diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml index ffe7c22fb..3d77c5800 100644 --- a/.github/workflows/build.yml +++ b/.github/workflows/build.yml @@ -26,14 +26,14 @@ jobs: runs-on: ubuntu-24.04 continue-on-error: true steps: - - uses: actions/checkout@v5 + - uses: actions/checkout@v6 - uses: fsfe/reuse-action@v5 clang-format: runs-on: ubuntu-24.04 continue-on-error: true steps: - - uses: actions/checkout@v5 + - uses: actions/checkout@v6 with: fetch-depth: 0 - name: Install @@ -46,7 +46,7 @@ jobs: env: COMMIT_RANGE: ${{ github.event.pull_request.base.sha }}..${{ github.event.pull_request.head.sha }} run: ./.ci/clang-format.sh - + get-info: runs-on: ubuntu-24.04 outputs: @@ -54,7 +54,7 @@ jobs: shorthash: ${{ steps.vars.outputs.shorthash }} fullhash: ${{ steps.vars.outputs.fullhash }} steps: - - uses: actions/checkout@v5 + - uses: actions/checkout@v6 - name: Get date and git hash id: vars run: | @@ -69,23 +69,23 @@ jobs: runs-on: windows-2025 needs: get-info steps: - - uses: actions/checkout@v5 + - uses: actions/checkout@v6 with: submodules: recursive - name: Cache CMake Configuration - uses: actions/cache@v4 + uses: actions/cache@v5 env: cache-name: ${{ runner.os }}-sdl-ninja-cache-cmake-configuration with: - path: | + path: | ${{github.workspace}}/build key: ${{ env.cache-name }}-${{ hashFiles('**/CMakeLists.txt', 'cmake/**') }} restore-keys: | ${{ env.cache-name }}- - name: Cache CMake Build - uses: hendrikmuhs/ccache-action@v1.2.19 + uses: hendrikmuhs/ccache-action@v1.2.21 env: cache-name: ${{ runner.os }}-sdl-cache-cmake-build with: @@ -99,7 +99,7 @@ jobs: run: cmake --build ${{github.workspace}}/build --config ${{env.BUILD_TYPE}} --parallel $env:NUMBER_OF_PROCESSORS - name: Upload Windows SDL artifact - uses: actions/upload-artifact@v4 + uses: actions/upload-artifact@v6 with: name: shadps4-win64-sdl-${{ needs.get-info.outputs.date }}-${{ needs.get-info.outputs.shorthash }} path: ${{github.workspace}}/build/shadPS4.exe @@ -108,7 +108,7 @@ jobs: runs-on: macos-15 needs: get-info steps: - - uses: actions/checkout@v5 + - uses: actions/checkout@v6 with: submodules: recursive @@ -118,18 +118,18 @@ jobs: xcode-version: latest - name: Cache CMake Configuration - uses: actions/cache@v4 - env: + uses: actions/cache@v5 + env: cache-name: ${{ runner.os }}-sdl-cache-cmake-configuration - with: - path: | - ${{github.workspace}}/build - key: ${{ env.cache-name }}-${{ hashFiles('**/CMakeLists.txt', 'cmake/**') }} - restore-keys: | - ${{ env.cache-name }}- + with: + path: | + ${{github.workspace}}/build + key: ${{ env.cache-name }}-${{ hashFiles('**/CMakeLists.txt', 'cmake/**') }} + restore-keys: | + ${{ env.cache-name }}- - name: Cache CMake Build - uses: hendrikmuhs/ccache-action@v1.2.19 + uses: hendrikmuhs/ccache-action@v1.2.21 env: cache-name: ${{runner.os}}-sdl-cache-cmake-build with: @@ -150,7 +150,7 @@ jobs: mv ${{github.workspace}}/build/shadps4 upload mv ${{github.workspace}}/build/MoltenVK_icd.json upload mv ${{github.workspace}}/build/libMoltenVK.dylib upload - - uses: actions/upload-artifact@v4 + - uses: actions/upload-artifact@v6 with: name: shadps4-macos-sdl-${{ needs.get-info.outputs.date }}-${{ needs.get-info.outputs.shorthash }} path: upload/ @@ -159,7 +159,7 @@ jobs: runs-on: ubuntu-24.04 needs: get-info steps: - - uses: actions/checkout@v5 + - uses: actions/checkout@v6 with: submodules: recursive @@ -172,18 +172,18 @@ jobs: run: sudo apt-get update && sudo apt install -y libx11-dev libxext-dev libwayland-dev libdecor-0-dev libxkbcommon-dev libglfw3-dev libgles2-mesa-dev libfuse2 clang-19 mold build-essential libasound2-dev libpulse-dev libopenal-dev libudev-dev libxcursor-dev libxi-dev libxss-dev libxtst-dev - name: Cache CMake Configuration - uses: actions/cache@v4 - env: + uses: actions/cache@v5 + env: cache-name: ${{ runner.os }}-sdl-cache-cmake-configuration - with: - path: | - ${{github.workspace}}/build - key: ${{ env.cache-name }}-${{ hashFiles('**/CMakeLists.txt', 'cmake/**') }} - restore-keys: | - ${{ env.cache-name }}- + with: + path: | + ${{github.workspace}}/build + key: ${{ env.cache-name }}-${{ hashFiles('**/CMakeLists.txt', 'cmake/**') }} + restore-keys: | + ${{ env.cache-name }}- - name: Cache CMake Build - uses: hendrikmuhs/ccache-action@v1.2.19 + uses: hendrikmuhs/ccache-action@v1.2.21 env: cache-name: ${{ runner.os }}-sdl-cache-cmake-build with: @@ -195,23 +195,23 @@ jobs: - name: Build run: cmake --build ${{github.workspace}}/build --config ${{env.BUILD_TYPE}} --parallel $(nproc) - - - name: Package and Upload Linux(ubuntu64) SDL artifact + + - name: Package and Upload