shader: Implement indexed attributes

src/shader_recompiler/backend/spirv/emit_context.cpp

@@ -82,6 +82,28 @@ Id GetAttributeType(EmitContext& ctx, AttributeType type) {
     }
     throw InvalidArgument("Invalid attribute type {}", type);
 }
+
+struct AttrInfo {
+    Id pointer;
+    Id id;
+    bool needs_cast;
+};
+
+std::optional<AttrInfo> AttrTypes(EmitContext& ctx, u32 index) {
+    const AttributeType type{ctx.profile.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);
+}
+
 } // Anonymous namespace
 
 void VectorTypes::Define(Sirit::Module& sirit_ctx, Id base_type, std::string_view name) {
@@ -107,6 +129,7 @@ EmitContext::EmitContext(const Profile& profile_, IR::Program& program, u32& bin
     DefineConstantBuffers(program.info, binding);
     DefineStorageBuffers(program.info, binding);
     DefineTextures(program.info, binding);
+    DefineAttributeMemAccess(program.info);
     DefineLabels(program);
 }
 
@@ -290,6 +313,107 @@ void EmitContext::DefineSharedMemory(const IR::Program& program) {
     }
 }
 
+void EmitContext::DefineAttributeMemAccess(const Info& info) {
+    const auto make_load{[&]() {
+        const Id end_block{OpLabel()};
+        const Id default_label{OpLabel()};
+
+        const Id func_type_load{TypeFunction(F32[1], U32[1])};
+        const Id func{OpFunction(F32[1], spv::FunctionControlMask::MaskNone, func_type_load)};
+        const Id offset{OpFunctionParameter(U32[1])};
+        AddLabel();
+        const Id base_index{OpShiftRightLogical(U32[1], offset, Constant(U32[1], 2U))};
+        const Id masked_index{OpBitwiseAnd(U32[1], base_index, Constant(U32[1], 3U))};
+        const Id compare_index{OpShiftRightLogical(U32[1], base_index, Constant(U32[1], 2U))};
+        std::vector<Sirit::Literal> literals;
+        std::vector<Id> labels;
+        const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2;
+        for (u32 i = 0; i < info.input_generics.size(); i++) {
+            if (!info.input_generics[i].used) {
+                continue;
+            }
+            literals.push_back(base_attribute_value + i);
+            labels.push_back(OpLabel());
+        }
+        OpSelectionMerge(end_block, spv::SelectionControlMask::MaskNone);
+        OpSwitch(compare_index, default_label, literals, labels);
+        AddLabel(default_label);
+        OpReturnValue(Constant(F32[1], 0.0f));
+        size_t label_index = 0;
+        for (u32 i = 0; i < info.input_generics.size(); i++) {
+            if (!info.input_generics[i].used) {
+                continue;
+            }
+            AddLabel(labels[label_index]);
+            const auto type{AttrTypes(*this, i)};
+            if (!type) {
+                OpReturnValue(Constant(F32[1], 0.0f));
+                label_index++;
+                continue;
+            }
+            const Id generic_id{input_generics.at(i)};
+            const Id pointer{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};
+            OpReturnValue(result);
+            label_index++;
+        }
+        AddLabel(end_block);
+        OpUnreachable();
+        OpFunctionEnd();
+        return func;
+    }};
+    const auto make_store{[&]() {
+        const Id end_block{OpLabel()};
+        const Id default_label{OpLabel()};
+
+        const Id func_type_store{TypeFunction(void_id, U32[1], F32[1])};
+        const Id func{OpFunction(void_id, spv::FunctionControlMask::MaskNone, func_type_store)};
+        const Id offset{OpFunctionParameter(U32[1])};
+        const Id store_value{OpFunctionParameter(F32[1])};
+        AddLabel();
+        const Id base_index{OpShiftRightLogical(U32[1], offset, Constant(U32[1], 2U))};
+        const Id masked_index{OpBitwiseAnd(U32[1], base_index, Constant(U32[1], 3U))};
+        const Id compare_index{OpShiftRightLogical(U32[1], base_index, Constant(U32[1], 2U))};
+        std::vector<Sirit::Literal> literals;
+        std::vector<Id> labels;
+        const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2;
+        for (u32 i = 0; i < info.stores_generics.size(); i++) {
+            if (!info.stores_generics[i]) {
+                continue;
+            }
+            literals.push_back(base_attribute_value + i);
+            labels.push_back(OpLabel());
+        }
+        OpSelectionMerge(end_block, spv::SelectionControlMask::MaskNone);
+        OpSwitch(compare_index, default_label, literals, labels);
+        AddLabel(default_label);
+        OpReturn();
+        size_t label_index = 0;
+        for (u32 i = 0; i < info.stores_generics.size(); i++) {
+            if (!info.stores_generics[i]) {
+                continue;
+            }
+            AddLabel(labels[label_index]);
+            const Id generic_id{output_generics.at(i)};
+            const Id pointer{OpAccessChain(output_f32, generic_id, masked_index)};
+            OpStore(pointer, store_value);
+            OpReturn();
+            label_index++;
+        }
+        AddLabel(end_block);
+        OpUnreachable();
+        OpFunctionEnd();
+        return func;
+    }};
+    if (info.loads_indexed_attributes) {
+        indexed_load_func = make_load();
+    }
+    if (info.stores_indexed_attributes) {
+        indexed_store_func = make_store();
+    }
+}
+
 void EmitContext::DefineConstantBuffers(const Info& info, u32& binding) {
     if (info.constant_buffer_descriptors.empty()) {
         return;

