// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <boost/container/static_vector.hpp>
#include "shader_recompiler/backend/spirv/emit_spirv.h"
#include "shader_recompiler/backend/spirv/emit_spirv_instructions.h"
#include "shader_recompiler/backend/spirv/spirv_emit_context.h"
#include "shader_recompiler/frontend/ir/modifiers.h"
namespace Shader::Backend::SPIRV {
namespace {
class ImageOperands {
public:
explicit ImageOperands(EmitContext& ctx, bool has_bias, bool has_lod, bool has_lod_clamp,
Id lod, const IR::Value& offset) {
if (has_bias) {
const Id bias{has_lod_clamp ? ctx.OpCompositeExtract(ctx.F32[1], lod, 0) : lod};
Add(spv::ImageOperandsMask::Bias, bias);
}
if (has_lod) {
const Id lod_value{has_lod_clamp ? ctx.OpCompositeExtract(ctx.F32[1], lod, 0) : lod};
Add(spv::ImageOperandsMask::Lod, lod_value);
}
AddOffset(ctx, offset);
if (has_lod_clamp) {
const Id lod_clamp{has_bias ? ctx.OpCompositeExtract(ctx.F32[1], lod, 1) : lod};
Add(spv::ImageOperandsMask::MinLod, lod_clamp);
}
}
explicit ImageOperands(EmitContext& ctx, const IR::Value& offset, const IR::Value& offset2) {
if (offset2.IsEmpty()) {
if (offset.IsEmpty()) {
return;
}
Add(spv::ImageOperandsMask::Offset, ctx.Def(offset));
return;
}
const std::array values{offset.InstRecursive(), offset2.InstRecursive()};
if (!values[0]->AreAllArgsImmediates() || !values[1]->AreAllArgsImmediates()) {
LOG_WARNING(Shader_SPIRV, "Not all arguments in PTP are immediate, ignoring");
return;
}
const IR::Opcode opcode{values[0]->GetOpcode()};
if (opcode != values[1]->GetOpcode() || opcode != IR::Opcode::CompositeConstructU32x4) {
throw LogicError("Invalid PTP arguments");
}
auto read{[&](unsigned int a, unsigned int b) { return values[a]->Arg(b).U32(); }};
const Id offsets{ctx.ConstantComposite(
ctx.TypeArray(ctx.U32[2], ctx.Const(4U)), ctx.Const(read(0, 0), read(0, 1)),
ctx.Const(read(0, 2), read(0, 3)), ctx.Const(read(1, 0), read(1, 1)),
ctx.Const(read(1, 2), read(1, 3)))};
Add(spv::ImageOperandsMask::ConstOffsets, offsets);
}
explicit ImageOperands(Id offset, Id lod, Id ms) {
if (Sirit::ValidId(lod)) {
Add(spv::ImageOperandsMask::Lod, lod);
}
if (Sirit::ValidId(offset)) {
Add(spv::ImageOperandsMask::Offset, offset);
}
if (Sirit::ValidId(ms)) {
Add(spv::ImageOperandsMask::Sample, ms);
}
}
explicit ImageOperands(EmitContext& ctx, bool has_lod_clamp, Id derivates, u32 num_derivates,
Id offset, Id lod_clamp) {
if (!Sirit::ValidId(derivates)) {
throw LogicError("Derivates must be present");
}
boost::container::static_vector<Id, 3> deriv_x_accum;
boost::container::static_vector<Id, 3> deriv_y_accum;
for (u32 i = 0; i < num_derivates; ++i) {
deriv_x_accum.push_back(ctx.OpCompositeExtract(ctx.F32[1], derivates, i * 2));
deriv_y_accum.push_back(ctx.OpCompositeExtract(ctx.F32[1], derivates, i * 2 + 1));
}
const Id derivates_X{ctx.OpCompositeConstruct(
ctx.F32[num_derivates], std::span{deriv_x_accum.data(), deriv_x_accum.size()})};
const Id derivates_Y{ctx.OpCompositeConstruct(
ctx.F32[num_derivates], std::span{deriv_y_accum.data(), deriv_y_accum.size()})};
Add(spv::ImageOperandsMask::Grad, derivates_X, derivates_Y);
if (Sirit::ValidId(offset)) {
Add(spv::ImageOperandsMask::Offset, offset);
}
if (has_lod_clamp) {
Add(spv::ImageOperandsMask::MinLod, lod_clamp);
}
}
std::span<const Id> Span() const noexcept {
return std::span{operands.data(), operands.size()};
}
std::optional<spv::ImageOperandsMask> MaskOptional() const noexcept {
return mask != spv::ImageOperandsMask{} ? std::make_optional(mask) : std::nullopt;
}
spv::ImageOperandsMask Mask() const noexcept {
return mask;
}
private:
void AddOffset(EmitContext& ctx, const IR::Value& offset) {
if (offset.IsEmpty()) {
return;
}
if (offset.IsImmediate()) {
