yuzu-emu
/
yuzu
Archived
1
0
Fork 0

shader: Address feedback

This commit is contained in:
FernandoS27 2021-04-05 04:03:12 +02:00 committed by ameerj
parent 73cb17f41b
commit dcaf0e9150
5 changed files with 54 additions and 53 deletions

View File

@ -314,7 +314,7 @@ void EmitContext::DefineSharedMemory(const IR::Program& program) {
} }
void EmitContext::DefineAttributeMemAccess(const Info& info) { void EmitContext::DefineAttributeMemAccess(const Info& info) {
const auto make_load{[&]() { const auto make_load{[&] {
const Id end_block{OpLabel()}; const Id end_block{OpLabel()};
const Id default_label{OpLabel()}; const Id default_label{OpLabel()};
@ -322,9 +322,9 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
const Id func{OpFunction(F32[1], spv::FunctionControlMask::MaskNone, func_type_load)}; const Id func{OpFunction(F32[1], spv::FunctionControlMask::MaskNone, func_type_load)};
const Id offset{OpFunctionParameter(U32[1])}; const Id offset{OpFunctionParameter(U32[1])};
AddLabel(); AddLabel();
const Id base_index{OpShiftRightLogical(U32[1], offset, Constant(U32[1], 2U))}; const Id base_index{OpShiftRightArithmetic(U32[1], offset, Constant(U32[1], 2U))};
const Id masked_index{OpBitwiseAnd(U32[1], base_index, Constant(U32[1], 3U))}; const Id masked_index{OpBitwiseAnd(U32[1], base_index, Constant(U32[1], 3U))};
const Id compare_index{OpShiftRightLogical(U32[1], base_index, Constant(U32[1], 2U))}; const Id compare_index{OpShiftRightArithmetic(U32[1], base_index, Constant(U32[1], 2U))};
std::vector<Sirit::Literal> literals; std::vector<Sirit::Literal> literals;
std::vector<Id> labels; std::vector<Id> labels;
if (info.loads_position) { if (info.loads_position) {
@ -343,22 +343,22 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
OpSwitch(compare_index, default_label, literals, labels); OpSwitch(compare_index, default_label, literals, labels);
AddLabel(default_label); AddLabel(default_label);
OpReturnValue(Constant(F32[1], 0.0f)); OpReturnValue(Constant(F32[1], 0.0f));
size_t label_index = 0; size_t label_index{0};
if (info.loads_position) { if (info.loads_position) {
AddLabel(labels[label_index]); AddLabel(labels[label_index]);
const Id result{OpLoad(F32[1], OpAccessChain(input_f32, input_position, masked_index))}; const Id result{OpLoad(F32[1], OpAccessChain(input_f32, input_position, masked_index))};
OpReturnValue(result); OpReturnValue(result);
label_index++; ++label_index;
} }
for (u32 i = 0; i < info.input_generics.size(); i++) { for (size_t i = 0; i < info.input_generics.size(); i++) {
if (!info.input_generics[i].used) { if (!info.input_generics[i].used) {
continue; continue;
} }
AddLabel(labels[label_index]); AddLabel(labels[label_index]);
const auto type{AttrTypes(*this, i)}; const auto type{AttrTypes(*this, static_cast<u32>(i))};
if (!type) { if (!type) {
OpReturnValue(Constant(F32[1], 0.0f)); OpReturnValue(Constant(F32[1], 0.0f));
label_index++; ++label_index;
continue; continue;
} }
const Id generic_id{input_generics.at(i)}; const Id generic_id{input_generics.at(i)};
@ -366,14 +366,14 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
const Id value{OpLoad(type->id, pointer)}; const Id value{OpLoad(type->id, pointer)};
const Id result{type->needs_cast ? OpBitcast(F32[1], value) : value}; const Id result{type->needs_cast ? OpBitcast(F32[1], value) : value};
OpReturnValue(result); OpReturnValue(result);
label_index++; ++label_index;
} }
AddLabel(end_block); AddLabel(end_block);
OpUnreachable(); OpUnreachable();
OpFunctionEnd(); OpFunctionEnd();
return func; return func;
}}; }};
const auto make_store{[&]() { const auto make_store{[&] {
const Id end_block{OpLabel()}; const Id end_block{OpLabel()};
const Id default_label{OpLabel()}; const Id default_label{OpLabel()};
@ -382,9 +382,9 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
