Add automatic sccache/ccache detection to CMake and remove deprecated Ninja setup action (#734)

* remove outdated ninja setup action

* Remove deprecated ninja setup action

The seanmiddleditch/gha-setup-ninja action has been archived by its
maintainer, who explicitly states that Ninja is now pre-installed on
all GitHub-hosted runners. This removes the action from all three
build jobs (Windows, macOS, Linux) and removes ninja from the package
install lists on macOS and Linux.

* Add automatic ccache/sccache detection to CMakeLists.txt

Detects sccache (preferred on Windows/MSVC) or ccache at configure time
and sets CMAKE_C_COMPILER_LAUNCHER / CMAKE_CXX_COMPILER_LAUNCHER
automatically. Works for local builds without manual flags.

Priority: sccache > ccache (sccache handles MSVC PDBs better).

* Delete github-actions-version-research.md

* Respect user-specified CMAKE_C_COMPILER_LAUNCHER / CMAKE_CXX_COMPILER_LAUNCHER

Only auto-detect sccache/ccache if no launcher was explicitly provided
via -DCMAKE_C_COMPILER_LAUNCHER=... or -DCMAKE_CXX_COMPILER_LAUNCHER=...

* Preserve explicit compiler launcher settings per language

---------

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

.github/workflows/build.yml

@@ -43,9 +43,6 @@ jobs:
       - name: Setup Visual Studio environment
         uses: ilammy/msvc-dev-cmd@v1
 
-      - name: Setup Ninja
-        uses: seanmiddleditch/gha-setup-ninja@v5
-
       - name: Locate vcpkg
         id: vcpkg
         shell: pwsh
@@ -155,7 +152,7 @@ jobs:
 
       - name: Install build dependencies
         shell: bash
-        run: brew install glslang ninja
+        run: brew install glslang
 
       - name: Install Qt 6.10.3
         uses: jurplel/install-qt-action@v4
@@ -318,7 +315,6 @@ jobs:
             libxrandr-dev \
             libxss-dev \
             lld \
-            ninja-build \
             pkg-config \
             wayland-protocols
 

CMakeLists.txt

@@ -12,6 +12,22 @@ endif()
 
 project(Kyty)
 
+find_program(SCCACHE_PROGRAM sccache)
+find_program(CCACHE_PROGRAM ccache)
+
+foreach(lang IN ITEMS C CXX)
+    # An explicitly empty launcher disables caching for this language.
+    if(NOT DEFINED CMAKE_${lang}_COMPILER_LAUNCHER)
+        if(SCCACHE_PROGRAM)
+            set(CMAKE_${lang}_COMPILER_LAUNCHER "${SCCACHE_PROGRAM}")
+            message(STATUS "sccache enabled for ${lang}: ${SCCACHE_PROGRAM}")
+        elseif(CCACHE_PROGRAM)
+            set(CMAKE_${lang}_COMPILER_LAUNCHER "${CCACHE_PROGRAM}")
+            message(STATUS "ccache enabled for ${lang}: ${CCACHE_PROGRAM}")
+        endif()
+    endif()
+endforeach()
+
 set(KYTY_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/src")
 set(KYTY_TESTS_DIR "${CMAKE_CURRENT_SOURCE_DIR}/tests")
 set(KYTY_THIRD_PARTY_DIR "${CMAKE_CURRENT_SOURCE_DIR}/3rdparty")

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/shader/recompiler/frontend/decode/ScalarAluOps.cpp

