lux/citra
1
0
Fork 0
mirror of https://git.h3cjp.net/H3cJP/citra.git synced 2024-12-04 09:52:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: Address Feedback

Browse source
This commit is contained in:
FernandoS27 2021-04-03 01:48:39 +02:00 committed by ameerj
parent 45d547af11
commit baec84247f
16 changed files with 60 additions and 211 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
src/shader_recompiler/backend/spirv/emit_context.cpp
View file

@ -510,7 +510,8 @@ void EmitContext::DefineOutputs(const Info& info) {
const Id type{TypeArray(F32[1], Constant(U32[1], 8U))};
clip_distances = DefineOutput(*this, type, spv::BuiltIn::ClipDistance);
}
if (info.stores_viewport_index && !ignore_viewport_layer) {
if (info.stores_viewport_index &&
(profile.support_viewport_index_layer_non_geometry || stage == Shader::Stage::Geometry)) {
if (stage == Stage::Fragment) {
throw NotImplementedException("Storing ViewportIndex in Fragment stage");
}

2
src/shader_recompiler/backend/spirv/emit_context.h
View file

@ -134,8 +134,6 @@ public:
std::vector<Id> interfaces;
bool ignore_viewport_layer{};
private:
void DefineCommonTypes(const Info& info);
void DefineCommonConstants();

4
src/shader_recompiler/backend/spirv/emit_spirv.cpp
View file

@ -228,11 +228,9 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
if (info.stores_viewport_index) {
ctx.AddCapability(spv::Capability::MultiViewport);
if (profile.support_viewport_index_layer_non_geometry &&
ctx.stage == Shader::Stage::VertexB) {
ctx.stage != Shader::Stage::Geometry) {
ctx.AddExtension("SPV_EXT_shader_viewport_index_layer");
ctx.AddCapability(spv::Capability::ShaderViewportIndexLayerEXT);
} else {
ctx.ignore_viewport_layer = true;
}
}
if (!profile.support_vertex_instance_id && (info.loads_instance_id || info.loads_vertex_id)) {

12
src/shader_recompiler/backend/spirv/emit_spirv.h
View file

@ -28,7 +28,9 @@ void EmitSelectionMerge(EmitContext& ctx, Id merge_label);
void EmitReturn(EmitContext& ctx);
void EmitUnreachable(EmitContext& ctx);
void EmitDemoteToHelperInvocation(EmitContext& ctx, Id continue_label);
void EmitMemoryBarrier(EmitContext& ctx, IR::Inst* inst);
void EmitMemoryBarrierWorkgroupLevel(EmitContext& ctx);
void EmitMemoryBarrierDeviceLevel(EmitContext& ctx);
void EmitMemoryBarrierSystemLevel(EmitContext& ctx);
void EmitPrologue(EmitContext& ctx);
void EmitEpilogue(EmitContext& ctx);
void EmitGetRegister(EmitContext& ctx);
@ -60,14 +62,6 @@ void EmitSetZFlag(EmitContext& ctx);
void EmitSetSFlag(EmitContext& ctx);
void EmitSetCFlag(EmitContext& ctx);
void EmitSetOFlag(EmitContext& ctx);
void EmitGetFCSMFlag(EmitContext& ctx);
void EmitGetTAFlag(EmitContext& ctx);
void EmitGetTRFlag(EmitContext& ctx);
void EmitGetMXFlag(EmitContext& ctx);
void EmitSetFCSMFlag(EmitContext& ctx);
void EmitSetTAFlag(EmitContext& ctx);
void EmitSetTRFlag(EmitContext& ctx);
void EmitSetMXFlag(EmitContext& ctx);
Id EmitWorkgroupId(EmitContext& ctx);
Id EmitLocalInvocationId(EmitContext& ctx);
Id EmitLoadLocal(EmitContext& ctx, Id word_offset);

