catto
/
shadPS4
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

spirv: fix image sample lod/clamp/offset translation (#402) 

* spirv: fix image sample lod/clamp translation

* spirv: fix image sample offsets

* fix ImageSample opcodes & offset emission
This commit is contained in:
Vinicius Rangel 2024-08-13 03:12:38 -03:00 committed by GitHub
parent 5eecd089ab
commit dfcfd62d4f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
7 changed files with 112 additions and 77 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
src/shader_recompiler/backend/spirv/emit_spirv_image.cpp
View File

@ -21,14 +21,19 @@ struct ImageOperands {
boost::container::static_vector<Id, 4> operands;
};
Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias_lc,
Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias,
Id offset) {
const auto& texture = ctx.images[handle & 0xFFFF];
const Id image = ctx.OpLoad(texture.image_type, texture.id);
const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]);
const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler);
ImageOperands operands;
operands.Add(spv::ImageOperandsMask::Offset, offset);
if (Sirit::ValidId(bias)) {
operands.Add(spv::ImageOperandsMask::Bias, bias);
}
if (Sirit::ValidId(offset)) {
operands.Add(spv::ImageOperandsMask::Offset, offset);
}
return ctx.OpImageSampleImplicitLod(ctx.F32[4], sampled_image, coords, operands.mask,
operands.operands);
}
@ -39,27 +44,49 @@ Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id c
const Id image = ctx.OpLoad(texture.image_type, texture.id);
const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]);
const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler);
return ctx.OpImageSampleExplicitLod(ctx.F32[4], sampled_image, coords,
spv::ImageOperandsMask::Lod, lod);
ImageOperands operands;
if (Sirit::ValidId(lod)) {
operands.Add(spv::ImageOperandsMask::Lod, lod);
}
if (Sirit::ValidId(offset)) {
operands.Add(spv::ImageOperandsMask::Offset, offset);
}
return ctx.OpImageSampleExplicitLod(ctx.F32[4], sampled_image, coords, operands.mask,
operands.operands);
}
Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref,
Id bias_lc, const IR::Value& offset) {
Id bias, Id offset) {
const auto& texture = ctx.images[handle & 0xFFFF];
const Id image = ctx.OpLoad(texture.image_type, texture.id);
const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]);
const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler);
return ctx.OpImageSampleDrefImplicitLod(ctx.F32[1], sampled_image, coords, dref);
ImageOperands operands;
if (Sirit::ValidId(bias)) {
operands.Add(spv::ImageOperandsMask::Bias, bias);
}
if (Sirit::ValidId(offset)) {
operands.Add(spv::ImageOperandsMask::Offset, offset);
}
return ctx.OpImageSampleDrefImplicitLod(ctx.F32[1], sampled_image, coords, dref, operands.mask,
operands.operands);
}
Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref,
Id bias_lc, Id offset) {
Id lod, Id offset) {
const auto& texture = ctx.images[handle & 0xFFFF];
const Id image = ctx.OpLoad(texture.image_type, texture.id);
const Id sampler = ctx.OpLoad(ctx.sampler_type, ctx.samplers[handle >> 16]);
const Id sampled_image = ctx.OpSampledImage(texture.sampled_type, image, sampler);
return ctx.OpImageSampleDrefExplicitLod(ctx.F32[1], sampled_image, coords, dref,
spv::ImageOperandsMask::Lod, ctx.ConstF32(0.f));
ImageOperands operands;
if (Sirit::ValidId(lod)) {
operands.Add(spv::ImageOperandsMask::Lod, lod);
}
if (Sirit::ValidId(offset)) {
operands.Add(spv::ImageOperandsMask::Offset, offset);
}
return ctx.OpImageSampleDrefExplicitLod(ctx.F32[1], sampled_image, coords, dref, operands.mask,
operands.operands);
}
Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id offset, Id offset2) {

