common: remove unused recursive file enumeration (#681)

common: don't follow reparse points in the Windows recursive file walk

SysFileFindFiles recurses into every entry carrying
FILE_ATTRIBUTE_DIRECTORY, which on Windows includes directory symlinks
and junctions. A junction pointing at one of its own ancestors makes the
walk recurse until the stack runs out.

The Linux implementation already guards against this: it uses lstat
rather than stat specifically so a symlink is never followed into a
cycle. Skip reparse-point entries on Windows for the same reason.

Co-authored-by: nmzik <Nmzik@mail.ru>

src/common/emulatorConfig.cpp

@@ -135,6 +135,10 @@ bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }
 
+bool TessellationEnabled() {
+	return g_config->tessellation_enabled;
+}
+
 bool PlayGoHackEnabled() {
 	return g_config->playgo_hack_enabled;
 }

src/common/emulatorConfig.h

@@ -66,6 +66,7 @@ struct ConfigOptions {
 	bool                   gpu_assisted_validation_enabled = false;
 	bool                   renderdoc_enabled           = false;
 	bool                   readback_linear_images      = false;
+	bool                   tessellation_enabled        = false;
 	bool                   playgo_hack_enabled         = false;
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	bool red_zone_protection_enabled = false;
@@ -110,6 +111,7 @@ bool GpuAssistedValidationEnabled();
 
 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
+bool TessellationEnabled();
 bool PlayGoHackEnabled();
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 bool RedZoneProtectionEnabled();

src/common/file.cpp

@@ -445,45 +445,6 @@ bool File::SetLastAccessAndWriteTimeUTC(const std::filesystem::path& name, const
 	return SysFileSetLastAccessAndWriteTimeUtc(name, af, wf);
 }
 
-std::vector<File::FindInfo> File::FindFiles(const std::filesystem::path& path) {
-	std::vector<sys_file_find_t> files;
-
-	SysFileFindFiles(path, files);
-
-	const auto normalized_path = path.lexically_normal();
-	std::vector<File::FindInfo> ret;
-	ret.reserve(files.size());
-
-	for (const auto& f: files) {
-		File::FindInfo r {};
-
-		r.path_with_name     = f.path_with_name;
-		r.rel_path_with_name = f.path_with_name.lexically_normal().lexically_relative(normalized_path);
-		r.size               = f.size;
-
-		SysTimeStruct at {};
-		SysTimeStruct wt {};
-
-		SysFileToSystemTimeUtc(f.last_access_time, at);
-		SysFileToSystemTimeUtc(f.last_write_time, wt);
-
-		if (!at.is_invalid && !wt.is_invalid) {
-			r.last_access_time = DateTime(Date(at.Year, at.Month, at.Day),
-			                              Time(at.Hour, at.Minute, at.Second, at.Milliseconds));
-			r.last_write_time  = DateTime(Date(wt.Year, wt.Month, wt.Day),
-			                              Time(wt.Hour, wt.Minute, wt.Second, wt.Milliseconds));
-		}
-
-		// printf("%s, %s, %" PRIu64", %s, %s\n", r.path_with_name.c_str(),
-		// r.rel_path_with_name.c_str(), r.size, r.last_access_time.ToString().c_str(),
-		// r.last_write_time.ToString().c_str());
-
-		ret.push_back(std::move(r));
-	}
-
-	return ret;
-}
-
 std::vector<File::DirEntry> File::GetDirEntries(const std::filesystem::path& path) {
 	std::vector<sys_dir_entry_t> files;
 

src/common/file.h

@@ -30,14 +30,6 @@ class File {
 public:
 	enum class Mode { Read, Write, ReadWrite, WriteRead };
 
-	struct FindInfo {
-		std::filesystem::path path_with_name;
-		std::filesystem::path rel_path_with_name;
-		DateTime              last_access_time;
-		DateTime              last_write_time;
-		uint64_t              size;
-	};
-
 	struct DirEntry {
 		std::string name;
 		bool        is_file;
@@ -98,7 +90,6 @@ public:
 	static bool SetLastAccessAndWriteTimeUTC(const std::filesystem::path& name,
 	                                         const DateTime& access, const DateTime& write);
 
-	static std::vector<FindInfo> FindFiles(const std::filesystem::path& path);
 	static std::vector<DirEntry> GetDirEntries(const std::filesystem::path& path);
 	static bool CopyFile(const std::filesystem::path& src, const std::filesystem::path& dst);
 	static bool RenameFile(const std::filesystem::path& src, const std::filesystem::path& dst);

src/common/platform/sysFileIO.h

@@ -18,14 +18,6 @@ enum sys_file_cache_type_t {
 	SYS_FILE_CACHE_SEQUENTIAL_SCAN = 2  // NOLINT(readability-identifier-naming)
 };
 
-// NOLINTNEXTLINE(readability-identifier-naming)
-struct sys_file_find_t {
-	std::filesystem::path path_with_name;
-	SysFileTimeStruct     last_access_time;
-	SysFileTimeStruct     last_write_time;
-	uint64_t              size;
-};
-
 // NOLINTNEXTLINE(readability-identifier-naming)
 struct sys_dir_entry_t {
 	std::string name;
@@ -75,8 +67,6 @@ bool SysFileSetLastWriteTimeUtc(const std::filesystem::path& name, SysFileTimeSt
 bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
                                          SysFileTimeStruct& access, SysFileTimeStruct& write);
 // NOLINTNEXTLINE(google-runtime-references)
-void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out);
-// NOLINTNEXTLINE(google-runtime-references)
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out);
 bool SysFileCopyFile(const std::filesystem::path& src, const std::filesystem::path& dst);
 bool SysFileRenameFile(const std::filesystem::path& src, const std::filesystem::path& dst);

src/common/platform/sysLinuxFileIO.cpp

@@ -610,51 +610,6 @@ bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
 	//	}
 }
 
