diff options
Diffstat (limited to 'src/shader_recompiler/backend')
18 files changed, 205 insertions, 109 deletions
diff --git a/src/shader_recompiler/backend/glasm/emit_glasm.cpp b/src/shader_recompiler/backend/glasm/emit_glasm.cpp index 0cb1e193e..fd4a61a4d 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm.cpp @@ -279,6 +279,8 @@ void SetupOptions(const IR::Program& program, const Profile& profile, header += "OPTION NV_internal;" "OPTION NV_shader_storage_buffer;" "OPTION NV_gpu_program_fp64;"; + // TODO: Enable only when MS is used + header += "OPTION NV_texture_multisample;"; if (info.uses_int64_bit_atomics) { header += "OPTION NV_shader_atomic_int64;"; } diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_bitwise_conversion.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_bitwise_conversion.cpp index 5bfdecc09..2fc2a0ac6 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_bitwise_conversion.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm_bitwise_conversion.cpp @@ -43,10 +43,6 @@ void EmitBitCastU64F64(EmitContext&, IR::Inst& inst, const IR::Value& value) { Alias(inst, value); } -void EmitBitCastS32F32(EmitContext&, IR::Inst& inst, const IR::Value& value) { - Alias(inst, value); -} - void EmitBitCastF16U16(EmitContext&, IR::Inst& inst, const IR::Value& value) { Alias(inst, value); } diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp index f0bd84ab2..c7d7d5fef 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp @@ -137,6 +137,15 @@ void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, Scal case IR::Attribute::VertexId: ctx.Add("MOV.F {}.x,{}.id;", inst, ctx.attrib_name); break; + case IR::Attribute::BaseInstance: + ctx.Add("MOV.F {}.x,{}.baseInstance;", inst, ctx.attrib_name); + break; + case IR::Attribute::BaseVertex: + ctx.Add("MOV.F {}.x,{}.baseVertex;", inst, ctx.attrib_name); + break; + case IR::Attribute::DrawID: + ctx.Add("MOV.F {}.x,{}.draw.id;", inst, ctx.attrib_name); + break; case IR::Attribute::FrontFace: ctx.Add("CMP.F {}.x,{}.facing.x,0,-1;", inst, ctx.attrib_name); break; @@ -156,6 +165,15 @@ void EmitGetAttributeU32(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, S case IR::Attribute::VertexId: ctx.Add("MOV.S {}.x,{}.id;", inst, ctx.attrib_name); break; + case IR::Attribute::BaseInstance: + ctx.Add("MOV.S {}.x,{}.baseInstance;", inst, ctx.attrib_name); + break; + case IR::Attribute::BaseVertex: + ctx.Add("MOV.S {}.x,{}.baseVertex;", inst, ctx.attrib_name); + break; + case IR::Attribute::DrawID: + ctx.Add("MOV.S {}.x,{}.draw.id;", inst, ctx.attrib_name); + break; default: throw NotImplementedException("Get U32 attribute {}", attr); } diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp index e67e80fac..b7bc11416 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp +++ b/src/shader_recompiler/backend/glasm/emit_glasm_image.cpp @@ -59,7 +59,7 @@ std::string Image(EmitContext& ctx, IR::TextureInstInfo info, } } -std::string_view TextureType(IR::TextureInstInfo info) { +std::string_view TextureType(IR::TextureInstInfo info, bool is_ms = false) { if (info.is_depth) { switch (info.type) { case TextureType::Color1D: @@ -88,9 +88,9 @@ std::string_view TextureType(IR::TextureInstInfo info) { return "ARRAY1D"; case TextureType::Color2D: case TextureType::Color2DRect: - return "2D"; + return is_ms ? "2DMS" : "2D"; case TextureType::ColorArray2D: - return "ARRAY2D"; + return is_ms ? "ARRAY2DMS" : "ARRAY2D"; case TextureType::Color3D: return "3D"; case TextureType::ColorCube: @@ -510,15 +510,16 @@ void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, const IR::Value& coord, const IR::Value& offset, ScalarS32 lod, ScalarS32 ms) { const auto info{inst.Flags<IR::TextureInstInfo>()}; const auto sparse_inst{PrepareSparse(inst)}; + const bool is_multisample{ms.type != Type::Void}; const std::string_view sparse_mod{sparse_inst ? ".SPARSE" : ""}; - const std::string_view type{TextureType(info)}; + const std::string_view type{TextureType(info, is_multisample)}; const std::string texture{Texture(ctx, info, index)}; const std::string offset_vec{Offset(ctx, offset)}; const