shader: Initial instruction support

src/shader_recompiler/CMakeLists.txt

@@ -39,18 +39,27 @@ add_executable(shader_recompiler
     frontend/maxwell/program.h
     frontend/maxwell/termination_code.cpp
     frontend/maxwell/termination_code.h
+    frontend/maxwell/translate/impl/common_encoding.h
+    frontend/maxwell/translate/impl/floating_point_add.cpp
     frontend/maxwell/translate/impl/floating_point_conversion_integer.cpp
+    frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp
     frontend/maxwell/translate/impl/floating_point_multi_function.cpp
+    frontend/maxwell/translate/impl/floating_point_multiply.cpp
     frontend/maxwell/translate/impl/impl.cpp
     frontend/maxwell/translate/impl/impl.h
+    frontend/maxwell/translate/impl/integer_add.cpp
+    frontend/maxwell/translate/impl/integer_scaled_add.cpp
+    frontend/maxwell/translate/impl/integer_set_predicate.cpp
+    frontend/maxwell/translate/impl/integer_shift_left.cpp
+    frontend/maxwell/translate/impl/integer_short_multiply_add.cpp
     frontend/maxwell/translate/impl/load_store_attribute.cpp
     frontend/maxwell/translate/impl/load_store_memory.cpp
     frontend/maxwell/translate/impl/not_implemented.cpp
-    frontend/maxwell/translate/impl/register_move.cpp
+    frontend/maxwell/translate/impl/move_register.cpp
+    frontend/maxwell/translate/impl/move_special_register.cpp
     frontend/maxwell/translate/translate.cpp
     frontend/maxwell/translate/translate.h
     ir_opt/dead_code_elimination_pass.cpp
-    ir_opt/get_set_elimination_pass.cpp
     ir_opt/identity_removal_pass.cpp
     ir_opt/passes.h
     ir_opt/ssa_rewrite_pass.cpp

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()) {

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);

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);
 }

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

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);

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

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

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;

src/shader_recompiler/frontend/maxwell/program.cpp

@@ -56,6 +56,7 @@ Program::Program(Environment& env, const Flow::CFG& cfg) {
         Optimization::Invoke(Optimization::IdentityRemovalPass, function);
         // Optimization::Invoke(Optimization::VerificationPass, function);
     }
+    //*/
 }
 
 std::string DumpProgram(const Program& program) {

src/shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h

@@ -0,0 +1,56 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "common/common_types.h"
+#include "shader_recompiler/exception.h"
+#include "shader_recompiler/frontend/ir/modifiers.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
+
+namespace Shader::Maxwell {
+
+enum class FpRounding : u64 {
+    RN,
+    RM,
+    RP,
+    RZ,
+};
+
+enum class FmzMode : u64 {
+    None,
+    FTZ,
+    FMZ,
+    INVALIDFMZ3,
+};
+
+inline IR::FpRounding CastFpRounding(FpRounding fp_rounding) {
+    switch (fp_rounding) {
+    case FpRounding::RN:
+        return IR::FpRounding::RN;
+    case FpRounding::RM:
+        return IR::FpRounding::RM;
+    case FpRounding::RP:
+        return IR::FpRounding::RP;
+    case FpRounding::RZ:
+        return IR::FpRounding::RZ;
+    }
+    throw NotImplementedException("Invalid floating-point rounding {}", fp_rounding);
+}
+
+inline IR::FmzMode CastFmzMode(FmzMode fmz_mode) {
+    switch (fmz_mode) {
+    case FmzMode::None:
+        return IR::FmzMode::None;
+    case FmzMode::FTZ:
+        return IR::FmzMode::FTZ;
+    case FmzMode::FMZ:
+        return IR::FmzMode::FMZ;
+    case FmzMode::INVALIDFMZ3:
+        break;
+    }
+    throw NotImplementedException("Invalid FMZ mode {}", fmz_mode);
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_add.cpp

@@ -0,0 +1,71 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/common_types.h"
+#include "shader_recompiler/exception.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+
+void FADD(TranslatorVisitor& v, u64 insn, bool sat, bool cc, bool ftz, FpRounding fp_rounding,
+          const IR::U32& src_b, bool abs_a, bool neg_a, bool abs_b, bool neg_b) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> src_a;
+    } const fadd{insn};
+
+    if (sat) {
+        throw NotImplementedException("FADD SAT");
+    }
+    if (cc) {
+        throw NotImplementedException("FADD CC");
+    }
+    const IR::U32 op_a{v.ir.FPAbsNeg(v.X(fadd.src_a), abs_a, neg_a)};
+    const IR::U32 op_b{v.ir.FPAbsNeg(src_b, abs_b, neg_b)};
+    IR::FpControl control{
+        .no_contraction{true},
+        .rounding{CastFpRounding(fp_rounding)},
+        .fmz_mode{ftz ? IR::FmzMode::FTZ : IR::FmzMode::None},
+    };
+    v.X(fadd.dest_reg, v.ir.FPAdd(op_a, op_b, control));
+}
+
+void FADD(TranslatorVisitor& v, u64 insn, const IR::U32& src_b) {
+    union {
+        u64 raw;
+        BitField<39, 2, FpRounding> fp_rounding;
+        BitField<44, 1, u64> ftz;
+        BitField<45, 1, u64> neg_b;
+        BitField<46, 1, u64> abs_a;
+        BitField<47, 1, u64> cc;
+        BitField<48, 1, u64> neg_a;
+        BitField<49, 1, u64> abs_b;
+        BitField<50, 1, u64> sat;
+    } const fadd{insn};
+
+    FADD(v, insn, fadd.sat != 0, fadd.cc != 0, fadd.ftz != 0, fadd.fp_rounding, src_b,
+         fadd.abs_a != 0, fadd.neg_a != 0, fadd.abs_b != 0, fadd.neg_b != 0);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::FADD_reg(u64 insn) {
+    FADD(*this, insn, GetReg20(insn));
+}
+
+void TranslatorVisitor::FADD_cbuf(u64) {
+    throw NotImplementedException("FADD (cbuf)");
+}
+
+void TranslatorVisitor::FADD_imm(u64) {
+    throw NotImplementedException("FADD (imm)");
+}
+
+void TranslatorVisitor::FADD32I(u64) {
+    throw NotImplementedException("FADD32I");
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_fused_multiply_add.cpp

