diff options
Diffstat (limited to 'src/shader_recompiler/backend/spirv/spirv_emit_context.cpp')
| -rw-r--r-- | src/shader_recompiler/backend/spirv/spirv_emit_context.cpp | 1585 |
1 files changed, 1585 insertions, 0 deletions
diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp new file mode 100644 index 000000000..723455462 --- /dev/null +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp | |||
| @@ -0,0 +1,1585 @@ | |||
| 1 | // Copyright 2021 yuzu Emulator Project | ||
| 2 | // Licensed under GPLv2 or any later version | ||
| 3 | // Refer to the license.txt file included. | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <array> | ||
| 7 | #include <climits> | ||
| 8 | #include <string_view> | ||
| 9 | |||
| 10 | #include <boost/container/static_vector.hpp> | ||
| 11 | |||
| 12 | #include <fmt/format.h> | ||
| 13 | |||
| 14 | #include "common/common_types.h" | ||
| 15 | #include "common/div_ceil.h" | ||
| 16 | #include "shader_recompiler/backend/spirv/emit_context.h" | ||
| 17 | #include "shader_recompiler/backend/spirv/emit_spirv.h" | ||
| 18 | |||
| 19 | namespace Shader::Backend::SPIRV { | ||
| 20 | namespace { | ||
| 21 | constexpr size_t NUM_FIXEDFNCTEXTURE = 10; | ||
| 22 | |||
| 23 | enum class Operation { | ||
| 24 | Increment, | ||
| 25 | Decrement, | ||
| 26 | FPAdd, | ||
| 27 | FPMin, | ||
| 28 | FPMax, | ||
| 29 | }; | ||
| 30 | |||
| 31 | struct AttrInfo { | ||
| 32 | Id pointer; | ||
| 33 | Id id; | ||
| 34 | bool needs_cast; | ||
| 35 | }; | ||
| 36 | |||
| 37 | Id ImageType(EmitContext& ctx, const TextureDescriptor& desc) { | ||
| 38 | const spv::ImageFormat format{spv::ImageFormat::Unknown}; | ||
| 39 | const Id type{ctx.F32[1]}; | ||
| 40 | const bool depth{desc.is_depth}; | ||
| 41 | switch (desc.type) { | ||
| 42 | case TextureType::Color1D: | ||
| 43 | return ctx.TypeImage(type, spv::Dim::Dim1D, depth, false, false, 1, format); | ||
| 44 | case TextureType::ColorArray1D: | ||
| 45 | return ctx.TypeImage(type, spv::Dim::Dim1D, depth, true, false, 1, format); | ||
| 46 | case TextureType::Color2D: | ||
| 47 | return ctx.TypeImage(type, spv::Dim::Dim2D, depth, false, false, 1, format); | ||
| 48 | case TextureType::ColorArray2D: | ||
| 49 | return ctx.TypeImage(type, spv::Dim::Dim2D, depth, true, false, 1, format); | ||
| 50 | case TextureType::Color3D: | ||
| 51 | return ctx.TypeImage(type, spv::Dim::Dim3D, depth, false, false, 1, format); | ||
| 52 | case TextureType::ColorCube: | ||
| 53 | return ctx.TypeImage(type, spv::Dim::Cube, depth, false, false, 1, format); | ||
| 54 | case TextureType::ColorArrayCube: | ||
| 55 | return ctx.TypeImage(type, spv::Dim::Cube, depth, true, false, 1, format); | ||
| 56 | case TextureType::Buffer: | ||
| 57 | break; | ||
| 58 | } | ||
| 59 | throw InvalidArgument("Invalid texture type {}", desc.type); | ||
| 60 | } | ||
| 61 | |||
| 62 | spv::ImageFormat GetImageFormat(ImageFormat format) { | ||
| 63 | switch (format) { | ||
| 64 | case ImageFormat::Typeless: | ||
| 65 | return spv::ImageFormat::Unknown; | ||
| 66 | case ImageFormat::R8_UINT: | ||
| 67 | return spv::ImageFormat::R8ui; | ||
| 68 | case ImageFormat::R8_SINT: | ||
| 69 | return spv::ImageFormat::R8i; | ||
| 70 | case ImageFormat::R16_UINT: | ||
| 71 | return spv::ImageFormat::R16ui; | ||
| 72 | case ImageFormat::R16_SINT: | ||
| 73 | return spv::ImageFormat::R16i; | ||
| 74 | case ImageFormat::R32_UINT: | ||
| 75 | return spv::ImageFormat::R32ui; | ||
| 76 | case ImageFormat::R32G32_UINT: | ||
| 77 | return spv::ImageFormat::Rg32ui; | ||
| 78 | case ImageFormat::R32G32B32A32_UINT: | ||
| 79 | return spv::ImageFormat::Rgba32ui; | ||
| 80 | } | ||
| 81 | throw InvalidArgument("Invalid image format {}", format); | ||
| 82 | } | ||
| 83 | |||
| 84 | Id ImageType(EmitContext& ctx, const ImageDescriptor& desc) { | ||
| 85 | const spv::ImageFormat format{GetImageFormat(desc.format)}; | ||
| 86 | const Id type{ctx.U32[1]}; | ||
| 87 | switch (desc.type) { | ||
| 88 | case TextureType::Color1D: | ||
| 89 | return ctx.TypeImage(type, spv::Dim::Dim1D, false, false, false, 2, format); | ||
| 90 | case TextureType::ColorArray1D: | ||
| 91 | return ctx.TypeImage(type, spv::Dim::Dim1D, false, true, false, 2, format); | ||
| 92 | case TextureType::Color2D: | ||
| 93 | return ctx.TypeImage(type, spv::Dim::Dim2D, false, false, false, 2, format); | ||
| 94 | case TextureType::ColorArray2D: | ||
| 95 | return ctx.TypeImage(type, spv::Dim::Dim2D, false, true, false, 2, format); | ||
| 96 | case TextureType::Color3D: | ||
| 97 | return ctx.TypeImage(type, spv::Dim::Dim3D, false, false, false, 2, format); | ||
| 98 | case TextureType::Buffer: | ||
| 99 | throw NotImplementedException("Image buffer"); | ||
| 100 | default: | ||
| 101 | break; | ||
| 102 | } | ||
| 103 | throw InvalidArgument("Invalid texture type {}", desc.type); | ||
| 104 | } | ||
| 105 | |||
| 106 | Id DefineVariable(EmitContext& ctx, Id type, std::optional<spv::BuiltIn> builtin, | ||
| 107 | spv::StorageClass storage_class) { | ||
| 108 | const Id pointer_type{ctx.TypePointer(storage_class, type)}; | ||
| 109 | const Id id{ctx.AddGlobalVariable(pointer_type, storage_class)}; | ||
| 110 | if (builtin) { | ||
| 111 | ctx.Decorate(id, spv::Decoration::BuiltIn, *builtin); | ||
| 112 | } | ||
| 113 | ctx.interfaces.push_back(id); | ||
| 114 | return id; | ||
| 115 | } | ||
| 116 | |||
| 117 | u32 NumVertices(InputTopology input_topology) { | ||
| 118 | switch (input_topology) { | ||
| 119 | case InputTopology::Points: | ||
| 120 | return 1; | ||
| 121 | case InputTopology::Lines: | ||
| 122 | return 2; | ||
| 123 | case InputTopology::LinesAdjacency: | ||
| 124 | return 4; | ||
| 125 | case InputTopology::Triangles: | ||
| 126 | return 3; | ||
| 127 | case InputTopology::TrianglesAdjacency: | ||
| 128 | return 6; | ||
| 129 | } | ||
| 130 | throw InvalidArgument("Invalid input topology {}", input_topology); | ||
| 131 | } | ||
| 132 | |||
| 133 | Id DefineInput(EmitContext& ctx, Id type, bool per_invocation, | ||
| 134 | std::optional<spv::BuiltIn> builtin = std::nullopt) { | ||
| 135 | switch (ctx.stage) { | ||
| 136 | case Stage::TessellationControl: | ||
| 137 | case Stage::TessellationEval: | ||
| 138 | if (per_invocation) { | ||
| 139 | type = ctx.TypeArray(type, ctx.Const(32u)); | ||
| 140 | } | ||
| 141 | break; | ||
| 142 | case Stage::Geometry: | ||
| 143 | if (per_invocation) { | ||
| 144 | const u32 num_vertices{NumVertices(ctx.runtime_info.input_topology)}; | ||
| 145 | type = ctx.TypeArray(type, ctx.Const(num_vertices)); | ||
| 146 | } | ||
| 147 | break; | ||
| 148 | default: | ||
| 149 | break; | ||
| 150 | } | ||
| 151 | return DefineVariable(ctx, type, builtin, spv::StorageClass::Input); | ||
| 152 | } | ||
| 153 | |||
| 154 | Id DefineOutput(EmitContext& ctx, Id type, std::optional<u32> invocations, | ||
| 155 | std::optional<spv::BuiltIn> builtin = std::nullopt) { | ||
| 156 | if (invocations && ctx.stage == Stage::TessellationControl) { | ||
| 157 | type = ctx.TypeArray(type, ctx.Const(*invocations)); | ||
| 158 | } | ||
| 159 | return DefineVariable(ctx, type, builtin, spv::StorageClass::Output); | ||
| 160 | } | ||
| 161 | |||
| 162 | void DefineGenericOutput(EmitContext& ctx, size_t index, std::optional<u32> invocations) { | ||
| 163 | static constexpr std::string_view swizzle{"xyzw"}; | ||
| 164 | const size_t base_attr_index{static_cast<size_t>(IR::Attribute::Generic0X) + index * 4}; | ||
| 165 | u32 element{0}; | ||
| 166 | while (element < 4) { | ||
| 167 | const u32 remainder{4 - element}; | ||
| 168 | const TransformFeedbackVarying* xfb_varying{}; | ||
| 169 | if (!ctx.runtime_info.xfb_varyings.empty()) { | ||
| 170 | xfb_varying = &ctx.runtime_info.xfb_varyings[base_attr_index + element]; | ||
| 171 | xfb_varying = xfb_varying && xfb_varying->components > 0 ? xfb_varying : nullptr; | ||
| 172 | } | ||
| 173 | const u32 num_components{xfb_varying ? xfb_varying->components : remainder}; | ||
| 174 | |||
| 175 | const Id id{DefineOutput(ctx, ctx.F32[num_components], invocations)}; | ||
| 176 | ctx.Decorate(id, spv::Decoration::Location, static_cast<u32>(index)); | ||
| 177 | if (element > 0) { | ||
| 178 | ctx.Decorate(id, spv::Decoration::Component, element); | ||
| 179 | } | ||
| 180 | if (xfb_varying) { | ||
| 181 | ctx.Decorate(id, spv::Decoration::XfbBuffer, xfb_varying->buffer); | ||
| 182 | ctx.Decorate(id, spv::Decoration::XfbStride, xfb_varying->stride); | ||
| 183 | ctx.Decorate(id, spv::Decoration::Offset, xfb_varying->offset); | ||
| 184 | } | ||
| 185 | if (num_components < 4 || element > 0) { | ||
| 186 | const std::string_view subswizzle{swizzle.substr(element, num_components)}; | ||
| 187 | ctx.Name(id, fmt::format("out_attr{}_{}", index, subswizzle)); | ||
| 188 | } else { | ||
| 189 | ctx.Name(id, fmt::format("out_attr{}", index)); | ||
| 190 | } | ||
| 191 | const GenericElementInfo info{ | ||
| 192 | .id = id, | ||
| 193 | .first_element = element, | ||
| 194 | .num_components = num_components, | ||
| 195 | }; | ||
| 196 | std::fill_n(ctx.output_generics[index].begin() + element, num_components, info); | ||
| 197 | element += num_components; | ||
| 198 | } | ||
| 199 | } | ||
| 200 | |||
| 201 | Id GetAttributeType(EmitContext& ctx, AttributeType type) { | ||
| 202 | switch (type) { | ||
| 203 | case AttributeType::Float: | ||
| 204 | return ctx.F32[4]; | ||
| 205 | case AttributeType::SignedInt: | ||
| 206 | return ctx.TypeVector(ctx.TypeInt(32, true), 4); | ||
| 207 | case AttributeType::UnsignedInt: | ||
| 208 | return ctx.U32[4]; | ||
| 209 | case AttributeType::Disabled: | ||
| 210 | break; | ||
| 211 | } | ||
| 212 | throw InvalidArgument("Invalid attribute type {}", type); | ||
| 213 | } | ||
| 214 | |||
| 215 | std::optional<AttrInfo> AttrTypes(EmitContext& ctx, u32 index) { | ||
| 216 | const AttributeType type{ctx.runtime_info.generic_input_types.at(index)}; | ||
| 217 | switch (type) { | ||