37
src/shader_recompiler/backend/spirv/emit_spirv_barriers.cpp
View file

@ -7,34 +7,27 @@
namespace Shader::Backend::SPIRV {
namespace {
spv::Scope MemoryScopeToSpirVScope(IR::MemoryScope scope) {
switch (scope) {
case IR::MemoryScope::Warp:
return spv::Scope::Subgroup;
case IR::MemoryScope::Workgroup:
return spv::Scope::Workgroup;
case IR::MemoryScope::Device:
return spv::Scope::Device;
case IR::MemoryScope::System:
return spv::Scope::CrossDevice;
case IR::MemoryScope::DontCare:
return spv::Scope::Invocation;
default:
throw NotImplementedException("Unknown memory scope!");
}
}
} // namespace
void EmitMemoryBarrier(EmitContext& ctx, IR::Inst* inst) {
const auto info{inst->Flags<IR::BarrierInstInfo>()};
void EmitMemoryBarrierImpl(EmitContext& ctx, spv::Scope scope) {
const auto semantics =
spv::MemorySemanticsMask::AcquireRelease | spv::MemorySemanticsMask::UniformMemory |
spv::MemorySemanticsMask::WorkgroupMemory | spv::MemorySemanticsMask::AtomicCounterMemory |
spv::MemorySemanticsMask::ImageMemory;
const auto scope = MemoryScopeToSpirVScope(info.scope);
ctx.OpMemoryBarrier(ctx.Constant(ctx.U32[1], static_cast<u32>(scope)),
ctx.Constant(ctx.U32[1], static_cast<u32>(semantics)));
}
} // Anonymous namespace
void EmitMemoryBarrierWorkgroupLevel(EmitContext& ctx) {
EmitMemoryBarrierImpl(ctx, spv::Scope::Workgroup);
}
void EmitMemoryBarrierDeviceLevel(EmitContext& ctx) {
EmitMemoryBarrierImpl(ctx, spv::Scope::Device);
}
void EmitMemoryBarrierSystemLevel(EmitContext& ctx) {
EmitMemoryBarrierImpl(ctx, spv::Scope::CrossDevice);
}
} // namespace Shader::Backend::SPIRV

39
src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
View file

@ -58,7 +58,10 @@ std::optional<Id> OutputAttrPointer(EmitContext& ctx, IR::Attribute attr) {
return ctx.OpAccessChain(ctx.output_f32, ctx.clip_distances, clip_num);
}
case IR::Attribute::ViewportIndex:
return ctx.ignore_viewport_layer ? std::nullopt : std::optional<Id>{ctx.viewport_index};
return (ctx.profile.support_viewport_index_layer_non_geometry ||
ctx.stage == Shader::Stage::Geometry)
? std::optional<Id>{ctx.viewport_index}
: std::nullopt;
default:
throw NotImplementedException("Read attribute {}", attr);
}
@ -206,7 +209,7 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr) {
}
void EmitSetAttribute(EmitContext& ctx, IR::Attribute attr, Id value) {
auto output = OutputAttrPointer(ctx, attr);
const std::optional<Id> output{OutputAttrPointer(ctx, attr)};
if (!output) {
return;
}
@ -263,38 +266,6 @@ void EmitSetOFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
void EmitGetFCSMFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
void EmitGetTAFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
void EmitGetTRFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
void EmitGetMXFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
void EmitSetFCSMFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
void EmitSetTAFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
void EmitSetTRFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
void EmitSetMXFlag(EmitContext&) {
throw NotImplementedException("SPIR-V Instruction");
}
Id EmitWorkgroupId(EmitContext& ctx) {
return ctx.OpLoad(ctx.U32[3], ctx.workgroup_id);
}

49
src/shader_recompiler/frontend/ir/ir_emitter.cpp
View file

@ -82,8 +82,17 @@ void IREmitter::SelectionMerge(Block* merge_block) {
Inst(Opcode::SelectionMerge, merge_block);
}
void IREmitter::MemoryBarrier(BarrierInstInfo info) {
Inst(Opcode::MemoryBarrier, Flags{info});
void IREmitter::MemoryBarrier(MemoryScope scope) {
switch (scope) {
case MemoryScope::Workgroup:
Inst(Opcode::MemoryBarrierWorkgroupLevel);
case MemoryScope::Device:
Inst(Opcode::MemoryBarrierDeviceLevel);
case MemoryScope::System:
Inst(Opcode::MemoryBarrierSystemLevel);
default:
throw InvalidArgument("Invalid memory scope {}", scope);
}
}
void IREmitter::Return() {
@ -202,38 +211,6 @@ void IREmitter::SetOFlag(const U1& value) {
Inst(Opcode::SetOFlag, value);
}
U1 IREmitter::GetFCSMFlag() {
return Inst<U1>(Opcode::GetFCSMFlag);
}
U1 IREmitter::GetTAFlag() {
return Inst<U1>(Opcode::GetTAFlag);
}
U1 IREmitter::GetTRFlag() {
return Inst<U1>(Opcode::GetTRFlag);
}
U1 IREmitter::GetMXFlag() {
return Inst<U1>(Opcode::GetMXFlag);
}
void IREmitter::SetFCSMFlag(const U1& value) {
Inst(Opcode::SetFCSMFlag, value);
}
void IREmitter::SetTAFlag(const U1& value) {
Inst(Opcode::SetTAFlag, value);
}
void IREmitter::SetTRFlag(const U1& value) {
Inst(Opcode::SetTRFlag, value);
}
void IREmitter::SetMXFlag(const U1& value) {
Inst(Opcode::SetMXFlag, value);
}
static U1 GetFlowTest(IREmitter& ir, FlowTest flow_test) {
switch (flow_test) {
case FlowTest::F:
@ -292,9 +269,9 @@ static U1 GetFlowTest(IREmitter& ir, FlowTest flow_test) {
return ir.LogicalOr(ir.GetSFlag(), ir.GetZFlag());
case FlowTest::RGT:
return ir.LogicalAnd(ir.LogicalNot(ir.GetSFlag()), ir.LogicalNot(ir.GetZFlag()));
case FlowTest::FCSM_TR:
return ir.LogicalAnd(ir.GetFCSMFlag(), ir.GetTRFlag());
// LOG_WARNING(ShaderDecompiler, "FCSM_TR CC State (Stubbed)");
return ir.Imm1(false);
case FlowTest::CSM_TA:
case FlowTest::CSM_TR:
case FlowTest::CSM_MX:

