Echoblaze/suyu
2
1
Fork 0
forked from suyu/suyu
Code Pull requests Projects Releases Packages Activity Actions 1

Shader_IR: Address Feedback.

Browse source
This commit is contained in:
Fernando Sahmkow 2019-10-17 10:35:16 -04:00 committed by FernandoS27
parent e3afd6595a
commit be856a38d6
9 changed files with 66 additions and 56 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/common/hash.h
View file

@ -72,7 +72,7 @@ struct HashableStruct {
struct PairHash {
template <class T1, class T2>
std::size_t operator()(const std::pair<T1, T2>& pair) const {
std::size_t operator()(const std::pair<T1, T2>& pair) const noexcept {
std::size_t seed = std::hash<T1>()(pair.first);
boost::hash_combine(seed, std::hash<T2>()(pair.second));
return seed;

9
src/video_core/engines/const_buffer_engine_interface.h
View file

@ -34,6 +34,10 @@ struct SamplerDescriptor {
return raw == rhs.raw;
}
bool operator!=(const SamplerDescriptor& rhs) const noexcept {
return !operator==(rhs);
}
static SamplerDescriptor FromTicTexture(Tegra::Texture::TextureType tic_texture_type) {
SamplerDescriptor result;
switch (tic_texture_type) {
@ -73,13 +77,12 @@ struct SamplerDescriptor {
result.is_buffer.Assign(0);
result.is_shadow.Assign(0);
return result;
case Tegra::Texture::TextureType::Texture1DBuffer: {
case Tegra::Texture::TextureType::Texture1DBuffer:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture1D);
result.is_array.Assign(0);
result.is_buffer.Assign(1);
result.is_shadow.Assign(0);
return result;
}
case Tegra::Texture::TextureType::Texture2DNoMipmap:
result.texture_type.Assign(Tegra::Shader::TextureType::Texture2D);
result.is_array.Assign(0);
@ -105,7 +108,7 @@ static_assert(std::is_trivially_copyable_v<SamplerDescriptor>);
class ConstBufferEngineInterface {
public:
virtual ~ConstBufferEngineInterface() {}
virtual ~ConstBufferEngineInterface() = default;
virtual u32 AccessConstBuffer32(ShaderType stage, u64 const_buffer, u64 offset) const = 0;
virtual SamplerDescriptor AccessBoundSampler(ShaderType stage, u64 offset) const = 0;
virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,

8
src/video_core/shader/const_buffer_locker.cpp
View file

@ -22,6 +22,8 @@ ConstBufferLocker::ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
Tegra::Engines::ConstBufferEngineInterface& engine)
: stage{shader_stage}, engine{&engine} {}
ConstBufferLocker::~ConstBufferLocker() = default;
std::optional<u32> ConstBufferLocker::ObtainKey(u32 buffer, u32 offset) {
const std::pair<u32, u32> key = {buffer, offset};
const auto iter = keys.find(key);
@ -29,7 +31,7 @@ std::optional<u32> ConstBufferLocker::ObtainKey(u32 buffer, u32 offset) {
return iter->second;
}
if (!engine) {
return {};
return std::nullopt;
}
const u32 value = engine->AccessConstBuffer32(stage, buffer, offset);
keys.emplace(key, value);
@ -43,7 +45,7 @@ std::optional<SamplerDescriptor> ConstBufferLocker::ObtainBoundSampler(u32 offse
return iter->second;
}
if (!engine) {
return {};
return std::nullopt;
}
const SamplerDescriptor value = engine->AccessBoundSampler(stage, offset);
bound_samplers.emplace(key, value);
@ -58,7 +60,7 @@ std::optional<Tegra::Engines::SamplerDescriptor> ConstBufferLocker::ObtainBindle
return iter->second;
}
if (!engine) {
return {};
return std::nullopt;
}
const SamplerDescriptor value = engine->AccessBindlessSampler(stage, buffer, offset);
bindless_samplers.emplace(key, value);

7
src/video_core/shader/const_buffer_locker.h
View file

@ -16,6 +16,11 @@ using BoundSamplerMap = std::unordered_map<u32, Tegra::Engines::SamplerDescripto
using BindlessSamplerMap =
std::unordered_map<std::pair<u32, u32>, Tegra::Engines::SamplerDescriptor, Common::PairHash>;
/**
* The ConstBufferLocker is a class use to interface the 3D and compute engines with the shader
* compiler. with it, the shader can obtain required data from GPU state and store it for disk
* shader compilation.
**/
class ConstBufferLocker {
public:
explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage);
@ -23,6 +28,8 @@ public:
explicit ConstBufferLocker(Tegra::Engines::ShaderType shader_stage,
Tegra::Engines::ConstBufferEngineInterface& engine);
~ConstBufferLocker();
/// Retrieves a key from the locker, if it's registered, it will give the registered value, if
/// not it will obtain it from maxwell3d and register it.
std::optional<u32> ObtainKey(u32 buffer, u32 offset);