-// Recursively collect regular files.
-void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out) {
-	auto real_path = get_internal_name(path);
-
-	DIR* dir = opendir(real_path.string().c_str());
-	if (dir == nullptr) {
-		return;
-	}
-
-	for (const dirent* entry = readdir(dir); entry != nullptr; entry = readdir(dir)) {
-		const std::string file_name(entry->d_name);
-
-		if (file_name == "." || file_name == "..") {
-			continue;
-		}
-
-		auto        child = real_path / file_name;
-		struct stat s {};
-
-		// lstat, so a symlink is never followed into a cycle during the recursive walk.
-		if (0 != lstat(child.string().c_str(), &s)) {
-			continue;
-		}
-
-		if (S_ISDIR(s.st_mode)) {
-			SysFileFindFiles(child, out);
-		} else if (S_ISREG(s.st_mode)) {
-			sys_file_find_t r {};
-
-			r.path_with_name              = child;
-			r.size                        = static_cast<uint64_t>(s.st_size);
-			r.last_access_time.is_invalid = false;
-			r.last_access_time.time       = s.st_atime;
-			r.last_access_time.nanos      = KYTY_STAT_ATIME_NS(s);
-			r.last_write_time.is_invalid  = false;
-			r.last_write_time.time        = s.st_mtime;
-			r.last_write_time.nanos       = KYTY_STAT_MTIME_NS(s);
-
-			out.push_back(r);
-		}
-	}
-
-	closedir(dir);
-}
-
 // Keep "." and ".." to match FindFirstFileW.
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out) {
 	auto real_path = get_internal_name(path);

src/common/platform/sysWindowsFileIO.cpp

@@ -57,10 +57,10 @@ static DWORD GetCacheAccessType(sys_file_cache_type_t t) {
 	}
 
 	if (t == SYS_FILE_CACHE_SEQUENTIAL_SCAN) {
-		return SYS_FILE_CACHE_SEQUENTIAL_SCAN;
+		return FILE_FLAG_SEQUENTIAL_SCAN;
 	}
 
-	return SYS_FILE_CACHE_AUTO;
+	return FILE_ATTRIBUTE_NORMAL;
 }
 
 void SysFileRead(void* data, uint32_t size, sys_file_t& f, uint32_t* bytes_read) {
@@ -476,46 +476,6 @@ bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
 	return ok;
 }
 
-void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out) {
-	const auto pattern = path / L"*";
-
-	WIN32_FIND_DATAW data {};
-	HANDLE h = FindFirstFileW(pattern.c_str(), &data);
-
-	if (h == INVALID_HANDLE_VALUE) {
-		return;
-	}
-
-	do {
-		std::filesystem::path file_name(data.cFileName);
-
-		if (file_name == L"." || file_name == L"..") {
-			continue;
-		}
-
-		const auto full_path = path / file_name;
-
-		if ((data.dwFileAttributes & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) != 0u) {
-			SysFileFindFiles(full_path, out);
-		} else {
-			sys_file_find_t r {};
-
-			r.path_with_name              = full_path;
-			r.size                        = (static_cast<uint64_t>(data.nFileSizeHigh) << 32u) +
-			                                static_cast<uint64_t>(data.nFileSizeLow);
-			r.last_access_time.is_invalid = false;
-			r.last_access_time.time       = data.ftLastAccessTime;
-			r.last_write_time.is_invalid  = false;
-			r.last_write_time.time        = data.ftLastWriteTime;
-
-			out.push_back(std::move(r));
-		}
-
-	} while (FindNextFileW(h, &data) != 0);
-
-	FindClose(h);
-}
-
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out) {
 	const auto pattern = path / L"*";
 

src/graphics/guest_gpu/gpu_defs.h

@@ -26,6 +26,10 @@ enum class PrimitiveType : uint32_t {
 	kPolygon            = 21,
 };
 
+constexpr bool IsRectList(PrimitiveType type) {
+	return type == PrimitiveType::kRectList || type == PrimitiveType::kRectListLegacy;
+}
+
 enum class IndexType : uint32_t {
 	kIndex16 = 0,
 	kIndex32 = 1,

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

@@ -755,8 +755,7 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
 	static_params.depth_bounds_test_enable = depth.depth_bounds_test_enable;
 	static_params.depth_min_bounds         = depth.depth_min_bounds;
 	static_params.depth_max_bounds         = depth.depth_max_bounds;
-	const bool rect_list =
-	    command.GetUserConfig().GetPrimType() == Prospero::PrimitiveType::kRectList;
+	const bool rect_list = Prospero::IsRectList(command.GetUserConfig().GetPrimType());
 	static_params.cull_back  = !rect_list && mc.cull_back;
 	static_params.cull_front = !rect_list && mc.cull_front;
 	static_params.face       = mc.face;

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -2,6 +2,7 @@
 
 #include "common/assert.h"
 #include "common/common.h"
+#include "common/emulatorConfig.h"
 #include "common/file.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
@@ -789,8 +790,7 @@ static void SetDrawDebugPhase(CommandBuffer& buffer, uint64_t submit_id, const D
 	                    draw.instance_count, draw.first_instance);
 }
 
-static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
-                            vk::PrimitiveTopology& topology) {
+static bool GetDrawTopology(const HW::UserConfig& ucfg, vk::PrimitiveTopology& topology) {
 
 	topology = vk::PrimitiveTopology::ePointList;
 
@@ -813,14 +813,11 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 			topology = vk::PrimitiveTopology::eTriangleStrip;
 			break;
 		case Prospero::PrimitiveType::kPatch:
+			if (!Config::TessellationEnabled()) return false;
+			[[fallthrough]];
 		case Prospero::PrimitiveType::kRectList:
-			topology = vk::PrimitiveTopology::ePatchList;
-			break;
 		case Prospero::PrimitiveType::kRectListLegacy:
-			if (!auto_draw) {
-				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
-			}
-			topology = vk::PrimitiveTopology::eTriangleStrip;
+			topology = vk::PrimitiveTopology::ePatchList;
 			break;
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			topology = vk::PrimitiveTopology::eTriangleFan;
@@ -999,8 +996,7 @@ static void LogDrawStateIfNeeded(const CommandBuffer& buffer, const DrawCallInfo
 		return;
 	}
 
-	if (!draw.IsIndexed() &&
-	    buffer.GetUserConfig().GetPrimType() != Prospero::PrimitiveType::kRectListLegacy) {
+	if (!draw.IsIndexed() && !Prospero::IsRectList(buffer.GetUserConfig().GetPrimType())) {
 		return;
 	}
 