const Id offset{OpFunctionParameter(U32[1])}; const Id offset{OpFunctionParameter(U32[1])};
const Id store_value{OpFunctionParameter(F32[1])}; const Id store_value{OpFunctionParameter(F32[1])};
AddLabel(); AddLabel();
const Id base_index{OpShiftRightLogical(U32[1], offset, Constant(U32[1], 2U))}; const Id base_index{OpShiftRightArithmetic(U32[1], offset, Constant(U32[1], 2U))};
const Id masked_index{OpBitwiseAnd(U32[1], base_index, Constant(U32[1], 3U))}; const Id masked_index{OpBitwiseAnd(U32[1], base_index, Constant(U32[1], 3U))};
const Id compare_index{OpShiftRightLogical(U32[1], base_index, Constant(U32[1], 2U))}; const Id compare_index{OpShiftRightArithmetic(U32[1], base_index, Constant(U32[1], 2U))};
std::vector<Sirit::Literal> literals; std::vector<Sirit::Literal> literals;
std::vector<Id> labels; std::vector<Id> labels;
if (info.stores_position) { if (info.stores_position) {
@ -392,11 +392,11 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
labels.push_back(OpLabel()); labels.push_back(OpLabel());
} }
const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2; const u32 base_attribute_value = static_cast<u32>(IR::Attribute::Generic0X) >> 2;
for (u32 i = 0; i < info.stores_generics.size(); i++) { for (size_t i = 0; i < info.stores_generics.size(); i++) {
if (!info.stores_generics[i]) { if (!info.stores_generics[i]) {
continue; continue;
} }
literals.push_back(base_attribute_value + i); literals.push_back(base_attribute_value + static_cast<u32>(i));
labels.push_back(OpLabel()); labels.push_back(OpLabel());
} }
if (info.stores_clip_distance) { if (info.stores_clip_distance) {
@ -409,15 +409,15 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
OpSwitch(compare_index, default_label, literals, labels); OpSwitch(compare_index, default_label, literals, labels);
AddLabel(default_label); AddLabel(default_label);
OpReturn(); OpReturn();
size_t label_index = 0; size_t label_index{0};
if (info.stores_position) { if (info.stores_position) {
AddLabel(labels[label_index]); AddLabel(labels[label_index]);
const Id pointer{OpAccessChain(output_f32, output_position, masked_index)}; const Id pointer{OpAccessChain(output_f32, output_position, masked_index)};
OpStore(pointer, store_value); OpStore(pointer, store_value);
OpReturn(); OpReturn();
label_index++; ++label_index;
} }
for (u32 i = 0; i < info.stores_generics.size(); i++) { for (size_t i = 0; i < info.stores_generics.size(); i++) {
if (!info.stores_generics[i]) { if (!info.stores_generics[i]) {
continue; continue;
} }
@ -426,20 +426,20 @@ void EmitContext::DefineAttributeMemAccess(const Info& info) {
const Id pointer{OpAccessChain(output_f32, generic_id, masked_index)}; const Id pointer{OpAccessChain(output_f32, generic_id, masked_index)};
OpStore(pointer, store_value); OpStore(pointer, store_value);
OpReturn(); OpReturn();
label_index++; ++label_index;
} }
if (info.stores_clip_distance) { if (info.stores_clip_distance) {
AddLabel(labels[label_index]); AddLabel(labels[label_index]);
const Id pointer{OpAccessChain(output_f32, clip_distances, masked_index)}; const Id pointer{OpAccessChain(output_f32, clip_distances, masked_index)};
OpStore(pointer, store_value); OpStore(pointer, store_value);
OpReturn(); OpReturn();
label_index++; ++label_index;
AddLabel(labels[label_index]); AddLabel(labels[label_index]);
const Id fixed_index{OpIAdd(U32[1], masked_index, Constant(U32[1], 4))}; const Id fixed_index{OpIAdd(U32[1], masked_index, Constant(U32[1], 4))};
const Id pointer2{OpAccessChain(output_f32, clip_distances, fixed_index)}; const Id pointer2{OpAccessChain(output_f32, clip_distances, fixed_index)};
OpStore(pointer2, store_value); OpStore(pointer2, store_value);
OpReturn(); OpReturn();
label_index++; ++label_index;
} }
AddLabel(end_block); AddLabel(end_block);
OpUnreachable(); OpUnreachable();

