summaryrefslogtreecommitdiffstats
path: root/src/shader_recompiler/frontend/ir
diff options
context:
space:
mode:
authorReinUsesLisp <reinuseslisp@airmail.cc>2021-02-03 20:43:04 +0100
committerameerj <52414509+ameerj@users.noreply.github.com>2021-07-23 03:51:21 +0200
commitd24a16045f0f6b0b873d5e3b5bf187c1a8c4343f (patch)
tree0108a028b437bc59dfe7864f333cf4c50a46d3b5 /src/shader_recompiler/frontend/ir
parentshader: SSA and dominance (diff)
downloadyuzu-d24a16045f0f6b0b873d5e3b5bf187c1a8c4343f.tar
yuzu-d24a16045f0f6b0b873d5e3b5bf187c1a8c4343f.tar.gz
yuzu-d24a16045f0f6b0b873d5e3b5bf187c1a8c4343f.tar.bz2
yuzu-d24a16045f0f6b0b873d5e3b5bf187c1a8c4343f.tar.lz
yuzu-d24a16045f0f6b0b873d5e3b5bf187c1a8c4343f.tar.xz
yuzu-d24a16045f0f6b0b873d5e3b5bf187c1a8c4343f.tar.zst
yuzu-d24a16045f0f6b0b873d5e3b5bf187c1a8c4343f.zip
Diffstat (limited to 'src/shader_recompiler/frontend/ir')
-rw-r--r--src/shader_recompiler/frontend/ir/basic_block.cpp4
-rw-r--r--src/shader_recompiler/frontend/ir/basic_block.h2
-rw-r--r--src/shader_recompiler/frontend/ir/ir_emitter.cpp200
-rw-r--r--src/shader_recompiler/frontend/ir/ir_emitter.h67
-rw-r--r--src/shader_recompiler/frontend/ir/microinstruction.h12
-rw-r--r--src/shader_recompiler/frontend/ir/modifiers.h28
-rw-r--r--src/shader_recompiler/frontend/ir/opcode.inc139
-rw-r--r--src/shader_recompiler/frontend/ir/pred.h11
8 files changed, 398 insertions, 65 deletions
diff --git a/src/shader_recompiler/frontend/ir/basic_block.cpp b/src/shader_recompiler/frontend/ir/basic_block.cpp
index e795618fc..249251dd0 100644
--- a/src/shader_recompiler/frontend/ir/basic_block.cpp
+++ b/src/shader_recompiler/frontend/ir/basic_block.cpp
@@ -23,8 +23,8 @@ void Block::AppendNewInst(Opcode op, std::initializer_list<Value> args) {
}
Block::iterator Block::PrependNewInst(iterator insertion_point, Opcode op,
- std::initializer_list<Value> args) {
- Inst* const inst{std::construct_at(instruction_alloc_pool.allocate(), op)};
+ std::initializer_list<Value> args, u64 flags) {
+ Inst* const inst{std::construct_at(instruction_alloc_pool.allocate(), op, flags)};
const auto result_it{instructions.insert(insertion_point, *inst)};
if (inst->NumArgs() != args.size()) {
diff --git a/src/shader_recompiler/frontend/ir/basic_block.h b/src/shader_recompiler/frontend/ir/basic_block.h
index 4b6b80c4b..ec4a41cb1 100644
--- a/src/shader_recompiler/frontend/ir/basic_block.h
+++ b/src/shader_recompiler/frontend/ir/basic_block.h
@@ -39,7 +39,7 @@ public:
/// Prepends a new instruction to this basic block before the insertion point.
iterator PrependNewInst(iterator insertion_point, Opcode op,
- std::initializer_list<Value> args = {});
+ std::initializer_list<Value> args = {}, u64 flags = 0);
/// Adds a new immediate predecessor to the basic block.