auto [coord_vec, coord_alloc]{Coord(ctx, coord)}; const Register ret{ctx.reg_alloc.Define(inst)}; if (info.type == TextureType::Buffer) { ctx.Add("TXF.F{} {},{},{},{}{};", sparse_mod, ret, coord_vec, texture, type, offset_vec); - } else if (ms.type != Type::Void) { + } else if (is_multisample) { ctx.Add("MOV.S {}.w,{};" "TXFMS.F{} {},{},{},{}{};", coord_vec, ms, sparse_mod, ret, coord_vec, texture, type, offset_vec); @@ -531,7 +532,7 @@ void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, } void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, - ScalarS32 lod) { + ScalarS32 lod, [[maybe_unused]] const IR::Value& skip_mips) { const auto info{inst.Flags<IR::TextureInstInfo>()}; const std::string texture{Texture(ctx, info, index)}; const std::string_view type{TextureType(info)}; diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_instructions.h b/src/shader_recompiler/backend/glasm/emit_glasm_instructions.h index eaaf9ba39..1a1ea61d5 100644 --- a/src/shader_recompiler/backend/glasm/emit_glasm_instructions.h +++ b/src/shader_recompiler/backend/glasm/emit_glasm_instructions.h @@ -197,7 +197,6 @@ void EmitSelectF64(EmitContext& ctx, ScalarS32 cond, Register true_value, Regist void EmitBitCastU16F16(EmitContext& ctx, IR::Inst& inst, const IR::Value& value); void EmitBitCastU32F32(EmitContext& ctx, IR::Inst& inst, const IR::Value& value); void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, const IR::Value& value); -void EmitBitCastS32F32(EmitContext& ctx, IR::Inst& inst, const IR::Value& value); void EmitBitCastF16U16(EmitContext& ctx, IR::Inst& inst, const IR::Value& value); void EmitBitCastF32U32(EmitContext& ctx, IR::Inst& inst, const IR::Value& value); void EmitBitCastF64U64(EmitContext& ctx, IR::Inst& inst, const IR::Value& value); @@ -582,7 +581,7 @@ void EmitImageGatherDref(EmitContext& ctx, IR::Inst& inst, const IR::Value& inde void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, const IR::Value& coord, const IR::Value& offset, ScalarS32 lod, ScalarS32 ms); void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, - ScalarS32 lod); + ScalarS32 lod, const IR::Value& skip_mips); void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, Register coord); void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, const IR::Value& coord, const IR::Value& derivatives, diff --git a/src/shader_recompiler/backend/glsl/emit_glsl.cpp b/src/shader_recompiler/backend/glsl/emit_glsl.cpp index e8a4390f6..d91e04446 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl.cpp @@ -219,7 +219,7 @@ std::string EmitGLSL(const Profile& profile, const RuntimeInfo& runtime_info, IR EmitContext ctx{program, bindings, profile, runtime_info}; Precolor(program); EmitCode(ctx, program); - const std::string version{fmt::format("#version 450{}\n", GlslVersionSpecifier(ctx))}; + const std::string version{fmt::format("#version 460{}\n", GlslVersionSpecifier(ctx))}; ctx.header.insert(0, version); if (program.shared_memory_size > 0) { const auto requested_size{program.shared_memory_size}; diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_bitwise_conversion.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_bitwise_conversion.cpp index 8e5e6cf1f..1be4a0f59 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_bitwise_conversion.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl_bitwise_conversion.cpp @@ -48,10 +48,6 @@ void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value) ctx.AddU64("{}=doubleBitsToUint64({});", inst, value); } -void EmitBitCastS32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value) { - ctx.AddF32("{}=ftoi({});", inst, value); -} - void EmitBitCastF16U16([[maybe_unused]] EmitContext& ctx, [[maybe_unused]] IR::Inst& inst) { NotImplemented(); } diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp index 39579cf5d..2e369ed72 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl_context_get_set.cpp @@ -234,6 +234,15 @@ void EmitGetAttribute(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, case IR::Attribute::FrontFace: ctx.AddF32("{}=itof(gl_FrontFacing?