor::DispatchIndirect(uint32_t data_offset, uint32_t mode) {
@@ -1352,6 +1341,9 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					              value >> 16u);
 					Sync::WriteAtEndOfPipeGds32(m_submit_id, CurrentBuffer(), dst, value & 0xffffu,
 					                            value >> 16u);
+					if (with_interrupt) {
+						m_renderer.TriggerEopEvent(interrupt_context_id);
+					}
 					return;
 				}
 			} else if (eop_event_type == 0x04 && cache_action == 0x00 && event_index == 0x05) {
@@ -1554,7 +1546,6 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index) {
 				     event_index);
 			}
 			EmitGlobalBarrier();
-			SynchronizeGpu();
 			break;
 		// DbDataWritebackInvalidate, DbMetadataWritebackInvalidate, CbMetadataWritebackInvalidate.
 		case 0x0000002a:

src/graphics/guest_gpu/hardwareContext.h

@@ -348,6 +348,16 @@ struct ModeControl {
 	bool    persp_corr_dis           = false;
 };
 
+struct PolyOffset {
+	int8_t neg_num_db_bits = -23;
+	bool   db_is_float_fmt = true;
+	float  clamp           = 0.0f;
+	float  front_scale     = 0.0f;
+	float  front_offset    = 0.0f;
+	float  back_scale      = 0.0f;
+	float  back_offset     = 0.0f;
+};
+
 struct BlendControl {
 	uint8_t color_srcblend       = 1;
 	uint8_t color_comb_fcn       = 0;
@@ -482,6 +492,26 @@ struct VsShaderResource2 {
 	uint8_t shared_vgprs      = 0;
 };
 
+struct HsShaderResource1 {
+	uint8_t vgprs                     = 0;
+	uint8_t priority                  = 0;
+	uint8_t float_mode                = 0;
+	bool    dx10_clamp                = false;
+	bool    debug_mode                = false;
+	bool    ieee_mode                 = false;
+	bool    require_forward_progress  = false;
+	bool    threadgroup_configuration = false;
+	uint8_t ls_vgpr_component_count   = 0;
+	bool    fp16_overflow             = false;
+};
+
+struct HsShaderResource2 {
+	bool     scratch_en   = false;
+	uint8_t  user_sgpr    = 0;
+	uint16_t lds_size     = 0;
+	uint8_t  shared_vgprs = 0;
+};
+
 struct PsShaderResource1 {
 	uint8_t vgprs                    = 0;
 	uint8_t sgprs                    = 0;
@@ -500,7 +530,7 @@ struct PsShaderResource2 {
 	uint8_t user_sgpr              = 0;
 	bool    wave_cnt_en            = false;
 	uint8_t extra_lds_size         = 0;
-	bool    raster_ordered_shading = false;
+	uint8_t raster_ordered_shading = 0;
 	uint8_t shared_vgprs           = 0;
 };
 
@@ -513,22 +543,27 @@ struct PsStageRegisters {
 
 struct CsStageRegisters {
 
-	uint64_t data_addr      = 0;
-	uint32_t num_thread_x   = 0;
-	uint32_t num_thread_y   = 0;
-	uint32_t num_thread_z   = 0;
-	uint8_t  vgprs          = 0;
-	uint8_t  sgprs          = 0;
-	uint8_t  bulky          = 0;
-	uint8_t  wave_size      = 64;
-	uint8_t  scratch_en     = 0;
-	uint8_t  user_sgpr      = 0;
-	uint8_t  tgid_x_en      = 0;
-	uint8_t  tgid_y_en      = 0;
-	uint8_t  tgid_z_en      = 0;
-	uint8_t  tg_size_en     = 0;
-	uint8_t  tidig_comp_cnt = 0;
-	uint8_t  lds_size       = 0;
+	uint64_t data_addr                 = 0;
+	uint32_t num_thread_x              = 0;
+	uint32_t num_thread_y              = 0;
+	uint32_t num_thread_z              = 0;
+	uint8_t  vgprs                     = 0;
+	uint8_t  sgprs                     = 0;
+	uint8_t  float_mode                = 0;
+	bool     dx10_clamp                = false;
+	bool     ieee_mode                 = false;
+	bool     fp16_overflow             = false;
+	bool     bulky                     = false;
+	bool     threadgroup_configuration = false;
+	uint8_t  wave_size                 = 64;
+	bool     scratch_en                = false;
+	uint8_t  user_sgpr                 = 0;
+	bool     tgid_x_en                 = false;
+	bool     tgid_y_en                 = false;
+	bool     tgid_z_en                 = false;
+	bool     tg_size_en                = false;
+	uint8_t  tidig_comp_cnt            = 0;
+	uint16_t lds_size                  = 0;
 };
 
 struct EsStageRegisters {
@@ -543,24 +578,24 @@ struct LsStageRegisters {
 
 struct HsStageRegisters {
 	uint64_t          data_addr = 0;
-	VsShaderResource1 rsrc1;
-	VsShaderResource2 rsrc2;
+	HsShaderResource1 rsrc1;
+	HsShaderResource2 rsrc2;
 	uint64_t          chksum = 0;
 };
 
 struct GsShaderResource1 {
-	uint8_t vgprs                    = 0;
-	uint8_t sgprs                    = 0;
-	uint8_t priority                 = 0;
-	uint8_t float_mode               = 0;
-	bool    dx10_clamp               = false;
-	bool    debug_mode               = false;
-	bool    ieee_mode                = false;
-	bool    cu_group_enable          = false;
-	bool    require_forward_progress = false;
-	bool    lds_configuration        = false;
-	uint8_t gs_vgpr_component_count  = 0;
-	bool    fp16_overflow            = false;
+	uint8_t vgprs                     = 0;
+	uint8_t sgprs                     = 0;
+	uint8_t priority                  = 0;
+	uint8_t float_mode                = 0;
+	bool    dx10_clamp                = false;
+	bool    debug_mode                = false;
+	bool    ieee_mode                 = false;
+	bool    cu_group_enable           = false;
+	bool    require_forward_progress  = false;
+	bool    threadgroup_configuration = false;
+	uint8_t gs_vgpr_component_count   = 0;
+	bool    fp16_overflow             = false;
 };
 
 struct GsShaderResource2 {
@@ -880,6 +915,8 @@ public:
 	void SetDepthControl(const DepthControl& control) { m_depth_control = control; }
 	[[nodiscard]] const ModeControl& GetModeControl() const { return m_mode_control; }
 	void SetModeControl(const ModeControl& control) { m_mode_control = control; }
+	[[nodiscard]] const PolyOffset& GetPolyOffset() const { return m_poly_offset; }
+	void SetPolyOffset(const PolyOffset& offset) { m_poly_offset = offset; }
 	[[nodiscard]] const EqaaControl& GetEqaaControl() const { return m_eqaa_control; }
 	void SetEqaaControl(const EqaaControl& control) { m_eqaa_control = control; }
 	[[nodiscard]] const StencilControl& GetStencilControl() const { return m_stencil_control; }
@@ -977,6 +1014,7 @@ private:
 	uint8_t           m_stencil_clear_value = 0;
 
 	ModeControl m_mode_control;
+	PolyOffset  m_poly_offset;
 	EqaaControl m_eqaa_control;
 
 	ShaderRegisters m_sh_regs;
@@ -1034,8 +1072,8 @@ public:
 	void SetLsShaderResource1(const VsShaderResource1& rsrc1) { m_vs.ls_regs.rsrc1 = rsrc1; }
 	void SetLsShaderResource2(const VsShaderResource2& rsrc2) { m_vs.ls_regs.rsrc2 = rsrc2; }
 	void SetHsShaderBase(uint64_t addr) { m_vs.hs_regs.data_addr = addr; }
-	void SetHsShaderResource1(const VsShaderResource1& rsrc1) { m_vs.hs_regs.rsrc1 = rsrc1; }
-	void SetHsShaderResource2(const VsShaderResource2& rsrc2) { m_vs.hs_regs.rsrc2 = rsrc2; }
+	void SetHsShaderResource1(const HsShaderResource1& rsrc1) { m_vs.hs_regs.rsrc1 = rsrc1; }
+	void SetHsShaderResource2(const HsShaderResource2& rsrc2) { m_vs.hs_regs.rsrc2 = rsrc2; }
 	void SetHsShaderChksum(uint32_t value) { m_vs.hs_regs.chksum = value; }
 	void SetGsShaderBase(uint64_t addr) { m_vs.gs_regs.data_addr = addr; }
 	void SetGsShaderResource1(const GsShaderResource1& rsrc1) { m_vs.gs_regs.rsrc1 = rsrc1; }

src/graphics/guest_gpu/pm4.h

@@ -578,24 +578,28 @@ constexpr uint32_t DB_HTILE_SURFACE_PREFETCH_HEIGHT_MASK          = 0x3F;
 constexpr uint32_t DB_HTILE_SURFACE_DST_OUTSIDE_ZERO_TO_ONE_SHIFT = 16;
 constexpr uint32_t DB_HTILE_SURFACE_DST_OUTSIDE_ZERO_TO_ONE_MASK  = 0x1;
 
-constexpr uint32_t DB_SRESULTS_COMPARE_STATE0     = 0x2B0;
-constexpr uint32_t DB_SRESULTS_COMPARE_STATE1     = 0x2B1;
-constexpr uint32_t VGT_GS_MAX_VERT_OUT            = 0x2CE;
-constexpr uint32_t GE_NGG_SUBGRP_CNTL             = 0x2D3;
-constexpr uint32_t VGT_TESS_DISTRIBUTION          = 0x2D4;
-constexpr uint32_t VGT_SHADER_STAGES_EN           = 0x2D5;
-constexpr uint32_t VGT_LS_HS_CONFIG               = 0x2D6;
-constexpr uint32_t VGT_TF_PARAM                   = 0x2DB;
-constexpr uint32_t DB_ALPHA_TO_MASK               = 0x2DC;
-constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL  = 0x2DE;
-constexpr uint32_t PA_SU_POLY_OFFSET_CLAMP        = 0x2DF;
-constexpr uint32_t PA_SU_POLY_OFFSET_FRONT_SCALE  = 0x2E0;
-constexpr uint32_t PA_SU_POLY_OFFSET_FRONT_OFFSET = 0x2E1;
-constexpr uint32_t PA_SU_POLY_OFFSET_BACK_SCALE   = 0x2E2;
-constexpr uint32_t PA_SU_POLY_OFFSET_BACK_OFFSET  = 0x2E3;
-constexpr uint32_t VGT_GS_INSTANCE_CNT            = 0x2E4;
-constexpr uint32_t PA_SC_CENTROID_PRIORITY_0      = 0x2F5;
-constexpr uint32_t PA_SC_CENTROID_PRIORITY_1      = 0x2F6;
+constexpr uint32_t DB_SRESULTS_COMPARE_STATE0                          = 0x2B0;
+constexpr uint32_t DB_SRESULTS_COMPARE_STATE1                          = 0x2B1;
+constexpr uint32_t VGT_GS_MAX_VERT_OUT                                 = 0x2CE;
+constexpr uint32_t GE_NGG_SUBGRP_CNTL                                  = 0x2D3;
+constexpr uint32_t VGT_TESS_DISTRIBUTION                               = 0x2D4;
+constexpr uint32_t VGT_SHADER_STAGES_EN                                = 0x2D5;
+constexpr uint32_t VGT_LS_HS_CONFIG                                    = 0x2D6;
+constexpr uint32_t VGT_TF_PARAM                                        = 0x2DB;
+constexpr uint32_t DB_ALPHA_TO_MASK                                    = 0x2DC;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL                       = 0x2DE;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL_NEG_NUM_DB_BITS_SHIFT = 0;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL_NEG_NUM_DB_BITS_MASK  = 0xFF;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL_DB_IS_FLOAT_FMT_SHIFT = 8;
+constexpr uint32_t PA_SU_POLY_OFFSET_DB_FMT_CNTL_DB_IS_FLOAT_FMT_MASK  = 0x1;
+constexpr uint32_t PA_SU_POLY_OFFSET_CLAMP                             = 0x2DF;
+constexpr uint32_t PA_SU_POLY_OFFSET_FRONT_SCALE                       = 0x2E0;
+constexpr uint32_t PA_SU_POLY_OFFSET_FRONT_OFFSET                      = 0x2E1;
+constexpr uint32_t PA_SU_POLY_OFFSET_BACK_SCALE                        = 0x2E2;
+constexpr uint32_t PA_SU_POLY_OFFSET_BACK_OFFSET                       = 0x2E3;
+constexpr uint32_t VGT_GS_INSTANCE_CNT                                 = 0x2E4;
+constexpr uint32_t PA_SC_CENTROID_PRIORITY_0                           = 0x2F5;
+constexpr uint32_t PA_SC_CENTROID_PRIORITY_1                           = 0x2F6;
 
 constexpr uint32_t PA_SC_AA_CONFIG                             = 0x2F8;
 constexpr uint32_t PA_SC_AA_CONFIG_MSAA_NUM_SAMPLES_SHIFT      = 0;
@@ -779,21 +783,21 @@ constexpr uint32_t SPI_SHADER_PGM_RSRC1_PS_FWD_PROGRESS_MASK      = 0x1;
 constexpr uint32_t SPI_SHADER_PGM_RSRC1_PS_FP16_OVFL_SHIFT        = 29;
 constexpr uint32_t SPI_SHADER_PGM_RSRC1_PS_FP16_OVFL_MASK         = 0x1;
 
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS                                = 0xB;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SCRATCH_EN_SHIFT               = 0;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SCRATCH_EN_MASK                = 0x1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_SHIFT                = 1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MASK                 = 0x1F;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_WAVE_CNT_EN_SHIFT              = 7;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_WAVE_CNT_EN_MASK               = 0x1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_EXTRA_LDS_SIZE_SHIFT           = 8;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_EXTRA_LDS_SIZE_MASK            = 0xFF;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_LOAD_INTRAWAVE_COLLISION_SHIFT = 26;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_LOAD_INTRAWAVE_COLLISION_MASK  = 0x1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MSB_SHIFT            = 27;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MSB_MASK             = 0x1;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SHARED_VGPR_CNT_SHIFT          = 28;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SHARED_VGPR_CNT_MASK           = 0xF;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS                              = 0xB;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SCRATCH_EN_SHIFT             = 0;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SCRATCH_EN_MASK              = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_SHIFT              = 1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MASK               = 0x1F;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_WAVE_CNT_EN_SHIFT            = 7;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_WAVE_CNT_EN_MASK             = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_EXTRA_LDS_SIZE_SHIFT         = 8;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_EXTRA_LDS_SIZE_MASK          = 0xFF;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_RASTER_ORDERED_SHADING_SHIFT = 25;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_RASTER_ORDERED_SHADING_MASK  = 0x3;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MSB_SHIFT          = 27;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_USER_SGPR_MSB_MASK           = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SHARED_VGPR_CNT_SHIFT        = 28;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_PS_SHARED_VGPR_CNT_MASK         = 0xF;
 
 constexpr uint32_t SPI_SHADER_USER_DATA_PS_0  = 0xC;
 constexpr uint32_t SPI_SHADER_USER_DATA_PS_15 = 0x1B;
@@ -886,61 +890,101 @@ constexpr uint32_t SPI_SHADER_PGM_RSRC2_GS_USER_SGPR_MSB_MASK     = 0x1;
 constexpr uint32_t SPI_SHADER_PGM_RSRC2_GS_SHARED_VGPR_CNT_SHIFT  = 28;
 constexpr uint32_t SPI_SHADER_PGM_RSRC2_GS_SHARED_VGPR_CNT_MASK   = 0xF;
 
-constexpr uint32_t SPI_SHADER_USER_DATA_GS_0       = 0x8C;
-constexpr uint32_t SPI_SHADER_USER_DATA_GS_15      = 0x9B;
-constexpr uint32_t SPI_SHADER_USER_DATA_GS_31      = 0xAB;
-constexpr uint32_t SPI_SHADER_REQ_CTRL_ESGS        = 0xB0;
-constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_0    = 0xB2;
-constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_3    = 0xB5;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_ES_GS      = 0xBC;
-constexpr uint32_t SPI_SHADER_PGM_LO_ES            = 0xC8;
-constexpr uint32_t SPI_SHADER_PGM_HI_ES            = 0xC9;
-constexpr uint32_t SPI_SHADER_PGM_RSRC1_ES         = 0xCA;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_ES         = 0xCB;
-constexpr uint32_t SPI_SHADER_USER_DATA_ES_0       = 0xCC;
-constexpr uint32_t SPI_SHADER_USER_DATA_ES_15      = 0xDB;
-constexpr uint32_t SPI_SHADER_USER_DATA_ES_31      = 0xEB;
-constexpr uint32_t SPI_SHADER_PGM_CHKSUM_HS        = 0x100;
-constexpr uint32_t SPI_SHADER_PGM_RSRC4_HS         = 0x101;
-constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_LO_HS = 0x102;
-constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_HI_HS = 0x103;
-constexpr uint32_t SPI_SHADER_PGM_RSRC3_HS         = 0x107;
-constexpr uint32_t SPI_SHADER_PGM_LO_HS            = 0x108;
-constexpr uint32_t SPI_SHADER_PGM_HI_HS            = 0x109;
-constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS         = 0x10A;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS         = 0x10B;
-constexpr uint32_t SPI_SHADER_USER_DATA_HS_0       = 0x10C;
-constexpr uint32_t SPI_SHADER_USER_DATA_HS_15      = 0x11B;
-constexpr uint32_t SPI_SHADER_USER_DATA_HS_31      = 0x12B;
-constexpr uint32_t SPI_SHADER_REQ_CTRL_LSHS        = 0x130;
-constexpr uint32_t SPI_SHADER_USER_ACCUM_LSHS_0    = 0x132;
-constexpr uint32_t SPI_SHADER_USER_ACCUM_LSHS_3    = 0x135;
-constexpr uint32_t SPI_SHADER_PGM_LO_LS            = 0x148;
-constexpr uint32_t SPI_SHADER_PGM_HI_LS            = 0x149;
-constexpr uint32_t SPI_SHADER_PGM_RSRC1_LS         = 0x14A;
-constexpr uint32_t SPI_SHADER_PGM_RSRC2_LS         = 0x14B;
-constexpr uint32_t COMPUTE_START_X                 = 0x204;
-constexpr uint32_t COMPUTE_START_Y                 = 0x205;
-constexpr uint32_t COMPUTE_START_Z                 = 0x206;
-constexpr uint32_t COMPUTE_NUM_THREAD_X            = 0x207;
-constexpr uint32_t COMPUTE_NUM_THREAD_Y            = 0x208;
-constexpr uint32_t COMPUTE_NUM_THREAD_Z            = 0x209;
-constexpr uint32_t COMPUTE_PGM_LO                  = 0x20C;
-constexpr uint32_t COMPUTE_PGM_HI                  = 0x20D;
-constexpr uint32_t COMPUTE_TBA_LO                  = 0x20E;
-constexpr uint32_t COMPUTE_TBA_HI                  = 0x20F;
-constexpr uint32_t COMPUTE_TMA_LO                  = 0x210;
-constexpr uint32_t COMPUTE_TMA_HI                  = 0x211;
+constexpr uint32_t SPI_SHADER_USER_DATA_GS_0                      = 0x8C;
+constexpr uint32_t SPI_SHADER_USER_DATA_GS_15                     = 0x9B;
+constexpr uint32_t SPI_SHADER_USER_DATA_GS_31                     = 0xAB;
+constexpr uint32_t SPI_SHADER_REQ_CTRL_ESGS                       = 0xB0;
+constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_0                   = 0xB2;
+constexpr uint32_t SPI_SHADER_USER_ACCUM_ESGS_3                   = 0xB5;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_ES_GS                     = 0xBC;
+constexpr uint32_t SPI_SHADER_PGM_LO_ES                           = 0xC8;
+constexpr uint32_t SPI_SHADER_PGM_HI_ES                           = 0xC9;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_ES                        = 0xCA;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_ES                        = 0xCB;
+constexpr uint32_t SPI_SHADER_USER_DATA_ES_0                      = 0xCC;
+constexpr uint32_t SPI_SHADER_USER_DATA_ES_15                     = 0xDB;
+constexpr uint32_t SPI_SHADER_USER_DATA_ES_31                     = 0xEB;
+constexpr uint32_t SPI_SHADER_PGM_CHKSUM_HS                       = 0x100;
+constexpr uint32_t SPI_SHADER_PGM_RSRC4_HS                        = 0x101;
+constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_LO_HS                = 0x102;
+constexpr uint32_t SPI_SHADER_USER_DATA_ADDR_HI_HS                = 0x103;
+constexpr uint32_t SPI_SHADER_PGM_RSRC3_HS                        = 0x107;
+constexpr uint32_t SPI_SHADER_PGM_LO_HS                           = 0x108;
+constexpr uint32_t SPI_SHADER_PGM_HI_HS                           = 0x109;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS                        = 0x10A;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS                        = 0x10B;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_VGPRS_SHIFT            = 0;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_VGPRS_MASK             = 0x3F;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_PRIORITY_SHIFT         = 10;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_PRIORITY_MASK          = 0x3;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FLOAT_MODE_SHIFT       = 12;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FLOAT_MODE_MASK        = 0xFF;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_DX10_CLAMP_SHIFT       = 21;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_DX10_CLAMP_MASK        = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_DEBUG_MODE_SHIFT       = 22;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_DEBUG_MODE_MASK        = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_IEEE_MODE_SHIFT        = 23;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_IEEE_MODE_MASK         = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FWD_PROGRESS_SHIFT     = 25;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FWD_PROGRESS_MASK      = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_WGP_MODE_SHIFT         = 26;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_WGP_MODE_MASK          = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_LS_VGPR_COMP_CNT_SHIFT = 28;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_LS_VGPR_COMP_CNT_MASK  = 0x3;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FP16_OVFL_SHIFT        = 30;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_HS_FP16_OVFL_MASK         = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_SCRATCH_EN_SHIFT       = 0;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_SCRATCH_EN_MASK        = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_USER_SGPR_SHIFT        = 1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_USER_SGPR_MASK         = 0x1F;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_LDS_SIZE_SHIFT         = 18;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_LDS_SIZE_MASK          = 0x1FF;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_USER_SGPR_MSB_SHIFT    = 27;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_USER_SGPR_MSB_MASK     = 0x1;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_SHARED_VGPR_CNT_SHIFT  = 28;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_HS_SHARED_VGPR_CNT_MASK   = 0xF;
+constexpr uint32_t SPI_SHADER_USER_DATA_HS_0                      = 0x10C;
+constexpr uint32_t SPI_SHADER_USER_DATA_HS_15                     = 0x11B;
+constexpr uint32_t SPI_SHADER_USER_DATA_HS_31                     = 0x12B;
+constexpr uint32_t SPI_SHADER_REQ_CTRL_LSHS                       = 0x130;
+constexpr uint32_t SPI_SHADER_USER_ACCUM_LSHS_0                   = 0x132;
+constexpr uint32_t SPI_SHADER_USER_ACCUM_LSHS_3                   = 0x135;
+constexpr uint32_t SPI_SHADER_PGM_LO_LS                           = 0x148;
+constexpr uint32_t SPI_SHADER_PGM_HI_LS                           = 0x149;
+constexpr uint32_t SPI_SHADER_PGM_RSRC1_LS                        = 0x14A;
+constexpr uint32_t SPI_SHADER_PGM_RSRC2_LS                        = 0x14B;
+constexpr uint32_t COMPUTE_START_X                                = 0x204;
+constexpr uint32_t COMPUTE_START_Y                                = 0x205;
+constexpr uint32_t COMPUTE_START_Z                                = 0x206;
+constexpr uint32_t COMPUTE_NUM_THREAD_X                           = 0x207;
+constexpr uint32_t COMPUTE_NUM_THREAD_Y                           = 0x208;
+constexpr uint32_t COMPUTE_NUM_THREAD_Z                           = 0x209;
+constexpr uint32_t COMPUTE_PGM_LO                                 = 0x20C;
+constexpr uint32_t COMPUTE_PGM_HI                                 = 0x20D;
+constexpr uint32_t COMPUTE_TBA_LO                                 = 0x20E;
+constexpr uint32_t COMPUTE_TBA_HI                                 = 0x20F;
+constexpr uint32_t COMPUTE_TMA_LO                                 = 0x210;
+constexpr uint32_t COMPUTE_TMA_HI                                 = 0x211;
 
-constexpr uint32_t COMPUTE_PGM_RSRC1              = 0x212;
-constexpr uint32_t COMPUTE_PGM_RSRC1_VGPRS_SHIFT  = 0;
-constexpr uint32_t COMPUTE_PGM_RSRC1_VGPRS_MASK   = 0x3F;
-constexpr uint32_t COMPUTE_PGM_RSRC1_SGPRS_SHIFT  = 6;
-constexpr uint32_t COMPUTE_PGM_RSRC1_SGPRS_MASK   = 0xF;
-constexpr uint32_t COMPUTE_PGM_RSRC1_BULKY_SHIFT  = 24;
-constexpr uint32_t COMPUTE_PGM_RSRC1_BULKY_MASK   = 0x1;
-constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_SHIFT = 30;
-constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_MASK  = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1                  = 0x212;
+constexpr uint32_t COMPUTE_PGM_RSRC1_VGPRS_SHIFT      = 0;
+constexpr uint32_t COMPUTE_PGM_RSRC1_VGPRS_MASK       = 0x3F;
+constexpr uint32_t COMPUTE_PGM_RSRC1_SGPRS_SHIFT      = 6;
+constexpr uint32_t COMPUTE_PGM_RSRC1_SGPRS_MASK       = 0xF;
+constexpr uint32_t COMPUTE_PGM_RSRC1_FLOAT_MODE_SHIFT = 12;
+constexpr uint32_t COMPUTE_PGM_RSRC1_FLOAT_MODE_MASK  = 0xFF;
+constexpr uint32_t COMPUTE_PGM_RSRC1_DX10_CLAMP_SHIFT = 21;
+constexpr uint32_t COMPUTE_PGM_RSRC1_DX10_CLAMP_MASK  = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_IEEE_MODE_SHIFT  = 23;
+constexpr uint32_t COMPUTE_PGM_RSRC1_IEEE_MODE_MASK   = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_BULKY_SHIFT      = 24;
+constexpr uint32_t COMPUTE_PGM_RSRC1_BULKY_MASK       = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_FP16_OVFL_SHIFT  = 26;
+constexpr uint32_t COMPUTE_PGM_RSRC1_FP16_OVFL_MASK   = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_WGP_MODE_SHIFT   = 29;
+constexpr uint32_t COMPUTE_PGM_RSRC1_WGP_MODE_MASK    = 0x1;
+constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_SHIFT     = 30;
+constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_MASK      = 0x1;
 
 constexpr uint32_t COMPUTE_PGM_RSRC2                      = 0x213;
 constexpr uint32_t COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT     = 0;

src/graphics/guest_gpu/tile.h

@@ -39,7 +39,7 @@ enum class TileBlockFamily : uint32_t {
 };
 
 struct TileBlockLayout {
-	TileBlockFamily family            = TileBlockFamily::Standard256B;
+	TileBlockFamily family            = TileBlockFamily::Count;
 	uint32_t        bytes_per_element = 0;
 	uint32_t        block_size        = 0;
 	uint32_t        block_width       = 0;
@@ -47,38 +47,69 @@ struct TileBlockLayout {
 	uint32_t        block_depth       = 0;
 };
 
