shadPS4/src/shader_recompiler/ir/passes/resource_tracking_pass.cpp

394 lines
15 KiB
C++

// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <algorithm>
#include <deque>
#include <boost/container/small_vector.hpp>
#include "shader_recompiler/ir/basic_block.h"
#include "shader_recompiler/ir/ir_emitter.h"
#include "shader_recompiler/ir/program.h"
#include "shader_recompiler/runtime_info.h"
#include "video_core/amdgpu/resource.h"
namespace Shader::Optimization {
namespace {
struct SharpLocation {
u32 sgpr_base;
u32 dword_offset;
auto operator<=>(const SharpLocation&) const = default;
};
bool IsBufferInstruction(const IR::Inst& inst) {
switch (inst.GetOpcode()) {
case IR::Opcode::LoadBufferF32:
case IR::Opcode::LoadBufferF32x2:
case IR::Opcode::LoadBufferF32x3:
case IR::Opcode::LoadBufferF32x4:
case IR::Opcode::LoadBufferU32:
case IR::Opcode::ReadConstBuffer:
case IR::Opcode::ReadConstBufferU32:
case IR::Opcode::StoreBufferF32:
case IR::Opcode::StoreBufferF32x2:
case IR::Opcode::StoreBufferF32x3:
case IR::Opcode::StoreBufferF32x4:
case IR::Opcode::StoreBufferU32:
return true;
default:
return false;
}
}
IR::Type BufferDataType(const IR::Inst& inst) {
switch (inst.GetOpcode()) {
case IR::Opcode::LoadBufferF32:
case IR::Opcode::LoadBufferF32x2:
case IR::Opcode::LoadBufferF32x3:
case IR::Opcode::LoadBufferF32x4:
case IR::Opcode::ReadConstBuffer:
case IR::Opcode::StoreBufferF32:
case IR::Opcode::StoreBufferF32x2:
case IR::Opcode::StoreBufferF32x3:
case IR::Opcode::StoreBufferF32x4:
return IR::Type::F32;
case IR::Opcode::LoadBufferU32:
case IR::Opcode::ReadConstBufferU32:
case IR::Opcode::StoreBufferU32:
return IR::Type::U32;
default:
UNREACHABLE();
}
}
bool IsBufferStore(const IR::Inst& inst) {
switch (inst.GetOpcode()) {
case IR::Opcode::StoreBufferF32:
case IR::Opcode::StoreBufferF32x2:
case IR::Opcode::StoreBufferF32x3:
case IR::Opcode::StoreBufferF32x4:
case IR::Opcode::StoreBufferU32:
return true;
default:
return false;
}
}
bool IsImageInstruction(const IR::Inst& inst) {
switch (inst.GetOpcode()) {
case IR::Opcode::ImageSampleExplicitLod:
case IR::Opcode::ImageSampleImplicitLod:
case IR::Opcode::ImageSampleDrefExplicitLod:
case IR::Opcode::ImageSampleDrefImplicitLod:
case IR::Opcode::ImageFetch:
case IR::Opcode::ImageGather:
case IR::Opcode::ImageGatherDref:
case IR::Opcode::ImageQueryDimensions:
case IR::Opcode::ImageQueryLod:
case IR::Opcode::ImageGradient:
case IR::Opcode::ImageRead:
case IR::Opcode::ImageWrite:
return true;
default:
return false;
}
}
bool IsImageStorageInstruction(const IR::Inst& inst) {
switch (inst.GetOpcode()) {
case IR::Opcode::ImageWrite:
case IR::Opcode::ImageRead:
return true;
default:
return false;
}
}
class Descriptors {
public:
explicit Descriptors(BufferResourceList& buffer_resources_, ImageResourceList& image_resources_,
SamplerResourceList& sampler_resources_)
: buffer_resources{buffer_resources_}, image_resources{image_resources_},
sampler_resources{sampler_resources_} {}
u32 Add(const BufferResource& desc) {
const u32 index{Add(buffer_resources, desc, [&desc](const auto& existing) {
return desc.sgpr_base == existing.sgpr_base &&
desc.dword_offset == existing.dword_offset;
})};