@@ -1013,8 +1009,7 @@ static void LogDrawStateIfNeeded(const CommandBuffer& buffer, const DrawCallInfo
 }
 
 static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_buffer,
-                               const ShaderVertexInputInfo& vs_input_info, const DrawCallInfo& draw,
-                               const DrawEmitInfo& emit) {
+                               const DrawCallInfo& draw, const DrawEmitInfo& emit) {
 	switch (ucfg.GetPrimType()) {
 		case Prospero::PrimitiveType::kPointList:
 		case Prospero::PrimitiveType::kLineList:
@@ -1023,6 +1018,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 		case Prospero::PrimitiveType::kTriFan:
 		case Prospero::PrimitiveType::kTriStrip:
 		case Prospero::PrimitiveType::kRectList:
+		case Prospero::PrimitiveType::kRectListLegacy:
 		case Prospero::PrimitiveType::kPatch:
 			if (draw.IsIndexed()) {
 				vk_buffer.drawIndexed(draw.index_count, draw.instance_count, 0, emit.vertex_offset,
@@ -1032,14 +1028,6 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 				               emit.first_instance);
 			}
 			break;
-		case Prospero::PrimitiveType::kRectListLegacy:
-			if (draw.IsIndexed()) {
-				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
-			}
-			// Sarah
-			EXIT_NOT_IMPLEMENTED(!(draw.index_count == 3 && vs_input_info.buffers_num == 0));
-			vk_buffer.draw(4, draw.instance_count, emit.first_vertex, emit.first_instance);
-			break;
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			EXIT_NOT_IMPLEMENTED((draw.index_count & 0x3u) != 0);
 			for (uint32_t i = 0; i < draw.index_count; i += 4) {
@@ -1182,7 +1170,7 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, CommandBuffer& buff
 	if (mesh_active) {
 		vk_buffer.drawMeshTasksEXT(mesh_groups, draw.instance_count, 1);
 	} else {
-		EmitDrawPrimitives(ucfg, vk_buffer, state.vertex_info[0], draw, emit);
+		EmitDrawPrimitives(ucfg, vk_buffer, draw, emit);
 	}
 
 	if (!draw.IsIndexed()) {
@@ -1258,7 +1246,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, CommandBuffer& buffer,
 	hw_check(buffer);
 
 	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
-	if (!GetDrawTopology(ucfg, false, topology)) {
+	if (!GetDrawTopology(ucfg, topology)) {
 		return;
 	}
 
@@ -1368,7 +1356,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 	                         args.vertex_count, args.instance_count, args.first_instance};
 
 	vk::PrimitiveTopology topology = vk::PrimitiveTopology::ePointList;
-	if (!GetDrawTopology(ucfg, true, topology)) {
+	if (!GetDrawTopology(ucfg, topology)) {
 		ResetBindings();
 		return;
 	}
@@ -1378,7 +1366,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, CommandBuffer& buffer, const D
 		return;
 	}
 
-	const bool rect_list = ucfg.GetPrimType() == Prospero::PrimitiveType::kRectList;
+	const bool rect_list = Prospero::IsRectList(ucfg.GetPrimType());
 	if (rect_list && state.vertex_info[0].buffers_num == 0 &&
 	    state.vertex_info[0].stage.program->param_export_mask == 0 &&
 	    state.ps_input_info.input_num != 0) {

src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp

@@ -521,7 +521,8 @@ constexpr Vop1SdwaRule VOP1_SDWA_RULES[] = {
     {Opcode::V_CVT_F32_U32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
     {Opcode::V_CVT_F32_I32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
     {Opcode::V_CVT_F32_UBYTE0, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
-    {Opcode::V_NOT_B32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
+    {Opcode::V_NOT_B32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelBytes() | SdwaSelWords(),
+     SdwaSelFull(), false},
     {Opcode::V_FFBL_B32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), 0, 0, false},
     {Opcode::V_CVT_F32_F16, SdwaSelWords() | SdwaSelFull(), 0, 0, true},
     {Opcode::V_CVT_F16_F32, SdwaSelBytes() | SdwaSelWords() | SdwaSelFull(), SdwaSelWords(),

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

@@ -125,6 +125,7 @@ bool IsRuntimeUniformOp(ValueOpcode op) {
 		case ValueOpcode::ULessThan32:
 		case ValueOpcode::IEqual32:
 		case ValueOpcode::UGreaterThan32:
+		case ValueOpcode::SGreaterThanEqual32:
 		case ValueOpcode::INotEqual32:
 		case ValueOpcode::LogicalOr:
 		case ValueOpcode::LogicalAnd:
@@ -930,6 +931,13 @@ private:
 					return true;
 				}
 				return false;
+			case ValueOpcode::SGreaterThanEqual32:
+				if (binary()) {
+					result = std::bit_cast<int32_t>(static_cast<uint32_t>(a)) >=
+					         std::bit_cast<int32_t>(static_cast<uint32_t>(b));
+					return true;
+				}
+				return false;
 			case ValueOpcode::LogicalAnd:
 				if (binary()) {
 					result = (a != 0u) && (b != 0u);

src/launcher/forms/configuration_edit_dialog.ui

@@ -167,6 +167,12 @@
           <property name="text"><string>Enable readback</string></property>
          </widget>
         </item>
+        <item row="7" column="1">
+         <widget class="QCheckBox" name="checkBox_tessellation">
+          <property name="toolTip"><string>Draw tessellated geometry. Disabled by default; unsupported tessellation may cause crashes.</string></property>
+          <property name="text"><string>Enable tessellation support (experimental)</string></property>
+         </widget>
+        </item>
        </layout>
       </widget>
      </item>
@@ -275,13 +281,13 @@
         <item>
          <widget class="QCheckBox" name="checkBox_amd_cpu">
           <property name="toolTip"><string>Patch CPU instructions that differ between AMD and the host CPU.</string></property>
-          <property name="text"><string>AMD CPU patch (EXPERIMENTAL)</string></property>
+          <property name="text"><string>AMD CPU patch (experimental)</string></property>
          </widget>
         </item>
         <item>
          <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>
+          <property name="text"><string>Windows SysV red zone crash protection (experimental)</string></property>
          </widget>
         </item>
        </layout>

src/launcher/include/configuration.h

@@ -92,6 +92,7 @@ public:
 	int                    gpu_index                   = -1;
 	bool                   fullscreen_enabled          = false;
 	bool                   readback_linear_images      = false;
+	bool                   tessellation_enabled        = false;
 	int                    vblank_frequency            = 60;
 	int                    console_language            = DEFAULT_CONSOLE_LANGUAGE;
 	bool                   vulkan_validation_enabled   = false;
@@ -122,6 +123,7 @@ public:
 		gpu_index                   = other.gpu_index;
 		fullscreen_enabled          = other.fullscreen_enabled;
 		readback_linear_images      = other.readback_linear_images;
+		tessellation_enabled        = other.tessellation_enabled;
 		vblank_frequency            = other.vblank_frequency;
 		console_language            = other.console_language;
 		vulkan_validation_enabled   = other.vulkan_validation_enabled;
@@ -169,6 +171,7 @@ public:
 		KYTY_CFG_SET(gpu_index);
 		KYTY_CFG_SET(fullscreen_enabled);
 		KYTY_CFG_SET(readback_linear_images);
+		KYTY_CFG_SET(tessellation_enabled);
 		KYTY_CFG_SET(vblank_frequency);
 		KYTY_CFG_SET(console_language);
 		KYTY_CFG_SET(vulkan_validation_enabled);
@@ -210,6 +213,7 @@ public:
 		}
 		KYTY_CFG_GET(fullscreen_enabled);
 		KYTY_CFG_GET(readback_linear_images);
+		KYTY_CFG_GET(tessellation_enabled);
 		vblank_frequency = s->value("vblank_frequency", vblank_frequency).toInt();
 		console_language = s->value("console_language", console_language).toInt();
 		if (console_language < 0 || console_language > MAX_CONSOLE_LANGUAGE) {

