diff options
Diffstat (limited to 'src/shader_recompiler/backend/spirv')
9 files changed, 168 insertions, 78 deletions
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.cpp b/src/shader_recompiler/backend/spirv/emit_spirv.cpp index 0f86a8004..34592a01f 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv.cpp @@ -387,7 +387,7 @@ void SetupSignedNanCapabilities(const Profile& profile, const IR::Program& progr } void SetupTransformFeedbackCapabilities(EmitContext& ctx, Id main_func) { - if (ctx.runtime_info.xfb_varyings.empty()) { + if (ctx.runtime_info.xfb_count == 0) { return; } ctx.AddCapability(spv::Capability::TransformFeedback); diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp index 4b3043b65..0ce73f289 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp @@ -69,6 +69,11 @@ Id StorageAtomicU32(EmitContext& ctx, const IR::Value& binding, const IR::Value& Id StorageAtomicU64(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset, Id value, Id (Sirit::Module::*atomic_func)(Id, Id, Id, Id, Id), Id (Sirit::Module::*non_atomic_func)(Id, Id, Id)) { + if (!ctx.profile.support_descriptor_aliasing) { + LOG_WARNING(Shader_SPIRV, "Descriptor aliasing not supported, this cannot be atomic."); + return ctx.ConstantNull(ctx.U64); + } + if (ctx.profile.support_int64_atomics) { const Id pointer{StoragePointer(ctx, ctx.storage_types.U64, &StorageDefinitions::U64, binding, offset, sizeof(u64))}; @@ -86,6 +91,11 @@ Id StorageAtomicU64(EmitContext& ctx, const IR::Value& binding, const IR::Value& Id StorageAtomicU32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset, Id value, Id (Sirit::Module::*non_atomic_func)(Id, Id, Id)) { + if (!ctx.profile.support_descriptor_aliasing) { + LOG_WARNING(Shader_SPIRV, "Descriptor aliasing not supported, this cannot be atomic."); + return ctx.ConstantNull(ctx.U32[2]); + } + LOG_WARNING(Shader_SPIRV, "Int64 atomics not supported, fallback to non-atomic"); const Id pointer{StoragePointer(ctx, ctx.storage_types.U32x2, &StorageDefinitions::U32x2, binding, offset, sizeof(u32[2]))}; 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 0cd87a48f..2868fc57d 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 @@ -10,27 +10,6 @@ namespace Shader::Backend::SPIRV { namespace { -struct AttrInfo { - Id pointer; - Id id; - bool needs_cast; -}; - -std::optional<AttrInfo> AttrTypes(EmitContext& ctx, u32 index) { - const AttributeType type{ctx.runtime_info.generic_input_types.at(index)}; - switch (type) { - case AttributeType::Float: - return AttrInfo{ctx.input_f32, ctx.F32[1], false}; - case AttributeType::UnsignedInt: - return AttrInfo{ctx.input_u32, ctx.U32[1], true}; - case AttributeType::SignedInt: - return AttrInfo{ctx.input_s32, ctx.TypeInt(32, true), true}; - case AttributeType::Disabled: - return std::nullopt; - } - throw InvalidArgument("Invalid attribute type {}", type); -} - template <typename... Args> Id AttrPointer(EmitContext& ctx, Id pointer_type, Id vertex, Id base, Args&&... args) { switch (ctx.stage) { @@ -302,15 +281,26 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) { const u32 element{static_cast<u32>(attr) % 4}; if (IR::IsGeneric(attr)) { const u32 index{IR::GenericAttributeIndex(attr)}; - const std::optional<AttrInfo> type{AttrTypes(ctx, index)}; - if (!type || !ctx.runtime_info.previous_stage_stores.Generic(index, element)) { + const auto& generic{ctx.input_generics.at(index)}; + if (!ValidId(generic.id)) { // Attribute is disabled or varying component is not written return ctx.Const(element == 3 ? 