src/shader_recompiler/backend/spirv/emit_context.h

@@ -116,6 +116,9 @@ public:
     Id fswzadd_lut_a{};
     Id fswzadd_lut_b{};
 
+    Id indexed_load_func{};
+    Id indexed_store_func{};
+
     Id local_memory{};
 
     Id shared_memory_u8{};
@@ -148,6 +151,7 @@ private:
     void DefineConstantBuffers(const Info& info, u32& binding);
     void DefineStorageBuffers(const Info& info, u32& binding);
     void DefineTextures(const Info& info, u32& binding);
+    void DefineAttributeMemAccess(const Info& info);
     void DefineLabels(IR::Program& program);
 
     void DefineConstantBuffers(const Info& info, Id UniformDefinitions::*member_type, u32 binding,

src/shader_recompiler/backend/spirv/emit_spirv.h

@@ -51,8 +51,8 @@ Id EmitGetCbufF32(EmitContext& ctx, const IR::Value& binding, const IR::Value& o
 Id EmitGetCbufU32x2(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset);
 Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr);
 void EmitSetAttribute(EmitContext& ctx, IR::Attribute attr, Id value);
-void EmitGetAttributeIndexed(EmitContext& ctx);
-void EmitSetAttributeIndexed(EmitContext& ctx);
+Id EmitGetAttributeIndexed(EmitContext& ctx, Id offset);
+void EmitSetAttributeIndexed(EmitContext& ctx, Id offset, Id value);
 void EmitSetFragColor(EmitContext& ctx, u32 index, u32 component, Id value);
 void EmitSetFragDepth(EmitContext& ctx, Id value);
 void EmitGetZFlag(EmitContext& ctx);

src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp

@@ -216,12 +216,12 @@ void EmitSetAttribute(EmitContext& ctx, IR::Attribute attr, Id value) {
     ctx.OpStore(*output, value);
 }
 
-void EmitGetAttributeIndexed(EmitContext&) {
-    throw NotImplementedException("SPIR-V Instruction");
+Id EmitGetAttributeIndexed(EmitContext& ctx, Id offset) {
+    return ctx.OpFunctionCall(ctx.F32[1], ctx.indexed_load_func, offset);
 }
 
-void EmitSetAttributeIndexed(EmitContext&) {
-    throw NotImplementedException("SPIR-V Instruction");
+void EmitSetAttributeIndexed(EmitContext& ctx, Id offset, Id value) {
+    ctx.OpFunctionCall(ctx.void_id, ctx.indexed_store_func, offset, value);
 }
 
 void EmitSetFragColor(EmitContext& ctx, u32 index, u32 component, Id value) {

src/shader_recompiler/frontend/ir/ir_emitter.cpp

@@ -307,6 +307,14 @@ void IREmitter::SetAttribute(IR::Attribute attribute, const F32& value) {
     Inst(Opcode::SetAttribute, attribute, value);
 }
 