src/launcher/src/configurationEditDialog.cpp

@@ -230,6 +230,7 @@ void ConfigurationEditDialog::Init(const Configuration& info) {
 	                                                                            : 0);
 	m_ui->checkBox_fullscreen->setChecked(info.fullscreen_enabled);
 	m_ui->checkBox_readback->setChecked(info.readback_linear_images);
+	m_ui->checkBox_tessellation->setChecked(info.tessellation_enabled);
 	m_ui->spinBox_vblank_frequency->setValue(info.vblank_frequency);
 	m_ui->comboBox_console_language->clear();
 	m_ui->comboBox_console_language->addItems(CONSOLE_LANGUAGE_NAMES);
@@ -375,6 +376,7 @@ static void UpdateInfo(Configuration& info, Ui::ConfigurationEditDialog& ui) {
 	info.gpu_index                 = ui.comboBox_gpu->currentIndex() - 1;
 	info.fullscreen_enabled        = ui.checkBox_fullscreen->isChecked();
 	info.readback_linear_images    = ui.checkBox_readback->isChecked();
+	info.tessellation_enabled      = ui.checkBox_tessellation->isChecked();
 	info.vblank_frequency          = ui.spinBox_vblank_frequency->value();
 	info.console_language          = ui.comboBox_console_language->currentIndex();
 	info.vulkan_validation_enabled = ui.checkBox_vulkan_validation->isChecked();

src/launcher/src/mainDialog.cpp

@@ -226,6 +226,9 @@ static QStringList CreateEmulatorArgs(const Configuration& info) {
 		args << "--fullscreen";
 	}
 	args << "--readback-linear-images" << BoolArg(info.readback_linear_images);
+	if (info.tessellation_enabled) {
+		args << "--tessellation";
+	}
 	args << "--vblank-frequency" << QString::number(info.vblank_frequency);
 	args << "--console-language" << QString::number(info.console_language);
 	args << "--vulkan-validation" << BoolArg(info.vulkan_validation_enabled);

src/launcher/src/patchesDialog.cpp

@@ -19,7 +19,7 @@
 PatchesDialog::PatchesDialog(const Configuration& game, QWidget* parent)
     : QDialog(parent), m_title_id(game.title_id.trimmed().toUpper()) {
 	setAttribute(Qt::WA_DeleteOnClose);
-	setWindowTitle(tr("Cheats (Experimental) - %1").arg(game.name));
+	setWindowTitle(tr("Cheats (experimental) - %1").arg(game.name));
 	resize(640, 480);
 
 	auto* layout = new QVBoxLayout(this);

src/libs/avPlayer.cpp

@@ -706,7 +706,7 @@ public:
 	int Enable(uint32_t id) {
 		std::scoped_lock lifecycle_lock(lifecycle_mutex);
 		std::lock_guard  state_lock(mutex);
-		if (!stopped) {
+		if (state == State::Playing) {
 			return AVPLAYER_ERROR_OPERATION_FAILED;
 		}
 		if (fmt == nullptr || id >= fmt->nb_streams) {
@@ -724,7 +724,7 @@ public:
 	int Disable(uint32_t id) {
 		std::scoped_lock lifecycle_lock(lifecycle_mutex);
 		std::lock_guard  state_lock(mutex);
-		if (!stopped) {
+		if (state == State::Playing) {
 			return AVPLAYER_ERROR_OPERATION_FAILED;
 		}
 		if (video_id == static_cast<int>(id)) {
@@ -755,9 +755,11 @@ public:
 				return AVPLAYER_ERROR_OPERATION_FAILED;
 			}
 			time       = CurrentTimeNoLock();
-			restart    = !stopped;
+			restart    = state == State::Playing;
 			was_paused = paused;
-			stopped    = true;
+			if (restart) {
+				state = State::Ready;
+			}
 		}
 		if (restart) {
 			StopWorkers();
@@ -778,7 +780,7 @@ public:
 	int StartImpl(uint64_t ms, bool paused_after_start = false, bool show_seek_frame = false) {
 		{
 			std::lock_guard lock(mutex);
-			stopped = true;
+			state = State::Ready;
 		}
 		StopWorkers();
 		interrupt_io = false;
@@ -791,14 +793,17 @@ public:
 				AutoEnable();
 			}
 			if (!video_id && !audio_id) {
+				state = State::Stopped;
 				return AVPLAYER_ERROR_OPERATION_FAILED;
 			}
 			if (!OpenCodecs()) {
 				ResetNoLock(true);
+				state = State::Stopped;
 				return AVPLAYER_ERROR_OPERATION_FAILED;
 			}
 			if (!AllocateBuffers()) {
 				ResetNoLock(true);
+				state = State::Stopped;
 				return AVPLAYER_ERROR_NO_MEMORY;
 			}
 			SeekNoLock(ms);
@@ -809,12 +814,11 @@ public:
 			pause_time    = clock_start;
 			seek_video_frame_pending =
 			    paused_after_start && show_seek_frame && video_id.has_value();
-			stopped = false;
 		}
 