View File

@ -307,11 +307,11 @@ void IREmitter::SetAttribute(IR::Attribute attribute, const F32& value) {
Inst(Opcode::SetAttribute, attribute, value); Inst(Opcode::SetAttribute, attribute, value);
} }
F32 IREmitter::GetAttributeIndexed(IR::U32 phys_address) { F32 IREmitter::GetAttributeIndexed(const U32& phys_address) {
return Inst<F32>(Opcode::GetAttributeIndexed, phys_address); return Inst<F32>(Opcode::GetAttributeIndexed, phys_address);
} }
void IREmitter::SetAttributeIndexed(IR::U32 phys_address, const F32& value) { void IREmitter::SetAttributeIndexed(const U32& phys_address, const F32& value) {
Inst(Opcode::SetAttributeIndexed, phys_address, value); Inst(Opcode::SetAttributeIndexed, phys_address, value);
} }

View File

@ -76,8 +76,8 @@ public:
[[nodiscard]] F32 GetAttribute(IR::Attribute attribute); [[nodiscard]] F32 GetAttribute(IR::Attribute attribute);
void SetAttribute(IR::Attribute attribute, const F32& value); void SetAttribute(IR::Attribute attribute, const F32& value);
[[nodiscard]] F32 GetAttributeIndexed(IR::U32 phys_address); [[nodiscard]] F32 GetAttributeIndexed(const U32& phys_address);
void SetAttributeIndexed(IR::U32 phys_address, const F32& value); void SetAttributeIndexed(const U32& phys_address, const F32& value);
void SetFragColor(u32 index, u32 component, const F32& value); void SetFragColor(u32 index, u32 component, const F32& value);
void SetFragDepth(const F32& value); void SetFragDepth(const F32& value);

View File

@ -44,6 +44,17 @@ u32 NumElements(Size size) {
} }
throw InvalidArgument("Invalid size {}", size); throw InvalidArgument("Invalid size {}", size);
} }
template <typename F>
void HandleIndexed(TranslatorVisitor& v, IR::Reg index_reg, u32 num_elements, F&& f) {
const IR::U32 index_value{v.X(index_reg)};
for (u32 element = 0; element < num_elements; ++element) {
const IR::U32 final_offset{
element == 0 ? index_value : IR::U32{v.ir.IAdd(index_value, v.ir.Imm32(element * 4U))}};
f(element, final_offset);
}
}
} // Anonymous namespace } // Anonymous namespace
void TranslatorVisitor::ALD(u64 insn) { void TranslatorVisitor::ALD(u64 insn) {
@ -70,18 +81,15 @@ void TranslatorVisitor::ALD(u64 insn) {
throw NotImplementedException("Unaligned absolute offset {}", offset); throw NotImplementedException("Unaligned absolute offset {}", offset);
} }
const u32 num_elements{NumElements(ald.size)}; const u32 num_elements{NumElements(ald.size)};
if (ald.index_reg != IR::Reg::RZ) { if (ald.index_reg == IR::Reg::RZ) {
const IR::U32 index_value = X(ald.index_reg);
for (u32 element = 0; element < num_elements; ++element) { for (u32 element = 0; element < num_elements; ++element) {
const IR::U32 final_offset = F(ald.dest_reg + element, ir.GetAttribute(IR::Attribute{offset / 4 + element}));
element == 0 ? index_value : IR::U32{ir.IAdd(index_value, ir.Imm32(element * 4U))};
F(ald.dest_reg + element, ir.GetAttributeIndexed(final_offset));
} }
return; return;
} }
for (u32 element = 0; element < num_elements; ++element) { HandleIndexed(*this, ald.index_reg, num_elements, [&](u32 element, IR::U32 final_offset) {
F(ald.dest_reg + element, ir.GetAttribute(IR::Attribute{offset / 4 + element})); F(ald.dest_reg + element, ir.GetAttributeIndexed(final_offset));
} });
} }
void TranslatorVisitor::AST(u64 insn) { void TranslatorVisitor::AST(u64 insn) {
@ -110,18 +118,15 @@ void TranslatorVisitor::AST(u64 insn) {
throw NotImplementedException("Unaligned absolute offset {}", offset); throw NotImplementedException("Unaligned absolute offset {}", offset);
} }
const u32 num_elements{NumElements(ast.size)}; const u32 num_elements{NumElements(ast.size)};
if (ast.index_reg != IR::Reg::RZ) { if (ast.index_reg == IR::Reg::RZ) {
const IR::U32 index_value = X(ast.index_reg);
for (u32 element = 0; element < num_elements; ++element) { for (u32 element = 0; element < num_elements; ++element) {
const IR::U32 final_offset = ir.SetAttribute(IR::Attribute{offset / 4 + element}, F(ast.src_reg + element));
element == 0 ? index_value : IR::U32{ir.IAdd(index_value, ir.Imm32(element * 4U))};
ir.SetAttributeIndexed(final_offset, F(ast.src_reg + element));
} }
return; return;
} }
for (u32 element = 0; element < num_elements; ++element) { HandleIndexed(*this, ast.index_reg, num_elements, [&](u32 element, IR::U32 final_offset) {
ir.SetAttribute(IR::Attribute{offset / 4 + element}, F(ast.src_reg + element)); ir.SetAttributeIndexed(final_offset, F(ast.src_reg + element));
} });
} }
void TranslatorVisitor::IPA(u64 insn) { void TranslatorVisitor::IPA(u64 insn) {