1.0f : 0.0f); } - const Id generic_id{ctx.input_generics.at(index)}; - const Id pointer{AttrPointer(ctx, type->pointer, vertex, generic_id, ctx.Const(element))}; - const Id value{ctx.OpLoad(type->id, pointer)}; - return type->needs_cast ? ctx.OpBitcast(ctx.F32[1], value) : value; + const Id pointer{ + AttrPointer(ctx, generic.pointer_type, vertex, generic.id, ctx.Const(element))}; + const Id value{ctx.OpLoad(generic.component_type, pointer)}; + return [&ctx, generic, value]() { + switch (generic.load_op) { + case InputGenericLoadOp::Bitcast: + return ctx.OpBitcast(ctx.F32[1], value); + case InputGenericLoadOp::SToF: + return ctx.OpConvertSToF(ctx.F32[1], value); + case InputGenericLoadOp::UToF: + return ctx.OpConvertUToF(ctx.F32[1], value); + default: + return value; + }; + }(); } switch (attr) { case IR::Attribute::PrimitiveId: @@ -339,9 +329,7 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, Id vertex) { if (ctx.profile.support_vertex_instance_id) { return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.vertex_id)); } else { - const Id index{ctx.OpLoad(ctx.U32[1], ctx.vertex_index)}; - const Id base{ctx.OpLoad(ctx.U32[1], ctx.base_vertex)}; - return ctx.OpBitcast(ctx.F32[1], ctx.OpISub(ctx.U32[1], index, base)); + return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.vertex_index)); } case IR::Attribute::BaseInstance: return ctx.OpBitcast(ctx.F32[1], ctx.OpLoad(ctx.U32[1], ctx.base_instance)); @@ -386,9 +374,7 @@ Id EmitGetAttributeU32(EmitContext& ctx, IR::Attribute attr, Id) { if (ctx.profile.support_vertex_instance_id) { return ctx.OpLoad(ctx.U32[1], ctx.vertex_id); } else { - const Id index{ctx.OpLoad(ctx.U32[1], ctx.vertex_index)}; - const Id base{ctx.OpLoad(ctx.U32[1], ctx.base_vertex)}; - return ctx.OpISub(ctx.U32[1], index, base); + return ctx.OpLoad(ctx.U32[1], ctx.vertex_index); } case IR::Attribute::BaseInstance: return ctx.OpLoad(ctx.U32[1], ctx.base_instance); @@ -473,7 +459,8 @@ void EmitSetFragColor(EmitContext& ctx, u32 index, u32 component, Id value) { } void EmitSetSampleMask(EmitContext& ctx, Id value) { - ctx.OpStore(ctx.sample_mask, value); + const Id pointer{ctx.OpAccessChain(ctx.output_u32, ctx.sample_mask, ctx.u32_zero_value)}; + ctx.OpStore(pointer, value); } void EmitSetFragDepth(EmitContext& ctx, Id value) { diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp index fb5799c42..7d901c04b 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_image.cpp @@ -201,6 +201,13 @@ Id Image(EmitContext& ctx, const IR::Value& index, IR::TextureInstInfo info) { } } +bool IsTextureMsaa(EmitContext& ctx, const IR::TextureInstInfo& info) { + if (info.type == TextureType::Buffer) { + return false; + } + return ctx.textures.at(info.descriptor_index).is_multisample; +} + Id Decorate(EmitContext& ctx, IR::Inst* inst, Id sample) { const auto info{inst->Flags<IR::TextureInstInfo>()}; if (info.relaxed_precision != 0) { @@ -254,6 +261,30 @@ Id BitTest(EmitContext& ctx, Id mask, Id bit) { const Id bit_value{ctx.OpBitwiseAnd(ctx.U32[1], shifted, ctx.Const(1u))}; return ctx.OpINotEqual(ctx.U1, bit_value, ctx.u32_zero_value); } + +Id ImageGatherSubpixelOffset(EmitContext& ctx, const IR::TextureInstInfo& info, Id texture, + Id coords) { + // Apply a subpixel offset of 1/512 the texel size of the texture to ensure same rounding on + // AMD hardware as on Maxwell or other Nvidia architectures. + const auto calculate_coords{[&](size_t dim) { + const Id nudge{ctx.Const(0x1p-9f)}; + const Id image_size{ctx.OpImageQuerySizeLod(ctx.U32[dim], texture, ctx.u32_zero_value)}; + Id offset{dim == 2 ? ctx.ConstantComposite(ctx.F32[dim], nudge, nudge) + : ctx.ConstantComposite(ctx.F32[dim], nudge, nudge, ctx.f32_zero_value)}; + offset = ctx.OpFDiv(ctx.F32[dim], offset, ctx.OpConvertUToF(ctx.F32[dim], image_size)); + return ctx.OpFAdd(ctx.F32[dim], coords, offset); + }}; + switch (info.type) { + case TextureType::Color2D: + case TextureType::Color2DRect: + return calculate_coords(2); + case TextureType::ColorArray2D: + case TextureType::ColorCube: + return calculate_coords(3); + default: + return coords; + } +} } // Anonymous namespace Id EmitBindlessImageSampleImplicitLod(EmitContext&) { @@ -416,6 +447,9 @@ Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id const IR::Value& offset, const IR::Value& offset2) { const auto info{inst->Flags<IR::TextureInstInfo>()}; const ImageOperands operands(ctx, offset, offset2); + if (ctx.profile.need_gather_subpixel_offset) { + coords = ImageGatherSubpixelOffset(ctx, info, TextureImage(ctx, info, index), coords); + } return Emit(&EmitContext::OpImageSparseGather, &EmitContext::OpImageGather, ctx, inst, ctx.F32[4], Texture(ctx, info, index), coords, ctx.Const(info.gather_component), operands.MaskOptional(), operands.Span()); @@ -425,6 +459,9 @@ Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, const IR::Value& offset, const IR::Value& offset2, Id dref) { const auto info{inst->Flags<IR::TextureInstInfo>()}; const ImageOperands operands(ctx, offset, offset2); + if (ctx.profile.need_gather_subpixel_offset) { + coords = ImageGatherSubpixelOffset(ctx, info, TextureImage(ctx, info, index), coords); + } return Emit(&EmitContext::OpImageSparseDrefGather, &EmitContext::OpImageDrefGather, ctx, inst, ctx.F32[4], Texture(ctx, info, index), coords, dref, operands.MaskOptional(), operands.Span()); @@ -436,34 +473,42 @@ 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); }}; + const bool is_msaa{IsTextureMsaa(ctx, info)}; + const bool uses_lod{!is_msaa && info.type != TextureType::Buffer}; + const auto query{[&](Id type) { + return uses_lod ? ctx.OpImageQuerySizeLod(type, image, lod) + : ctx.OpImageQuerySize(type, image); + }}; switch (info.type) { case TextureType::Color1D: - return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[1], image, lod), - zero, zero, mips()); + return ctx.OpCompositeConstruct(ctx.U32[4], query(ctx.U32[1]), zero, zero, mips()); case TextureType::ColorArray1D: case TextureType::Color2D: case TextureType::ColorCube: case TextureType::Color2DRect: - return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[2], image, lod), - zero, mips()); + return ctx.OpCompositeConstruct(ctx.U32[4], query(ctx.U32[2]), zero, mips()); case TextureType::ColorArray2D: case TextureType::Color3D: case TextureType::ColorArrayCube: - return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[3], image, lod), - mips()); + return ctx.OpCompositeConstruct(ctx.U32[4], query(ctx.U32[3]), mips()); case TextureType::Buffer: - return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySize(ctx.U32[1], image), zero, - zero, mips()); + return ctx.OpCompositeConstruct(ctx.U32[4], query(ctx.U32[1]), zero, zero, mips()); } throw LogicError("Unspecified image type {}", info.type.Value()); } 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 c5db19d09..77ff8c573 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp @@ -17,7 +17,22 @@ Id GetThreadId(EmitContext& ctx) { Id WarpExtract(EmitContext& ctx, Id value) { const Id thread_id{GetThreadId(ctx)}; const Id local_index{ctx.OpShiftRightArithmetic(ctx.U32[1], thread_id, ctx.Const(5U))}; - return ctx.OpVectorExtractDynamic(ctx.U32[1], value, local_index); + if (ctx.profile.has_broken_spirv_subgroup_mask_vector_extract_dynamic) { + const Id