+F32 IREmitter::GetAttributeIndexed(IR::U32 phys_address) {
+    return Inst<F32>(Opcode::GetAttributeIndexed, phys_address);
+}
+
+void IREmitter::SetAttributeIndexed(IR::U32 phys_address, const F32& value) {
+    Inst(Opcode::SetAttributeIndexed, phys_address, value);
+}
+
 void IREmitter::SetFragColor(u32 index, u32 component, const F32& value) {
     Inst(Opcode::SetFragColor, Imm32(index), Imm32(component), value);
 }

src/shader_recompiler/frontend/ir/ir_emitter.h

@@ -76,6 +76,9 @@ public:
     [[nodiscard]] F32 GetAttribute(IR::Attribute attribute);
     void SetAttribute(IR::Attribute attribute, const F32& value);
 
+    [[nodiscard]] F32 GetAttributeIndexed(IR::U32 phys_address);
+    void SetAttributeIndexed(IR::U32 phys_address, const F32& value);
+
     void SetFragColor(u32 index, u32 component, const F32& value);
     void SetFragDepth(const F32& value);
 

src/shader_recompiler/frontend/maxwell/program.cpp

@@ -87,7 +87,7 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
     Optimization::DeadCodeEliminationPass(program);
     Optimization::IdentityRemovalPass(program);
     Optimization::VerificationPass(program);
-    Optimization::CollectShaderInfoPass(program);
+    Optimization::CollectShaderInfoPass(env, program);
     CollectInterpolationInfo(env, program);
     return program;
 }

src/shader_recompiler/frontend/maxwell/translate/impl/load_store_attribute.cpp

@@ -31,7 +31,7 @@ enum class SampleMode : u64 {
     Offset,
 };
 
-int NumElements(Size size) {
+u32 NumElements(Size size) {
     switch (size) {
     case Size::B32:
         return 1;
@@ -65,15 +65,21 @@ void TranslatorVisitor::ALD(u64 insn) {
     if (ald.patch != 0) {
         throw NotImplementedException("P");
     }
-    if (ald.index_reg != IR::Reg::RZ) {
-        throw NotImplementedException("Indexed");
-    }
     const u64 offset{ald.absolute_offset.Value()};
     if (offset % 4 != 0) {
         throw NotImplementedException("Unaligned absolute offset {}", offset);
     }
-    const int num_elements{NumElements(ald.size)};
-    for (int element = 0; element < num_elements; ++element) {
+    const u32 num_elements{NumElements(ald.size)};
+    if (ald.index_reg != IR::Reg::RZ) {
+        const IR::U32 index_value = X(ald.index_reg);
+        for (u32 element = 0; element < num_elements; ++element) {
+            const IR::U32 final_offset =
+                element == 0 ? index_value : IR::U32{ir.IAdd(index_value, ir.Imm32(element * 4U))};
+            F(ald.dest_reg + element, ir.GetAttributeIndexed(final_offset));
+        }
+        return;
+    }
+    for (u32 element = 0; element < num_elements; ++element) {
         F(ald.dest_reg + element, ir.GetAttribute(IR::Attribute{offset / 4 + element}));
     }
 }
@@ -103,8 +109,17 @@ void TranslatorVisitor::AST(u64 insn) {
     if (offset % 4 != 0) {
         throw NotImplementedException("Unaligned absolute offset {}", offset);
     }
-    const int num_elements{NumElements(ast.size)};
-    for (int element = 0; element < num_elements; ++element) {
+    const u32 num_elements{NumElements(ast.size)};
+    if (ast.index_reg != IR::Reg::RZ) {
+        const IR::U32 index_value = X(ast.index_reg);
+        for (u32 element = 0; element < num_elements; ++element) {
+            const IR::U32 final_offset =
+                element == 0 ? index_value : IR::U32{ir.IAdd(index_value, ir.Imm32(element * 4U))};
+            ir.SetAttributeIndexed(final_offset, F(ast.src_reg + element));
+        }
+        return;
+    }
+    for (u32 element = 0; element < num_elements; ++element) {
         ir.SetAttribute(IR::Attribute{offset / 4 + element}, F(ast.src_reg + element));
     }
 }
@@ -134,12 +149,9 @@ void TranslatorVisitor::IPA(u64 insn) {
     //     gl_FragColor = colors[idx];
     // }
     const bool is_indexed{ipa.idx != 0 && ipa.index_reg != IR::Reg::RZ};
-    if (is_indexed) {
-        throw NotImplementedException("IDX");
-    }
-
     const IR::Attribute attribute{ipa.attribute};
-    IR::F32 value{ir.GetAttribute(attribute)};
+    IR::F32 value{is_indexed ? ir.GetAttributeIndexed(X(ipa.index_reg))
+                             : ir.GetAttribute(attribute)};
     if (IR::IsGeneric(attribute)) {
         const ProgramHeader& sph{env.SPH()};
         const u32 attr_index{IR::GenericAttributeIndex(attribute)};