| 218 | case AttributeType::Float: | ||
| 219 | return AttrInfo{ctx.input_f32, ctx.F32[1], false}; | ||
| 220 | case AttributeType::UnsignedInt: | ||
| 221 | return AttrInfo{ctx.input_u32, ctx.U32[1], true}; | ||
| 222 | case AttributeType::SignedInt: | ||
| 223 | return AttrInfo{ctx.input_s32, ctx.TypeInt(32, true), true}; | ||
| 224 | case AttributeType::Disabled: | ||
| 225 | return std::nullopt; | ||
| 226 | } | ||
| 227 | throw InvalidArgument("Invalid attribute type {}", type); | ||
| 228 | } | ||
| 229 | |||
| 230 | std::string_view StageName(Stage stage) { | ||
| 231 | switch (stage) { | ||
| 232 | case Stage::VertexA: | ||
| 233 | return "vs_a"; | ||
| 234 | case Stage::VertexB: | ||
| 235 | return "vs"; | ||
| 236 | case Stage::TessellationControl: | ||
| 237 | return "tcs"; | ||
| 238 | case Stage::TessellationEval: | ||
| 239 | return "tes"; | ||
| 240 | case Stage::Geometry: | ||
| 241 | return "gs"; | ||
| 242 | case Stage::Fragment: | ||
| 243 | return "fs"; | ||
| 244 | case Stage::Compute: | ||
| 245 | return "cs"; | ||
| 246 | } | ||
| 247 | throw InvalidArgument("Invalid stage {}", stage); | ||
| 248 | } | ||
| 249 | |||
| 250 | template <typename... Args> | ||
| 251 | void Name(EmitContext& ctx, Id object, std::string_view format_str, Args&&... args) { | ||
| 252 | ctx.Name(object, fmt::format(fmt::runtime(format_str), StageName(ctx.stage), | ||
| 253 | std::forward<Args>(args)...) | ||
| 254 | .c_str()); | ||
| 255 | } | ||
| 256 | |||
| 257 | void DefineConstBuffers(EmitContext& ctx, const Info& info, Id UniformDefinitions::*member_type, | ||
| 258 | u32 binding, Id type, char type_char, u32 element_size) { | ||
| 259 | const Id array_type{ctx.TypeArray(type, ctx.Const(65536U / element_size))}; | ||
| 260 | ctx.Decorate(array_type, spv::Decoration::ArrayStride, element_size); | ||
| 261 | |||
| 262 | const Id struct_type{ctx.TypeStruct(array_type)}; | ||
| 263 | Name(ctx, struct_type, "{}_cbuf_block_{}{}", ctx.stage, type_char, element_size * CHAR_BIT); | ||
| 264 | ctx.Decorate(struct_type, spv::Decoration::Block); | ||
| 265 | ctx.MemberName(struct_type, 0, "data"); | ||
| 266 | ctx.MemberDecorate(struct_type, 0, spv::Decoration::Offset, 0U); | ||
| 267 | |||
| 268 | const Id struct_pointer_type{ctx.TypePointer(spv::StorageClass::Uniform, struct_type)}; | ||
| 269 | const Id uniform_type{ctx.TypePointer(spv::StorageClass::Uniform, type)}; | ||
| 270 | ctx.uniform_types.*member_type = uniform_type; | ||
| 271 | |||
| 272 | for (const ConstantBufferDescriptor& desc : info.constant_buffer_descriptors) { | ||
| 273 | const Id id{ctx.AddGlobalVariable(struct_pointer_type, spv::StorageClass::Uniform)}; | ||
| 274 | ctx.Decorate(id, spv::Decoration::Binding, binding); | ||
| 275 | ctx.Decorate(id, spv::Decoration::DescriptorSet, 0U); | ||
| 276 | ctx.Name(id, fmt::format("c{}", desc.index)); | ||
| 277 | for (size_t i = 0; i < desc.count; ++i) { | ||
| 278 | ctx.cbufs[desc.index + i].*member_type = id; | ||
| 279 | } | ||
| 280 | if (ctx.profile.supported_spirv >= 0x00010400) { | ||
| 281 | ctx.interfaces.push_back(id); | ||
| 282 | } | ||
| 283 | binding += desc.count; | ||
| 284 | } | ||
| 285 | } | ||
| 286 | |||
| 287 | void DefineSsbos(EmitContext& ctx, StorageTypeDefinition& type_def, | ||
| 288 | Id StorageDefinitions::*member_type, const Info& info, u32 binding, Id type, | ||
| 289 | u32 stride) { | ||
| 290 | const Id array_type{ctx.TypeRuntimeArray(type)}; | ||
| 291 | ctx.Decorate(array_type, spv::Decoration::ArrayStride, stride); | ||
| 292 | |||
| 293 | const Id struct_type{ctx.TypeStruct(array_type)}; | ||
| 294 | ctx.Decorate(struct_type, spv::Decoration::Block); | ||
| 295 | ctx.MemberDecorate(struct_type, 0, spv::Decoration::Offset, 0U); | ||
| 296 | |||
| 297 | const Id struct_pointer{ctx.TypePointer(spv::StorageClass::StorageBuffer, struct_type)}; | ||
| 298 | type_def.array = struct_pointer; | ||
| 299 | type_def.element = ctx.TypePointer(spv::StorageClass::StorageBuffer, type); | ||
| 300 | |||
| 301 | u32 index{}; | ||
| 302 | for (const StorageBufferDescriptor& desc : info.storage_buffers_descriptors) { | ||
| 303 | const Id id{ctx.AddGlobalVariable(struct_pointer, spv::StorageClass::StorageBuffer)}; | ||
| 304 | ctx.Decorate(id, spv::Decoration::Binding, binding); | ||
| 305 | ctx.Decorate(id, spv::Decoration::DescriptorSet, 0U); | ||
| 306 | ctx.Name(id, fmt::format("ssbo{}", index)); | ||
| 307 | if (ctx.profile.supported_spirv >= 0x00010400) { | ||
| 308 | ctx.interfaces.push_back(id); | ||
| 309 | } | ||
| 310 | for (size_t i = 0; i < desc.count; ++i) { | ||
| 311 | ctx.ssbos[index + i].*member_type = id; | ||
| 312 | } | ||
| 313 | index += desc.count; | ||
| 314 | binding += desc.count; | ||
| 315 | } | ||
| 316 | } | ||
| 317 | |||
| 318 | Id CasFunction(EmitContext& ctx, Operation operation, Id value_type) { | ||
| 319 | const Id func_type{ctx.TypeFunction(value_type, value_type, value_type)}; | ||
| 320 | const Id func{ctx.OpFunction(value_type, spv::FunctionControlMask::MaskNone, func_type)}; | ||
| 321 | const Id op_a{ctx.OpFunctionParameter(value_type)}; | ||
| 322 | const Id op_b{ctx.OpFunctionParameter(value_type)}; | ||
| 323 | ctx.AddLabel(); | ||
| 324 | Id result{}; | ||
| 325 | switch (operation) { | ||
| 326 | case Operation::Increment: { | ||
| 327 | const Id pred{ctx.OpUGreaterThanEqual(ctx.U1, op_a, op_b)}; | ||
| 328 | const Id incr{ctx.OpIAdd(value_type, op_a, ctx.Constant(value_type, 1))}; | ||
| 329 | result = ctx.OpSelect(value_type, pred, ctx.u32_zero_value, incr); | ||
| 330 | break; | ||
| 331 | } | ||
| 332 | case Operation::Decrement: { | ||
| 333 | const Id lhs{ctx.OpIEqual(ctx.U1, op_a, ctx.Constant(value_type, 0u))}; | ||
| 334 | const Id rhs{ctx.OpUGreaterThan(ctx.U1, op_a, op_b)}; | ||
| 335 | const Id pred{ctx.OpLogicalOr(ctx.U1, lhs, rhs)}; | ||
| 336 | const Id decr{ctx.OpISub(value_type, op_a, ctx.Constant(value_type, 1))}; | ||
| 337 | result = ctx.OpSelect(value_type, pred, op_b, decr); | ||
| 338 | break; | ||
| 339 | } | ||
| 340 | case Operation::FPAdd: | ||
| 341 | result = ctx.OpFAdd(value_type, op_a, op_b); | ||
| 342 | break; | ||
| 343 | case Operation::FPMin: | ||
| 344 | result = ctx.OpFMin(value_type, op_a, op_b); | ||
| 345 | break; | ||
| 346 | case Operation::FPMax: | ||
| 347 | result = ctx.OpFMax(value_type, op_a, op_b); | ||
| 348 | break; | ||
| 349 | default: | ||
| 350 | break; | ||
| 351 | } | ||
| 352 | ctx.OpReturnValue(result); | ||
| 353 | ctx.OpFunctionEnd(); | ||
| 354 | return func; | ||
| 355 | } | ||
| 356 | |||
| 357 | Id CasLoop(EmitContext& ctx, Operation operation, Id array_pointer, Id element_pointer, | ||
| 358 | Id value_type, Id memory_type, spv::Scope scope) { | ||
| 359 | const bool is_shared{scope == spv::Scope::Workgroup}; | ||
| 360 | const bool is_struct{!is_shared || ctx.profile.support_explicit_workgroup_layout}; | ||
| 361 | const Id cas_func{CasFunction(ctx, operation, value_type)}; | ||
| 362 | const Id zero{ctx.u32_zero_value}; | ||
| 363 | const Id scope_id{ctx.Const(static_cast<u32>(scope))}; | ||
| 364 | |||
| 365 | const Id loop_header{ctx.OpLabel()}; | ||
| 366 | const Id continue_block{ctx.OpLabel()}; | ||
| 367 | const Id merge_block{ctx.OpLabel()}; | ||
| 368 | const Id func_type{is_shared | ||
| 369 | ? ctx.TypeFunction(value_type, ctx.U32[1], value_type) | ||
| 370 | : ctx.TypeFunction(value_type, ctx.U32[1], value_type, array_pointer)}; | ||
| 371 | |||
| 372 | const Id func{ctx.OpFunction(value_type, spv::FunctionControlMask::MaskNone, func_type)}; | ||
| 373 | const Id index{ctx.OpFunctionParameter(ctx.U32[1])}; | ||
| 374 | const Id op_b{ctx.OpFunctionParameter(value_type)}; | ||
| 375 | const Id base{is_shared ? ctx.shared_memory_u32 : ctx.OpFunctionParameter(array_pointer)}; | ||
| 376 | ctx.AddLabel(); | ||
| 377 | ctx.OpBranch(loop_header); | ||
| 378 | ctx.AddLabel(loop_header); | ||
| 379 | |||
| 380 | ctx.OpLoopMerge(merge_block, continue_block, spv::LoopControlMask::MaskNone); | ||
| 381 | ctx.OpBranch(continue_block); | ||
| 382 | |||
| 383 | ctx.AddLabel(continue_block); | ||
| 384 | const Id word_pointer{is_struct ? ctx.OpAccessChain(element_pointer, base, zero, index) | ||
| 385 | : ctx.OpAccessChain(element_pointer, base, index)}; | ||
| 386 | if (value_type.value == ctx.F32[2].value) { | ||
| 387 | const Id u32_value{ctx.OpLoad(ctx.U32[1], word_pointer)}; | ||
| 388 | const Id value{ctx.OpUnpackHalf2x16(ctx.F32[2], u32_value)}; | ||
| 389 | const Id new_value{ctx.OpFunctionCall(value_type, cas_func, value, op_b)}; | ||
| 390 | const Id u32_new_value{ctx.OpPackHalf2x16(ctx.U32[1], new_value)}; | ||
| 391 | const Id atomic_res{ctx.OpAtomicCompareExchange(ctx.U32[1], word_pointer, scope_id, zero, | ||
| 392 | zero, u32_new_value, u32_value)}; | ||
| 393 | const Id success{ctx.OpIEqual(ctx.U1, atomic_res, u32_value)}; | ||
| 394 | ctx.OpBranchConditional(success, merge_block, loop_header); | ||
| 395 | |||
| 396 | ctx.AddLabel(merge_block); | ||
| 397 | ctx.OpReturnValue(ctx.OpUnpackHalf2x16(ctx.F32[2], atomic_res)); | ||
| 398 | } else { | ||
| 399 | const Id value{ctx.OpLoad(memory_type, word_pointer)}; | ||
| 400 | const bool matching_type{value_type.value == memory_type.value}; | ||
| 401 | const Id bitcast_value{matching_type ? value : ctx.OpBitcast(value_type, value)}; | ||
| 402 | const Id cal_res{ctx.OpFunctionCall(value_type, cas_func, bitcast_value, op_b)}; | ||
| 403 | const Id new_value{matching_type ? cal_res : ctx.OpBitcast(memory_type, cal_res)}; | ||
| 404 | const Id atomic_res{ctx.OpAtomicCompareExchange(ctx.U32[1], word_pointer, scope_id, zero, | ||
| 405 | zero, new_value, value)}; | ||
| 406 | const Id success{ctx.OpIEqual(ctx.U1, atomic_res, value)}; | ||
| 407 | ctx.OpBranchConditional(success, merge_block, loop_header); | ||
| 408 | |||
| 409 | ctx.AddLabel(merge_block); | ||
| 410 | ctx.OpReturnValue(ctx.OpBitcast(value_type, atomic_res)); | ||