@@ -0,0 +1,73 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/common_types.h"
+#include "shader_recompiler/exception.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+void FFMA(TranslatorVisitor& v, u64 insn, const IR::U32& src_b, const IR::U32& src_c, bool neg_a,
+          bool neg_b, bool neg_c, bool sat, bool cc, FmzMode fmz_mode, FpRounding fp_rounding) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> src_a;
+    } const ffma{insn};
+
+    if (sat) {
+        throw NotImplementedException("FFMA SAT");
+    }
+    if (cc) {
+        throw NotImplementedException("FFMA CC");
+    }
+    const IR::U32 op_a{v.ir.FPAbsNeg(v.X(ffma.src_a), false, neg_a)};
+    const IR::U32 op_b{v.ir.FPAbsNeg(src_b, false, neg_b)};
+    const IR::U32 op_c{v.ir.FPAbsNeg(src_c, false, neg_c)};
+    const IR::FpControl fp_control{
+        .no_contraction{true},
+        .rounding{CastFpRounding(fp_rounding)},
+        .fmz_mode{CastFmzMode(fmz_mode)},
+    };
+    v.X(ffma.dest_reg, v.ir.FPFma(op_a, op_b, op_c, fp_control));
+}
+
+void FFMA(TranslatorVisitor& v, u64 insn, const IR::U32& src_b, const IR::U32& src_c) {
+    union {
+        u64 raw;
+        BitField<47, 1, u64> cc;
+        BitField<48, 1, u64> neg_b;
+        BitField<49, 1, u64> neg_c;
+        BitField<50, 1, u64> sat;
+        BitField<51, 2, FpRounding> fp_rounding;
+        BitField<53, 2, FmzMode> fmz_mode;
+    } const ffma{insn};
+
+    FFMA(v, insn, src_b, src_c, false, ffma.neg_b != 0, ffma.neg_c != 0, ffma.sat != 0,
+         ffma.cc != 0, ffma.fmz_mode, ffma.fp_rounding);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::FFMA_reg(u64 insn) {
+    FFMA(*this, insn, GetReg20(insn), GetReg39(insn));
+}
+
+void TranslatorVisitor::FFMA_rc(u64) {
+    throw NotImplementedException("FFMA (rc)");
+}
+
+void TranslatorVisitor::FFMA_cr(u64 insn) {
+    FFMA(*this, insn, GetCbuf(insn), GetReg39(insn));
+}
+
+void TranslatorVisitor::FFMA_imm(u64) {
+    throw NotImplementedException("FFMA (imm)");
+}
+
+void TranslatorVisitor::FFMA32I(u64) {
+    throw NotImplementedException("FFMA32I");
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/floating_point_multiply.cpp

@@ -0,0 +1,108 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/ir/modifiers.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+enum class Scale : u64 {
+    None,
+    D2,
+    D4,
+    D8,
+    M8,
+    M4,
+    M2,
+    INVALIDSCALE37,
+};
+
+float ScaleFactor(Scale scale) {
+    switch (scale) {
+    case Scale::None:
+        return 1.0f;
+    case Scale::D2:
+        return 1.0f / 2.0f;
+    case Scale::D4:
+        return 1.0f / 4.0f;
+    case Scale::D8:
+        return 1.0f / 8.0f;
+    case Scale::M8:
+        return 8.0f;
+    case Scale::M4:
+        return 4.0f;
+    case Scale::M2:
+        return 2.0f;
+    case Scale::INVALIDSCALE37:
+        break;
+    }
+    throw NotImplementedException("Invalid FMUL scale {}", scale);
+}
+
+void FMUL(TranslatorVisitor& v, u64 insn, const IR::U32& src_b, FmzMode fmz_mode,
+          FpRounding fp_rounding, Scale scale, bool sat, bool cc, bool neg_b) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> src_a;
+    } const fmul{insn};
+
+    if (cc) {
+        throw NotImplementedException("FMUL CC");
+    }
+    if (sat) {
+        throw NotImplementedException("FMUL SAT");
+    }
+    IR::U32 op_a{v.X(fmul.src_a)};
+    if (scale != Scale::None) {
+        if (fmz_mode != FmzMode::FTZ || fp_rounding != FpRounding::RN) {
+            throw NotImplementedException("FMUL scale with non-FMZ or non-RN modifiers");
+        }
+        op_a = v.ir.FPMul(op_a, v.ir.Imm32(ScaleFactor(scale)));
+    }
+    const IR::U32 op_b{v.ir.FPAbsNeg(src_b, false, neg_b)};
+    const IR::FpControl fp_control{
+        .no_contraction{true},
+        .rounding{CastFpRounding(fp_rounding)},
+        .fmz_mode{CastFmzMode(fmz_mode)},
+    };
+    v.X(fmul.dest_reg, v.ir.FPMul(op_a, op_b, fp_control));
+}
+
+void FMUL(TranslatorVisitor& v, u64 insn, const IR::U32& src_b) {
+    union {
+        u64 raw;
+        BitField<39, 2, FpRounding> fp_rounding;
+        BitField<41, 3, Scale> scale;
+        BitField<44, 2, FmzMode> fmz;
+        BitField<47, 1, u64> cc;
+        BitField<48, 1, u64> neg_b;
+        BitField<50, 1, u64> sat;
+    } fmul{insn};
+
+    FMUL(v, insn, src_b, fmul.fmz, fmul.fp_rounding, fmul.scale, fmul.sat != 0, fmul.cc != 0,
+         fmul.neg_b != 0);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::FMUL_reg(u64 insn) {
+    return FMUL(*this, insn, GetReg20(insn));
+}
+
+void TranslatorVisitor::FMUL_cbuf(u64) {
+    throw NotImplementedException("FMUL (cbuf)");
+}
+
+void TranslatorVisitor::FMUL_imm(u64) {
+    throw NotImplementedException("FMUL (imm)");
+}
+
+void TranslatorVisitor::FMUL32I(u64) {
+    throw NotImplementedException("FMUL32I");
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp

@@ -16,6 +16,22 @@ void TranslatorVisitor::X(IR::Reg dest_reg, const IR::U32& value) {
     ir.SetReg(dest_reg, value);
 }
 
+IR::U32 TranslatorVisitor::GetReg20(u64 insn) {
+    union {
+        u64 raw;
+        BitField<20, 8, IR::Reg> index;
+    } const reg{insn};
+    return X(reg.index);
+}
+
+IR::U32 TranslatorVisitor::GetReg39(u64 insn) {
+    union {
+        u64 raw;
+        BitField<39, 8, IR::Reg> index;
+    } const reg{insn};
+    return X(reg.index);
+}
+
 IR::U32 TranslatorVisitor::GetCbuf(u64 insn) {
     union {
         u64 raw;
@@ -33,7 +49,7 @@ IR::U32 TranslatorVisitor::GetCbuf(u64 insn) {
     return ir.GetCbuf(binding, byte_offset);
 }
 
-IR::U32 TranslatorVisitor::GetImm(u64 insn) {
+IR::U32 TranslatorVisitor::GetImm20(u64 insn) {
     union {
         u64 raw;
         BitField<20, 19, u64> value;
@@ -44,6 +60,14 @@ IR::U32 TranslatorVisitor::GetImm(u64 insn) {
     return ir.Imm32(value);
 }
 
+IR::U32 TranslatorVisitor::GetImm32(u64 insn) {
+    union {
+        u64 raw;
+        BitField<20, 32, u64> value;
+    } const imm{insn};
+    return ir.Imm32(static_cast<u32>(imm.value));
+}
+
 void TranslatorVisitor::SetZFlag(const IR::U1& value) {
     ir.SetZFlag(value);
 }

src/shader_recompiler/frontend/maxwell/translate/impl/impl.h

@@ -46,7 +46,7 @@ public:
     void DADD_reg(u64 insn);
     void DADD_cbuf(u64 insn);
     void DADD_imm(u64 insn);
-    void DEPBAR(u64 insn);
+    void DEPBAR();
     void DFMA_reg(u64 insn);
     void DFMA_rc(u64 insn);
     void DFMA_cr(u64 insn);
@@ -298,9 +298,14 @@ public:
     [[nodiscard]] IR::U32 X(IR::Reg reg);
     void X(IR::Reg dest_reg, const IR::U32& value);
 