void AddImmediatePredecessor(IR::Block* immediate_predecessor);
diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.cpp b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
index 6450e4b2c..87b253c9a 100644
--- a/src/shader_recompiler/frontend/ir/ir_emitter.cpp
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.cpp
@@ -129,6 +129,58 @@ void IREmitter::SetAttribute(IR::Attribute attribute, const U32& value) {
Inst(Opcode::SetAttribute, attribute, value);
}
+U32 IREmitter::WorkgroupIdX() {
+ return Inst<U32>(Opcode::WorkgroupIdX);
+}
+
+U32 IREmitter::WorkgroupIdY() {
+ return Inst<U32>(Opcode::WorkgroupIdY);
+}
+
+U32 IREmitter::WorkgroupIdZ() {
+ return Inst<U32>(Opcode::WorkgroupIdZ);
+}
+
+U32 IREmitter::LocalInvocationIdX() {
+ return Inst<U32>(Opcode::LocalInvocationIdX);
+}
+
+U32 IREmitter::LocalInvocationIdY() {
+ return Inst<U32>(Opcode::LocalInvocationIdY);
+}
+
+U32 IREmitter::LocalInvocationIdZ() {
+ return Inst<U32>(Opcode::LocalInvocationIdZ);
+}
+
+U32 IREmitter::LoadGlobalU8(const U64& address) {
+ return Inst<U32>(Opcode::LoadGlobalU8, address);
+}
+
+U32 IREmitter::LoadGlobalS8(const U64& address) {
+ return Inst<U32>(Opcode::LoadGlobalS8, address);
+}
+
+U32 IREmitter::LoadGlobalU16(const U64& address) {
+ return Inst<U32>(Opcode::LoadGlobalU16, address);
+}
+
+U32 IREmitter::LoadGlobalS16(const U64& address) {
+ return Inst<U32>(Opcode::LoadGlobalS16, address);
+}
+
+U32 IREmitter::LoadGlobal32(const U64& address) {
+ return Inst<U32>(Opcode::LoadGlobal32, address);
+}
+
+Value IREmitter::LoadGlobal64(const U64& address) {
+ return Inst<Value>(Opcode::LoadGlobal64, address);
+}
+
+Value IREmitter::LoadGlobal128(const U64& address) {
+ return Inst<Value>(Opcode::LoadGlobal128, address);
+}
+
void IREmitter::WriteGlobalU8(const U64& address, const U32& value) {
Inst(Opcode::WriteGlobalU8, address, value);
}
@@ -173,17 +225,17 @@ U1 IREmitter::GetOverflowFromOp(const Value& op) {
return Inst<U1>(Opcode::GetOverflowFromOp, op);
}
-U16U32U64 IREmitter::FPAdd(const U16U32U64& a, const U16U32U64& b) {
+U16U32U64 IREmitter::FPAdd(const U16U32U64& a, const U16U32U64& b, FpControl control) {
if (a.Type() != a.Type()) {
throw InvalidArgument("Mismatching types {} and {}", a.Type(), b.Type());
}
switch (a.Type()) {
case Type::U16:
- return Inst<U16>(Opcode::FPAdd16, a, b);
+ return Inst<U16>(Opcode::FPAdd16, Flags{control}, a, b);
case Type::U32:
- return Inst<U32>(Opcode::FPAdd32, a, b);
+ return Inst<U32>(Opcode::FPAdd32, Flags{control}, a, b);
case Type::U64:
- return Inst<U64>(Opcode::FPAdd64, a, b);
+ return Inst<U64>(Opcode::FPAdd64, Flags{control}, a, b);
default:
ThrowInvalidType(a.Type());
}
@@ -191,14 +243,14 @@ U16U32U64 IREmitter::FPAdd(const U16U32U64& a, const U16U32U64& b) {
Value IREmitter::CompositeConstruct(const UAny& e1, const UAny& e2) {
if (e1.Type() != e2.Type()) {
- throw InvalidArgument("Incompatible types {} {}", e1.Type(), e2.Type());
+ throw InvalidArgument("Mismatching types {} and {}", e1.Type(), e2.Type());
}
return Inst(Opcode::CompositeConstruct2, e1, e2);
}
Value IREmitter::CompositeConstruct(const UAny& e1, const UAny& e2, const UAny& e3) {
if (e1.Type() != e2.Type() || e1.Type() != e3.Type()) {