| 411 | } | ||
| 412 | ctx.OpFunctionEnd(); | ||
| 413 | return func; | ||
| 414 | } | ||
| 415 | |||
| 416 | template <typename Desc> | ||
| 417 | std::string NameOf(Stage stage, const Desc& desc, std::string_view prefix) { | ||
| 418 | if (desc.count > 1) { | ||
| 419 | return fmt::format("{}_{}{}_{:02x}x{}", StageName(stage), prefix, desc.cbuf_index, | ||
| 420 | desc.cbuf_offset, desc.count); | ||
| 421 | } else { | ||
| 422 | return fmt::format("{}_{}{}_{:02x}", StageName(stage), prefix, desc.cbuf_index, | ||
| 423 | desc.cbuf_offset); | ||
| 424 | } | ||
| 425 | } | ||
| 426 | |||
| 427 | Id DescType(EmitContext& ctx, Id sampled_type, Id pointer_type, u32 count) { | ||
| 428 | if (count > 1) { | ||
| 429 | const Id array_type{ctx.TypeArray(sampled_type, ctx.Const(count))}; | ||
| 430 | return ctx.TypePointer(spv::StorageClass::UniformConstant, array_type); | ||
| 431 | } else { | ||
| 432 | return pointer_type; | ||
| 433 | } | ||
| 434 | } | ||
| 435 | |||
| 436 | size_t FindAndSetNextUnusedLocation(std::bitset<IR::NUM_GENERICS>& used_locations, | ||
| 437 | size_t& start_offset) { | ||
| 438 | for (size_t location = start_offset; location < used_locations.size(); ++location) { | ||
| 439 | if (!used_locations.test(location)) { | ||
| 440 | start_offset = location; | ||
| 441 | used_locations.set(location); | ||
| 442 | return location; | ||
| 443 | } | ||
| 444 | } | ||
| 445 | throw RuntimeError("Unable to get an unused location for legacy attribute"); | ||
| 446 | } | ||
| 447 | |||
| 448 | Id DefineLegacyInput(EmitContext& ctx, std::bitset<IR::NUM_GENERICS>& used_locations, | ||
| 449 | size_t& start_offset) { | ||
| 450 | const Id id{DefineInput(ctx, ctx.F32[4], true)}; | ||
| 451 | const size_t location = FindAndSetNextUnusedLocation(used_locations, start_offset); | ||
| 452 | ctx.Decorate(id, spv::Decoration::Location, location); | ||
| 453 | return id; | ||
| 454 | } | ||
| 455 | |||
| 456 | Id DefineLegacyOutput(EmitContext& ctx, std::bitset<IR::NUM_GENERICS>& used_locations, | ||
| 457 | size_t& start_offset, std::optional<u32> invocations) { | ||
| 458 | const Id id{DefineOutput(ctx, ctx.F32[4], invocations)}; | ||
| 459 | const size_t location = FindAndSetNextUnusedLocation(used_locations, start_offset); | ||
| 460 | ctx.Decorate(id, spv::Decoration::Location, location); | ||
| 461 | return id; | ||
| 462 | } | ||
| 463 | } // Anonymous namespace | ||
| 464 | |||
| 465 | void VectorTypes::Define(Sirit::Module& sirit_ctx, Id base_type, std::string_view name) { | ||
| 466 | defs[0] = sirit_ctx.Name(base_type, name); | ||
| 467 | |||
| 468 | std::array<char, 6> def_name; | ||
| 469 | for (int i = 1; i < 4; ++i) { | ||
| 470 | const std::string_view def_name_view( | ||
| 471 | def_name.data(), | ||
| 472 | fmt::format_to_n(def_name.data(), def_name.size(), "{}x{}", name, i + 1).size); | ||
| 473 | defs[static_cast<size_t>(i)] = | ||
| 474 | sirit_ctx.Name(sirit_ctx.TypeVector(base_type, i + 1), def_name_view); | ||
| 475 | } | ||
| 476 | } | ||
| 477 | |||
| 478 | EmitContext::EmitContext(const Profile& profile_, const RuntimeInfo& runtime_info_, | ||
| 479 | IR::Program& program, Bindings& bindings) | ||
| 480 | : Sirit::Module(profile_.supported_spirv), profile{profile_}, runtime_info{runtime_info_}, | ||
| 481 | stage{program.stage}, texture_rescaling_index{bindings.texture_scaling_index}, | ||
| 482 | image_rescaling_index{bindings.image_scaling_index} { | ||
| 483 | const bool is_unified{profile.unified_descriptor_binding}; | ||
| 484 | u32& uniform_binding{is_unified ? bindings.unified : bindings.uniform_buffer}; | ||
| 485 | u32& storage_binding{is_unified ? bindings.unified : bindings.storage_buffer}; | ||
| 486 | u32& texture_binding{is_unified ? bindings.unified : bindings.texture}; | ||
| 487 | u32& image_binding{is_unified ? bindings.unified : bindings.image}; | ||
| 488 | AddCapability(spv::Capability::Shader); | ||
| 489 | DefineCommonTypes(program.info); | ||
| 490 | DefineCommonConstants(); | ||
| 491 | DefineInterfaces(program); | ||
| 492 | DefineLocalMemory(program); | ||
| 493 | DefineSharedMemory(program); | ||
| 494 | DefineSharedMemoryFunctions(program); | ||
| 495 | DefineConstantBuffers(program.info, uniform_binding); | ||
| 496 | DefineStorageBuffers(program.info, storage_binding); | ||
| 497 | DefineTextureBuffers(program.info, texture_binding); | ||
| 498 | DefineImageBuffers(program.info, image_binding); | ||
| 499 | DefineTextures(program.info, texture_binding, bindings.texture_scaling_index); | ||
| 500 | DefineImages(program.info, image_binding, bindings.image_scaling_index); | ||
| 501 | DefineAttributeMemAccess(program.info); | ||
| 502 | DefineGlobalMemoryFunctions(program.info); | ||
| 503 | DefineRescalingInput(program.info); | ||
| 504 | } | ||
| 505 | |||
| 506 | EmitContext::~EmitContext() = default; | ||
| 507 | |||
| 508 | Id EmitContext::Def(const IR::Value& value) { | ||
| 509 | if (!value.IsImmediate()) { | ||
| 510 | return value.InstRecursive()->Definition<Id>(); | ||
| 511 | } | ||
| 512 | switch (value.Type()) { | ||
| 513 | case IR::Type::Void: | ||
| 514 | // Void instructions are used for optional arguments (e.g. texture offsets) | ||
| 515 | // They are not meant to be used in the SPIR-V module | ||
| 516 | return Id{}; | ||
| 517 | case IR::Type::U1: | ||
| 518 | return value.U1() ? true_value : false_value; | ||
| 519 | case IR::Type::U32: | ||
| 520 | return Const(value.U32()); | ||
| 521 | case IR::Type::U64: | ||
| 522 | return Constant(U64, value.U64()); | ||
| 523 | case IR::Type::F32: | ||
| 524 | return Const(value.F32()); | ||
| 525 | case IR::Type::F64: | ||
| 526 | return Constant(F64[1], value.F64()); | ||
| 527 | default: | ||
| 528 | throw NotImplementedException("Immediate type {}", value.Type()); | ||
| 529 | } | ||
| 530 | } | ||
| 531 | |||
| 532 | Id EmitContext::BitOffset8(const IR::Value& offset) { | ||
| 533 | if (offset.IsImmediate()) { | ||
| 534 | return Const((offset.U32() % 4) * 8); | ||
| 535 | } | ||
| 536 | return OpBitwiseAnd(U32[1], OpShiftLeftLogical(U32[1], Def(offset), Const(3u)), Const(24u)); | ||
| 537 | } | ||
| 538 | |||
| 539 | Id EmitContext::BitOffset16(const IR::Value& offset) { | ||
| 540 | if (offset.IsImmediate()) { | ||
| 541 | return Const(((offset.U32() / 2) % 2) * 16); | ||
| 542 | } | ||
| 543 | return OpBitwiseAnd(U32[1], OpShiftLeftLogical(U32[1], Def(offset), Const(3u)), Const(16u)); | ||
| 544 | } | ||
| 545 | |||
| 546 | Id EmitContext::InputLegacyAttribute(IR::Attribute attribute) { | ||
| 547 | if (attribute >= IR::Attribute::ColorFrontDiffuseR && | ||
| 548 | attribute <= IR::Attribute::ColorFrontDiffuseA) { | ||
| 549 | return input_front_color; | ||
| 550 | } | ||
| 551 | if (attribute >= IR::Attribute::ColorFrontSpecularR && | ||
| 552 | attribute <= IR::Attribute::ColorFrontSpecularA) { | ||
| 553 | return input_front_secondary_color; | ||
| 554 | } | ||
| 555 | if (attribute >= IR::Attribute::ColorBackDiffuseR && | ||
| 556 | attribute <= IR::Attribute::ColorBackDiffuseA) { | ||
| 557 | return input_back_color; | ||
| 558 | } | ||
| 559 | if (attribute >= IR::Attribute::ColorBackSpecularR && | ||
| 560 | attribute <= IR::Attribute::ColorBackSpecularA) { | ||
| 561 | return input_back_secondary_color; | ||
| 562 | } | ||
| 563 | if (attribute == IR::Attribute::FogCoordinate) { | ||
| 564 | return input_fog_frag_coord; | ||
| 565 | } | ||
| 566 | if (attribute >= IR::Attribute::FixedFncTexture0S && | ||
| 567 | attribute <= IR::Attribute::FixedFncTexture9Q) { | ||
| 568 | u32 index = | ||
| 569 | (static_cast<u32>(attribute) - static_cast<u32>(IR::Attribute::FixedFncTexture0S)) / 4; | ||
| 570 | return input_fixed_fnc_textures[index]; | ||
| 571 | } | ||
| 572 | throw InvalidArgument("Attribute is not legacy attribute {}", attribute); | ||
| 573 | } | ||
| 574 | |||
| 575 | Id EmitContext::OutputLegacyAttribute(IR::Attribute attribute) { | ||
| 576 | if (attribute >= IR::Attribute::ColorFrontDiffuseR && | ||
| 577 | attribute <= IR::Attribute::ColorFrontDiffuseA) { | ||
| 578 | return output_front_color; | ||
| 579 | } | ||
| 580 | if (attribute >= IR::Attribute::ColorFrontSpecularR && | ||
| 581 | attribute <= IR::Attribute::ColorFrontSpecularA) { | ||
| 582 | return output_front_secondary_color; | ||
| 583 | } | ||
| 584 | if (attribute >= IR::Attribute::ColorBackDiffuseR && | ||
| 585 | attribute <= IR::Attribute::ColorBackDiffuseA) { | ||
| 586 | return output_back_color; | ||
| 587 | } | ||
| 588 | if (attribute >= IR::Attribute::ColorBackSpecularR && | ||
| 589 | attribute <= IR::Attribute::ColorBackSpecularA) { | ||
| 590 | return output_back_secondary_color; | ||
| 591 | } | ||
| 592 | if (attribute == IR::Attribute::FogCoordinate) { | ||
| 593 | return output_fog_frag_coord; | ||
| 594 | } | ||
| 595 | if (attribute >= IR::Attribute::FixedFncTexture0S && | ||
| 596 | attribute <= IR::Attribute::FixedFncTexture9Q) { | ||
| 597 | u32 index = | ||
| 598 | (static_cast<u32>(attribute) - static_cast<u32>(IR::Attribute::FixedFncTexture0S)) / 4; | ||
| 599 | return output_fixed_fnc_textures[index]; | ||
| 600 | } | ||
| 601 | throw InvalidArgument("Attribute is not legacy attribute {}", attribute); | ||
| 602 | } | ||
| 603 | |||
| 604 | void EmitContext::DefineCommonTypes(const Info& info) { | ||
| 605 | void_id = TypeVoid(); | ||
| 606 | |||
| 607 | U1 = Name(TypeBool(), "u1"); | ||
| 608 | |||
| 609 | F32.Define(*this, TypeFloat(32), "f32"); | ||
| 610 | U32.Define(*this, TypeInt(32, false), "u32"); | ||
| 611 | S32.Define(*this, TypeInt(32, true), "s32"); | ||
| 612 | |||
| 613 | private_u32 = Name(TypePointer(spv::StorageClass::Private, U32[1]), "private_u32"); | ||
| 614 | |||
| 615 | input_f32 = Name(TypePointer(spv::StorageClass::Input, F32[1]), "input_f32"); | ||
| 616 | input_u32 = Name(TypePointer(spv::StorageClass::Input, U32[1]), "input_u32"); | ||