+    [[nodiscard]] IR::U32 GetReg20(u64 insn);
+    [[nodiscard]] IR::U32 GetReg39(u64 insn);
+
     [[nodiscard]] IR::U32 GetCbuf(u64 insn);
 
-    [[nodiscard]] IR::U32 GetImm(u64 insn);
+    [[nodiscard]] IR::U32 GetImm20(u64 insn);
+
+    [[nodiscard]] IR::U32 GetImm32(u64 insn);
 
     void SetZFlag(const IR::U1& value);
     void SetSFlag(const IR::U1& value);

src/shader_recompiler/frontend/maxwell/translate/impl/integer_add.cpp

@@ -0,0 +1,106 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+void IADD(TranslatorVisitor& v, u64 insn, const IR::U32 op_b, bool neg_a, bool po, bool sat, bool x,
+          bool cc) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> src_a;
+    } const iadd{insn};
+
+    if (sat) {
+        throw NotImplementedException("IADD SAT");
+    }
+    if (x && po) {
+        throw NotImplementedException("IADD X+PO");
+    }
+    // Operand A is always read from here, negated if needed
+    IR::U32 op_a{v.X(iadd.src_a)};
+    if (neg_a) {
+        op_a = v.ir.INeg(op_a);
+    }
+    // Add both operands
+    IR::U32 result{v.ir.IAdd(op_a, op_b)};
+    if (x) {
+        const IR::U32 carry{v.ir.Select(v.ir.GetCFlag(), v.ir.Imm32(1), v.ir.Imm32(0))};
+        result = v.ir.IAdd(result, carry);
+    }
+    if (po) {
+        // .PO adds one to the result
+        result = v.ir.IAdd(result, v.ir.Imm32(1));
+    }
+    if (cc) {
+        // Store flags
+        // TODO: Does this grab the result pre-PO or after?
+        if (po) {
+            throw NotImplementedException("IADD CC+PO");
+        }
+        // TODO: How does CC behave when X is set?
+        if (x) {
+            throw NotImplementedException("IADD X+CC");
+        }
+        v.SetZFlag(v.ir.GetZeroFromOp(result));
+        v.SetSFlag(v.ir.GetSignFromOp(result));
+        v.SetCFlag(v.ir.GetCarryFromOp(result));
+        v.SetOFlag(v.ir.GetOverflowFromOp(result));
+    }
+    // Store result
+    v.X(iadd.dest_reg, result);
+}
+
+void IADD(TranslatorVisitor& v, u64 insn, IR::U32 op_b) {
+    union {
+        u64 insn;
+        BitField<43, 1, u64> x;
+        BitField<47, 1, u64> cc;
+        BitField<48, 2, u64> three_for_po;
+        BitField<48, 1, u64> neg_b;
+        BitField<49, 1, u64> neg_a;
+        BitField<50, 1, u64> sat;
+    } const iadd{insn};
+
+    const bool po{iadd.three_for_po == 3};
+    const bool neg_a{!po && iadd.neg_a != 0};
+    if (!po && iadd.neg_b != 0) {
+        op_b = v.ir.INeg(op_b);
+    }
+    IADD(v, insn, op_b, iadd.neg_a != 0, po, iadd.sat != 0, iadd.x != 0, iadd.cc != 0);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::IADD_reg(u64) {
+    throw NotImplementedException("IADD (reg)");
+}
+
+void TranslatorVisitor::IADD_cbuf(u64 insn) {
+    IADD(*this, insn, GetCbuf(insn));
+}
+
+void TranslatorVisitor::IADD_imm(u64) {
+    throw NotImplementedException("IADD (imm)");
+}
+
+void TranslatorVisitor::IADD32I(u64 insn) {
+    union {
+        u64 raw;
+        BitField<52, 1, u64> cc;
+        BitField<53, 1, u64> x;
+        BitField<54, 1, u64> sat;
+        BitField<55, 2, u64> three_for_po;
+        BitField<56, 1, u64> neg_a;
+    } const iadd32i{insn};
+
+    const bool po{iadd32i.three_for_po == 3};
+    const bool neg_a{!po && iadd32i.neg_a != 0};
+    IADD(*this, insn, GetImm32(insn), neg_a, po, iadd32i.sat != 0, iadd32i.x != 0, iadd32i.cc != 0);
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/integer_scaled_add.cpp

@@ -0,0 +1,73 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+void ISCADD(TranslatorVisitor& v, u64 insn, IR::U32 op_b) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> op_a;
+        BitField<47, 1, u64> cc;
+        BitField<48, 2, u64> three_for_po;
+        BitField<48, 1, u64> neg_b;
+        BitField<49, 1, u64> neg_a;
+        BitField<39, 5, u64> scale;
+    } const iscadd{insn};
+
+    const bool po{iscadd.three_for_po == 3};
+    IR::U32 op_a{v.X(iscadd.op_a)};
+    if (!po) {
+        // When PO is not present, the bits are interpreted as negation
+        if (iscadd.neg_a != 0) {
+            op_a = v.ir.INeg(op_a);
+        }
+        if (iscadd.neg_b != 0) {
+            op_b = v.ir.INeg(op_b);
+        }
+    }
+    // With the operands already processed, scale A
+    const IR::U32 scale{v.ir.Imm32(static_cast<u32>(iscadd.scale))};
+    const IR::U32 scaled_a{v.ir.ShiftLeftLogical(op_a, scale)};
+
+    IR::U32 result{v.ir.IAdd(scaled_a, op_b)};
+    if (po) {
+        // .PO adds one to the final result
+        result = v.ir.IAdd(result, v.ir.Imm32(1));
+    }
+    v.X(iscadd.dest_reg, result);
+
+    if (iscadd.cc != 0) {
+        throw NotImplementedException("ISCADD CC");
+    }
+}
+
+} // Anonymous namespace
+
+void TranslatorVisitor::ISCADD_reg(u64 insn) {
+    union {
+        u64 raw;
+        BitField<20, 8, IR::Reg> op_b;
+    } const iscadd{insn};
+
+    ISCADD(*this, insn, X(iscadd.op_b));
+}
+
+void TranslatorVisitor::ISCADD_cbuf(u64) {
+    throw NotImplementedException("ISCADD (cbuf)");
+}
+
+void TranslatorVisitor::ISCADD_imm(u64) {
+    throw NotImplementedException("ISCADD (imm)");
+}
+
+void TranslatorVisitor::ISCADD32I(u64) {
+    throw NotImplementedException("ISCADD32I");
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/integer_set_predicate.cpp