Linux(ubuntu64) SDL artifact run: | ls -la ${{ github.workspace }}/build/shadps4 - - - uses: actions/upload-artifact@v4 + + - uses: actions/upload-artifact@v6 with: name: shadps4-ubuntu64-${{ needs.get-info.outputs.date }}-${{ needs.get-info.outputs.shorthash }} path: ${{ github.workspace }}/build/shadps4 - name: Run AppImage packaging script run: ./.github/linux-appimage-sdl.sh - + - name: Package and Upload Linux SDL artifact run: | tar cf shadps4-linux-sdl.tar.gz -C ${{github.workspace}}/build shadps4 - - uses: actions/upload-artifact@v4 + - uses: actions/upload-artifact@v6 with: name: shadps4-linux-sdl-${{ needs.get-info.outputs.date }}-${{ needs.get-info.outputs.shorthash }} path: Shadps4-sdl.AppImage @@ -220,7 +220,7 @@ jobs: runs-on: ubuntu-24.04 needs: get-info steps: - - uses: actions/checkout@v5 + - uses: actions/checkout@v6 with: submodules: recursive @@ -228,18 +228,18 @@ jobs: run: sudo apt-get update && sudo apt install -y libx11-dev libxext-dev libwayland-dev libdecor-0-dev libxkbcommon-dev libglfw3-dev libgles2-mesa-dev libfuse2 gcc-14 mold build-essential libasound2-dev libpulse-dev libopenal-dev libudev-dev libxcursor-dev libxi-dev libxss-dev libxtst-dev - name: Cache CMake Configuration - uses: actions/cache@v4 - env: + uses: actions/cache@v5 + env: cache-name: ${{ runner.os }}-sdl-gcc-cache-cmake-configuration - with: - path: | - ${{github.workspace}}/build - key: ${{ env.cache-name }}-${{ hashFiles('**/CMakeLists.txt', 'cmake/**') }} - restore-keys: | - ${{ env.cache-name }}- + with: + path: | + ${{github.workspace}}/build + key: ${{ env.cache-name }}-${{ hashFiles('**/CMakeLists.txt', 'cmake/**') }} + restore-keys: | + ${{ env.cache-name }}- - name: Cache CMake Build - uses: hendrikmuhs/ccache-action@v1.2.19 + uses: hendrikmuhs/ccache-action@v1.2.21 env: cache-name: ${{ runner.os }}-sdl-gcc-cache-cmake-build with: @@ -258,7 +258,7 @@ jobs: runs-on: ubuntu-latest steps: - name: Download all artifacts - uses: actions/download-artifact@v5 + uses: actions/download-artifact@v8 with: path: ./artifacts @@ -266,7 +266,7 @@ jobs: run: | chmod -R a+x ./artifacts/shadps4-linux-sdl-* chmod -R a+x ./artifacts/shadps4-macos-sdl-* - + - name: Compress individual directories (without parent directory) run: | cd ./artifacts @@ -277,7 +277,7 @@ jobs: (cd "$dir_name" && zip -r "../${dir_name}.zip" .) fi done - + - name: Get latest release information id: get_latest_release env: @@ -351,52 +351,52 @@ jobs: upload_url="https://uploads.github.com/repos/$REPO/releases/$release_id/assets?name=$filename" curl -X POST -H "Authorization: token $GITHUB_TOKEN" -H "Content-Type: application/octet-stream" --data-binary @"$file" "$upload_url" done - + - name: Get current pre-release information env: GITHUB_TOKEN: ${{ secrets.SHADPS4_TOKEN_REPO }} run: | api_url="https://api.github.com/repos/${{ github.repository }}/releases" - + # Get all releases (sorted by date) releases=$(curl -H "Authorization: token $GITHUB_TOKEN" "$api_url") - + # Capture the most recent pre-release (assuming the first one is the latest) current_release=$(echo "$releases" | jq -c '.[] | select(.prerelease == true) | .published_at' | sort -r | head -n 1) - + # Remove extra quotes from captured date current_release=$(echo $current_release | tr -d '"') - + # Export the current published_at to be available for the next step echo "CURRENT_PUBLISHED_AT=$current_release" >> $GITHUB_ENV - + - name: Delete old pre-releases and tags env: GITHUB_TOKEN: ${{ secrets.SHADPS4_TOKEN_REPO }} run: | api_url="https://api.github.com/repos/${{ github.repository }}/releases" - + # Get current pre-releases releases=$(curl -H "Authorization: token $GITHUB_TOKEN" "$api_url") - + # Remove extra quotes from captured date CURRENT_PUBLISHED_AT=$(echo $CURRENT_PUBLISHED_AT | tr -d '"') - + # Convert CURRENT_PUBLISHED_AT para timestamp Unix current_published_ts=$(date -d "$CURRENT_PUBLISHED_AT" +%s) - + # Identify pre-releases echo "$releases" | jq -c '.