@@ -74,7 +74,8 @@ constexpr OpcodeMap SOPC_OPCODE_LIST[] = {
     {0x03u, Opcode::S_CMP_GE_I32},  {0x04u, Opcode::S_CMP_LT_I32},  {0x05u, Opcode::S_CMP_LE_I32},
     {0x06u, Opcode::S_CMP_EQ_U32},  {0x07u, Opcode::S_CMP_LG_U32},  {0x08u, Opcode::S_CMP_GT_U32},
     {0x09u, Opcode::S_CMP_GE_U32},  {0x0au, Opcode::S_CMP_LT_U32},  {0x0bu, Opcode::S_CMP_LE_U32},
-    {0x0cu, Opcode::S_BITCMP0_B32}, {0x0du, Opcode::S_BITCMP1_B32}, {0x12u, Opcode::S_CMP_EQ_U64},
+    {0x0cu, Opcode::S_BITCMP0_B32}, {0x0du, Opcode::S_BITCMP1_B32},
+    {0x0eu, Opcode::S_BITCMP0_B64}, {0x0fu, Opcode::S_BITCMP1_B64}, {0x12u, Opcode::S_CMP_EQ_U64},
     {0x13u, Opcode::S_CMP_LG_U64},
 };
 

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h

@@ -73,6 +73,8 @@ enum class Opcode {
 	S_SUB_I32,
 	S_BITCMP0_B32,
 	S_BITCMP1_B32,
+	S_BITCMP0_B64,
+	S_BITCMP1_B64,
 	S_BITSET0_B32,
 	S_BITSET1_B32,
 	S_BITSET0_B64,

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/frontend/translate/Integer.cpp

@@ -507,6 +507,18 @@ bool Translator::S_BITCMP_B32(const Decoder::Instruction& inst, bool expected) {
 	return true;
 }
 
+bool Translator::S_BITCMP_B64(const Decoder::Instruction& inst, bool expected) {
+	const auto value    = ReadU32Pair(inst.src0);
+	const auto offset   = ir.BitwiseAnd(ReadU32(inst.src1), IR::U32(IR::Value(63u)));
+	const auto word_bit = ir.BitwiseAnd(offset, IR::U32(IR::Value(31u)));
+	const auto word =
+	    ir.Select(ir.ULessThan(offset, IR::U32(IR::Value(32u))), value[0], value[1]);
+	const auto bit =
+	    IR::U32(ir.Emit(IR::ValueOpcode::BitFieldUExtract, {word, word_bit, IR::Value(1u)}));
+	WriteCompareResult(inst.dst, ir.IEqual(bit, IR::U32(IR::Value(expected ? 1u : 0u))));
+	return true;
+}
+
 bool Translator::V_ALIGNBIT_B32(const Decoder::Instruction& inst) {
 	const auto hi      = ReadU32(inst.src0);
 	const auto lo      = ReadU32(inst.src1);

src/graphics/shader/recompiler/frontend/translate/Scalar.cpp

@@ -208,6 +208,8 @@ bool Translator::EmitScalar(const Decoder::Instruction& inst) {
 		case O::S_BFE_U64: return S_BFE_U64(inst);
 		case O::S_BITCMP0_B32: return S_BITCMP_B32(inst, false);
 		case O::S_BITCMP1_B32: return S_BITCMP_B32(inst, true);
+		case O::S_BITCMP0_B64: return S_BITCMP_B64(inst, false);
+		case O::S_BITCMP1_B64: return S_BITCMP_B64(inst, true);
 		case O::S_PACK_LL_B32_B16: return PackB16(inst, false, false);
 		case O::S_PACK_LH_B32_B16: return PackB16(inst, false, true);
 		case O::S_PACK_HH_B32_B16: return PackB16(inst, true, true);

src/graphics/shader/recompiler/frontend/translate/Translator.h

@@ -205,6 +205,7 @@ private:
 	bool    V_BFE_U32(const Decoder::Instruction& inst, bool sign);
 	bool    V_BFI_B32(const Decoder::Instruction& inst);
 	bool    S_BITCMP_B32(const Decoder::Instruction& inst, bool expected);
+	bool    S_BITCMP_B64(const Decoder::Instruction& inst, bool expected);
 	bool    V_ALIGNBIT_B32(const Decoder::Instruction& inst);
 	bool    V_ALIGNBYTE_B32(const Decoder::Instruction& inst);
 	bool    V_LSHL_ADD_U32(const Decoder::Instruction& inst);