| 617 | input_s32 = Name(TypePointer(spv::StorageClass::Input, TypeInt(32, true)), "input_s32"); | ||
| 618 | |||
| 619 | output_f32 = Name(TypePointer(spv::StorageClass::Output, F32[1]), "output_f32"); | ||
| 620 | output_u32 = Name(TypePointer(spv::StorageClass::Output, U32[1]), "output_u32"); | ||
| 621 | |||
| 622 | if (info.uses_int8 && profile.support_int8) { | ||
| 623 | AddCapability(spv::Capability::Int8); | ||
| 624 | U8 = Name(TypeInt(8, false), "u8"); | ||
| 625 | S8 = Name(TypeInt(8, true), "s8"); | ||
| 626 | } | ||
| 627 | if (info.uses_int16 && profile.support_int16) { | ||
| 628 | AddCapability(spv::Capability::Int16); | ||
| 629 | U16 = Name(TypeInt(16, false), "u16"); | ||
| 630 | S16 = Name(TypeInt(16, true), "s16"); | ||
| 631 | } | ||
| 632 | if (info.uses_int64) { | ||
| 633 | AddCapability(spv::Capability::Int64); | ||
| 634 | U64 = Name(TypeInt(64, false), "u64"); | ||
| 635 | } | ||
| 636 | if (info.uses_fp16) { | ||
| 637 | AddCapability(spv::Capability::Float16); | ||
| 638 | F16.Define(*this, TypeFloat(16), "f16"); | ||
| 639 | } | ||
| 640 | if (info.uses_fp64) { | ||
| 641 | AddCapability(spv::Capability::Float64); | ||
| 642 | F64.Define(*this, TypeFloat(64), "f64"); | ||
| 643 | } | ||
| 644 | } | ||
| 645 | |||
| 646 | void EmitContext::DefineCommonConstants() { | ||
| 647 | true_value = ConstantTrue(U1); | ||
| 648 | false_value = ConstantFalse(U1); | ||
| 649 | u32_zero_value = Const(0U); | ||
| 650 | f32_zero_value = Const(0.0f); | ||
| 651 | } | ||
| 652 | |||
| 653 | void EmitContext::DefineInterfaces(const IR::Program& program) { | ||
| 654 | DefineInputs(program); | ||
| 655 | DefineOutputs(program); | ||
| 656 | } | ||
| 657 | |||
| 658 | void EmitContext::DefineLocalMemory(const IR::Program& program) { | ||
| 659 | if (program.local_memory_size == 0) { | ||
| 660 | return; | ||
| 661 | } | ||
| 662 | const u32 num_elements{Common::DivCeil(program.local_memory_size, 4U)}; | ||
| 663 | const Id type{TypeArray(U32[1], Const(num_elements))}; | ||
| 664 | const Id pointer{TypePointer(spv::StorageClass::Private, type)}; | ||
| 665 | local_memory = AddGlobalVariable(pointer, spv::StorageClass::Private); | ||
| 666 | if (profile.supported_spirv >= 0x00010400) { | ||
| 667 | interfaces.push_back(local_memory); | ||
| 668 | } | ||
| 669 | } | ||
| 670 | |||
| 671 | void EmitContext::DefineSharedMemory(const IR::Program& program) { | ||
| 672 | if (program.shared_memory_size == 0) { | ||
| 673 | return; | ||
| 674 | } | ||
| 675 | const auto make{[&](Id element_type, u32 element_size) { | ||
| 676 | const u32 num_elements{Common::DivCeil(program.shared_memory_size, element_size)}; | ||
| 677 | const Id array_type{TypeArray(element_type, Const(num_elements))}; | ||
| 678 | Decorate(array_type, spv::Decoration::ArrayStride, element_size); | ||
| 679 | |||
| 680 | const Id struct_type{TypeStruct(array_type)}; | ||
| 681 | MemberDecorate(struct_type, 0U, spv::Decoration::Offset, 0U); | ||
| 682 | Decorate(struct_type, spv::Decoration::Block); | ||
| 683 | |||
| 684 | const Id pointer{TypePointer(spv::StorageClass::Workgroup, struct_type)}; | ||
| 685 | const Id element_pointer{TypePointer(spv::StorageClass::Workgroup, element_type)}; | ||
| 686 | const Id variable{AddGlobalVariable(pointer, spv::StorageClass::Workgroup)}; | ||
| 687 | Decorate(variable, spv::Decoration::Aliased); | ||
| 688 | interfaces.push_back(variable); | ||
| 689 | |||
| 690 | return std::make_tuple(variable, element_pointer, pointer); | ||
| 691 | }}; | ||
| 692 | if (profile.support_explicit_workgroup_layout) { | ||
| 693 | AddExtension("SPV_KHR_workgroup_memory_explicit_layout"); | ||
| 694 | AddCapability(spv::Capability::WorkgroupMemoryExplicitLayoutKHR); | ||
| 695 | if (program.info.uses_int8) { | ||
| 696 | AddCapability(spv::Capability::WorkgroupMemoryExplicitLayout8BitAccessKHR); | ||
| 697 | std::tie(shared_memory_u8, shared_u8, std::ignore) = make(U8, 1); | ||
| 698 | } | ||
| 699 | if (program.info.uses_int16) { | ||
| 700 | AddCapability(spv::Capability::WorkgroupMemoryExplicitLayout16BitAccessKHR); | ||
| 701 | std::tie(shared_memory_u16, shared_u16, std::ignore) = make(U16, 2); | ||
| 702 | } | ||
| 703 | if (program.info.uses_int64) { | ||
| 704 | std::tie(shared_memory_u64, shared_u64, std::ignore) = make(U64, 8); | ||
| 705 | } | ||
| 706 | std::tie(shared_memory_u32, shared_u32, shared_memory_u32_type) = make(U32[1], 4); | ||
| 707 | std::tie(shared_memory_u32x2, shared_u32x2, std::ignore) = make(U32[2], 8); | ||
| 708 | std::tie(shared_memory_u32x4, shared_u32x4, std::ignore) = make(U32[4], 16); | ||
| 709 | return; | ||
| 710 | } | ||
| 711 | const u32 num_elements{Common::DivCeil(program.shared_memory_size, 4U)}; | ||
| 712 | const Id type{TypeArray(U32[1], Const(num_elements))}; | ||
| 713 | shared_memory_u32_type = TypePointer(spv::StorageClass::Workgroup, type); | ||
| 714 | |||
| 715 | shared_u32 = TypePointer(spv::StorageClass::Workgroup, U32[1]); | ||
| 716 | shared_memory_u32 = AddGlobalVariable(shared_memory_u32_type, spv::StorageClass::Workgroup); | ||
| 717 | interfaces.push_back(shared_memory_u32); | ||
| 718 | |||
| 719 | const Id func_type{TypeFunction(void_id, U32[1], U32[1])}; | ||
| 720 | const auto make_function{[&](u32 mask, u32 size) { | ||
| 721 | const Id loop_header{OpLabel()}; | ||
| 722 | const Id continue_block{OpLabel()}; | ||
| 723 | const Id merge_block{OpLabel()}; | ||
| 724 | |||
| 725 | const Id func{OpFunction(void_id, spv::FunctionControlMask::MaskNone, func_type)}; | ||
| 726 | const Id offset{OpFunctionParameter(U32[1])}; | ||
| 727 | const Id insert_value{OpFunctionParameter(U32[1])}; | ||
| 728 | AddLabel(); | ||
| 729 | OpBranch(loop_header); | ||
| 730 | |||
| 731 | AddLabel(loop_header); | ||
| 732 | const Id word_offset{OpShiftRightArithmetic(U32[1], offset, Const(2U))}; | ||
| 733 | const Id shift_offset{OpShiftLeftLogical(U32[1], offset, Const(3U))}; | ||
| 734 | const Id bit_offset{OpBitwiseAnd(U32[1], shift_offset, Const(mask))}; | ||
| 735 | const Id count{Const(size)}; | ||
| 736 | OpLoopMerge(merge_block, continue_block, spv::LoopControlMask::MaskNone); | ||
| 737 | OpBranch(continue_block); | ||
| 738 | |||
| 739 | AddLabel(continue_block); | ||
| 740 | const Id word_pointer{OpAccessChain(shared_u32, shared_memory_u32, word_offset)}; | ||
| 741 | const Id old_value{OpLoad(U32[1], word_pointer)}; | ||
| 742 | const Id new_value{OpBitFieldInsert(U32[1], old_value, insert_value, bit_offset, count)}; | ||
| 743 | const Id atomic_res{OpAtomicCompareExchange(U32[1], word_pointer, Const(1U), u32_zero_value, | ||
| 744 | u32_zero_value, new_value, old_value)}; | ||
| 745 | const Id success{OpIEqual(U1, atomic_res, old_value)}; | ||
| 746 | OpBranchConditional(success, merge_block, loop_header); | ||
| 747 | |||
| 748 | AddLabel(merge_block); | ||
| 749 | OpReturn(); | ||
| 750 | OpFunctionEnd(); | ||
| 751 | return func; | ||
| 752 | }}; | ||
| 753 | if (program.info.uses_int8) { | ||
| 754 | shared_store_u8_func = make_function(24, 8); | ||
| 755 | } | ||
| 756 | if (program.info.uses_int16) { | ||
| 757 | shared_store_u16_func = make_function(16, 16); | ||
| 758 | } | ||
| 759 | } | ||
| 760 | |||
| 761 | void EmitContext::DefineSharedMemoryFunctions(const IR::Program& program) { | ||
| 762 | if (program.info.uses_shared_increment) { | ||
| 763 | increment_cas_shared = CasLoop(*this, Operation::Increment, shared_memory_u32_type, | ||
| 764 | shared_u32, U32[1], U32[1], spv::Scope::Workgroup); | ||
| 765 | } | ||
| 766 | if (program.info.uses_shared_decrement) { | ||
| 767 | decrement_cas_shared = CasLoop(*this, Operation::Decrement, shared_memory_u32_type, | ||
| 768 | shared_u32, U32[1], U32[1], spv::Scope::Workgroup); | ||
| 769 | } | ||
| 770 | } | ||
| 771 | |||
| 772 | void EmitContext::DefineAttributeMemAccess(const Info& info) { | ||
| 773 | const auto make_load{[&] { | ||
| 774 | const bool is_array{stage == Stage::Geometry}; | ||
| 775 | const Id end_block{OpLabel()}; | ||
| 776 | const Id default_label{OpLabel()}; | ||
| 777 | |||
| 778 | const Id func_type_load{is_array ? TypeFunction(F32[1], U32[1], U32[1]) | ||
| 779 | : TypeFunction(F32[1], U32[1])}; | ||
| 780 | const Id func{OpFunction(F32[1], spv::FunctionControlMask::MaskNone, func_type_load)}; | ||
| 781 | const Id offset{OpFunctionParameter(U32[1])}; | ||
| 782 | const Id vertex{is_array ? OpFunctionParameter(U32[1]) : Id{}}; | ||
| 783 | |||
| 784 | AddLabel(); | ||
| 785 | const Id base_index{OpShiftRightArithmetic(U32[1], offset, Const(2U))}; | ||
| 786 | const Id masked_index{OpBitwiseAnd(U32[1], base_index, Const(3U))}; | ||
| 787 | const Id compare_index{OpShiftRightArithmetic(U32[1], base_index, Const(2U))}; | ||
| 788 | std::vector<Sirit::Literal> literals; | ||
| 789 | std::vector<Id> labels; | ||
| 790 | if (info.loads.AnyComponent(IR::Attribute::PositionX)) { | ||
| 791 | literals.push_back(static_cast<u32>(IR::Attribute::PositionX) >> 2); | ||
| 792 | labels.push_back(OpLabel()); | ||
| 793 | } | ||
| 794 | const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2; | ||
| 795 | for (u32 index = 0; index < static_cast<u32>(IR::NUM_GENERICS); ++index) { | ||
| 796 | if (!info.loads.Generic(index)) { | ||
| 797 | continue; | ||
| 798 | } | ||
| 799 | literals.push_back(base_attribute_value + index); | ||
| 800 | labels.push_back(OpLabel()); | ||
| 801 | } | ||
| 802 | OpSelectionMerge(end_block, spv::SelectionControlMask::MaskNone); | ||
| 803 | OpSwitch(compare_index, default_label, literals, labels); | ||
| 804 | AddLabel(default_label); | ||
| 805 | OpReturnValue(Const(0.0f)); | ||
| 806 | size_t label_index{0}; | ||
| 807 | if (info.loads.AnyComponent(IR::Attribute::PositionX)) { | ||
| 808 | AddLabel(labels[label_index]); | ||
| 809 | const Id pointer{is_array | ||
| 810 | ? OpAccessChain(input_f32, input_position, vertex, masked_index) | ||
| 811 | : OpAccessChain(input_f32, input_position, masked_index)}; | ||