 		if (!StartWorkers()) {
 			std::lock_guard lock(mutex);
-			stopped = true;
+			state = State::Stopped;
 			ResetNoLock(true);
 			result = AVPLAYER_ERROR_OPERATION_FAILED;
 		}
@@ -825,20 +829,22 @@ public:
 	}
 	int Stop() {
 		bool                     was_playing_video = false;
-		AvPlayerEventReplacement stop_event;
+		AvPlayerEventReplacement stop_event {};
 		{
 			std::scoped_lock lifecycle_lock(lifecycle_mutex);
 			{
 				std::lock_guard lock(mutex);
-				was_playing_video = !stopped && video_id.has_value();
-				stopped           = true;
+				if (state == State::Playing) {
+					was_playing_video = video_id.has_value();
+					stop_event        = event;
+				}
+				state = State::Stopped;
 			}
 			StopWorkers();
 			{
 				std::lock_guard lock(mutex);
 				ResetNoLock();
 			}
-			stop_event = TakeStopEvent();
 		}
 		if (was_playing_video) {
 			::printf("AvPlayer video stopped\n");
@@ -872,7 +878,8 @@ public:
 	}
 	bool Active() const {
 		std::lock_guard lock(mutex);
-		return !stopped && !pipeline_failed && !DrainedNoLock();
+		return state == State::Ready ||
+		       (state == State::Playing && !pipeline_failed && !DrainedNoLock());
 	}
 	uint64_t CurrentTime() const {
 		std::lock_guard lock(mutex);
@@ -888,7 +895,7 @@ public:
 		return StartImpl(ms, was_paused, true);
 	}
 	uint64_t CurrentTimeNoLock() const {
-		if (stopped) {
+		if (state != State::Playing) {
 			return 0;
 		}
 		using namespace std::chrono;
@@ -925,7 +932,7 @@ public:
 			return false;
 		}
 		std::lock_guard lock(mutex);
-		if (stopped || !video_id) {
+		if (state != State::Playing || !video_id) {
 			return false;
 		}
 		const bool deliver_seek_frame = paused && seek_video_frame_pending;
@@ -955,7 +962,6 @@ public:
 		current_video = std::move(*frame);
 		*out          = current_video->info;
 		RecordLoopBoundary(*current_video);
-		completion_changed.notify_all();
 		if (deliver_seek_frame) {
 			seek_video_frame_pending = false;
 		}
@@ -966,7 +972,8 @@ public:
 			return false;
 		}
 		std::lock_guard lock(mutex);
-		if (paused || stopped || !audio_id || trick_speed != AVPLAYER_TRICK_SPEED_NORMAL) {
+		if (paused || state != State::Playing || !audio_id ||
+		    trick_speed != AVPLAYER_TRICK_SPEED_NORMAL) {
 			return false;
 		}
 		auto frame = audio_frames.TryPop();
@@ -987,7 +994,6 @@ public:
 		            current_audio->info.details.audio.language_code, 4);
 		last_audio_ts = out->time_stamp;
 		RecordLoopBoundary(*current_audio);
-		completion_changed.notify_all();
 		return true;
 	}
 	std::optional<int32_t> TakeWarning() {
@@ -1000,13 +1006,12 @@ public:
 	}
 
 private:
+	enum class State { Ready, Playing, Stopped };
+
 	bool DrainedNoLock() const {
 		return demux_eof && video_done && audio_done && video_frames.Empty() &&
 		       audio_frames.Empty();
 	}
-	AvPlayerEventReplacement TakeStopEvent() {
-		return stop_event_sent.exchange(true) ? AvPlayerEventReplacement {} : event;
-	}
 	void RecordLoopBoundary(const ReadyFrame& frame) {
 		if (frame.timestamp_offset > last_output_loop_offset) {
 			last_output_loop_offset = frame.timestamp_offset;
@@ -1198,11 +1203,10 @@ private:
 			StopWorkers();
 			return false;
 		}
-		stop_event_sent = false;
+		state = State::Playing;
 		return true;
 	}
 	void NotifyWorkers() {
-		completion_changed.notify_all();
 		video_packets.Notify();
 		audio_packets.Notify();
 		video_buffers.Notify();
@@ -1289,13 +1293,6 @@ private:
 		demux_eof = true;
 		video_packets.Notify();
 		audio_packets.Notify();
-		std::unique_lock lock(mutex);
-		completion_changed.wait(lock, [&] { return worker_stop || DrainedNoLock(); });
-		if (!worker_stop) {
-			const auto stop_event = TakeStopEvent();
-			lock.unlock();
-			emit_event(stop_event, AVPLAYER_EVENT_STATE_STOP);
-		}
 	}
 	void VideoDecoder() {
 		Decoder(video_packets, video_ctx, video_buffers, video_frames, video_done,
@@ -1343,11 +1340,7 @@ private:
 				break;
 			}
 		}
-		{
-			std::lock_guard lock(mutex);
-			done = true;
-		}
-		completion_changed.notify_all();
+		done = true;
 	}
 	bool DecodePacket(AVCodecContext* codec, AVPacket* packet, uint64_t timestamp_offset,
 	                  WorkQueue<std::unique_ptr<GuestBuffer>>& buffers,
@@ -1412,15 +1405,8 @@ private:
 		return use_vdec2 ? static_cast<uint32_t>(s->codecpar->height)
 		                 : align_up(static_cast<uint32_t>(s->codecpar->height), 16);
 	}
-	float Aspect(AVStream* s) const {
-		if (s->codecpar->height == 0) {
-			return 0.0f;
-		}
-		double r = static_cast<double>(s->codecpar->width) / s->codecpar->height;
-		if (s->sample_aspect_ratio.num > 0 && s->sample_aspect_ratio.den > 0) {
-			r *= av_q2d(s->sample_aspect_ratio);
-		}
-		return static_cast<float>(r);
+	float Aspect(AVRational ratio) const {
+		return ratio.num > 0 && ratio.den > 0 ? static_cast<float>(av_q2d(ratio)) : 0.0f;
 	}
 	uint64_t Duration(AVStream* s) const {
 		return s->duration != AV_NOPTS_VALUE
@@ -1433,14 +1419,14 @@ private:
 		std::memset(v, 0, sizeof(*v));
 		v->width        = Width(s);
 		v->height       = Height(s);
-		v->aspect_ratio = Aspect(s);
+		v->aspect_ratio = Aspect(s->sample_aspect_ratio);
 		lang(v->language_code, s->metadata);
 	}
-	void FillVideoEx(AVStream* s, AvPlayerVideoEx* v) const {
+	void FillVideoEx(AVStream* s, AvPlayerVideoEx* v, AVRational aspect) const {
 		std::memset(v, 0, sizeof(*v));
 		v->width                 = Width(s);
 		v->height                = Height(s);
-		v->aspect_ratio          = Aspect(s);
+		v->aspect_ratio          = Aspect(aspect);
 		v->pitch                 = VideoPitch(s);
 		v->luma_bit_depth        = 8;
 		v->chroma_bit_depth      = 8;
@@ -1476,7 +1462,7 @@ private:
 					FillVideo(s, &d->video);
 				}
 				if (ex) {
-					FillVideoEx(s, &ex->video);
+					FillVideoEx(s, &ex->video, s->sample_aspect_ratio);
 				}
 				break;
 			case AVMEDIA_TYPE_AUDIO:
@@ -1561,7 +1547,7 @@ private:
 		info->time_stamp = to_ms(
 		    src->best_effort_timestamp != AV_NOPTS_VALUE ? src->best_effort_timestamp : src->pts,
 		    s->time_base);
-		FillVideoEx(s, &info->details.video);
+		FillVideoEx(s, &info->details.video, src->sample_aspect_ratio);
 		info->details.video.crop_left_offset = static_cast<uint32_t>(src->crop_left);
 		info->details.video.crop_right_offset =
 		    static_cast<uint32_t>(src->crop_right + (pitch - src->width));
@@ -1672,11 +1658,9 @@ private:
 	std::atomic_bool                         demux_eof {true};
 	std::atomic_bool                         video_done {true};
 	std::atomic_bool                         audio_done {true};
-	std::atomic_bool                         stop_event_sent {true};
-	std::condition_variable                  completion_changed;
 	std::mutex                               lifecycle_mutex;
 	mutable std::mutex                       mutex;
-	bool                                     stopped                  = true;
+	State                                    state                    = State::Ready;
 	bool                                     paused                   = false;
 	bool                                     seek_video_frame_pending = false;
 	std::atomic_bool                         loop {false};
@@ -2029,7 +2013,7 @@ Bool KYTY_SYSV_ABI AvPlayerGetAudioData(AvPlayerInternal* h, AvPlayerFrameInfo*
 }
 Bool KYTY_SYSV_ABI AvPlayerIsActive(AvPlayerInternal* h) {
 	PRINT_NAME();
-	return h != nullptr && h->source != nullptr && h->source->Active() ? 1 : 0;
+	return h != nullptr && (h->source == nullptr || h->source->Active());
 }
 uint64_t KYTY_SYSV_ABI AvPlayerCurrentTime(AvPlayerInternal* h) {
 	PRINT_NAME();