- throw InvalidArgument("Incompatible types {} {} {}", e1.Type(), e2.Type(), e3.Type());
+ throw InvalidArgument("Mismatching types {}, {}, and {}", e1.Type(), e2.Type(), e3.Type());
}
return Inst(Opcode::CompositeConstruct3, e1, e2, e3);
}
@@ -206,8 +258,8 @@ Value IREmitter::CompositeConstruct(const UAny& e1, const UAny& e2, const UAny&
Value IREmitter::CompositeConstruct(const UAny& e1, const UAny& e2, const UAny& e3,
const UAny& e4) {
if (e1.Type() != e2.Type() || e1.Type() != e3.Type() || e1.Type() != e4.Type()) {
- throw InvalidArgument("Incompatible types {} {} {}", e1.Type(), e2.Type(), e3.Type(),
- e4.Type());
+ throw InvalidArgument("Mismatching types {}, {}, {}, and {}", e1.Type(), e2.Type(),
+ e3.Type(), e4.Type());
}
return Inst(Opcode::CompositeConstruct4, e1, e2, e3, e4);
}
@@ -219,6 +271,24 @@ UAny IREmitter::CompositeExtract(const Value& vector, size_t element) {
return Inst<UAny>(Opcode::CompositeExtract, vector, Imm32(static_cast<u32>(element)));
}
+UAny IREmitter::Select(const U1& condition, const UAny& true_value, const UAny& false_value) {
+ if (true_value.Type() != false_value.Type()) {
+ throw InvalidArgument("Mismatching types {} and {}", true_value.Type(), false_value.Type());
+ }
+ switch (true_value.Type()) {
+ case Type::U8:
+ return Inst<UAny>(Opcode::Select8, condition, true_value, false_value);
+ case Type::U16:
+ return Inst<UAny>(Opcode::Select16, condition, true_value, false_value);
+ case Type::U32:
+ return Inst<UAny>(Opcode::Select32, condition, true_value, false_value);
+ case Type::U64:
+ return Inst<UAny>(Opcode::Select64, condition, true_value, false_value);
+ default:
+ throw InvalidArgument("Invalid type {}", true_value.Type());
+ }
+}
+
U64 IREmitter::PackUint2x32(const Value& vector) {
return Inst<U64>(Opcode::PackUint2x32, vector);
}
@@ -243,17 +313,34 @@ Value IREmitter::UnpackDouble2x32(const U64& value) {
return Inst<Value>(Opcode::UnpackDouble2x32, value);
}
-U16U32U64 IREmitter::FPMul(const U16U32U64& a, const U16U32U64& b) {
+U16U32U64 IREmitter::FPMul(const U16U32U64& a, const U16U32U64& b, FpControl control) {
if (a.Type() != b.Type()) {
throw InvalidArgument("Mismatching types {} and {}", a.Type(), b.Type());
}
switch (a.Type()) {
case Type::U16:
- return Inst<U16>(Opcode::FPMul16, a, b);
+ return Inst<U16>(Opcode::FPMul16, Flags{control}, a, b);
case Type::U32:
- return Inst<U32>(Opcode::FPMul32, a, b);
+ return Inst<U32>(Opcode::FPMul32, Flags{control}, a, b);
case Type::U64:
- return Inst<U64>(Opcode::FPMul64, a, b);
+ return Inst<U64>(Opcode::FPMul64, Flags{control}, a, b);
+ default:
+ ThrowInvalidType(a.Type());
+ }
+}
+
+U16U32U64 IREmitter::FPFma(const U16U32U64& a, const U16U32U64& b, const U16U32U64& c,
+ FpControl control) {
+ if (a.Type() != b.Type() || a.Type() != c.Type()) {
+ throw InvalidArgument("Mismatching types {}, {}, and {}", a.Type(), b.Type(), c.Type());
+ }
+ switch (a.Type()) {
+ case Type::U16:
+ return Inst<U16>(Opcode::FPFma16, Flags{control}, a, b, c);
+ case Type::U32:
+ return Inst<U32>(Opcode::FPFma32, Flags{control}, a, b, c);
+ case Type::U64:
+ return Inst<U64>(Opcode::FPFma64, Flags{control}, a, b, c);
default:
ThrowInvalidType(a.Type());
}
@@ -403,6 +490,91 @@ U16U32U64 IREmitter::FPTrunc(const U16U32U64& value) {
}
}