Add(spv::ImageOperandsMask::ConstOffset, ctx.SConst(static_cast<s32>(offset.U32())));
return;
}
IR::Inst* const inst{offset.InstRecursive()};
if (inst->AreAllArgsImmediates()) {
switch (inst->GetOpcode()) {
case IR::Opcode::CompositeConstructU32x2:
Add(spv::ImageOperandsMask::ConstOffset,
ctx.SConst(static_cast<s32>(inst->Arg(0).U32()),
static_cast<s32>(inst->Arg(1).U32())));
return;
case IR::Opcode::CompositeConstructU32x3:
Add(spv::ImageOperandsMask::ConstOffset,
ctx.SConst(static_cast<s32>(inst->Arg(0).U32()),
static_cast<s32>(inst->Arg(1).U32()),
static_cast<s32>(inst->Arg(2).U32())));
return;
case IR::Opcode::CompositeConstructU32x4:
Add(spv::ImageOperandsMask::ConstOffset,
ctx.SConst(static_cast<s32>(inst->Arg(0).U32()),
static_cast<s32>(inst->Arg(1).U32()),
static_cast<s32>(inst->Arg(2).U32()),
static_cast<s32>(inst->Arg(3).U32())));
return;
default:
break;
}
}
Add(spv::ImageOperandsMask::Offset, ctx.Def(offset));
}
void Add(spv::ImageOperandsMask new_mask, Id value) {
mask = static_cast<spv::ImageOperandsMask>(static_cast<unsigned>(mask) |
static_cast<unsigned>(new_mask));
operands.push_back(value);
}
void Add(spv::ImageOperandsMask new_mask, Id value_1, Id value_2) {
mask = static_cast<spv::ImageOperandsMask>(static_cast<unsigned>(mask) |
static_cast<unsigned>(new_mask));
operands.push_back(value_1);
operands.push_back(value_2);
}
boost::container::static_vector<Id, 4> operands;
spv::ImageOperandsMask mask{};
};
Id Texture(EmitContext& ctx, IR::TextureInstInfo info, [[maybe_unused]] const IR::Value& index) {
const TextureDefinition& def{ctx.textures.at(info.descriptor_index)};
if (def.count > 1) {
const Id pointer{ctx.OpAccessChain(def.pointer_type, def.id, ctx.Def(index))};
return ctx.OpLoad(def.sampled_type, pointer);
} else {
return ctx.OpLoad(def.sampled_type, def.id);
}
}
Id TextureImage(EmitContext& ctx, IR::TextureInstInfo info, const IR::Value& index) {
if (!index.IsImmediate() || index.U32() != 0) {
throw NotImplementedException("Indirect image indexing");
}
if (info.type == TextureType::Buffer) {
const TextureBufferDefinition& def{ctx.texture_buffers.at(info.descriptor_index)};
if (def.count > 1) {
throw NotImplementedException("Indirect texture sample");
}
const Id sampler_id{def.id};
const Id id{ctx.OpLoad(ctx.sampled_texture_buffer_type, sampler_id)};
return ctx.OpImage(ctx.image_buffer_type, id);
} else {
const TextureDefinition& def{ctx.textures.at(info.descriptor_index)};
if (def.count > 1) {
throw NotImplementedException("Indirect texture sample");
}
return ctx.OpImage(def.image_type, ctx.OpLoad(def.sampled_type, def.id));
}
}
Id Image(EmitContext& ctx, const IR::Value& index, IR::TextureInstInfo info) {
if (!index.IsImmediate() || index.U32() != 0) {
throw NotImplementedException("Indirect image indexing");
}
if (info.type == TextureType::Buffer) {
const ImageBufferDefinition def{ctx.image_buffers.at(info.descriptor_index)};
return ctx.OpLoad(def.image_type, def.id);
} else {
const ImageDefinition def{ctx.images.at(info.descriptor_index)};
return ctx.OpLoad(def.image_type, def.id);
}
}
Id Decorate(EmitContext& ctx, IR::Inst* inst, Id sample) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
if (info.relaxed_precision != 0) {
ctx.Decorate(sample, spv::Decoration::RelaxedPrecision);
}
return sample;
}
template <typename MethodPtrType, typename... Args>
Id Emit(MethodPtrType sparse_ptr, MethodPtrType non_sparse_ptr, EmitContext& ctx, IR::Inst* inst,
Id result_type, Args&&... args) {
IR::Inst* const sparse{inst->GetAssociatedPseudoOperation(IR::Opcode::GetSparseFromOp)};
if (!sparse) {
return Decorate(ctx, inst, (ctx.*non_sparse_ptr)(result_type, std::forward<Args>(args)...));
}