6
src/shader_recompiler/backend/spirv/emit_spirv_instructions.h
View File

@ -357,14 +357,14 @@ Id EmitConvertF64U64(EmitContext& ctx, Id value);
Id EmitConvertU16U32(EmitContext& ctx, Id value);
Id EmitConvertU32U16(EmitContext& ctx, Id value);
Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias_lc,
Id EmitImageSampleImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id bias,
Id offset);
Id EmitImageSampleExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id lod,
Id offset);
Id EmitImageSampleDrefImplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref,
Id bias_lc, const IR::Value& offset);
Id bias, Id offset);
Id EmitImageSampleDrefExplicitLod(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id dref,
Id bias_lc, Id offset);
Id lod, Id offset);
Id EmitImageGather(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id offset, Id offset2);
Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, u32 handle, Id coords, Id offset,
Id offset2, Id dref);

13
src/shader_recompiler/frontend/translate/vector_memory.cpp
View File

@ -135,8 +135,8 @@ void Translator::IMAGE_SAMPLE(const GcnInst& inst) {
// Load first address components as denoted in 8.2.4 VGPR Usage Sea Islands Series Instruction
// Set Architecture
const IR::Value offset =
flags.test(MimgModifier::Offset) ? ir.GetVectorReg(addr_reg++) : IR::Value{};
const IR::U32 offset =
flags.test(MimgModifier::Offset) ? ir.GetVectorReg<IR::U32>(addr_reg++) : IR::U32{};
const IR::F32 bias =
flags.test(MimgModifier::LodBias) ? ir.GetVectorReg<IR::F32>(addr_reg++) : IR::F32{};
const IR::F32 dref =
@ -168,18 +168,17 @@ void Translator::IMAGE_SAMPLE(const GcnInst& inst) {
// Issue IR instruction, leaving unknown fields blank to patch later.
const IR::Value texel = [&]() -> IR::Value {
const IR::F32 lod = flags.test(MimgModifier::Level0) ? ir.Imm32(0.f) : IR::F32{};
if (!flags.test(MimgModifier::Pcf)) {
if (explicit_lod) {
return ir.ImageSampleExplicitLod(handle, body, lod, offset, info);
return ir.ImageSampleExplicitLod(handle, body, offset, info);
} else {
return ir.ImageSampleImplicitLod(handle, body, bias, offset, {}, info);
return ir.ImageSampleImplicitLod(handle, body, bias, offset, info);
}
}
if (explicit_lod) {
return ir.ImageSampleDrefExplicitLod(handle, body, dref, lod, offset, info);
return ir.ImageSampleDrefExplicitLod(handle, body, dref, offset, info);
}
return ir.ImageSampleDrefImplicitLod(handle, body, dref, bias, offset, {}, info);
return ir.ImageSampleDrefImplicitLod(handle, body, dref, bias, offset, info);
}();
for (u32 i = 0; i < 4; i++) {