[] | select(.prerelease == true)' | while read -r release; do release_date=$(echo "$release" | jq -r '.published_at') release_id=$(echo "$release" | jq -r '.id') release_tag=$(echo "$release" | jq -r '.tag_name') - + # Remove extra quotes from captured date release_date=$(echo $release_date | tr -d '"') - + # Convert release_date para timestamp Unix release_date_ts=$(date -d "$release_date" +%s) - + # Compare timestamps and delete old pre-releases if [[ "$release_date_ts" -lt "$current_published_ts" ]]; then echo "Deleting old pre-release: $release_id from $release_date with tag: $release_tag" diff --git a/CMakeLists.txt b/CMakeLists.txt index e1d05cdce..a61bc2370 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -202,7 +202,7 @@ execute_process( # Set Version set(EMULATOR_VERSION_MAJOR "0") -set(EMULATOR_VERSION_MINOR "14") +set(EMULATOR_VERSION_MINOR "15") set(EMULATOR_VERSION_PATCH "1") set_source_files_properties(src/shadps4.rc PROPERTIES COMPILE_DEFINITIONS "EMULATOR_VERSION_MAJOR=${EMULATOR_VERSION_MAJOR};EMULATOR_VERSION_MINOR=${EMULATOR_VERSION_MINOR};EMULATOR_VERSION_PATCH=${EMULATOR_VERSION_PATCH}") diff --git a/dist/net.shadps4.shadPS4.metainfo.xml b/dist/net.shadps4.shadPS4.metainfo.xml index 210ca1c5e..8a7fa852b 100644 --- a/dist/net.shadps4.shadPS4.metainfo.xml +++ b/dist/net.shadps4.shadPS4.metainfo.xml @@ -38,7 +38,10 @@ Game - + + https://github.com/shadps4-emu/shadPS4/releases/tag/v.0.15.0 + + https://github.com/shadps4-emu/shadPS4/releases/tag/v.0.14.0 diff --git a/src/core/file_format/psf.cpp b/src/core/file_format/psf.cpp index e647059f0..c5be7410a 100644 --- a/src/core/file_format/psf.cpp +++ b/src/core/file_format/psf.cpp @@ -113,6 +113,7 @@ bool PSF::Encode(const std::filesystem::path& filepath) const { LOG_ERROR(Core, "Failed to write PSF file. Written {} Expected {}", written, psf_buffer.size()); } + file.Close(); return written == psf_buffer.size(); } diff --git a/src/core/libraries/kernel/threads/pthread.cpp b/src/core/libraries/kernel/threads/pthread.cpp index f97451154..3742db5cf 100644 --- a/src/core/libraries/kernel/threads/pthread.cpp +++ b/src/core/libraries/kernel/threads/pthread.cpp @@ -242,7 +242,7 @@ int PS4_SYSV_ABI posix_pthread_create_name_np(PthreadT* thread, const PthreadAtt new_thread->attr.sched_policy = curthread->attr.sched_policy; } - static int TidCounter = 1; + static std::atomic TidCounter = 1; new_thread->tid = ++TidCounter; if (new_thread->attr.stackaddr_attr == nullptr) { diff --git a/src/core/libraries/network/ssl2.cpp b/src/core/libraries/network/ssl2.cpp index 0b408d094..3a7fd71e5 100644 --- a/src/core/libraries/network/ssl2.cpp +++ b/src/core/libraries/network/ssl2.cpp @@ -114,7 +114,13 @@ int PS4_SYSV_ABI sceSslFreeCaCerts(s32 ssl_ctx_id, OrbisSslCaCerts* certs) { if (certs == nullptr) { return ORBIS_SSL_ERROR_INVALID_ARGUMENT; } - delete (certs->certs); + if (certs->certs != nullptr) { + for (s32 data = 0; data < certs->num; data++) { + free(certs->certs[data].ptr); + } + delete (certs->certs); + } + // delete (certs->pool); return ORBIS_OK; } @@ -139,7 +145,12 @@ int PS4_SYSV_ABI sceSslGetCaCerts(s32 ssl_ctx_id, OrbisSslCaCerts* certs) { if (certs == nullptr) { return ORBIS_SSL_ERROR_INVALID_ARGUMENT; } - certs->certs = new OrbisSslData{nullptr, 0}; + // Allocate a buffer to store dummy data in. + const char* dummy_data = "dummy"; + u64 dummy_length = strlen(dummy_data) + 1; + char* data = static_cast(malloc(dummy_length)); + strncpy(data, dummy_data, dummy_length); + certs->certs = new OrbisSslData{data, dummy_length}; certs->num = 1; certs->pool = nullptr; return ORBIS_OK; diff --git a/src/core/linker.cpp b/src/core/linker.cpp index a8a97b2a6..889f3a298 100644 --- a/src/core/linker.cpp +++ b/src/core/linker.cpp @@ -361,8 +361,10 @@ bool Linker::Resolve(const std::string& name, Loader::SymbolType sym_type, Modul return_info->virtual_address = AeroLib::GetStub(sr.name.c_str()); return_info->name = "Unknown !!!"; } - LOG_WARNING(Core_Linker, "Linker: Stub resolved {} as {} (lib: {}, mod: {})", sr.name, - return_info->name, library->name, module->name); + if (library->name != "libc" && library->name != "libSceFios2") { + LOG_WARNING(Core_Linker, "Linker: Stub resolved {} as {} (lib: {}, mod: {})", sr.name, + return_info->name, library->name, module->name); + } return false; } diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp index e2a969b61..0b05dcef4 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp @@ -220,20 +220,33 @@ Id EmitImageGradient(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id Id EmitImageRead(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id lod, Id ms) { const auto& texture = ctx.images[handle & 0xFFFF]; - const Id image = ctx.OpLoad(texture.image_type, texture.id); const Id color_type = texture.data_types->Get(4); ImageOperands operands; operands.Add(spv::ImageOperandsMask::Sample, ms); Id texel; if (!texture.is_storage) { + const Id image = ctx.OpLoad(texture.image_type, texture.id); operands.Add(spv::ImageOperandsMask::Lod, lod); texel = ctx.OpImageFetch(color_type, image, coords, operands.mask, operands.operands); } else { + Id image_ptr = texture.id; if (ctx.profile.supports_image_load_store_lod) { operands.Add(spv::ImageOperandsMask::Lod, lod); } else