-struct TileVolumeLayout {
-	TileBlockFamily family            = TileBlockFamily::Count;
-	uint32_t        bytes_per_element = 0;
-	uint32_t        texel_width       = 1;
-	uint32_t        texel_height      = 1;
-	uint32_t        first_tail_level  = 0;
-	uint32_t        block_depth       = 1;
-	uint64_t        block_slice_size  = 0;
-	uint64_t        total_size        = 0;
-	uint64_t        level_offsets[16] = {};
-	uint64_t        level_sizes[16]   = {};
-	uint32_t        tail_x[16]        = {};
-	uint32_t        tail_y[16]        = {};
-	uint32_t        level_widths[16]  = {};
-	uint32_t        level_heights[16] = {};
+struct TileTextureElementLayout {
+	uint32_t bytes        = 0;
+	uint32_t texel_width  = 1;
+	uint32_t texel_height = 1;
+};
+
+struct TileTextureBlockLayout {
+	TileBlockLayout block;
+	uint32_t        texel_width  = 1;
+	uint32_t        texel_height = 1;
+};
+
+enum class TileSurfaceDimension : uint32_t {
+	Dim2D,
+	Dim3D,
+};
+
+struct TileSurfaceDescription {
+	uint32_t             format    = 0;
+	uint32_t             tile_mode = 0;
+	TileSurfaceDimension dimension = TileSurfaceDimension::Dim2D;
+	uint32_t             width     = 0;
+	uint32_t             height    = 0;
+	uint32_t             depth     = 1;
+	uint32_t             levels    = 1;
+	uint32_t             layers    = 1;
+};
+
+// Width/height fields are measured in addressable elements, not texture texels.
+struct TileMipLayout {
+	uint64_t offset        = 0;
+	uint64_t size          = 0;
+	uint32_t width         = 0;
+	uint32_t height        = 0;
+	uint32_t padded_width  = 0;
+	uint32_t padded_height = 0;
+	uint32_t tail_x        = 0;
+	uint32_t tail_y        = 0;
+};
+
+struct TileSurfaceLayout {
+	TileSurfaceDescription description;
+	TileTextureBlockLayout texture;
+	uint32_t               first_tail_level = 0;
+	uint64_t               block_slice_size = 0;
+	uint64_t               total_size       = 0;
+	TileMipLayout          mips[16]         = {};
 };
 
 bool TileGetBlockLayout(TileBlockFamily family, uint32_t bytes_per_element,
                         TileBlockLayout& layout);
+bool TileGetTextureElementLayout(uint32_t format, TileTextureElementLayout& layout);
+bool TileGetTextureBlockLayout(uint32_t format, uint32_t tile, bool volume,
+                               TileTextureBlockLayout& layout);
+bool TileGetTiledTextureLayout(const TileSurfaceDescription& description,
+                               TileSurfaceLayout&            layout);
 bool TileGetBlockOffset(const TileBlockLayout& layout, uint32_t x, uint32_t y, uint32_t z,
                         uint32_t& byte_offset);
 bool TileGetBlockXor(const TileBlockLayout& layout, uint32_t block_x, uint32_t block_y,
                      uint32_t& byte_offset);
 bool TileGetBlockXor(const TileBlockLayout& layout, uint32_t block_x, uint32_t block_y,
                      uint32_t block_z, uint32_t& byte_offset);
-bool TileIsStandard256BTextureSupported(uint32_t format);
-bool TileIsStandard4KBTextureSupported(uint32_t format);
-bool TileIsStandard64KBTextureSupported(uint32_t format);
-bool TileGetTextureVolumeLayout(uint32_t format, uint32_t width, uint32_t height, uint32_t depth,
-                                uint32_t levels, uint32_t tile, TileVolumeLayout& layout);
 
-bool     TileGetHtileSize(uint32_t width, uint32_t height, TileSizeAlign& htile_size);
 bool     TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch, uint32_t z_format,
                           uint32_t stencil_format, bool htile, TileSizeAlign& stencil_size,
                           TileSizeAlign& htile_size, TileSizeAlign& depth_size,
@@ -94,13 +125,13 @@ bool     TileGetRenderTargetMipLayout(uint32_t width, uint32_t height, uint32_t
                                       uint32_t bytes_per_element, uint32_t levels,
                                       TileSizeAlign& total_size, TileSizeOffset* level_sizes,
                                       TilePaddedSize* padded_size);
-void     TileGetTextureSize(uint32_t format, uint32_t width, uint32_t height, uint32_t pitch,
-                            uint32_t levels, uint32_t tile, TileSizeAlign* total_size,
-                            TileSizeOffset* level_sizes, TilePaddedSize* padded_size);
-void TileGetTextureTotalSize(uint32_t format, uint32_t width, uint32_t height, uint32_t depth,
-                             uint32_t pitch, uint32_t levels, uint32_t tile, bool volume_texture,
-                             TileSizeAlign& total_size);
-uint32_t TileGetTexturePitch(uint32_t format, uint32_t width, uint32_t levels, uint32_t tile);
+void     TileGetTextureSize(uint32_t format, uint32_t width, uint32_t height, uint32_t levels,
+                            uint32_t tile, TileSizeAlign* total_size, TileSizeOffset* level_sizes,
+                            TilePaddedSize* padded_size);
+void     TileGetTextureTotalSize(uint32_t format, uint32_t width, uint32_t height, uint32_t depth,
+                                 uint32_t levels, uint32_t tile, bool volume_texture,
+                                 TileSizeAlign& total_size);
+uint32_t TileGetTexturePitch(uint32_t format, uint32_t width, uint32_t tile);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/memoryTracker.cpp

@@ -79,7 +79,6 @@ RegionManager* MemoryTracker::GetOrCreateRegion(uint64_t index) {
 
 bool MemoryTracker::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	return Iterate<true>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		return manager->IsModified<DirtySource::Cpu>(offset, bytes);
@@ -88,7 +87,6 @@ bool MemoryTracker::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 
 bool MemoryTracker::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	return Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		return manager->IsModified<DirtySource::Gpu>(offset, bytes);
@@ -97,7 +95,6 @@ bool MemoryTracker::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
 
 void MemoryTracker::MarkRegionAsCpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	Iterate<true>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		manager->ChangeState<DirtySource::Cpu, true>(manager->GetCpuAddr() + offset, bytes);
@@ -106,7 +103,6 @@ void MemoryTracker::MarkRegionAsCpuModified(uint64_t vaddr, uint64_t size) {
 
 void MemoryTracker::MarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	Iterate<true>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		manager->ChangeState<DirtySource::Gpu, true>(manager->GetCpuAddr() + offset, bytes);
@@ -115,14 +111,13 @@ void MemoryTracker::MarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 
 void MemoryTracker::UnmarkRegionAsGpuModified(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
 	Iterate<false>(vaddr, size, [](RegionManager* manager, uint64_t offset, uint64_t bytes) {
 		std::scoped_lock lock(manager->lock);
 		manager->ChangeState<DirtySource::Gpu, false>(manager->GetCpuAddr() + offset, bytes);
 	});
 }
 
-void MemoryTracker::UntrackMemoryLocked(uint64_t vaddr, uint64_t size) {
+void MemoryTracker::UntrackMemoryImpl(uint64_t vaddr, uint64_t size) {
 	std::vector<RegionManager*> managers;
 	managers.reserve((vaddr % TRACKER_REGION_SIZE + size + TRACKER_REGION_SIZE - 1) /
 	                 TRACKER_REGION_SIZE);
@@ -148,8 +143,7 @@ void MemoryTracker::UntrackMemoryLocked(uint64_t vaddr, uint64_t size) {
 
 void MemoryTracker::UntrackMemory(uint64_t vaddr, uint64_t size) {
 	CheckNotInUploadCallback();
-	std::lock_guard access(m_access_mutex);
-	UntrackMemoryLocked(vaddr, size);
+	UntrackMemoryImpl(vaddr, size);
 }
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/memoryTracker.h

@@ -68,7 +68,6 @@ public:
 		static_assert(std::is_nothrow_invocable_v<Preflight&, uint64_t, uint64_t>);
 		static_assert(std::is_nothrow_invocable_v<Func&, uint64_t, uint64_t>);
 		CheckNotInUploadCallback();
-		std::lock_guard             access(m_access_mutex);
 		std::vector<RegionManager*> managers;
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t, uint64_t) {
 			managers.push_back(manager);
@@ -109,7 +108,6 @@ public:
 		static_assert(std::is_nothrow_invocable_v<RangeFunc&, uint64_t, uint64_t>);
 		static_assert(std::is_nothrow_invocable_v<UploadFunc&>);
 		CheckNotInUploadCallback();
-		std::unique_lock access(m_access_mutex);
 		Iterate<true>(vaddr, size, [](RegionManager*, uint64_t, uint64_t) {});
 		const auto* previous_upload_owner = std::exchange(s_upload_owner, this);
 		Iterate<false>(vaddr, size, [&](RegionManager* manager, uint64_t offset, uint64_t bytes) {
@@ -173,13 +171,12 @@ private:
 	}
 
 	static void    ValidateRange(uint64_t vaddr, uint64_t size);
-	void           UntrackMemoryLocked(uint64_t vaddr, uint64_t size);
+	void           UntrackMemoryImpl(uint64_t vaddr, uint64_t size);
 	RegionManager* GetOrCreateRegion(uint64_t index);
 
 	std::unique_ptr<std::atomic<RegionManager*>[]> m_regions;
 	std::vector<std::unique_ptr<RegionManager>>    m_region_storage;
 	std::mutex                                     m_region_mutex;
-	std::mutex                                     m_access_mutex;
 	PageManager&                                   m_page_manager;
 };
 

src/graphics/host_gpu/renderer/cache/bufferCache.cpp

@@ -5,7 +5,6 @@
 #include "common/profiler.h"
 #include "graphics/guest_gpu/graphicsRun.h"
 #include "graphics/host_gpu/graphicContext.h"
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
 #include "graphics/host_gpu/renderer/render.h"
@@ -23,30 +22,9 @@ namespace Libs::Graphics {
 
 namespace {
 
-thread_local const void* g_cache_lock_owner = nullptr;
-
 constexpr uint64_t MiB           = 1024 * 1024;
 constexpr uint64_t GdsBufferSize = 64 * 1024;
 
-class FaultSafeCacheLock final {
-public:
-	FaultSafeCacheLock(const void* owner, Common::Mutex& mutex): m_mutex(mutex) {
-		if (g_cache_lock_owner != nullptr) {
-			EXIT("BufferCache: recursive cache lock acquisition\n");
-		}
-		g_cache_lock_owner = owner;
-		m_mutex.Lock();
-	}
-
-	~FaultSafeCacheLock() {
-		m_mutex.Unlock();
-		g_cache_lock_owner = nullptr;
-	}
-
-private:
-	Common::Mutex& m_mutex;
-};
-
 } // namespace
 
 uint64_t BufferCache::AlignDown(uint64_t value) noexcept {
@@ -60,15 +38,6 @@ uint64_t BufferCache::AlignUp(uint64_t value) {
 	return (value + CACHING_PAGE_SIZE - 1) & ~(CACHING_PAGE_SIZE - 1);
 }
 
-bool BufferCache::PageOverlaps(uint64_t left, uint64_t left_size, uint64_t right,
-                               uint64_t right_size) noexcept {
-	const auto left_begin  = left & ~(TRACKER_PAGE_SIZE - 1);
-	const auto left_end    = (left + left_size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
-	const auto right_begin = right & ~(TRACKER_PAGE_SIZE - 1);
-	const auto right_end = (right + right_size + TRACKER_PAGE_SIZE - 1) & ~(TRACKER_PAGE_SIZE - 1);
-	return left_begin < right_end && right_begin < left_end;
-}
-
 void BufferCache::Upload(CommandBuffer& command, Buffer& destination, uint64_t destination_offset,
                          const void* source, uint64_t size) {
 	auto* bytes = static_cast<const uint8_t*>(source);
@@ -201,27 +170,23 @@ void BufferCache::QueueGarbageDownload(std::span<const DownloadCopy> copies, Ret
 	m_scheduler.DeferOperation(
 	    [this, downloads = std::move(downloads), retire = std::move(retire)]() mutable {
 		    PublishDownloads(downloads);
-		    {
-			    FaultSafeCacheLock lock(this, m_mutex);
-			    for (const auto& range: downloads) {
-				    m_gpu_modified_ranges.Subtract(range.address, range.size);
-			    }
-			    // ForEachDownloadRange reports full tracker pages, and every exact GPU-owned
-			    // interval on those pages was downloaded and removed. Clearing the original
-			    // query therefore cannot orphan a dirty sibling on an edge page.
-			    m_memory_tracker.UnmarkRegionAsGpuModified(retire.address, retire.size);
-			    if (m_memory_tracker.IsRegionGpuModified(retire.address, retire.size) ||
-			        !m_gpu_modified_ranges.Intersections(retire.address, retire.size).empty()) {
-				    EXIT("BufferCache: asynchronous garbage collection retained GPU ownership\n");
-			    }
-			    m_memory_tracker.UntrackMemory(retire.address, retire.size);
+		    for (const auto& range: downloads) {
+			    m_gpu_modified_ranges.Subtract(range.address, range.size);
 		    }
+		    // ForEachDownloadRange reports full tracker pages, and every exact GPU-owned
+		    // interval on those pages was downloaded and removed. Clearing the original
+		    // query therefore cannot orphan a dirty sibling on an edge page.
+		    m_memory_tracker.UnmarkRegionAsGpuModified(retire.address, retire.size);
+		    if (m_memory_tracker.IsRegionGpuModified(retire.address, retire.size) ||
+		        !m_gpu_modified_ranges.Intersections(retire.address, retire.size).empty()) {
+			    EXIT("BufferCache: asynchronous garbage collection retained GPU ownership\n");
+		    }
+		    m_memory_tracker.UntrackMemory(retire.address, retire.size);
 	    });
 }
 
 BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
-                         PageManager& page_manager, TextureCache& texture_cache,
-                         ResourceMutex& resource_mutex)
+                         PageManager& page_manager, TextureCache& texture_cache)
     : m_graphics(graphics), m_scheduler(scheduler),
       m_gds_buffer(graphics, scheduler, MemoryUsage::Stream, 0, AllFlags, GdsBufferSize),
       m_memory_tracker(page_manager),
@@ -229,7 +194,7 @@ BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
       m_stream_buffer(graphics, scheduler, MemoryUsage::Stream, 64 * MiB),
       m_download_buffer(graphics, scheduler, MemoryUsage::Download, 32 * MiB),
       m_device_buffer(graphics, scheduler, MemoryUsage::DeviceLocal, 128 * MiB),
-      m_texture_cache(texture_cache), m_resource_mutex(resource_mutex) {
+      m_texture_cache(texture_cache) {
 	std::memset(m_gds_buffer.Mapped().data(), 0, static_cast<size_t>(m_gds_buffer.Size()));
 	m_gds_buffer.Flush(0, m_gds_buffer.Size());
 	if (!m_graphics.CanReportMemoryUsage()) {
@@ -287,9 +252,6 @@ void BufferCache::InvalidateMemory(uint64_t vaddr, uint64_t size) {
 	    size > TRACKER_ADDRESS_SIZE - vaddr) {
 		EXIT("BufferCache: invalid memory-invalidation range\n");
 	}
-	if (!HasPageOverlap(vaddr, size)) {
-		return;
-	}
 	m_memory_tracker.InvalidateRegion(vaddr, size,
 	                                  [this, vaddr, size] { ReadMemory(vaddr, size, true); });
 }
@@ -304,70 +266,63 @@ void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size, bool is_write) {
 		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
 		     vaddr, size);
 	}
-	if (m_resource_mutex.IsOwnedByCurrentThread()) {
-		EXIT("unsupported buffer readback from a pre-owned resource transaction, "
-		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
-		     vaddr, size);
-	}
 	m_scheduler.Context().GetGpu().SendCommandSync(
 	    [this, vaddr, size, is_write] { ReadMemoryOnGpu(vaddr, size, is_write); });
 }
 
 void BufferCache::ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write) {
+	// CPU invalidation reaches this point only for a GPU-owned tracker page. Resolve the exact
+	// Buffer owner on the GPU thread so the cache index remains single-thread-owned.
+	if (is_write && !IsRegionRegistered(vaddr, size)) {
+		return;
+	}
 	std::vector<DownloadCopy> copies;
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		m_memory_tracker.ForEachDownloadRange<false>(
-		    vaddr, size,
-		    [&](uint64_t address, uint64_t bytes) noexcept {
-			    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
-			                                           "memory invalidation");
-		    },
-		    [&](uint64_t address, uint64_t bytes) noexcept {
-			    for (const auto range: m_gpu_modified_ranges.Intersections(address, bytes)) {
-				    for (uint64_t copied = 0; copied < range.size;) {
-					    const auto copy_address = range.address + copied;
-					    auto       owner        = m_buffers.upper_bound(copy_address);
-					    if (owner == m_buffers.begin()) {
-						    EXIT("BufferCache: invalidation readback has no buffer owner\n");
-					    }
-					    auto& cached = *std::prev(owner)->second;
-					    if (!cached.buffer->IsInBounds(copy_address, 1)) {
-						    EXIT(
-						        "BufferCache: invalidation readback is outside its buffer owner\n");
-					    }
-					    const auto copy_size = std::min(range.size - copied,
-					                                    cached.vaddr + cached.size - copy_address);
-					    copies.push_back({cached.buffer, cached.buffer->Offset(copy_address),
-					                      copy_address, copy_size});
-					    copied += copy_size;
+	m_memory_tracker.ForEachDownloadRange<false>(
+	    vaddr, size,
+	    [&](uint64_t address, uint64_t bytes) noexcept {
+		    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
+		                                           "memory invalidation");
+	    },
+	    [&](uint64_t address, uint64_t bytes) noexcept {
+		    for (const auto range: m_gpu_modified_ranges.Intersections(address, bytes)) {
+			    for (uint64_t copied = 0; copied < range.size;) {
+				    const auto copy_address = range.address + copied;
+				    auto       owner        = m_buffers.upper_bound(copy_address);
+				    if (owner == m_buffers.begin()) {
+					    EXIT("BufferCache: invalidation readback has no buffer owner\n");
 				    }
+				    auto& cached = *std::prev(owner)->second;
+				    if (!cached.buffer->IsInBounds(copy_address, 1)) {
+					    EXIT("BufferCache: invalidation readback is outside its buffer owner\n");
+				    }
+				    const auto copy_size =
+				        std::min(range.size - copied, cached.vaddr + cached.size - copy_address);
+				    copies.push_back({cached.buffer, cached.buffer->Offset(copy_address),
+				                      copy_address, copy_size});
+				    copied += copy_size;
 			    }
-		    });
-		if (copies.empty()) {
-			if (!is_write) {
-				return;
-			}
-			// A preceding read fault can consume the last GPU-owned copy after this write
-			// invalidation has already chosen to flush. Complete the CPU ownership handoff even
-			// though this callback no longer has bytes to download.
-			m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
+		    }
+	    });
+	if (copies.empty()) {
+		if (!is_write) {
 			return;
 		}
+		// A preceding read fault can consume the last GPU-owned copy after this write invalidation
+		// has already chosen to flush. Complete the CPU ownership handoff even though this callback
+		// no longer has bytes to download.
+		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
+		return;
 	}
 	auto downloads = RecordDownloads(copies);
 	m_scheduler.FinishCurrent();
 	PublishDownloads(downloads);
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		for (const auto& range: downloads) {
-			m_gpu_modified_ranges.Subtract(range.address, range.size);
-		}
-		// The enumeration above covered whole dirty pages and every exact interval on them.
-		m_memory_tracker.UnmarkRegionAsGpuModified(vaddr, size);
-		if (is_write) {
-			m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
-		}
+	for (const auto& range: downloads) {
+		m_gpu_modified_ranges.Subtract(range.address, range.size);
+	}
+	// The enumeration above covered whole dirty pages and every exact interval on them.
+	m_memory_tracker.UnmarkRegionAsGpuModified(vaddr, size);
+	if (is_write) {
+		m_memory_tracker.MarkRegionAsCpuModified(vaddr, size);
 	}
 }
 
@@ -378,43 +333,40 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 	std::vector<DownloadCopy>                  copies;
 	std::vector<std::pair<uint64_t, uint64_t>> modified_buffers;
 	std::vector<std::pair<uint64_t, uint64_t>> retired_buffers;
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		for (const auto& [begin, cached]: m_buffers) {
-			if (vaddr < begin + cached->size && begin < vaddr + size) {
-				retired_buffers.emplace_back(begin, cached->size);
-			}
+	for (const auto& [begin, cached]: m_buffers) {
+		if (vaddr < begin + cached->size && begin < vaddr + size) {
+			retired_buffers.emplace_back(begin, cached->size);
 		}
-		for (const auto& [begin, cached]: m_buffers) {
-			if (vaddr >= begin + cached->size || begin >= vaddr + size ||
-			    !m_memory_tracker.IsRegionGpuModified(begin, cached->size)) {
-				continue;
-			}
-			const auto dirty = m_gpu_modified_ranges.Intersections(begin, cached->size);
-			if (dirty.empty()) {
-				EXIT("BufferCache: GPU-modified buffer has no dirty ranges\n");
-			}
-			modified_buffers.emplace_back(begin, cached->size);
+	}
+	for (const auto& [begin, cached]: m_buffers) {
+		if (vaddr >= begin + cached->size || begin >= vaddr + size ||
+		    !m_memory_tracker.IsRegionGpuModified(begin, cached->size)) {
+			continue;
 		}
-		for (const auto& [begin, bytes]: modified_buffers) {
-			auto owner = m_buffers.find(begin);
-			if (owner == m_buffers.end() || owner->second->size != bytes) {
-				EXIT("BufferCache: unmap owner changed during collection\n");
-			}
-			auto& cached = *owner->second;
-			m_memory_tracker.ForEachDownloadRange<false>(
-			    begin, cached.size,
-			    [&](uint64_t address, uint64_t bytes) noexcept {
-				    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
-				                                           "unmap");
-			    },
-			    [&](uint64_t address, uint64_t bytes) noexcept {
-				    for (const auto& range: m_gpu_modified_ranges.Intersections(address, bytes)) {
-					    copies.push_back(
-					        {cached.buffer, range.address - begin, range.address, range.size});
-				    }
-			    });
+		const auto dirty = m_gpu_modified_ranges.Intersections(begin, cached->size);
+		if (dirty.empty()) {
+			EXIT("BufferCache: GPU-modified buffer has no dirty ranges\n");
 		}
+		modified_buffers.emplace_back(begin, cached->size);
+	}
+	for (const auto& [begin, bytes]: modified_buffers) {
+		auto owner = m_buffers.find(begin);
+		if (owner == m_buffers.end() || owner->second->size != bytes) {
+			EXIT("BufferCache: unmap owner changed during collection\n");
+		}
+		auto& cached = *owner->second;
+		m_memory_tracker.ForEachDownloadRange<false>(
+		    begin, cached.size,
+		    [&](uint64_t address, uint64_t bytes) noexcept {
+			    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
+			                                           "unmap");
+		    },
+		    [&](uint64_t address, uint64_t bytes) noexcept {
+			    for (const auto& range: m_gpu_modified_ranges.Intersections(address, bytes)) {
+				    copies.push_back(
+				        {cached.buffer, range.address - begin, range.address, range.size});
+			    }
+		    });
 	}
 	if (!copies.empty()) {
 		auto downloads = RecordDownloads(copies);
@@ -425,29 +377,26 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		// command stream before removing such backing.
 		m_scheduler.FinishCurrent();
 	}
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		for (const auto& [begin, bytes]: modified_buffers) {
-			m_gpu_modified_ranges.Subtract(begin, bytes);
-			m_memory_tracker.UnmarkRegionAsGpuModified(begin, bytes);
-		}
-		for (const auto& [begin, bytes]: retired_buffers) {
-			m_memory_tracker.MarkRegionAsCpuModified(begin, bytes);
-		}
-		if (!m_gpu_modified_ranges.Intersections(vaddr, size).empty()) {
-			EXIT("BufferCache: unmap retained dirty byte ranges\n");
-		}
-		m_memory_tracker.UntrackMemory(vaddr, size);
-		for (auto it = m_buffers.begin(); it != m_buffers.end();) {
-			if (vaddr < it->first + it->second->size && it->first < vaddr + size) {
-				if (it->second->size > m_total_used_memory) {
-					EXIT("BufferCache: allocation accounting underflow\n");
-				}
-				m_total_used_memory -= it->second->size;
-				it = m_buffers.erase(it);
-			} else {
-				++it;
+	for (const auto& [begin, bytes]: modified_buffers) {
+		m_gpu_modified_ranges.Subtract(begin, bytes);
+		m_memory_tracker.UnmarkRegionAsGpuModified(begin, bytes);
+	}
+	for (const auto& [begin, bytes]: retired_buffers) {
+		m_memory_tracker.MarkRegionAsCpuModified(begin, bytes);
+	}
+	if (!m_gpu_modified_ranges.Intersections(vaddr, size).empty()) {
+		EXIT("BufferCache: unmap retained dirty byte ranges\n");
+	}
+	m_memory_tracker.UntrackMemory(vaddr, size);
+	for (auto it = m_buffers.begin(); it != m_buffers.end();) {
+		if (vaddr < it->first + it->second->size && it->first < vaddr + size) {
+			if (it->second->size > m_total_used_memory) {
+				EXIT("BufferCache: allocation accounting underflow\n");
 			}
+			m_total_used_memory -= it->second->size;
+			it = m_buffers.erase(it);
+		} else {
+			++it;
 		}
 	}
 }
@@ -529,7 +478,6 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 	if (command.IsInvalid() || command.IsExecute()) {
 		EXIT("BufferCache: buffer request requires a recording command buffer\n");
 	}
-	std::lock_guard transaction(m_resource_mutex);
 
 	if (is_read && !is_written && size <= CACHING_PAGE_SIZE &&
 	    !m_memory_tracker.IsRegionGpuModified(vaddr, size) &&
@@ -556,7 +504,6 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 		(void)m_texture_cache.InvalidateMemoryFromGPU(vaddr, size, true);
 	}
 
-	FaultSafeCacheLock                         lock(this, m_mutex);
 	auto&                                      cached = GetOrCreateBuffer(command, vaddr, size);
 	std::vector<std::pair<uint64_t, uint64_t>> uploads;
 	m_memory_tracker.ForEachUploadRange(
@@ -571,26 +518,23 @@ BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr,
 	if (is_written) {
 		m_gpu_modified_ranges.Add(vaddr, size);
 	}
+	auto     owner  = cached.buffer;
+	uint64_t offset = cached.buffer->Offset(vaddr);
 	if (is_formatted && is_read && !is_written) {
-		(void)SynchronizeBufferFromImage(*cached.buffer, vaddr, size);
+		(void)SynchronizeBufferFromImage(*owner, vaddr, size);
 	}
-	return {cached.buffer, cached.buffer->Handle(), cached.buffer->Offset(vaddr)};
+	return {owner, owner->Handle(), offset};
 }
 
 std::shared_ptr<Buffer> BufferCache::ObtainNullBuffer() {
-	std::shared_ptr<Buffer> buffer;
-	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		if (m_null_buffer != nullptr) {
-			return m_null_buffer;
-		}
-		m_null_buffer = std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::DeviceLocal,
-		                                         0, AllFlags, 16);
-		buffer        = m_null_buffer;
+	if (m_null_buffer != nullptr) {
+		return m_null_buffer;
 	}
+	m_null_buffer = std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::DeviceLocal, 0,
+	                                         AllFlags, 16);
 	const std::array<uint8_t, 16> zeros {};
-	Upload(m_scheduler.Current(), *buffer, 0, zeros.data(), zeros.size());
-	return buffer;
+	Upload(m_scheduler.Current(), *m_null_buffer, 0, zeros.data(), zeros.size());
+	return m_null_buffer;
 }
 
 ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t size) {
@@ -608,11 +552,10 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 	};
 
 	{
-		FaultSafeCacheLock lock(this, m_mutex);
-		const bool         cpu_modified = m_memory_tracker.IsRegionCpuModified(vaddr, size);
-		const bool         gpu_modified = m_memory_tracker.IsRegionGpuModified(vaddr, size);
-		const auto         dirty        = m_gpu_modified_ranges.Intersections(vaddr, size);
-		const bool         has_dirty_buffer_source = !dirty.empty();
+		const bool cpu_modified            = m_memory_tracker.IsRegionCpuModified(vaddr, size);
+		const bool gpu_modified            = m_memory_tracker.IsRegionGpuModified(vaddr, size);
+		const auto dirty                   = m_gpu_modified_ranges.Intersections(vaddr, size);
+		const bool has_dirty_buffer_source = !dirty.empty();
 		m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size,
 		                                           "image source");
 
@@ -683,10 +626,9 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 	}
 	m_staging_buffer.Commit();
 
-	FaultSafeCacheLock lock(this, m_mutex);
-	const auto         dirty                   = m_gpu_modified_ranges.Intersections(vaddr, size);
-	const bool         has_dirty_buffer_source = !dirty.empty();
-	auto               owner                   = find_owner();
+	const auto dirty                   = m_gpu_modified_ranges.Intersections(vaddr, size);
+	const bool has_dirty_buffer_source = !dirty.empty();
+	auto       owner                   = find_owner();
 	if (has_dirty_buffer_source && owner == m_buffers.end()) {
 		EXIT("BufferCache: GPU-dirty image source lost its native owner\n");
 	}
@@ -719,8 +661,7 @@ void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data)
 	}
 	Libs::LibKernel::Memory::WriteBacking(vaddr, data.data(), data.size());
 