src/libs/libUserService.cpp

@@ -92,7 +92,9 @@ static KYTY_SYSV_ABI int UserServiceGetLoginUserIdList(UserServiceLoginUserIdLis
 }
 
 static KYTY_SYSV_ABI int UserServiceGetUserName(int user_id, char* name, size_t size) {
-	EXIT_NOT_IMPLEMENTED(user_id != Config::GetUserId());
+	if (user_id != Config::GetUserId()) {
+		return USER_SERVICE_ERROR_NOT_LOGGED_IN;
+	}
 	EXIT_NOT_IMPLEMENTED(size < 5);
 
 	const auto& user_name = Config::GetUserName();

src/main.cpp

@@ -64,6 +64,7 @@ static void PrintUsage() {
 	::printf("  --gpu-assisted-validation <t|f>      Bounds-check shader accesses on the GPU.\n"
 	         "                                       Implies --vulkan-validation; very slow.\n");
 	::printf("  --shader-validation <true|false>     Enable shader validation.\n");
+	::printf("  --tessellation                      Draw tessellation patches; skipped by default.\n");
 	::printf("  --shader-optimization-type <value>   None, Size, or Performance.\n");
 	::printf("  --shader-log-direction <value>       Silent, Console, or File.\n");
 	::printf("  --shader-log-folder <path>           Shader log output folder.\n");
@@ -180,6 +181,11 @@ static bool ParseArgs(int argc, char* argv[], RunOptions& options, bool& show_he
 			continue;
 		}
 
+		if (arg == "--tessellation") {
+			options.config.tessellation_enabled = true;
+			continue;
+		}
+
 		if (arg == "--profile") {
 			options.config.profiler_enabled = true;
 			continue;

tests/KernelFileSystemTests.cpp

@@ -189,20 +189,6 @@ void CheckUnicodePaths(const std::filesystem::path &root) {
   Check(fixture.Create(native_file), "create Unicode filename");
   fixture.Close();
 
-  for (const auto &directory : {unicode_root, std::filesystem::relative(unicode_root)}) {
-    // A Windows temporary directory may be on a different drive from the working directory.
-    if (directory.empty()) {
-      continue;
-    }
-    const auto files = Common::File::FindFiles(directory);
-    const auto relative_file =
-        std::filesystem::path(u8"nested/asset-\u00e9-\u65e5\u672c\u8a9e.bin");
-    Check(std::any_of(files.begin(), files.end(), [&](const auto &file) {
-            return file.path_with_name.lexically_normal() ==
-                       (directory / relative_file).lexically_normal() &&
-                   file.rel_path_with_name == relative_file;
-          }), "recursive enumeration preserves native and relative Unicode paths");
-  }
   const auto entries = Common::File::GetDirEntries(nested_root);
   Check(std::any_of(entries.begin(), entries.end(), [&](const auto &entry) {
           return entry.is_file && entry.name == GuestFilename;

tests/ResourceTrackingTests.cpp

@@ -1246,7 +1246,7 @@ ResourcePlan ConditionalSamplerPlan(bool diamond, bool reverse, bool reverse_phi
     flag = fixture.Emit(ValueOpcode::LaneId);
   }
   fixture.program.block_info[0].condition =
-      fixture.Emit(ValueOpcode::IEqual32, {flag, Value(0u)});
+      fixture.Emit(ValueOpcode::SGreaterThanEqual32, {flag, Value(0u)});
   if (writable) {
     MemoryInfo memory;
     memory.kind = ResourceKind::Buffer;
@@ -1324,9 +1324,10 @@ void TestConditionalSamplerPhi() {
                                  .userdata = &memory,
                                  .read_specialization_memory =
                                      ReadLinearTestMemory};
-        for (uint32_t flag = 0; flag < 2; ++flag) {
-          memory.words[(0x1000 + 196) / 4] = flag;
-          const uint32_t first = (flag != 0u) != reverse ? 480 / 4 : 448 / 4;
+        // Sampler selection compares a signed value against zero.
+        for (const auto flag : {-1, 0, 1, INT32_MIN, INT32_MAX}) {
+          memory.words[(0x1000 + 196) / 4] = std::bit_cast<uint32_t>(flag);
+          const uint32_t first = (flag < 0) != reverse ? 480 / 4 : 448 / 4;
           memory.fail_address = 0x1000 + (first == 448 / 4 ? 480u : 448u);
           DescriptorValue selected;
           Check(EvaluateDescriptorSource(plan, source, runtime, selected),