c0_sel{ctx.OpSelect(ctx.U32[1], ctx.OpIEqual(ctx.U1, local_index, ctx.Const(0U)), + ctx.OpCompositeExtract(ctx.U32[1], value, 0U), ctx.Const(0U))}; + const Id c1_sel{ctx.OpSelect(ctx.U32[1], ctx.OpIEqual(ctx.U1, local_index, ctx.Const(1U)), + ctx.OpCompositeExtract(ctx.U32[1], value, 1U), ctx.Const(0U))}; + const Id c2_sel{ctx.OpSelect(ctx.U32[1], ctx.OpIEqual(ctx.U1, local_index, ctx.Const(2U)), + ctx.OpCompositeExtract(ctx.U32[1], value, 2U), ctx.Const(0U))}; + const Id c3_sel{ctx.OpSelect(ctx.U32[1], ctx.OpIEqual(ctx.U1, local_index, ctx.Const(3U)), + ctx.OpCompositeExtract(ctx.U32[1], value, 3U), ctx.Const(0U))}; + const Id c0_or_c1{ctx.OpBitwiseOr(ctx.U32[1], c0_sel, c1_sel)}; + const Id c2_or_c3{ctx.OpBitwiseOr(ctx.U32[1], c2_sel, c3_sel)}; + const Id c0_or_c1_or_c2_or_c3{ctx.OpBitwiseOr(ctx.U32[1], c0_or_c1, c2_or_c3)}; + return c0_or_c1_or_c2_or_c3; + } else { + return ctx.OpVectorExtractDynamic(ctx.U32[1], value, local_index); + } } Id LoadMask(EmitContext& ctx, Id mask) { diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp index a0c155fdb..bec5db173 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp @@ -25,16 +25,11 @@ enum class Operation { FPMax, }; -struct AttrInfo { - Id pointer; - Id id; - bool needs_cast; -}; - 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 +37,9 @@ 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: @@ -165,7 +160,7 @@ void DefineGenericOutput(EmitContext& ctx, size_t index, std::optional<u32> invo const u32 remainder{4 - element}; const TransformFeedbackVarying* xfb_varying{}; const size_t xfb_varying_index{base_attr_index + element}; - if (xfb_varying_index < ctx.runtime_info.xfb_varyings.size()) { + if (xfb_varying_index < ctx.runtime_info.xfb_count) { xfb_varying = &ctx.runtime_info.xfb_varyings[xfb_varying_index]; xfb_varying = xfb_varying->components > 0 ? xfb_varying : nullptr; } @@ -205,23 +200,37 @@ Id GetAttributeType(EmitContext& ctx, AttributeType type) { return ctx.TypeVector(ctx.TypeInt(32, true), 4); case AttributeType::UnsignedInt: return ctx.U32[4]; + case AttributeType::SignedScaled: + return ctx.profile.support_scaled_attributes ? ctx.F32[4] + : ctx.TypeVector(ctx.TypeInt(32, true), 4); + case AttributeType::UnsignedScaled: + return ctx.profile.support_scaled_attributes ? ctx.F32[4] : ctx.U32[4]; case AttributeType::Disabled: break; } throw InvalidArgument("Invalid attribute type {}", type); } -std::optional<AttrInfo> AttrTypes(EmitContext& ctx, u32 index) { - const AttributeType type{ctx.runtime_info.generic_input_types.at(index)}; +InputGenericInfo GetAttributeInfo(EmitContext& ctx, AttributeType type, Id id) { switch (type) { case AttributeType::Float: - return AttrInfo{ctx.input_f32, ctx.F32[1], false}; + return InputGenericInfo{id, ctx.input_f32, ctx.F32[1], InputGenericLoadOp::None}; case AttributeType::UnsignedInt: - return AttrInfo{ctx.input_u32, ctx.U32[1], true}; + return InputGenericInfo{id, ctx.input_u32, ctx.U32[1], InputGenericLoadOp::Bitcast}; case AttributeType::SignedInt: - return AttrInfo{ctx.input_s32, ctx.TypeInt(32, true), true}; + return InputGenericInfo{id, ctx.input_s32, ctx.TypeInt(32, true), + InputGenericLoadOp::Bitcast}; + case AttributeType::SignedScaled: + return ctx.profile.support_scaled_attributes + ? InputGenericInfo{id, ctx.input_f32, ctx.F32[1], InputGenericLoadOp::None} + : InputGenericInfo{id, ctx.input_s32, ctx.TypeInt(32, true), + InputGenericLoadOp::SToF}; + case AttributeType::UnsignedScaled: + return ctx.profile.support_scaled_attributes + ? InputGenericInfo{id, ctx.input_f32, ctx.F32[1], InputGenericLoadOp::None} + : InputGenericInfo{id, ctx.input_u32, ctx.U32[1], InputGenericLoadOp::UToF}; case