const Id struct_type{ctx.TypeStruct(ctx.U32[1], result_type)};
const Id sample{(ctx.*sparse_ptr)(struct_type, std::forward<Args>(args)...)};
const Id resident_code{ctx.OpCompositeExtract(ctx.U32[1], sample, 0U)};
sparse->SetDefinition(ctx.OpImageSparseTexelsResident(ctx.U1, resident_code));
sparse->Invalidate();
Decorate(ctx, inst, sample);
return ctx.OpCompositeExtract(result_type, sample, 1U);
}
Id IsScaled(EmitContext& ctx, const IR::Value& index, Id member_index, u32 base_index) {
const Id push_constant_u32{ctx.TypePointer(spv::StorageClass::PushConstant, ctx.U32[1])};
Id bit{};
if (index.IsImmediate()) {
// Use BitwiseAnd instead of BitfieldExtract for better codegen on Nvidia OpenGL.
// LOP32I.NZ is used to set the predicate rather than BFE+ISETP.
const u32 index_value{index.U32() + base_index};
const Id word_index{ctx.Const(index_value / 32)};
const Id bit_index_mask{ctx.Const(1u << (index_value % 32))};
const Id pointer{ctx.OpAccessChain(push_constant_u32, ctx.rescaling_push_constants,
member_index, word_index)};
const Id word{ctx.OpLoad(ctx.U32[1], pointer)};
bit = ctx.OpBitwiseAnd(ctx.U32[1], word, bit_index_mask);
} else {
Id index_value{ctx.Def(index)};
if (base_index != 0) {
index_value = ctx.OpIAdd(ctx.U32[1], index_value, ctx.Const(base_index));
}
const Id bit_index{ctx.OpBitwiseAnd(ctx.U32[1], index_value, ctx.Const(31u))};
bit = ctx.OpBitFieldUExtract(ctx.U32[1], index_value, bit_index, ctx.Const(1u));
}
return ctx.OpINotEqual(ctx.U1, bit, ctx.u32_zero_value);
}
Id BitTest(EmitContext& ctx, Id mask, Id bit) {
const Id shifted{ctx.OpShiftRightLogical(ctx.U32[1], mask, bit)};
const Id bit_value{ctx.OpBitwiseAnd(ctx.U32[1], shifted, ctx.Const(1u))};
return ctx.OpINotEqual(ctx.U1, bit_value, ctx.u32_zero_value);
}
} // Anonymous namespace
Id EmitBindlessImageSampleImplicitLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageSampleExplicitLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageSampleDrefImplicitLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageSampleDrefExplicitLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageGather(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageGatherDref(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageFetch(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageQueryDimensions(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageQueryLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageGradient(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageRead(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBindlessImageWrite(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageSampleImplicitLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageSampleExplicitLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageSampleDrefImplicitLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageSampleDrefExplicitLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageGather(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageGatherDref(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageFetch(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageQueryDimensions(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageQueryLod(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageGradient(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageRead(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitBoundImageWrite(EmitContext&) {
throw LogicError("Unreachable instruction");
}
Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
Id bias_lc, const IR::Value& offset) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
if (ctx.stage == Stage::Fragment) {
const ImageOperands operands(ctx, info.has_bias != 0, false, info.has_lod_clamp != 0,
bias_lc, offset);