auto& buffer = buffer_resources[index];
ASSERT(buffer.stride == desc.stride && buffer.num_records == desc.num_records);
buffer.is_storage |= desc.is_storage;
buffer.used_types |= desc.used_types;
return index;
}
u32 Add(const ImageResource& desc) {
const u32 index{Add(image_resources, desc, [&desc](const auto& existing) {
return desc.sgpr_base == existing.sgpr_base &&
desc.dword_offset == existing.dword_offset && desc.type == existing.type &&
desc.is_storage == existing.is_storage;
})};
return index;
}
u32 Add(const SamplerResource& desc) {
const u32 index{Add(sampler_resources, desc, [&desc](const auto& existing) {
return desc.sgpr_base == existing.sgpr_base &&
desc.dword_offset == existing.dword_offset;
})};
return index;
}
private:
template <typename Descriptors, typename Descriptor, typename Func>
static u32 Add(Descriptors& descriptors, const Descriptor& desc, Func&& pred) {
const auto it{std::ranges::find_if(descriptors, pred)};
if (it != descriptors.end()) {
return static_cast<u32>(std::distance(descriptors.begin(), it));
}
descriptors.push_back(desc);
return static_cast<u32>(descriptors.size()) - 1;
}
BufferResourceList& buffer_resources;
ImageResourceList& image_resources;
SamplerResourceList& sampler_resources;
};
} // Anonymous namespace
SharpLocation TrackSharp(const IR::Inst* inst) {
while (inst->GetOpcode() == IR::Opcode::Phi) {
inst = inst->Arg(0).InstRecursive();
}
if (inst->GetOpcode() == IR::Opcode::GetUserData) {
return SharpLocation{
.sgpr_base = u32(IR::ScalarReg::Max),
.dword_offset = u32(inst->Arg(0).ScalarReg()),
};
}
ASSERT_MSG(inst->GetOpcode() == IR::Opcode::ReadConst, "Sharp load not from constant memory");
// Retrieve offset from base.
const u32 dword_offset = inst->Arg(1).U32();
const IR::Inst* spgpr_base = inst->Arg(0).InstRecursive();
// Retrieve SGPR pair that holds sbase
const IR::Inst* sbase0 = spgpr_base->Arg(0).InstRecursive();
const IR::Inst* sbase1 = spgpr_base->Arg(1).InstRecursive();
while (sbase0->GetOpcode() == IR::Opcode::Phi) {
sbase0 = sbase0->Arg(0).TryInstRecursive();
}
while (sbase1->GetOpcode() == IR::Opcode::Phi) {
sbase1 = sbase1->Arg(0).TryInstRecursive();
}
ASSERT_MSG(sbase0->GetOpcode() == IR::Opcode::GetUserData &&
sbase1->GetOpcode() == IR::Opcode::GetUserData,
"Nested resource loads not supported");
const IR::ScalarReg base = sbase0->Arg(0).ScalarReg();
// Return retrieved location.
return SharpLocation{
.sgpr_base = u32(base),
.dword_offset = dword_offset,
};
}
void PatchBufferInstruction(IR::Block& block, IR::Inst& inst, Info& info,
Descriptors& descriptors) {
IR::Inst* producer = inst.Arg(0).InstRecursive();
const auto sharp = TrackSharp(producer);
const auto buffer = info.ReadUd<AmdGpu::Buffer>(sharp.sgpr_base, sharp.dword_offset);
const u32 binding = descriptors.Add(BufferResource{
.sgpr_base = sharp.sgpr_base,
.dword_offset = sharp.dword_offset,
.stride = buffer.GetStride(),
.num_records = u32(buffer.num_records),
.used_types = BufferDataType(inst),
.is_storage = IsBufferStore(inst),
});
const auto inst_info = inst.Flags<IR::BufferInstInfo>();
IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
// Replace handle with binding index in buffer resource list.