@@ -0,0 +1,99 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+enum class CompareOp : u64 {
+    F,  // Always false
+    LT, // Less than
+    EQ, // Equal
+    LE, // Less than or equal
+    GT, // Greater than
+    NE, // Not equal
+    GE, // Greater than or equal
+    T,  // Always true
+};
+
+enum class Bop : u64 {
+    AND,
+    OR,
+    XOR,
+};
+
+IR::U1 Compare(IR::IREmitter& ir, CompareOp op, const IR::U32& lhs, const IR::U32& rhs,
+               bool is_signed) {
+    switch (op) {
+    case CompareOp::F:
+        return ir.Imm1(false);
+    case CompareOp::LT:
+        return ir.ILessThan(lhs, rhs, is_signed);
+    case CompareOp::EQ:
+        return ir.IEqual(lhs, rhs);
+    case CompareOp::LE:
+        return ir.ILessThanEqual(lhs, rhs, is_signed);
+    case CompareOp::GT:
+        return ir.IGreaterThan(lhs, rhs, is_signed);
+    case CompareOp::NE:
+        return ir.INotEqual(lhs, rhs);
+    case CompareOp::GE:
+        return ir.IGreaterThanEqual(lhs, rhs, is_signed);
+    case CompareOp::T:
+        return ir.Imm1(true);
+    }
+    throw NotImplementedException("Invalid ISETP compare op {}", op);
+}
+
+IR::U1 Combine(IR::IREmitter& ir, Bop bop, const IR::U1& comparison, const IR::U1& bop_pred) {
+    switch (bop) {
+    case Bop::AND:
+        return ir.LogicalAnd(comparison, bop_pred);
+    case Bop::OR:
+        return ir.LogicalOr(comparison, bop_pred);
+    case Bop::XOR:
+        return ir.LogicalXor(comparison, bop_pred);
+    }
+    throw NotImplementedException("Invalid ISETP bop {}", bop);
+}
+
+void ISETP(TranslatorVisitor& v, u64 insn, const IR::U32& op_b) {
+    union {
+        u64 raw;
+        BitField<0, 3, IR::Pred> dest_pred_b;
+        BitField<3, 3, IR::Pred> dest_pred_a;
+        BitField<8, 8, IR::Reg> src_reg_a;
+        BitField<39, 3, IR::Pred> bop_pred;
+        BitField<42, 1, u64> neg_bop_pred;
+        BitField<45, 2, Bop> bop;
+        BitField<48, 1, u64> is_signed;
+        BitField<49, 3, CompareOp> compare_op;
+    } const isetp{insn};
+
+    const Bop bop{isetp.bop};
+    const IR::U32 op_a{v.X(isetp.src_reg_a)};
+    const IR::U1 comparison{Compare(v.ir, isetp.compare_op, op_a, op_b, isetp.is_signed != 0)};
+    const IR::U1 bop_pred{v.ir.GetPred(isetp.bop_pred, isetp.neg_bop_pred != 0)};
+    const IR::U1 result_a{Combine(v.ir, bop, comparison, bop_pred)};
+    const IR::U1 result_b{Combine(v.ir, bop, v.ir.LogicalNot(comparison), bop_pred)};
+    v.ir.SetPred(isetp.dest_pred_a, result_a);
+    v.ir.SetPred(isetp.dest_pred_b, result_b);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::ISETP_reg(u64 insn) {
+    ISETP(*this, insn, GetReg20(insn));
+}
+
+void TranslatorVisitor::ISETP_cbuf(u64 insn) {
+    ISETP(*this, insn, GetCbuf(insn));
+}
+
+void TranslatorVisitor::ISETP_imm(u64) {
+    throw NotImplementedException("ISETP_imm");
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/integer_shift_left.cpp

@@ -0,0 +1,71 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+void SHL(TranslatorVisitor& v, u64 insn, const IR::U32& unsafe_shift) {
+    union {
+        u64 insn;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> src_reg_a;
+        BitField<39, 1, u64> w;
+        BitField<43, 1, u64> x;
+        BitField<47, 1, u64> cc;
+    } const shl{insn};
+
+    if (shl.x != 0) {
+        throw NotImplementedException("SHL.X");
+    }
+    if (shl.cc != 0) {
+        throw NotImplementedException("SHL.CC");
+    }
+    const IR::U32 base{v.X(shl.src_reg_a)};
+    IR::U32 result;
+    if (shl.w != 0) {
+        // When .W is set, the shift value is wrapped
+        // To emulate this we just have to clamp it ourselves.
+        const IR::U32 shift{v.ir.BitwiseAnd(unsafe_shift, v.ir.Imm32(31))};
+        result = v.ir.ShiftLeftLogical(base, shift);
+    } else {
+        // When .W is not set, the shift value is clamped between 0 and 32.
+        // To emulate this we have to have in mind the special shift of 32, that evaluates as 0.
+        // We can safely evaluate an out of bounds shift according to the SPIR-V specification:
+        //
+        // https://www.khronos.org/registry/spir-v/specs/unified1/SPIRV.html#OpShiftLeftLogical
+        // "Shift is treated as unsigned. The resulting value is undefined if Shift is greater than
+        //  or equal to the bit width of the components of Base."
+        //
+        // And on the GLASM specification it is also safe to evaluate out of bounds:
+        //
+        // https://www.khronos.org/registry/OpenGL/extensions/NV/NV_gpu_program4.txt
+        // "The results of a shift operation ("<<") are undefined if the value of the second operand
+        //  is negative, or greater than or equal to the number of bits in the first operand."
+        //
+        // Emphasis on undefined results in contrast to undefined behavior.
+        //
+        const IR::U1 is_safe{v.ir.ILessThan(unsafe_shift, v.ir.Imm32(32), false)};
+        const IR::U32 unsafe_result{v.ir.ShiftLeftLogical(base, unsafe_shift)};
+        result = v.ir.Select(is_safe, unsafe_result, v.ir.Imm32(0));
+    }
+    v.X(shl.dest_reg, result);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::SHL_reg(u64) {
+    throw NotImplementedException("SHL_reg");
+}
+
+void TranslatorVisitor::SHL_cbuf(u64) {
+    throw NotImplementedException("SHL_cbuf");
+}
+
+void TranslatorVisitor::SHL_imm(u64 insn) {
+    SHL(*this, insn, GetImm20(insn));
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/integer_short_multiply_add.cpp