return Emit(&EmitContext::OpImageSparseSampleImplicitLod,
&EmitContext::OpImageSampleImplicitLod, ctx, inst, ctx.F32[4],
Texture(ctx, info, index), coords, operands.MaskOptional(), operands.Span());
} else {
// We can't use implicit lods on non-fragment stages on SPIR-V. Maxwell hardware behaves as
// if the lod was explicitly zero. This may change on Turing with implicit compute
// derivatives
const Id lod{ctx.Const(0.0f)};
const ImageOperands operands(ctx, false, true, info.has_lod_clamp != 0, lod, offset);
return Emit(&EmitContext::OpImageSparseSampleExplicitLod,
&EmitContext::OpImageSampleExplicitLod, ctx, inst, ctx.F32[4],
Texture(ctx, info, index), coords, operands.Mask(), operands.Span());
}
}
Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
Id lod, const IR::Value& offset) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
const ImageOperands operands(ctx, false, true, false, lod, offset);
return Emit(&EmitContext::OpImageSparseSampleExplicitLod,
&EmitContext::OpImageSampleExplicitLod, ctx, inst, ctx.F32[4],
Texture(ctx, info, index), coords, operands.Mask(), operands.Span());
}
Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index,
Id coords, Id dref, Id bias_lc, const IR::Value& offset) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
if (ctx.stage == Stage::Fragment) {
const ImageOperands operands(ctx, info.has_bias != 0, false, info.has_lod_clamp != 0,
bias_lc, offset);
return Emit(&EmitContext::OpImageSparseSampleDrefImplicitLod,
&EmitContext::OpImageSampleDrefImplicitLod, ctx, inst, ctx.F32[1],
Texture(ctx, info, index), coords, dref, operands.MaskOptional(),
operands.Span());
} else {
// Implicit lods in compute behave on hardware as if sampling from LOD 0.
// This check is to ensure all drivers behave this way.
const Id lod{ctx.Const(0.0f)};
const ImageOperands operands(ctx, false, true, false, lod, offset);
return Emit(&EmitContext::OpImageSparseSampleDrefExplicitLod,
&EmitContext::OpImageSampleDrefExplicitLod, ctx, inst, ctx.F32[1],
Texture(ctx, info, index), coords, dref, operands.Mask(), operands.Span());
}
}
Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index,
Id coords, Id dref, Id lod, const IR::Value& offset) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
const ImageOperands operands(ctx, false, true, false, lod, offset);
return Emit(&EmitContext::OpImageSparseSampleDrefExplicitLod,
&EmitContext::OpImageSampleDrefExplicitLod, ctx, inst, ctx.F32[1],
Texture(ctx, info, index), coords, dref, operands.Mask(), operands.Span());
}
Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
const IR::Value& offset, const IR::Value& offset2) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
const ImageOperands operands(ctx, offset, offset2);
return Emit(&EmitContext::OpImageSparseGather, &EmitContext::OpImageGather, ctx, inst,
ctx.F32[4], Texture(ctx, info, index), coords, ctx.Const(info.gather_component),
operands.MaskOptional(), operands.Span());
}
Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
const IR::Value& offset, const IR::Value& offset2, Id dref) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
const ImageOperands operands(ctx, offset, offset2);
return Emit(&EmitContext::OpImageSparseDrefGather, &EmitContext::OpImageDrefGather, ctx, inst,
ctx.F32[4], Texture(ctx, info, index), coords, dref, operands.MaskOptional(),
operands.Span());
}
Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, Id offset,
Id lod, Id ms) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
if (info.type == TextureType::Buffer) {
lod = Id{};
}
if (Sirit::ValidId(ms)) {
// This image is multisampled, lod must be implicit
lod = Id{};
}
const ImageOperands operands(offset, lod, ms);