+U32U64 IREmitter::IAdd(const U32U64& a, const U32U64& b) {
+ if (a.Type() != b.Type()) {
+ throw InvalidArgument("Mismatching types {} and {}", a.Type(), b.Type());
+ }
+ switch (a.Type()) {
+ case Type::U32:
+ return Inst<U32>(Opcode::IAdd32, a, b);
+ case Type::U64:
+ return Inst<U64>(Opcode::IAdd64, a, b);
+ default:
+ ThrowInvalidType(a.Type());
+ }
+}
+
+U32 IREmitter::IMul(const U32& a, const U32& b) {
+ return Inst<U32>(Opcode::IMul32, a, b);
+}
+
+U32 IREmitter::INeg(const U32& value) {
+ return Inst<U32>(Opcode::INeg32, value);
+}
+
+U32 IREmitter::IAbs(const U32& value) {
+ return Inst<U32>(Opcode::IAbs32, value);
+}
+
+U32 IREmitter::ShiftLeftLogical(const U32& base, const U32& shift) {
+ return Inst<U32>(Opcode::ShiftLeftLogical32, base, shift);
+}
+
+U32 IREmitter::ShiftRightLogical(const U32& base, const U32& shift) {
+ return Inst<U32>(Opcode::ShiftRightLogical32, base, shift);
+}
+
+U32 IREmitter::ShiftRightArithmetic(const U32& base, const U32& shift) {
+ return Inst<U32>(Opcode::ShiftRightArithmetic32, base, shift);
+}
+
+U32 IREmitter::BitwiseAnd(const U32& a, const U32& b) {
+ return Inst<U32>(Opcode::BitwiseAnd32, a, b);
+}
+
+U32 IREmitter::BitwiseOr(const U32& a, const U32& b) {
+ return Inst<U32>(Opcode::BitwiseOr32, a, b);
+}
+
+U32 IREmitter::BitwiseXor(const U32& a, const U32& b) {
+ return Inst<U32>(Opcode::BitwiseXor32, a, b);
+}
+
+U32 IREmitter::BitFieldInsert(const U32& base, const U32& insert, const U32& offset,
+ const U32& count) {
+ return Inst<U32>(Opcode::BitFieldInsert, base, insert, offset, count);
+}
+
+U32 IREmitter::BitFieldExtract(const U32& base, const U32& offset, const U32& count,
+ bool is_signed) {
+ return Inst<U32>(is_signed ? Opcode::BitFieldSExtract : Opcode::BitFieldUExtract, base, offset,
+ count);
+}
+
+U1 IREmitter::ILessThan(const U32& lhs, const U32& rhs, bool is_signed) {
+ return Inst<U1>(is_signed ? Opcode::SLessThan : Opcode::ULessThan, lhs, rhs);
+}
+
+U1 IREmitter::IEqual(const U32& lhs, const U32& rhs) {
+ return Inst<U1>(Opcode::IEqual, lhs, rhs);
+}
+
+U1 IREmitter::ILessThanEqual(const U32& lhs, const U32& rhs, bool is_signed) {
+ return Inst<U1>(is_signed ? Opcode::SLessThanEqual : Opcode::ULessThanEqual, lhs, rhs);
+}
+
+U1 IREmitter::IGreaterThan(const U32& lhs, const U32& rhs, bool is_signed) {
+ return Inst<U1>(is_signed ? Opcode::SGreaterThan : Opcode::UGreaterThan, lhs, rhs);
+}
+
+U1 IREmitter::INotEqual(const U32& lhs, const U32& rhs) {
+ return Inst<U1>(Opcode::INotEqual, lhs, rhs);
+}
+
+U1 IREmitter::IGreaterThanEqual(const U32& lhs, const U32& rhs, bool is_signed) {
+ return Inst<U1>(is_signed ? Opcode::SGreaterThanEqual : Opcode::UGreaterThanEqual, lhs, rhs);
+}
+
U1 IREmitter::LogicalOr(const U1& a, const U1& b) {
return Inst<U1>(Opcode::LogicalOr, a, b);
}
@@ -411,6 +583,10 @@ U1 IREmitter::LogicalAnd(const U1& a, const U1& b) {
return Inst<U1>(Opcode::LogicalAnd, a, b);
}
+U1 IREmitter::LogicalXor(const U1& a, const U1& b) {
+ return Inst<U1>(Opcode::LogicalXor, a, b);
+}
+
U1 IREmitter::LogicalNot(const U1& value) {
return Inst<U1>(Opcode::LogicalNot, value);
}
diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.h b/src/shader_recompiler/frontend/ir/ir_emitter.h
index 1af79f41c..7ff763ecf 100644