-1:0);", inst); break; + case IR::Attribute::BaseInstance: + ctx.AddF32("{}=itof(gl_BaseInstance);", inst); + break; + case IR::Attribute::BaseVertex: + ctx.AddF32("{}=itof(gl_BaseVertex);", inst); + break; + case IR::Attribute::DrawID: + ctx.AddF32("{}=itof(gl_DrawID);", inst); + break; default: throw NotImplementedException("Get attribute {}", attr); } @@ -250,6 +259,15 @@ void EmitGetAttributeU32(EmitContext& ctx, IR::Inst& inst, IR::Attribute attr, s case IR::Attribute::VertexId: ctx.AddU32("{}=uint(gl_VertexID);", inst); break; + case IR::Attribute::BaseInstance: + ctx.AddU32("{}=uint(gl_BaseInstance);", inst); + break; + case IR::Attribute::BaseVertex: + ctx.AddU32("{}=uint(gl_BaseVertex);", inst); + break; + case IR::Attribute::DrawID: + ctx.AddU32("{}=uint(gl_DrawID);", inst); + break; default: throw NotImplementedException("Get U32 attribute {}", attr); } diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp b/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp index cecdbb9d6..4be2c25ec 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp +++ b/src/shader_recompiler/backend/glsl/emit_glsl_image.cpp @@ -414,7 +414,7 @@ void EmitImageGatherDref(EmitContext& ctx, IR::Inst& inst, const IR::Value& inde void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, std::string_view coords, std::string_view offset, std::string_view lod, - [[maybe_unused]] std::string_view ms) { + std::string_view ms) { const auto info{inst.Flags<IR::TextureInstInfo>()}; if (info.has_bias) { throw NotImplementedException("EmitImageFetch Bias texture samples"); @@ -431,19 +431,24 @@ void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, ctx.AddU1("{}=true;", *sparse_inst); } if (!sparse_inst || !supports_sparse) { - if (!offset.empty()) { - ctx.Add("{}=texelFetchOffset({},{},int({}),{});", texel, texture, - CoordsCastToInt(coords, info), lod, CoordsCastToInt(offset, info)); + const auto int_coords{CoordsCastToInt(coords, info)}; + if (!ms.empty()) { + ctx.Add("{}=texelFetch({},{},int({}));", texel, texture, int_coords, ms); + } else if (!offset.empty()) { + ctx.Add("{}=texelFetchOffset({},{},int({}),{});", texel, texture, int_coords, lod, + CoordsCastToInt(offset, info)); } else { if (info.type == TextureType::Buffer) { ctx.Add("{}=texelFetch({},int({}));", texel, texture, coords); } else { - ctx.Add("{}=texelFetch({},{},int({}));", texel, texture, - CoordsCastToInt(coords, info), lod); + ctx.Add("{}=texelFetch({},{},int({}));", texel, texture, int_coords, lod); } } return; } + if (!ms.empty()) { + throw NotImplementedException("EmitImageFetch Sparse MSAA samples"); + } if (!offset.empty()) { ctx.AddU1("{}=sparseTexelsResidentARB(sparseTexelFetchOffsetARB({},{},int({}),{},{}));", *sparse_inst, texture, CastToIntVec(coords, info), lod, @@ -455,27 +460,27 @@ void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, } void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, - std::string_view lod) { + std::string_view lod, const IR::Value& skip_mips_val) { const auto info{inst.Flags<IR::TextureInstInfo>()}; const auto texture{Texture(ctx, info, index)}; + const bool skip_mips{skip_mips_val.U1()}; + const auto mips{ + [&] { return skip_mips ? "0u" : fmt::format("uint(textureQueryLevels({}))", texture); }}; switch (info.type) { case TextureType::Color1D: - return ctx.AddU32x4( - "{}=uvec4(uint(textureSize({},int({}))),0u,0u,uint(textureQueryLevels({})));", inst, - texture, lod, texture); + return ctx.AddU32x4("{}=uvec4(uint(textureSize({},int({}))),0u,0u,{});", inst, texture, lod, + mips()); case TextureType::ColorArray1D: case TextureType::Color2D: case TextureType::ColorCube: case TextureType::Color2DRect: - return ctx.AddU32x4( - "{}=uvec4(uvec2(textureSize({},int({}))),0u,uint(textureQueryLevels({})));", inst, - texture, lod, texture); + return ctx.AddU32x4("{}=uvec4(uvec2(textureSize({},int({}))),0u,{});", inst, texture, lod, + mips()); case TextureType::ColorArray2D: case TextureType::Color3D: case TextureType::ColorArrayCube: - return ctx.AddU32x4( - "{}=uvec4(uvec3(textureSize({},int({}))),uint(textureQueryLevels({})));", inst, texture, - lod, texture); + return