src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp

@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "shader_recompiler/environment.h"
 #include "shader_recompiler/frontend/ir/microinstruction.h"
 #include "shader_recompiler/frontend/ir/modifiers.h"
 #include "shader_recompiler/frontend/ir/program.h"
@@ -323,6 +324,12 @@ void VisitUsages(Info& info, IR::Inst& inst) {
     case IR::Opcode::SetAttribute:
         SetAttribute(info, inst.Arg(0).Attribute());
         break;
+    case IR::Opcode::GetAttributeIndexed:
+        info.loads_indexed_attributes = true;
+        break;
+    case IR::Opcode::SetAttributeIndexed:
+        info.stores_indexed_attributes = true;
+        break;
     case IR::Opcode::SetFragColor:
         info.stores_frag_color[inst.Arg(0).U32()] = true;
         break;
@@ -502,15 +509,42 @@ void Visit(Info& info, IR::Inst& inst) {
     VisitUsages(info, inst);
     VisitFpModifiers(info, inst);
 }
+
+void GatherInfoFromHeader(Environment& env, Info& info) {
+    auto stage = env.ShaderStage();
+    if (stage == Stage::Compute) {
+        return;
+    }
+    const auto& header = env.SPH();
+    if (stage == Stage::Fragment) {
+        for (size_t i = 0; i < info.input_generics.size(); i++) {
+            info.input_generics[i].used =
+                info.input_generics[i].used || header.ps.IsGenericVectorActive(i);
+        }
+        return;
+    }
+    for (size_t i = 0; i < info.input_generics.size(); i++) {
+        info.input_generics[i].used =
+            info.input_generics[i].used || header.vtg.IsInputGenericVectorActive(i);
+    }
+    for (size_t i = 0; i < info.stores_generics.size(); i++) {
+        info.stores_generics[i] =
+            info.stores_generics[i] || header.vtg.IsOutputGenericVectorActive(i);
+    }
+    info.stores_clip_distance =
+        info.stores_clip_distance || header.vtg.omap_systemc.clip_distances != 0;
+}
+
 } // Anonymous namespace
 
-void CollectShaderInfoPass(IR::Program& program) {
+void CollectShaderInfoPass(Environment& env, IR::Program& program) {
     Info& info{program.info};
     for (IR::Block* const block : program.post_order_blocks) {
         for (IR::Inst& inst : block->Instructions()) {
             Visit(info, inst);
         }
     }
+    GatherInfoFromHeader(env, info);
 }
 
 } // namespace Shader::Optimization

src/shader_recompiler/ir_opt/passes.h

@@ -12,7 +12,7 @@
 
 namespace Shader::Optimization {
 
-void CollectShaderInfoPass(IR::Program& program);
+void CollectShaderInfoPass(Environment& env, IR::Program& program);
 void ConstantPropagationPass(IR::Program& program);
 void DeadCodeEliminationPass(IR::Program& program);
 void GlobalMemoryToStorageBufferPass(IR::Program& program);