--- a/src/shader_recompiler/frontend/ir/ir_emitter.h
+++ b/src/shader_recompiler/frontend/ir/ir_emitter.h
@@ -4,8 +4,12 @@
#pragma once
+#include <cstring>
+#include <type_traits>
+
#include "shader_recompiler/frontend/ir/attribute.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
+#include "shader_recompiler/frontend/ir/modifiers.h"
#include "shader_recompiler/frontend/ir/value.h"
namespace Shader::IR {
@@ -52,6 +56,22 @@ public:
[[nodiscard]] U32 GetAttribute(IR::Attribute attribute);
void SetAttribute(IR::Attribute attribute, const U32& value);
+ [[nodiscard]] U32 WorkgroupIdX();
+ [[nodiscard]] U32 WorkgroupIdY();
+ [[nodiscard]] U32 WorkgroupIdZ();
+
+ [[nodiscard]] U32 LocalInvocationIdX();
+ [[nodiscard]] U32 LocalInvocationIdY();
+ [[nodiscard]] U32 LocalInvocationIdZ();
+
+ [[nodiscard]] U32 LoadGlobalU8(const U64& address);
+ [[nodiscard]] U32 LoadGlobalS8(const U64& address);
+ [[nodiscard]] U32 LoadGlobalU16(const U64& address);
+ [[nodiscard]] U32 LoadGlobalS16(const U64& address);
+ [[nodiscard]] U32 LoadGlobal32(const U64& address);
+ [[nodiscard]] Value LoadGlobal64(const U64& address);
+ [[nodiscard]] Value LoadGlobal128(const U64& address);
+
void WriteGlobalU8(const U64& address, const U32& value);
void WriteGlobalS8(const U64& address, const U32& value);
void WriteGlobalU16(const U64& address, const U32& value);
@@ -71,6 +91,8 @@ public:
const UAny& e4);
[[nodiscard]] UAny CompositeExtract(const Value& vector, size_t element);
+ [[nodiscard]] UAny Select(const U1& condition, const UAny& true_value, const UAny& false_value);
+
[[nodiscard]] U64 PackUint2x32(const Value& vector);
[[nodiscard]] Value UnpackUint2x32(const U64& value);
@@ -80,8 +102,10 @@ public:
[[nodiscard]] U64 PackDouble2x32(const Value& vector);
[[nodiscard]] Value UnpackDouble2x32(const U64& value);
- [[nodiscard]] U16U32U64 FPAdd(const U16U32U64& a, const U16U32U64& b);
- [[nodiscard]] U16U32U64 FPMul(const U16U32U64& a, const U16U32U64& b);
+ [[nodiscard]] U16U32U64 FPAdd(const U16U32U64& a, const U16U32U64& b, FpControl control = {});
+ [[nodiscard]] U16U32U64 FPMul(const U16U32U64& a, const U16U32U64& b, FpControl control = {});
+ [[nodiscard]] U16U32U64 FPFma(const U16U32U64& a, const U16U32U64& b, const U16U32U64& c,
+ FpControl control = {});
[[nodiscard]] U16U32U64 FPAbs(const U16U32U64& value);
[[nodiscard]] U16U32U64 FPNeg(const U16U32U64& value);
@@ -100,8 +124,31 @@ public:
[[nodiscard]] U16U32U64 FPCeil(const U16U32U64& value);
[[nodiscard]] U16U32U64 FPTrunc(const U16U32U64& value);
+ [[nodiscard]] U32U64 IAdd(const U32U64& a, const U32U64& b);
+ [[nodiscard]] U32 IMul(const U32& a, const U32& b);
+ [[nodiscard]] U32 INeg(const U32& value);
+ [[nodiscard]] U32 IAbs(const U32& value);
+ [[nodiscard]] U32 ShiftLeftLogical(const U32& base, const U32& shift);
+ [[nodiscard]] U32 ShiftRightLogical(const U32& base, const U32& shift);
+ [[nodiscard]] U32 ShiftRightArithmetic(const U32& base, const U32& shift);
+ [[nodiscard]] U32 BitwiseAnd(const U32& a, const U32& b);
+ [[nodiscard]] U32 BitwiseOr(const U32& a, const U32& b);
+ [[nodiscard]] U32 BitwiseXor(const U32& a, const U32& b);
+ [[nodiscard]] U32 BitFieldInsert(const U32& base, const U32& insert, const U32& offset,
+ const U32& count);