if (Sirit::ValidId(lod)) { - LOG_WARNING(Render, "Image read with LOD not supported by driver"); +#if 1 + // It's confusing what interactions will cause this code path so leave it as + // unreachable until a case is found. + // Normally IMAGE_LOAD_MIP should translate -> OpImageFetch + UNREACHABLE_MSG("Unsupported ImageRead with Lod"); +#else + LOG_WARNING(Render, "Fallback for ImageRead with LOD"); + ASSERT(texture.mip_fallback_mode == MipStorageFallbackMode::DynamicIndex); + const Id single_image_ptr_type = + ctx.TypePointer(spv::StorageClass::UniformConstant, texture.image_type); + image_ptr = ctx.OpAccessChain(single_image_ptr_type, image_ptr, std::array{lod}); +#endif } + const Id image = ctx.OpLoad(texture.image_type, image_ptr); texel = ctx.OpImageRead(color_type, image, coords, operands.mask, operands.operands); } return texture.is_integer ? ctx.OpBitcast(ctx.F32[4], texel) : texel; @@ -242,15 +255,20 @@ Id EmitImageRead(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id lod void EmitImageWrite(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id lod, Id ms, Id color) { const auto& texture = ctx.images[handle & 0xFFFF]; - const Id image = ctx.OpLoad(texture.image_type, texture.id); + Id image_ptr = texture.id; const Id color_type = texture.data_types->Get(4); ImageOperands operands; operands.Add(spv::ImageOperandsMask::Sample, ms); if (ctx.profile.supports_image_load_store_lod) { operands.Add(spv::ImageOperandsMask::Lod, lod); } else if (Sirit::ValidId(lod)) { - LOG_WARNING(Render, "Image write with LOD not supported by driver"); + LOG_WARNING(Render, "Fallback for ImageWrite with LOD"); + ASSERT(texture.mip_fallback_mode == MipStorageFallbackMode::DynamicIndex); + const Id single_image_ptr_type = + ctx.TypePointer(spv::StorageClass::UniformConstant, texture.image_type); + image_ptr = ctx.OpAccessChain(single_image_ptr_type, image_ptr, std::array{lod}); } + const Id image = ctx.OpLoad(texture.image_type, image_ptr); const Id texel = texture.is_integer ? ctx.OpBitcast(color_type, color) : color; ctx.OpImageWrite(image, coords, texel, operands.mask, operands.operands); } diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp index 261155ab5..c0e469964 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp @@ -961,23 +961,33 @@ void EmitContext::DefineImagesAndSamplers() { const auto nfmt = sharp.GetNumberFmt(); const bool is_integer = AmdGpu::IsInteger(nfmt); const bool is_storage = image_desc.is_written; + const MipStorageFallbackMode mip_fallback_mode = image_desc.mip_fallback_mode; const VectorIds& data_types = GetAttributeType(*this, nfmt); const Id sampled_type = data_types[1]; const Id image_type{ImageType(*this, image_desc, sampled_type)}; - const Id pointer_type{TypePointer(spv::StorageClass::UniformConstant, image_type)}; + + const u32 num_bindings = image_desc.NumBindings(info); + Id pointee_type = image_type; + if (mip_fallback_mode == MipStorageFallbackMode::DynamicIndex) { + pointee_type = TypeArray(pointee_type, ConstU32(num_bindings)); + } + + const Id pointer_type{TypePointer(spv::StorageClass::UniformConstant, pointee_type)}; const Id id{AddGlobalVariable(pointer_type, spv::StorageClass::UniformConstant)}; - Decorate(id, spv::Decoration::Binding, binding.unified++); + Decorate(id, spv::Decoration::Binding, binding.unified); + binding.unified += num_bindings; Decorate(id, spv::Decoration::DescriptorSet, 0U); + // TODO better naming for resources (flattened sharp_idx is not informative) Name(id, fmt::format("{}_{}{}", stage, "img", image_desc.sharp_idx)); images.push_back({ .data_types = &data_types, .id = id, .sampled_type = is_storage ? sampled_type : TypeSampledImage(image_type), - .pointer_type = pointer_type, .image_type = image_type, .view_type = sharp.GetViewType(image_desc.is_array), .is_integer = is_integer, .is_storage = is_storage, + .mip_fallback_mode = mip_fallback_mode, }); interfaces.push_back(id); } diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.h b/src/shader_recompiler/backend/spirv/spirv_emit_context.h index 9bb2b7d7a..a9c6f0968 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.h +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.h @@ -293,11 +293,11 @@ public: const VectorIds* data_types; Id id; Id sampled_type; - Id pointer_type; Id image_type; AmdGpu::ImageType view_type; bool is_integer = false; bool is_storage = false; + MipStorageFallbackMode mip_fallback_mode{}; }; enum class PointerType : u32 { diff --git a/src/shader_recompiler/frontend/translate/translate.cpp b/src/shader_recompiler/frontend/translate/translate.cpp