src/shader_recompiler/program_header.h

@@ -68,10 +68,24 @@ struct ProgramHeader {
 
     union {
         struct {
-            INSERT_PADDING_BYTES_NOINIT(3);  // ImapSystemValuesA
-            INSERT_PADDING_BYTES_NOINIT(1);  // ImapSystemValuesB
-            INSERT_PADDING_BYTES_NOINIT(16); // ImapGenericVector[32]
-            INSERT_PADDING_BYTES_NOINIT(2);  // ImapColor
+            INSERT_PADDING_BYTES_NOINIT(3); // ImapSystemValuesA
+            INSERT_PADDING_BYTES_NOINIT(1); // ImapSystemValuesB
+
+            union {
+                BitField<0, 1, u8> x;
+                BitField<1, 1, u8> y;
+                BitField<2, 1, u8> z;
+                BitField<3, 1, u8> w;
+                BitField<4, 1, u8> x2;
+                BitField<5, 1, u8> y2;
+                BitField<6, 1, u8> z2;
+                BitField<7, 1, u8> w2;
+                BitField<0, 4, u8> first;
+                BitField<4, 4, u8> second;
+                u8 raw;
+            } imap_generic_vector[16];
+
+            INSERT_PADDING_BYTES_NOINIT(2); // ImapColor
             union {
                 BitField<0, 8, u16> clip_distances;
                 BitField<8, 1, u16> point_sprite_s;
@@ -82,15 +96,54 @@ struct ProgramHeader {
                 BitField<14, 1, u16> instance_id;
                 BitField<15, 1, u16> vertex_id;
             };
-            INSERT_PADDING_BYTES_NOINIT(5);  // ImapFixedFncTexture[10]
-            INSERT_PADDING_BYTES_NOINIT(1);  // ImapReserved
-            INSERT_PADDING_BYTES_NOINIT(3);  // OmapSystemValuesA
-            INSERT_PADDING_BYTES_NOINIT(1);  // OmapSystemValuesB
-            INSERT_PADDING_BYTES_NOINIT(16); // OmapGenericVector[32]
-            INSERT_PADDING_BYTES_NOINIT(2);  // OmapColor
-            INSERT_PADDING_BYTES_NOINIT(2);  // OmapSystemValuesC
-            INSERT_PADDING_BYTES_NOINIT(5);  // OmapFixedFncTexture[10]
-            INSERT_PADDING_BYTES_NOINIT(1);  // OmapReserved
+            INSERT_PADDING_BYTES_NOINIT(5); // ImapFixedFncTexture[10]
+            INSERT_PADDING_BYTES_NOINIT(1); // ImapReserved
+            INSERT_PADDING_BYTES_NOINIT(3); // OmapSystemValuesA
+            INSERT_PADDING_BYTES_NOINIT(1); // OmapSystemValuesB
+
+            union {
+                BitField<0, 1, u8> x;
+                BitField<1, 1, u8> y;
+                BitField<2, 1, u8> z;
+                BitField<3, 1, u8> w;
+                BitField<4, 1, u8> x2;
+                BitField<5, 1, u8> y2;
+                BitField<6, 1, u8> z2;
+                BitField<7, 1, u8> w2;
+                BitField<0, 4, u8> first;
+                BitField<4, 4, u8> second;
+                u8 raw;
+            } omap_generic_vector[16];
+
+            INSERT_PADDING_BYTES_NOINIT(2); // OmapColor
+
+            union {
+                BitField<0, 8, u16> clip_distances;
+                BitField<8, 1, u16> point_sprite_s;
+                BitField<9, 1, u16> point_sprite_t;
+                BitField<10, 1, u16> fog_coordinate;
+                BitField<12, 1, u16> tessellation_eval_point_u;
+                BitField<13, 1, u16> tessellation_eval_point_v;
+                BitField<14, 1, u16> instance_id;
+                BitField<15, 1, u16> vertex_id;
+            } omap_systemc;
+
+            INSERT_PADDING_BYTES_NOINIT(5); // OmapFixedFncTexture[10]
+            INSERT_PADDING_BYTES_NOINIT(1); // OmapReserved
+
+            [[nodiscard]] bool IsInputGenericVectorActive(size_t index) const {
+                if ((index & 1) == 0) {
+                    return imap_generic_vector[index >> 1].first != 0;
+                }
+                return imap_generic_vector[index >> 1].second != 0;
+            }
+
+            [[nodiscard]] bool IsOutputGenericVectorActive(size_t index) const {
+                if ((index & 1) == 0) {
+                    return omap_generic_vector[index >> 1].first != 0;
+                }
+                return omap_generic_vector[index >> 1].second != 0;
+            }
         } vtg;
 
         struct {
@@ -128,6 +181,10 @@ struct ProgramHeader {
                 const auto& vector{imap_generic_vector[attribute]};
                 return {vector.x, vector.y, vector.z, vector.w};
             }
+
+            [[nodiscard]] bool IsGenericVectorActive(size_t index) const {
+                return imap_generic_vector[index].raw != 0;
+            }
         } ps;
 
         std::array<u32, 0xf> raw;

src/shader_recompiler/shader_info.h

@@ -76,6 +76,7 @@ struct Info {
     bool loads_vertex_id{};
     bool loads_front_face{};
     bool loads_point_coord{};
+    bool loads_indexed_attributes{};
 
     std::array<bool, 8> stores_frag_color{};
     bool stores_frag_depth{};
@@ -84,6 +85,7 @@ struct Info {
     bool stores_point_size{};
     bool stores_clip_distance{};
     bool stores_viewport_index{};
+    bool stores_indexed_attributes{};
 
     bool uses_fp16{};
     bool uses_fp64{};