+ [[nodiscard]] U32 BitFieldExtract(const U32& base, const U32& offset, const U32& count,
+ bool is_signed);
+
+ [[nodiscard]] U1 ILessThan(const U32& lhs, const U32& rhs, bool is_signed);
+ [[nodiscard]] U1 IEqual(const U32& lhs, const U32& rhs);
+ [[nodiscard]] U1 ILessThanEqual(const U32& lhs, const U32& rhs, bool is_signed);
+ [[nodiscard]] U1 IGreaterThan(const U32& lhs, const U32& rhs, bool is_signed);
+ [[nodiscard]] U1 INotEqual(const U32& lhs, const U32& rhs);
+ [[nodiscard]] U1 IGreaterThanEqual(const U32& lhs, const U32& rhs, bool is_signed);
+
[[nodiscard]] U1 LogicalOr(const U1& a, const U1& b);
[[nodiscard]] U1 LogicalAnd(const U1& a, const U1& b);
+ [[nodiscard]] U1 LogicalXor(const U1& a, const U1& b);
[[nodiscard]] U1 LogicalNot(const U1& value);
[[nodiscard]] U32U64 ConvertFToS(size_t bitsize, const U16U32U64& value);
@@ -118,6 +165,22 @@ private:
auto it{block.PrependNewInst(insertion_point, op, {Value{args}...})};
return T{Value{&*it}};
}
+
+ template <typename T>
+ requires(sizeof(T) <= sizeof(u64) && std::is_trivially_copyable_v<T>) struct Flags {
+ Flags() = default;
+ Flags(T proxy_) : proxy{proxy_} {}
+
+ T proxy;
+ };
+
+ template <typename T = Value, typename FlagType, typename... Args>
+ T Inst(Opcode op, Flags<FlagType> flags, Args... args) {
+ u64 raw_flags{};
+ std::memcpy(&raw_flags, &flags.proxy, sizeof(flags.proxy));
+ auto it{block.PrependNewInst(insertion_point, op, {Value{args}...}, raw_flags)};
+ return T{Value{&*it}};
+ }
};
} // namespace Shader::IR
diff --git a/src/shader_recompiler/frontend/ir/microinstruction.h b/src/shader_recompiler/frontend/ir/microinstruction.h
index 7f1ed6710..61849695a 100644
--- a/src/shader_recompiler/frontend/ir/microinstruction.h
+++ b/src/shader_recompiler/frontend/ir/microinstruction.h
@@ -5,7 +5,9 @@
#pragma once
#include <array>
+#include <cstring>
#include <span>
+#include <type_traits>
#include <vector>
#include <boost/intrusive/list.hpp>
@@ -23,7 +25,7 @@ constexpr size_t MAX_ARG_COUNT = 4;
class Inst : public boost::intrusive::list_base_hook<> {
public:
- explicit Inst(Opcode op_) noexcept : op(op_) {}
+ explicit Inst(Opcode op_, u64 flags_) noexcept : op{op_}, flags{flags_} {}
/// Get the number of uses this instruction has.
[[nodiscard]] int UseCount() const noexcept {
@@ -73,6 +75,14 @@ public:
void ReplaceUsesWith(Value replacement);
+ template <typename FlagsType>
+ requires(sizeof(FlagsType) <= sizeof(u64) && std::is_trivially_copyable_v<FlagsType>)
+ [[nodiscard]] FlagsType Flags() const noexcept {
+ FlagsType ret;
+ std::memcpy(&ret, &flags, sizeof(ret));
+ return ret;
+ }
+
private:
void Use(const Value& value);
void UndoUse(const Value& value);
diff --git a/src/shader_recompiler/frontend/ir/modifiers.h b/src/shader_recompiler/frontend/ir/modifiers.h
new file mode 100644
index 000000000..28bb9e798
--- /dev/null
+++ b/src/shader_recompiler/frontend/ir/modifiers.h
@@ -0,0 +1,28 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+namespace Shader::IR {
+
+enum class FmzMode {
+ None, // Denorms are not flushed, NAN is propagated (nouveau)
+ FTZ, // Flush denorms to zero, NAN is propagated (D3D11, NVN, GL, VK)
+ FMZ, // Flush denorms to zero, x * 0 == 0 (D3D9)
+};
+
+enum class FpRounding {
+ RN, // Round to nearest even,