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/ShaderRecompilerComputeTests.cpp

@@ -17057,6 +17057,117 @@ TestCase ScalarGetpcWritesNextInstructionPc() {
           {O::S_GETPC_B64, O::V_MOV_B32, O::BUFFER_STORE_DWORD, O::S_ENDPGM}};
 }
 
+TestCase ScalarBitcmpB64DynamicOperands() {
+  using O = ShaderOpcode;
+
+  struct Input {
+    u32 low;
+    u32 high;
+    u32 index;
+  };
+  const Input inputs[] = {
+      {1u, 0u, 0u}, {0x80000000u, 0u, 31u}, {0u, 1u, 32u},
+      {0u, 0x80000000u, 63u}, {1u, 0u, 64u},
+      {0x80000000u, 0u, 95u}, {0u, 1u, 96u},
+      {0u, 0x80000000u, 127u}, {1u, 0u, 128u},
+      {0u, 0x80000000u, 0xffffffffu}, {0u, 0u, 17u},
+      {0xffffffffu, 0xffffffffu, 42u},
+      {0x89abcdefu, 0x76543210u, 4u},
+      {0x89abcdefu, 0x76543210u, 36u},
+  };
+  TestCase test;
+  test.name = "ScalarBitcmpB64DynamicOperands";
+  for (const auto &input : inputs) {
+    test.initial.insert(test.initial.end(), {input.low, input.high, input.index});
+  }
+  test.expected = test.initial;
+  for (u32 i = 0; i < std::size(inputs); ++i) {
+    // Load the value and index on the GPU so the compares cannot constant-fold.
+    for (u32 component = 0; component < 3; ++component) {
+      AppendVMovU32(&test.code, 30, (i * 3 + component) * 4);
+      AppendBufferLoadDword(&test.code, 0, 30);
+      test.code.push_back(EncodeVop1(0x02, 20 + component, Vgpr(0)));
+    }
+    const uint64_t value = uint64_t{inputs[i].low} | (uint64_t{inputs[i].high} << 32);
+    const u32 bit = static_cast<u32>((value >> (inputs[i].index & 63u)) & 1u);
+    for (u32 expected_bit = 0; expected_bit < 2; ++expected_bit) {
+      const bool scc = bit == expected_bit;
+      test.code.push_back(EncodeSopc(0x0e + expected_bit, 20, 22));
+      test.code.push_back(EncodeSop2(0x0a, 24, InlineU32(1), InlineU32(0)));
+      AppendStoreSgpr(&test.code, 24, static_cast<u32>(test.expected.size()));
+      test.expected.push_back(scc ? 1u : 0u);
+
+      // Exercise SCC-driven control flow as well as S_CSELECT_B32.
+      test.code.push_back(EncodeSopp(0x05, 2)); // s_cbranch_scc1 to the true arm.
+      test.code.push_back(EncodeSMovB32(24, InlineU32(3)));
+      test.code.push_back(EncodeSopp(0x02, 1)); // s_branch past the true arm.
+      test.code.push_back(EncodeSMovB32(24, InlineU32(7)));
+      AppendStoreSgpr(&test.code, 24, static_cast<u32>(test.expected.size()));
+      test.expected.push_back(scc ? 7u : 3u);
+    }
+  }
+  AppendEnd(&test.code);
+  test.initial.resize(test.expected.size());
+  test.opcodes = {O::V_MOV_B32, O::BUFFER_LOAD_DWORD, O::V_READFIRSTLANE_B32,
+                  O::S_BITCMP0_B64, O::S_BITCMP1_B64, O::S_CSELECT_B32,
+                  O::S_CBRANCH_SCC1, O::S_BRANCH, O::S_MOV_B32,
+                  O::BUFFER_STORE_DWORD, O::S_ENDPGM};
+  test.required_spirv = {"OpBitFieldUExtract", "OpULessThan", "OpSelect"};
+  return test;
+}
+
+TestCase ScalarBitcmpB64IntegerConstants() {