@@ -0,0 +1,110 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+enum class SelectMode : u64 {
+    Default,
+    CLO,
+    CHI,
+    CSFU,
+    CBCC,
+};
+
+enum class Half : u64 {
+    H0, // Least-significant bits (15:0)
+    H1, // Most-significant bits (31:16)
+};
+
+IR::U32 ExtractHalf(TranslatorVisitor& v, const IR::U32& src, Half half, bool is_signed) {
+    const IR::U32 offset{v.ir.Imm32(half == Half::H1 ? 16 : 0)};
+    return v.ir.BitFieldExtract(src, offset, v.ir.Imm32(16), is_signed);
+}
+
+void XMAD(TranslatorVisitor& v, u64 insn, const IR::U32& src_b, const IR::U32& src_c,
+          SelectMode select_mode, Half half_b, bool psl, bool mrg, bool x) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<8, 8, IR::Reg> src_reg_a;
+        BitField<47, 1, u64> cc;
+        BitField<48, 1, u64> is_a_signed;
+        BitField<49, 1, u64> is_b_signed;
+        BitField<53, 1, Half> half_a;
+    } const xmad{insn};
+
+    if (x) {
+        throw NotImplementedException("XMAD X");
+    }
+    const IR::U32 op_a{ExtractHalf(v, v.X(xmad.src_reg_a), xmad.half_a, xmad.is_a_signed != 0)};
+    const IR::U32 op_b{ExtractHalf(v, src_b, half_b, xmad.is_b_signed != 0)};
+
+    IR::U32 product{v.ir.IMul(op_a, op_b)};
+    if (psl) {
+        // .PSL shifts the product 16 bits
+        product = v.ir.ShiftLeftLogical(product, v.ir.Imm32(16));
+    }
+    const IR::U32 op_c{[&]() -> IR::U32 {
+        switch (select_mode) {
+        case SelectMode::Default:
+            return src_c;
+        case SelectMode::CLO:
+            return ExtractHalf(v, src_c, Half::H0, false);
+        case SelectMode::CHI:
+            return ExtractHalf(v, src_c, Half::H1, false);
+        case SelectMode::CBCC:
+            return v.ir.IAdd(v.ir.ShiftLeftLogical(src_b, v.ir.Imm32(16)), src_b);
+        case SelectMode::CSFU:
+            throw NotImplementedException("XMAD CSFU");
+        }
+        throw NotImplementedException("Invalid XMAD select mode {}", select_mode);
+    }()};
+    IR::U32 result{v.ir.IAdd(product, op_c)};
+    if (mrg) {
+        // .MRG inserts src_b [15:0] into result's [31:16].
+        const IR::U32 lsb_b{ExtractHalf(v, src_b, Half::H0, false)};
+        result = v.ir.BitFieldInsert(result, lsb_b, v.ir.Imm32(16), v.ir.Imm32(16));
+    }
+    if (xmad.cc) {
+        throw NotImplementedException("XMAD CC");
+    }
+    // Store result
+    v.X(xmad.dest_reg, result);
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::XMAD_reg(u64) {
+    throw NotImplementedException("XMAD (reg)");
+}
+
+void TranslatorVisitor::XMAD_rc(u64) {
+    throw NotImplementedException("XMAD (rc)");
+}
+
+void TranslatorVisitor::XMAD_cr(u64) {
+    throw NotImplementedException("XMAD (cr)");
+}
+
+void TranslatorVisitor::XMAD_imm(u64 insn) {
+    union {
+        u64 raw;
+        BitField<20, 16, u64> src_b;
+        BitField<36, 1, u64> psl;
+        BitField<37, 1, u64> mrg;
+        BitField<38, 1, u64> x;
+        BitField<39, 8, IR::Reg> src_c;
+        BitField<50, 3, SelectMode> select_mode;
+    } const xmad{insn};
+
+    const IR::U32 src_b{ir.Imm32(static_cast<u32>(xmad.src_b))};
+    const IR::U32 src_c{X(xmad.src_c)};
+    XMAD(*this, insn, src_b, src_c, xmad.select_mode, Half::H0, xmad.psl != 0, xmad.mrg != 0,
+         xmad.x != 0);
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/load_store_memory.cpp

@@ -10,16 +10,35 @@
 
 namespace Shader::Maxwell {
 namespace {
+enum class LoadSize : u64 {
+    U8,  // Zero-extend
+    S8,  // Sign-extend
+    U16, // Zero-extend
+    S16, // Sign-extend
+    B32,
+    B64,
+    B128,
+    U128, // ???
+};
+
 enum class StoreSize : u64 {
-    U8,
-    S8,
-    U16,
-    S16,
+    U8,  // Zero-extend
+    S8,  // Sign-extend
+    U16, // Zero-extend
+    S16, // Sign-extend
     B32,
     B64,
     B128,
 };
 
+// See Table 27 in https://docs.nvidia.com/cuda/parallel-thread-execution/index.html
+enum class LoadCache : u64 {
+    CA, // Cache at all levels, likely to be accessed again
+    CG, // Cache at global level (cache in L2 and below, not L1)
+    CI, // ???
+    CV, // Don't cache and fetch again (consider cached system memory lines stale, fetch again)
+};
+
 // See Table 28 in https://docs.nvidia.com/cuda/parallel-thread-execution/index.html
 enum class StoreCache : u64 {
     WB, // Cache write-back all coherent levels
@@ -27,61 +46,137 @@ enum class StoreCache : u64 {
     CS, // Cache streaming, likely to be accessed once
     WT, // Cache write-through (to system memory)
 };
-} // Anonymous namespace
 
-void TranslatorVisitor::STG(u64 insn) {
-    // STG stores registers into global memory.
+IR::U64 Address(TranslatorVisitor& v, u64 insn) {
     union {
         u64 raw;
-        BitField<0, 8, IR::Reg> data_reg;
         BitField<8, 8, IR::Reg> addr_reg;
+        BitField<20, 24, s64> addr_offset;
+        BitField<20, 24, u64> rz_addr_offset;
         BitField<45, 1, u64> e;
-        BitField<46, 2, StoreCache> cache;
-        BitField<48, 3, StoreSize> size;
-    } const stg{insn};
+    } const mem{insn};
 
     const IR::U64 address{[&]() -> IR::U64 {
-        if (stg.e == 0) {
-            // STG without .E uses a 32-bit pointer, zero-extend it
-            return ir.ConvertU(64, X(stg.addr_reg));
+        if (mem.e == 0) {
+            // LDG/STG without .E uses a 32-bit pointer, zero-extend it
+            return v.ir.ConvertU(64, v.X(mem.addr_reg));
         }
-        if (!IR::IsAligned(stg.addr_reg, 2)) {
+        if (!IR::IsAligned(mem.addr_reg, 2)) {
             throw NotImplementedException("Unaligned address register");
         }
-        // Pack two registers to build the 32-bit address
-        return ir.PackUint2x32(ir.CompositeConstruct(X(stg.addr_reg), X(stg.addr_reg + 1)));
+        // Pack two registers to build the 64-bit address
+        return v.ir.PackUint2x32(v.ir.CompositeConstruct(v.X(mem.addr_reg), v.X(mem.addr_reg + 1)));
+    }()};
+    const u64 addr_offset{[&]() -> u64 {
+        if (mem.addr_reg == IR::Reg::RZ) {
+            // When RZ is used, the address is an absolute address
+            return static_cast<u64>(mem.rz_addr_offset.Value());
+        } else {
+            return static_cast<u64>(mem.addr_offset.Value());
+        }
     }()};
+    // Apply the offset
+    return v.ir.IAdd(address, v.ir.Imm64(addr_offset));
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::LDG(u64 insn) {
+    // LDG loads global memory into registers
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<46, 2, LoadCache> cache;
+        BitField<48, 3, LoadSize> size;
+    } const ldg{insn};
+
+    // Pointer to load data from
+    const IR::U64 address{Address(*this, insn)};
+    const IR::Reg dest_reg{ldg.dest_reg};
+    switch (ldg.size) {
+    case LoadSize::U8:
+        X(dest_reg, ir.LoadGlobalU8(address));
+        break;
+    case LoadSize::S8:
+        X(dest_reg, ir.LoadGlobalS8(address));
+        break;
+    case LoadSize::U16:
+        X(dest_reg, ir.LoadGlobalU16(address));
+        break;
+    case LoadSize::S16:
+        X(dest_reg, ir.LoadGlobalS16(address));
+        break;
+    case LoadSize::B32:
+        X(dest_reg, ir.LoadGlobal32(address));
+        break;
+    case LoadSize::B64: {
+        if (!IR::IsAligned(dest_reg, 2)) {
+            throw NotImplementedException("Unaligned data registers");
+        }
+        const IR::Value vector{ir.LoadGlobal64(address)};
+        for (int i = 0; i < 2; ++i) {
+            X(dest_reg + i, ir.CompositeExtract(vector, i));
+        }
+        break;
+    }
+    case LoadSize::B128: {
+        if (!IR::IsAligned(dest_reg, 4)) {
+            throw NotImplementedException("Unaligned data registers");
+        }
+        const IR::Value vector{ir.LoadGlobal128(address)};
+        for (int i = 0; i < 4; ++i) {
+            X(dest_reg + i, ir.CompositeExtract(vector, i));
+        }
+        break;
+    }
+    case LoadSize::U128:
+        throw NotImplementedException("LDG U.128");
+    default:
+        throw NotImplementedException("Invalid LDG size {}", ldg.size.Value());
+    }
+}
+
+void TranslatorVisitor::STG(u64 insn) {
+    // STG stores registers into global memory.
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> data_reg;
+        BitField<46, 2, StoreCache> cache;
+        BitField<48, 3, StoreSize> size;
+    } const stg{insn};
 