12
src/shader_recompiler/frontend/ir/ir_emitter.h
View file

@ -70,16 +70,6 @@ public:
void SetCFlag(const U1& value);
void SetOFlag(const U1& value);
[[nodiscard]] U1 GetFCSMFlag();
[[nodiscard]] U1 GetTAFlag();
[[nodiscard]] U1 GetTRFlag();
[[nodiscard]] U1 GetMXFlag();
void SetFCSMFlag(const U1& value);
void SetTAFlag(const U1& value);
void SetTRFlag(const U1& value);
void SetMXFlag(const U1& value);
[[nodiscard]] U1 Condition(IR::Condition cond);
[[nodiscard]] U1 GetFlowTestResult(FlowTest test);
@ -138,7 +128,7 @@ public:
[[nodiscard]] Value Select(const U1& condition, const Value& true_value,
const Value& false_value);
[[nodiscard]] void MemoryBarrier(BarrierInstInfo info);
[[nodiscard]] void MemoryBarrier(MemoryScope scope);
template <typename Dest, typename Source>
[[nodiscard]] Dest BitCast(const Source& value);

13
src/shader_recompiler/frontend/ir/modifiers.h
View file

@ -25,13 +25,7 @@ enum class FpRounding : u8 {
RZ, // Round towards zero
};
enum class MemoryScope : u32 {
DontCare,
Warp,
Workgroup,
Device,
System
};
enum class MemoryScope : u32 { DontCare, Warp, Workgroup, Device, System };
struct FpControl {
bool no_contraction{false};
@ -40,11 +34,6 @@ struct FpControl {
};
static_assert(sizeof(FpControl) <= sizeof(u32));
union BarrierInstInfo {
u32 raw;
BitField<0, 3, MemoryScope> scope;
};
union TextureInstInfo {
u32 raw;
BitField<0, 8, TextureType> type;

12
src/shader_recompiler/frontend/ir/opcodes.inc
View file

@ -17,7 +17,9 @@ OPCODE(Unreachable, Void,
OPCODE(DemoteToHelperInvocation, Void, Label, )
// Barriers
OPCODE(MemoryBarrier, Void, )
OPCODE(MemoryBarrierWorkgroupLevel, Void, )
OPCODE(MemoryBarrierDeviceLevel, Void, )
OPCODE(MemoryBarrierSystemLevel, Void, )
// Special operations
OPCODE(Prologue, Void, )
@ -49,18 +51,10 @@ OPCODE(GetZFlag, U1, Void
OPCODE(GetSFlag, U1, Void, )
OPCODE(GetCFlag, U1, Void, )
OPCODE(GetOFlag, U1, Void, )
OPCODE(GetFCSMFlag, U1, Void, )
OPCODE(GetTAFlag, U1, Void, )
OPCODE(GetTRFlag, U1, Void, )
OPCODE(GetMXFlag, U1, Void, )
OPCODE(SetZFlag, Void, U1, )
OPCODE(SetSFlag, Void, U1, )
OPCODE(SetCFlag, Void, U1, )
OPCODE(SetOFlag, Void, U1, )
OPCODE(SetFCSMFlag, Void, U1, )
OPCODE(SetTAFlag, Void, U1, )
OPCODE(SetTRFlag, Void, U1, )
OPCODE(SetMXFlag, Void, U1, )
OPCODE(WorkgroupId, U32x3, )
OPCODE(LocalInvocationId, U32x3, )
OPCODE(LaneId, U32, )