| 812 | const Id result{OpLoad(F32[1], pointer)}; | ||
| 813 | OpReturnValue(result); | ||
| 814 | ++label_index; | ||
| 815 | } | ||
| 816 | for (size_t index = 0; index < IR::NUM_GENERICS; ++index) { | ||
| 817 | if (!info.loads.Generic(index)) { | ||
| 818 | continue; | ||
| 819 | } | ||
| 820 | AddLabel(labels[label_index]); | ||
| 821 | const auto type{AttrTypes(*this, static_cast<u32>(index))}; | ||
| 822 | if (!type) { | ||
| 823 | OpReturnValue(Const(0.0f)); | ||
| 824 | ++label_index; | ||
| 825 | continue; | ||
| 826 | } | ||
| 827 | const Id generic_id{input_generics.at(index)}; | ||
| 828 | const Id pointer{is_array | ||
| 829 | ? OpAccessChain(type->pointer, generic_id, vertex, masked_index) | ||
| 830 | : OpAccessChain(type->pointer, generic_id, masked_index)}; | ||
| 831 | const Id value{OpLoad(type->id, pointer)}; | ||
| 832 | const Id result{type->needs_cast ? OpBitcast(F32[1], value) : value}; | ||
| 833 | OpReturnValue(result); | ||
| 834 | ++label_index; | ||
| 835 | } | ||
| 836 | AddLabel(end_block); | ||
| 837 | OpUnreachable(); | ||
| 838 | OpFunctionEnd(); | ||
| 839 | return func; | ||
| 840 | }}; | ||
| 841 | const auto make_store{[&] { | ||
| 842 | const Id end_block{OpLabel()}; | ||
| 843 | const Id default_label{OpLabel()}; | ||
| 844 | |||
| 845 | const Id func_type_store{TypeFunction(void_id, U32[1], F32[1])}; | ||
| 846 | const Id func{OpFunction(void_id, spv::FunctionControlMask::MaskNone, func_type_store)}; | ||
| 847 | const Id offset{OpFunctionParameter(U32[1])}; | ||
| 848 | const Id store_value{OpFunctionParameter(F32[1])}; | ||
| 849 | AddLabel(); | ||
| 850 | const Id base_index{OpShiftRightArithmetic(U32[1], offset, Const(2U))}; | ||
| 851 | const Id masked_index{OpBitwiseAnd(U32[1], base_index, Const(3U))}; | ||
| 852 | const Id compare_index{OpShiftRightArithmetic(U32[1], base_index, Const(2U))}; | ||
| 853 | std::vector<Sirit::Literal> literals; | ||
| 854 | std::vector<Id> labels; | ||
| 855 | if (info.stores.AnyComponent(IR::Attribute::PositionX)) { | ||
| 856 | literals.push_back(static_cast<u32>(IR::Attribute::PositionX) >> 2); | ||
| 857 | labels.push_back(OpLabel()); | ||
| 858 | } | ||
| 859 | const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2; | ||
| 860 | for (size_t index = 0; index < IR::NUM_GENERICS; ++index) { | ||
| 861 | if (!info.stores.Generic(index)) { | ||
| 862 | continue; | ||
| 863 | } | ||
| 864 | literals.push_back(base_attribute_value + static_cast<u32>(index)); | ||
| 865 | labels.push_back(OpLabel()); | ||
| 866 | } | ||
| 867 | if (info.stores.ClipDistances()) { | ||
| 868 | literals.push_back(static_cast<u32>(IR::Attribute::ClipDistance0) >> 2); | ||
| 869 | labels.push_back(OpLabel()); | ||
| 870 | literals.push_back(static_cast<u32>(IR::Attribute::ClipDistance4) >> 2); | ||
| 871 | labels.push_back(OpLabel()); | ||
| 872 | } | ||
| 873 | OpSelectionMerge(end_block, spv::SelectionControlMask::MaskNone); | ||
| 874 | OpSwitch(compare_index, default_label, literals, labels); | ||
| 875 | AddLabel(default_label); | ||
| 876 | OpReturn(); | ||
| 877 | size_t label_index{0}; | ||
| 878 | if (info.stores.AnyComponent(IR::Attribute::PositionX)) { | ||
| 879 | AddLabel(labels[label_index]); | ||
| 880 | const Id pointer{OpAccessChain(output_f32, output_position, masked_index)}; | ||
| 881 | OpStore(pointer, store_value); | ||
| 882 | OpReturn(); | ||
| 883 | ++label_index; | ||
| 884 | } | ||
| 885 | for (size_t index = 0; index < IR::NUM_GENERICS; ++index) { | ||
| 886 | if (!info.stores.Generic(index)) { | ||
| 887 | continue; | ||
| 888 | } | ||
| 889 | if (output_generics[index][0].num_components != 4) { | ||
| 890 | throw NotImplementedException("Physical stores and transform feedbacks"); | ||
| 891 | } | ||
| 892 | AddLabel(labels[label_index]); | ||
| 893 | const Id generic_id{output_generics[index][0].id}; | ||
| 894 | const Id pointer{OpAccessChain(output_f32, generic_id, masked_index)}; | ||
| 895 | OpStore(pointer, store_value); | ||
| 896 | OpReturn(); | ||
| 897 | ++label_index; | ||
| 898 | } | ||
| 899 | if (info.stores.ClipDistances()) { | ||
| 900 | AddLabel(labels[label_index]); | ||
| 901 | const Id pointer{OpAccessChain(output_f32, clip_distances, masked_index)}; | ||
| 902 | OpStore(pointer, store_value); | ||
| 903 | OpReturn(); | ||
| 904 | ++label_index; | ||
| 905 | AddLabel(labels[label_index]); | ||
| 906 | const Id fixed_index{OpIAdd(U32[1], masked_index, Const(4U))}; | ||
| 907 | const Id pointer2{OpAccessChain(output_f32, clip_distances, fixed_index)}; | ||
| 908 | OpStore(pointer2, store_value); | ||
| 909 | OpReturn(); | ||
| 910 | ++label_index; | ||
| 911 | } | ||
| 912 | AddLabel(end_block); | ||
| 913 | OpUnreachable(); | ||
| 914 | OpFunctionEnd(); | ||
| 915 | return func; | ||
| 916 | }}; | ||
| 917 | if (info.loads_indexed_attributes) { | ||
| 918 | indexed_load_func = make_load(); | ||
| 919 | } | ||
| 920 | if (info.stores_indexed_attributes) { | ||
| 921 | indexed_store_func = make_store(); | ||
| 922 | } | ||
| 923 | } | ||
| 924 | |||
| 925 | void EmitContext::DefineGlobalMemoryFunctions(const Info& info) { | ||
| 926 | if (!info.uses_global_memory || !profile.support_int64) { | ||
| 927 | return; | ||
| 928 | } | ||
| 929 | using DefPtr = Id StorageDefinitions::*; | ||
| 930 | const Id zero{u32_zero_value}; | ||
| 931 | const auto define_body{[&](DefPtr ssbo_member, Id addr, Id element_pointer, u32 shift, | ||
| 932 | auto&& callback) { | ||
| 933 | AddLabel(); | ||
| 934 | const size_t num_buffers{info.storage_buffers_descriptors.size()}; | ||
| 935 | for (size_t index = 0; index < num_buffers; ++index) { | ||
| 936 | if (!info.nvn_buffer_used[index]) { | ||
| 937 | continue; | ||
| 938 | } | ||
| 939 | const auto& ssbo{info.storage_buffers_descriptors[index]}; | ||
| 940 | const Id ssbo_addr_cbuf_offset{Const(ssbo.cbuf_offset / 8)}; | ||
| 941 | const Id ssbo_size_cbuf_offset{Const(ssbo.cbuf_offset / 4 + 2)}; | ||
| 942 | const Id ssbo_addr_pointer{OpAccessChain( | ||
| 943 | uniform_types.U32x2, cbufs[ssbo.cbuf_index].U32x2, zero, ssbo_addr_cbuf_offset)}; | ||
| 944 | const Id ssbo_size_pointer{OpAccessChain(uniform_types.U32, cbufs[ssbo.cbuf_index].U32, | ||
| 945 | zero, ssbo_size_cbuf_offset)}; | ||
| 946 | |||
| 947 | const Id ssbo_addr{OpBitcast(U64, OpLoad(U32[2], ssbo_addr_pointer))}; | ||
| 948 | const Id ssbo_size{OpUConvert(U64, OpLoad(U32[1], ssbo_size_pointer))}; | ||
| 949 | const Id ssbo_end{OpIAdd(U64, ssbo_addr, ssbo_size)}; | ||
| 950 | const Id cond{OpLogicalAnd(U1, OpUGreaterThanEqual(U1, addr, ssbo_addr), | ||
| 951 | OpULessThan(U1, addr, ssbo_end))}; | ||
| 952 | const Id then_label{OpLabel()}; | ||
| 953 | const Id else_label{OpLabel()}; | ||
| 954 | OpSelectionMerge(else_label, spv::SelectionControlMask::MaskNone); | ||
| 955 | OpBranchConditional(cond, then_label, else_label); | ||
| 956 | AddLabel(then_label); | ||
| 957 | const Id ssbo_id{ssbos[index].*ssbo_member}; | ||
| 958 | const Id ssbo_offset{OpUConvert(U32[1], OpISub(U64, addr, ssbo_addr))}; | ||
| 959 | const Id ssbo_index{OpShiftRightLogical(U32[1], ssbo_offset, Const(shift))}; | ||
| 960 | const Id ssbo_pointer{OpAccessChain(element_pointer, ssbo_id, zero, ssbo_index)}; | ||
| 961 | callback(ssbo_pointer); | ||
| 962 | AddLabel(else_label); | ||
| 963 | } | ||
| 964 | }}; | ||
| 965 | const auto define_load{[&](DefPtr ssbo_member, Id element_pointer, Id type, u32 shift) { | ||
| 966 | const Id function_type{TypeFunction(type, U64)}; | ||
| 967 | const Id func_id{OpFunction(type, spv::FunctionControlMask::MaskNone, function_type)}; | ||
| 968 | const Id addr{OpFunctionParameter(U64)}; | ||
| 969 | define_body(ssbo_member, addr, element_pointer, shift, | ||
| 970 | [&](Id ssbo_pointer) { OpReturnValue(OpLoad(type, ssbo_pointer)); }); | ||
| 971 | OpReturnValue(ConstantNull(type)); | ||
| 972 | OpFunctionEnd(); | ||
| 973 | return func_id; | ||
| 974 | }}; | ||
| 975 | const auto define_write{[&](DefPtr ssbo_member, Id element_pointer, Id type, u32 shift) { | ||
| 976 | const Id function_type{TypeFunction(void_id, U64, type)}; | ||
| 977 | const Id func_id{OpFunction(void_id, spv::FunctionControlMask::MaskNone, function_type)}; | ||
| 978 | const Id addr{OpFunctionParameter(U64)}; | ||
| 979 | const Id data{OpFunctionParameter(type)}; | ||
| 980 | define_body(ssbo_member, addr, element_pointer, shift, [&](Id ssbo_pointer) { | ||
| 981 | OpStore(ssbo_pointer, data); | ||
| 982 | OpReturn(); | ||
| 983 | }); | ||
| 984 | OpReturn(); | ||
| 985 | OpFunctionEnd(); | ||
| 986 | return func_id; | ||
| 987 | }}; | ||
| 988 | const auto define{ | ||
| 989 | [&](DefPtr ssbo_member, const StorageTypeDefinition& type_def, Id type, size_t size) { | ||
| 990 | const Id element_type{type_def.element}; | ||
| 991 | const u32 shift{static_cast<u32>(std::countr_zero(size))}; | ||
| 992 | const Id load_func{define_load(ssbo_member, element_type, type, shift)}; | ||
| 993 | const Id write_func{define_write(ssbo_member, element_type, type, shift)}; | ||
| 994 | return std::make_pair(load_func, write_func); | ||
| 995 | }}; | ||
| 996 | std::tie(load_global_func_u32, write_global_func_u32) = | ||
| 997 | define(&StorageDefinitions::U32, storage_types.U32, U32[1], sizeof(u32)); | ||
| 998 | std::tie(load_global_func_u32x2, write_global_func_u32x2) = | ||
| 999 | define(&StorageDefinitions::U32x2, storage_types.U32x2, U32[2], sizeof(u32[2])); | ||
| 1000 | std::tie(load_global_func_u32x4, write_global_func_u32x4) = | ||
| 1001 | define(&StorageDefinitions::U32x4, storage_types.U32x4, U32[4], sizeof(u32[4])); | ||
| 1002 | } | ||
| 1003 | |||
| 1004 | void EmitContext::DefineRescalingInput(const Info& info) { | ||
| 1005 | if (!info.uses_rescaling_uniform) { | ||
| 1006 | return; | ||
| 1007 | } | ||
| 1008 | if (profile.unified_descriptor_binding) { | ||
| 1009 | DefineRescalingInputPushConstant(); | ||