+  using O = ShaderOpcode;
+
+  struct Source {
+    u32 encoding;
+    uint64_t bits;
+  };
+  // Negative integer inlines sign-extend; B64 literals zero-extend.
+  const Source sources[] = {
+      {128u, 0u}, {129u, 1u}, {192u, 64u},
+      {193u, 0xffffffffffffffffull}, {208u, 0xfffffffffffffff0ull},
+      {255u, 0xffffffffu}, {255u, 0x80000000u},
+  };
+  const u32 indices[] = {0u, 4u, 6u, 31u, 32u, 63u, 64u, 127u};
+  TestCase test;
+  test.name = "ScalarBitcmpB64IntegerConstants";
+  test.initial.assign(std::begin(indices), std::end(indices));
+  test.expected = test.initial;
+  for (u32 i = 0; i < std::size(indices); ++i) {
+    AppendVMovU32(&test.code, 30, i * 4);
+    AppendBufferLoadDword(&test.code, 0, 30);
+    test.code.push_back(EncodeVop1(0x02, 22, Vgpr(0)));
+    for (const auto &source : sources) {
+      const u32 bit = static_cast<u32>((source.bits >> (indices[i] & 63u)) & 1u);
+      for (u32 expected_bit = 0; expected_bit < 2; ++expected_bit) {
+        test.code.push_back(EncodeSopc(0x0e + expected_bit, source.encoding, 22));
+        if (source.encoding == 255u) {
+          test.code.push_back(static_cast<u32>(source.bits));
+        }
+        test.code.push_back(EncodeSop2(0x0a, 24, InlineU32(1), InlineU32(0)));
+        AppendStoreSgpr(&test.code, 24, static_cast<u32>(test.expected.size()));
+        test.expected.push_back(bit == expected_bit ? 1u : 0u);
+      }
+    }
+  }
+  // An instruction has one literal word even when both sources select it.
+  for (u32 expected_bit = 0; expected_bit < 2; ++expected_bit) {
+    test.code.push_back(EncodeSopc(0x0e + expected_bit, 255u, 255u));
+    test.code.push_back(0x8000001fu); // Select bit 31 of the same zero-extended literal.
+    test.code.push_back(EncodeSop2(0x0a, 24, InlineU32(1), InlineU32(0)));
+    AppendStoreSgpr(&test.code, 24, static_cast<u32>(test.expected.size()));
+    test.expected.push_back(expected_bit);
+  }
+  AppendEnd(&test.code);
+  test.initial.resize(test.expected.size());
+  test.opcodes = {O::V_MOV_B32, O::BUFFER_LOAD_DWORD, O::V_READFIRSTLANE_B32,
+                  O::S_BITCMP0_B64, O::S_BITCMP1_B64, O::S_CSELECT_B32,
+                  O::BUFFER_STORE_DWORD, O::S_ENDPGM};
+  test.required_spirv = {"OpBitFieldUExtract"};
+  return test;
+}
+
 TestCase ScalarBitfieldPack() {
   using O = ShaderOpcode;
 
@@ -18065,6 +18176,29 @@ 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;
 
@@ -27149,6 +27283,8 @@ std::vector<TestCase> MakeCases() {
   cases.push_back(ScalarSubvectorLoops(64));
   AddCase(ScalarGetpcWritesNextInstructionPc);
   AddCase(ScalarBitfieldPack);
+  AddCase(ScalarBitcmpB64DynamicOperands);
+  AddCase(ScalarBitcmpB64IntegerConstants);
   AddCase(ScalarBrevB32PreservesScc);
   AddCase(ScalarBfeI32CapturedRawSignExtends);
   AddCase(BitfieldExtractWidthPastEndEdges);
@@ -27176,6 +27312,7 @@ 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