-	FaultSafeCacheLock lock(this, m_mutex);
-	const auto         end = vaddr + data.size();
+	const auto end = vaddr + data.size();
 	for (auto& [address, cached]: m_buffers) {
 		const auto cached_end = address + cached->size;
 		const auto begin      = std::max(vaddr, address);
@@ -750,8 +691,7 @@ void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool
 	}
 	(void)m_texture_cache.ClearMeta(vaddr);
 	{
-		std::lock_guard transaction(m_resource_mutex);
-		const auto      region = m_texture_cache.QueryRegion(vaddr, size);
+		const auto region = m_texture_cache.QueryRegion(vaddr, size);
 		if (!HasGpuDirtyBytes(vaddr, size) && !region.gpu_image_bytes) {
 			if (region.image_bytes) {
 				m_texture_cache.InvalidateMemory(vaddr, size);
@@ -790,8 +730,7 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 		EXIT("BufferCache: invalid or overlapping copy range\n");
 	}
 	if (src_memory || dst_memory) {
-		std::lock_guard transaction(m_resource_mutex);
-		const auto      src_region =
+		const auto src_region =
 		    src_memory ? m_texture_cache.QueryRegion(src_vaddr, size) : TextureCache::RegionInfo {};
 		const auto dst_region =
 		    dst_memory ? m_texture_cache.QueryRegion(dst_vaddr, size) : TextureCache::RegionInfo {};
@@ -842,18 +781,19 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	destination.CopyFrom(command, source, src.offset, dst.offset, size);
 }
 
-bool BufferCache::HasPageOverlap(uint64_t vaddr, uint64_t size) {
+bool BufferCache::IsRegionRegistered(uint64_t vaddr, uint64_t size) {
 	if (vaddr == 0 || size == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
 	    size > TRACKER_ADDRESS_SIZE - vaddr) {
-		EXIT("BufferCache: invalid page-overlap query\n");
+		EXIT("BufferCache: invalid registered-region query\n");
 	}
-	FaultSafeCacheLock lock(this, m_mutex);
-	for (const auto& [address, cached]: m_buffers) {
-		if (PageOverlaps(vaddr, size, address, cached->size)) {
-			return true;
-		}
+	// Cached buffers are ordered and non-overlapping. The last buffer beginning before the query
+	// end is therefore the only possible intersection.
+	const auto candidate = m_buffers.lower_bound(vaddr + size);
+	if (candidate == m_buffers.begin()) {
+		return false;
 	}
-	return false;
+	const auto& [address, cached] = *std::prev(candidate);
+	return address + cached->size > vaddr;
 }
 
 bool BufferCache::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
@@ -861,7 +801,6 @@ bool BufferCache::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
 }
 
 bool BufferCache::HasGpuDirtyBytes(uint64_t vaddr, uint64_t size) {
-	FaultSafeCacheLock lock(this, m_mutex);
 	return !m_gpu_modified_ranges.Intersections(vaddr, size).empty();
 }
 
@@ -870,8 +809,7 @@ bool BufferCache::IsRegionCpuModified(uint64_t vaddr, uint64_t size) {
 }
 
 void BufferCache::RunGarbageCollector() {
-	std::lock_guard transaction(m_resource_mutex);
-	const auto      tick = m_gc_tick++;
+	const auto tick = m_gc_tick++;
 	if (m_graphics.CanReportMemoryUsage()) {
 		m_total_used_memory = m_graphics.GetDeviceMemoryUsage();
 	}
@@ -886,7 +824,6 @@ void BufferCache::RunGarbageCollector() {
 	std::vector<RetiredBuffer>                                       retires;
 	std::vector<std::pair<RetiredBuffer, std::vector<DownloadCopy>>> dirty_retires;
 	{
-		FaultSafeCacheLock    lock(this, m_mutex);
 		std::vector<uint64_t> candidates;
 		for (const auto& [address, owner]: m_buffers) {
 			const auto& cached = *owner;
@@ -934,7 +871,6 @@ void BufferCache::RunGarbageCollector() {
 		QueueGarbageDownload(copies, std::move(retire));
 	}
 
-	FaultSafeCacheLock lock(this, m_mutex);
 	for (const auto& retire: retires) {
 		auto found = m_buffers.find(retire.address);
 		if (found == m_buffers.end() || found->second->size != retire.size ||

src/graphics/host_gpu/renderer/cache/bufferCache.h

@@ -3,7 +3,6 @@
 
 #include "common/abi.h"
 #include "common/common.h"
-#include "common/threads.h"
 #include "graphics/host_gpu/memoryTracker.h"
 #include "graphics/host_gpu/rangeSet.h"
 #include "graphics/host_gpu/renderer/cache/streamBuffer.h"
@@ -20,7 +19,6 @@ struct GraphicContext;
 class CommandBuffer;
 class CommandScheduler;
 class TextureCache;
-class ResourceMutex;
 
 struct BufferBinding {
 	std::shared_ptr<void> owner;
@@ -41,7 +39,7 @@ public:
 	}
 
 	BufferCache(GraphicContext& graphics, CommandScheduler& scheduler, PageManager& page_manager,
-	            TextureCache& texture_cache, ResourceMutex& resource_mutex);
+	            TextureCache& texture_cache);
 	~BufferCache();
 	KYTY_CLASS_NO_COPY(BufferCache);
 
@@ -61,7 +59,8 @@ public:
 	void FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool is_gds = false);
 	void CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t size, bool dst_gds = false,
 	                bool src_gds = false);
-	[[nodiscard]] bool HasPageOverlap(uint64_t vaddr, uint64_t size);
+	// Cache-index and exact dirty-range queries require GPU-thread serialization.
+	[[nodiscard]] bool IsRegionRegistered(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] bool HasGpuDirtyBytes(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] bool IsRegionCpuModified(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] bool IsRegionGpuModified(uint64_t vaddr, uint64_t size);
@@ -84,8 +83,6 @@ private:
 	[[nodiscard]] static constexpr uint64_t AlignDownload(uint64_t size) noexcept {
 		return (size + DOWNLOAD_ALIGNMENT - 1) & ~(DOWNLOAD_ALIGNMENT - 1);
 	}
-	[[nodiscard]] static bool PageOverlaps(uint64_t left, uint64_t left_size, uint64_t right,
-	                                       uint64_t right_size) noexcept;
 	[[nodiscard]] static std::pair<uint64_t, uint64_t> DownloadEnvelope(const DownloadCopy& copy);
 	[[nodiscard]] static bool ResolveOverlap(CacheRange& merged, CacheRange candidate) noexcept;
 	void Upload(CommandBuffer& command, Buffer& destination, uint64_t destination_offset,
@@ -102,7 +99,6 @@ private:
 	GraphicContext&                                   m_graphics;
 	CommandScheduler&                                 m_scheduler;
 	Buffer                                            m_gds_buffer;
-	Common::Mutex                                     m_mutex;
 	std::shared_ptr<Buffer>                           m_null_buffer;
 	std::map<uint64_t, std::unique_ptr<CachedBuffer>> m_buffers;
 	RangeSet                                          m_gpu_modified_ranges;
@@ -112,7 +108,6 @@ private:
 	StreamBuffer                                      m_download_buffer;
 	StreamBuffer                                      m_device_buffer;
 	TextureCache&                                     m_texture_cache;
-	ResourceMutex&                                    m_resource_mutex;
 	uint64_t                                          m_total_used_memory = 0;
 	uint64_t m_trigger_gc_memory  = 1ull * 1024 * 1024 * 1024;
 	uint64_t m_critical_gc_memory = 2ull * 1024 * 1024 * 1024;

src/graphics/host_gpu/renderer/cache/gpuResourceManager.cpp

@@ -6,9 +6,8 @@
 namespace Libs::Graphics {
 
 GpuResourceManager::GpuResourceManager(GraphicContext& graphics, CommandScheduler& scheduler)
-    : m_scheduler(scheduler),
-      m_buffer_cache(graphics, scheduler, m_page_manager, m_texture_cache, m_resource_mutex),
-      m_texture_cache(graphics, scheduler, m_page_manager, m_buffer_cache, m_resource_mutex) {}
+    : m_scheduler(scheduler), m_buffer_cache(graphics, scheduler, m_page_manager, m_texture_cache),
+      m_texture_cache(graphics, scheduler, m_page_manager, m_buffer_cache) {}
 
 GpuResourceManager::~GpuResourceManager() = default;
 
@@ -58,11 +57,6 @@ void GpuResourceManager::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
 		     vaddr, size);
 	}
-	if (m_resource_mutex.IsOwnedByCurrentThread()) {
-		EXIT("unsupported memory unmap from a pre-owned resource transaction, "
-		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
-		     vaddr, size);
-	}
 	const auto unmap = [this, vaddr, size] {
 		if (m_scheduler.Active()) {
 			const auto tick = m_scheduler.CurrentTick();

src/graphics/host_gpu/renderer/cache/gpuResourceManager.h

@@ -5,7 +5,6 @@
 #include "common/common.h"
 #include "graphics/host_gpu/pageManager.h"
 #include "graphics/host_gpu/renderer/cache/bufferCache.h"
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
 
 #include <cstdint>
@@ -35,7 +34,6 @@ public:
 
 private:
 	PageManager               m_page_manager;
-	ResourceMutex             m_resource_mutex;
 	CommandScheduler&         m_scheduler;
 	BufferCache               m_buffer_cache;
 	TextureCache              m_texture_cache;

src/graphics/host_gpu/renderer/cache/resourceMutex.cpp

@@ -1,47 +0,0 @@
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
-
-#include "common/assert.h"
-
-namespace Libs::Graphics {
-
-ResourceMutex::~ResourceMutex() {
-	std::lock_guard state(m_state);
-	if (m_resource_owner != std::thread::id {}) {
-		EXIT("ResourceMutex destroyed with an active owner\n");
-	}
-}
-
-void ResourceMutex::lock() {
-	const auto current = std::this_thread::get_id();
-	{
-		std::lock_guard state(m_state);
-		if (m_resource_owner == current) {
-			EXIT("recursive resource transaction\n");
-		}
-	}
-	m_resource.lock();
-	std::lock_guard state(m_state);
-	if (m_resource_owner != std::thread::id {}) {
-		EXIT("resource mutex acquired with a stale owner\n");
-	}
-	m_resource_owner = current;
-}
-
-void ResourceMutex::unlock() {
-	const auto current = std::this_thread::get_id();
-	{
-		std::lock_guard state(m_state);
-		if (m_resource_owner != current) {
-			EXIT("resource transaction released without ownership\n");
-		}
-		m_resource_owner = {};
-	}
-	m_resource.unlock();
-}
-
-bool ResourceMutex::IsOwnedByCurrentThread() {
-	std::lock_guard state(m_state);
-	return m_resource_owner == std::this_thread::get_id();
-}
-
-} // namespace Libs::Graphics

src/graphics/host_gpu/renderer/cache/resourceMutex.h

@@ -1,30 +0,0 @@
-#ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RESOURCEMUTEX_H_
-#define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RESOURCEMUTEX_H_
-
-#include "common/common.h"
-
-#include <mutex>
-#include <thread>
-
-namespace Libs::Graphics {
-
-// Owner-tracked shared buffer/image transaction.
-class ResourceMutex final {
-public:
-	ResourceMutex() = default;
-	~ResourceMutex();
-	KYTY_CLASS_NO_COPY(ResourceMutex);
-
-	void               lock();
-	void               unlock();
-	[[nodiscard]] bool IsOwnedByCurrentThread();
-
-private:
-	std::mutex      m_resource;
-	std::mutex      m_state;
-	std::thread::id m_resource_owner;
-};
-
-} // namespace Libs::Graphics
-
-#endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RESOURCEMUTEX_H_

src/graphics/host_gpu/renderer/cache/textureCache.cpp

@@ -8,7 +8,6 @@
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/cache/bufferCache.h"
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
@@ -56,13 +55,12 @@ private:
 } // namespace
 
 TextureCache::TextureCache(GraphicContext& graphics, CommandScheduler& scheduler,
-                           PageManager& page_manager, BufferCache& buffer_cache,
-                           ResourceMutex& resource_mutex)
+                           PageManager& page_manager, BufferCache& buffer_cache)
     : m_graphics(graphics), m_scheduler(scheduler), m_page_manager(page_manager),
       m_blit_helper(graphics, scheduler),
       m_tiler(std::make_unique<TileManager>(graphics, scheduler,
                                             buffer_cache.GetUtilityBuffer(MemoryUsage::Stream))),
-      m_buffer_cache(buffer_cache), m_resource_mutex(resource_mutex),
+      m_buffer_cache(buffer_cache),
       m_readback_linear_images(Config::ReadbackLinearImagesEnabled()) {
 	if (m_graphics.CanReportMemoryUsage()) {
 		constexpr int64_t GiB = 1024ll * 1024 * 1024;
@@ -434,9 +432,8 @@ void TextureCache::UntrackImageTail(ImageId id) {
 }
 
 void TextureCache::TrackImageDownload(ImageId id) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	TrackImageDownloadLocked(id, image);
 }
 
@@ -911,7 +908,6 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 	uint32_t    format           = info.guest_format;
 	uint32_t    layers           = info.TransferLayers();
 	bool        volume           = info.IsVolume();
-	bool        layered          = info.IsLayered();
 	bool        allow_depth_tile = direction == TransferDirection::Upload;
 	const char* owner =
 	    direction == TransferDirection::Upload ? "TextureCache" : "TextureCache readback";
@@ -940,7 +936,6 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 				format           = info.guest_format;
 				layers           = info.resources.layers;
 				volume           = false;
-				layered          = layers > 1;
 				allow_depth_tile = false;
 				plan.swap_bgra16 = info.bgra16;
 				owner            = "VideoOut";
@@ -958,19 +953,14 @@ TextureCache::BuildColorTransfer(const Image& image, BindingType binding,
 		plan.swap_bgra16 = info.bgra16;
 	}
 
-	plan.layout = TextureCalcUploadLayout(format, info.extent.width, info.extent.height,
-	                                      info.resources.levels, layers, info.pitch, info.tile_mode,
-	                                      info.data.size, allow_depth_tile, volume, owner);
-	plan.regions = TextureBuildImageCopies(plan.layout, info.extent.width, info.extent.height,
-	                                       layers, info.resources.levels, layered, volume);
-	plan.tiled   = static_cast<Prospero::TileMode>(plan.layout.tile) != Prospero::TileMode::kLinear;
+	plan.layout  = TextureCalcUploadLayout(format, info.extent.width, info.extent.height,
+	                                       info.resources.levels, layers, info.tile_mode,
+	                                       info.data.size, allow_depth_tile, volume, owner);
+	plan.regions = TextureBuildImageCopies(plan.layout);
+	plan.tiled   = static_cast<Prospero::TileMode>(plan.layout.surface.description.tile_mode) !=
+	               Prospero::TileMode::kLinear;
 	if (plan.tiled) {
-		// CMASK/FMASK decoding is intentionally deferred.
-		if (plan.layout.tile_family == TileBlockFamily::Depth64KB &&
-		    Prospero::IsFmaskTextureFormat(format)) {
-			return plan;
-		}
-		if (!TextureBuildGpuTileInfos(info.data.size, plan.regions, plan.layout, format, layers,
+		if (!TextureBuildGpuTileInfos(info.data.size, plan.regions, plan.layout,
 		                              info.resources.levels, plan.tiles)) {
 			return plan;
 		}
@@ -1017,9 +1007,10 @@ void TextureCache::UploadImage(Image& image, const ImageDesc& desc, Buffer& sour
 			     " size=0x%016" PRIx64 " format=%u tile=%u family=%u extent=%ux%ux%u "
 			     "pitch=%u levels=%u layers=%u samples=%u\n",
 			     static_cast<uint32_t>(desc.type), info.data.address, info.data.size,
-			     info.guest_format, info.tile_mode, static_cast<uint32_t>(plan.layout.tile_family),
-			     info.extent.width, info.extent.height, info.extent.depth, info.pitch,
-			     info.resources.levels, info.resources.layers, info.samples);
+			     info.guest_format, info.tile_mode,
+			     static_cast<uint32_t>(plan.layout.surface.texture.block.family), info.extent.width,
+			     info.extent.height, info.extent.depth, info.pitch, info.resources.levels,
+			     info.resources.layers, info.samples);
 		}
 		TileManager::Result linear {source.Handle(), source_offset, info.data.size};
 		if (plan.tiled) {
@@ -1152,8 +1143,7 @@ void TextureCache::RefreshImage(ImageId id, const ImageDesc& desc) {
 }
 
 void TextureCache::AssociateStencil(ImageId depth_id, GuestRange stencil) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
+	CacheLock lock(*this, m_lock);
 	AssociateStencilLocked(depth_id, stencil);
 }
 
@@ -1196,11 +1186,9 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 	}
 
 	ImageId result {};
-	bool    inserted_new = false;
 	{
-		std::lock_guard transaction(m_resource_mutex);
-		CacheLock       lock(*this, m_lock);
-		const auto      candidates =
+		CacheLock  lock(*this, m_lock);
+		const auto candidates =
 		    FindImagesInRegion(desc.info.data.address, desc.info.data.size, false);
 
 		for (const auto id: candidates) {
@@ -1243,19 +1231,12 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 		}
 		if (!result) {
 			result         = InsertImage(desc.info);
-			inserted_new   = true;
 			auto& inserted = ResolveImage(result);
 			if (m_buffer_cache.HasGpuDirtyBytes(inserted.info.data.address,
 			                                    inserted.info.data.size)) {
 				inserted.MarkBufferModified();
 			}
 		}
-		if (inserted_new) {
-			InitializeImage(result, desc);
-		} else {
-			RefreshImage(result, desc);
-		}
-
 		auto& image = ResolveImage(result);
 		if (desc.type == BindingType::VideoOut &&
 		    desc.info.metadata.compression != VideoOutCompression::Uncompressed) {
@@ -1280,6 +1261,13 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {
 	return result;
 }
 
+void TextureCache::UpdateImage(ImageId id) {
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
+	TouchImage(image);
+	RefreshImage(id, ImageDesc {.info = image.info, .type = UploadBinding(image)});
+}
+
 ImageId TextureCache::FindImageFromRange(uint64_t address, uint64_t size, bool ensure_valid) {
 	if (!GuestRange {address, size}.Valid()) {
 		return {};
@@ -1324,22 +1312,24 @@ ImageId TextureCache::FindImageFromRange(uint64_t address, uint64_t size, bool e
 }
 
 vk::ImageView TextureCache::FindTexture(ImageId id, const ImageDesc& desc) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	TouchImage(image);
 	if (!image.info.data.Empty()) {
 		if (!image.registered || image.depth_id || image.binding.needs_rebind) {
 			EXIT("TextureCache: texture requires rediscovery before final acquisition\n");
 		}
+	}
+	if (desc.type == BindingType::Storage) {
+		image.MarkGpuModified();
+	}
+	if (!image.info.data.Empty()) {
 		RefreshImage(id, desc);
 	}
 	switch (desc.type) {
 		case BindingType::Texture: break;
 		case BindingType::Storage:
-			if (image.info.data.Empty()) {
-				image.MarkGpuModified();
-			} else {
+			if (!image.info.data.Empty()) {
 				if (!image.registered || image.depth_id) {
 					EXIT("TextureCache: cannot acquire an unavailable storage image\n");
 				}
@@ -1358,16 +1348,16 @@ vk::ImageView TextureCache::FindRenderTarget(ImageId id, const ImageDesc& desc)
 	if (desc.type != BindingType::RenderTarget) {
 		EXIT("TextureCache: invalid color-target binding\n");
 	}
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	if (!image.registered || image.depth_id || image.binding.needs_rebind) {
 		EXIT("TextureCache: color target requires rediscovery before final acquisition\n");
 	}
 	TouchImage(image);
+	image.MarkGpuModified();
+	image.usage.render_target = true;
 	RefreshImage(id, desc);
 	CommitGpuWrite(image);
-	image.usage.render_target = true;
 	TrackImageDownloadLocked(id, image);
 	const auto view = image.FindView(desc.view_info);
 	RetainImage(m_scheduler.Current(), id);
@@ -1378,13 +1368,14 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 	if (desc.type != BindingType::DepthTarget) {
 		EXIT("TextureCache: invalid depth-target binding\n");
 	}
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	if (!image.registered || image.depth_id || image.binding.needs_rebind) {
 		EXIT("TextureCache: depth target requires rediscovery before final acquisition\n");
 	}
 	TouchImage(image);
+	image.MarkGpuModified();
+	image.usage.depth_target = true;
 	RefreshImage(id, desc);
 	if (desc.info.HasMetadata()) {
 		image.info.metadata = desc.info.metadata;
@@ -1392,7 +1383,6 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 		                            MetaDataInfo {.clear_mask = image.info.htile_clear_mask});
 	}
 	CommitGpuWrite(image);
-	image.usage.depth_target = true;
 	if (desc.info.HasStencil()) {
 		AssociateStencilLocked(id, desc.info.stencil);
 	}
@@ -1402,9 +1392,8 @@ vk::ImageView TextureCache::FindDepthTarget(ImageId id, const ImageDesc& desc) {
 }
 
 void TextureCache::MarkGpuWritten(ImageId id) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto&           image = ResolveImage(id);
+	CacheLock lock(*this, m_lock);
+	auto&     image = ResolveImage(id);
 	if (!image.registered || image.depth_id) {
 		EXIT("TextureCache: cannot mark an unavailable image GPU-written\n");
 	}
@@ -1428,7 +1417,6 @@ bool TextureCache::ClearImageFromBuffer(CommandBuffer& command, uint64_t address
 	if (command.IsInvalid() || !GuestRange {address, size}.Valid()) {
 		EXIT("TextureCache: invalid image clear\n");
 	}
-	std::lock_guard      transaction(m_resource_mutex);
 	CacheLock            lock(*this, m_lock);
 	ImageId              selected {};
 	vk::ImageAspectFlags aspect {};
@@ -1713,14 +1701,9 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uin
 			return false;
 		}
 		if (color.tiled) {
-			const auto binding = m_texture_cache.UploadBinding(image);
-			const auto format =
-			    binding == TextureCache::BindingType::RenderTarget
-			        ? ImageOps::RenderTargetTransferFormat(image.info.bytes_per_block)
-			        : image.info.guest_format;
 			color.tiles.clear();
-			if (!TextureBuildGpuTileInfos(copy_size, color.regions, color.layout, format,
-			                              image.info.TransferLayers(), levels, color.tiles)) {
+			if (!TextureBuildGpuTileInfos(copy_size, color.regions, color.layout, levels,
+			                              color.tiles)) {
 				return false;
 			}
 			color.linear_size = GetLinearSize(color.tiles);
@@ -1891,9 +1874,8 @@ bool TextureCache::IsMetaCleared(uint64_t address, uint32_t slice) {
 }
 
 bool TextureCache::ClearMeta(uint64_t address) {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	const auto      found = m_surface_metas.find(address);
+	CacheLock  lock(*this, m_lock);
+	const auto found = m_surface_metas.find(address);
 	if (found == m_surface_metas.end()) {
 		return false;
 	}
@@ -1919,9 +1901,8 @@ void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 	if (!GuestRange {address, size}.Valid()) {
 		EXIT("TextureCache: invalid unmap range\n");
 	}
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	auto            images = FindImagesInRegion(address, size, false);
+	CacheLock lock(*this, m_lock);
+	auto      images = FindImagesInRegion(address, size, false);
 	if (!images.empty()) {
 		m_scheduler.Finish();
 	}
@@ -1938,9 +1919,8 @@ void TextureCache::UnmapMemory(uint64_t address, uint64_t size) {
 }
 
 void TextureCache::RunGarbageCollector() {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
-	const uint64_t  tick = m_gc_tick++;
+	CacheLock      lock(*this, m_lock);
+	const uint64_t tick = m_gc_tick++;
 	if (m_graphics.CanReportMemoryUsage()) {
 		m_total_used_memory = m_graphics.GetDeviceMemoryUsage();
 	}
@@ -2000,8 +1980,7 @@ void TextureCache::RunGarbageCollector() {
 }
 
 void TextureCache::ProcessDownloadImages() {
-	std::lock_guard transaction(m_resource_mutex);
-	CacheLock       lock(*this, m_lock);
+	CacheLock lock(*this, m_lock);
 	for (const auto id: m_download_images) {
 		const auto owner = ResolveOwner(id);
 		if (owner != nullptr && owner->registered && owner->IsGpuModified()) {

src/graphics/host_gpu/renderer/cache/textureCache.h

@@ -26,7 +26,6 @@ class Buffer;
 class BufferCache;
 class CommandBuffer;
 class CommandScheduler;
-class ResourceMutex;
 class RenderExecutor;
 class StreamBuffer;
 class TileManager;
@@ -49,11 +48,12 @@ public:
 	};
 
 	TextureCache(GraphicContext& graphics, CommandScheduler& scheduler, PageManager& page_manager,
-	             BufferCache& buffer_cache, ResourceMutex& resource_mutex);
+	             BufferCache& buffer_cache);
 	~TextureCache();
 	KYTY_CLASS_NO_COPY(TextureCache);
 
 	[[nodiscard]] ImageId       FindImage(ImageDesc& desc, bool exact_format = false);
+	void                        UpdateImage(ImageId id);
 	[[nodiscard]] ImageId       FindImageFromRange(uint64_t address, uint64_t size,
 	                                               bool ensure_valid = true);
 	[[nodiscard]] vk::ImageView FindTexture(ImageId id, const ImageDesc& desc);
@@ -168,7 +168,6 @@ private:
 	BlitHelper                                        m_blit_helper;
 	std::unique_ptr<TileManager>                      m_tiler;
 	BufferCache&                                      m_buffer_cache;
-	ResourceMutex&                                    m_resource_mutex;
 	std::vector<Slot>                                 m_slots;
 	std::vector<uint32_t>                             m_free_slots;
 	ImagePageTable                                    m_image_page_table;

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -178,10 +178,10 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		pitch = (rt.pitch.pitch_div8_minus1 + 1u) << 3u;
 	} else if (tile) {
 		if (volume) {
-			pitch = TileGetTexturePitch(transfer_format, width, levels, rt.attrib3.tile_mode);
+			pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
 		} else if (standard64) {
 			pitch = TileGetTexturePitch(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float),
-			                            width, levels, rt.attrib3.tile_mode);
+			                            width, rt.attrib3.tile_mode);
 		} else {
 			pitch = TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
 		}
@@ -192,13 +192,20 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		pitch = width;
 	}
 
-	TileSizeOffset   mip_sizes[16] {};
-	TilePaddedSize   mip_padded[16] {};
-	TileVolumeLayout volume_layout {};
-	uint64_t         backing_size = 0;
+	TileSizeOffset    mip_sizes[16] {};
+	TilePaddedSize    mip_padded[16] {};
+	TileSurfaceLayout volume_layout {};
+	uint64_t          backing_size = 0;
 	if (volume) {
-		if (!tile || !TileGetTextureVolumeLayout(transfer_format, width, height, depth, levels,
-		                                         rt.attrib3.tile_mode, volume_layout)) {
+		const TileSurfaceDescription description {transfer_format,
+		                                          rt.attrib3.tile_mode,
+		                                          TileSurfaceDimension::Dim3D,
+		                                          width,
+		                                          height,
+		                                          depth,
+		                                          levels,
+		                                          1};
+		if (!tile || !TileGetTiledTextureLayout(description, volume_layout)) {
 			EXIT("unsupported 3D render-target layout: %ux%ux%u levels=%u tile=%u\n", width, height,
 			     depth, levels, rt.attrib3.tile_mode);
 		}
@@ -209,7 +216,7 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		bool          valid_layout = false;
 		if (standard64) {
 			TileGetTextureSize(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float), width,
-			                   height, pitch, levels, rt.attrib3.tile_mode, &layout, mip_sizes,
+			                   height, levels, rt.attrib3.tile_mode, &layout, mip_sizes,
 			                   mip_padded);
 			valid_layout = layout.size != 0 && layout.align == 65536;
 		} else {
@@ -289,11 +296,12 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	desc.info.tile_mode       = rt.attrib3.tile_mode;
 	for (uint32_t level = 0; level < levels; level++) {
 		if (volume) {
+			const auto& mip             = volume_layout.mips[level];
 			desc.info.mip_layout[level] = {
-			    volume_layout.level_offsets[level],
-			    volume_layout.level_sizes[level],
-			    volume_layout.level_widths[level],
-			    volume_layout.level_heights[level],
+			    mip.offset,
+			    mip.size,
+			    mip.padded_width,
+			    mip.padded_height,
 			};
 			continue;
 		}

src/graphics/host_gpu/renderer/commandScheduler.cpp

@@ -99,7 +99,10 @@ bool CommandScheduler::InDeferredOperation() noexcept {
 
 CommandScheduler::CommandScheduler(RenderContext& context, GraphicContext& graphics)
     : m_master(graphics), m_context(context), m_graphics(graphics),
-      m_priority_thread([this](std::stop_token stop) { PriorityOperationsThread(stop); }) {}
+      m_priority_thread([this](std::stop_token stop) { PriorityOperationsThread(stop); }) {
+	m_buffers.reserve(CommandBufferGrowStep);
+	m_buffer_ticks.reserve(CommandBufferGrowStep);
+}
 
 CommandScheduler::~CommandScheduler() {
 	Shutdown();
@@ -154,10 +157,7 @@ void CommandScheduler::Begin(HW::Context& registers, HW::UserConfig& user_config
 	m_shaders     = &shaders;
 
 	if (!Active()) {
-		for (auto& buffer: m_buffers) {
-			buffer = std::make_unique<RenderCommandBuffer>(*this);
-		}
-		m_current = 0;
+		m_current = static_cast<int>(GrowCommandBuffers());
 	}
 