12
src/shader_recompiler/frontend/maxwell/translate/impl/barrier_operations.cpp
View file

@ -5,8 +5,8 @@
#include "common/bit_field.h"
#include "common/common_types.h"
#include "shader_recompiler/frontend/ir/modifiers.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
#include "shader_recompiler/frontend/maxwell/opcodes.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
namespace Shader::Maxwell {
namespace {
@ -21,28 +21,24 @@ enum class LocalScope : u64 {
IR::MemoryScope LocalScopeToMemoryScope(LocalScope scope) {
switch (scope) {
case LocalScope::CTG:
return IR::MemoryScope::Warp;
return IR::MemoryScope::Workgroup;
case LocalScope::GL:
return IR::MemoryScope::Device;
case LocalScope::SYS:
return IR::MemoryScope::System;
case LocalScope::VC:
return IR::MemoryScope::Workgroup; // or should be device?
default:
throw NotImplementedException("Unimplemented Local Scope {}", scope);
}
}
} // namespace
} // Anonymous namespace
void TranslatorVisitor::MEMBAR(u64 inst) {
union {
u64 raw;
BitField<8, 2, LocalScope> scope;
} membar{inst};
IR::BarrierInstInfo info{};
info.scope.Assign(LocalScopeToMemoryScope(membar.scope));
ir.MemoryBarrier(info);
ir.MemoryBarrier(LocalScopeToMemoryScope(membar.scope));
}
void TranslatorVisitor::DEPBAR() {

2
src/shader_recompiler/frontend/maxwell/translate/impl/move_special_register.cpp
View file

@ -96,8 +96,10 @@ enum class SpecialRegister : u64 {
case SpecialRegister::SR_CTAID_Z:
return ir.WorkgroupIdZ();
case SpecialRegister::SR_WSCALEFACTOR_XY:
// LOG_WARNING(ShaderDecompiler, "SR_WSCALEFACTOR_XY (Stubbed)");
return ir.Imm32(Common::BitCast<u32>(1.0f));
case SpecialRegister::SR_WSCALEFACTOR_Z:
// LOG_WARNING(ShaderDecompiler, "SR_WSCALEFACTOR_Z (Stubbed)");
return ir.Imm32(Common::BitCast<u32>(1.0f));
case SpecialRegister::SR_LANEID:
return ir.LaneId();

5
src/shader_recompiler/frontend/maxwell/translate/impl/vote.cpp
View file

@ -50,10 +50,7 @@ void TranslatorVisitor::VOTE(u64 insn) {
}
void TranslatorVisitor::VOTE_vtg(u64) {
// LOG_WARNING("VOTE.VTG: Stubbed!");
auto imm = ir.Imm1(false);
ir.SetFCSMFlag(imm);
ir.SetTRFlag(imm);
// LOG_WARNING(ShaderDecompiler, "VOTE.VTG: Stubbed!");
}
} // namespace Shader::Maxwell

9
src/shader_recompiler/ir_opt/constant_propagation_pass.cpp
View file

@ -374,17 +374,14 @@ std::optional<IR::Value> FoldCompositeExtractImpl(IR::Value inst_value, IR::Opco
if (inst->Opcode() == construct) {
return inst->Arg(first_index);
}
if (inst->Opcode() != insert) {
return std::nullopt;
}
IR::Value value_index{inst->Arg(2)};
if (!value_index.IsImmediate()) {
return std::nullopt;
}
const u32 second_index = value_index.U32();
const u32 second_index{value_index.U32()};
if (first_index != second_index) {
IR::Value value_composite{inst->Arg(0)};
if (value_composite.IsImmediate()) {
@ -404,8 +401,8 @@ void FoldCompositeExtract(IR::Inst& inst, IR::Opcode construct, IR::Opcode inser
if (!value_2.IsImmediate()) {
return;
}
const u32 first_index = value_2.U32();
auto result = FoldCompositeExtractImpl(value_1, insert, construct, first_index);
const u32 first_index{value_2.U32()};
const std::optional result{FoldCompositeExtractImpl(value_1, insert, construct, first_index)};
if (!result) {
return;
}

10
src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp
View file

@ -4,9 +4,9 @@
#include <algorithm>
#include <compare>
#include <map>
#include <optional>
#include <ranges>
#include <map>
#include <boost/container/flat_set.hpp>
#include <boost/container/small_vector.hpp>
@ -295,12 +295,12 @@ void CollectStorageBuffers(IR::Block& block, IR::Inst& inst, StorageBufferSet& s
}
}
// Collect storage buffer and the instruction
const bool is_a_write = IsGlobalMemoryWrite(inst);
auto it = writes_map.find(*storage_buffer);
const bool is_a_write{IsGlobalMemoryWrite(inst)};
auto it{writes_map.find(*storage_buffer)};
if (it == writes_map.end()) {
writes_map[*storage_buffer] = is_a_write;
writes_map[*storage_buffer] = is_a_write;
} else {
it->second = it->second || is_a_write;
it->second = it->second || is_a_write;
}
storage_buffer_set.insert(*storage_buffer);
to_replace.push_back(StorageInst{