AttributeType::Disabled: - return std::nullopt; + return InputGenericInfo{}; } throw InvalidArgument("Invalid attribute type {}", type); } @@ -745,18 +754,29 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) { continue; } AddLabel(labels[label_index]); - const auto type{AttrTypes(*this, static_cast<u32>(index))}; - if (!type) { + const auto& generic{input_generics.at(index)}; + const Id generic_id{generic.id}; + if (!ValidId(generic_id)) { OpReturnValue(Const(0.0f)); ++label_index; continue; } - const Id generic_id{input_generics.at(index)}; - const Id pointer{is_array - ? OpAccessChain(type->pointer, generic_id, vertex, masked_index) - : OpAccessChain(type->pointer, generic_id, masked_index)}; - const Id value{OpLoad(type->id, pointer)}; - const Id result{type->needs_cast ? OpBitcast(F32[1], value) : value}; + const Id pointer{ + is_array ? OpAccessChain(generic.pointer_type, generic_id, vertex, masked_index) + : OpAccessChain(generic.pointer_type, generic_id, masked_index)}; + const Id value{OpLoad(generic.component_type, pointer)}; + const Id result{[this, generic, value]() { + switch (generic.load_op) { + case InputGenericLoadOp::Bitcast: + return OpBitcast(F32[1], value); + case InputGenericLoadOp::SToF: + return OpConvertSToF(F32[1], value); + case InputGenericLoadOp::UToF: + return OpConvertUToF(F32[1], value); + default: + return value; + }; + }()}; OpReturnValue(result); ++label_index; } @@ -1287,6 +1307,7 @@ void EmitContext::DefineTextures(const Info& info, u32& binding, u32& scaling_in .pointer_type = pointer_type, .image_type = image_type, .count = desc.count, + .is_multisample = desc.is_multisample, }); if (profile.supported_spirv >= 0x00010400) { interfaces.push_back(id); @@ -1455,7 +1476,7 @@ void EmitContext::DefineInputs(const IR::Program& program) { const Id id{DefineInput(*this, type, true)}; Decorate(id, spv::Decoration::Location, static_cast<u32>(index)); Name(id, fmt::format("in_attr{}", index)); - input_generics[index] = id; + input_generics[index] = GetAttributeInfo(*this, input_type, id); if (info.passthrough.Generic(index) && profile.support_geometry_shader_passthrough) { Decorate(id, spv::Decoration::PassthroughNV); @@ -1570,7 +1591,8 @@ void EmitContext::DefineOutputs(const IR::Program& program) { Decorate(frag_depth, spv::Decoration::BuiltIn, spv::BuiltIn::FragDepth); } if (info.stores_sample_mask) { - sample_mask = DefineOutput(*this, U32[1], std::nullopt); + const Id array_type{TypeArray(U32[1], Const(1U))}; + sample_mask = DefineOutput(*this, array_type, std::nullopt); Decorate(sample_mask, spv::Decoration::BuiltIn, spv::BuiltIn::SampleMask); } break; diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.h b/src/shader_recompiler/backend/spirv/spirv_emit_context.h index dbc5c55b9..e63330f11 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.h +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.h @@ -35,6 +35,7 @@ struct TextureDefinition { Id pointer_type; Id image_type; u32 count; + bool is_multisample; }; struct TextureBufferDefinition { @@ -94,6 +95,20 @@ struct StorageDefinitions { Id U32x4{}; }; +enum class InputGenericLoadOp { + None, + Bitcast, + SToF, + UToF, +}; + +struct InputGenericInfo { + Id id; + Id pointer_type; + Id component_type; + InputGenericLoadOp load_op; +}; + struct GenericElementInfo { Id id{}; u32 first_element{}; @@ -282,7 +297,7 @@ public: bool need_input_position_indirect{}; Id input_position{}; - std::array<Id, 32> input_generics{}; + std::array<InputGenericInfo, 32> input_generics{}; Id output_point_size{}; Id output_position{}; |