+    // Pointer to store data into
+    const IR::U64 address{Address(*this, insn)};
+    const IR::Reg data_reg{stg.data_reg};
     switch (stg.size) {
     case StoreSize::U8:
-        ir.WriteGlobalU8(address, X(stg.data_reg));
+        ir.WriteGlobalU8(address, X(data_reg));
         break;
     case StoreSize::S8:
-        ir.WriteGlobalS8(address, X(stg.data_reg));
+        ir.WriteGlobalS8(address, X(data_reg));
         break;
     case StoreSize::U16:
-        ir.WriteGlobalU16(address, X(stg.data_reg));
+        ir.WriteGlobalU16(address, X(data_reg));
         break;
     case StoreSize::S16:
-        ir.WriteGlobalS16(address, X(stg.data_reg));
+        ir.WriteGlobalS16(address, X(data_reg));
         break;
     case StoreSize::B32:
-        ir.WriteGlobal32(address, X(stg.data_reg));
+        ir.WriteGlobal32(address, X(data_reg));
         break;
     case StoreSize::B64: {
-        if (!IR::IsAligned(stg.data_reg, 2)) {
+        if (!IR::IsAligned(data_reg, 2)) {
             throw NotImplementedException("Unaligned data registers");
         }
-        const IR::Value vector{ir.CompositeConstruct(X(stg.data_reg), X(stg.data_reg + 1))};
+        const IR::Value vector{ir.CompositeConstruct(X(data_reg), X(data_reg + 1))};
         ir.WriteGlobal64(address, vector);
         break;
     }
     case StoreSize::B128:
-        if (!IR::IsAligned(stg.data_reg, 4)) {
+        if (!IR::IsAligned(data_reg, 4)) {
             throw NotImplementedException("Unaligned data registers");
         }
-        const IR::Value vector{ir.CompositeConstruct(X(stg.data_reg), X(stg.data_reg + 1),
-                                                     X(stg.data_reg + 2), X(stg.data_reg + 3))};
+        const IR::Value vector{
+            ir.CompositeConstruct(X(data_reg), X(data_reg + 1), X(data_reg + 2), X(data_reg + 3))};
         ir.WriteGlobal128(address, vector);
         break;
     }

src/shader_recompiler/frontend/maxwell/translate/impl/register_move.cpp → src/shader_recompiler/frontend/maxwell/translate/impl/move_register.cpp

@@ -39,7 +39,7 @@ void TranslatorVisitor::MOV_cbuf(u64 insn) {
 void TranslatorVisitor::MOV_imm(u64 insn) {
     const MOV mov{insn};
     CheckMask(mov);
-    X(mov.dest_reg, GetImm(insn));
+    X(mov.dest_reg, GetImm20(insn));
 }
 
 } // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/move_special_register.cpp

@@ -0,0 +1,114 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/bit_field.h"
+#include "common/common_types.h"
+#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
+
+namespace Shader::Maxwell {
+namespace {
+enum class SpecialRegister : u64 {
+    SR_LANEID = 0,
+    SR_VIRTCFG = 2,
+    SR_VIRTID = 3,
+    SR_PM0 = 4,
+    SR_PM1 = 5,
+    SR_PM2 = 6,
+    SR_PM3 = 7,
+    SR_PM4 = 8,
+    SR_PM5 = 9,
+    SR_PM6 = 10,
+    SR_PM7 = 11,
+    SR_ORDERING_TICKET = 15,
+    SR_PRIM_TYPE = 16,
+    SR_INVOCATION_ID = 17,
+    SR_Y_DIRECTION = 18,
+    SR_THREAD_KILL = 19,
+    SM_SHADER_TYPE = 20,
+    SR_DIRECTCBEWRITEADDRESSLOW = 21,
+    SR_DIRECTCBEWRITEADDRESSHIGH = 22,
+    SR_DIRECTCBEWRITEENABLE = 23,
+    SR_MACHINE_ID_0 = 24,
+    SR_MACHINE_ID_1 = 25,
+    SR_MACHINE_ID_2 = 26,
+    SR_MACHINE_ID_3 = 27,
+    SR_AFFINITY = 28,
+    SR_INVOCATION_INFO = 29,
+    SR_WSCALEFACTOR_XY = 30,
+    SR_WSCALEFACTOR_Z = 31,
+    SR_TID = 32,
+    SR_TID_X = 33,
+    SR_TID_Y = 34,
+    SR_TID_Z = 35,
+    SR_CTAID_X = 37,
+    SR_CTAID_Y = 38,
+    SR_CTAID_Z = 39,
+    SR_NTID = 49,
+    SR_CirQueueIncrMinusOne = 50,
+    SR_NLATC = 51,
+    SR_SWINLO = 57,
+    SR_SWINSZ = 58,
+    SR_SMEMSZ = 59,
+    SR_SMEMBANKS = 60,
+    SR_LWINLO = 61,
+    SR_LWINSZ = 62,
+    SR_LMEMLOSZ = 63,
+    SR_LMEMHIOFF = 64,
+    SR_EQMASK = 65,
+    SR_LTMASK = 66,
+    SR_LEMASK = 67,
+    SR_GTMASK = 68,
+    SR_GEMASK = 69,
+    SR_REGALLOC = 70,
+    SR_GLOBALERRORSTATUS = 73,
+    SR_WARPERRORSTATUS = 75,
+    SR_PM_HI0 = 81,
+    SR_PM_HI1 = 82,
+    SR_PM_HI2 = 83,
+    SR_PM_HI3 = 84,
+    SR_PM_HI4 = 85,
+    SR_PM_HI5 = 86,
+    SR_PM_HI6 = 87,
+    SR_PM_HI7 = 88,
+    SR_CLOCKLO = 89,
+    SR_CLOCKHI = 90,
+    SR_GLOBALTIMERLO = 91,
+    SR_GLOBALTIMERHI = 92,
+    SR_HWTASKID = 105,
+    SR_CIRCULARQUEUEENTRYINDEX = 106,
+    SR_CIRCULARQUEUEENTRYADDRESSLOW = 107,
+    SR_CIRCULARQUEUEENTRYADDRESSHIGH = 108,
+};
+
+[[nodiscard]] IR::U32 Read(IR::IREmitter& ir, SpecialRegister special_register) {
+    switch (special_register) {
+    case SpecialRegister::SR_TID_X:
+        return ir.LocalInvocationIdX();
+    case SpecialRegister::SR_TID_Y:
+        return ir.LocalInvocationIdY();
+    case SpecialRegister::SR_TID_Z:
+        return ir.LocalInvocationIdZ();
+    case SpecialRegister::SR_CTAID_X:
+        return ir.WorkgroupIdX();
+    case SpecialRegister::SR_CTAID_Y:
+        return ir.WorkgroupIdY();
+    case SpecialRegister::SR_CTAID_Z:
+        return ir.WorkgroupIdZ();
+    default:
+        throw NotImplementedException("S2R special register {}", special_register);
+    }
+}
+} // Anonymous namespace
+
+void TranslatorVisitor::S2R(u64 insn) {
+    union {
+        u64 raw;
+        BitField<0, 8, IR::Reg> dest_reg;
+        BitField<20, 8, SpecialRegister> src_reg;
+    } const s2r{insn};
+
+    X(s2r.dest_reg, Read(ir, s2r.src_reg));
+}
+
+} // namespace Shader::Maxwell