inst.SetArg(0, ir.Imm32(binding));
ASSERT(!buffer.swizzle_enable && !buffer.add_tid_enable);
if (inst_info.is_typed) {
ASSERT(inst_info.nfmt == AmdGpu::NumberFormat::Float &&
inst_info.dmft == AmdGpu::DataFormat::Format32_32_32_32);
}
if (inst.GetOpcode() == IR::Opcode::ReadConstBuffer ||
inst.GetOpcode() == IR::Opcode::ReadConstBufferU32) {
return;
}
// Calculate buffer address.
const u32 dword_stride = buffer.GetStrideElements(sizeof(u32));
const u32 dword_offset = inst_info.inst_offset.Value() / sizeof(u32);
IR::U32 address = ir.Imm32(dword_offset);
if (inst_info.index_enable && inst_info.offset_enable) {
const IR::U32 offset{ir.CompositeExtract(inst.Arg(1), 1)};
const IR::U32 index{ir.CompositeExtract(inst.Arg(1), 0)};
address = ir.IAdd(ir.IMul(index, ir.Imm32(dword_stride)), address);
address = ir.IAdd(address, ir.ShiftRightLogical(offset, ir.Imm32(2)));
} else if (inst_info.index_enable) {
const IR::U32 index{inst.Arg(1)};
address = ir.IAdd(ir.IMul(index, ir.Imm32(dword_stride)), address);
} else if (inst_info.offset_enable) {
const IR::U32 offset{inst.Arg(1)};
}
inst.SetArg(1, address);
}
IR::Value PatchCubeCoord(IR::IREmitter& ir, const IR::Value& s, const IR::Value& t,
const IR::Value& z) {
// We need to fix x and y coordinate,
// because the s and t coordinate will be scaled and plus 1.5 by v_madak_f32.
// We already force the scale value to be 1.0 when handling v_cubema_f32,
// here we subtract 1.5 to recover the original value.
const IR::Value x = ir.FPSub(IR::F32{s}, ir.Imm32(1.5f));
const IR::Value y = ir.FPSub(IR::F32{t}, ir.Imm32(1.5f));
return ir.CompositeConstruct(x, y, z);
}
void PatchImageInstruction(IR::Block& block, IR::Inst& inst, Info& info, Descriptors& descriptors) {
std::deque<IR::Inst*> insts{&inst};
const auto& pred = [](auto opcode) -> bool {
return (opcode == IR::Opcode::CompositeConstructU32x2 || // IMAGE_SAMPLE (image+sampler)
opcode == IR::Opcode::ReadConst || // IMAGE_LOAD (image only)
opcode == IR::Opcode::GetUserData);
};
IR::Inst* producer{};
while (!insts.empty() && (producer = insts.front(), !pred(producer->GetOpcode()))) {
for (auto arg_idx = 0u; arg_idx < producer->NumArgs(); ++arg_idx) {
const auto arg = producer->Arg(arg_idx);
if (arg.TryInstRecursive()) {
insts.push_back(arg.InstRecursive());
}
}
insts.pop_front();
}
ASSERT(pred(producer->GetOpcode()));
const auto [tsharp_handle, ssharp_handle] = [&] -> std::pair<IR::Inst*, IR::Inst*> {
if (producer->GetOpcode() == IR::Opcode::CompositeConstructU32x2) {
return std::make_pair(producer->Arg(0).InstRecursive(),
producer->Arg(1).InstRecursive());
}
return std::make_pair(producer, nullptr);
}();
// Read image sharp.
const auto tsharp = TrackSharp(tsharp_handle);
const auto image = info.ReadUd<AmdGpu::Image>(tsharp.sgpr_base, tsharp.dword_offset);
const auto inst_info = inst.Flags<IR::TextureInstInfo>();
u32 image_binding = descriptors.Add(ImageResource{
.sgpr_base = tsharp.sgpr_base,
.dword_offset = tsharp.dword_offset,
.type = image.GetType(),
.nfmt = static_cast<AmdGpu::NumberFormat>(image.GetNumberFmt()),
.is_storage = IsImageStorageInstruction(inst),
.is_depth = bool(inst_info.is_depth),
});
// Read sampler sharp. This doesn't exist for IMAGE_LOAD/IMAGE_STORE instructions
if (ssharp_handle) {
const auto ssharp = TrackSharp(ssharp_handle);
const u32 sampler_binding = descriptors.Add(SamplerResource{
.sgpr_base = ssharp.sgpr_base,
.dword_offset = ssharp.dword_offset,
});
image_binding |= (sampler_binding << 16);
}
// Patch image handle
IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
inst.SetArg(0, ir.Imm32(image_binding));
// Now that we know the image type, adjust texture coordinate vector.