return Emit(&EmitContext::OpImageSparseFetch, &EmitContext::OpImageFetch, ctx, inst, ctx.F32[4],
TextureImage(ctx, info, index), coords, operands.MaskOptional(), operands.Span());
}
Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod,
const IR::Value& skip_mips_val) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
const Id image{TextureImage(ctx, info, index)};
const Id zero{ctx.u32_zero_value};
const bool skip_mips{skip_mips_val.U1()};
const auto mips{[&] { return skip_mips ? zero : ctx.OpImageQueryLevels(ctx.U32[1], image); }};
switch (info.type) {
case TextureType::Color1D:
return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[1], image, lod),
zero, zero, mips());
case TextureType::ColorArray1D:
case TextureType::Color2D:
case TextureType::ColorCube:
case TextureType::Color2DRect:
return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[2], image, lod),
zero, mips());
case TextureType::ColorArray2D:
case TextureType::Color3D:
case TextureType::ColorArrayCube:
return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[3], image, lod),
mips());
case TextureType::Buffer:
return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySize(ctx.U32[1], image), zero,
zero, mips());
}
throw LogicError("Unspecified image type {}", info.type.Value());
}
Id EmitImageQueryLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
const Id zero{ctx.f32_zero_value};
const Id sampler{Texture(ctx, info, index)};
return ctx.OpCompositeConstruct(ctx.F32[4], ctx.OpImageQueryLod(ctx.F32[2], sampler, coords),
zero, zero);
}
Id EmitImageGradient(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
Id derivates, Id offset, Id lod_clamp) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
const ImageOperands operands(ctx, info.has_lod_clamp != 0, derivates, info.num_derivates,
offset, lod_clamp);
return Emit(&EmitContext::OpImageSparseSampleExplicitLod,
&EmitContext::OpImageSampleExplicitLod, ctx, inst, ctx.F32[4],
Texture(ctx, info, index), coords, operands.Mask(), operands.Span());
}
Id EmitImageRead(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
if (info.image_format == ImageFormat::Typeless && !ctx.profile.support_typeless_image_loads) {
LOG_WARNING(Shader_SPIRV, "Typeless image read not supported by host");
return ctx.ConstantNull(ctx.U32[4]);
}
return Emit(&EmitContext::OpImageSparseRead, &EmitContext::OpImageRead, ctx, inst, ctx.U32[4],
Image(ctx, index, info), coords, std::nullopt, std::span<const Id>{});
}
void EmitImageWrite(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, Id color) {
const auto info{inst->Flags<IR::TextureInstInfo>()};
ctx.OpImageWrite(Image(ctx, index, info), coords, color);
}
Id EmitIsTextureScaled(EmitContext& ctx, const IR::Value& index) {
if (ctx.profile.unified_descriptor_binding) {
const Id member_index{ctx.Const(ctx.rescaling_textures_member_index)};
return IsScaled(ctx, index, member_index, ctx.texture_rescaling_index);
} else {
const Id composite{ctx.OpLoad(ctx.F32[4], ctx.rescaling_uniform_constant)};
const Id mask_f32{ctx.OpCompositeExtract(ctx.F32[1], composite, 0u)};
const Id mask{ctx.OpBitcast(ctx.U32[1], mask_f32)};
return BitTest(ctx, mask, ctx.Def(index));
}
}
Id EmitIsImageScaled(EmitContext& ctx, const IR::Value& index) {
if (ctx.profile.unified_descriptor_binding) {
const Id member_index{ctx.Const(ctx.rescaling_images_member_index)};
return IsScaled(ctx, index, member_index, ctx.image_rescaling_index);
} else {
const Id composite{ctx.OpLoad(ctx.F32[4], ctx.rescaling_uniform_constant)};
const Id mask_f32{ctx.OpCompositeExtract(ctx.F32[1], composite, 1u)};
const Id mask{ctx.OpBitcast(ctx.U32[1], mask_f32)};
return BitTest(ctx, mask, ctx.Def(index));
}
}
} // namespace Shader::Backend::SPIRV