src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp

@@ -7,21 +7,8 @@
 #include "shader_recompiler/frontend/maxwell/opcode.h"
 #include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
 
-#include "shader_recompiler/ir_opt/passes.h"
-
 namespace Shader::Maxwell {
 
-[[maybe_unused]] static inline void DumpOptimized(IR::Block& block) {
-    auto raw{IR::DumpBlock(block)};
-
-    Optimization::GetSetElimination(block);
-    Optimization::DeadCodeEliminationPass(block);
-    Optimization::IdentityRemovalPass(block);
-    auto dumped{IR::DumpBlock(block)};
-
-    fmt::print(stderr, "{}", dumped);
-}
-
 [[noreturn]] static void ThrowNotImplemented(Opcode opcode) {
     throw NotImplementedException("Instruction {} is not implemented", opcode);
 }
@@ -146,8 +133,8 @@ void TranslatorVisitor::DADD_imm(u64) {
     ThrowNotImplemented(Opcode::DADD_imm);
 }
 
-void TranslatorVisitor::DEPBAR(u64) {
-    ThrowNotImplemented(Opcode::DEPBAR);
+void TranslatorVisitor::DEPBAR() {
+    // DEPBAR is a no-op
 }
 
 void TranslatorVisitor::DFMA_reg(u64) {
@@ -230,22 +217,6 @@ void TranslatorVisitor::F2F_imm(u64) {
     ThrowNotImplemented(Opcode::F2F_imm);
 }
 
-void TranslatorVisitor::FADD_reg(u64) {
-    ThrowNotImplemented(Opcode::FADD_reg);
-}
-
-void TranslatorVisitor::FADD_cbuf(u64) {
-    ThrowNotImplemented(Opcode::FADD_cbuf);
-}
-
-void TranslatorVisitor::FADD_imm(u64) {
-    ThrowNotImplemented(Opcode::FADD_imm);
-}
-
-void TranslatorVisitor::FADD32I(u64) {
-    ThrowNotImplemented(Opcode::FADD32I);
-}
-
 void TranslatorVisitor::FCHK_reg(u64) {
     ThrowNotImplemented(Opcode::FCHK_reg);
 }
@@ -274,26 +245,6 @@ void TranslatorVisitor::FCMP_imm(u64) {
     ThrowNotImplemented(Opcode::FCMP_imm);
 }
 
-void TranslatorVisitor::FFMA_reg(u64) {
-    ThrowNotImplemented(Opcode::FFMA_reg);
-}
-
-void TranslatorVisitor::FFMA_rc(u64) {
-    ThrowNotImplemented(Opcode::FFMA_rc);
-}
-
-void TranslatorVisitor::FFMA_cr(u64) {
-    ThrowNotImplemented(Opcode::FFMA_cr);
-}
-
-void TranslatorVisitor::FFMA_imm(u64) {
-    ThrowNotImplemented(Opcode::FFMA_imm);
-}
-
-void TranslatorVisitor::FFMA32I(u64) {
-    ThrowNotImplemented(Opcode::FFMA32I);
-}
-
 void TranslatorVisitor::FLO_reg(u64) {
     ThrowNotImplemented(Opcode::FLO_reg);
 }
@@ -318,22 +269,6 @@ void TranslatorVisitor::FMNMX_imm(u64) {
     ThrowNotImplemented(Opcode::FMNMX_imm);
 }
 
-void TranslatorVisitor::FMUL_reg(u64) {
-    ThrowNotImplemented(Opcode::FMUL_reg);
-}
-
-void TranslatorVisitor::FMUL_cbuf(u64) {
-    ThrowNotImplemented(Opcode::FMUL_cbuf);
-}
-
-void TranslatorVisitor::FMUL_imm(u64) {
-    ThrowNotImplemented(Opcode::FMUL_imm);
-}
-
-void TranslatorVisitor::FMUL32I(u64) {
-    ThrowNotImplemented(Opcode::FMUL32I);
-}
-
 void TranslatorVisitor::FSET_reg(u64) {
     ThrowNotImplemented(Opcode::FSET_reg);
 }
@@ -470,18 +405,6 @@ void TranslatorVisitor::I2I_imm(u64) {
     ThrowNotImplemented(Opcode::I2I_imm);
 }
 
-void TranslatorVisitor::IADD_reg(u64) {
-    ThrowNotImplemented(Opcode::IADD_reg);
-}
-
-void TranslatorVisitor::IADD_cbuf(u64) {
-    ThrowNotImplemented(Opcode::IADD_cbuf);
-}
-
-void TranslatorVisitor::IADD_imm(u64) {
-    ThrowNotImplemented(Opcode::IADD_imm);
-}
-
 void TranslatorVisitor::IADD3_reg(u64) {
     ThrowNotImplemented(Opcode::IADD3_reg);
 }
@@ -494,10 +417,6 @@ void TranslatorVisitor::IADD3_imm(u64) {
     ThrowNotImplemented(Opcode::IADD3_imm);
 }
 
-void TranslatorVisitor::IADD32I(u64) {
-    ThrowNotImplemented(Opcode::IADD32I);
-}
-
 void TranslatorVisitor::ICMP_reg(u64) {
     ThrowNotImplemented(Opcode::ICMP_reg);
 }
@@ -594,22 +513,6 @@ void TranslatorVisitor::ISBERD(u64) {
     ThrowNotImplemented(Opcode::ISBERD);
 }
 