 	BindCurrent();
@@ -235,7 +235,13 @@ void CommandScheduler::Wait(uint64_t tick) {
 }
 
 void CommandScheduler::PopPendingOperations() {
-	m_master.Refresh();
+	PopPendingOperations(true);
+}
+
+void CommandScheduler::PopPendingOperations(bool refresh_gpu_tick) {
+	if (refresh_gpu_tick) {
+		m_master.Refresh();
+	}
 	for (;;) {
 		Common::UniqueFunction<void> callback;
 		{
@@ -362,7 +368,7 @@ uint64_t CommandScheduler::NextSubmitSequence() noexcept {
 }
 
 void CommandScheduler::CheckActive() const {
-	EXIT_IF(!Active() || m_current >= BufferCount);
+	EXIT_IF(!Active() || static_cast<size_t>(m_current) >= m_buffers.size());
 }
 
 RenderCommandBuffer& CommandScheduler::Current() const {
@@ -384,15 +390,52 @@ CommandBuffer& CommandScheduler::SubmitCurrent(SubmitInfo& submit) {
 	const auto signal_tick = m_master.NextTick();
 	submit.AddSignal(m_master.Handle(), signal_tick);
 	submitted.Execute(submit);
-	m_recording = false;
+	m_buffer_ticks[static_cast<size_t>(m_current)] = signal_tick;
+	m_recording                                    = false;
 	return submitted;
 }
 
+int CommandScheduler::FindReusableBuffer(uint64_t gpu_tick) const {
+	EXIT_IF(m_buffer_ticks.size() != m_buffers.size());
+	const auto buffer_count = m_buffers.size();
+	for (size_t offset = 1; offset <= buffer_count; ++offset) {
+		const auto candidate = (static_cast<size_t>(m_current) + offset) % buffer_count;
+		if (gpu_tick >= m_buffer_ticks[candidate]) {
+			return static_cast<int>(candidate);
+		}
+	}
+	return -1;
+}
+
+size_t CommandScheduler::GrowCommandBuffers() {
+	const auto first = m_buffers.size();
+	const auto end   = first + CommandBufferGrowStep;
+	m_buffers.reserve(end);
+	m_buffer_ticks.reserve(end);
+	for (size_t i = first; i < end; ++i) {
+		m_buffers.emplace_back(std::make_unique<RenderCommandBuffer>(*this));
+		m_buffer_ticks.push_back(0);
+	}
+	return first;
+}
+
 void CommandScheduler::BeginNext() {
 	EXIT_IF(m_recording);
-	m_current = (m_current + 1) % BufferCount;
+
+	auto candidate = FindReusableBuffer(m_master.KnownGpuTick());
+	bool refreshed = false;
+	if (candidate < 0) {
+		m_master.Refresh();
+		refreshed = true;
+		candidate = FindReusableBuffer(m_master.KnownGpuTick());
+	}
+	if (candidate < 0) {
+		candidate = static_cast<int>(GrowCommandBuffers());
+	}
+
+	m_current = candidate;
 	Current().WaitForFenceAndReset();
-	PopPendingOperations();
+	PopPendingOperations(!refreshed);
 	BindCurrent();
 	Current().Begin();
 	m_recording = true;

src/graphics/host_gpu/renderer/commandScheduler.h

@@ -16,6 +16,7 @@
 #include <queue>
 
 #include <thread>
+#include <vector>
 
 namespace Libs::Graphics {
 
@@ -31,8 +32,6 @@ struct CommandSlot {
 
 class CommandScheduler {
 public:
-	static constexpr int BufferCount = 8;
-
 	CommandScheduler(RenderContext& context, GraphicContext& graphics);
 	~CommandScheduler();
 	KYTY_CLASS_NO_COPY(CommandScheduler);
@@ -65,6 +64,8 @@ public:
 	[[nodiscard]] GraphicContext& Graphics() const noexcept { return m_graphics; }
 
 private:
+	static constexpr size_t CommandBufferGrowStep = 4;
+
 	class CommandPool {
 	public:
 		CommandPool() = default;
@@ -94,30 +95,34 @@ private:
 	void                       BindCurrent() const;
 	CommandBuffer&             SubmitCurrent(SubmitInfo& submit);
 	void                       BeginNext();
+	void                       PopPendingOperations(bool refresh_gpu_tick);
+	[[nodiscard]] int          FindReusableBuffer(uint64_t gpu_tick) const;
+	[[nodiscard]] size_t       GrowCommandBuffers();
 	void                       PriorityOperationsThread(std::stop_token stop);
 	void                       RunOperation(Common::UniqueFunction<void>&& operation);
 	[[nodiscard]] CommandSlot* AllocateCommandBuffer();
 	[[nodiscard]] uint64_t     NextSubmitSequence() noexcept;
 
-	MasterSemaphore                                               m_master;
-	RenderContext&                                                m_context;
-	GraphicContext&                                               m_graphics;
-	CommandPool                                                   m_command_pool;
-	std::array<std::unique_ptr<RenderCommandBuffer>, BufferCount> m_buffers;
-	std::queue<PendingOperation>                                  m_pending_operations;
-	std::queue<PendingOperation>                                  m_priority_operations;
-	std::mutex                                                    m_operation_mutex;
-	std::condition_variable                                       m_operation_available;
-	std::jthread                                                  m_priority_thread;
-	bool                                                          m_priority_active      = false;
-	uint64_t                                                      m_priority_active_tick = 0;
-	OperationState        m_operation_state = OperationState::Open;
-	int                   m_current         = -1;
-	bool                  m_recording       = false;
-	HW::Context*          m_registers       = nullptr;
-	HW::UserConfig*       m_user_config     = nullptr;
-	HW::Shader*           m_shaders         = nullptr;
-	std::atomic<uint64_t> m_submit_sequence = 0;
+	MasterSemaphore                                   m_master;
+	RenderContext&                                    m_context;
+	GraphicContext&                                   m_graphics;
+	CommandPool                                       m_command_pool;
+	std::vector<std::unique_ptr<RenderCommandBuffer>> m_buffers;
+	std::vector<uint64_t>                             m_buffer_ticks;
+	std::queue<PendingOperation>                      m_pending_operations;
+	std::queue<PendingOperation>                      m_priority_operations;
+	std::mutex                                        m_operation_mutex;
+	std::condition_variable                           m_operation_available;
+	std::jthread                                      m_priority_thread;
+	bool                                              m_priority_active      = false;
+	uint64_t                                          m_priority_active_tick = 0;
+	OperationState                                    m_operation_state      = OperationState::Open;
+	int                                               m_current              = -1;
+	bool                                              m_recording            = false;
+	HW::Context*                                      m_registers            = nullptr;
+	HW::UserConfig*                                   m_user_config          = nullptr;
+	HW::Shader*                                       m_shaders              = nullptr;
+	std::atomic<uint64_t>                             m_submit_sequence      = 0;
 
 	friend class CommandBuffer;
 };

src/graphics/host_gpu/renderer/debug.cpp

@@ -10,6 +10,7 @@
 #include "graphics/host_gpu/renderer/render.h"
 
 #include <algorithm>
+#include <array>
 #include <atomic>
 #include <cmath>
 #include <fmt/format.h>
@@ -686,14 +687,6 @@ static void McCheck(const HW::ModeControl& c) {
 	}
 	EXIT_NOT_IMPLEMENTED(c.polymode_front_ptype != 0 && c.polymode_front_ptype != 2);
 	EXIT_NOT_IMPLEMENTED(c.polymode_back_ptype != 0 && c.polymode_back_ptype != 2);
-	if (c.poly_offset_front_enable || c.poly_offset_back_enable) {
-		static bool logged = false;
-		if (!logged) {
-			LOGF("\t temporary: PA_SU_SC_MODE_CNTL.POLY_OFFSET_*_ENABLE is not implemented; "
-			     "continuing without depth bias\n");
-			logged = true;
-		}
-	}
 	if (c.vtx_window_offset_enable) {
 		static bool logged = false;
 		if (!logged) {
@@ -1067,18 +1060,24 @@ static ScissorRect ScissorRectClamp(ScissorRect r, uint32_t width, uint32_t heig
 	return r;
 }
 
+static constexpr std::array<uint16_t, 16> MakeScissorIntersectionRules() {
+	std::array<uint16_t, 16> rules {};
+	for (uint32_t candidate = 0; candidate < rules.size(); candidate++) {
+		for (uint32_t combination = 0; combination < 16; combination++) {
+			if ((combination & candidate) == candidate) {
+				rules[candidate] |= static_cast<uint16_t>(1u << combination);
+			}
+		}
+	}
+	return rules;
+}
+
 static bool ScissorClipRuleToIntersectionMask(uint16_t rule, uint8_t* mask) {
 	EXIT_IF(mask == nullptr);
 
-	for (uint32_t candidate = 0; candidate < 16; candidate++) {
-		uint16_t expected = 0;
-		for (uint32_t combination = 0; combination < 16; combination++) {
-			if ((combination & candidate) == candidate) {
-				expected |= static_cast<uint16_t>(1u << combination);
-			}
-		}
-
-		if (expected == rule) {
+	static constexpr auto rules = MakeScissorIntersectionRules();
+	for (uint32_t candidate = 0; candidate < rules.size(); candidate++) {
+		if (rules[candidate] == rule) {
 			*mask = static_cast<uint8_t>(candidate);
 			return true;
 		}

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -38,7 +38,10 @@ namespace Libs::Graphics {
 	EXIT("unsupported render state; details were printed above\n");
 }
 
-static vk::StencilOp ConvertStencilOp(uint8_t value) {
+static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t op_value) {
+	if (write_mask == 0) {
+		return vk::StencilOp::eKeep;
+	}
 	switch (static_cast<Prospero::StencilOp>(value)) {
 		case Prospero::StencilOp::kKeep: return vk::StencilOp::eKeep;
 		case Prospero::StencilOp::kZero: return vk::StencilOp::eZero;
@@ -49,7 +52,17 @@ static vk::StencilOp ConvertStencilOp(uint8_t value) {
 		case Prospero::StencilOp::kInvert: return vk::StencilOp::eInvert;
 		case Prospero::StencilOp::kAddWrap: return vk::StencilOp::eIncrementAndWrap;
 		case Prospero::StencilOp::kSubWrap: return vk::StencilOp::eDecrementAndWrap;
-		default: DepthFatal("unsupported stencil operation");
+		case Prospero::StencilOp::kXor:
+			if ((write_mask & op_value) == 0) {
+				return vk::StencilOp::eKeep;
+			}
+			if ((write_mask & ~op_value) != 0) {
+				DepthFatal("unsupported stencil XOR operands: write mask=0x%02" PRIx8
+				           ", operation value=0x%02" PRIx8,
+				           write_mask, op_value);
+			}
+			return vk::StencilOp::eInvert;
+		default: DepthFatal("unsupported stencil operation: 0x%02" PRIx8, value);
 	}
 }
 
@@ -283,6 +296,8 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 	r.stencil_clear_value  = hw.GetStencilClearValue();
 	r.stencil_test_enable  = has_stencil && dc.stencil_enable;
 	if (r.stencil_test_enable) {
+		const uint8_t front_write_mask = rc.stencil_clear_enable ? 0 : sm.stencil_writemask;
+		const uint8_t back_write_mask  = rc.stencil_clear_enable ? 0 : sm.stencil_writemask_bf;
 		if (dc.stencilfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
 		    (dc.backface_enable &&
 		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways)) ||
@@ -294,19 +309,18 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 			DepthFatal("unsupported stencil compare or replacement state");
 		}
 		r.stencil_static_front = {
-		    ConvertStencilOp(sc.stencil_fail), ConvertStencilOp(sc.stencil_zpass),
-		    ConvertStencilOp(sc.stencil_zfail), static_cast<vk::CompareOp>(dc.stencilfunc)};
-		r.stencil_dynamic_front = {sm.stencil_mask,
-		                           rc.stencil_clear_enable ? 0u : sm.stencil_writemask,
-		                           sm.stencil_testval};
+		    ConvertStencilOp(sc.stencil_fail, front_write_mask, sm.stencil_opval),
+		    ConvertStencilOp(sc.stencil_zpass, front_write_mask, sm.stencil_opval),
+		    ConvertStencilOp(sc.stencil_zfail, front_write_mask, sm.stencil_opval),
+		    static_cast<vk::CompareOp>(dc.stencilfunc)};
+		r.stencil_dynamic_front = {sm.stencil_mask, front_write_mask, sm.stencil_testval};
 		if (dc.backface_enable) {
-			r.stencil_static_back  = {ConvertStencilOp(sc.stencil_fail_bf),
-			                          ConvertStencilOp(sc.stencil_zpass_bf),
-			                          ConvertStencilOp(sc.stencil_zfail_bf),
-			                          static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
-			r.stencil_dynamic_back = {sm.stencil_mask_bf,
-			                          rc.stencil_clear_enable ? 0u : sm.stencil_writemask_bf,
-			                          sm.stencil_testval_bf};
+			r.stencil_static_back = {
+			    ConvertStencilOp(sc.stencil_fail_bf, back_write_mask, sm.stencil_opval_bf),
+			    ConvertStencilOp(sc.stencil_zpass_bf, back_write_mask, sm.stencil_opval_bf),
+			    ConvertStencilOp(sc.stencil_zfail_bf, back_write_mask, sm.stencil_opval_bf),
+			    static_cast<vk::CompareOp>(dc.stencilfunc_bf)};
+			r.stencil_dynamic_back = {sm.stencil_mask_bf, back_write_mask, sm.stencil_testval_bf};
 		} else {
 			r.stencil_static_back  = r.stencil_static_front;
 			r.stencil_dynamic_back = r.stencil_dynamic_front;

src/graphics/host_gpu/renderer/image/image.cpp

@@ -663,7 +663,8 @@ Image::Image(GraphicContext& graphics, CommandScheduler& scheduler, const ImageI
     : info(image_info), m_graphics(&graphics), m_scheduler(&scheduler) {
 	KYTY_PROFILER_FUNCTION();
 	ImageOps::Validate(info);
-	m_cpu_dirty = !info.data.Empty();
+	m_cpu_dirty =
+	    !info.data.Empty() && info.metadata.compression == VideoOutCompression::Uncompressed;
 	if (info.pixel_format == vk::Format::eUndefined) {
 		return;
 	}

src/graphics/host_gpu/renderer/image/textureCommon.cpp

@@ -168,14 +168,6 @@ RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t raw_layout, uint32_
 
 namespace {
 
-static uint64_t GetLevelSrcOffset(const TileSizeOffset& level_size) {
-	return (level_size.src_size != 0 ? level_size.src_offset : level_size.offset);
-}
-
-static uint64_t GetLevelSrcSize(const TileSizeOffset& level_size) {
-	return (level_size.src_size != 0 ? level_size.src_size : level_size.size);
-}
-
 static uint32_t GetTextureLevelDepth(uint32_t depth, uint32_t level, bool volume_texture) {
 	return volume_texture ? std::max(depth >> level, 1u) : depth;
 }
@@ -188,20 +180,6 @@ static size_t GetTextureRegionCount(uint32_t depth, uint64_t levels, bool volume
 	return count;
 }
 
-uint64_t TextureUploadSliceSourceOffset(const TextureUploadLayout& layout, uint32_t level,
-                                        uint32_t slice) {
-	if (level >= 16 || layout.level_sizes[level].size == 0) {
-		EXIT("invalid texture upload slice source, level=%u slice=%u\n", level, slice);
-	}
-	const auto level_offset = GetLevelSrcOffset(layout.level_sizes[level]);
-	const auto slice_stride =
-	    layout.source_slice_stride != 0 ? layout.source_slice_stride : layout.slice_stride;
-	if (slice_stride != 0 && slice > (UINT64_MAX - level_offset) / slice_stride) {
-		EXIT("texture upload slice source offset overflow, level=%u slice=%u\n", level, slice);
-	}
-	return level_offset + static_cast<uint64_t>(slice) * slice_stride;
-}
-
 vk::ComponentSwizzle TextureGetComponentSwizzle(uint8_t s) {
 	switch (static_cast<Prospero::CompSwizzle>(s)) {
 		case Prospero::CompSwizzle::kZero: return vk::ComponentSwizzle::eZero;
@@ -241,12 +219,11 @@ vk::Format TextureGetFormat(uint32_t fmt) {
 
 namespace {
 
-uint64_t CalcTextureSliceStride(const TileSizeOffset* level_sizes, uint64_t levels,
+uint64_t CalcTextureSliceStride(const TextureUploadMipLayout* mips, uint64_t levels,
                                 uint64_t total_size, uint32_t depth) {
 	uint64_t stride = 0;
 	for (uint32_t i = 0; i < levels; i++) {
-		stride =
-		    std::max(stride, static_cast<uint64_t>(level_sizes[i].offset) + level_sizes[i].size);
+		stride = std::max(stride, mips[i].offset + mips[i].size);
 	}
 
 	if (depth > 1 && total_size != 0 && total_size % depth == 0) {
@@ -259,39 +236,26 @@ uint64_t CalcTextureSliceStride(const TileSizeOffset* level_sizes, uint64_t leve
 	return stride;
 }
 
-uint64_t CalcLinearUploadLevelSize(uint32_t fmt, uint32_t pitch, uint32_t height) {
-	if (const uint32_t bytes_per_element = Prospero::NumBytesPerElement(fmt);
-	    bytes_per_element != 0) {
-		return static_cast<uint64_t>(pitch) * height * bytes_per_element;
-	}
-
-	if (const uint32_t bytes_per_block = Prospero::BlockCompressedBytesPerBlock(fmt);
-	    bytes_per_block != 0) {
-		const uint32_t blocks_w = std::max((pitch + 3u) / 4u, 1u);
-		const uint32_t blocks_h = std::max((height + 3u) / 4u, 1u);
-		return static_cast<uint64_t>(blocks_w) * blocks_h * bytes_per_block;
-	}
-
-	return 0;
+uint64_t CalcLinearUploadLevelSize(const TileTextureElementLayout& element, uint32_t pitch,
+                                   uint32_t height) {
+	const uint32_t elements_w =
+	    std::max((pitch + element.texel_width - 1u) / element.texel_width, 1u);
+	const uint32_t elements_h =
+	    std::max((height + element.texel_height - 1u) / element.texel_height, 1u);
+	return static_cast<uint64_t>(elements_w) * elements_h * element.bytes;
 }
 
-uint64_t SetLinearUploadLevels(TileSizeOffset* level_sizes, uint32_t fmt, uint64_t height,
+uint64_t SetLinearUploadLevels(TextureUploadMipLayout*         mips,
+                               const TileTextureElementLayout& element, uint64_t height,
                                uint64_t levels, uint32_t base_pitch) {
 	uint64_t offset = 0;
 	auto     pitch  = base_pitch;
 	auto     h      = static_cast<uint32_t>(height);
 
 	for (uint32_t i = 0; i < levels; i++) {
-		const auto size = CalcLinearUploadLevelSize(fmt, pitch, h);
-		EXIT_NOT_IMPLEMENTED(size > 0xffffffffull);
-		EXIT_NOT_IMPLEMENTED(offset > 0xffffffffull);
-
-		level_sizes[i].size       = static_cast<uint32_t>(size);
-		level_sizes[i].offset     = static_cast<uint32_t>(offset);
-		level_sizes[i].src_size   = 0;
-		level_sizes[i].src_offset = 0;
-		level_sizes[i].x          = 0;
-		level_sizes[i].y          = 0;
+		const auto size = CalcLinearUploadLevelSize(element, pitch, h);
+		mips[i].size    = size;
+		mips[i].offset  = offset;
 
 		offset += size;
 		if (pitch > 1) {
@@ -307,141 +271,123 @@ uint64_t SetLinearUploadLevels(TileSizeOffset* level_sizes, uint32_t fmt, uint64
 
 } // namespace
 
-TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint64_t width, uint64_t height,
-                                            uint64_t levels, uint32_t depth, uint64_t pitch,
-                                            uint64_t tile, uint64_t upload_size,
-                                            bool allow_depth_tile, bool volume_texture,
-                                            const char* owner) {
+TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32_t height,
+                                            uint32_t levels, uint32_t depth, uint32_t tile,
+                                            uint64_t upload_size, bool allow_depth_tile,
+                                            bool volume_texture, const char* owner) {
 	TextureUploadLayout layout {};
-	layout.tile           = static_cast<uint32_t>(tile);
-	layout.pitch          = static_cast<uint32_t>(pitch);
-	layout.texel_block    = Prospero::BlockCompressedBytesPerBlock(fmt) != 0 ? 4u : 1u;
-	layout.volume_texture = volume_texture;
+	layout.surface.description = {
+	    fmt,
+	    tile,
+	    volume_texture ? TileSurfaceDimension::Dim3D : TileSurfaceDimension::Dim2D,
+	    width,
+	    height,
+	    volume_texture ? depth : 1u,
+	    levels,
+	    volume_texture ? 1u : depth,
+	};
+	const auto&              description = layout.surface.description;
+	const auto               tile_mode   = static_cast<Prospero::TileMode>(description.tile_mode);
+	TileTextureElementLayout element {};
 
-	if (fmt != 0) {
-		if (layout.tile != 0) {
-			const auto tile_mode = static_cast<Prospero::TileMode>(layout.tile);
-			switch (tile_mode) {
-				case Prospero::TileMode::kStandard256B:
-					if (TileIsStandard256BTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard256B;
-					}
-					break;
-				case Prospero::TileMode::kStandard4KB:
-					if (TileIsStandard4KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard4KB;
-					}
-					break;
-				case Prospero::TileMode::kStandard64KB:
-					if (TileIsStandard64KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Standard64KB;
-					}
-					break;
-				case Prospero::TileMode::kPrt:
-					if (TileIsStandard64KBTextureSupported(fmt)) {
-						layout.tile_family = TileBlockFamily::Prt64KB;
-					}
-					break;
-				case Prospero::TileMode::kRenderTarget:
-					if (Prospero::RenderTargetBytesPerElement(fmt) != 0) {
-						layout.tile_family = TileBlockFamily::RenderTarget64KB;
-					}
-					break;
-				case Prospero::TileMode::kDepth:
-					if (allow_depth_tile && Prospero::RenderTargetBytesPerElement(fmt) != 0) {
-						layout.tile_family = TileBlockFamily::Depth64KB;
-					}
-					break;
-				default: break;
-			}
-			if (layout.tile_family == TileBlockFamily::Count) {
-				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
-				     "size=%" PRIu64 " extent=%" PRIu64 "x%" PRIu64 " pitch=%" PRIu64
-				     " levels=%" PRIu64 "\n",
-				     owner, static_cast<uint32_t>(fmt), layout.tile, upload_size, width, height,
-				     pitch, levels);
-			}
-		}
-
-		layout.pitch = TileGetTexturePitch(fmt, width, levels, layout.tile);
-
-		TileGetTextureSize(fmt, width, height, layout.pitch, levels, layout.tile, nullptr,
-		                   layout.level_sizes, layout.padded_sizes);
-
-		if (static_cast<Prospero::TileMode>(layout.tile) != Prospero::TileMode::kLinear) {
-			if (layout.volume_texture) {
-				layout.slice_stride = SetLinearUploadLevels(layout.level_sizes, fmt, height, levels,
-				                                            static_cast<uint32_t>(width));
-			} else {
-				TileSizeOffset tiled_levels[16] {};
-				std::copy_n(layout.level_sizes, levels, tiled_levels);
-				layout.source_slice_stride =
-				    CalcTextureSliceStride(tiled_levels, levels, upload_size, depth);
-				SetLinearUploadLevels(layout.level_sizes, fmt, height, levels, layout.pitch);
-				for (uint32_t i = 0; i < levels; ++i) {
-					if (tiled_levels[i].src_size > tiled_levels[i].size) {
-						layout.first_tail_level = std::min(layout.first_tail_level, i);
-					}
-					layout.level_sizes[i].src_offset = GetLevelSrcOffset(tiled_levels[i]);
-					layout.level_sizes[i].src_size   = GetLevelSrcSize(tiled_levels[i]);
-					layout.level_sizes[i].x          = tiled_levels[i].x;
-					layout.level_sizes[i].y          = tiled_levels[i].y;
-				}
-			}
-		} else if (layout.volume_texture) {
-			layout.slice_stride =
-			    CalcTextureSliceStride(layout.level_sizes, levels, upload_size, depth);
-		}
-	} else {
+	if (fmt == 0) {
 		EXIT("%s: legacy texture upload format unsupported: fmt=0 tile=%u size=%" PRIu64
-		     " extent=%" PRIu64 "x%" PRIu64 " pitch=%" PRIu64 " levels=%" PRIu64 "\n",
-		     owner, layout.tile, upload_size, width, height, pitch, levels);
+		     " extent=%ux%u levels=%u\n",
+		     owner, description.tile_mode, upload_size, width, height, levels);
 	}
 
-	if (!layout.volume_texture) {
-		layout.slice_stride =
-		    CalcTextureSliceStride(layout.level_sizes, levels, upload_size, depth);
+	switch (tile_mode) {
+		case Prospero::TileMode::kLinear:
+			if (!TileGetTextureElementLayout(fmt, element)) {
+				EXIT("%s: unsupported linear texture format: fmt=%u\n", owner, fmt);
+			}
+			break;
+		case Prospero::TileMode::kDepth:
+			if (!allow_depth_tile || !TileGetTiledTextureLayout(description, layout.surface)) {
+				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
+				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
+				     owner, static_cast<uint32_t>(fmt), description.tile_mode, upload_size, width,
+				     height, levels);
+			}
+			break;
+		default:
+			if (!TileGetTiledTextureLayout(description, layout.surface)) {
+				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
+				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
+				     owner, static_cast<uint32_t>(fmt), description.tile_mode, upload_size, width,
+				     height, levels);
+			}
+			break;
+	}
+
+	if (tile_mode != Prospero::TileMode::kLinear) {
+		element = {layout.surface.texture.block.bytes_per_element,
+		           layout.surface.texture.texel_width, layout.surface.texture.texel_height};
+	}
+	layout.pitch = TileGetTexturePitch(fmt, width, description.tile_mode);
+	if (tile_mode == Prospero::TileMode::kLinear) {
+		TileSizeOffset level_sizes[16] {};
+		TilePaddedSize padded_sizes[16] {};
+		TileGetTextureSize(fmt, width, height, levels, description.tile_mode, nullptr, level_sizes,
+		                   padded_sizes);
+		for (uint32_t level = 0; level < levels; ++level) {
+			layout.mips[level] = {level_sizes[level].offset, level_sizes[level].size,
+			                      padded_sizes[level].width, padded_sizes[level].height};
+		}
+	} else if (volume_texture) {
+		layout.slice_stride = SetLinearUploadLevels(layout.mips, element, height, levels, width);
+	} else {
+		layout.source_slice_stride = layout.surface.block_slice_size;
+		if (depth > 1 && upload_size != 0 && upload_size % depth == 0) {
+			layout.source_slice_stride = std::max(layout.source_slice_stride, upload_size / depth);
+		}
+		SetLinearUploadLevels(layout.mips, element, height, levels, layout.pitch);
+	}
+
+	if (!volume_texture || tile_mode == Prospero::TileMode::kLinear) {
+		layout.slice_stride = CalcTextureSliceStride(layout.mips, levels, upload_size, depth);
 	}
 	return layout;
 }
 