+ RM, // Round towards negative infinity
+ RP, // Round towards positive infinity
+ RZ, // Round towards zero
+};
+
+struct FpControl {
+ bool no_contraction{false};
+ FpRounding rounding : 8 = FpRounding::RN;
+ FmzMode fmz_mode : 8 = FmzMode::FTZ;
+};
+static_assert(sizeof(FpControl) <= sizeof(u64));
+} // namespace Shader::IR
diff --git a/src/shader_recompiler/frontend/ir/opcode.inc b/src/shader_recompiler/frontend/ir/opcode.inc
index 40759e96a..4ecb5e936 100644
--- a/src/shader_recompiler/frontend/ir/opcode.inc
+++ b/src/shader_recompiler/frontend/ir/opcode.inc
@@ -35,6 +35,12 @@ OPCODE(SetZFlag, Void, U1,
OPCODE(SetSFlag, Void, U1, )
OPCODE(SetCFlag, Void, U1, )
OPCODE(SetOFlag, Void, U1, )
+OPCODE(WorkgroupIdX, U32, )
+OPCODE(WorkgroupIdY, U32, )
+OPCODE(WorkgroupIdZ, U32, )
+OPCODE(LocalInvocationIdX, U32, )
+OPCODE(LocalInvocationIdY, U32, )
+OPCODE(LocalInvocationIdZ, U32, )
// Undefined
OPCODE(Undef1, U1, )
@@ -44,6 +50,13 @@ OPCODE(Undef32, U32,
OPCODE(Undef64, U64, )
// Memory operations
+OPCODE(LoadGlobalU8, U32, U64, )
+OPCODE(LoadGlobalS8, U32, U64, )
+OPCODE(LoadGlobalU16, U32, U64, )
+OPCODE(LoadGlobalS16, U32, U64, )
+OPCODE(LoadGlobal32, U32, U64, )
+OPCODE(LoadGlobal64, Opaque, U64, )
+OPCODE(LoadGlobal128, Opaque, U64, )
OPCODE(WriteGlobalU8, Void, U64, U32, )
OPCODE(WriteGlobalS8, Void, U64, U32, )
OPCODE(WriteGlobalU16, Void, U64, U32, )
@@ -58,6 +71,12 @@ OPCODE(CompositeConstruct3, Opaque, Opaq
OPCODE(CompositeConstruct4, Opaque, Opaque, Opaque, Opaque, Opaque, )
OPCODE(CompositeExtract, Opaque, Opaque, U32, )
+// Select operations
+OPCODE(Select8, U8, U1, U8, U8, )
+OPCODE(Select16, U16, U1, U16, U16, )
+OPCODE(Select32, U32, U1, U32, U32, )
+OPCODE(Select64, U64, U1, U64, U64, )
+
// Bitwise conversions
OPCODE(PackUint2x32, U64, Opaque, )
OPCODE(UnpackUint2x32, Opaque, U64, )
@@ -74,56 +93,84 @@ OPCODE(GetOverflowFromOp, U1, Opaq
OPCODE(GetZSCOFromOp, ZSCO, Opaque, )
// Floating-point operations
-OPCODE(FPAbs16, U16, U16 )
-OPCODE(FPAbs32, U32, U32 )
-OPCODE(FPAbs64, U64, U64 )
-OPCODE(FPAdd16, U16, U16, U16 )
-OPCODE(FPAdd32, U32, U32, U32 )
-OPCODE(FPAdd64, U64, U64, U64 )
-OPCODE(FPFma16, U16, U16, U16 )
-OPCODE(FPFma32, U32, U32, U32 )
-OPCODE(FPFma64, U64, U64, U64 )
-OPCODE(FPMax32, U32, U32, U32 )
-OPCODE(FPMax64, U64, U64, U64 )
-OPCODE(FPMin32, U32, U32, U32 )
-OPCODE(FPMin64, U64, U64, U64 )
-OPCODE(FPMul16, U16, U16, U16 )
-OPCODE(FPMul32, U32, U32, U32 )
-OPCODE(FPMul64, U64, U64, U64 )
-OPCODE(FPNeg16, U16, U16 )
-OPCODE(FPNeg32, U32, U32 )
-OPCODE(FPNeg64, U64, U64 )
-OPCODE(FPRecip32, U32, U32 )
-OPCODE(FPRecip64, U64, U64 )
-OPCODE(FPRecipSqrt32, U32, U32 )
-OPCODE(FPRecipSqrt64, U64, U64 )
-OPCODE(FPSqrt, U32, U32 )
-OPCODE(FPSin, U32, U32 )
-OPCODE(FPSinNotReduced, U32, U32 )
-OPCODE(FPExp2, U32, U32 )
-OPCODE(FPExp2NotReduced, U32, U32 )
-OPCODE(FPCos, U32, U32 )
-OPCODE(FPCosNotReduced, U32, U32 )
-OPCODE(FPLog2, U32, U32 )
-OPCODE(FPSaturate16, U16, U16 )
-OPCODE(FPSaturate32, U32, U32 )
-OPCODE(FPSaturate64, U64, U64 )
-OPCODE(FPRoundEven16, U16, U16 )
-OPCODE(FPRoundEven32, U32, U32 )
-OPCODE(FPRoundEven64, U64, U64 )
-OPCODE(FPFloor16, U16, U16 )
-OPCODE(FPFloor32, U32, U32 )
-OPCODE(FPFloor64, U64, U64 )
-OPCODE(FPCeil16, U16, U16 )