| 1010 | } else { | ||
| 1011 | DefineRescalingInputUniformConstant(); | ||
| 1012 | } | ||
| 1013 | } | ||
| 1014 | |||
| 1015 | void EmitContext::DefineRescalingInputPushConstant() { | ||
| 1016 | boost::container::static_vector<Id, 3> members{}; | ||
| 1017 | u32 member_index{0}; | ||
| 1018 | |||
| 1019 | rescaling_textures_type = TypeArray(U32[1], Const(4u)); | ||
| 1020 | Decorate(rescaling_textures_type, spv::Decoration::ArrayStride, 4u); | ||
| 1021 | members.push_back(rescaling_textures_type); | ||
| 1022 | rescaling_textures_member_index = member_index++; | ||
| 1023 | |||
| 1024 | rescaling_images_type = TypeArray(U32[1], Const(NUM_IMAGE_SCALING_WORDS)); | ||
| 1025 | Decorate(rescaling_images_type, spv::Decoration::ArrayStride, 4u); | ||
| 1026 | members.push_back(rescaling_images_type); | ||
| 1027 | rescaling_images_member_index = member_index++; | ||
| 1028 | |||
| 1029 | if (stage != Stage::Compute) { | ||
| 1030 | members.push_back(F32[1]); | ||
| 1031 | rescaling_downfactor_member_index = member_index++; | ||
| 1032 | } | ||
| 1033 | const Id push_constant_struct{TypeStruct(std::span(members.data(), members.size()))}; | ||
| 1034 | Decorate(push_constant_struct, spv::Decoration::Block); | ||
| 1035 | Name(push_constant_struct, "ResolutionInfo"); | ||
| 1036 | |||
| 1037 | MemberDecorate(push_constant_struct, rescaling_textures_member_index, spv::Decoration::Offset, | ||
| 1038 | static_cast<u32>(offsetof(RescalingLayout, rescaling_textures))); | ||
| 1039 | MemberName(push_constant_struct, rescaling_textures_member_index, "rescaling_textures"); | ||
| 1040 | |||
| 1041 | MemberDecorate(push_constant_struct, rescaling_images_member_index, spv::Decoration::Offset, | ||
| 1042 | static_cast<u32>(offsetof(RescalingLayout, rescaling_images))); | ||
| 1043 | MemberName(push_constant_struct, rescaling_images_member_index, "rescaling_images"); | ||
| 1044 | |||
| 1045 | if (stage != Stage::Compute) { | ||
| 1046 | MemberDecorate(push_constant_struct, rescaling_downfactor_member_index, | ||
| 1047 | spv::Decoration::Offset, | ||
| 1048 | static_cast<u32>(offsetof(RescalingLayout, down_factor))); | ||
| 1049 | MemberName(push_constant_struct, rescaling_downfactor_member_index, "down_factor"); | ||
| 1050 | } | ||
| 1051 | const Id pointer_type{TypePointer(spv::StorageClass::PushConstant, push_constant_struct)}; | ||
| 1052 | rescaling_push_constants = AddGlobalVariable(pointer_type, spv::StorageClass::PushConstant); | ||
| 1053 | Name(rescaling_push_constants, "rescaling_push_constants"); | ||
| 1054 | |||
| 1055 | if (profile.supported_spirv >= 0x00010400) { | ||
| 1056 | interfaces.push_back(rescaling_push_constants); | ||
| 1057 | } | ||
| 1058 | } | ||
| 1059 | |||
| 1060 | void EmitContext::DefineRescalingInputUniformConstant() { | ||
| 1061 | const Id pointer_type{TypePointer(spv::StorageClass::UniformConstant, F32[4])}; | ||
| 1062 | rescaling_uniform_constant = | ||
| 1063 | AddGlobalVariable(pointer_type, spv::StorageClass::UniformConstant); | ||
| 1064 | Decorate(rescaling_uniform_constant, spv::Decoration::Location, 0u); | ||
| 1065 | |||
| 1066 | if (profile.supported_spirv >= 0x00010400) { | ||
| 1067 | interfaces.push_back(rescaling_uniform_constant); | ||
| 1068 | } | ||
| 1069 | } | ||
| 1070 | |||
| 1071 | void EmitContext::DefineConstantBuffers(const Info& info, u32& binding) { | ||
| 1072 | if (info.constant_buffer_descriptors.empty()) { | ||
| 1073 | return; | ||
| 1074 | } | ||
| 1075 | if (!profile.support_descriptor_aliasing) { | ||
| 1076 | DefineConstBuffers(*this, info, &UniformDefinitions::U32x4, binding, U32[4], 'u', | ||
| 1077 | sizeof(u32[4])); | ||
| 1078 | for (const ConstantBufferDescriptor& desc : info.constant_buffer_descriptors) { | ||
| 1079 | binding += desc.count; | ||
| 1080 | } | ||
| 1081 | return; | ||
| 1082 | } | ||
| 1083 | IR::Type types{info.used_constant_buffer_types}; | ||
| 1084 | if (True(types & IR::Type::U8)) { | ||
| 1085 | if (profile.support_int8) { | ||
| 1086 | DefineConstBuffers(*this, info, &UniformDefinitions::U8, binding, U8, 'u', sizeof(u8)); | ||
| 1087 | DefineConstBuffers(*this, info, &UniformDefinitions::S8, binding, S8, 's', sizeof(s8)); | ||
| 1088 | } else { | ||
| 1089 | types |= IR::Type::U32; | ||
| 1090 | } | ||
| 1091 | } | ||
| 1092 | if (True(types & IR::Type::U16)) { | ||
| 1093 | if (profile.support_int16) { | ||
| 1094 | DefineConstBuffers(*this, info, &UniformDefinitions::U16, binding, U16, 'u', | ||
| 1095 | sizeof(u16)); | ||
| 1096 | DefineConstBuffers(*this, info, &UniformDefinitions::S16, binding, S16, 's', | ||
| 1097 | sizeof(s16)); | ||
| 1098 | } else { | ||
| 1099 | types |= IR::Type::U32; | ||
| 1100 | } | ||
| 1101 | } | ||
| 1102 | if (True(types & IR::Type::U32)) { | ||
| 1103 | DefineConstBuffers(*this, info, &UniformDefinitions::U32, binding, U32[1], 'u', | ||
| 1104 | sizeof(u32)); | ||
| 1105 | } | ||
| 1106 | if (True(types & IR::Type::F32)) { | ||
| 1107 | DefineConstBuffers(*this, info, &UniformDefinitions::F32, binding, F32[1], 'f', | ||
| 1108 | sizeof(f32)); | ||
| 1109 | } | ||
| 1110 | if (True(types & IR::Type::U32x2)) { | ||
| 1111 | DefineConstBuffers(*this, info, &UniformDefinitions::U32x2, binding, U32[2], 'u', | ||
| 1112 | sizeof(u32[2])); | ||
| 1113 | } | ||
| 1114 | binding += static_cast<u32>(info.constant_buffer_descriptors.size()); | ||
| 1115 | } | ||
| 1116 | |||
| 1117 | void EmitContext::DefineStorageBuffers(const Info& info, u32& binding) { | ||
| 1118 | if (info.storage_buffers_descriptors.empty()) { | ||
| 1119 | return; | ||
| 1120 | } | ||
| 1121 | AddExtension("SPV_KHR_storage_buffer_storage_class"); | ||
| 1122 | |||
| 1123 | const IR::Type used_types{profile.support_descriptor_aliasing ? info.used_storage_buffer_types | ||
| 1124 | : IR::Type::U32}; | ||
| 1125 | if (profile.support_int8 && True(used_types & IR::Type::U8)) { | ||
| 1126 | DefineSsbos(*this, storage_types.U8, &StorageDefinitions::U8, info, binding, U8, | ||
| 1127 | sizeof(u8)); | ||
| 1128 | DefineSsbos(*this, storage_types.S8, &StorageDefinitions::S8, info, binding, S8, | ||
| 1129 | sizeof(u8)); | ||
| 1130 | } | ||
| 1131 | if (profile.support_int16 && True(used_types & IR::Type::U16)) { | ||
| 1132 | DefineSsbos(*this, storage_types.U16, &StorageDefinitions::U16, info, binding, U16, | ||
| 1133 | sizeof(u16)); | ||
| 1134 | DefineSsbos(*this, storage_types.S16, &StorageDefinitions::S16, info, binding, S16, | ||
| 1135 | sizeof(u16)); | ||
| 1136 | } | ||
| 1137 | if (True(used_types & IR::Type::U32)) { | ||
| 1138 | DefineSsbos(*this, storage_types.U32, &StorageDefinitions::U32, info, binding, U32[1], | ||
| 1139 | sizeof(u32)); | ||
| 1140 | } | ||
| 1141 | if (True(used_types & IR::Type::F32)) { | ||
| 1142 | DefineSsbos(*this, storage_types.F32, &StorageDefinitions::F32, info, binding, F32[1], | ||
| 1143 | sizeof(f32)); | ||
| 1144 | } | ||
| 1145 | if (True(used_types & IR::Type::U64)) { | ||
| 1146 | DefineSsbos(*this, storage_types.U64, &StorageDefinitions::U64, info, binding, U64, | ||
| 1147 | sizeof(u64)); | ||
| 1148 | } | ||
| 1149 | if (True(used_types & IR::Type::U32x2)) { | ||
| 1150 | DefineSsbos(*this, storage_types.U32x2, &StorageDefinitions::U32x2, info, binding, U32[2], | ||
| 1151 | sizeof(u32[2])); | ||
| 1152 | } | ||
| 1153 | if (True(used_types & IR::Type::U32x4)) { | ||
| 1154 | DefineSsbos(*this, storage_types.U32x4, &StorageDefinitions::U32x4, info, binding, U32[4], | ||
| 1155 | sizeof(u32[4])); | ||
| 1156 | } | ||
| 1157 | for (const StorageBufferDescriptor& desc : info.storage_buffers_descriptors) { | ||
| 1158 | binding += desc.count; | ||
| 1159 | } | ||
| 1160 | const bool needs_function{ | ||
| 1161 | info.uses_global_increment || info.uses_global_decrement || info.uses_atomic_f32_add || | ||
| 1162 | info.uses_atomic_f16x2_add || info.uses_atomic_f16x2_min || info.uses_atomic_f16x2_max || | ||
| 1163 | info.uses_atomic_f32x2_add || info.uses_atomic_f32x2_min || info.uses_atomic_f32x2_max}; | ||
| 1164 | if (needs_function) { | ||
| 1165 | AddCapability(spv::Capability::VariablePointersStorageBuffer); | ||
| 1166 | } | ||
| 1167 | if (info.uses_global_increment) { | ||
| 1168 | increment_cas_ssbo = CasLoop(*this, Operation::Increment, storage_types.U32.array, | ||
| 1169 | storage_types.U32.element, U32[1], U32[1], spv::Scope::Device); | ||
| 1170 | } | ||
| 1171 | if (info.uses_global_decrement) { | ||
| 1172 | decrement_cas_ssbo = CasLoop(*this, Operation::Decrement, storage_types.U32.array, | ||
| 1173 | storage_types.U32.element, U32[1], U32[1], spv::Scope::Device); | ||
| 1174 | } | ||
| 1175 | if (info.uses_atomic_f32_add) { | ||
| 1176 | f32_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array, | ||
| 1177 | storage_types.U32.element, F32[1], U32[1], spv::Scope::Device); | ||
| 1178 | } | ||
| 1179 | if (info.uses_atomic_f16x2_add) { | ||
| 1180 | f16x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array, | ||
| 1181 | storage_types.U32.element, F16[2], F16[2], spv::Scope::Device); | ||
| 1182 | } | ||
| 1183 | if (info.uses_atomic_f16x2_min) { | ||
| 1184 | f16x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.array, | ||
| 1185 | storage_types.U32.element, F16[2], F16[2], spv::Scope::Device); | ||
| 1186 | } | ||
| 1187 | if (info.uses_atomic_f16x2_max) { | ||
| 1188 | f16x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.array, | ||
| 1189 | storage_types.U32.element, F16[2], F16[2], spv::Scope::Device); | ||
| 1190 | } | ||
| 1191 | if (info.uses_atomic_f32x2_add) { | ||
| 1192 | f32x2_add_cas = CasLoop(*this, Operation::FPAdd, storage_types.U32.array, | ||
| 1193 | storage_types.U32.element, F32[2], F32[2], spv::Scope::Device); | ||
| 1194 | } | ||
| 1195 | if (info.uses_atomic_f32x2_min) { | ||
| 1196 | f32x2_min_cas = CasLoop(*this, Operation::FPMin, storage_types.U32.array, | ||
| 1197 | storage_types.U32.element, F32[2], F32[2], spv::Scope::Device); | ||