-std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout,
-                                                         uint32_t width, uint32_t height,
-                                                         uint32_t depth, uint64_t levels,
-                                                         bool array_texture, bool volume_texture) {
-	uint32_t mip_width  = width;
-	uint32_t mip_height = height;
-	uint32_t mip_pitch  = volume_texture && static_cast<Prospero::TileMode>(layout.tile) !=
-	                                            Prospero::TileMode::kLinear
-	                          ? width
-	                          : layout.pitch;
+std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout) {
+	const auto& description    = layout.surface.description;
+	const bool  volume_texture = description.dimension == TileSurfaceDimension::Dim3D;
+	const auto  depth          = volume_texture ? description.depth : description.layers;
+	uint32_t    mip_width      = description.width;
+	uint32_t    mip_height     = description.height;
+	uint32_t mip_pitch = volume_texture && static_cast<Prospero::TileMode>(description.tile_mode) !=
+	                                           Prospero::TileMode::kLinear
+	                         ? description.width
+	                         : layout.pitch;
 
 	std::vector<vk::BufferImageCopy> regions;
-	regions.reserve(GetTextureRegionCount(depth, levels, volume_texture));
-	for (uint32_t i = 0; i < levels; i++) {
-		EXIT_NOT_IMPLEMENTED(layout.level_sizes[i].size == 0);
+	regions.reserve(GetTextureRegionCount(depth, description.levels, volume_texture));
+	for (uint32_t i = 0; i < description.levels; i++) {
+		EXIT_NOT_IMPLEMENTED(layout.mips[i].size == 0);
 
 		const auto mip_depth = GetTextureLevelDepth(depth, i, volume_texture);
 
 		for (uint32_t z = 0; z < mip_depth; z++) {
 			const auto          slice_offset = z * layout.slice_stride;
 			vk::BufferImageCopy region {};
-			region.bufferOffset     = layout.level_sizes[i].offset + slice_offset;
-			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i, array_texture ? z : 0,
+			region.bufferOffset     = layout.mips[i].offset + slice_offset;
+			region.imageSubresource = {vk::ImageAspectFlagBits::eColor, i, volume_texture ? 0u : z,
 			                           1};
 			region.imageOffset.z    = volume_texture ? static_cast<int>(z) : 0;
 			region.imageExtent      = {mip_width, mip_height, 1};
-			const bool linear =
-			    static_cast<Prospero::TileMode>(layout.tile) == Prospero::TileMode::kLinear;
+			const bool linear       = static_cast<Prospero::TileMode>(description.tile_mode) ==
+			                          Prospero::TileMode::kLinear;
 			if (linear) {
-				region.bufferRowLength   = layout.padded_sizes[i].width;
-				region.bufferImageHeight = layout.padded_sizes[i].height;
+				region.bufferRowLength   = layout.mips[i].row_length;
+				region.bufferImageHeight = layout.mips[i].image_height;
 			} else {
-				const auto align = [](uint32_t value, uint32_t block) {
+				const auto texel_width = layout.surface.texture.texel_width;
+				const auto align       = [](uint32_t value, uint32_t block) {
 					return ((value + block - 1u) / block) * block;
 				};
-				const auto pitch       = align(mip_pitch, layout.texel_block);
-				region.bufferRowLength = pitch > align(mip_width, layout.texel_block) ? pitch : 0;
+				const auto pitch       = align(mip_pitch, texel_width);
+				region.bufferRowLength = pitch > align(mip_width, texel_width) ? pitch : 0;
 			}
 			regions.push_back(region);
 		}
@@ -460,24 +406,6 @@ std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayo
 	return regions;
 }
 
-struct GpuTileElementLayout {
-	uint32_t bytes = 0;
-	uint32_t wide  = 1;
-	uint32_t tall  = 1;
-};
-
-static bool GetGpuTileElementLayout(uint32_t fmt, GpuTileElementLayout& out) {
-	if (const auto bytes = Prospero::NumBytesPerElement(fmt); bytes != 0) {
-		out = {bytes, 1, 1};
-		return true;
-	}
-	if (const auto bytes = Prospero::BlockCompressedBytesPerBlock(fmt); bytes != 0) {
-		out = {bytes, 4, 4};
-		return true;
-	}
-	return false;
-}
-
 static bool SetGpuTileSize(uint64_t offset, uint64_t length, uint64_t capacity, uint64_t& size) {
 	if (offset > capacity || length > capacity - offset) {
 		return false;
@@ -487,42 +415,32 @@ static bool SetGpuTileSize(uint64_t offset, uint64_t length, uint64_t capacity,
 }
 
 bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
-                              const TextureUploadLayout& layout, uint32_t fmt, uint32_t depth,
-                              uint64_t levels, std::vector<GpuTileInfo>& out_infos) {
+                              const TextureUploadLayout& layout, uint32_t levels,
+                              std::vector<GpuTileInfo>& out_infos) {
+	const auto& description    = layout.surface.description;
+	const bool  volume_texture = description.dimension == TileSurfaceDimension::Dim3D;
+	const auto  depth          = volume_texture ? description.depth : description.layers;
 	if (tiled_size == 0 || levels == 0 || levels > 16 || depth == 0 ||
-	    regions.size() != GetTextureRegionCount(depth, levels, layout.volume_texture) ||
-	    Prospero::IsFmaskTextureFormat(fmt)) {
+	    regions.size() != GetTextureRegionCount(depth, levels, volume_texture) ||
+	    Prospero::IsFmaskTextureFormat(description.format)) {
 		return false;
 	}
 
-	GpuTileElementLayout element {};
-	if (layout.tile_family == TileBlockFamily::RenderTarget64KB ||
-	    layout.tile_family == TileBlockFamily::Depth64KB) {
-		element.bytes = Prospero::RenderTargetBytesPerElement(fmt);
-	} else if (!GetGpuTileElementLayout(fmt, element)) {
-		return false;
-	}
-	if (element.bytes == 0) {
+	const auto& surface = layout.surface;
+	const auto& texture = surface.texture;
+	const auto& block   = texture.block;
+	if (surface.description.levels < levels || block.bytes_per_element == 0 ||
+	    block.family == TileBlockFamily::Count) {
 		return false;
 	}
 
 	std::vector<GpuTileInfo> infos;
 	infos.reserve(regions.size());
-	if (layout.volume_texture) {
-		TileVolumeLayout volume {};
-		if (!TileGetTextureVolumeLayout(fmt, regions[0].imageExtent.width,
-		                                regions[0].imageExtent.height, depth,
-		                                static_cast<uint32_t>(levels), layout.tile, volume)) {
-			return false;
-		}
-		element = {volume.bytes_per_element, volume.texel_width, volume.texel_height};
-		TileBlockLayout block {};
-		if (!TileGetBlockLayout(volume.family, element.bytes, block)) return false;
-
+	if (volume_texture) {
 		size_t region_base = 0;
 		for (uint32_t level = 0; level < levels; ++level) {
 			const uint32_t mip_depth     = GetTextureLevelDepth(depth, level, true);
-			const bool     tail          = level >= volume.first_tail_level;
+			const bool     tail          = level >= surface.first_tail_level;
 			const uint64_t linear_stride = layout.slice_stride;
 			for (uint32_t z = 0; z < mip_depth; z += block.block_depth) {
 				const uint32_t copy_depth = std::min(block.block_depth, mip_depth - z);
@@ -537,50 +455,43 @@ bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferI
 				info.bytes_per_element = block.bytes_per_element;
 				info.linear_offset     = region.bufferOffset;
 				info.tiled_offset =
-				    static_cast<uint64_t>(z / block.block_depth) * volume.block_slice_size +
-				    volume.level_offsets[level];
+				    static_cast<uint64_t>(z / block.block_depth) * surface.block_slice_size +
+				    surface.mips[level].offset;
 				const uint64_t linear_span =
 				    static_cast<uint64_t>(copy_depth - 1u) * linear_stride +
-				    layout.level_sizes[level].size;
+				    layout.mips[level].size;
 				if (!SetGpuTileSize(info.linear_offset, linear_span, UINT64_MAX,
 				                    info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, volume.level_sizes[level], tiled_size,
+				    !SetGpuTileSize(info.tiled_offset, surface.mips[level].size, tiled_size,
 				                    info.tiled_size)) {
 					return false;
 				}
 				info.linear_slice_stride = linear_stride;
-				info.width =
-				    std::max((region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
-				info.height = std::max((logical_height + element.tall - 1u) / element.tall, 1u);
-				info.depth  = copy_depth;
+				info.width  = std::max((region.imageExtent.width + texture.texel_width - 1u) /
+				                           texture.texel_width,
+				                       1u);
+				info.height = std::max(
+				    (logical_height + texture.texel_height - 1u) / texture.texel_height, 1u);
+				info.depth = copy_depth;
 				info.surface_z =
 				    block.block_depth == 1 ? static_cast<uint32_t>(region.imageOffset.z) : 0;
-				info.pitch        = std::max((pitch + element.wide - 1u) / element.wide, 1u);
-				info.tail_x       = tail ? volume.tail_x[level] : 0;
-				info.tail_y       = tail ? volume.tail_y[level] : 0;
+				info.pitch = std::max((pitch + texture.texel_width - 1u) / texture.texel_width, 1u);
+				info.tail_x       = tail ? surface.mips[level].tail_x : 0;
+				info.tail_y       = tail ? surface.mips[level].tail_y : 0;
 				info.tail         = tail;
-				info.tiled_width  = volume.level_widths[level];
-				info.tiled_height = volume.level_heights[level];
+				info.tiled_width  = surface.mips[level].padded_width;
+				info.tiled_height = surface.mips[level].padded_height;
 				infos.push_back(info);
 			}
 			region_base += mip_depth;
 		}
 	} else {
-		const auto base_family = layout.tile_family;
-		if (base_family == TileBlockFamily::Count) {
-			return false;
-		}
-
 		size_t region_index = 0;
 		for (uint32_t level = 0; level < levels; level++) {
-			const auto&     level_size = layout.level_sizes[level];
-			const bool      tail       = level >= layout.first_tail_level;
-			const auto      family     = base_family;
-			TileBlockLayout block {};
-			if (!TileGetBlockLayout(family, element.bytes, block)) {
-				return false;
-			}
-			const auto level_depth = GetTextureLevelDepth(depth, level, layout.volume_texture);
+			const auto& level_size  = layout.mips[level];
+			const auto& mip         = surface.mips[level];
+			const bool  tail        = level >= surface.first_tail_level;
+			const auto  level_depth = GetTextureLevelDepth(depth, level, false);
 			for (uint32_t z = 0; z < level_depth; z++) {
 				const auto& region = regions[region_index++];
 				const auto  pitch =
@@ -589,39 +500,36 @@ bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferI
 				                                 ? region.bufferImageHeight
 				                                 : region.imageExtent.height;
 				GpuTileInfo info {};
-				info.family            = block.family;
-				info.bytes_per_element = block.bytes_per_element;
-				info.linear_offset     = region.bufferOffset;
-				info.tiled_offset      = TextureUploadSliceSourceOffset(layout, level, z);
-				if (!SetGpuTileSize(info.linear_offset, level_size.size, UINT64_MAX,
-				                    info.linear_size) ||
-				    !SetGpuTileSize(info.tiled_offset, GetLevelSrcSize(level_size), tiled_size,
-				                    info.tiled_size)) {
+				info.family              = block.family;
+				info.bytes_per_element   = block.bytes_per_element;
+				info.linear_offset       = region.bufferOffset;
+				const auto source_stride = layout.source_slice_stride != 0
+				                               ? layout.source_slice_stride
+				                               : surface.block_slice_size;
+				if (z > (UINT64_MAX - mip.offset) / source_stride) {
 					return false;
 				}
-				info.width =
-				    std::max((region.imageExtent.width + element.wide - 1u) / element.wide, 1u);
-				info.height    = std::max((logical_height + element.tall - 1u) / element.tall, 1u);
-				info.surface_z = base_family == TileBlockFamily::RenderTarget64KB ||
-				                         base_family == TileBlockFamily::Depth64KB
+				info.tiled_offset = mip.offset + static_cast<uint64_t>(z) * source_stride;
+				if (!SetGpuTileSize(info.linear_offset, level_size.size, UINT64_MAX,
+				                    info.linear_size) ||
+				    !SetGpuTileSize(info.tiled_offset, mip.size, tiled_size, info.tiled_size)) {
+					return false;
+				}
+				info.width  = std::max((region.imageExtent.width + texture.texel_width - 1u) /
+				                           texture.texel_width,
+				                       1u);
+				info.height = std::max(
+				    (logical_height + texture.texel_height - 1u) / texture.texel_height, 1u);
+				info.surface_z = block.family == TileBlockFamily::RenderTarget64KB ||
+				                         block.family == TileBlockFamily::Depth64KB
 				                     ? region.imageSubresource.baseArrayLayer
 				                     : 0;
-				info.pitch     = std::max((pitch + element.wide - 1u) / element.wide, 1u);
-				info.tail      = tail;
-				info.tail_x    = tail ? level_size.x : 0;
-				info.tail_y    = tail ? level_size.y : 0;
-				info.tiled_width =
-				    layout.padded_sizes[level].width != 0
-				        ? std::max((layout.padded_sizes[level].width + element.wide - 1u) /
-				                       element.wide,
-				                   1u)
-				        : info.pitch;
-				info.tiled_height =
-				    layout.padded_sizes[level].height != 0
-				        ? std::max((layout.padded_sizes[level].height + element.tall - 1u) /
-				                       element.tall,
-				                   1u)
-				        : info.height;
+				info.pitch = std::max((pitch + texture.texel_width - 1u) / texture.texel_width, 1u);
+				info.tail  = tail;
+				info.tail_x       = tail ? mip.tail_x : 0;
+				info.tail_y       = tail ? mip.tail_y : 0;
+				info.tiled_width  = mip.padded_width;
+				info.tiled_height = mip.padded_height;
 				infos.push_back(info);
 			}
 		}

src/graphics/host_gpu/renderer/image/textureCommon.h

@@ -19,34 +19,32 @@ struct RenderTargetFormatInfo {
 	Prospero::ColorComponentMapping export_mapping;
 };
 
+struct TextureUploadMipLayout {
+	uint64_t offset       = 0;
+	uint64_t size         = 0;
+	uint32_t row_length   = 0;
+	uint32_t image_height = 0;
+};
+
 struct TextureUploadLayout {
-	uint32_t        tile                = 0;
-	uint32_t        pitch               = 0;
-	uint32_t        texel_block         = 1;
-	uint64_t        slice_stride        = 0;
-	uint64_t        source_slice_stride = 0;
-	uint32_t        first_tail_level    = 16;
-	TileBlockFamily tile_family         = TileBlockFamily::Count;
-	bool            volume_texture      = false;
-	TileSizeOffset  level_sizes[16]     = {};
-	TilePaddedSize  padded_sizes[16]    = {};
+	uint32_t               pitch               = 0;
+	uint64_t               slice_stride        = 0;
+	uint64_t               source_slice_stride = 0;
+	TileSurfaceLayout      surface;
+	TextureUploadMipLayout mips[16] = {};
 };
 
 vk::ComponentMapping   TextureGetComponentMapping(uint32_t swizzle);
 vk::Format             TextureGetFormat(uint32_t fmt);
 RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t layout, uint32_t type, uint32_t order);
-TextureUploadLayout    TextureCalcUploadLayout(uint32_t fmt, uint64_t width, uint64_t height,
-                                               uint64_t levels, uint32_t depth, uint64_t pitch,
-                                               uint64_t tile, uint64_t upload_size,
-                                               bool allow_depth_tile, bool volume_texture,
-                                               const char* owner);
-std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout,
-                                                         uint32_t width, uint32_t height,
-                                                         uint32_t depth, uint64_t levels,
-                                                         bool array_texture, bool volume_texture);
+TextureUploadLayout    TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32_t height,
+                                               uint32_t levels, uint32_t depth, uint32_t tile,
+                                               uint64_t upload_size, bool allow_depth_tile,
+                                               bool volume_texture, const char* owner);
+std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout);
 bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
-                              const TextureUploadLayout& layout, uint32_t fmt, uint32_t depth,
-                              uint64_t levels, std::vector<GpuTileInfo>& infos);
+                              const TextureUploadLayout& layout, uint32_t levels,
+                              std::vector<GpuTileInfo>& infos);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/renderer/pipeline/descriptors.cpp

@@ -239,7 +239,7 @@ bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor, c
 	const auto samples      = multisampled ? 1u << descriptor.LastLevel() : 1u;
 	const auto pitch =
 	    multisampled ? TileGetDepthPitch(width, image.info.bytes_per_block, descriptor.LastLevel())
-	                 : TileGetTexturePitch(descriptor.Format(), width, 1, descriptor.TileMode());
+	                 : TileGetTexturePitch(descriptor.Format(), width, descriptor.TileMode());
 	const bool supported_2d    = type == Prospero::ImageType::kColor2D &&
 	                             image.info.resources.layers == 1 && descriptor.Depth() == 0 &&
 	                             descriptor.BaseArray5() == 0;
@@ -321,7 +321,7 @@ static void ValidateDepthTargetBinding(const ShaderRecompiler::IR::ImageResource
 		return;
 	}
 	const auto descriptor_pitch =
-	    TileGetTexturePitch(descriptor.Format(), static_cast<uint32_t>(descriptor.Width5()) + 1u, 1,
+	    TileGetTexturePitch(descriptor.Format(), static_cast<uint32_t>(descriptor.Width5()) + 1u,
 	                        descriptor.TileMode());
 	EXIT("unsupported sampled depth target: resource=%d descriptor=%d encoding=%d format=%d "
 	     "kind=%u dimension=%u mip_mode=%u read=%d written=%d atomic=%d compare=%d "
@@ -373,26 +373,33 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	const bool is_3d = resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim3D &&
 	                   descriptor.Type() == Prospero::GpuEnumValue(Prospero::ImageType::kColor3D) &&
 	                   descriptor.BaseArray5() == 0;
-	TileBlockLayout depth_block {};
-	const auto      depth_bpe = Prospero::RenderTargetBytesPerElement(descriptor.Format());
-	const bool      supported_depth_tile =
-	    tile == Prospero::GpuEnumValue(Prospero::TileMode::kDepth) && !resource.read &&
-	    !Prospero::IsFmaskTextureFormat(descriptor.Format()) && (is_2d || is_2d_array) &&
-	    TileGetBlockLayout(TileBlockFamily::Depth64KB, depth_bpe, depth_block);
-	const bool supported_standard_256b =
-	    (is_2d || is_2d_array) &&
-	    tile == Prospero::GpuEnumValue(Prospero::TileMode::kStandard256B) &&
-	    TileIsStandard256BTextureSupported(descriptor.Format());
-	const bool supported_standard_tile =
-	    (tile == Prospero::GpuEnumValue(Prospero::TileMode::kStandard4KB) &&
-	     TileIsStandard4KBTextureSupported(descriptor.Format())) ||
-	    (tile == Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB) &&
-	     TileIsStandard64KBTextureSupported(descriptor.Format()));
-	const bool supported_tile = tile == Prospero::GpuEnumValue(Prospero::TileMode::kLinear) ||
-	                            tile == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget) ||
-	                            supported_depth_tile || supported_standard_256b ||
-	                            supported_standard_tile;
-	const auto swizzle        = descriptor.DstSelXYZW();
+	TileTextureBlockLayout tile_layout {};
+	bool                   supported_tile = false;
+	switch (static_cast<Prospero::TileMode>(tile)) {
+		case Prospero::TileMode::kLinear: supported_tile = true; break;
+		case Prospero::TileMode::kDepth:
+			supported_tile =
+			    !resource.read && !Prospero::IsFmaskTextureFormat(descriptor.Format()) &&
+			    (is_2d || is_2d_array) &&
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		case Prospero::TileMode::kStandard256B:
+			supported_tile =
+			    (is_2d || is_2d_array) &&
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		case Prospero::TileMode::kStandard4KB:
+		case Prospero::TileMode::kStandard64KB:
+			supported_tile =
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, is_3d, tile_layout);
+			break;
+		case Prospero::TileMode::kRenderTarget:
+			supported_tile =
+			    TileGetTextureBlockLayout(descriptor.Format(), tile, false, tile_layout);
+			break;
+		default: break;
+	}
+	const auto swizzle = descriptor.DstSelXYZW();
 	const bool supported_swizzle =
 	    IsValidImageSwizzle(swizzle) &&
 	    (swizzle == DstSel(4, 5, 6, 7) || !resource.read || resource.atomic);
@@ -498,18 +505,20 @@ static TextureCache::ImageDesc NullTextureDesc(const ShaderRecompiler::IR::Image
 
 static void PopulateTextureMipLayout(ImageInfo& info) {
 	if (info.IsVolume() && info.tile_mode != Prospero::GpuEnumValue(Prospero::TileMode::kLinear)) {
-		TileVolumeLayout volume {};
-		if (!TileGetTextureVolumeLayout(info.guest_format, info.extent.width, info.extent.height,
-		                                info.extent.depth, info.resources.levels, info.tile_mode,
-		                                volume)) {
+		TileSurfaceLayout            surface {};
+		const TileSurfaceDescription description {
+		    info.guest_format,  info.tile_mode,    TileSurfaceDimension::Dim3D, info.extent.width,
+		    info.extent.height, info.extent.depth, info.resources.levels,       1};
+		if (!TileGetTiledTextureLayout(description, surface)) {
 			EXIT("unsupported normalized volume texture layout\n");
 		}
 		for (uint32_t level = 0; level < info.resources.levels; level++) {
+			const auto& mip        = surface.mips[level];
 			info.mip_layout[level] = {
-			    volume.level_offsets[level],
-			    volume.level_sizes[level],
-			    volume.level_widths[level],
-			    volume.level_heights[level],
+			    mip.offset,
+			    mip.size,
+			    mip.padded_width,
+			    mip.padded_height,
 			};
 		}
 		return;
@@ -517,7 +526,7 @@ static void PopulateTextureMipLayout(ImageInfo& info) {
 
 	TileSizeOffset levels[16] {};
 	TilePaddedSize padded[16] {};
-	TileGetTextureSize(info.guest_format, info.extent.width, info.extent.height, info.pitch,
+	TileGetTextureSize(info.guest_format, info.extent.width, info.extent.height,
 	                   info.resources.levels, info.tile_mode, nullptr, levels, padded);
 	for (uint32_t level = 0; level < info.resources.levels; level++) {
 		const auto offset =
@@ -675,9 +684,9 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso
 		}
 		size.size *= image_layers;
 	} else {
-		pitch = TileGetTexturePitch(format, width, levels, tile);
-		TileGetTextureTotalSize(format, width, height, volume ? depth : image_layers, pitch, levels,
-		                        tile, volume, size);
+		pitch = TileGetTexturePitch(format, width, tile);
+		TileGetTextureTotalSize(format, width, height, volume ? depth : image_layers, levels, tile,
+		                        volume, size);
 	}
 	EXIT_NOT_IMPLEMENTED(size.size == 0 || size.align == 0 ||
 	                     (address & (static_cast<uint64_t>(size.align) - 1u)) != 0);
@@ -924,22 +933,21 @@ RenderExecutor::PrepareGraphicsBindings(CommandBuffer& buffer, const ShaderStage
 	return bindings;
 }
 
-static void RetainBindings(CommandBuffer&                            buffer,
-                           const DescriptorCache::NativeDescriptors& resources) {
+static void RetainBindings(CommandBuffer& buffer, DescriptorCache::NativeDescriptors& resources) {
 	if (resources.flattened_srt.owner != nullptr) {
-		buffer.RetainResourceUntilFence(resources.flattened_srt.owner);
+		buffer.RetainResourceUntilFence(std::move(resources.flattened_srt.owner));
 	}
 	if (resources.user_data.owner != nullptr) {
-		buffer.RetainResourceUntilFence(resources.user_data.owner);
+		buffer.RetainResourceUntilFence(std::move(resources.user_data.owner));
 	}
-	for (const auto& view: resources.buffers) {
+	for (auto& view: resources.buffers) {
 		if (view.owner != nullptr) {
-			buffer.RetainResourceUntilFence(view.owner);
+			buffer.RetainResourceUntilFence(std::move(view.owner));
 		}
 	}
-	for (const auto& view: resources.addresses) {
+	for (auto& view: resources.addresses) {
 		if (view.owner != nullptr) {
-			buffer.RetainResourceUntilFence(view.owner);
+			buffer.RetainResourceUntilFence(std::move(view.owner));
 		}
 	}
 }

src/graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.cpp

@@ -36,6 +36,22 @@ VulkanMemoryBarrier MakeShaderWriteDependency() {
 	return barrier;
 }
 
+VulkanMemoryBarrier MakeShaderAccessDependency() {
+	VulkanMemoryBarrier barrier {};
+	barrier.sType         = vk::StructureType::eMemoryBarrier;
+	barrier.srcAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
+	barrier.dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
+	return barrier;
+}
+
+VulkanMemoryBarrier MakeShaderWriteHazardDependency() {
+	VulkanMemoryBarrier barrier {};
+	barrier.sType         = vk::StructureType::eMemoryBarrier;
+	barrier.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
+	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
+	return barrier;
+}
+
 vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer) {
 	EXIT_IF(buffer == nullptr);
 
@@ -84,6 +100,21 @@ bool HasShaderBufferWrites(const ShaderStageRuntime& runtime) {
 	return !CollectShaderBufferWrites(*runtime.program, *runtime.resources).empty();
 }
 
+void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages) {
+	EXIT_IF(vk_buffer == nullptr || !source_stages);
+	const auto barrier = MakeShaderAccessDependency();
+	vk_buffer.pipelineBarrier(source_stages, vk::PipelineStageFlagBits::eAllCommands,
+	                          vk::DependencyFlags {}, 1, &barrier, 0, nullptr, 0, nullptr);
+}
+
+void ShaderWriteHazardBarrier(vk::CommandBuffer      vk_buffer,
+                              vk::PipelineStageFlags destination_stages) {
+	EXIT_IF(vk_buffer == nullptr || !destination_stages);
+	const auto barrier = MakeShaderWriteHazardDependency();
+	vk_buffer.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands, destination_stages,
+	                          vk::DependencyFlags {}, 1, &barrier, 0, nullptr, 0, nullptr);
+}
+
 void ShaderWriteBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages) {
 	EXIT_IF(vk_buffer == nullptr || !source_stages);
 	const auto barrier = MakeShaderWriteDependency();

src/graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h

@@ -18,12 +18,17 @@ struct ShaderBufferWriteRange {
 };
 
 vk::PipelineStageFlags  ShaderPipelineStages(vk::ShaderStageFlags stages);
+VulkanMemoryBarrier     MakeShaderAccessDependency();
+VulkanMemoryBarrier     MakeShaderWriteHazardDependency();
 VulkanMemoryBarrier     MakeShaderWriteDependency();
 vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer);
 std::vector<ShaderBufferWriteRange>
 CollectShaderBufferWrites(const ShaderRecompiler::IR::Program&          program,
                           const ShaderRecompiler::IR::ResourceSnapshot& resources);
 bool HasShaderBufferWrites(const ShaderStageRuntime& runtime);
+void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages);
+void ShaderWriteHazardBarrier(vk::CommandBuffer      vk_buffer,
+                              vk::PipelineStageFlags destination_stages);
 void ShaderWriteBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages);
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/renderer/pipeline/shaders.cpp

@@ -947,15 +947,23 @@ void CreatePipelineInternal(
 	    vk::DynamicState::eViewport,
 	    vk::DynamicState::eScissor,
 	    vk::DynamicState::eLineWidth,
+	    vk::DynamicState::eDepthBiasEnable,
+	    vk::DynamicState::eDepthBias,
 	    vk::DynamicState::eStencilCompareMask,
 	    vk::DynamicState::eStencilReference,
 	    vk::DynamicState::eStencilWriteMask,
 #if !defined(__APPLE__)
+	    // Keep last so depth-only pipelines can omit this dynamic state.
 	    vk::DynamicState::eColorWriteEnableEXT, // unsupported by MoltenVK; static mask instead
 #endif
 	};
-	const auto dynamic_states_count =
+	auto dynamic_states_count =
 	    static_cast<uint32_t>(sizeof(dynamic_states) / sizeof(dynamic_states[0]));
+#if !defined(__APPLE__)
+	if (static_params.color_count == 0) {
+		dynamic_states_count--;
+	}
+#endif
 
 	vk::PipelineDynamicStateCreateInfo dynamic_state {};
 	dynamic_state.sType             = vk::StructureType::ePipelineDynamicStateCreateInfo;

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -325,9 +325,6 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 
 	auto vk_buffer = buffer.Handle();
 	CommitBindings(buffer, vk::PipelineBindPoint::eCompute, pipeline.pipeline_layout, bindings);
-	vk_buffer.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline.pipeline);
-	vk_buffer.dispatch(thread_group_x, thread_group_y, thread_group_z);
-
 	bool has_storage_writes = HasShaderBufferWrites(input_info.stage);
 	has_storage_writes =
 	    std::any_of(program.info.images.begin(), program.info.images.end(),
@@ -338,8 +335,15 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 	                }) ||
 	    has_storage_writes;
 	if (has_storage_writes) {
-		ShaderWriteBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
+		// A host fence used to serialize every dispatch. Preserve its read-before-write ordering
+		// while allowing the queue to execute asynchronously.
+		ShaderWriteHazardBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
 	}
+	vk_buffer.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline.pipeline);
+	vk_buffer.dispatch(thread_group_x, thread_group_y, thread_group_z);
+
+	// The removed host fence also ordered read-only dispatches before later writers.
+	ShaderAccessBarrier(vk_buffer, vk::PipelineStageFlagBits::eComputeShader);
 	ResetBindings();
 }
 

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -64,12 +64,32 @@ int32_t ResolveVertexOffset(uint32_t index_offset, const ShaderVertexInputInfo&
 	return 0;
 }
 