ctx.AddU32x4("{}=uvec4(uvec3(textureSize({},int({}))),{});", inst, texture, lod, + mips()); case TextureType::Buffer: throw NotImplementedException("EmitImageQueryDimensions Texture buffers"); } diff --git a/src/shader_recompiler/backend/glsl/emit_glsl_instructions.h b/src/shader_recompiler/backend/glsl/emit_glsl_instructions.h index 4151c89de..8d0a65047 100644 --- a/src/shader_recompiler/backend/glsl/emit_glsl_instructions.h +++ b/src/shader_recompiler/backend/glsl/emit_glsl_instructions.h @@ -231,7 +231,6 @@ void EmitSelectF64(EmitContext& ctx, IR::Inst& inst, std::string_view cond, void EmitBitCastU16F16(EmitContext& ctx, IR::Inst& inst); void EmitBitCastU32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value); void EmitBitCastU64F64(EmitContext& ctx, IR::Inst& inst, std::string_view value); -void EmitBitCastS32F32(EmitContext& ctx, IR::Inst& inst, std::string_view value); void EmitBitCastF16U16(EmitContext& ctx, IR::Inst& inst); void EmitBitCastF32U32(EmitContext& ctx, IR::Inst& inst, std::string_view value); void EmitBitCastF64U64(EmitContext& ctx, IR::Inst& inst, std::string_view value); @@ -655,7 +654,7 @@ void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, std::string_view coords, std::string_view offset, std::string_view lod, std::string_view ms); void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, - std::string_view lod); + std::string_view lod, const IR::Value& skip_mips); void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, std::string_view coords); void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, diff --git a/src/shader_recompiler/backend/glsl/glsl_emit_context.cpp b/src/shader_recompiler/backend/glsl/glsl_emit_context.cpp index 5d01ec0cd..1b006e811 100644 --- a/src/shader_recompiler/backend/glsl/glsl_emit_context.cpp +++ b/src/shader_recompiler/backend/glsl/glsl_emit_context.cpp @@ -61,24 +61,28 @@ std::string OutputDecorator(Stage stage, u32 size) { } } -std::string_view SamplerType(TextureType type, bool is_depth) { - if (is_depth) { - switch (type) { - case TextureType::Color1D: - return "sampler1DShadow"; - case TextureType::ColorArray1D: - return "sampler1DArrayShadow"; - case TextureType::Color2D: - return "sampler2DShadow"; - case TextureType::ColorArray2D: - return "sampler2DArrayShadow"; - case TextureType::ColorCube: - return "samplerCubeShadow"; - case TextureType::ColorArrayCube: - return "samplerCubeArrayShadow"; - default: - throw NotImplementedException("Texture type: {}", type); - } +std::string_view DepthSamplerType(TextureType type) { + switch (type) { + case TextureType::Color1D: + return "sampler1DShadow"; + case TextureType::ColorArray1D: + return "sampler1DArrayShadow"; + case TextureType::Color2D: + return "sampler2DShadow"; + case TextureType::ColorArray2D: + return "sampler2DArrayShadow"; + case TextureType::ColorCube: + return "samplerCubeShadow"; + case TextureType::ColorArrayCube: + return "samplerCubeArrayShadow"; + default: + throw NotImplementedException("Texture type: {}", type); + } +} + +std::string_view ColorSamplerType(TextureType type, bool is_multisample = false) { + if (is_multisample) { + ASSERT(type == TextureType::Color2D || type == TextureType::ColorArray2D); } switch (type) { case TextureType::Color1D: @@ -87,9 +91,9 @@ std::string_view SamplerType(TextureType type, bool is_depth) { return "sampler1DArray"; case TextureType::Color2D: case TextureType::Color2DRect: - return "sampler2D"; + return is_multisample ? "sampler2DMS" : "sampler2D"; case TextureType::ColorArray2D: - return "sampler2DArray"; + return is_multisample ? "sampler2DMSArray" : "sampler2DArray"; case TextureType::Color3D: return "sampler3D"; case TextureType::ColorCube: @@ -677,7 +681,7 @@ void EmitContext::SetupTextures(Bindings& bindings) { texture_buffers.reserve(info.texture_buffer_descriptors.size()); for (const auto& desc : info.texture_buffer_descriptors) { texture_buffers.push_back({bindings.texture, desc.count}); - const auto sampler_type{SamplerType(TextureType::Buffer, false)}; + const auto sampler_type{ColorSamplerType(TextureType::Buffer)}; const