44
src/shader_recompiler/ir/ir_emitter.cpp
View File

@ -16,18 +16,6 @@ namespace {
UNREACHABLE_MSG("Invalid type = {}, functionName = {}, line = {}", u32(type), functionName,
lineNumber);
}
Value MakeLodClampPair(IREmitter& ir, const F32& bias_lod, const F32& lod_clamp) {
if (!bias_lod.IsEmpty() && !lod_clamp.IsEmpty()) {
return ir.CompositeConstruct(bias_lod, lod_clamp);
} else if (!bias_lod.IsEmpty()) {
return bias_lod;
} else if (!lod_clamp.IsEmpty()) {
return lod_clamp;
} else {
return Value{};
}
}
} // Anonymous namespace
U1 IREmitter::Imm1(bool value) const {
@ -1386,30 +1374,26 @@ Value IREmitter::ImageAtomicExchange(const Value& handle, const Value& coords, c
return Inst(Opcode::ImageAtomicExchange32, Flags{info}, handle, coords, value);
}
Value IREmitter::ImageSampleImplicitLod(const Value& handle, const Value& coords, const F32& bias,
const Value& offset, const F32& lod_clamp,
Value IREmitter::ImageSampleImplicitLod(const Value& handle, const Value& body, const F32& bias,
const U32& offset, TextureInstInfo info) {
return Inst(Opcode::ImageSampleImplicitLod, Flags{info}, handle, body, bias, offset);
}
Value IREmitter::ImageSampleExplicitLod(const Value& handle, const Value& body, const U32& offset,
TextureInstInfo info) {
const Value bias_lc{MakeLodClampPair(*this, bias, lod_clamp)};
return Inst(Opcode::ImageSampleImplicitLod, Flags{info}, handle, coords, bias_lc, offset);
return Inst(Opcode::ImageSampleExplicitLod, Flags{info}, handle, body, IR::F32{}, offset);
}
Value IREmitter::ImageSampleExplicitLod(const Value& handle, const Value& coords, const F32& lod,
const Value& offset, TextureInstInfo info) {
return Inst(Opcode::ImageSampleExplicitLod, Flags{info}, handle, coords, lod, offset);
}
F32 IREmitter::ImageSampleDrefImplicitLod(const Value& handle, const Value& coords, const F32& dref,
const F32& bias, const Value& offset,
const F32& lod_clamp, TextureInstInfo info) {
const Value bias_lc{MakeLodClampPair(*this, bias, lod_clamp)};
return Inst<F32>(Opcode::ImageSampleDrefImplicitLod, Flags{info}, handle, coords, dref, bias_lc,
F32 IREmitter::ImageSampleDrefImplicitLod(const Value& handle, const Value& body, const F32& dref,
const F32& bias, const U32& offset,
TextureInstInfo info) {
return Inst<F32>(Opcode::ImageSampleDrefImplicitLod, Flags{info}, handle, body, dref, bias,
offset);
}
F32 IREmitter::ImageSampleDrefExplicitLod(const Value& handle, const Value& coords, const F32& dref,
const F32& lod, const Value& offset,
TextureInstInfo info) {
return Inst<F32>(Opcode::ImageSampleDrefExplicitLod, Flags{info}, handle, coords, dref, lod,
F32 IREmitter::ImageSampleDrefExplicitLod(const Value& handle, const Value& body, const F32& dref,
const U32& offset, TextureInstInfo info) {
return Inst<F32>(Opcode::ImageSampleDrefExplicitLod, Flags{info}, handle, body, dref, IR::F32{},
offset);
}

22
src/shader_recompiler/ir/ir_emitter.h
View File

@ -241,19 +241,21 @@ public:
[[nodiscard]] Value ImageAtomicExchange(const Value& handle, const Value& coords,
const Value& value, TextureInstInfo info);
[[nodiscard]] Value ImageSampleImplicitLod(const Value& handle, const Value& coords,
const F32& bias, const Value& offset,
const F32& lod_clamp, TextureInstInfo info);
[[nodiscard]] Value ImageSampleExplicitLod(const Value& handle, const Value& coords,
const F32& lod, const Value& offset,
[[nodiscard]] Value ImageSampleImplicitLod(const Value& handle, const Value& body,
const F32& bias, const U32& offset,
TextureInstInfo info);
[[nodiscard]] F32 ImageSampleDrefImplicitLod(const Value& handle, const Value& coords,
[[nodiscard]] Value ImageSampleExplicitLod(const Value& handle, const Value& body,
const U32& offset, TextureInstInfo info);
[[nodiscard]] F32 ImageSampleDrefImplicitLod(const Value& handle, const Value& body,
const F32& dref, const F32& bias,
const Value& offset, const F32& lod_clamp,
const U32& offset, TextureInstInfo info);
[[nodiscard]] F32 ImageSampleDrefExplicitLod(const Value& handle, const Value& body,
const F32& dref, const U32& offset,
TextureInstInfo info);
[[nodiscard]] F32 ImageSampleDrefExplicitLod(const Value& handle, const Value& coords,
const F32& dref, const F32& lod,
const Value& offset, TextureInstInfo info);
[[nodiscard]] Value ImageQueryDimension(const Value& handle, const IR::U32& lod,
const IR::U1& skip_mips);
[[nodiscard]] Value ImageQueryDimension(const Value& handle, const IR::U32& lod,