-static std::atomic<uint32_t> g_draw_state_log_count       = 0;
-static std::atomic<uint32_t> g_draw_input_log_count       = 0;
-static std::atomic<uint32_t> g_mrt_state_log_count        = 0;
-static std::atomic<uint32_t> g_shader_stage_log_count     = 0;
+static std::atomic<uint32_t> g_draw_state_log_count   = 0;
+static std::atomic<uint32_t> g_draw_input_log_count   = 0;
+static std::atomic<uint32_t> g_mrt_state_log_count    = 0;
+static std::atomic<uint32_t> g_shader_stage_log_count = 0;
+
 static std::atomic<uint32_t> g_framebuffer_skip_log_count = 0;
 
+static float ConvertPolygonOffsetConstantFactor(float guest_factor, const HW::PolyOffset& offset,
+                                                vk::Format host_depth_format) {
+	if (offset.db_is_float_fmt) {
+		return guest_factor;
+	}
+
+	int host_depth_bits = 0;
+	switch (host_depth_format) {
+		case vk::Format::eD16Unorm:
+		case vk::Format::eD16UnormS8Uint: host_depth_bits = 16; break;
+		case vk::Format::eD24UnormS8Uint: host_depth_bits = 24; break;
+		default:
+			// A fixed-point guest bias cannot be represented exactly by a floating-point host
+			// attachment without VK_EXT_depth_bias_control.
+			return guest_factor;
+	}
+	return std::ldexp(guest_factor, host_depth_bits + offset.neg_num_db_bits);
+}
+
 static const char* RenderColorTypeName(RenderColorType type) {
 	switch (type) {
 		case RenderColorType::NoColorOutput: return "NoColorOutput";
@@ -335,6 +355,22 @@ static PipelineDynamicParameters BuildGraphicsDynamicParams(const RenderCommandB
 	ret.stencil_front      = depth.stencil_dynamic_front;
 	ret.stencil_back       = depth.stencil_dynamic_back;
 
+	const auto& mode        = ctx.GetModeControl();
+	const auto& poly_offset = ctx.GetPolyOffset();
+	const bool  use_front   = mode.poly_offset_front_enable && !mode.cull_front;
+	const bool  use_back    = mode.poly_offset_back_enable && !mode.cull_back;
+	ret.depth_bias_enable   = use_front || use_back;
+	if (ret.depth_bias_enable) {
+		// Vulkan has one bias for both faces. Prefer a visible front face when both are enabled.
+		const float guest_constant_factor =
+		    use_front ? poly_offset.front_offset : poly_offset.back_offset;
+		ret.depth_bias_constant_factor =
+		    ConvertPolygonOffsetConstantFactor(guest_constant_factor, poly_offset, depth.format);
+		ret.depth_bias_clamp = poly_offset.clamp;
+		ret.depth_bias_slope_factor =
+		    (use_front ? poly_offset.front_scale : poly_offset.back_scale) / 16.0f;
+	}
+
 	// CB_COLOR_CONTROL.operation controls special CB operations, not the normal color component
 	// write mask. Use CB_TARGET_MASK here so scanout passes with mode=Disable do not go black.
 	for (uint32_t i = 0; i < RENDER_COLOR_ATTACHMENTS_MAX; i++) {
@@ -379,6 +415,13 @@ static void SetDynamicParams(const RenderCommandBuffer& buffer, vk::CommandBuffe
 	}
 	vk_buffer.setLineWidth(line_width);
 
+	vk_buffer.setDepthBiasEnable(dynamic_params.depth_bias_enable ? VK_TRUE : VK_FALSE);
+	if (dynamic_params.depth_bias_enable) {
+		vk_buffer.setDepthBias(dynamic_params.depth_bias_constant_factor,
+		                       dynamic_params.depth_bias_clamp,
+		                       dynamic_params.depth_bias_slope_factor);
+	}
+
 	if (dynamic_params.stencil_test_enable) {
 		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eFront,
 		                                dynamic_params.stencil_front.compareMask);
@@ -684,11 +727,24 @@ struct VertexBufferRange {
 	[[nodiscard]] uint64_t RequestedSize() const { return requested_end - base_address; }
 };
 
-static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&         buffer,
-                                                       const ShaderVertexInputInfo& vs_input_info) {
+struct PreparedVertexBuffers {
+	static constexpr uint32_t MaxBuffers = ShaderVertexInputInfo::RES_MAX;
+
+	std::array<vk::Buffer, MaxBuffers>            buffers {};
+	std::array<vk::DeviceSize, MaxBuffers>        offsets {};
+	std::array<std::shared_ptr<void>, MaxBuffers> owners {};
+	uint32_t                                      count       = 0;
+	uint32_t                                      owner_count = 0;
+};
+
+static PreparedVertexBuffers AcquireVertexBuffers(RenderCommandBuffer&         buffer,
+                                                  const ShaderVertexInputInfo& vs_input_info) {
+	EXIT_IF(vs_input_info.buffers_num < 0 ||
+	        vs_input_info.buffers_num > ShaderVertexInputInfo::RES_MAX);
+
 	// Collect the non-empty guest vertex ranges.
-	std::vector<VertexBufferRange> ranges;
-	ranges.reserve(vs_input_info.buffers_num);
+	std::array<VertexBufferRange, ShaderVertexInputInfo::RES_MAX> ranges {};
+	uint32_t                                                      range_count = 0;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
 		const auto  size   = VertexBufferDescriptorSize(vertex);
@@ -699,27 +755,31 @@ static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&
 			EXIT("invalid vertex buffer range: addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
 			     vertex.addr, size);
 		}
-		ranges.push_back({vertex.addr, vertex.addr + size});
+		ranges[range_count++] = {vertex.addr, vertex.addr + size};
 	}
 
-	std::ranges::sort(ranges, [](const VertexBufferRange& left, const VertexBufferRange& right) {
-		return left.base_address < right.base_address;
-	});
+	std::sort(ranges.begin(), ranges.begin() + range_count,
+	          [](const VertexBufferRange& left, const VertexBufferRange& right) {
+		          return left.base_address < right.base_address;
+	          });
 
 	// Merge overlapping or touching ranges before acquiring host buffers.
-	std::vector<VertexBufferRange> merged_ranges;
-	merged_ranges.reserve(ranges.size());
-	for (const auto& range: ranges) {
-		if (!merged_ranges.empty() && merged_ranges.back().requested_end >= range.base_address) {
-			merged_ranges.back().requested_end =
-			    std::max(merged_ranges.back().requested_end, range.requested_end);
+	std::array<VertexBufferRange, ShaderVertexInputInfo::RES_MAX> merged_ranges {};
+	uint32_t                                                      merged_count = 0;
+	for (uint32_t i = 0; i < range_count; i++) {
+		const auto& range = ranges[i];
+		if (merged_count != 0 &&
+		    merged_ranges[merged_count - 1].requested_end >= range.base_address) {
+			merged_ranges[merged_count - 1].requested_end =
+			    std::max(merged_ranges[merged_count - 1].requested_end, range.requested_end);
 			continue;
 		}
-		merged_ranges.push_back(range);
+		merged_ranges[merged_count++] = {range.base_address, range.requested_end};
 	}
 
 	auto& cache = buffer.GetContext().GetBufferCache();
-	for (auto& range: merged_ranges) {
+	for (uint32_t i = 0; i < merged_count; i++) {
+		auto& range = merged_ranges[i];
 		// PPSA20298
 		const auto size =
 		    Libs::LibKernel::Memory::ClampRangeSize(range.base_address, range.RequestedSize());
@@ -728,30 +788,47 @@ static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&
 	}
 
 	// Rebuild slot bindings, offsetting non-empty slots into their acquired merged range.
-	std::vector<BufferBinding> bindings;
-	bindings.reserve(vs_input_info.buffers_num);
+	PreparedVertexBuffers prepared;
+	prepared.count = static_cast<uint32_t>(vs_input_info.buffers_num);
+	std::shared_ptr<Buffer> null_owner;
+	vk::Buffer              null_buffer = nullptr;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
 		const auto  size   = VertexBufferDescriptorSize(vertex);
 		if (size == 0) {
-			auto owner = cache.ObtainNullBuffer();
-			bindings.push_back({owner, owner->Handle(), 0});
+			if (null_owner == nullptr) {
+				null_owner  = cache.ObtainNullBuffer();
+				null_buffer = null_owner->Handle();
+			}
+			prepared.buffers[i] = null_buffer;
+			prepared.offsets[i] = 0;
 			continue;
 		}
 
-		const auto range = std::ranges::find_if(merged_ranges, [&](const VertexBufferRange& value) {
-			return vertex.addr >= value.base_address && vertex.addr < value.acquired_end;
-		});
-		if (range == merged_ranges.end()) {
+		const auto range = std::find_if(merged_ranges.begin(), merged_ranges.begin() + merged_count,
+		                                [&](const VertexBufferRange& value) {
+			                                return vertex.addr >= value.base_address &&
+			                                       vertex.addr < value.acquired_end;
+		                                });
+		if (range == merged_ranges.begin() + merged_count) {
 			EXIT("vertex buffer address is outside the acquired range: addr=0x%016" PRIx64 "\n",
 			     vertex.addr);
 		}
 
-		auto binding = range->binding;
-		binding.offset += vertex.addr - range->base_address;
-		bindings.push_back(std::move(binding));
+		prepared.buffers[i] = range->binding.buffer;
+		prepared.offsets[i] = range->binding.offset + vertex.addr - range->base_address;
 	}
-	return bindings;
+
+	if (null_owner != nullptr) {
+		prepared.owners[prepared.owner_count++] = std::move(null_owner);
+	}
+	for (uint32_t i = 0; i < merged_count; i++) {
+		if (merged_ranges[i].binding.owner != nullptr) {
+			EXIT_IF(prepared.owner_count >= PreparedVertexBuffers::MaxBuffers);
+			prepared.owners[prepared.owner_count++] = std::move(merged_ranges[i].binding.owner);
+		}
+	}
+	return prepared;
 }
 
 static void SetDrawDebugPhase(RenderCommandBuffer& buffer, uint64_t submit_id,
@@ -883,10 +960,9 @@ static void RefreshShaders(RenderCommandBuffer& buffer, const DrawCallInfo& draw
 	}
 }
 
-static std::vector<BufferBinding> PrepareVertexBuffers(uint64_t                     submit_id,
-                                                       RenderCommandBuffer&         buffer,
-                                                       const DrawCallInfo&          draw,
-                                                       const ShaderVertexInputInfo& vs_input_info) {
+static PreparedVertexBuffers PrepareVertexBuffers(uint64_t submit_id, RenderCommandBuffer& buffer,
+                                                  const DrawCallInfo&          draw,
+                                                  const ShaderVertexInputInfo& vs_input_info) {
 	EXIT_IF(draw.name == nullptr);
 	(void)submit_id;
 
@@ -920,24 +996,28 @@ static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffe
 }
 
 static void CommitVertexBuffers(RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                                std::vector<BufferBinding>& bindings) {
-	for (uint32_t slot = 0; slot < bindings.size(); slot++) {
-		auto& binding = bindings[slot];
-		if (binding.owner != nullptr) {
-			buffer.RetainResourceUntilFence(binding.owner);
+                                PreparedVertexBuffers& prepared) {
+	for (uint32_t i = 0; i < prepared.owner_count; i++) {
+		if (prepared.owners[i] != nullptr) {
+			buffer.RetainResourceUntilFence(std::move(prepared.owners[i]));
 		}
-		EXIT_IF(binding.buffer == nullptr);
-		vk_buffer.bindVertexBuffers(slot, 1, &binding.buffer, &binding.offset);
+	}
+	for (uint32_t i = 0; i < prepared.count; i++) {
+		EXIT_IF(prepared.buffers[i] == nullptr);
+	}
+	if (prepared.count != 0) {
+		vk_buffer.bindVertexBuffers(0, prepared.count, prepared.buffers.data(),
+		                            prepared.offsets.data());
 	}
 }
 
 static void CommitIndexBuffer(RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                              const PreparedIndexBuffer& prepared) {
+                              PreparedIndexBuffer& prepared) {
 	if (prepared.size == 0) {
 		return;
 	}
 	if (prepared.owner != nullptr) {
-		buffer.RetainResourceUntilFence(prepared.owner);
+		buffer.RetainResourceUntilFence(std::move(prepared.owner));
 	}
 	EXIT_IF(prepared.buffer == nullptr);
 	vk_buffer.bindIndexBuffer(prepared.buffer, prepared.offset, prepared.type);

src/graphics/host_gpu/renderer/renderDraw.h

@@ -15,12 +15,16 @@ struct ShaderVertexInputInfo;
 
 struct PipelineDynamicParameters {
 	bool stencil_test_enable = false;
+	bool depth_bias_enable   = false;
 
 	float viewport_scale[3]  = {};
 	float viewport_offset[3] = {};
 	int   scissor_ltrb[4]    = {0};
 
 	float    line_width                                       = 1.0f;
+	float    depth_bias_constant_factor                       = 0.0f;
+	float    depth_bias_clamp                                 = 0.0f;
+	float    depth_bias_slope_factor                          = 0.0f;
 	uint32_t color_write_count                                = 1;
 	bool     color_write_enable[RENDER_COLOR_ATTACHMENTS_MAX] = {true, true};
 
@@ -33,10 +37,6 @@ struct PipelineDynamicParameters {
 static_assert(std::is_trivially_copyable_v<PipelineDynamicParameters>);
 static_assert(std::is_standard_layout_v<PipelineDynamicParameters>);
 static_assert(alignof(PipelineDynamicParameters) == 1);
-static_assert(sizeof(PipelineDynamicParameters) ==
-              sizeof(bool) + sizeof(float[3]) + sizeof(float[3]) + sizeof(int[4]) + sizeof(float) +
-                  sizeof(uint32_t) + sizeof(bool[RENDER_COLOR_ATTACHMENTS_MAX]) +
-                  sizeof(PipelineStencilDynamicState) * 2);
 
 [[nodiscard]] int32_t ResolveVertexOffset(uint32_t                     index_offset,
                                           const ShaderVertexInputInfo& vs_input_info);

src/graphics/presentation/renderDoc.cpp

@@ -128,8 +128,7 @@ static bool BindRenderDocApi(HMODULE module) {
 	g_api    = static_cast<RenderDocApi*>(api);
 
 	g_api->SetCaptureFilePathTemplate("_RenderDoc/kyty");
-	static RenderDocInputButton capture_keys[] = {eRENDERDOC_Key_F1};
-	g_api->SetCaptureKeys(capture_keys, 1);
+	g_api->SetCaptureKeys(nullptr, 0);
 	g_api->UnloadCrashHandler();
 
 	char module_path[MAX_PATH] = {};
@@ -170,8 +169,7 @@ static bool BindRenderDocApi(HMODULE module) {
 	g_api    = static_cast<RenderDocApi*>(api);
 
 	g_api->SetCaptureFilePathTemplate("_RenderDoc/kyty");
-	static RenderDocInputButton capture_keys[] = {eRENDERDOC_Key_F1};
-	g_api->SetCaptureKeys(capture_keys, 1);
+	g_api->SetCaptureKeys(nullptr, 0);
 	g_api->UnloadCrashHandler();
 
 	Dl_info info {};

src/graphics/presentation/videoOut.cpp

@@ -537,10 +537,10 @@ Graphics::ImageInfo BufferAttributeGroup::ImageInfo(const VideoOutBuffer& buffer
 	const auto tile_mode =
 	    Graphics::Prospero::GpuEnumValue(Graphics::Prospero::TileMode::kRenderTarget);
 	const auto pitch =
-	    Graphics::TileGetTexturePitch(pixel_format.guest_format, attribute.width, 1, tile_mode);
+	    Graphics::TileGetTexturePitch(pixel_format.guest_format, attribute.width, tile_mode);
 	Graphics::TileSizeAlign total {};
 	Graphics::TileGetTextureTotalSize(pixel_format.guest_format, attribute.width, attribute.height,
-	                                  1, pitch, 1, tile_mode, false, total);
+	                                  1, 1, tile_mode, false, total);
 	if (total.size == 0 || total.align != 65536 ||
 	    (buffer.data_address & (total.align - 1u)) != 0) {
 		EXIT("invalid video-out surface footprint or alignment\n");
@@ -788,14 +788,10 @@ void VideoOutDriver::Impl::PresentThread(std::stop_token token) {
 	while (!token.stop_requested()) {
 		const auto sleep_begin = Common::Timer::QueryPerformanceCounter();
 		if (total_wait > 0) {
-			const auto sleep_end = sleep_begin + static_cast<uint64_t>(total_wait);
-			auto       now       = sleep_begin;
-			while (!token.stop_requested() && now < sleep_end) {
-				const auto remaining_us = (sleep_end - now) * 1000000u / frequency;
-				Common::Thread::SleepMicro(
-				    static_cast<uint32_t>(std::clamp<uint64_t>(remaining_us, 1, 1000)));
-				now = Common::Timer::QueryPerformanceCounter();
-			}
+			const auto remaining_us =
+			    (static_cast<uint64_t>(total_wait) * 1000000u + frequency - 1) / frequency;
+			Common::Thread::SleepMicro(static_cast<uint32_t>(
+			    std::clamp<uint64_t>(remaining_us, 1, std::numeric_limits<uint32_t>::max())));
 		}
 		if (token.stop_requested()) {
 			break;

src/graphics/presentation/window/swapchain.cpp

@@ -387,9 +387,11 @@ struct Presenter::Impl {
 		desc.view_info.usage       = vk::ImageUsageFlagBits::eTransferSrc;
 		desc.type                  = TextureCache::BindingType::VideoOut;
 
-		auto& cache           = renderer.GetTextureCache();
-		auto& image           = cache.GetImage(cache.FindImage(desc));
+		auto&      cache      = renderer.GetTextureCache();
+		const auto image_id   = cache.FindImage(desc);
+		auto&      image      = cache.GetImage(image_id);
 		image.usage.video_out = true;
+		cache.UpdateImage(image_id);
 		return image;
 	}
 

src/graphics/presentation/window/vulkanWindow.cpp

@@ -260,6 +260,10 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 			LOGF("sampleRateShading is not supported\n");
 			skip_device = true;
 		}
+		if (device_features2.features.depthBiasClamp != VK_TRUE) {
+			LOGF("depthBiasClamp is not supported\n");
+			skip_device = true;
+		}
 
 		if (device_features2.features.fragmentStoresAndAtomics != VK_TRUE) {
 			LOGF("fragmentStoresAndAtomics is not supported\n");
@@ -549,6 +553,7 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	EXIT_NOT_IMPLEMENTED(required_features13.synchronization2 == VK_TRUE &&
 	                     supported_features13.synchronization2 != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.sampleRateShading != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features2.features.depthBiasClamp != VK_TRUE);
 	features12.timelineSemaphore = VK_TRUE;
 
 	vk::PhysicalDeviceFeatures device_features {};
@@ -564,6 +569,7 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	device_features.independentBlend                     = VK_TRUE;
 	device_features.tessellationShader                   = VK_TRUE;
 	device_features.sampleRateShading                    = VK_TRUE;
+	device_features.depthBiasClamp                       = VK_TRUE;
 	graphics.sample_rate_shading_enabled                 = true;
 	device_features.vertexPipelineStoresAndAtomics =
 	    supported_features2.features.vertexPipelineStoresAndAtomics;

src/graphics/presentation/window/window.cpp

@@ -752,26 +752,18 @@ void WindowContext::Run() {
 #if defined(__APPLE__)
 		DrainMainThreadTasks();
 #endif
-		if (SDL_PollEvent(&loop.event) != 0) {
-			ProcessEvent(timer.GetTimeS());
-			continue;
-		}
-
 		if (loop.paused.load(std::memory_order_acquire)) {
 			if (!timer.IsPaused()) {
 				timer.Pause();
 			}
-			if (SDL_WaitEvent(&loop.event) == 0) {
-				EXIT("%s\n", SDL_GetError());
-			}
-			ProcessEvent(timer.GetTimeS());
-			continue;
-		}
-
-		if (timer.IsPaused()) {
+		} else if (timer.IsPaused()) {
 			timer.Resume();
 		}
-		Common::Thread::SleepMicro(1000);
+
+		if (SDL_WaitEvent(&loop.event) == 0) {
+			EXIT("%s\n", SDL_GetError());
+		}
+		ProcessEvent(timer.GetTimeS());
 	}
 }
 
@@ -946,13 +938,6 @@ void WindowContext::UpdateTitle() {
 	static constexpr auto build_type = "Unknown";
 #endif
 
-#if defined(KYTY_OFFICIAL_BUILD)
-	static constexpr auto build_label = "Official build " KYTY_RELEASE_TAG;
-#elif defined(KYTY_FORK_BUILD)
-	static constexpr auto build_label = "Fork build " KYTY_BUILD_REPOSITORY " " KYTY_GIT_HASH;
-#else
-	static constexpr auto build_label = "Source build " KYTY_GIT_HASH;
-#endif
 
 	const auto now       = Common::Timer::QueryPerformanceCounter();
 	const auto frequency = Common::Timer::QueryPerformanceFrequency();
@@ -965,7 +950,7 @@ void WindowContext::UpdateTitle() {
 		fps_frames  = 0;
 	}
 
-	auto fps = fmt::format("[{} | {}] {}{}{}{}{}{}[{}] [{}], frame: {}, fps: {:f}", build_label,
+	auto fps = fmt::format("[{} | {}] {}{}{}{}{}{}[{}] [{}], frame: {}, fps: {:f}", KYTY_BUILD_LABEL,
 	                       build_type, (has_title ? title : ""), (has_title ? ", " : ""),
 	                       (has_title_id ? title_id : ""), (has_title_id ? ", " : ""),
 	                       (has_app_ver ? app_ver : ""), (has_app_ver ? " " : ""), device_name,

src/graphics/shader/recompiler/ir/SrtWalker.cpp

@@ -562,8 +562,7 @@ static bool EvaluateRuntimeSourcesImpl(const Program&
 		const auto  dynamic =
 		    std::find(program.srt.dynamic_sources.begin(), program.srt.dynamic_sources.end(),
 		              request.source) != program.srt.dynamic_sources.end();
-		if (descriptor == nullptr || dynamic ||
-		    !DescriptorSourceResolved(program, request.source)) {
+		if (descriptor == nullptr || dynamic) {
 			if (error != nullptr) {
 				*error = Diagnostic(program, request.use_pc,
 				                    fmt::format("descriptor source {} is {}", request.source,

src/graphics/shader/shader.cpp

@@ -302,7 +302,7 @@ static void vs_check(const HW::VertexShaderInfo& vs, const HW::ShaderRegisters&
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.ieee_mode != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.cu_group_enable != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.require_forward_progress != false);
-		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.lds_configuration != false);
+		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.threadgroup_configuration != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.gs_vgpr_component_count != 3);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc1.fp16_overflow != false);
 		EXIT_NOT_IMPLEMENTED(vs.gs_regs.rsrc2.scratch_en != false);
@@ -374,7 +374,7 @@ static void ps_check(const HW::PsStageRegisters& ps, const HW::ShaderRegisters&
 			     ps.rsrc2.extra_lds_size);
 		}
 	}
-	EXIT_NOT_IMPLEMENTED(ps.rsrc2.raster_ordered_shading != false);
+	EXIT_NOT_IMPLEMENTED(ps.rsrc2.raster_ordered_shading != 0);
 	EXIT_NOT_IMPLEMENTED(ps.rsrc2.shared_vgprs != 0);
 
 	if (sh.shader_z_format != 0x00000000 && sh.shader_z_format != 0x00000001 &&

src/kernel/memoryAddressSpace.inc

@@ -22,6 +22,13 @@ static bool GuestFreeRangeContains(uint64_t base, uint64_t range_size, uint64_t
 	return candidate <= range_end && size <= range_end - candidate;
 }
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+static bool CoalescePlaceholders(uint64_t vaddr, uint64_t size) {
+	return size != 0 && VirtualFree(reinterpret_cast<void*>(vaddr), size,
+	                                MEM_RELEASE | MEM_COALESCE_PLACEHOLDERS) != 0;
+}
+#endif
+
 class GuestBackingStore {
 public:
 	enum class FailureReason {
@@ -264,6 +271,19 @@ public:
 			removed.push_back(entry);
 		}
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+		// Each unmapped file view becomes a separate Windows placeholder. The address-space owner
+		// records this contiguous guest span as one free range, so normalize fragmented backing
+		// views into one placeholder before publishing it there.
+		if (removed.size() > 1 && !CoalescePlaceholders(vaddr, size)) {
+			for (auto it = removed.rbegin(); it != removed.rend(); ++it) {
+				EXIT_IF(!MapExistingPlaceholderFixed(it->vaddr, it->size, it->backing_offset,
+				                                     it->mode));
+			}
+			return false;
+		}
+#endif
+
 		if (placeholder_preserved != nullptr) {
 			*placeholder_preserved = true;
 		}
@@ -832,6 +852,8 @@ public:
 			EXIT_IF(VirtualFree(reinterpret_cast<void*>(region_vaddr), region_size,
 			                    MEM_RELEASE | MEM_PRESERVE_PLACEHOLDER) == 0);
 		}
+		// ReleaseCommitted also publishes the aggregate span as one logical free range.
+		EXIT_IF(regions.size() > 1 && !CoalescePlaceholders(vaddr, size));
 #else
 		for (const auto& [region_vaddr, region_size]: regions) {
 			RestorePlaceholder(region_vaddr, region_size);
@@ -1252,8 +1274,7 @@ private:
 
 		if (base != it->first || end != it->first + it->second) {
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-			if (VirtualFree(reinterpret_cast<void*>(base), end - base,
-			                MEM_RELEASE | MEM_COALESCE_PLACEHOLDERS) != 0) {
+			if (CoalescePlaceholders(base, end - base)) {
 				m_free.erase(first, last);
 				m_free.emplace(base, end - base);
 			}

src/kernel/pthread.cpp

@@ -1849,8 +1849,9 @@ static int PthreadMutexInitNamed(PthreadMutex* mutex, const PthreadMutexattr* at
 	}
 }
 
-int KYTY_SYSV_ABI PthreadMutexInit(PthreadMutex* mutex, const PthreadMutexattr* attr) {
-	return PthreadMutexInitNamed(mutex, attr, "");
+int KYTY_SYSV_ABI PthreadMutexInit(PthreadMutex* mutex, const PthreadMutexattr* attr,
+                                   const char* name) {
+	return PthreadMutexInitNamed(mutex, attr, name);
 }
 
 int KYTY_SYSV_ABI PthreadMutexDestroy(PthreadMutex* mutex) {
@@ -2443,8 +2444,9 @@ static int PthreadRwlockInitNamed(PthreadRwlock* rwlock, const PthreadRwlockattr
 	return OK;
 }
 
-int KYTY_SYSV_ABI PthreadRwlockInit(PthreadRwlock* rwlock, const PthreadRwlockattr* attr) {
-	return PthreadRwlockInitNamed(rwlock, attr, "");
+int KYTY_SYSV_ABI PthreadRwlockInit(PthreadRwlock* rwlock, const PthreadRwlockattr* attr,
+                                    const char* name) {
+	return PthreadRwlockInitNamed(rwlock, attr, name);
 }
 
 static PthreadRwlockPrivate::Reader* RwlockFindReader(PthreadRwlock rwlock, Pthread thread) {
@@ -2938,8 +2940,9 @@ static int PthreadCondInitNamed(PthreadCond* cond, const PthreadCondattr* attr,
 	}
 }
 
-int KYTY_SYSV_ABI PthreadCondInit(PthreadCond* cond, const PthreadCondattr* attr) {
-	return PthreadCondInitNamed(cond, attr, "");
+int KYTY_SYSV_ABI PthreadCondInit(PthreadCond* cond, const PthreadCondattr* attr,
+                                  const char* name) {
+	return PthreadCondInitNamed(cond, attr, name);
 }
 
 int KYTY_SYSV_ABI PthreadCondSignal(PthreadCond* cond) {
@@ -4417,6 +4420,13 @@ int KYTY_SYSV_ABI pthread_cond_signal(LibKernel::PthreadCond* cond) {
 	return POSIX_PTHREAD_CALL(LibKernel::PthreadCondSignal(cond));
 }
 
+int KYTY_SYSV_ABI pthread_cond_init(LibKernel::PthreadCond*           cond,
+                                    const LibKernel::PthreadCondattr* attr) {
+	PRINT_NAME();
+
+	return POSIX_PTHREAD_CALL(LibKernel::PthreadCondInit(cond, attr, nullptr));
+}
+
 int KYTY_SYSV_ABI pthread_condattr_setclock(LibKernel::PthreadCondattr* attr,
                                             LibKernel::KernelClockid    clock_id) {
 	PRINT_NAME();
@@ -4525,6 +4535,13 @@ int KYTY_SYSV_ABI pthread_rwlock_destroy(LibKernel::PthreadRwlock* rwlock) {
 	return POSIX_PTHREAD_CALL(LibKernel::PthreadRwlockDestroy(rwlock));
 }
 