index 0489a5d88..de3822296 100644 --- a/src/shader_recompiler/frontend/translate/translate.cpp +++ b/src/shader_recompiler/frontend/translate/translate.cpp @@ -352,10 +352,10 @@ T Translator::GetSrc(const InstOperand& operand) { } } else { if (operand.input_modifier.abs) { - value = ir.IAbs(value); + value = ir.BitwiseAnd(value, ir.Imm32(0x7FFFFFFFu)); } if (operand.input_modifier.neg) { - value = ir.INeg(value); + value = ir.BitwiseXor(value, ir.Imm32(0x80000000u)); } } return value; @@ -453,6 +453,23 @@ T Translator::GetSrc64(const InstOperand& operand) { if (operand.input_modifier.neg) { value = ir.FPNeg(value); } + } else { + // GCN VOP3 abs/neg modifier bits operate on the sign bit (bit 63 for + // 64-bit values). Unpack, modify the high dword's bit 31, repack. + if (operand.input_modifier.abs) { + const auto unpacked = ir.UnpackUint2x32(value); + const auto lo = IR::U32{ir.CompositeExtract(unpacked, 0)}; + const auto hi = IR::U32{ir.CompositeExtract(unpacked, 1)}; + const auto hi_abs = ir.BitwiseAnd(hi, ir.Imm32(0x7FFFFFFFu)); + value = ir.PackUint2x32(ir.CompositeConstruct(lo, hi_abs)); + } + if (operand.input_modifier.neg) { + const auto unpacked = ir.UnpackUint2x32(value); + const auto lo = IR::U32{ir.CompositeExtract(unpacked, 0)}; + const auto hi = IR::U32{ir.CompositeExtract(unpacked, 1)}; + const auto hi_neg = ir.BitwiseXor(hi, ir.Imm32(0x80000000u)); + value = ir.PackUint2x32(ir.CompositeConstruct(lo, hi_neg)); + } } return value; } diff --git a/src/shader_recompiler/frontend/translate/translate.h b/src/shader_recompiler/frontend/translate/translate.h index 08b0192f5..5ee75e336 100644 --- a/src/shader_recompiler/frontend/translate/translate.h +++ b/src/shader_recompiler/frontend/translate/translate.h @@ -153,6 +153,7 @@ public: void V_SUB_F32(const GcnInst& inst); void V_SUBREV_F32(const GcnInst& inst); void V_MUL_F32(const GcnInst& inst); + void V_MUL_LEGACY_F32(const GcnInst& inst); void V_MUL_I32_I24(const GcnInst& inst, bool is_signed); void V_MIN_F32(const GcnInst& inst, bool is_legacy = false); void V_MAX_F32(const GcnInst& inst, bool is_legacy = false); diff --git a/src/shader_recompiler/frontend/translate/vector_alu.cpp b/src/shader_recompiler/frontend/translate/vector_alu.cpp index 08a0f6527..23236b702 100644 --- a/src/shader_recompiler/frontend/translate/vector_alu.cpp +++ b/src/shader_recompiler/frontend/translate/vector_alu.cpp @@ -25,7 +25,7 @@ void Translator::EmitVectorAlu(const GcnInst& inst) { case Opcode::V_MAC_LEGACY_F32: return V_MAC_F32(inst); case Opcode::V_MUL_LEGACY_F32: - return V_MUL_F32(inst); + return V_MUL_LEGACY_F32(inst); case Opcode::V_MUL_F32: return V_MUL_F32(inst); case Opcode::V_MUL_I32_I24: @@ -493,6 +493,19 @@ void Translator::V_MUL_F32(const GcnInst& inst) { SetDst(inst.dst[0], ir.FPMul(GetSrc(inst.src[0]), GetSrc(inst.src[1]))); } +void Translator::V_MUL_LEGACY_F32(const GcnInst& inst) { + // GCN V_MUL_LEGACY_F32: if either source is zero, the result is +0.0 + // regardless of the other operand (even if NaN or Inf). + // Standard IEEE multiply would produce NaN for 0 * Inf. + const IR::F32 src0{GetSrc(inst.src[0])}; + const IR::F32 src1{GetSrc(inst.src[1])}; + const IR::F32 zero{ir.Imm32(0.0f)}; + const IR::U1 src0_zero{ir.FPEqual(src0, zero)}; + const IR::U1 src1_zero{ir.FPEqual(src1, zero)}; + const IR::U1 either_zero{ir.LogicalOr(src0_zero, src1_zero)}; + SetDst(inst.dst[0], IR::F32{ir.Select(either_zero, zero, ir.FPMul(src0, src1))}); +} + void Translator::V_MUL_I32_I24(const GcnInst& inst, bool is_signed) { const IR::U32 src0{ ir.BitFieldExtract(GetSrc(inst.src[0]), ir.Imm32(0), ir.Imm32(24), is_signed)}; diff --git a/src/shader_recompiler/ir/passes/ir_passes.h b/src/shader_recompiler/ir/passes/ir_passes.h index f103b6736..1b14a1c6b 100644 --- a/src/shader_recompiler/ir/passes/ir_passes.h +++ b/src/shader_recompiler/ir/passes/ir_passes.h @@ -19,7 +19,7 @@ void DeadCodeEliminationPass(IR::Program& program); void ConstantPropagationPass(IR::BlockList& program); void FlattenExtendedUserdataPass(IR::Program& program); void ReadLaneEliminationPass(IR::Program& program); -void ResourceTrackingPass(IR::Program& program); +void ResourceTrackingPass(IR::Program& program, const Profile& profile); void CollectShaderInfoPass(IR::Program& program, const Profile& profile); void LowerBufferFormatToRaw(IR::Program& program); void LowerFp64ToFp32(IR::Program& program); diff --git a/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp b/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp index 4c41e94e9..3b7888ab3 100644 --- a/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp +++ b/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp @@ -9,6 +9,7 @@ #include "shader_recompiler/ir/operand_helper.h" #include "shader_recompiler/ir/program.h" #include "shader_recompiler/ir/reinterpret.h" +#include "shader_recompiler/profile.h" #include "video_core/amdgpu/resource.h" namespace Shader::Optimization { @@ -255,7 +256,9 @@ public: u32 Add(const ImageResource& desc) { const u32 index{Add(image_resources, desc, [&desc](const auto& existing) { - return desc.sharp_idx == existing.sharp_idx && desc.is_array == existing.is_array; + return desc.sharp_idx == existing.sharp_idx && desc.is_array == existing.is_array && + desc.mip_fallback_mode == existing.mip_fallback_mode && + desc.constant_mip_index == existing.constant_mip_index; })}; auto& image = image_resources[index]; image.is_atomic |= desc.is_atomic; @@ -529,14 +532,21 @@ void PatchBufferSharp(IR::Block& block, IR::Inst& inst, Info& info, Descriptors& inst.SetArg(0, ir.Imm32(buffer_binding)); } -void PatchImageSharp(IR::Block& block, IR::Inst& inst, Info& info, Descriptors& descriptors) { +void PatchImageSharp(IR::Block& block, IR::Inst& inst, Info& info, Descriptors& descriptors, + const Profile& profile) { // Read image sharp. const auto inst_info = inst.Flags(); const IR::Inst* image_handle = inst.Arg(0).InstRecursive(); const auto tsharp = TrackSharp(image_handle, block, inst_info.pc); const bool is_atomic = IsImageAtomicInstruction(inst); const bool is_written = inst.GetOpcode() == IR::Opcode::ImageWrite || is_atomic; - const ImageResource image_res = { + const bool is_storage = + inst.GetOpcode() == IR::Opcode::ImageRead || inst.GetOpcode() == IR::Opcode::ImageWrite; + // ImageRead with !is_written gets emitted as OpImageFetch with LOD operand, doesn't + // need fallback (TODO is this 100% true?) + const bool needs_mip_storage_fallback = + inst_info.has_lod && is_written && !profile.supports_image_load_store_lod; + ImageResource image_res = { .sharp_idx = tsharp, .is_depth = bool(inst_info.is_depth), .is_atomic = is_atomic, @@ -544,9 +554,42 @@ void PatchImageSharp(IR::Block& block, IR::Inst& inst, Info& info, Descriptors& .is_written = is_written, .is_r128 = bool(inst_info.is_r128), }; + auto image = image_res.GetSharp(info); ASSERT(image.GetType() != AmdGpu::ImageType::Invalid); + if (needs_mip_storage_fallback) { + // If the mip level to IMAGE_(LOAD/STORE)_MIP is a constant, set up ImageResource + // so that we will only bind a single level. + // If index is dynamic, we will bind levels as an array + const auto view_type = image.GetViewType(image_res.is_array); + + IR::Inst* body = inst.Arg(1).InstRecursive(); + const auto lod_arg = [&] -> IR::Value { + switch (view_type) { + case AmdGpu::ImageType::Color1D: // x, [lod] + return body->Arg(1); + case AmdGpu::ImageType::Color1DArray: // x, slice, [lod] + case AmdGpu::ImageType::Color2D: // x, y, [lod] + return body->Arg(2); + case AmdGpu::ImageType::Color2DArray: // x, y, slice, [lod] + case AmdGpu::ImageType::Color3D: // x, y, z, [lod] + return body->Arg(3); + case AmdGpu::ImageType::Color2DMsaa: + case AmdGpu::ImageType::Color2DMsaaArray: + default: + UNREACHABLE_MSG("Invalid image type {}", view_type); + } + }(); + + if (lod_arg.IsImmediate()) { + image_res.mip_fallback_mode = MipStorageFallbackMode::ConstantIndex; + image_res.constant_mip_index = lod_arg.U32(); + } else { + image_res.mip_fallback_mode = MipStorageFallbackMode::DynamicIndex; + } + } + // Patch image instruction if image is FMask. if (AmdGpu::IsFmask(image.GetDataFmt())) { ASSERT_MSG(!is_written, "FMask storage instructions are not supported"); @@ -1080,7 +1123,11 @@ void PatchImageArgs(IR::Block& block, IR::Inst& inst, Info& info) { const auto has_ms = view_type == AmdGpu::ImageType::Color2DMsaa || view_type == AmdGpu::ImageType::Color2DMsaaArray; ASSERT(!inst_info.has_lod || !has_ms); - const auto lod = inst_info.has_lod ? IR::U32{arg} : IR::U32{}; + // If we are binding a single mip level as fallback, drop the argument + const auto lod = + (inst_info.has_lod && image_res.mip_fallback_mode != MipStorageFallbackMode::ConstantIndex) + ? IR::U32{arg} + : IR::U32{}; const auto ms = has_ms ? IR::U32{arg} : IR::U32{}; const auto is_storage = image_res.is_written; @@ -1111,7 +1158,7 @@ void PatchImageArgs(IR::Block& block, IR::Inst& inst, Info& info) { } } -void ResourceTrackingPass(IR::Program& program) { +void ResourceTrackingPass(IR::Program& program, const Profile& profile) { // Iterate resource instructions and patch them after finding the sharp. auto& info = program.info; @@ -1122,7 +1169,7 @@ void