-OPCODE(FPCeil32, U32, U32 )
-OPCODE(FPCeil64, U64, U64 )
-OPCODE(FPTrunc16, U16, U16 )
-OPCODE(FPTrunc32, U32, U32 )
-OPCODE(FPTrunc64, U64, U64 )
+OPCODE(FPAbs16, U16, U16, )
+OPCODE(FPAbs32, U32, U32, )
+OPCODE(FPAbs64, U64, U64, )
+OPCODE(FPAdd16, U16, U16, U16, )
+OPCODE(FPAdd32, U32, U32, U32, )
+OPCODE(FPAdd64, U64, U64, U64, )
+OPCODE(FPFma16, U16, U16, U16, U16, )
+OPCODE(FPFma32, U32, U32, U32, U32, )
+OPCODE(FPFma64, U64, U64, U64, U64, )
+OPCODE(FPMax32, U32, U32, U32, )
+OPCODE(FPMax64, U64, U64, U64, )
+OPCODE(FPMin32, U32, U32, U32, )
+OPCODE(FPMin64, U64, U64, U64, )
+OPCODE(FPMul16, U16, U16, U16, )
+OPCODE(FPMul32, U32, U32, U32, )
+OPCODE(FPMul64, U64, U64, U64, )
+OPCODE(FPNeg16, U16, U16, )
+OPCODE(FPNeg32, U32, U32, )
+OPCODE(FPNeg64, U64, U64, )
+OPCODE(FPRecip32, U32, U32, )
+OPCODE(FPRecip64, U64, U64, )
+OPCODE(FPRecipSqrt32, U32, U32, )
+OPCODE(FPRecipSqrt64, U64, U64, )
+OPCODE(FPSqrt, U32, U32, )
+OPCODE(FPSin, U32, U32, )
+OPCODE(FPSinNotReduced, U32, U32, )
+OPCODE(FPExp2, U32, U32, )
+OPCODE(FPExp2NotReduced, U32, U32, )
+OPCODE(FPCos, U32, U32, )
+OPCODE(FPCosNotReduced, U32, U32, )
+OPCODE(FPLog2, U32, U32, )
+OPCODE(FPSaturate16, U16, U16, )
+OPCODE(FPSaturate32, U32, U32, )
+OPCODE(FPSaturate64, U64, U64, )
+OPCODE(FPRoundEven16, U16, U16, )
+OPCODE(FPRoundEven32, U32, U32, )
+OPCODE(FPRoundEven64, U64, U64, )
+OPCODE(FPFloor16, U16, U16, )
+OPCODE(FPFloor32, U32, U32, )
+OPCODE(FPFloor64, U64, U64, )
+OPCODE(FPCeil16, U16, U16, )
+OPCODE(FPCeil32, U32, U32, )
+OPCODE(FPCeil64, U64, U64, )
+OPCODE(FPTrunc16, U16, U16, )
+OPCODE(FPTrunc32, U32, U32, )
+OPCODE(FPTrunc64, U64, U64, )
+
+// Integer operations
+OPCODE(IAdd32, U32, U32, U32, )
+OPCODE(IAdd64, U64, U64, U64, )
+OPCODE(IMul32, U32, U32, U32, )
+OPCODE(INeg32, U32, U32, )
+OPCODE(IAbs32, U32, U32, )
+OPCODE(ShiftLeftLogical32, U32, U32, U32, )
+OPCODE(ShiftRightLogical32, U32, U32, U32, )
+OPCODE(ShiftRightArithmetic32, U32, U32, U32, )
+OPCODE(BitwiseAnd32, U32, U32, U32, )
+OPCODE(BitwiseOr32, U32, U32, U32, )
+OPCODE(BitwiseXor32, U32, U32, U32, )
+OPCODE(BitFieldInsert, U32, U32, U32, U32, U32, )
+OPCODE(BitFieldSExtract, U32, U32, U32, U32, )
+OPCODE(BitFieldUExtract, U32, U32, U32, U32, )
+
+OPCODE(SLessThan, U1, U32, U32, )
+OPCODE(ULessThan, U1, U32, U32, )
+OPCODE(IEqual, U1, U32, U32, )
+OPCODE(SLessThanEqual, U1, U32, U32, )
+OPCODE(ULessThanEqual, U1, U32, U32, )
+OPCODE(SGreaterThan, U1, U32, U32, )
+OPCODE(UGreaterThan, U1, U32, U32, )
+OPCODE(INotEqual, U1, U32, U32, )
+OPCODE(SGreaterThanEqual, U1, U32, U32, )
+OPCODE(UGreaterThanEqual, U1, U32, U32, )
// Logical operations
OPCODE(LogicalOr, U1, U1, U1, )
OPCODE(LogicalAnd, U1, U1, U1, )
+OPCODE(LogicalXor, U1, U1, U1, )
OPCODE(LogicalNot, U1, U1, )
// Conversion operations
diff --git a/src/shader_recompiler/frontend/ir/pred.h b/src/shader_recompiler/frontend/ir/pred.h
index daf23193f..c6f2f82bf 100644
--- a/src/shader_recompiler/frontend/ir/pred.h
+++ b/src/shader_recompiler/frontend/ir/pred.h
@@ -8,7 +8,16 @@
namespace Shader::IR {
-enum class Pred { P0, P1, P2, P3, P4, P5, P6, PT };
+enum class Pred : u64 {
+ P0,
+ P1,
+ P2,
+ P3,
+ P4,
+ P5,
+ P6,
+ PT,
+};
constexpr size_t NUM_USER_PREDS = 6;
constexpr size_t NUM_PREDS = 7;