| 1198 | } | ||
| 1199 | if (info.uses_atomic_f32x2_max) { | ||
| 1200 | f32x2_max_cas = CasLoop(*this, Operation::FPMax, storage_types.U32.array, | ||
| 1201 | storage_types.U32.element, F32[2], F32[2], spv::Scope::Device); | ||
| 1202 | } | ||
| 1203 | } | ||
| 1204 | |||
| 1205 | void EmitContext::DefineTextureBuffers(const Info& info, u32& binding) { | ||
| 1206 | if (info.texture_buffer_descriptors.empty()) { | ||
| 1207 | return; | ||
| 1208 | } | ||
| 1209 | const spv::ImageFormat format{spv::ImageFormat::Unknown}; | ||
| 1210 | image_buffer_type = TypeImage(F32[1], spv::Dim::Buffer, 0U, false, false, 1, format); | ||
| 1211 | sampled_texture_buffer_type = TypeSampledImage(image_buffer_type); | ||
| 1212 | |||
| 1213 | const Id type{TypePointer(spv::StorageClass::UniformConstant, sampled_texture_buffer_type)}; | ||
| 1214 | texture_buffers.reserve(info.texture_buffer_descriptors.size()); | ||
| 1215 | for (const TextureBufferDescriptor& desc : info.texture_buffer_descriptors) { | ||
| 1216 | if (desc.count != 1) { | ||
| 1217 | throw NotImplementedException("Array of texture buffers"); | ||
| 1218 | } | ||
| 1219 | const Id id{AddGlobalVariable(type, spv::StorageClass::UniformConstant)}; | ||
| 1220 | Decorate(id, spv::Decoration::Binding, binding); | ||
| 1221 | Decorate(id, spv::Decoration::DescriptorSet, 0U); | ||
| 1222 | Name(id, NameOf(stage, desc, "texbuf")); | ||
| 1223 | texture_buffers.push_back({ | ||
| 1224 | .id = id, | ||
| 1225 | .count = desc.count, | ||
| 1226 | }); | ||
| 1227 | if (profile.supported_spirv >= 0x00010400) { | ||
| 1228 | interfaces.push_back(id); | ||
| 1229 | } | ||
| 1230 | ++binding; | ||
| 1231 | } | ||
| 1232 | } | ||
| 1233 | |||
| 1234 | void EmitContext::DefineImageBuffers(const Info& info, u32& binding) { | ||
| 1235 | image_buffers.reserve(info.image_buffer_descriptors.size()); | ||
| 1236 | for (const ImageBufferDescriptor& desc : info.image_buffer_descriptors) { | ||
| 1237 | if (desc.count != 1) { | ||
| 1238 | throw NotImplementedException("Array of image buffers"); | ||
| 1239 | } | ||
| 1240 | const spv::ImageFormat format{GetImageFormat(desc.format)}; | ||
| 1241 | const Id image_type{TypeImage(U32[1], spv::Dim::Buffer, false, false, false, 2, format)}; | ||
| 1242 | const Id pointer_type{TypePointer(spv::StorageClass::UniformConstant, image_type)}; | ||
| 1243 | const Id id{AddGlobalVariable(pointer_type, spv::StorageClass::UniformConstant)}; | ||
| 1244 | Decorate(id, spv::Decoration::Binding, binding); | ||
| 1245 | Decorate(id, spv::Decoration::DescriptorSet, 0U); | ||
| 1246 | Name(id, NameOf(stage, desc, "imgbuf")); | ||
| 1247 | image_buffers.push_back({ | ||
| 1248 | .id = id, | ||
| 1249 | .image_type = image_type, | ||
| 1250 | .count = desc.count, | ||
| 1251 | }); | ||
| 1252 | if (profile.supported_spirv >= 0x00010400) { | ||
| 1253 | interfaces.push_back(id); | ||
| 1254 | } | ||
| 1255 | ++binding; | ||
| 1256 | } | ||
| 1257 | } | ||
| 1258 | |||
| 1259 | void EmitContext::DefineTextures(const Info& info, u32& binding, u32& scaling_index) { | ||
| 1260 | textures.reserve(info.texture_descriptors.size()); | ||
| 1261 | for (const TextureDescriptor& desc : info.texture_descriptors) { | ||
| 1262 | const Id image_type{ImageType(*this, desc)}; | ||
| 1263 | const Id sampled_type{TypeSampledImage(image_type)}; | ||
| 1264 | const Id pointer_type{TypePointer(spv::StorageClass::UniformConstant, sampled_type)}; | ||
| 1265 | const Id desc_type{DescType(*this, sampled_type, pointer_type, desc.count)}; | ||
| 1266 | const Id id{AddGlobalVariable(desc_type, spv::StorageClass::UniformConstant)}; | ||
| 1267 | Decorate(id, spv::Decoration::Binding, binding); | ||
| 1268 | Decorate(id, spv::Decoration::DescriptorSet, 0U); | ||
| 1269 | Name(id, NameOf(stage, desc, "tex")); | ||
| 1270 | textures.push_back({ | ||
| 1271 | .id = id, | ||
| 1272 | .sampled_type = sampled_type, | ||
| 1273 | .pointer_type = pointer_type, | ||
| 1274 | .image_type = image_type, | ||
| 1275 | .count = desc.count, | ||
| 1276 | }); | ||
| 1277 | if (profile.supported_spirv >= 0x00010400) { | ||
| 1278 | interfaces.push_back(id); | ||
| 1279 | } | ||
| 1280 | ++binding; | ||
| 1281 | ++scaling_index; | ||
| 1282 | } | ||
| 1283 | if (info.uses_atomic_image_u32) { | ||
| 1284 | image_u32 = TypePointer(spv::StorageClass::Image, U32[1]); | ||
| 1285 | } | ||
| 1286 | } | ||
| 1287 | |||
| 1288 | void EmitContext::DefineImages(const Info& info, u32& binding, u32& scaling_index) { | ||
| 1289 | images.reserve(info.image_descriptors.size()); | ||
| 1290 | for (const ImageDescriptor& desc : info.image_descriptors) { | ||
| 1291 | if (desc.count != 1) { | ||
| 1292 | throw NotImplementedException("Array of images"); | ||
| 1293 | } | ||
| 1294 | const Id image_type{ImageType(*this, desc)}; | ||
| 1295 | const Id pointer_type{TypePointer(spv::StorageClass::UniformConstant, image_type)}; | ||
| 1296 | const Id id{AddGlobalVariable(pointer_type, spv::StorageClass::UniformConstant)}; | ||
| 1297 | Decorate(id, spv::Decoration::Binding, binding); | ||
| 1298 | Decorate(id, spv::Decoration::DescriptorSet, 0U); | ||
| 1299 | Name(id, NameOf(stage, desc, "img")); | ||
| 1300 | images.push_back({ | ||
| 1301 | .id = id, | ||
| 1302 | .image_type = image_type, | ||
| 1303 | .count = desc.count, | ||
| 1304 | }); | ||
| 1305 | if (profile.supported_spirv >= 0x00010400) { | ||
| 1306 | interfaces.push_back(id); | ||
| 1307 | } | ||
| 1308 | ++binding; | ||
| 1309 | ++scaling_index; | ||
| 1310 | } | ||
| 1311 | } | ||
| 1312 | |||
| 1313 | void EmitContext::DefineInputs(const IR::Program& program) { | ||
| 1314 | const Info& info{program.info}; | ||
| 1315 | const VaryingState loads{info.loads.mask | info.passthrough.mask}; | ||
| 1316 | |||
| 1317 | if (info.uses_workgroup_id) { | ||
| 1318 | workgroup_id = DefineInput(*this, U32[3], false, spv::BuiltIn::WorkgroupId); | ||
| 1319 | } | ||
| 1320 | if (info.uses_local_invocation_id) { | ||
| 1321 | local_invocation_id = DefineInput(*this, U32[3], false, spv::BuiltIn::LocalInvocationId); | ||
| 1322 | } | ||
| 1323 | if (info.uses_invocation_id) { | ||
| 1324 | invocation_id = DefineInput(*this, U32[1], false, spv::BuiltIn::InvocationId); | ||
| 1325 | } | ||
| 1326 | if (info.uses_sample_id) { | ||
| 1327 | sample_id = DefineInput(*this, U32[1], false, spv::BuiltIn::SampleId); | ||
| 1328 | } | ||
| 1329 | if (info.uses_is_helper_invocation) { | ||
| 1330 | is_helper_invocation = DefineInput(*this, U1, false, spv::BuiltIn::HelperInvocation); | ||
| 1331 | } | ||
| 1332 | if (info.uses_subgroup_mask) { | ||
| 1333 | subgroup_mask_eq = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupEqMaskKHR); | ||
| 1334 | subgroup_mask_lt = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupLtMaskKHR); | ||
| 1335 | subgroup_mask_le = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupLeMaskKHR); | ||
| 1336 | subgroup_mask_gt = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupGtMaskKHR); | ||
| 1337 | subgroup_mask_ge = DefineInput(*this, U32[4], false, spv::BuiltIn::SubgroupGeMaskKHR); | ||
| 1338 | } | ||
| 1339 | if (info.uses_subgroup_invocation_id || info.uses_subgroup_shuffles || | ||
| 1340 | (profile.warp_size_potentially_larger_than_guest && | ||
| 1341 | (info.uses_subgroup_vote || info.uses_subgroup_mask))) { | ||
| 1342 | subgroup_local_invocation_id = | ||
| 1343 | DefineInput(*this, U32[1], false, spv::BuiltIn::SubgroupLocalInvocationId); | ||
| 1344 | } | ||
| 1345 | if (info.uses_fswzadd) { | ||
| 1346 | const Id f32_one{Const(1.0f)}; | ||
| 1347 | const Id f32_minus_one{Const(-1.0f)}; | ||
| 1348 | const Id f32_zero{Const(0.0f)}; | ||
| 1349 | fswzadd_lut_a = ConstantComposite(F32[4], f32_minus_one, f32_one, f32_minus_one, f32_zero); | ||
| 1350 | fswzadd_lut_b = | ||
| 1351 | ConstantComposite(F32[4], f32_minus_one, f32_minus_one, f32_one, f32_minus_one); | ||
| 1352 | } | ||
| 1353 | if (loads[IR::Attribute::PrimitiveId]) { | ||
| 1354 | primitive_id = DefineInput(*this, U32[1], false, spv::BuiltIn::PrimitiveId); | ||
| 1355 | } | ||
| 1356 | if (loads.AnyComponent(IR::Attribute::PositionX)) { | ||
| 1357 | const bool is_fragment{stage != Stage::Fragment}; | ||
| 1358 | const spv::BuiltIn built_in{is_fragment ? spv::BuiltIn::Position : spv::BuiltIn::FragCoord}; | ||
| 1359 | input_position = DefineInput(*this, F32[4], true, built_in); | ||
| 1360 | if (profile.support_geometry_shader_passthrough) { | ||
| 1361 | if (info.passthrough.AnyComponent(IR::Attribute::PositionX)) { | ||
| 1362 | Decorate(input_position, spv::Decoration::PassthroughNV); | ||
| 1363 | } | ||
| 1364 | } | ||
| 1365 | } | ||
| 1366 | if (loads[IR::Attribute::InstanceId]) { | ||
| 1367 | if (profile.support_vertex_instance_id) { | ||
| 1368 | instance_id = DefineInput(*this, U32[1], true, spv::BuiltIn::InstanceId); | ||
| 1369 | } else { | ||
| 1370 | instance_index = DefineInput(*this, U32[1], true, spv::BuiltIn::InstanceIndex); | ||
| 1371 | base_instance = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseInstance); | ||
| 1372 | } | ||
| 1373 | } | ||
| 1374 | if (loads[IR::Attribute::VertexId]) { | ||
| 1375 | if (profile.support_vertex_instance_id) { | ||
| 1376 | vertex_id = DefineInput(*this, U32[1], true, spv::BuiltIn::VertexId); | ||
| 1377 | } else { | ||
| 1378 | vertex_index = DefineInput(*this, U32[1], true, spv::BuiltIn::VertexIndex); | ||
| 1379 | base_vertex = DefineInput(*this, U32[1], true, spv::BuiltIn::BaseVertex); | ||
| 1380 | } | ||
| 1381 | } | ||
| 1382 | if (loads[IR::Attribute::FrontFace]) { | ||
| 1383 | front_face = DefineInput(*this, U1, true, spv::BuiltIn::FrontFacing); | ||
| 1384 | } | ||
| 1385 | if (loads[IR::Attribute::PointSpriteS] || loads[IR::Attribute::PointSpriteT]) { | ||
| 1386 | point_coord = DefineInput(*this, F32[2], true, spv::BuiltIn::PointCoord); | ||
| 1387 | } | ||