46
src/video_core/shader/control_flow.cpp
View file

@ -41,14 +41,10 @@ BlockBranchInfo MakeBranchInfo(Args&&... args) {
return std::make_shared<BranchData>(T(std::forward<Args>(args)...));
}
bool BlockBranchInfoAreEqual(BlockBranchInfo first, BlockBranchInfo second) {
return false; //(*first) == (*second);
}
bool BlockBranchIsIgnored(BlockBranchInfo first) {
bool ignore = false;
if (std::holds_alternative<SingleBranch>(*first)) {
auto branch = std::get_if<SingleBranch>(first.get());
const auto branch = std::get_if<SingleBranch>(first.get());
ignore = branch->ignore;
}
return ignore;
@ -151,10 +147,10 @@ std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState&
const Instruction instr = {state.program_code[pos]};
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() != OpCode::Id::BRX) {
return {};
return std::nullopt;
}
if (instr.brx.constant_buffer != 0) {
return {};
return std::nullopt;
}
track_register = instr.gpr8.Value();
result.relative_position = instr.brx.GetBranchExtend();
@ -172,8 +168,8 @@ std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState&
if (opcode->get().GetId() == OpCode::Id::LD_C) {
if (instr.gpr0.Value() == track_register &&
instr.ld_c.type.Value() == Tegra::Shader::UniformType::Single) {
result.buffer = instr.cbuf36.index;
result.offset = instr.cbuf36.GetOffset();
result.buffer = instr.cbuf36.index.Value();
result.offset = static_cast<u32>(instr.cbuf36.GetOffset());
track_register = instr.gpr8.Value();
pos--;
found_track = true;
@ -184,7 +180,7 @@ std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState&
}
if (!found_track) {
return {};
return std::nullopt;
}
found_track = false;
@ -194,7 +190,7 @@ std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState&
pos--;
continue;
}
const Instruction instr = {state.program_code[pos]};
const Instruction instr = state.program_code[pos];
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::SHL_IMM) {
if (instr.gpr0.Value() == track_register) {
@ -208,7 +204,7 @@ std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState&
}
if (!found_track) {
return {};
return std::nullopt;
}
found_track = false;
@ -218,7 +214,7 @@ std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState&
pos--;
continue;
}
const Instruction instr = {state.program_code[pos]};
const Instruction instr = state.program_code[pos];
const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::IMNMX_IMM) {
if (instr.gpr0.Value() == track_register) {
@ -233,9 +229,9 @@ std::optional<BranchIndirectInfo> TrackBranchIndirectInfo(const CFGRebuildState&
}
if (!found_track) {
return {};
return std::nullopt;
}
return {result};
return result;
}
std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address) {
@ -440,8 +436,8 @@ std::pair<ParseResult, ParseInfo> ParseCode(CFGRebuildState& state, u32 address)
branches.emplace_back(value, target);
}
parse_info.end_address = offset;
parse_info.branch_info =
MakeBranchInfo<MultiBranch>(static_cast<u32>(instr.gpr8.Value()), branches);
parse_info.branch_info = MakeBranchInfo<MultiBranch>(
static_cast<u32>(instr.gpr8.Value()), std::move(branches));
return {ParseResult::ControlCaught, parse_info};
} else {
@ -486,7 +482,7 @@ bool TryInspectAddress(CFGRebuildState& state) {
current_block.end = address - 1;
new_block.branch = current_block.branch;
BlockBranchInfo forward_branch = MakeBranchInfo<SingleBranch>();
auto branch = std::get_if<SingleBranch>(forward_branch.get());
const auto branch = std::get_if<SingleBranch>(forward_branch.get());
branch->address = address;
branch->ignore = true;
current_block.branch = forward_branch;
@ -504,7 +500,7 @@ bool TryInspectAddress(CFGRebuildState& state) {
BlockInfo& block_info = CreateBlockInfo(state, address, parse_info.end_address);
block_info.branch = parse_info.branch_info;
if (std::holds_alternative<SingleBranch>(*block_info.branch)) {
auto branch = std::get_if<SingleBranch>(block_info.branch.get());
const auto branch = std::get_if<SingleBranch>(block_info.branch.get());
if (branch->condition.IsUnconditional()) {
return true;
}
@ -550,7 +546,7 @@ bool TryQuery(CFGRebuildState& state) {
gather_labels(q2.ssy_stack, state.ssy_labels, block);
gather_labels(q2.pbk_stack, state.pbk_labels, block);
if (std::holds_alternative<SingleBranch>(*block.branch)) {
auto branch = std::get_if<SingleBranch>(block.branch.get());
const auto branch = std::get_if<SingleBranch>(block.branch.get());
if (!branch->condition.IsUnconditional()) {
q2.address = block.end + 1;
state.queries.push_back(q2);
@ -573,8 +569,8 @@ bool TryQuery(CFGRebuildState& state) {
state.queries.push_back(std::move(conditional_query));
return true;
}
auto multi_branch = std::get_if<MultiBranch>(block.branch.get());
for (auto& branch_case : multi_branch->branches) {
const auto multi_branch = std::get_if<MultiBranch>(block.branch.get());
for (const auto& branch_case : multi_branch->branches) {
Query conditional_query{q2};
conditional_query.address = branch_case.address;
state.queries.push_back(std::move(conditional_query));
@ -612,7 +608,7 @@ void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch_info) {
return MakeExpr<ExprBoolean>(true);
});
if (std::holds_alternative<SingleBranch>(*branch_info)) {
auto branch = std::get_if<SingleBranch>(branch_info.get());
const auto branch = std::get_if<SingleBranch>(branch_info.get());
if (branch->address < 0) {
if (branch->kill) {
mm.InsertReturn(get_expr(branch->condition), true);
@ -624,8 +620,8 @@ void InsertBranch(ASTManager& mm, const BlockBranchInfo& branch_info) {
mm.InsertGoto(get_expr(branch->condition), branch->address);
return;
}
auto multi_branch = std::get_if<MultiBranch>(branch_info.get());
for (auto& branch_case : multi_branch->branches) {
const auto multi_branch = std::get_if<MultiBranch>(branch_info.get());
for (const auto& branch_case : multi_branch->branches) {
mm.InsertGoto(MakeExpr<ExprGprEqual>(multi_branch->gpr, branch_case.cmp_value),
branch_case.address);
}