auto array_decorator{desc.count > 1 ? fmt::format("[{}]", desc.count) : ""}; header += fmt::format("layout(binding={}) uniform {} tex{}{};", bindings.texture, sampler_type, bindings.texture, array_decorator); @@ -686,7 +690,8 @@ void EmitContext::SetupTextures(Bindings& bindings) { textures.reserve(info.texture_descriptors.size()); for (const auto& desc : info.texture_descriptors) { textures.push_back({bindings.texture, desc.count}); - const auto sampler_type{SamplerType(desc.type, desc.is_depth)}; + const auto sampler_type{desc.is_depth ? DepthSamplerType(desc.type) + : ColorSamplerType(desc.type, desc.is_multisample)}; const auto array_decorator{desc.count > 1 ? fmt::format("[{}]", desc.count) : ""}; header += fmt::format("layout(binding={}) uniform {} tex{}{};", bindings.texture, sampler_type, bindings.texture, array_decorator); diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_bitwise_conversion.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_bitwise_conversion.cpp index 50daacd95..c4ca28d11 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_bitwise_conversion.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_bitwise_conversion.cpp @@ -18,10 +18,6 @@ void EmitBitCastU64F64(EmitContext&) { throw NotImplementedException("SPIR-V Instruction"); } -void EmitBitCastS32F32(EmitContext&) { - throw NotImplementedException("SPIR-V Instruction"); -} - void EmitBitCastF16U16(EmitContext&) { throw NotImplementedException("SPIR-V Instruction"); } diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp index 73b67f0af..0cd87a48f 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp @@ -321,8 +321,12 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) { case IR::Attribute::PositionY: case IR::Attribute::PositionZ: case IR::Attribute::PositionW: - return ctx.OpLoad(ctx.F32[1], AttrPointer(ctx, ctx.input_f32, vertex, ctx.input_position, - ctx.Const(element))); + return ctx.OpLoad( + ctx.F32[1], + ctx.need_input_position_indirect + ? AttrPointer(ctx, ctx.input_f32, vertex, ctx.input_position, ctx.u32_zero_value, + ctx.Const(element)) + : AttrPointer(ctx, ctx.input_f32, vertex, ctx.input_position, ctx.Const(element))); case IR::Attribute::InstanceId: if (ctx.profile.support_vertex_instance_id) { return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.instance_id)); @@ -339,6 +343,12 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) { const Id base{ctx.OpLoad(ctx.U32[1], ctx.base_vertex)}; return ctx.OpBitcast(ctx.F32[1], ctx.OpISub(ctx.U32[1], index, base)); } + case IR::Attribute::BaseInstance: + return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.base_instance)); + case IR::Attribute::BaseVertex: + return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.base_vertex)); + case IR::Attribute::DrawID: + return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.draw_index)); case IR::Attribute::FrontFace: return ctx.OpSelect(ctx.F32[1], ctx.OpLoad(ctx.U1, ctx.front_face), ctx.OpBitcast(ctx.F32[1], ctx.Const(std::numeric_limits<u32>::max())), @@ -380,6 +390,12 @@ Id EmitGetAttributeU32(EmitContext& ctx, IR::Attribute attr, Id) { const Id base{ctx.OpLoad(ctx.U32[1], ctx.base_vertex)}; return ctx.OpISub(ctx.U32[1], index, base); } + case IR::Attribute::BaseInstance: + return ctx.OpLoad(ctx.U32[1], ctx.base_instance); + case IR::Attribute::BaseVertex: + return ctx.OpLoad(ctx.U32[1], ctx.base_vertex); + case IR::Attribute::DrawID: + return ctx.OpLoad(ctx.U32[1], ctx.draw_index); default: throw NotImplementedException("Read U32 attribute {}", attr); } diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp index fb5799c42..3b969d915 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp @@ -436,16 +436,22 @@ Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id c if (info.type == TextureType::Buffer) { lod = Id{}; } + if (Sirit::ValidId(ms)) { + // This image is multisampled, lod must be implicit + lod = Id{}; + } const ImageOperands operands(offset, lod, ms); return