| 1388 | if (loads[IR::Attribute::TessellationEvaluationPointU] || | ||
| 1389 | loads[IR::Attribute::TessellationEvaluationPointV]) { | ||
| 1390 | tess_coord = DefineInput(*this, F32[3], false, spv::BuiltIn::TessCoord); | ||
| 1391 | } | ||
| 1392 | std::bitset<IR::NUM_GENERICS> used_locations{}; | ||
| 1393 | for (size_t index = 0; index < IR::NUM_GENERICS; ++index) { | ||
| 1394 | const AttributeType input_type{runtime_info.generic_input_types[index]}; | ||
| 1395 | if (!runtime_info.previous_stage_stores.Generic(index)) { | ||
| 1396 | continue; | ||
| 1397 | } | ||
| 1398 | if (!loads.Generic(index)) { | ||
| 1399 | continue; | ||
| 1400 | } | ||
| 1401 | if (input_type == AttributeType::Disabled) { | ||
| 1402 | continue; | ||
| 1403 | } | ||
| 1404 | used_locations.set(index); | ||
| 1405 | const Id type{GetAttributeType(*this, input_type)}; | ||
| 1406 | const Id id{DefineInput(*this, type, true)}; | ||
| 1407 | Decorate(id, spv::Decoration::Location, static_cast<u32>(index)); | ||
| 1408 | Name(id, fmt::format("in_attr{}", index)); | ||
| 1409 | input_generics[index] = id; | ||
| 1410 | |||
| 1411 | if (info.passthrough.Generic(index) && profile.support_geometry_shader_passthrough) { | ||
| 1412 | Decorate(id, spv::Decoration::PassthroughNV); | ||
| 1413 | } | ||
| 1414 | if (stage != Stage::Fragment) { | ||
| 1415 | continue; | ||
| 1416 | } | ||
| 1417 | switch (info.interpolation[index]) { | ||
| 1418 | case Interpolation::Smooth: | ||
| 1419 | // Default | ||
| 1420 | // Decorate(id, spv::Decoration::Smooth); | ||
| 1421 | break; | ||
| 1422 | case Interpolation::NoPerspective: | ||
| 1423 | Decorate(id, spv::Decoration::NoPerspective); | ||
| 1424 | break; | ||
| 1425 | case Interpolation::Flat: | ||
| 1426 | Decorate(id, spv::Decoration::Flat); | ||
| 1427 | break; | ||
| 1428 | } | ||
| 1429 | } | ||
| 1430 | size_t previous_unused_location = 0; | ||
| 1431 | if (loads.AnyComponent(IR::Attribute::ColorFrontDiffuseR)) { | ||
| 1432 | input_front_color = DefineLegacyInput(*this, used_locations, previous_unused_location); | ||
| 1433 | } | ||
| 1434 | if (loads.AnyComponent(IR::Attribute::ColorFrontSpecularR)) { | ||
| 1435 | input_front_secondary_color = | ||
| 1436 | DefineLegacyInput(*this, used_locations, previous_unused_location); | ||
| 1437 | } | ||
| 1438 | if (loads.AnyComponent(IR::Attribute::ColorBackDiffuseR)) { | ||
| 1439 | input_back_color = DefineLegacyInput(*this, used_locations, previous_unused_location); | ||
| 1440 | } | ||
| 1441 | if (loads.AnyComponent(IR::Attribute::ColorBackSpecularR)) { | ||
| 1442 | input_back_secondary_color = | ||
| 1443 | DefineLegacyInput(*this, used_locations, previous_unused_location); | ||
| 1444 | } | ||
| 1445 | if (loads.AnyComponent(IR::Attribute::FogCoordinate)) { | ||
| 1446 | input_fog_frag_coord = DefineLegacyInput(*this, used_locations, previous_unused_location); | ||
| 1447 | } | ||
| 1448 | for (size_t index = 0; index < NUM_FIXEDFNCTEXTURE; ++index) { | ||
| 1449 | if (loads.AnyComponent(IR::Attribute::FixedFncTexture0S + index * 4)) { | ||
| 1450 | input_fixed_fnc_textures[index] = | ||
| 1451 | DefineLegacyInput(*this, used_locations, previous_unused_location); | ||
| 1452 | } | ||
| 1453 | } | ||
| 1454 | if (stage == Stage::TessellationEval) { | ||
| 1455 | for (size_t index = 0; index < info.uses_patches.size(); ++index) { | ||
| 1456 | if (!info.uses_patches[index]) { | ||
| 1457 | continue; | ||
| 1458 | } | ||
| 1459 | const Id id{DefineInput(*this, F32[4], false)}; | ||
| 1460 | Decorate(id, spv::Decoration::Patch); | ||
| 1461 | Decorate(id, spv::Decoration::Location, static_cast<u32>(index)); | ||
| 1462 | patches[index] = id; | ||
| 1463 | } | ||
| 1464 | } | ||
| 1465 | } | ||
| 1466 | |||
| 1467 | void EmitContext::DefineOutputs(const IR::Program& program) { | ||
| 1468 | const Info& info{program.info}; | ||
| 1469 | const std::optional<u32> invocations{program.invocations}; | ||
| 1470 | if (info.stores.AnyComponent(IR::Attribute::PositionX) || stage == Stage::VertexB) { | ||
| 1471 | output_position = DefineOutput(*this, F32[4], invocations, spv::BuiltIn::Position); | ||
| 1472 | } | ||
| 1473 | if (info.stores[IR::Attribute::PointSize] || runtime_info.fixed_state_point_size) { | ||
| 1474 | if (stage == Stage::Fragment) { | ||
| 1475 | throw NotImplementedException("Storing PointSize in fragment stage"); | ||
| 1476 | } | ||
| 1477 | output_point_size = DefineOutput(*this, F32[1], invocations, spv::BuiltIn::PointSize); | ||
| 1478 | } | ||
| 1479 | if (info.stores.ClipDistances()) { | ||
| 1480 | if (stage == Stage::Fragment) { | ||
| 1481 | throw NotImplementedException("Storing ClipDistance in fragment stage"); | ||
| 1482 | } | ||
| 1483 | const Id type{TypeArray(F32[1], Const(8U))}; | ||
| 1484 | clip_distances = DefineOutput(*this, type, invocations, spv::BuiltIn::ClipDistance); | ||
| 1485 | } | ||
| 1486 | if (info.stores[IR::Attribute::Layer] && | ||
| 1487 | (profile.support_viewport_index_layer_non_geometry || stage == Stage::Geometry)) { | ||
| 1488 | if (stage == Stage::Fragment) { | ||
| 1489 | throw NotImplementedException("Storing Layer in fragment stage"); | ||
| 1490 | } | ||
| 1491 | layer = DefineOutput(*this, U32[1], invocations, spv::BuiltIn::Layer); | ||
| 1492 | } | ||
| 1493 | if (info.stores[IR::Attribute::ViewportIndex] && | ||
| 1494 | (profile.support_viewport_index_layer_non_geometry || stage == Stage::Geometry)) { | ||
| 1495 | if (stage == Stage::Fragment) { | ||
| 1496 | throw NotImplementedException("Storing ViewportIndex in fragment stage"); | ||
| 1497 | } | ||
| 1498 | viewport_index = DefineOutput(*this, U32[1], invocations, spv::BuiltIn::ViewportIndex); | ||
| 1499 | } | ||
| 1500 | if (info.stores[IR::Attribute::ViewportMask] && profile.support_viewport_mask) { | ||
| 1501 | viewport_mask = DefineOutput(*this, TypeArray(U32[1], Const(1u)), std::nullopt, | ||
| 1502 | spv::BuiltIn::ViewportMaskNV); | ||
| 1503 | } | ||
| 1504 | std::bitset<IR::NUM_GENERICS> used_locations{}; | ||
| 1505 | for (size_t index = 0; index < IR::NUM_GENERICS; ++index) { | ||
| 1506 | if (info.stores.Generic(index)) { | ||
| 1507 | DefineGenericOutput(*this, index, invocations); | ||
| 1508 | used_locations.set(index); | ||
| 1509 | } | ||
| 1510 | } | ||
| 1511 | size_t previous_unused_location = 0; | ||
| 1512 | if (info.stores.AnyComponent(IR::Attribute::ColorFrontDiffuseR)) { | ||
| 1513 | output_front_color = | ||
| 1514 | DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); | ||
| 1515 | } | ||
| 1516 | if (info.stores.AnyComponent(IR::Attribute::ColorFrontSpecularR)) { | ||
| 1517 | output_front_secondary_color = | ||
| 1518 | DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); | ||
| 1519 | } | ||
| 1520 | if (info.stores.AnyComponent(IR::Attribute::ColorBackDiffuseR)) { | ||
| 1521 | output_back_color = | ||
| 1522 | DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); | ||
| 1523 | } | ||
| 1524 | if (info.stores.AnyComponent(IR::Attribute::ColorBackSpecularR)) { | ||
| 1525 | output_back_secondary_color = | ||
| 1526 | DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); | ||
| 1527 | } | ||
| 1528 | if (info.stores.AnyComponent(IR::Attribute::FogCoordinate)) { | ||
| 1529 | output_fog_frag_coord = | ||
| 1530 | DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); | ||
| 1531 | } | ||
| 1532 | for (size_t index = 0; index < NUM_FIXEDFNCTEXTURE; ++index) { | ||
| 1533 | if (info.stores.AnyComponent(IR::Attribute::FixedFncTexture0S + index * 4)) { | ||
| 1534 | output_fixed_fnc_textures[index] = | ||
| 1535 | DefineLegacyOutput(*this, used_locations, previous_unused_location, invocations); | ||
| 1536 | } | ||
| 1537 | } | ||
| 1538 | switch (stage) { | ||
| 1539 | case Stage::TessellationControl: | ||
| 1540 | if (info.stores_tess_level_outer) { | ||
| 1541 | const Id type{TypeArray(F32[1], Const(4U))}; | ||
| 1542 | output_tess_level_outer = | ||
| 1543 | DefineOutput(*this, type, std::nullopt, spv::BuiltIn::TessLevelOuter); | ||
| 1544 | Decorate(output_tess_level_outer, spv::Decoration::Patch); | ||
| 1545 | } | ||
| 1546 | if (info.stores_tess_level_inner) { | ||
| 1547 | const Id type{TypeArray(F32[1], Const(2U))}; | ||
| 1548 | output_tess_level_inner = | ||
| 1549 | DefineOutput(*this, type, std::nullopt, spv::BuiltIn::TessLevelInner); | ||
| 1550 | Decorate(output_tess_level_inner, spv::Decoration::Patch); | ||
| 1551 | } | ||
| 1552 | for (size_t index = 0; index < info.uses_patches.size(); ++index) { | ||
| 1553 | if (!info.uses_patches[index]) { | ||
| 1554 | continue; | ||
| 1555 | } | ||
| 1556 | const Id id{DefineOutput(*this, F32[4], std::nullopt)}; | ||
| 1557 | Decorate(id, spv::Decoration::Patch); | ||
| 1558 | Decorate(id, spv::Decoration::Location, static_cast<u32>(index)); | ||
| 1559 | patches[index] = id; | ||
| 1560 | } | ||
| 1561 | break; | ||
| 1562 | case Stage::Fragment: | ||
| 1563 | for (u32 index = 0; index < 8; ++index) { | ||
| 1564 | if (!info.stores_frag_color[index] && !profile.need_declared_frag_colors) { | ||
| 1565 | continue; | ||
| 1566 | } | ||
| 1567 | frag_color[index] = DefineOutput(*this, F32[4], std::nullopt); | ||
| 1568 | Decorate(frag_color[index], spv::Decoration::Location, index); | ||
| 1569 | Name(frag_color[index], fmt::format("frag_color{}", index)); | ||
| 1570 | } | ||
| 1571 | if (info.stores_frag_depth) { | ||
| 1572 | frag_depth = DefineOutput(*this, F32[1], std::nullopt); | ||
| 1573 | Decorate(frag_depth, spv::Decoration::BuiltIn, spv::BuiltIn::FragDepth); | ||
| 1574 | } | ||
| 1575 | if (info.stores_sample_mask) { | ||
| 1576 | sample_mask = DefineOutput(*this, U32[1], std::nullopt); | ||
| 1577 | Decorate(sample_mask, spv::Decoration::BuiltIn, spv::BuiltIn::SampleMask); | ||
| 1578 | } | ||
| 1579 | break; | ||
| 1580 | default: | ||
| 1581 | break; | ||
| 1582 | } | ||
| 1583 | } | ||
| 1584 | |||
| 1585 | } // namespace Shader::Backend::SPIRV | ||