-void TranslatorVisitor::ISCADD_reg(u64) {
-    ThrowNotImplemented(Opcode::ISCADD_reg);
-}
-
-void TranslatorVisitor::ISCADD_cbuf(u64) {
-    ThrowNotImplemented(Opcode::ISCADD_cbuf);
-}
-
-void TranslatorVisitor::ISCADD_imm(u64) {
-    ThrowNotImplemented(Opcode::ISCADD_imm);
-}
-
-void TranslatorVisitor::ISCADD32I(u64) {
-    ThrowNotImplemented(Opcode::ISCADD32I);
-}
-
 void TranslatorVisitor::ISET_reg(u64) {
     ThrowNotImplemented(Opcode::ISET_reg);
 }
@@ -622,18 +525,6 @@ void TranslatorVisitor::ISET_imm(u64) {
     ThrowNotImplemented(Opcode::ISET_imm);
 }
 
-void TranslatorVisitor::ISETP_reg(u64) {
-    ThrowNotImplemented(Opcode::ISETP_reg);
-}
-
-void TranslatorVisitor::ISETP_cbuf(u64) {
-    ThrowNotImplemented(Opcode::ISETP_cbuf);
-}
-
-void TranslatorVisitor::ISETP_imm(u64) {
-    ThrowNotImplemented(Opcode::ISETP_imm);
-}
-
 void TranslatorVisitor::JCAL(u64) {
     ThrowNotImplemented(Opcode::JCAL);
 }
@@ -658,10 +549,6 @@ void TranslatorVisitor::LDC(u64) {
     ThrowNotImplemented(Opcode::LDC);
 }
 
-void TranslatorVisitor::LDG(u64) {
-    ThrowNotImplemented(Opcode::LDG);
-}
-
 void TranslatorVisitor::LDL(u64) {
     ThrowNotImplemented(Opcode::LDL);
 }
@@ -866,10 +753,6 @@ void TranslatorVisitor::RTT(u64) {
     ThrowNotImplemented(Opcode::RTT);
 }
 
-void TranslatorVisitor::S2R(u64) {
-    ThrowNotImplemented(Opcode::S2R);
-}
-
 void TranslatorVisitor::SAM(u64) {
     ThrowNotImplemented(Opcode::SAM);
 }
@@ -914,18 +797,6 @@ void TranslatorVisitor::SHFL(u64) {
     ThrowNotImplemented(Opcode::SHFL);
 }
 
-void TranslatorVisitor::SHL_reg(u64) {
-    ThrowNotImplemented(Opcode::SHL_reg);
-}
-
-void TranslatorVisitor::SHL_cbuf(u64) {
-    ThrowNotImplemented(Opcode::SHL_cbuf);
-}
-
-void TranslatorVisitor::SHL_imm(u64) {
-    ThrowNotImplemented(Opcode::SHL_imm);
-}
-
 void TranslatorVisitor::SHR_reg(u64) {
     ThrowNotImplemented(Opcode::SHR_reg);
 }
@@ -1086,20 +957,4 @@ void TranslatorVisitor::VSHR(u64) {
     ThrowNotImplemented(Opcode::VSHR);
 }
 
-void TranslatorVisitor::XMAD_reg(u64) {
-    ThrowNotImplemented(Opcode::XMAD_reg);
-}
-
-void TranslatorVisitor::XMAD_rc(u64) {
-    ThrowNotImplemented(Opcode::XMAD_rc);
-}
-
-void TranslatorVisitor::XMAD_cr(u64) {
-    ThrowNotImplemented(Opcode::XMAD_cr);
-}
-
-void TranslatorVisitor::XMAD_imm(u64) {
-    ThrowNotImplemented(Opcode::XMAD_imm);
-}
-
 } // namespace Shader::Maxwell

src/shader_recompiler/ir_opt/get_set_elimination_pass.cpp

@@ -1,87 +0,0 @@
-// Copyright 2021 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <array>
-
-#include "shader_recompiler/frontend/ir/basic_block.h"
-#include "shader_recompiler/frontend/ir/microinstruction.h"
-#include "shader_recompiler/ir_opt/passes.h"
-
-namespace Shader::Optimization {
-namespace {
-using Iterator = IR::Block::iterator;
-
-enum class TrackingType {
-    Reg,
-};
-
-struct RegisterInfo {
-    IR::Value register_value;
-    TrackingType tracking_type;
-    Iterator last_set_instruction;
-    bool set_instruction_present = false;
-};
-
-void DoSet(IR::Block& block, RegisterInfo& info, IR::Value value, Iterator set_inst,
-           TrackingType tracking_type) {
-    if (info.set_instruction_present) {
-        info.last_set_instruction->Invalidate();
-        block.Instructions().erase(info.last_set_instruction);
-    }
-    info.register_value = value;
-    info.tracking_type = tracking_type;
-    info.set_instruction_present = true;
-    info.last_set_instruction = set_inst;
-}
-
-RegisterInfo Nothing(Iterator get_inst, TrackingType tracking_type) {
-    RegisterInfo info{};
-    info.register_value = IR::Value{&*get_inst};
-    info.tracking_type = tracking_type;
-    return info;
-}
-
-void DoGet(RegisterInfo& info, Iterator get_inst, TrackingType tracking_type) {
-    if (info.register_value.IsEmpty()) {
-        info = Nothing(get_inst, tracking_type);
-        return;
-    }
-    if (info.tracking_type == tracking_type) {
-        get_inst->ReplaceUsesWith(info.register_value);
-        return;
-    }
-    info = Nothing(get_inst, tracking_type);
-}
-} // Anonymous namespace
-
-void GetSetElimination(IR::Block& block) {
-    std::array<RegisterInfo, 255> reg_info;
-
-    for (Iterator inst = block.begin(); inst != block.end(); ++inst) {
-        switch (inst->Opcode()) {
-        case IR::Opcode::GetRegister: {
-            const IR::Reg reg{inst->Arg(0).Reg()};
-            if (reg == IR::Reg::RZ) {
-                break;
-            }
-            const size_t index{static_cast<size_t>(reg)};
-            DoGet(reg_info.at(index), inst, TrackingType::Reg);
-            break;
-        }
-        case IR::Opcode::SetRegister: {
-            const IR::Reg reg{inst->Arg(0).Reg()};
-            if (reg == IR::Reg::RZ) {
-                break;
-            }
-            const size_t index{static_cast<size_t>(reg)};
-            DoSet(block, reg_info.at(index), inst->Arg(1), inst, TrackingType::Reg);
-            break;
-        }
-        default:
-            break;
-        }
-    }
-}
-
-} // namespace Shader::Optimization

src/shader_recompiler/ir_opt/passes.h

@@ -17,7 +17,6 @@ void Invoke(Func&& func, IR::Function& function) {
 }
 
 void DeadCodeEliminationPass(IR::Block& block);
-void GetSetElimination(IR::Block& block);
 void IdentityRemovalPass(IR::Block& block);
 void SsaRewritePass(IR::Function& function);
 void VerificationPass(const IR::Block& block);

src/shader_recompiler/main.cpp

@@ -51,7 +51,8 @@ void RunDatabase() {
 int main() {
     // RunDatabase();
 
-    FileEnvironment env{"D:\\Shaders\\Database\\test.bin"};
+    // FileEnvironment env{"D:\\Shaders\\Database\\test.bin"};
+    FileEnvironment env{"D:\\Shaders\\Database\\Oninaki\\CS8F146B41DB6BD826.bin"};
     auto cfg{std::make_unique<Flow::CFG>(env, 0)};
     // fmt::print(stdout, "{}\n", cfg->Dot());