50
src/shader_recompiler/ir_opt/ssa_rewrite_pass.cpp
View file

@ -38,10 +38,6 @@ struct ZeroFlagTag : FlagTag {};
struct SignFlagTag : FlagTag {};
struct CarryFlagTag : FlagTag {};
struct OverflowFlagTag : FlagTag {};
struct FCSMFlagTag : FlagTag {};
struct TAFlagTag : FlagTag {};
struct TRFlagTag : FlagTag {};
struct MXFlagTag : FlagTag {};
struct GotoVariable : FlagTag {
GotoVariable() = default;
@ -57,8 +53,7 @@ struct IndirectBranchVariable {
};
using Variant = std::variant<IR::Reg, IR::Pred, ZeroFlagTag, SignFlagTag, CarryFlagTag,
OverflowFlagTag, FCSMFlagTag, TAFlagTag, TRFlagTag, MXFlagTag,
GotoVariable, IndirectBranchVariable>;
OverflowFlagTag, GotoVariable, IndirectBranchVariable>;
using ValueMap = boost::container::flat_map<IR::Block*, IR::Value, std::less<IR::Block*>>;
struct DefTable {
@ -94,22 +89,6 @@ struct DefTable {
return overflow_flag;
}
[[nodiscard]] ValueMap& operator[](FCSMFlagTag) noexcept {
return fcsm_flag;
}
[[nodiscard]] ValueMap& operator[](TAFlagTag) noexcept {
return ta_flag;
}
[[nodiscard]] ValueMap& operator[](TRFlagTag) noexcept {
return tr_flag;
}
[[nodiscard]] ValueMap& operator[](MXFlagTag) noexcept {
return mr_flag;
}
std::array<ValueMap, IR::NUM_USER_REGS> regs;
std::array<ValueMap, IR::NUM_USER_PREDS> preds;
boost::container::flat_map<u32, ValueMap> goto_vars;
@ -118,10 +97,6 @@ struct DefTable {
ValueMap sign_flag;
ValueMap carry_flag;
ValueMap overflow_flag;
ValueMap fcsm_flag;
ValueMap ta_flag;
ValueMap tr_flag;
ValueMap mr_flag;
};
IR::Opcode UndefOpcode(IR::Reg) noexcept {
@ -272,18 +247,6 @@ void VisitInst(Pass& pass, IR::Block* block, IR::Inst& inst) {
case IR::Opcode::SetOFlag:
pass.WriteVariable(OverflowFlagTag{}, block, inst.Arg(0));
break;
case IR::Opcode::SetFCSMFlag:
pass.WriteVariable(FCSMFlagTag{}, block, inst.Arg(0));
break;
case IR::Opcode::SetTAFlag:
pass.WriteVariable(TAFlagTag{}, block, inst.Arg(0));
break;
case IR::Opcode::SetTRFlag:
pass.WriteVariable(TRFlagTag{}, block, inst.Arg(0));
break;
case IR::Opcode::SetMXFlag:
pass.WriteVariable(MXFlagTag{}, block, inst.Arg(0));
break;
case IR::Opcode::GetRegister:
if (const IR::Reg reg{inst.Arg(0).Reg()}; reg != IR::Reg::RZ) {
inst.ReplaceUsesWith(pass.ReadVariable(reg, block));
@ -312,17 +275,6 @@ void VisitInst(Pass& pass, IR::Block* block, IR::Inst& inst) {
case IR::Opcode::GetOFlag:
inst.ReplaceUsesWith(pass.ReadVariable(OverflowFlagTag{}, block));
break;
case IR::Opcode::GetFCSMFlag:
inst.ReplaceUsesWith(pass.ReadVariable(FCSMFlagTag{}, block));
break;
case IR::Opcode::GetTAFlag:
inst.ReplaceUsesWith(pass.ReadVariable(TAFlagTag{}, block));
break;
case IR::Opcode::GetTRFlag:
inst.ReplaceUsesWith(pass.ReadVariable(TRFlagTag{}, block));
break;
case IR::Opcode::GetMXFlag:
inst.ReplaceUsesWith(pass.ReadVariable(MXFlagTag{}, block));
break;
default:
break;