6
src/video_core/shader/control_flow.h
View file

@ -51,6 +51,10 @@ public:
std::tie(b.condition, b.address, b.kill, b.is_sync, b.is_brk, b.ignore);
}
bool operator!=(const SingleBranch& b) const {
return !operator==(b);
}
Condition condition{};
s32 address{exit_branch};
bool kill{};
@ -67,7 +71,7 @@ struct CaseBranch {
class MultiBranch {
public:
MultiBranch(u32 gpr, std::vector<CaseBranch>& branches)
MultiBranch(u32 gpr, std::vector<CaseBranch>&& branches)
: gpr{gpr}, branches{std::move(branches)} {}
u32 gpr{};

2
src/video_core/shader/decode.cpp
View file

@ -155,7 +155,7 @@ void ShaderIR::Decode() {
[[fallthrough]];
case CompileDepth::BruteForce: {
coverage_begin = main_offset;
const u32 shader_end = program_code.size();
const std::size_t shader_end = program_code.size();
coverage_end = shader_end;
for (u32 label = main_offset; label < shader_end; label++) {
basic_blocks.insert({label, DecodeRange(label, label + 1)});

14
src/video_core/shader/decode/texture.cpp
View file

@ -284,7 +284,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
std::optional<SamplerInfo> sampler_info) {
const auto offset = static_cast<std::size_t>(sampler.index.Value());
const auto offset = static_cast<u32>(sampler.index.Value());
Tegra::Shader::TextureType type;
bool is_array;
@ -293,9 +293,7 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
type = sampler_info->type;
is_array = sampler_info->is_array;
is_shadow = sampler_info->is_shadow;
} else {
auto sampler = locker.ObtainBoundSampler(offset);
if (sampler) {
} else if (auto sampler = locker.ObtainBoundSampler(offset); sampler) {
type = sampler->texture_type.Value();
is_array = sampler->is_array.Value() != 0;
is_shadow = sampler->is_shadow.Value() != 0;
@ -304,7 +302,6 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
is_array = false;
is_shadow = false;
}
}
// If this sampler has already been used, return the existing mapping.
const auto itr =
@ -320,7 +317,7 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
const std::size_t next_index = used_samplers.size();
const Sampler entry{offset, next_index, type, is_array, is_shadow};
return *used_samplers.emplace(entry).first;
}
} // namespace VideoCommon::Shader
const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg,
std::optional<SamplerInfo> sampler_info) {
@ -336,9 +333,7 @@ const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg,
type = sampler_info->type;
is_array = sampler_info->is_array;
is_shadow = sampler_info->is_shadow;
} else {
auto sampler = locker.ObtainBindlessSampler(cbuf_index, cbuf_offset);
if (sampler) {
} else if (auto sampler = locker.ObtainBindlessSampler(cbuf_index, cbuf_offset); sampler) {
type = sampler->texture_type.Value();
is_array = sampler->is_array.Value() != 0;
is_shadow = sampler->is_shadow.Value() != 0;
@ -347,7 +342,6 @@ const Sampler& ShaderIR::GetBindlessSampler(const Tegra::Shader::Register& reg,
is_array = false;
is_shadow = false;
}
}
// If this sampler has already been used, return the existing mapping.
const auto itr =

4
src/video_core/shader/expr.h
View file

@ -127,6 +127,10 @@ public:
return gpr == b.gpr && value == b.value;
}
bool operator!=(const ExprGprEqual& b) const {
return !operator==(b);
}
u32 gpr;
u32 value;
};