tests/ShaderRecompilerComputeTests.cpp

@@ -12831,6 +12831,7 @@ public:
     const uint32_t extent = depth_feedback ? 8 : 32;
     constexpr uintptr_t depth_address = 0x0000000204400000ull;
     constexpr uint64_t allocation_size = 0x40000;
+    constexpr uint64_t rect_address = depth_address + 0x8000;
     EnsureRuntimeContext();
     RenderContext context(m_runtime_context);
     auto &scheduler = context.GetCommandScheduler();
@@ -12865,6 +12866,15 @@ public:
         static_cast<float *>(mapped)[y * 64 + x] = initial_depth(x, y);
       }
     }
+    if (!depth_feedback && !packed_vertex_color) {
+      // Rectangle input contains three position/UV records.
+      constexpr std::array<float, 24> rect_vertices{
+          -0.75f, -0.75f, 0, 0, 0, 0, 0, 0,
+           0.75f, -0.75f, 0, 0, 1, 0, 0, 0,
+          -0.75f,  0.75f, 0, 0, 0, 1, 0, 0};
+      std::memcpy(reinterpret_cast<void *>(rect_address), rect_vertices.data(),
+                  sizeof(rect_vertices));
+    }
     resources.MapMemory(depth_address, allocation_size);
     if (depth_feedback) {
       depth.desc.type = BindingType::DepthTarget;
@@ -13521,6 +13531,82 @@ public:
         Require(name, "sparse MRT3 output", sparse_pixels[component] == expected,
                 "MRT3 was compacted to a different slot or clipped by unused MRT0");
       }
+
+      // A fourth VS invocation reads beyond the descriptor instead of reconstructing the corner.
+      const ShaderBufferResource rect_buffer{{
+          static_cast<u32>(rect_address),
+          static_cast<u32>(rect_address >> 32u) | (32u << 16u), 3,
+          DstSel(4, 5, 6, 7) |
+              (static_cast<u32>(Prospero::BufferFormat::k32Float) << 12u)}};
+      static const auto rect_vertex_code = [] {
+        std::vector<u32> code{
+            EncodeMubuf0(0x0d, 0, true, false), EncodeMubuf1(0, 2, 5),
+            EncodeMubuf0(0x0d, 16, true, false), EncodeMubuf1(2, 2, 5)};
+        AppendVMovU32(&code, 4, 0);
+        AppendVMovLiteral(&code, 6, 0x3f800000u);
+        code.insert(code.end(), {EncodeExp0(0x0c, 0xf), EncodeExp1(0, 1, 4, 6),
+                                 EncodeExp0(0x20, 0xf), EncodeExp1(2, 3, 4, 6)});
+        AppendEnd(&code);
+        return code;
+      }();
+      static const auto rect_pixel_code = [] {
+        std::vector<u32> code;
+        for (u32 component = 0; component < 2; component++) {
+          code.push_back(EncodeVintrp(0, 20 + component, 0, component, 0));
+          code.push_back(EncodeVintrp(1, 20 + component, 0, component, 1));
+        }
+        AppendVMovU32(&code, 22, 0);
+        AppendVMovLiteral(&code, 23, 0x3f800000u);
+        code.insert(code.end(), {EncodeExp0(0x03, 0xf), EncodeExp1(20, 21, 22, 23)});
+        AppendEnd(&code);
+        return code;
+      }();
+      const auto rect_vs = reinterpret_cast<uint64_t>(rect_vertex_code.data());
+      const auto rect_ps = reinterpret_cast<uint64_t>(rect_pixel_code.data());
+      ShaderMapUserData(rect_vs,
+          {.type = Prospero::ShaderBinaryType::kGs, .user_data = &native_user_data,
+           .code_size_bytes = static_cast<u32>(rect_vertex_code.size() * sizeof(u32))});
+      ShaderMapUserData(rect_ps,
+          {.type = Prospero::ShaderBinaryType::kPs,
+           .code_size_bytes = static_cast<u32>(rect_pixel_code.size() * sizeof(u32))});
+      shaders.SetEsShaderBase(rect_vs);
+      shaders.SetPsShaderBase(rect_ps);
+      shaders.SetGsShaderResource2({.user_sgpr = 4});
+      for (u32 i = 0; i < 4; i++) {
+        shaders.SetGsUserSgpr(i, rect_buffer.fields[i], HW::UserSgprType::Vsharp);
+      }
+      registers.SetPsInControl(0x8001);
+      registers.SetPsInputEna(2);
+      registers.SetPsInputAddr(2);
+      registers.SetPsInputSettings(0, 0);
+      for (const auto primitive : {Prospero::PrimitiveType::kRectList,
+                                    Prospero::PrimitiveType::kRectListLegacy}) {
+        const char *rect_name = primitive == Prospero::PrimitiveType::kRectList
+                                    ? "RectListThreeRecords" : "LegacyRectListThreeRecords";
+        user_config.SetPrimitiveType(primitive);
+        registers.SetModeControl({});
+        TextureCacheTestAccess::ClearImage(cache, scheduler.Current(), sparse_color.image_id,
+            {vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1}, {});
+        RenderExecutorTestAccess::DrawAuto(
+            executor, scheduler.Current(), {.vertex_count = 3, .instance_count = 1});
+        const auto pixels = ReadCachedTexel(name, context, sparse_color.image_id,
+                                            {}, {extent, extent, 1});
+        for (u32 y = 0; y < extent; y++) {
+          for (u32 x = 0; x < extent; x++) {
+            const bool inside = x >= 4 && x < 28 && y >= 4 && y < 28;
+            const auto offset = 4 * (y * extent + x);
+            const std::array<float, 4> expected{
+                inside ? (x + 0.5f - 4) / 24 : 0,
+                inside ? (y + 0.5f - 4) / 24 : 0, 0, inside ? 1.0f : 0};
+            for (u32 component = 0; component < 4; component++) {
+              Require(rect_name, "three-vertex rectangle coverage and UV",
+                  std::abs(std::bit_cast<float>(pixels[offset + component]) -
+                           expected[component]) < 0.0001f,
+                  "rectangle reconstruction lost its fourth corner or interpolated UVs");
+            }
+          }
+        }
+      }
       DestroyBuffer(&stencil_readback);
 
       vertex_shader = owned_vertex_shader;
@@ -15011,6 +15097,7 @@ private:
     Require("VulkanHarness", "graphics", available_features12.shaderOutputLayer == true,
             "vertex layer output is not supported");
     Require("VulkanHarness", "graphics", available_features.fillModeNonSolid &&
+                available_features.tessellationShader &&
                 available_depth_clip.depthClipEnable && available_clip_control.depthClipControl &&
                 available_color_write.colorWriteEnable &&
                 available_feedback_layout.attachmentFeedbackLoopLayout &&
@@ -15076,6 +15163,7 @@ private:
     device_features.sampleRateShading = true;
     device_features.shaderInt64 = true;
     device_features.fillModeNonSolid = true;
+    device_features.tessellationShader = true;
     device_info.pEnabledFeatures = &device_features;
     constexpr const char *device_extensions[] = {
         VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME,
@@ -17977,6 +18065,29 @@ TestCase Vop1SdwaNotCapturedByte0Source() {
   return test;
 }
 