+int KYTY_SYSV_ABI pthread_rwlock_init(LibKernel::PthreadRwlock*           rwlock,
+                                      const LibKernel::PthreadRwlockattr* attr) {
+	PRINT_NAME();
+
+	return POSIX_PTHREAD_CALL(LibKernel::PthreadRwlockInit(rwlock, attr, nullptr));
+}
+
 int KYTY_SYSV_ABI pthread_key_create(LibKernel::PthreadKey*                   key,
                                      LibKernel::pthread_key_destructor_func_t destructor) {
 	PRINT_NAME();
@@ -4560,7 +4577,7 @@ int KYTY_SYSV_ABI pthread_mutex_init(LibKernel::PthreadMutex*           mutex,
                                      const LibKernel::PthreadMutexattr* attr) {
 	PRINT_NAME();
 
-	return POSIX_PTHREAD_CALL(LibKernel::PthreadMutexInit(mutex, attr));
+	return POSIX_PTHREAD_CALL(LibKernel::PthreadMutexInit(mutex, attr, nullptr));
 }
 
 int KYTY_SYSV_ABI pthread_mutexattr_init(LibKernel::PthreadMutexattr* attr) {

src/kernel/pthread.h

@@ -77,7 +77,8 @@ int KYTY_SYSV_ABI PthreadMutexattrInit(PthreadMutexattr* attr);
 int KYTY_SYSV_ABI PthreadMutexattrDestroy(PthreadMutexattr* attr);
 int KYTY_SYSV_ABI PthreadMutexattrSettype(PthreadMutexattr* attr, int type);
 int KYTY_SYSV_ABI PthreadMutexattrSetprotocol(PthreadMutexattr* attr, int protocol);
-int KYTY_SYSV_ABI PthreadMutexInit(PthreadMutex* mutex, const PthreadMutexattr* attr);
+int KYTY_SYSV_ABI PthreadMutexInit(PthreadMutex* mutex, const PthreadMutexattr* attr,
+                                   const char* name);
 int KYTY_SYSV_ABI PthreadMutexDestroy(PthreadMutex* mutex);
 int KYTY_SYSV_ABI PthreadMutexLock(PthreadMutex* mutex);
 int KYTY_SYSV_ABI PthreadMutexTrylock(PthreadMutex* mutex);
@@ -159,7 +160,8 @@ int KYTY_SYSV_ABI PthreadAttrSetstackaddr(PthreadAttr* attr, void* addr);
 int KYTY_SYSV_ABI PthreadAttrSetstacksize(PthreadAttr* attr, size_t stack_size);
 
 int KYTY_SYSV_ABI PthreadRwlockDestroy(PthreadRwlock* rwlock);
-int KYTY_SYSV_ABI PthreadRwlockInit(PthreadRwlock* rwlock, const PthreadRwlockattr* attr);
+int KYTY_SYSV_ABI PthreadRwlockInit(PthreadRwlock* rwlock, const PthreadRwlockattr* attr,
+                                    const char* name);
 int KYTY_SYSV_ABI PthreadRwlockRdlock(PthreadRwlock* rwlock);
 int KYTY_SYSV_ABI PthreadRwlockTimedrdlock(PthreadRwlock* rwlock, KernelUseconds usec);
 int KYTY_SYSV_ABI PthreadRwlockTimedwrlock(PthreadRwlock* rwlock, KernelUseconds usec);
@@ -177,7 +179,7 @@ int KYTY_SYSV_ABI PthreadCondattrInit(PthreadCondattr* attr);
 int KYTY_SYSV_ABI PthreadCondattrSetclock(PthreadCondattr* attr, KernelClockid clock_id);
 int KYTY_SYSV_ABI PthreadCondBroadcast(PthreadCond* cond);
 int KYTY_SYSV_ABI PthreadCondDestroy(PthreadCond* cond);
-int KYTY_SYSV_ABI PthreadCondInit(PthreadCond* cond, const PthreadCondattr* attr);
+int KYTY_SYSV_ABI PthreadCondInit(PthreadCond* cond, const PthreadCondattr* attr, const char* name);
 int KYTY_SYSV_ABI PthreadCondSignal(PthreadCond* cond);
 int KYTY_SYSV_ABI PthreadCondSignalto(PthreadCond* cond, Pthread thread);
 int KYTY_SYSV_ABI PthreadCondTimedwait(PthreadCond* cond, PthreadMutex* mutex, KernelUseconds usec);
@@ -267,6 +269,8 @@ int KYTY_SYSV_ABI pthread_rwlock_rdlock(LibKernel::PthreadRwlock* rwlock);
 int KYTY_SYSV_ABI pthread_rwlock_unlock(LibKernel::PthreadRwlock* rwlock);
 int KYTY_SYSV_ABI pthread_rwlock_wrlock(LibKernel::PthreadRwlock* rwlock);
 int KYTY_SYSV_ABI pthread_rwlock_destroy(LibKernel::PthreadRwlock* rwlock);
+int KYTY_SYSV_ABI pthread_rwlock_init(LibKernel::PthreadRwlock*           rwlock,
+                                      const LibKernel::PthreadRwlockattr* attr);
 int KYTY_SYSV_ABI pthread_key_create(LibKernel::PthreadKey*                   key,
                                      LibKernel::pthread_key_destructor_func_t destructor);
 int KYTY_SYSV_ABI pthread_key_delete(LibKernel::PthreadKey key);

src/kernel/syncOnAddress.cpp

@@ -119,6 +119,11 @@ int WaitLinux(volatile T* address, T expected, const uint32_t* timeout_micros,
 }
 
 int WakeLinux(volatile void* address, int32_t count) {
+	// The legacy FUTEX_WAKE path may wake one waiter when count is zero.
+	if (count == 0) {
+		return OK;
+	}
+
 	const auto result = syscall(SYS_futex, const_cast<void*>(address), FUTEX_WAKE_PRIVATE, count,
 	                            nullptr, nullptr, 0);
 	return result < 0 ? KERNEL_ERROR_EINVAL : OK;

src/kytyGitVersion.h.in

@@ -1,2 +1,13 @@
+#include "cmake_config.h"
+
 #define KYTY_GIT_VERSION "@KYTY_GIT_VERSION@"
 #define KYTY_GIT_HASH "@KYTY_GIT_HASH@"
+
+/* Build provenance, shown in the window title and in crash reports. */
+#if defined(KYTY_OFFICIAL_BUILD)
+#define KYTY_BUILD_LABEL "Official build " KYTY_RELEASE_TAG
+#elif defined(KYTY_FORK_BUILD)
+#define KYTY_BUILD_LABEL "Fork build " KYTY_BUILD_REPOSITORY " " KYTY_GIT_HASH
+#else
+#define KYTY_BUILD_LABEL "Source build " KYTY_GIT_HASH
+#endif

src/launcher/forms/configuration_edit_dialog.ui

@@ -88,6 +88,16 @@
            </property>
           </widget>
          </item>
+         <item row="2" column="1">
+          <widget class="QCheckBox" name="checkBox_red_zone_protection">
+           <property name="toolTip">
+            <string>Helps prevent crashes in some games on Windows</string>
+           </property>
+           <property name="text">
+            <string>Windows SysV red zone crash protection (Experimental)</string>
+           </property>
+          </widget>
+         </item>
         </layout>
        </widget>
       </item>

src/launcher/include/configuration.h

@@ -99,7 +99,10 @@ public:
 	QString                printf_output_file          = "_kyty.txt";
 	ProfilerDirection      profiler_direction          = ProfilerDirection::None;
 	bool                   renderdoc_enabled           = false;
-	QStringList            host_input_mapping;
+#if defined(_WIN32)
+	bool red_zone_protection_enabled = false;
+#endif
+	QStringList host_input_mapping;
 
 	QString elf = QStringLiteral("eboot.bin");
 
@@ -119,7 +122,10 @@ public:
 		printf_output_file          = other.printf_output_file;
 		profiler_direction          = other.profiler_direction;
 		renderdoc_enabled           = other.renderdoc_enabled;
-		host_input_mapping          = other.host_input_mapping;
+#if defined(_WIN32)
+		red_zone_protection_enabled = other.red_zone_protection_enabled;
+#endif
+		host_input_mapping = other.host_input_mapping;
 	}
 
 	void CopyFrom(const Configuration& other) {
@@ -156,6 +162,9 @@ public:
 		KYTY_CFG_SET(printf_output_file);
 		KYTY_CFG_SET(profiler_direction);
 		KYTY_CFG_SET(renderdoc_enabled);
+#if defined(_WIN32)
+		KYTY_CFG_SET(red_zone_protection_enabled);
+#endif
 		s->setValue("host_input_mapping", host_input_mapping);
 		KYTY_CFG_SET(elf);
 	}
@@ -183,6 +192,10 @@ public:
 		KYTY_CFG_GET(printf_output_file);
 		KYTY_CFG_GET(profiler_direction);
 		KYTY_CFG_GET(renderdoc_enabled);
+#if defined(_WIN32)
+		red_zone_protection_enabled =
+		    s->value("red_zone_protection_enabled", red_zone_protection_enabled).toBool();
+#endif
 		host_input_mapping = s->value("host_input_mapping", host_input_mapping).toStringList();
 		elf                = s->value("elf", elf).toString();
 	}

src/launcher/src/configurationEditDialog.cpp

@@ -164,6 +164,11 @@ void ConfigurationEditDialog::Init(const Configuration& info) {
 	m_ui->checkBox_shader_validation->setChecked(info.shader_validation_enabled);
 	m_ui->checkBox_vulkan_validation->setChecked(info.vulkan_validation_enabled);
 	m_ui->checkBox_renderdoc_capture->setChecked(info.renderdoc_enabled);
+#if defined(_WIN32)
+	m_ui->checkBox_red_zone_protection->setChecked(info.red_zone_protection_enabled);
+#else
+	m_ui->checkBox_red_zone_protection->setVisible(false);
+#endif
 	ListInit(m_ui->comboBox_shader_optimization_type, info.shader_optimization_type);
 	ListInit(m_ui->comboBox_shader_log_direction, info.shader_log_direction);
 	m_ui->lineEdit_shader_log_folder->setText(info.shader_log_folder);
@@ -286,7 +291,10 @@ static void UpdateInfo(Configuration& info, Ui::ConfigurationEditDialog& ui) {
 	info.vulkan_validation_enabled = ui.checkBox_vulkan_validation->isChecked();
 	info.shader_validation_enabled = ui.checkBox_shader_validation->isChecked();
 	info.renderdoc_enabled         = ui.checkBox_renderdoc_capture->isChecked();
-	info.shader_optimization_type  = TextToEnum<Configuration::ShaderOptimizationType>(
+#if defined(_WIN32)
+	info.red_zone_protection_enabled = ui.checkBox_red_zone_protection->isChecked();
+#endif
+	info.shader_optimization_type = TextToEnum<Configuration::ShaderOptimizationType>(
 	    ui.comboBox_shader_optimization_type->currentText());
 	info.shader_log_direction = TextToEnum<Configuration::ShaderLogDirection>(
 	    ui.comboBox_shader_log_direction->currentText());

src/launcher/src/mainDialog.cpp

@@ -220,6 +220,11 @@ static QStringList CreateEmulatorArgs(const Configuration& info) {
 	args << "--printf-output-file" << info.printf_output_file;
 	args << "--profiler-direction" << EnumToText(info.profiler_direction);
 	args << "--spirv-debug-printf" << "false";
+#if defined(_WIN32)
+	if (info.red_zone_protection_enabled) {
+		args << "--redzone";
+	}
+#endif
 	for (const auto& binding: info.host_input_mapping) {
 		args << "--keymap" << binding;
 	}

src/libs/controller.cpp

@@ -208,7 +208,8 @@ void GameController::AddState(const ControllerState& state) {
 void GameController::Button(int id, uint32_t button, bool down) {
 	Common::LockGuard lock(m_mutex);
 
-	if (m_active_id == id) {
+	// The keyboard shares the player-1 pad with the active gamepad.
+	if (m_active_id == id || id == HOST_INPUT_CONTROLLER_ID) {
 		auto state = GetLastState();
 
 		state.time = LibKernel::KernelGetProcessTime();
@@ -226,7 +227,7 @@ void GameController::Button(int id, uint32_t button, bool down) {
 void GameController::Axis(int id, Controller::Axis axis, int value) {
 	Common::LockGuard lock(m_mutex);
 
-	if (m_active_id == id) {
+	if (m_active_id == id || id == HOST_INPUT_CONTROLLER_ID) {
 		auto state = GetLastState();
 
 		state.time = LibKernel::KernelGetProcessTime();

src/libs/libC.cpp

@@ -481,13 +481,13 @@ static KYTY_SYSV_ABI int MtxInit(LibKernel::PthreadMutex* mtx, int type) {
 
 	constexpr int mtx_recursive = 0x100;
 	if ((type & mtx_recursive) == 0) {
-		return mtx_result(LibKernel::PthreadMutexInit(mtx, nullptr));
+		return mtx_result(LibKernel::PthreadMutexInit(mtx, nullptr, nullptr));
 	}
 
 	LibKernel::PthreadMutexattr attr   = nullptr;
 	int                         result = LibKernel::PthreadMutexattrInit(&attr);
 	result = (result == OK ? LibKernel::PthreadMutexattrSettype(&attr, 2) : result);
-	result = (result == OK ? LibKernel::PthreadMutexInit(mtx, &attr) : result);
+	result = (result == OK ? LibKernel::PthreadMutexInit(mtx, &attr, nullptr) : result);
 	(void)LibKernel::PthreadMutexattrDestroy(&attr);
 
 	return mtx_result(result);
@@ -659,8 +659,8 @@ void KYTY_SYSV_ABI LibcHeapGetTraceInfo(Info* info) {
 	info->mstate_table          = g_mstate_table;
 }
 
-int KYTY_SYSV_ABI LibcInternalExtCxaThreadAtexit(thread_atexit_destructor_t destructor, void* object,
-                                                 void* /*module_id*/) {
+int KYTY_SYSV_ABI LibcInternalExtCxaThreadAtexit(thread_atexit_destructor_t destructor,
+                                                 void* object, void* /*module_id*/) {
 	PRINT_NAME();
 
 	g_thread_atexit_destructors.push_back({destructor, object});

src/libs/libKernel.cpp

@@ -516,6 +516,7 @@ static void QueuePendingSignal(Pthread thread, int signum) {
 	PthreadQueuePendingSignal(thread, signum);
 }
 
+#if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
 static void WaitForSignalDispatch(Pthread thread, int signum) {
 	constexpr auto DISPATCH_WAIT_STEP = std::chrono::microseconds(1000);
 	constexpr auto DISPATCH_WAIT_MAX  = std::chrono::milliseconds(2);
@@ -526,6 +527,7 @@ static void WaitForSignalDispatch(Pthread thread, int signum) {
 		waited += DISPATCH_WAIT_STEP;
 	}
 }
+#endif
 
 static bool TakePendingSignal(Pthread thread, int signum) {
 	return PthreadTakePendingSignal(thread, signum);
@@ -1070,60 +1072,6 @@ static void SignalApcHandler(void* arg1, void* arg2, void* /*arg3*/, PCONTEXT co
 	}
 }
 
-static bool DispatchSignalWithSuspendedThreadContext(HANDLE target_thread, Pthread thread,
-                                                     int signum, exception_handler_func_t handler) {
-	if (target_thread == nullptr || thread == nullptr || handler == nullptr) {
-		return false;
-	}
-
-	if (!PthreadHasPendingSignal(thread, signum)) {
-		return true;
-	}
-
-	if (SuspendThread(target_thread) == static_cast<DWORD>(-1)) {
-		return false;
-	}
-
-	bool    dispatched = false;
-	CONTEXT host_ctx {};
-	host_ctx.ContextFlags = CONTEXT_FULL | CONTEXT_SEGMENTS;
-	if (GetThreadContext(target_thread, &host_ctx) != FALSE) {
-		if (TakePendingSignal(thread, signum)) {
-			auto   ctx                  = CreateSignalUcontext(&host_ctx);
-			auto   guest_context        = IsGuestCodeAddress(ctx.uc_mcontext.mc_rip);
-			HANDLE helper_thread_handle = nullptr;
-			if (guest_context) {
-				if (DuplicateHandle(GetCurrentProcess(), target_thread, GetCurrentProcess(),
-				                    &helper_thread_handle, THREAD_SUSPEND_RESUME, FALSE,
-				                    0) == FALSE) {
-					ResumeThread(target_thread);
-					return false;
-				}
-			} else {
-				SanitizeNonGuestSignalUcontext(&ctx, thread);
-				ResumeThread(target_thread);
-			}
-
-			std::thread([helper_thread_handle, thread, signum, handler, ctx]() mutable {
-				SignalDispatchScope scope;
-				auto*               previous_self = PthreadSwapSelfForSignal(thread);
-				handler(signum, &ctx);
-				PthreadSwapSelfForSignal(previous_self);
-				if (helper_thread_handle != nullptr) {
-					ResumeThread(helper_thread_handle);
-					CloseHandle(helper_thread_handle);
-				}
-			}).detach();
-			dispatched = true;
-			return true;
-		}
-		dispatched = true;
-	}
-
-	ResumeThread(target_thread);
-
-	return dispatched;
-}
 #endif
 
 void KernelDispatchPendingSignalForCurrentThread() {
@@ -1206,8 +1154,7 @@ static int KYTY_SYSV_ABI KernelRaiseException(Pthread thread, int signum) {
 		}
 
 		HANDLE target_thread =
-		    OpenThread(THREAD_SET_CONTEXT | THREAD_SUSPEND_RESUME | THREAD_GET_CONTEXT, FALSE,
-		               static_cast<DWORD>(target_thread_id));
+		    OpenThread(THREAD_SET_CONTEXT, FALSE, static_cast<DWORD>(target_thread_id));
 		if (target_thread == nullptr) {
 			return KERNEL_ERROR_EINVAL;
 		}
@@ -1231,10 +1178,6 @@ static int KYTY_SYSV_ABI KernelRaiseException(Pthread thread, int signum) {
 
 		Common::CondVar::SignalThread(PthreadGetUniqueId(thread));
 		PthreadWakeForSignal(thread);
-		WaitForSignalDispatch(thread, signum);
-		if (PthreadHasPendingSignal(thread, signum)) {
-			DispatchSignalWithSuspendedThreadContext(target_thread, thread, signum, handler);
-		}
 		CloseHandle(target_thread);
 		return OK;
 #elif defined(__x86_64__)
@@ -2229,9 +2172,11 @@ LIB_DEFINE(InitLibKernel_1_Posix) {
 	LIB_FUNC("ltCfaGr2JGE", Posix::pthread_mutex_destroy);
 	LIB_FUNC("mkx2fVhNMsg", Posix::pthread_cond_broadcast);
 	LIB_FUNC("2MOy+rUfuhQ", Posix::pthread_cond_signal);
+	LIB_FUNC("0TyVk4MSLt0", Posix::pthread_cond_init);
 	LIB_FUNC("Op8TBGY5KHg", Posix::pthread_cond_wait);
 	LIB_FUNC("Z4QosVuAsA0", Posix::pthread_once);
 	LIB_FUNC("1471ajPzxh0", Posix::pthread_rwlock_destroy);
+	LIB_FUNC("ytQULN-nhL4", Posix::pthread_rwlock_init);
 	LIB_FUNC("sIlRvQqsN2Y", Posix::pthread_rwlock_wrlock);
 	LIB_FUNC("SFxTMOfuCkE", LibKernel::PthreadRwlockTryrdlock);
 	LIB_FUNC("XhWHn6P5R7U", LibKernel::PthreadRwlockTrywrlock);
@@ -3268,7 +3213,6 @@ LIB_DEFINE(InitLibKernel_1_Pthread) {
 	LIB_FUNC("FXPWHNk8Of0", LibKernel::PthreadAttrGetschedparam);
 
 	LIB_FUNC("6ULAa0fq4jA", LibKernel::PthreadRwlockInit);
-	LIB_FUNC("ytQULN-nhL4", LibKernel::PthreadRwlockInit);
 	LIB_FUNC("BB+kb08Tl9A", LibKernel::PthreadRwlockDestroy);
 	LIB_FUNC("Ox9i0c7L5w0", LibKernel::PthreadRwlockRdlock);
 	LIB_FUNC("iGjsr1WAtI0", LibKernel::PthreadRwlockRdlock);
@@ -3284,7 +3228,6 @@ LIB_DEFINE(InitLibKernel_1_Pthread) {
 	LIB_FUNC("h-OifiouBd8", LibKernel::PthreadRwlockattrSettype);
 
 	LIB_FUNC("2Tb92quprl0", LibKernel::PthreadCondInit);
-	LIB_FUNC("0TyVk4MSLt0", LibKernel::PthreadCondInit);
 	LIB_FUNC("g+PZd2hiacg", LibKernel::PthreadCondDestroy);
 	LIB_FUNC("RXXqi4CtF8w", LibKernel::PthreadCondDestroy);
 	LIB_FUNC("WKAXJ4XBPQ4", LibKernel::PthreadCondWait);

src/loader/elf.h

@@ -53,6 +53,7 @@ constexpr Elf64_Half ET_DYNAMIC = 0xfe18; // Shared
 constexpr Elf64_Word PT_LOAD          = 1;
 constexpr Elf64_Word PT_DYNAMIC       = 2;
 constexpr Elf64_Word PT_TLS           = 7;
+constexpr Elf64_Word PT_GNU_EH_FRAME  = 0x6474e550;
 constexpr Elf64_Word PT_OS_DYNLIBDATA = 0x61000000;
 constexpr Elf64_Word PT_OS_PROCPARAM  = 0x61000001;
 constexpr Elf64_Word PT_OS_RELRO      = 0x61000010;

src/loader/redZonePatcher.h

@@ -0,0 +1,39 @@
+// SPDX-FileCopyrightText: Copyright 2026 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#ifndef KYTY_LOADER_RED_ZONE_PATCHER_H_
+#define KYTY_LOADER_RED_ZONE_PATCHER_H_
+
+#include "common/common.h"
+
+#include <span>
+#include <vector>
+
+namespace Loader {
+
+struct RedZonePatchResult {
+	uint64_t function_count                           = 0;
+	uint64_t instruction_count                        = 0;
+	uint64_t red_zone_function_count                  = 0;
+	uint64_t memory_instruction_count                 = 0;
+	uint64_t short_memory_instruction_count           = 0;
+	uint64_t patched_memory_instruction_count         = 0;
+	uint64_t stack_dependent_memory_instruction_count = 0;
+	uint64_t control_flow_memory_instruction_count    = 0;
+	uint64_t unrelocatable_memory_instruction_count   = 0;
+	uint64_t indirect_red_zone_function_count         = 0;
+};
+
+void RegisterRedZonePatchModule(void* module_ptr, uint64_t module_size, void* trampoline_area_ptr,
+                                uint64_t trampoline_area_size);
+void UnregisterRedZonePatchModule(void* module_ptr);
+
+RedZonePatchResult PatchRedZoneMemoryInstructions(uint64_t segment_addr, uint64_t segment_size,
+                                                  std::span<const uintptr_t> function_starts);
+
+bool DecodeEhFrameFunctionStarts(uint64_t eh_frame_header_addr, uint64_t eh_frame_header_size,
+                                 std::vector<uintptr_t>* function_starts);
+
+} // namespace Loader
+
+#endif /* KYTY_LOADER_RED_ZONE_PATCHER_H_ */

src/loader/runtimeLinker.cpp

@@ -19,6 +19,7 @@
 #include "loader/elf.h"
 #include "loader/gamePatch.h"
 #include "loader/jit.h"
+#include "loader/redZonePatcher.h"
 #include "loader/symbolDatabase.h"
 #include "loader/x64InstructionEmulator.h"
 
@@ -2202,17 +2203,39 @@ void RuntimeLinker::LoadProgramToMemory(Program* program) {
 	EXIT_IF(tls_handler_size > UINT64_MAX - program->base_size_aligned);
 	program->mapped_size = program->base_size_aligned + tls_handler_size;
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	const bool         use_red_zone_protection  = Config::RedZoneProtectionEnabled();
+	constexpr uint64_t RED_ZONE_TRAMPOLINE_SIZE = 8u * 1024u * 1024u;
+	if (use_red_zone_protection) {
+		EXIT_IF(RED_ZONE_TRAMPOLINE_SIZE > UINT64_MAX - program->mapped_size);
+		program->mapped_size += RED_ZONE_TRAMPOLINE_SIZE;
+	}
+#endif
+
 	program->base_vaddr = Libs::LibKernel::Memory::AllocateProgramMemory(
 	    g_desired_base_addr, program->mapped_size, Common::VirtualMemory::Mode::ExecuteReadWrite,
 	    Common::PathToString(program->file_name.filename()).c_str());
+	EXIT_IF(program->base_vaddr == 0);
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	if (use_red_zone_protection) {
+		program->red_zone_trampoline_vaddr = program->base_vaddr + program->base_size_aligned;
+		program->red_zone_trampoline_size  = RED_ZONE_TRAMPOLINE_SIZE;
+		RegisterRedZonePatchModule(reinterpret_cast<void*>(program->base_vaddr),
+		                           program->base_size_aligned,
+		                           reinterpret_cast<void*>(program->red_zone_trampoline_vaddr),
+		                           program->red_zone_trampoline_size);
+	}
+#endif
 	if (!is_shared) {
 		program->tls.handler_vaddr = program->base_vaddr + program->base_size_aligned;
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+		program->tls.handler_vaddr += program->red_zone_trampoline_size;
+#endif
 	}
 
 	g_desired_base_addr += CODE_BASE_INCR * (1 + program->mapped_size / CODE_BASE_INCR);
 
-	EXIT_IF(program->base_vaddr == 0);
 	EXIT_IF(program->base_size_aligned < program->base_size);
 	LOGF("base_vaddr             = 0x%016" PRIx64 "\n"
 	     "base_size              = 0x%016" PRIx64 "\n"
@@ -2228,6 +2251,11 @@ void RuntimeLinker::LoadProgramToMemory(Program* program) {
 	}
 
 	// program->elf->SetBaseVAddr(program->base_vaddr);
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	std::vector<std::pair<uint64_t, uint64_t>> executable_segments;
+	uint64_t                                   eh_frame_header_addr = 0;
+	uint64_t                                   eh_frame_header_size = 0;
+#endif
 
 	for (Elf64_Half i = 0; i < ehdr->e_phnum; i++) {
 		if (phdr[i].p_memsz != 0 && (phdr[i].p_type == PT_LOAD || phdr[i].p_type == PT_OS_RELRO)) {
@@ -2250,6 +2278,11 @@ void RuntimeLinker::LoadProgramToMemory(Program* program) {
 
 			if (Common::VirtualMemory::IsExecute(mode)) {
 				PatchProgram(program, segment_addr, segment_memory_size);
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+				if (use_red_zone_protection) {
+					executable_segments.emplace_back(segment_addr, segment_file_size);
+				}
+#endif
 			}
 
 			if (!skip_protect) {
@@ -2284,8 +2317,42 @@ void RuntimeLinker::LoadProgramToMemory(Program* program) {
 
 			program->proc_param_vaddr = phdr[i].p_vaddr + program->base_vaddr;
 		}
+
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+		if (use_red_zone_protection && phdr[i].p_type == PT_GNU_EH_FRAME) {
+			eh_frame_header_addr = phdr[i].p_vaddr + program->base_vaddr;
+			eh_frame_header_size = phdr[i].p_memsz;
+		}
+#endif
 	}
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	if (use_red_zone_protection) {
+		std::vector<uintptr_t> function_starts;
+		if (!DecodeEhFrameFunctionStarts(eh_frame_header_addr, eh_frame_header_size,
+		                                 &function_starts)) {
+			LOGF("Windows guest red-zone patching could not decode function boundaries for %s\n",
+			     Common::PathToString(program->file_name).c_str());
+		}
+		for (const auto& [segment_addr, segment_size]: executable_segments) {
+			const auto result =
+			    PatchRedZoneMemoryInstructions(segment_addr, segment_size, function_starts);
+			LOGF("Windows guest red-zone patching: %s, functions=%" PRIu64 ", red_zone=%" PRIu64
+			     ", memory=%" PRIu64 ", patched=%" PRIu64 ", short=%" PRIu64 ", stack=%" PRIu64
+			     ", control=%" PRIu64 ", unrelocatable=%" PRIu64 "\n",
+			     Common::PathToString(program->file_name.filename()).c_str(), result.function_count,
+			     result.red_zone_function_count, result.memory_instruction_count,
+			     result.patched_memory_instruction_count, result.short_memory_instruction_count,
+			     result.stack_dependent_memory_instruction_count,
+			     result.control_flow_memory_instruction_count,
+			     result.unrelocatable_memory_instruction_count);
+			Common::VirtualMemory::FlushInstructionCache(segment_addr, segment_size);
+		}
+		Common::VirtualMemory::FlushInstructionCache(program->red_zone_trampoline_vaddr,
+		                                             program->red_zone_trampoline_size);
+	}
+#endif
+
 	if (!is_shared) {
 		SetupTlsHandler(program);
 	}
@@ -2311,6 +2378,11 @@ void RuntimeLinker::DeleteProgram(Program* p) {
 
 	if (program->base_vaddr != 0 || program->mapped_size != 0) {
 		EXIT_IF(program->base_vaddr == 0 || program->mapped_size == 0);
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+		if (program->red_zone_trampoline_size != 0) {
+			UnregisterRedZonePatchModule(reinterpret_cast<void*>(program->base_vaddr));
+		}
+#endif
 		EXIT_IF(
 		    !Libs::LibKernel::Memory::FreeGuestMemory(program->base_vaddr, program->mapped_size));
 	}

src/loader/runtimeLinker.h

@@ -114,15 +114,19 @@ struct Program {
 	Program();
 	~Program();
 