Emit(&EmitContext::OpImageSparseFetch, &EmitContext::OpImageFetch, ctx, inst, ctx.F32[4], TextureImage(ctx, info, index), coords, operands.MaskOptional(), operands.Span()); } -Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod) { +Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod, + const IR::Value& skip_mips_val) { const auto info{inst->Flags<IR::TextureInstInfo>()}; const Id image{TextureImage(ctx, info, index)}; const Id zero{ctx.u32_zero_value}; - const auto mips{[&] { return ctx.OpImageQueryLevels(ctx.U32[1], image); }}; + const bool skip_mips{skip_mips_val.U1()}; + const auto mips{[&] { return skip_mips ? zero : ctx.OpImageQueryLevels(ctx.U32[1], image); }}; switch (info.type) { case TextureType::Color1D: return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[1], image, lod), diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h b/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h index e31cdc5e8..a440b557d 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h +++ b/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h @@ -179,7 +179,6 @@ Id EmitSelectF64(EmitContext& ctx, Id cond, Id true_value, Id false_value); void EmitBitCastU16F16(EmitContext& ctx); Id EmitBitCastU32F32(EmitContext& ctx, Id value); void EmitBitCastU64F64(EmitContext& ctx); -void EmitBitCastS32F32(EmitContext& ctx); void EmitBitCastF16U16(EmitContext&); Id EmitBitCastF32U32(EmitContext& ctx, Id value); void EmitBitCastF64U64(EmitContext& ctx); @@ -540,7 +539,8 @@ Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, const IR::Value& offset, const IR::Value& offset2, Id dref); Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, Id offset, Id lod, Id ms); -Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod); +Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod, + const IR::Value& skip_mips); Id EmitImageQueryLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords); Id EmitImageGradient(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, Id derivates, Id offset, Id lod_clamp); diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp index 2c90f2368..c5db19d09 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp @@ -58,11 +58,10 @@ Id SelectValue(EmitContext& ctx, Id in_range, Id value, Id src_thread_id) { ctx.OpGroupNonUniformShuffle(ctx.U32[1], SubgroupScope(ctx), value, src_thread_id), value); } -Id GetUpperClamp(EmitContext& ctx, Id invocation_id, Id clamp) { - const Id thirty_two{ctx.Const(32u)}; - const Id is_upper_partition{ctx.OpSGreaterThanEqual(ctx.U1, invocation_id, thirty_two)}; - const Id upper_clamp{ctx.OpIAdd(ctx.U32[1], thirty_two, clamp)}; - return ctx.OpSelect(ctx.U32[1], is_upper_partition, upper_clamp, clamp); +Id AddPartitionBase(EmitContext& ctx, Id thread_id) { + const Id partition_idx{ctx.OpShiftRightLogical(ctx.U32[1], GetThreadId(ctx), ctx.Const(5u))}; + const Id partition_base{ctx.OpShiftLeftLogical(ctx.U32[1], partition_idx, ctx.Const(5u))}; + return ctx.OpIAdd(ctx.U32[1], thread_id, partition_base); } } // Anonymous namespace @@ -145,64 +144,63 @@ Id EmitSubgroupGeMask(EmitContext& ctx) { Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { const Id not_seg_mask{ctx.OpNot(ctx.U32[1], segmentation_mask)}; - const Id thread_id{GetThreadId(ctx)}; - if (ctx.profile.warp_size_potentially_larger_than_guest) { - const Id thirty_two{ctx.Const(32u)}; - const Id is_upper_partition{ctx.OpSGreaterThanEqual(ctx.U1, thread_id, thirty_two)}; - const Id upper_index{ctx.OpIAdd(ctx.U32[1], thirty_two, index)}; - const Id upper_clamp{ctx.OpIAdd(ctx.U32[1], thirty_two, clamp)}; - index = ctx.OpSelect(ctx.U32[1], is_upper_partition, upper_index, index); - clamp = ctx.OpSelect(ctx.U32[1], is_upper_partition, upper_clamp, clamp); - } + const Id thread_id{EmitLaneId(ctx)}; const Id min_thread_id{ComputeMinThreadId(ctx, thread_id, segmentation_mask)}; const Id max_thread_id{ComputeMaxThreadId(ctx, min_thread_id, clamp, not_seg_mask)}; const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], index, not_seg_mask)}; - const Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], lhs, min_thread_id)}; + Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], lhs, min_thread_id)}; const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)}; + if (ctx.profile.warp_size_potentially_larger_than_guest) { + src_thread_id = AddPartitionBase(ctx, src_thread_id); + } + SetInBoundsFlag(inst, in_range); return SelectValue(ctx, in_range, value, src_thread_id); } Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { - const Id thread_id{GetThreadId(ctx)}; - if (ctx.profile.warp_size_potentially_larger_than_guest) { - clamp = GetUpperClamp(ctx, thread_id, clamp); - } + const Id thread_id{EmitLaneId(ctx)}; const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; - const Id src_thread_id{ctx.OpISub(ctx.U32[1], thread_id, index)}; + Id src_thread_id{ctx.OpISub(ctx.U32[1], thread_id, index)}; const Id in_range{ctx.OpSGreaterThanEqual(ctx.U1, src_thread_id, max_thread_id)}; + if (ctx.profile.warp_size_potentially_larger_than_guest) { + src_thread_id = AddPartitionBase(ctx, src_thread_id); + } + SetInBoundsFlag(inst, in_range); return SelectValue(ctx, in_range, value, src_thread_id); } Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { - const Id thread_id{GetThreadId(ctx)}; - if (ctx.profile.warp_size_potentially_larger_than_guest) { - clamp = GetUpperClamp(ctx, thread_id, clamp); - } + const Id thread_id{EmitLaneId(ctx)}; const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; - const Id src_thread_id{ctx.OpIAdd(ctx.U32[1], thread_id, index)}; + Id src_thread_id{ctx.OpIAdd(ctx.U32[1], thread_id, index)}; const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)}; + if (ctx.profile.warp_size_potentially_larger_than_guest) { + src_thread_id = AddPartitionBase(ctx, src_thread_id); + } + SetInBoundsFlag(inst, in_range); return SelectValue(ctx, in_range, value, src_thread_id); } Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp, Id segmentation_mask) { - const Id thread_id{GetThreadId(ctx)}; - if (ctx.profile.warp_size_potentially_larger_than_guest) { - clamp = GetUpperClamp(ctx, thread_id, clamp); - } + const Id thread_id{EmitLaneId(ctx)}; const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)}; - const Id src_thread_id{ctx.OpBitwiseXor(ctx.U32[1], thread_id, index)}; + Id src_thread_id{ctx.OpBitwiseXor(ctx.U32[1], thread_id, index)}; const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)}; + if (ctx.profile.warp_size_potentially_larger_than_guest) { + src_thread_id = AddPartitionBase(ctx, src_thread_id); + } + SetInBoundsFlag(inst, in_range); return SelectValue(ctx, in_range, value, src_thread_id); } diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp index 41dc6d031..3b97721e1 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp @@ -35,6 +35,7 @@ Id ImageType(EmitContext& ctx, const TextureDescriptor& desc) { const spv::ImageFormat format{spv::ImageFormat::Unknown}; const Id type{ctx.F32[1]}; const bool depth{desc.is_depth}; + const bool ms{desc.is_multisample}; switch (desc.type) { case TextureType::Color1D: return ctx.TypeImage(type, spv::Dim::Dim1D, depth, false, false, 1, format); @@ -42,9 +43,9 @@ Id ImageType(EmitContext& ctx, const TextureDescriptor& desc) { return ctx.TypeImage(type, spv::Dim::Dim1D, depth, true, false, 1, format); case TextureType::Color2D: case TextureType::Color2DRect: - return ctx.TypeImage(type, spv::Dim::Dim2D, depth, false, false, 1, format); + return ctx.TypeImage(type, spv::Dim::Dim2D, depth, false, ms, 1, format); case TextureType::ColorArray2D: - return ctx.TypeImage(type, spv::Dim::Dim2D, depth, true, false, 1, format); + return ctx.TypeImage(type, spv::Dim::Dim2D, depth, true, ms, 1, format); case TextureType::Color3D: return ctx.TypeImage(type, spv::Dim::Dim3D, depth, false, false, 1, format); case TextureType::ColorCube: @@ -544,7 +545,7 @@ void EmitContext::DefineCommonTypes(const Info& info) { U16 = Name(TypeInt(16, false), "u16"); S16 = Name(TypeInt(16, true), "s16"); } - if (info.uses_int64) { + if (info.uses_int64 && profile.support_int64) { AddCapability(spv::Capability::Int64); U64 = Name(TypeInt(64, false), "u64"); } @@ -721,9 +722,21 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) { size_t label_index{0}; if (info.loads.AnyComponent(IR::Attribute::PositionX)) { AddLabel(labels[label_index]); - const