View File

@ -511,37 +511,33 @@ void Visit(Info& info, IR::Inst& inst) {
} }
void GatherInfoFromHeader(Environment& env, Info& info) { void GatherInfoFromHeader(Environment& env, Info& info) {
auto stage = env.ShaderStage(); Stage stage{env.ShaderStage()};
if (stage == Stage::Compute) { if (stage == Stage::Compute) {
return; return;
} }
const auto& header = env.SPH(); const auto& header{env.SPH()};
if (stage == Stage::Fragment) { if (stage == Stage::Fragment) {
if (!info.loads_indexed_attributes) { if (!info.loads_indexed_attributes) {
return; return;
} }
for (size_t i = 0; i < info.input_generics.size(); i++) { for (size_t i = 0; i < info.input_generics.size(); i++) {
info.input_generics[i].used = info.input_generics[i].used |= header.ps.IsGenericVectorActive(i);
info.input_generics[i].used || header.ps.IsGenericVectorActive(i);
} }
info.loads_position = info.loads_position || header.ps.imap_systemb.position != 0; info.loads_position |= header.ps.imap_systemb.position != 0;
return; return;
} }
if (info.loads_indexed_attributes) { if (info.loads_indexed_attributes) {
for (size_t i = 0; i < info.input_generics.size(); i++) { for (size_t i = 0; i < info.input_generics.size(); i++) {
info.input_generics[i].used = info.input_generics[i].used |= header.vtg.IsInputGenericVectorActive(i);
info.input_generics[i].used || header.vtg.IsInputGenericVectorActive(i);
} }
info.loads_position |= header.vtg.imap_systemb.position != 0;
} }
if (info.stores_indexed_attributes) { if (info.stores_indexed_attributes) {
info.loads_position = info.loads_position || header.vtg.imap_systemb.position != 0;
for (size_t i = 0; i < info.stores_generics.size(); i++) { for (size_t i = 0; i < info.stores_generics.size(); i++) {
info.stores_generics[i] = info.stores_generics[i] |= header.vtg.IsOutputGenericVectorActive(i);
info.stores_generics[i] || header.vtg.IsOutputGenericVectorActive(i);
} }
info.stores_clip_distance = info.stores_clip_distance |= header.vtg.omap_systemc.clip_distances != 0;
info.stores_clip_distance || header.vtg.omap_systemc.clip_distances != 0; info.stores_position |= header.vtg.omap_systemb.position != 0;
info.stores_position = info.stores_position || header.vtg.omap_systemb.position != 0;
} }
} }