8
src/shader_recompiler/ir/opcodes.inc
View File

@ -298,10 +298,10 @@ OPCODE(ConvertU16U32, U16, U32,
OPCODE(ConvertU32U16, U32, U16, )
// Image operations
OPCODE(ImageSampleImplicitLod, F32x4, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageSampleExplicitLod, F32x4, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageSampleDrefImplicitLod, F32, Opaque, Opaque, F32, Opaque, Opaque, )
OPCODE(ImageSampleDrefExplicitLod, F32, Opaque, Opaque, F32, Opaque, Opaque, )
OPCODE(ImageSampleImplicitLod, F32x4, Opaque, Opaque, F32, U32, )
OPCODE(ImageSampleExplicitLod, F32x4, Opaque, Opaque, U32, U32, )
OPCODE(ImageSampleDrefImplicitLod, F32, Opaque, Opaque, Opaque, F32, U32, )
OPCODE(ImageSampleDrefExplicitLod, F32, Opaque, Opaque, Opaque, U32, U32, )
OPCODE(ImageGather, F32x4, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageGatherDref, F32x4, Opaque, Opaque, Opaque, Opaque, F32, )
OPCODE(ImageFetch, F32x4, Opaque, Opaque, Opaque, U32, Opaque, )

51
src/shader_recompiler/ir/passes/resource_tracking_pass.cpp
View File

@ -567,25 +567,47 @@ void PatchImageInstruction(IR::Block& block, IR::Inst& inst, Info& info, Descrip
if (inst_info.has_offset) {
// The offsets are six-bit signed integers: X=[5:0], Y=[13:8], and Z=[21:16].
const bool is_gather = inst.GetOpcode() == IR::Opcode::ImageGather ||
inst.GetOpcode() == IR::Opcode::ImageGatherDref;
const u32 arg_pos = is_gather ? 2 : (inst_info.is_depth ? 4 : 3);
const u32 arg_pos = [&]() -> u32 {
switch (inst.GetOpcode()) {
case IR::Opcode::ImageGather:
case IR::Opcode::ImageGatherDref:
return 2;
case IR::Opcode::ImageSampleExplicitLod:
case IR::Opcode::ImageSampleImplicitLod:
return 3;
case IR::Opcode::ImageSampleDrefExplicitLod:
case IR::Opcode::ImageSampleDrefImplicitLod:
return 4;
default:
break;
}
return inst_info.is_depth ? 4 : 3;
}();
const IR::Value arg = inst.Arg(arg_pos);
ASSERT_MSG(arg.Type() == IR::Type::U32, "Unexpected offset type");
const auto sign_ext = [&](u32 value) { return ir.Imm32(s32(value << 24) >> 24); };
union {
u32 raw;
BitField<0, 6, u32> x;
BitField<8, 6, u32> y;
BitField<16, 6, u32> z;
} offset{arg.U32()};
const IR::Value value = ir.CompositeConstruct(sign_ext(offset.x), sign_ext(offset.y));
const auto f = [&](IR::Value value, u32 offset) -> auto {
return ir.BitFieldExtract(IR::U32{arg}, ir.Imm32(offset), ir.Imm32(6), true);
};
const auto x = f(arg, 0);
const auto y = f(arg, 8);
const auto z = f(arg, 16);
const IR::Value value = ir.CompositeConstruct(x, y, z);
inst.SetArg(arg_pos, value);
}
if (inst_info.has_lod_clamp) {
// Final argument contains lod_clamp
const u32 arg_pos = inst_info.is_depth ? 5 : 4;
const u32 arg_pos = [&]() -> u32 {
switch (inst.GetOpcode()) {
case IR::Opcode::ImageSampleImplicitLod:
return 2;
case IR::Opcode::ImageSampleDrefImplicitLod:
return 3;
default:
break;
}
return inst_info.is_depth ? 5 : 4;
}();
inst.SetArg(arg_pos, arg);
}
if (inst_info.explicit_lod) {
@ -593,7 +615,8 @@ void PatchImageInstruction(IR::Block& block, IR::Inst& inst, Info& info, Descrip
inst.GetOpcode() == IR::Opcode::ImageSampleExplicitLod ||
inst.GetOpcode() == IR::Opcode::ImageSampleDrefExplicitLod);
const u32 pos = inst.GetOpcode() == IR::Opcode::ImageSampleExplicitLod ? 2 : 3;
inst.SetArg(pos, arg);
const IR::Value value = inst_info.force_level0 ? ir.Imm32(0.f) : arg;
inst.SetArg(pos, value);
}
}