Id pointer{is_array - ? OpAccessChain(input_f32, input_position, vertex, masked_index) - : OpAccessChain(input_f32, input_position, masked_index)}; + const Id pointer{[&]() { + if (need_input_position_indirect) { + if (is_array) + return OpAccessChain(input_f32, input_position, vertex, u32_zero_value, + masked_index); + else + return OpAccessChain(input_f32, input_position, u32_zero_value, + masked_index); + } else { + if (is_array) + return OpAccessChain(input_f32, input_position, vertex, masked_index); + else + return OpAccessChain(input_f32, input_position, masked_index); + } + }()}; const Id result{OpLoad(F32[1], pointer)}; OpReturnValue(result); ++label_index; @@ -1367,30 +1380,56 @@ void EmitContext::DefineInputs(const IR::Program& program) { Decorate(layer, spv::Decoration::Flat); } if (loads.AnyComponent(IR::Attribute::PositionX)) { - const bool is_fragment{stage != Stage::Fragment}; - const spv::BuiltIn built_in{is_fragment ? spv::BuiltIn::Position : spv::BuiltIn::FragCoord}; - input_position = DefineInput(*this, F32[4], true, built_in); - if (profile.support_geometry_shader_passthrough) { - if (info.passthrough.AnyComponent(IR::Attribute::PositionX)) { - Decorate(input_position, spv::Decoration::PassthroughNV); + const bool is_fragment{stage == Stage::Fragment}; + if (!is_fragment && profile.has_broken_spirv_position_input) { + need_input_position_indirect = true; + + const Id input_position_struct = TypeStruct(F32[4]); + input_position = DefineInput(*this, input_position_struct, true); + + MemberDecorate(input_position_struct, 0, spv::Decoration::BuiltIn, + static_cast<unsigned>(spv::BuiltIn::Position)); + Decorate(input_position_struct, spv::Decoration::Block); + } else { + const spv::BuiltIn built_in{is_fragment ? spv::BuiltIn::FragCoord + : spv::BuiltIn::Position}; + input_position = DefineInput(*this, F32[4], true, built_in); + + if (profile.support_geometry_shader_passthrough) { + if (info.passthrough.AnyComponent(IR::Attribute::PositionX)) { + Decorate(input_position, spv::Decoration::PassthroughNV); + } } } } if (loads[IR::Attribute::InstanceId]) { if (profile.support_vertex_instance_id) { instance_id = DefineInput(*this, U32[1], true, spv::BuiltIn::InstanceId); + if (loads[IR::Attribute::BaseInstance]) { + base_instance = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseVertex); + } } else { instance_index = DefineInput(*this, U32[1], true, spv::BuiltIn::InstanceIndex); base_instance = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseInstance); } + } else if (loads[IR::Attribute::BaseInstance]) { + base_instance = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseInstance); } if (loads[IR::Attribute::VertexId]) { if (profile.support_vertex_instance_id) { vertex_id = DefineInput(*this, U32[1], true, spv::BuiltIn::VertexId); + if (loads[IR::Attribute::BaseVertex]) { + base_vertex = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseVertex); + } } else { vertex_index = DefineInput(*this, U32[1], true, spv::BuiltIn::VertexIndex); base_vertex = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseVertex); } + } else if (loads[IR::Attribute::BaseVertex]) { + base_vertex = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseVertex); + } + if (loads[IR::Attribute::DrawID]) { + draw_index = DefineInput(*this, U32[1], true, spv::BuiltIn::DrawIndex); } if (loads[IR::Attribute::FrontFace]) { front_face = DefineInput(*this, U1, true, spv::BuiltIn::FrontFacing); diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.h b/src/shader_recompiler/backend/spirv/spirv_emit_context.h index dde45b4bc..dbc5c55b9 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.h +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.h @@ -218,6 +218,7 @@ public: Id base_instance{}; Id vertex_id{}; Id vertex_index{}; + Id draw_index{}; Id base_vertex{}; Id front_face{}; Id point_coord{}; @@ -279,6 +280,7 @@ public: Id write_global_func_u32x2{}; Id write_global_func_u32x4{}; + bool need_input_position_indirect{}; Id input_position{}; std::array<Id, 32> input_generics{}; |