-	int32_t                         unique_id = -1;
-	RuntimeLinker*                  rt        = nullptr;
-	std::filesystem::path           file_name;
-	std::unique_ptr<Elf64>          elf;
-	std::unique_ptr<DynamicInfo>    dynamic_info;
-	uint64_t                        base_vaddr        = 0;
-	uint64_t                        base_size         = 0;
-	uint64_t                        base_size_aligned = 0;
-	uint64_t                        mapped_size       = 0;
+	int32_t                      unique_id = -1;
+	RuntimeLinker*               rt        = nullptr;
+	std::filesystem::path        file_name;
+	std::unique_ptr<Elf64>       elf;
+	std::unique_ptr<DynamicInfo> dynamic_info;
+	uint64_t                     base_vaddr        = 0;
+	uint64_t                     base_size         = 0;
+	uint64_t                     base_size_aligned = 0;
+	uint64_t                     mapped_size       = 0;
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	uint64_t red_zone_trampoline_vaddr = 0;
+	uint64_t red_zone_trampoline_size  = 0;
+#endif
 	std::unique_ptr<SymbolDatabase> export_symbols;
 	std::unique_ptr<SymbolDatabase> import_symbols;
 	ThreadLocalStorage              tls;

src/main.cpp

@@ -64,6 +64,9 @@ static void PrintUsage() {
 	::printf("  --spirv-debug-printf <true|false>    Enable SPIR-V debug printf.\n");
 	::printf(
 	    "  --readback-linear-images <true|false> Read back writable linear images on submit.\n");
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+	::printf("  --redzone                            Protect the guest SysV red zone.\n");
+#endif
 	::printf("  --keymap <Pad=Key>                   DualSense mapping; may be repeated.\n");
 	::printf("  --rd                                 Enable RenderDoc capture.\n");
 }
@@ -138,6 +141,13 @@ static bool ParseArgs(int argc, char* argv[], RunOptions& options, bool& show_he
 			continue;
 		}
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+		if (arg == "--redzone") {
+			options.config.red_zone_protection_enabled = true;
+			continue;
+		}
+#endif
+
 		if (!Common::StartsWith(arg, "--")) {
 			::printf("game input must be provided with --game\n");
 			return false;

tests/MemoryTrackerTests.cpp

@@ -2,6 +2,8 @@
 #include "graphics/host_gpu/memoryTracker.h"
 #include "graphics/host_gpu/rangeSet.h"
 
+#include <atomic>
+#include <chrono>
 #include <cstdint>
 #include <cstdio>
 #include <cstdlib>
@@ -211,6 +213,40 @@ void TestQueriesDoNotRequireMappedOwnership() {
         "unowned tracker range did not expose its initial CPU-dirty state");
 }
 
+void TestConcurrentRegionPublication() {
+  TrackerHarness harness;
+  auto &tracker = harness.tracker;
+  auto &page_manager = harness.page_manager;
+  const auto page_size = page_manager.GetPageSize();
+  auto *memory = Allocate(page_manager, 1);
+  const auto address = reinterpret_cast<uint64_t>(memory);
+
+  std::binary_semaphore start_first{0};
+  std::binary_semaphore start_second{0};
+  std::atomic_uint32_t cpu_dirty_results{0};
+  std::jthread first([&] {
+    start_first.acquire();
+    if (tracker.IsRegionCpuModified(address, page_size)) {
+      cpu_dirty_results.fetch_add(1, std::memory_order_relaxed);
+    }
+  });
+  std::jthread second([&] {
+    start_second.acquire();
+    if (tracker.IsRegionCpuModified(address, page_size)) {
+      cpu_dirty_results.fetch_add(1, std::memory_order_relaxed);
+    }
+  });
+  start_first.release();
+  start_second.release();
+  first.join();
+  second.join();
+
+  tracker.UntrackMemory(address, page_size);
+  Release(page_manager, memory, page_size);
+  Check(cpu_dirty_results.load(std::memory_order_relaxed) == 2,
+        "concurrent region publication lost initial CPU ownership");
+}
+
 void TestCpuDirtyUpload() {
   TrackerHarness harness;
   auto &tracker = harness.tracker;
@@ -497,6 +533,114 @@ void TestCrossRegionUpload() {
   Release(page_manager, memory, region_size * 2);
 }
 
+void TestUploadDoesNotSerializeDisjointRegion() {
+  constexpr auto region_size = Libs::Graphics::TRACKER_REGION_SIZE;
+  constexpr auto page_size = Libs::Graphics::TRACKER_PAGE_SIZE;
+  TrackerHarness harness;
+  auto &tracker = harness.tracker;
+  auto &page_manager = harness.page_manager;
+  auto *memory = Allocate(page_manager, region_size * 2 / page_size);
+  const auto allocation_base = reinterpret_cast<uint64_t>(memory);
+  const auto second_region =
+      (allocation_base & ~(region_size - 1)) + region_size;
+  Check(second_region + page_size <= allocation_base + region_size * 2,
+        "test allocation does not span two tracker regions");
+
+  // Publish both managers before the concurrent section so this test measures
+  // tracker access serialization rather than manager allocation.
+  Check(tracker.IsRegionCpuModified(allocation_base, page_size) &&
+            tracker.IsRegionCpuModified(second_region, page_size),
+        "disjoint upload setup did not initialize both regions");
+
+  std::binary_semaphore upload_entered{0};
+  std::binary_semaphore finish_upload{0};
+  std::binary_semaphore query_finished{0};
+  std::atomic_bool query_result{false};
+  std::jthread uploader([&] {
+    tracker.ForEachUploadRange(
+        allocation_base, page_size, true, [](uint64_t, uint64_t) noexcept {},
+        [&]() noexcept {
+          upload_entered.release();
+          finish_upload.acquire();
+        });
+  });
+  upload_entered.acquire();
+  std::jthread query([&] {
+    query_result.store(tracker.IsRegionCpuModified(second_region, page_size),
+                       std::memory_order_relaxed);
+    query_finished.release();
+  });
+
+  const bool completed_while_upload_blocked =
+      query_finished.try_acquire_for(std::chrono::seconds(5));
+  finish_upload.release();
+  uploader.join();
+  query.join();
+
+  tracker.UnmarkRegionAsGpuModified(allocation_base, page_size);
+  tracker.MarkRegionAsCpuModified(allocation_base, page_size);
+  tracker.UntrackMemory(allocation_base, region_size * 2);
+  Release(page_manager, memory, region_size * 2);
+  Check(completed_while_upload_blocked &&
+            query_result.load(std::memory_order_relaxed),
+        "upload callback serialized an unrelated tracker region");
+}
+
+void TestDownloadDoesNotSerializeDisjointRegion() {
+  constexpr auto region_size = Libs::Graphics::TRACKER_REGION_SIZE;
+  constexpr auto page_size = Libs::Graphics::TRACKER_PAGE_SIZE;
+  TrackerHarness harness;
+  auto &tracker = harness.tracker;
+  auto &page_manager = harness.page_manager;
+  auto *memory = Allocate(page_manager, region_size * 2 / page_size);
+  const auto allocation_base = reinterpret_cast<uint64_t>(memory);
+  const auto second_region =
+      (allocation_base & ~(region_size - 1)) + region_size;
+  Check(second_region + page_size <= allocation_base + region_size * 2,
+        "test allocation does not span two tracker regions");
+
+  tracker.ForEachUploadRange(
+      allocation_base, page_size, true, [](uint64_t, uint64_t) noexcept {},
+      []() noexcept {});
+  tracker.ForEachUploadRange(
+      second_region, page_size, false, [](uint64_t, uint64_t) noexcept {},
+      []() noexcept {});
+
+  std::binary_semaphore download_entered{0};
+  std::binary_semaphore finish_download{0};
+  std::binary_semaphore mutation_finished{0};
+  std::jthread downloader([&] {
+    tracker.ForEachDownloadRange<false>(allocation_base, page_size,
+                                        [&](uint64_t, uint64_t) noexcept {
+                                          download_entered.release();
+                                          finish_download.acquire();
+                                        });
+  });
+  download_entered.acquire();
+  std::jthread mutation([&] {
+    tracker.MarkRegionAsGpuModified(second_region, page_size);
+    mutation_finished.release();
+  });
+
+  const bool completed_while_download_blocked =
+      mutation_finished.try_acquire_for(std::chrono::seconds(5));
+  finish_download.release();
+  downloader.join();
+  mutation.join();
+
+  const bool both_gpu_owned =
+      tracker.IsRegionGpuModified(allocation_base, page_size) &&
+      tracker.IsRegionGpuModified(second_region, page_size);
+  tracker.UnmarkRegionAsGpuModified(allocation_base, page_size);
+  tracker.UnmarkRegionAsGpuModified(second_region, page_size);
+  tracker.MarkRegionAsCpuModified(allocation_base, page_size);
+  tracker.MarkRegionAsCpuModified(second_region, page_size);
+  tracker.UntrackMemory(allocation_base, region_size * 2);
+  Release(page_manager, memory, region_size * 2);
+  Check(completed_while_download_blocked && both_gpu_owned,
+        "download callback serialized an unrelated tracker region");
+}
+
 void TestGpuUnmarkUsesRegionMask() {
   constexpr auto region_size = Libs::Graphics::TRACKER_REGION_SIZE;
   constexpr auto page_size = Libs::Graphics::TRACKER_PAGE_SIZE;
@@ -683,6 +827,7 @@ int main(int argc, char **argv) {
   }
   TestRangeSet();
   TestQueriesDoNotRequireMappedOwnership();
+  TestConcurrentRegionPublication();
   TestCpuDirtyUpload();
   TestRangeInvalidation();
   TestGpuReacquisitionAfterInvalidation();
@@ -690,6 +835,8 @@ int main(int argc, char **argv) {
   TestExactDirtyIntervalsSharingTrackerPage();
   TestGpuDownloadProtectionMirrors();
   TestCrossRegionUpload();
+  TestUploadDoesNotSerializeDisjointRegion();
+  TestDownloadDoesNotSerializeDisjointRegion();
   TestGpuUnmarkUsesRegionMask();
   TestFullRegionGpuUnmarkBatching();
   TestFatalPaths();

tests/ResourceMutexTests.cpp

@@ -1,120 +0,0 @@
-#include "graphics/host_gpu/renderer/cache/resourceMutex.h"
-
-#include <atomic>
-#include <cstdio>
-#include <cstdlib>
-#include <cstring>
-#include <mutex>
-#include <string>
-#include <thread>
-#include <vector>
-
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-#ifndef NOMINMAX
-#define NOMINMAX
-#endif
-#include <windows.h>
-#undef min
-#undef max
-#endif
-
-namespace {
-
-using Libs::Graphics::ResourceMutex;
-
-void Check(bool value, const char* text) {
-	if (!value) {
-		std::fprintf(stderr, "ResourceMutexTests: failed: %s\n", text);
-		std::abort();
-	}
-}
-
-void YieldMany() {
-	for (uint32_t i = 0; i < 4096; i++) {
-		std::this_thread::yield();
-	}
-}
-
-void TestOwnership() {
-	ResourceMutex mutex;
-	Check(!mutex.IsOwnedByCurrentThread(), "fresh mutex reported an owner");
-	{
-		std::lock_guard lock(mutex);
-		Check(mutex.IsOwnedByCurrentThread(), "owner tracking was not installed");
-	}
-	Check(!mutex.IsOwnedByCurrentThread(), "owner tracking survived unlock");
-}
-
-void TestSerializesTransactions() {
-	ResourceMutex    mutex;
-	std::unique_lock owner(mutex);
-	std::atomic_bool contender_started {false};
-	std::atomic_bool contender_entered {false};
-	std::thread      contender([&] {
-		contender_started.store(true, std::memory_order_release);
-		std::lock_guard lock(mutex);
-		contender_entered.store(true, std::memory_order_release);
-	});
-	while (!contender_started.load(std::memory_order_acquire)) {
-		std::this_thread::yield();
-	}
-	YieldMany();
-	Check(!contender_entered.load(std::memory_order_acquire),
-	      "contender entered an active resource transaction");
-	owner.unlock();
-	contender.join();
-	Check(contender_entered.load(std::memory_order_acquire),
-	      "contender did not resume after resource transaction");
-}
-
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-[[noreturn]] void RunDeathCase() {
-	ResourceMutex mutex;
-	std::lock_guard first(mutex);
-	std::lock_guard second(mutex);
-	std::_Exit(0x7f);
-}
-
-void CheckDeathCase() {
-	char path[MAX_PATH] {};
-	Check(GetModuleFileNameA(nullptr, path, MAX_PATH) != 0, "GetModuleFileName failed");
-	std::string       command = std::string("\"") + path + "\" --death";
-	std::vector<char> mutable_command(command.begin(), command.end());
-	mutable_command.push_back('\0');
-	STARTUPINFOA        startup {sizeof(startup)};
-	PROCESS_INFORMATION process {};
-	Check(CreateProcessA(nullptr, mutable_command.data(), nullptr, nullptr, FALSE, CREATE_NO_WINDOW,
-	                     nullptr, nullptr, &startup, &process) != 0,
-	      "CreateProcess failed");
-	const auto wait = WaitForSingleObject(process.hProcess, 10000);
-	if (wait != WAIT_OBJECT_0) {
-		TerminateProcess(process.hProcess, 0x7e);
-	}
-	Check(wait == WAIT_OBJECT_0, "ResourceMutex death test timed out");
-	DWORD exit_code = 0;
-	Check(GetExitCodeProcess(process.hProcess, &exit_code) != 0 && exit_code == 321,
-	      "ResourceMutex death path used the wrong exit");
-	CloseHandle(process.hThread);
-	CloseHandle(process.hProcess);
-}
-#endif
-
-} // namespace
-
-int main(int argc, char** argv) {
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-	if (argc == 2 && std::strcmp(argv[1], "--death") == 0) {
-		RunDeathCase();
-	}
-#else
-	(void)argc;
-	(void)argv;
-#endif
-	TestOwnership();
-	TestSerializesTransactions();
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-	CheckDeathCase();
-#endif
-	std::puts("ResourceMutexTests: all cases passed");
-	return 0;
-}

tests/VirtualMemoryAllocationTests.cpp

@@ -8,15 +8,28 @@
 #include "kernel/memory.h"
 #include "kernel/pthread.h"
 #include "libs/errno.h"
+#include "loader/redZonePatcher.h"
 #include "loader/runtimeLinker.h"
 #include "loader/systemContent.h"
 
+#include <array>
 #include <cinttypes>
 #include <cstdint>
 #include <cstdio>
 #include <cstring>
 #include <filesystem>
 #include <string>
+#include <vector>
+
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+#ifndef NOMINMAX
+#define NOMINMAX
+#endif
+#include <windows.h>
+#ifdef DeleteFile
+#undef DeleteFile
+#endif
+#endif
 
 namespace {
 
@@ -133,6 +146,105 @@ void InitSubsystems() {
 	initialized = true;
 }
 
+#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
+void* g_red_zone_fault_page = nullptr;
+
+LONG CALLBACK RedZoneFaultHandler(EXCEPTION_POINTERS* exception) {
+	if (exception == nullptr || exception->ExceptionRecord == nullptr ||
+	    exception->ContextRecord == nullptr ||
+	    exception->ExceptionRecord->ExceptionCode != EXCEPTION_ACCESS_VIOLATION ||
+	    reinterpret_cast<void*>(exception->ExceptionRecord->ExceptionInformation[1]) !=
+	        g_red_zone_fault_page) {
+		return EXCEPTION_CONTINUE_SEARCH;
+	}
+
+	// Model the Windows exception stack footprint that prompted the static patch:
+	// data below the interrupted RSP is not part of the Windows ABI contract.
+	*reinterpret_cast<uint64_t*>(exception->ContextRecord->Rsp - 0x18) = 0;
+	DWORD old_protection = 0;
+	if (VirtualProtect(g_red_zone_fault_page, 0x1000, PAGE_READWRITE, &old_protection) == FALSE) {
+		return EXCEPTION_CONTINUE_SEARCH;
+	}
+	return EXCEPTION_CONTINUE_EXECUTION;
+}
+
+void TestWindowsGuestRedZoneStaticPatcher() {
+	const char* test = "WindowsGuestRedZoneStaticPatcher";
+	constexpr uint64_t SENTINEL = 0x1122334455667788ull;
+	constexpr uint64_t CODE_SIZE = 0x4000;
+	constexpr uint64_t TRAMPOLINE_SIZE = 0x4000;
+	const auto mapping = Libs::LibKernel::Memory::AllocateProgramMemory(
+	    0x0000000902000000ull, CODE_SIZE + TRAMPOLINE_SIZE,
+	    Common::VirtualMemory::Mode::ExecuteReadWrite, "red_zone_patcher_test");
+	Check(test, mapping != 0, "failed to allocate patch test code");
+
+	std::vector<uint8_t> code;
+	const auto emit = [&code](std::initializer_list<uint8_t> bytes) {
+		code.insert(code.end(), bytes.begin(), bytes.end());
+	};
+	const auto emit64 = [&code](uint64_t value) {
+		const auto offset = code.size();
+		code.resize(offset + sizeof(value));
+		std::memcpy(code.data() + offset, &value, sizeof(value));
+	};
+	emit({0x48, 0xb8});
+	emit64(SENTINEL);                         // movabs rax, sentinel
+	emit({0x48, 0x89, 0x44, 0x24, 0xe8});     // mov [rsp-0x18], rax
+	emit({0x48, 0x8b, 0x07});                 // mov rax, [rdi] (faultable, 3 bytes)
+	emit({0x48, 0x8b, 0x44, 0x24, 0xe8});     // mov rax, [rsp-0x18]
+	emit({0x48, 0xb9});
+	emit64(SENTINEL);                         // movabs rcx, sentinel
+	emit({0x48, 0x39, 0xc8});                 // cmp rax, rcx
+	emit({0x0f, 0x94, 0xc0});                 // sete al
+	emit({0x0f, 0xb6, 0xc0, 0xc3});           // movzx eax, al; ret
+	Check(test, code.size() < CODE_SIZE, "generated patch test code is too large");
+	std::memcpy(reinterpret_cast<void*>(mapping), code.data(), code.size());
+	Check(test, Common::VirtualMemory::FlushInstructionCache(mapping, code.size()),
+	      "failed to flush generated test code");
+
+	g_red_zone_fault_page = VirtualAlloc(nullptr, 0x1000, MEM_RESERVE | MEM_COMMIT, PAGE_NOACCESS);
+	Check(test, g_red_zone_fault_page != nullptr, "failed to allocate fault page");
+	auto* handler = AddVectoredExceptionHandler(1, RedZoneFaultHandler);
+	Check(test, handler != nullptr, "failed to install test exception handler");
+
+	using GuestFunction = uint64_t(KYTY_SYSV_ABI*)(const uint64_t*);
+	const auto function = reinterpret_cast<GuestFunction>(mapping);
+	const bool unpatched_was_corrupted = function(static_cast<const uint64_t*>(g_red_zone_fault_page)) == 0;
+
+	DWORD old_protection = 0;
+	Check(test, VirtualProtect(g_red_zone_fault_page, 0x1000, PAGE_NOACCESS, &old_protection) != FALSE,
+	      "failed to reset fault page protection");
+	Loader::RegisterRedZonePatchModule(reinterpret_cast<void*>(mapping), CODE_SIZE,
+	                                   reinterpret_cast<void*>(mapping + CODE_SIZE),
+	                                   TRAMPOLINE_SIZE);
+	const std::array<uintptr_t, 1> function_starts = {static_cast<uintptr_t>(mapping)};
+	const auto result = Loader::PatchRedZoneMemoryInstructions(
+	    mapping, code.size(), function_starts);
+	Check(test, Common::VirtualMemory::FlushInstructionCache(mapping, CODE_SIZE + TRAMPOLINE_SIZE),
+	      "failed to flush patched test code");
+	const bool patched_preserved = function(static_cast<const uint64_t*>(g_red_zone_fault_page)) == 1;
+
+	Loader::UnregisterRedZonePatchModule(reinterpret_cast<void*>(mapping));
+	RemoveVectoredExceptionHandler(handler);
+	VirtualFree(g_red_zone_fault_page, 0, MEM_RELEASE);
+	g_red_zone_fault_page = nullptr;
+	const bool freed = Libs::LibKernel::Memory::FreeGuestMemory(mapping, CODE_SIZE + TRAMPOLINE_SIZE);
+
+	Check(test, unpatched_was_corrupted, "test harness did not reproduce red-zone corruption");
+	Check(test, result.red_zone_function_count == 1 && result.memory_instruction_count >= 1 &&
+	                result.patched_memory_instruction_count >= 1 &&
+	                result.unrelocatable_memory_instruction_count == 0,
+	      "static patcher did not cover the faultable instruction");
+	Check(test, patched_preserved, "patched fault still corrupted the guest red zone");
+	Check(test, freed, "failed to free patch test code");
+	std::printf("[host]    %-48s ok\n", test);
+}
+#else
+void TestWindowsGuestRedZoneStaticPatcher() {
+	std::printf("[host]    %-48s skipped\n", "WindowsGuestRedZoneStaticPatcher");
+}
+#endif
+
 void RunTest(void (*test_func)()) {
 	if (g_failed_tests != 0) {
 		return;
@@ -495,6 +607,85 @@ void TestFlexibleMemoryReuseIsZeroFilled() {
 	std::printf("[host]    %-48s ok\n", test);
 }
 
+void TestSmallerFlexibleMapReusesReleasedHole() {
+	const char*        test       = "SmallerFlexibleMapReusesReleasedHole";
+	const auto         baseline   = AvailableFlexibleMemory(test);
+	constexpr uint64_t SmallSize  = SceKernelPageSize * 12;
+	constexpr uint64_t LargeSize  = SceKernelPageSize * 16;
+	constexpr uint64_t BlockSize  = SceKernelPageSize * 4;
+	constexpr int      RuntimeMap = 0x8000;
+
+	// Leave a 0x30000 hole before an allocated backing block. The large mapping must then be
+	// assembled from two flexible-backing extents, matching the allocation that exposed the
+	// PPSA30528 regression.
+	void* seed_hole = nullptr;
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMapNamedFlexibleMemory(
+	            &seed_hole, SmallSize, SceKernelProtCpuRw, RuntimeMap, "backing_seed_hole"),
+	        "KernelMapNamedFlexibleMemory(seed hole)");
+	void* seed_blocker = nullptr;
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMapNamedFlexibleMemory(
+	            &seed_blocker, BlockSize, SceKernelProtCpuRw, RuntimeMap, "backing_seed_blocker"),
+	        "KernelMapNamedFlexibleMemory(seed blocker)");
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMunmap(reinterpret_cast<uint64_t>(seed_hole), SmallSize),
+	        "KernelMunmap(seed hole)");
+	void* virtual_blocker = seed_hole;
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelReserveVirtualRange(&virtual_blocker, SmallSize, 0,
+	                                                           SceKernelPageSize),
+	        "KernelReserveVirtualRange(seed virtual blocker)");
+	Check(test, virtual_blocker == seed_hole, "seed virtual hole was not reserved in place");
+
+	void* large = nullptr;
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMapNamedFlexibleMemory(
+	            &large, LargeSize, SceKernelProtCpuRw, RuntimeMap, "released_large_hole"),
+	        "KernelMapNamedFlexibleMemory(large)");
+	void* blocker = nullptr;
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMapNamedFlexibleMemory(
+	            &blocker, SmallSize, SceKernelProtCpuRw, RuntimeMap, "adjacent_blocker"),
+	        "KernelMapNamedFlexibleMemory(blocker)");
+	const auto large_base   = reinterpret_cast<uint64_t>(large);
+	const auto blocker_base = reinterpret_cast<uint64_t>(blocker);
+	Check(test, blocker_base == large_base + LargeSize,
+	      "hint-less flexible maps were not adjacent");
+
+	CheckOk(test, Libs::LibKernel::Memory::KernelMunmap(large_base, LargeSize),
+	        "KernelMunmap(large)");
+
+	void* reused = nullptr;
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMapNamedFlexibleMemory(
+	            &reused, SmallSize, SceKernelProtCpuRw, RuntimeMap, "smaller_hole_reuse"),
+	        "KernelMapNamedFlexibleMemory(smaller reuse)");
+	Check(test, reinterpret_cast<uint64_t>(reused) == large_base,
+	      "smaller flexible map did not reuse the released hole");
+	*reinterpret_cast<uint64_t*>(reused) = 0x4b59545952455553ull; // "KYTYREUS"
+	Check(test, *reinterpret_cast<const uint64_t*>(reused) == 0x4b59545952455553ull,
+	      "reused flexible hole is not writable");
+
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMunmap(reinterpret_cast<uint64_t>(reused), SmallSize),
+	        "KernelMunmap(reused)");
+	CheckOk(test, Libs::LibKernel::Memory::KernelMunmap(blocker_base, SmallSize),
+	        "KernelMunmap(blocker)");
+	CheckOk(
+	    test,
+	    Libs::LibKernel::Memory::KernelMunmap(reinterpret_cast<uint64_t>(seed_blocker), BlockSize),
+	    "KernelMunmap(seed blocker)");
+	CheckOk(test,
+	        Libs::LibKernel::Memory::KernelMunmap(reinterpret_cast<uint64_t>(virtual_blocker),
+	                                              SmallSize),
+	        "KernelMunmap(seed virtual blocker)");
+	Check(test, AvailableFlexibleMemory(test) == baseline,
+	      "released-hole reuse leaked flexible memory capacity");
+
+	std::printf("[host]    %-48s ok\n", test);
+}
+
 void TestGuestStackUsesPrivateOwnerMemoryAndCache() {
 	const char* test     = "GuestStackUsesPrivateOwnerMemoryAndCache";
 	const auto  baseline = AvailableFlexibleMemory(test);
@@ -605,6 +796,13 @@ void TestRuntimeMemoryOwnerLifecycle() {
 	    test,
 	    Libs::LibKernel::Memory::TestPlaceholderRangeIsFree(adjacent_first, SceKernelPageSize * 2),
 	    "combined adjacent runtime free did not restore one owner placeholder");
+	const auto adjacent_reused = Libs::LibKernel::Memory::AllocateRuntimeMemory(
+	    adjacent_first, SceKernelPageSize, Common::VirtualMemory::Mode::ReadWrite,
+	    "runtime_adjacent_reuse", true);
+	Check(test, adjacent_reused == adjacent_first,
+	      "combined adjacent runtime placeholder could not be split for reuse");
+	Check(test, Libs::LibKernel::Memory::FreeGuestMemory(adjacent_reused, SceKernelPageSize),
+	      "adjacent runtime reuse cleanup failed");
 
 	std::printf("[host]    %-48s ok\n", test);
 }
@@ -2246,9 +2444,14 @@ void TestModuleRelocationUsesWritableHostMapping() {
 
 } // namespace
 
-int main() {
+int main(int argc, char** argv) {
 	InitSubsystems();
+	if (argc == 2 && std::strcmp(argv[1], "--red-zone-patcher-only") == 0) {
+		RunTest(TestWindowsGuestRedZoneStaticPatcher);
+		return g_failed_tests == 0 ? 0 : 1;
+	}
 
+	RunTest(TestWindowsGuestRedZoneStaticPatcher);
 	RunTest(TestProsperoArgumentAndInfoSizeContracts);
 	RunTest(TestGuestAddressSpaceOwnsReservationsBeforeBacking);
 	RunTest(TestGuestAddressSpaceHasNoFixedFallback);
@@ -2258,6 +2461,7 @@ int main() {
 	RunTest(TestFlexibleDmemCompatAndAlignmentFlags);
 	RunTest(TestFlexibleNoCoalescePreservesBoundaries);
 	RunTest(TestFlexibleMemoryReuseIsZeroFilled);
+	RunTest(TestSmallerFlexibleMapReusesReleasedHole);
 	RunTest(TestGuestStackUsesPrivateOwnerMemoryAndCache);
 	RunTest(TestMainEntryUsesGuestStackAndDisablesHostChecks);
 	RunTest(TestFragmentedBackingUnmapRollback);