const IR::Inst* body = inst.Arg(1).InstRecursive();
const auto [coords, arg] = [&] -> std::pair<IR::Value, IR::Value> {
switch (image.GetType()) {
case AmdGpu::ImageType::Color1D: // x
return {body->Arg(0), body->Arg(1)};
case AmdGpu::ImageType::Color1DArray: // x, slice
[[fallthrough]];
case AmdGpu::ImageType::Color2D: // x, y
return {ir.CompositeConstruct(body->Arg(0), body->Arg(1)), body->Arg(2)};
case AmdGpu::ImageType::Color2DArray: // x, y, slice
[[fallthrough]];
case AmdGpu::ImageType::Color2DMsaa: // x, y, frag
[[fallthrough]];
case AmdGpu::ImageType::Color3D: // x, y, z
return {ir.CompositeConstruct(body->Arg(0), body->Arg(1), body->Arg(2)), body->Arg(3)};
case AmdGpu::ImageType::Cube: // x, y, face
return {PatchCubeCoord(ir, body->Arg(0), body->Arg(1), body->Arg(2)), body->Arg(3)};
default:
UNREACHABLE_MSG("Unknown image type {}", image.GetType());
}
}();
inst.SetArg(1, coords);
if (inst_info.has_lod_clamp) {
// Final argument contains lod_clamp
const u32 arg_pos = inst_info.is_depth ? 5 : 4;
inst.SetArg(arg_pos, arg);
}
if (inst_info.explicit_lod) {
ASSERT(inst.GetOpcode() == IR::Opcode::ImageFetch ||
inst.GetOpcode() == IR::Opcode::ImageSampleExplicitLod);
const u32 pos = inst.GetOpcode() == IR::Opcode::ImageFetch ? 3 : 2;
inst.SetArg(pos, arg);
}
}
void ResourceTrackingPass(IR::Program& program) {
// When loading data from untyped buffer we don't have if it is float or integer.
// Most of the time it is float so that is the default. This pass detects float buffer loads
// combined with bitcasts and patches them to be integer loads.
for (IR::Block* const block : program.post_order_blocks) {
break;
for (IR::Inst& inst : block->Instructions()) {
if (inst.GetOpcode() != IR::Opcode::BitCastU32F32) {
continue;
}
// Replace the bitcast with a typed buffer read
IR::Inst* const arg_inst{inst.Arg(0).TryInstRecursive()};
if (!arg_inst) {
continue;
}
const auto replace{[&](IR::Opcode new_opcode) {
inst.ReplaceOpcode(new_opcode);
inst.SetArg(0, arg_inst->Arg(0));
inst.SetArg(1, arg_inst->Arg(1));
inst.SetFlags(arg_inst->Flags<u32>());
arg_inst->Invalidate();
}};
if (arg_inst->GetOpcode() == IR::Opcode::ReadConstBuffer) {
replace(IR::Opcode::ReadConstBufferU32);
}
if (arg_inst->GetOpcode() == IR::Opcode::LoadBufferF32) {
replace(IR::Opcode::LoadBufferU32);
}
}
}
// Iterate resource instructions and patch them after finding the sharp.
auto& info = program.info;
Descriptors descriptors{info.buffers, info.images, info.samplers};
for (IR::Block* const block : program.post_order_blocks) {
for (IR::Inst& inst : block->Instructions()) {
if (IsBufferInstruction(inst)) {
PatchBufferInstruction(*block, inst, info, descriptors);
continue;
}
if (IsImageInstruction(inst)) {
PatchImageInstruction(*block, inst, info, descriptors);
}
}
}
}
} // namespace Shader::Optimization