+TestCase Vop1SdwaNotPreservesHighWordDestination() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendBufferLoadDword(&code, 0, 30);
+  AppendVMovLiteral(&code, 3, 0xabcd5555u);
+  code.push_back(0x7e066ef9u);
+  code.push_back(0x00061400u);
+  AppendStoreVgpr(&code, 3, 0);
+  AppendEnd(&code);
+
+  TestCase test;
+  test.name = "Vop1SdwaNotPreservesHighWordDestination";
+  test.code = std::move(code);
+  test.initial = {0x12345678u};
+  test.expected = {0xabcda987u};
+  test.opcodes = {O::BUFFER_LOAD_DWORD, O::V_MOV_B32, O::V_NOT_B32,
+                  O::BUFFER_STORE_DWORD, O::S_ENDPGM};
+  test.ir_counts = {{" = BitwiseNot32 ", 1}};
+  test.required_spirv = {"OpNot"};
+  return test;
+}
+
 TestCase Vop1SdwaMovByteDestinations() {
   using O = ShaderOpcode;
 
@@ -27088,6 +27199,7 @@ std::vector<TestCase> MakeCases() {
   AddCase(VectorFfbhI32NativeAndVop3OnGpu);
   AddCase(Vop1SdwaFfblCapturedHighWordSource);
   AddCase(Vop1SdwaNotCapturedByte0Source);
+  AddCase(Vop1SdwaNotPreservesHighWordDestination);
   AddCase(Vop1SdwaMovByteDestinations);
   AddCase(Vop2SdwaSubNcExactByte2Destination);
   AddCase(Vop2SdwaAddNcCapturedHighWordDestination);

tests/shaderCfgTests.cpp

@@ -3435,6 +3435,62 @@ void TestNewShaderRecompilerCapturedVop1SdwaByteConvert() {
   CheckSpirvBinaryValidates(result.spirv);
 }
 
+void TestNewShaderRecompilerVop1SdwaNotDestination() {
+  auto options = MakeCompileOptions(ShaderType::Pixel);
+
+  const auto check_destination = [&](uint32_t modifier, uint32_t dst_sel,
+                                      uint32_t dst_unused) {
+    const uint32_t shader[] = {
+        0x7e066ef9u, modifier, EncodeExp0(0x00, 0x1),
+        EncodeExp1(3, 0, 0, 0), EncodeSopp(0x01),
+    };
+
+    ShaderRecompiler::Decoder::Instruction decoded;
+    ShaderRecompiler::Decoder::DecodeInstruction(shader, 0, decoded);
+    Check(decoded.opcode == ShaderRecompiler::Decoder::Opcode::V_NOT_B32,
+          "V_NOT_B32 SDWA opcode did not decode");
+    Check(decoded.word_count == 2u,
+          "V_NOT_B32 SDWA instruction has the wrong word count");
+    Check(decoded.dst.kind == ShaderRecompiler::Decoder::OperandKind::Vgpr &&
+              decoded.dst.reg == 3u && decoded.dst.sdwa_sel == dst_sel &&
+              decoded.dst.sdwa_dst_unused == dst_unused &&
+              decoded.dst.explicit_sdwa_dst,
+          "V_NOT_B32 SDWA destination metadata is incorrect");
+    Check(decoded.src0.kind == ShaderRecompiler::Decoder::OperandKind::Vgpr &&
+              decoded.src0.reg == 0u && decoded.src0.sdwa_sel == 6u &&
+              !decoded.src0.negate && !decoded.src0.absolute,
+          "V_NOT_B32 SDWA source metadata is incorrect");
+
+    auto result = RecompileForTest(shader, options);
+    CheckSpirvBinaryValidates(result.spirv);
+  };
+
+  check_destination(0x00061400u, 4u, 2u);
+  check_destination(0x00060500u, 5u, 0u);
+  check_destination(0x00060000u, 0u, 0u);
+
+  const auto check_rejected = [&](uint32_t modifier, const char *message) {
+    const uint32_t instruction[] = {0x7e066ef9u, modifier};
+
+    ShaderRecompiler::Decoder::Instruction decoded;
+    ShaderRecompiler::Decoder::DecodeInstruction(instruction, 0, decoded);
+    Check(decoded.word_count == 2u &&
+              decoded.opcode == ShaderRecompiler::Decoder::Opcode::UNSUPPORTED,
+          message);
+  };
+
+  check_rejected(0x00060700u, "V_NOT_B32 SDWA accepted destination selector 7");
+  check_rejected(0x00070600u, "V_NOT_B32 SDWA accepted source selector 7");
+  check_rejected(0x00061c00u,
+                 "V_NOT_B32 SDWA accepted destination unused value 3");
+  check_rejected(0x00160400u, "V_NOT_B32 SDWA accepted source negation");
+  check_rejected(0x00260400u, "V_NOT_B32 SDWA accepted source absolute");
+  check_rejected(0x00062400u, "V_NOT_B32 SDWA accepted clamp");
+  check_rejected(0x00064400u, "V_NOT_B32 SDWA accepted output modifier");
+  check_rejected(0x00000400u,
+                 "V_NOT_B32 SDWA partial destination accepted a byte source");
+}
+
 void TestNewShaderRecompilerBootB16PackedAndSdwaOpcodes() {
   const uint32_t shader[] = {
       0xd7070001u,
@@ -13397,6 +13453,7 @@ int main() {
   TestNewShaderRecompilerDsReadWrite2Translation();
   TestNewShaderRecompilerDsWideAndAtomicTranslation();
   TestNewShaderRecompilerCapturedVop1SdwaByteConvert();
+  TestNewShaderRecompilerVop1SdwaNotDestination();
   TestNewShaderRecompilerScalarMemoryBindingDomains();
   // Opcode semantics and optimized SPIR-V are exercised by
   // ShaderRecompilerComputeTests; keep the distinct decoder contract checks