ResourceTrackingPass(IR::Program& program) { if (IsBufferInstruction(inst)) { PatchBufferSharp(*block, inst, info, descriptors); } else if (IsImageInstruction(inst)) { - PatchImageSharp(*block, inst, info, descriptors); + PatchImageSharp(*block, inst, info, descriptors, profile); } } } diff --git a/src/shader_recompiler/recompiler.cpp b/src/shader_recompiler/recompiler.cpp index f4fa45afc..d6efa2890 100644 --- a/src/shader_recompiler/recompiler.cpp +++ b/src/shader_recompiler/recompiler.cpp @@ -80,7 +80,7 @@ IR::Program TranslateProgram(const std::span& code, Pools& pools, Inf Shader::Optimization::RingAccessElimination(program, runtime_info); Shader::Optimization::ReadLaneEliminationPass(program); Shader::Optimization::FlattenExtendedUserdataPass(program); - Shader::Optimization::ResourceTrackingPass(program); + Shader::Optimization::ResourceTrackingPass(program, profile); Shader::Optimization::LowerBufferFormatToRaw(program); Shader::Optimization::SharedMemorySimplifyPass(program, profile); Shader::Optimization::SharedMemoryToStoragePass(program, runtime_info, profile); diff --git a/src/shader_recompiler/resource.h b/src/shader_recompiler/resource.h index 5ae3179f6..82a861e2a 100644 --- a/src/shader_recompiler/resource.h +++ b/src/shader_recompiler/resource.h @@ -71,6 +71,8 @@ struct BufferResource { }; using BufferResourceList = boost::container::static_vector; +enum class MipStorageFallbackMode : u32 { None, DynamicIndex, ConstantIndex }; + struct ImageResource { u32 sharp_idx; bool is_depth{}; @@ -78,6 +80,8 @@ struct ImageResource { bool is_array{}; bool is_written{}; bool is_r128{}; + MipStorageFallbackMode mip_fallback_mode{}; + u32 constant_mip_index{}; constexpr AmdGpu::Image GetSharp(const auto& info) const noexcept { AmdGpu::Image image{}; @@ -102,6 +106,13 @@ struct ImageResource { } return image; } + + u32 NumBindings(const auto& info) const { + const AmdGpu::Image tsharp = GetSharp(info); + return (mip_fallback_mode == MipStorageFallbackMode::DynamicIndex) + ? (tsharp.last_level - tsharp.base_level + 1) + : 1; + } }; using ImageResourceList = boost::container::static_vector; diff --git a/src/shader_recompiler/specialization.h b/src/shader_recompiler/specialization.h index 4f6bb44bf..fa14583af 100644 --- a/src/shader_recompiler/specialization.h +++ b/src/shader_recompiler/specialization.h @@ -52,6 +52,8 @@ struct ImageSpecialization { bool is_srgb = false; AmdGpu::CompMapping dst_select{}; AmdGpu::NumberConversion num_conversion{}; + // FIXME any pipeline cache changes needed? + u32 num_bindings = 0; bool operator==(const ImageSpecialization&) const = default; }; @@ -133,7 +135,7 @@ struct StageSpecialization { } }); ForEachSharp(binding, images, info->images, - [](auto& spec, const auto& desc, AmdGpu::Image sharp) { + [&](auto& spec, const auto& desc, AmdGpu::Image sharp) { spec.type = sharp.GetViewType(desc.is_array); spec.is_integer = AmdGpu::IsInteger(sharp.GetNumberFmt()); spec.is_storage = desc.is_written; @@ -144,6 +146,7 @@ struct StageSpecialization { spec.is_srgb = sharp.GetNumberFmt() == AmdGpu::NumberFormat::Srgb; } spec.num_conversion = sharp.GetNumberConversion(); + spec.num_bindings = desc.NumBindings(*info); }); ForEachSharp(binding, fmasks, info->fmasks, [](auto& spec, const auto& desc, AmdGpu::Image sharp) { diff --git a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp index eecd416d1..ba0a3afa2 100644 --- a/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_compute_pipeline.cpp @@ -48,13 +48,15 @@ ComputePipeline::ComputePipeline(const Instance& instance, Scheduler& scheduler, }); } for (const auto& image : info->images) { + const u32 num_bindings = image.NumBindings(*info); bindings.push_back({ - .binding = binding++, + .binding = binding, .descriptorType = image.is_written ? vk::DescriptorType::eStorageImage : vk::DescriptorType::eSampledImage, - .descriptorCount = 1, + .descriptorCount = num_bindings, .stageFlags = vk::ShaderStageFlagBits::eCompute, }); + binding += num_bindings; } for (const auto& sampler : info->samplers) { bindings.push_back({ diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp index 242c9b6f2..bc9ef571b 100644 --- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp +++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp @@ -457,13 +457,15 @@ void GraphicsPipeline::BuildDescSetLayout(bool preloading) { }); } for (const auto& image : stage->images) { + const u32 num_bindings = image.NumBindings(*stage); bindings.push_back({ - .binding = binding++, + .binding = binding, .descriptorType = image.is_written ? vk::DescriptorType::eStorageImage : vk::DescriptorType::eSampledImage, - .descriptorCount = 1, + .descriptorCount = num_bindings, .stageFlags = stage_bit, }); + binding += num_bindings; } for (const auto& sampler : stage->samplers) { bindings.push_back({ diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 41dbcc7d4..7f0bd43e9 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -662,6 +662,13 @@ void Rasterizer::BindBuffers(const Shader::Info& stage, Shader::Backend::Binding void Rasterizer::BindTextures(const Shader::Info& stage, Shader::Backend::Bindings& binding) { image_bindings.clear(); + const u32 first_image_idx = image_infos.size(); + // For loading/storing to explicit mip levels, when no native instruction support, bind an array + // of descriptors consecutively, 1 for each mip level. The shader can index this with LOD + // operand. + // This array holds the size of each consecutive array with the number of bindings consumed. + // This is currently always 1 for anything other than mip fallback arrays. + boost::container::small_vector image_descriptor_array_sizes; for (const auto& image_desc : stage.images) { const auto tsharp = image_desc.GetSharp(stage); @@ -671,25 +678,43 @@ void Rasterizer::BindTextures(const Shader::Info& stage, Shader::Backend::Bindin if (tsharp.GetDataFmt() == AmdGpu::DataFormat::FormatInvalid) { image_bindings.emplace_back(std::piecewise_construct, std::tuple{}, std::tuple{}); + image_descriptor_array_sizes.push_back(1); continue; } - auto& [image_id, desc] = image_bindings.emplace_back(std::piecewise_construct, std::tuple{}, - std::tuple{tsharp, image_desc}); - image_id = texture_cache.FindImage(desc); - auto* image = &texture_cache.GetImage(image_id); - if (image->depth_id) { - // If this image has an associated depth image, it's a stencil attachment. - // Redirect the access to the actual depth-stencil buffer. - image_id = image->depth_id; - image = &texture_cache.GetImage(image_id); + const Shader::MipStorageFallbackMode mip_fallback_mode = image_desc.mip_fallback_mode; + const u32 num_bindings = image_desc.NumBindings(stage); + + for (auto i = 0; i < num_bindings; i++) { + auto& [image_id, desc] = image_bindings.emplace_back( + std::piecewise_construct, std::tuple{}, std::tuple{tsharp, image_desc}); + + if (mip_fallback_mode == Shader::MipStorageFallbackMode::ConstantIndex) { + ASSERT(num_bindings == 1); + desc.view_info.range.base.level += image_desc.constant_mip_index; + desc.view_info.range.extent.levels = 1; + } else if (mip_fallback_mode == Shader::MipStorageFallbackMode::DynamicIndex) { + desc.view_info.range.base.level += i; + desc.view_info.range.extent.levels = 1; + } + + image_id = texture_cache.FindImage(desc); + auto* image = &texture_cache.GetImage(image_id); + if (image->depth_id) { + // If this image has an associated depth image, it's a stencil attachment. + // Redirect the access to the actual depth-stencil buffer. + image_id = image->depth_id; + image = &texture_cache.GetImage(image_id); + } + if (image->binding.is_bound) { + // The image is already bound. In case if it is about to be used as storage we + // need to force general layout on it. + image->binding.force_general |= image_desc.is_written; + } + image->binding.is_bound = 1u; } - if (image->binding.is_bound) { - // The image is already bound. In case if it is about to be used as storage we need - // to force general layout on it. - image->binding.force_general |= image_desc.is_written; - } - image->binding.is_bound = 1u; + + image_descriptor_array_sizes.push_back(num_bindings); } // Second pass to re-bind images that were updated after binding @@ -749,16 +774,26 @@ void Rasterizer::BindTextures(const Shader::Info& stage, Shader::Backend::Bindin image_infos.emplace_back(VK_NULL_HANDLE, *image_view.image_view, image.backing->state.layout); } + } + u32 image_info_idx = first_image_idx; + u32 image_binding_idx = 0; + for (u32 array_size : image_descriptor_array_sizes) { + const auto& [_, desc] = image_bindings[image_binding_idx]; + const bool is_storage = desc.type == VideoCore::TextureCache::BindingType::Storage; set_writes.push_back({ .dstSet = VK_NULL_HANDLE, - .dstBinding = binding.unified++, + .dstBinding = binding.unified, .dstArrayElement = 0, - .descriptorCount = 1, + .descriptorCount = array_size, .descriptorType = is_storage ? vk::DescriptorType::eStorageImage : vk::DescriptorType::eSampledImage, - .pImageInfo = &image_infos.back(), + .pImageInfo = &image_infos[image_info_idx], }); + + image_info_idx += array_size; + image_binding_idx += array_size; + binding.unified += array_size; } for (const auto& sampler : stage.samplers) {