VERY experimental Nintendo Switch 2 emulator
C++
[dynarmic] merge IR opt pass into single TU (#2561)
lizzief5bb07341ab5
diff --git a/src/dynarmic/src/dynarmic/CMakeLists.txt b/src/dynarmic/src/dynarmic/CMakeLists.txtindex a5234609cf..9e6bd25913 100644--- a/src/dynarmic/src/dynarmic/CMakeLists.txt+++ b/src/dynarmic/src/dynarmic/CMakeLists.txt@@ -89,14 +89,8 @@ add_library(dynarmicir/opcodes.cppir/opcodes.hir/opcodes.inc- ir/opt/constant_propagation_pass.cpp- ir/opt/dead_code_elimination_pass.cpp- ir/opt/identity_removal_pass.cpp- ir/opt/ir_matcher.h- ir/opt/naming_pass.cpp- ir/opt/passes.h- ir/opt/polyfill_pass.cpp- ir/opt/verification_pass.cpp+ ir/opt_passes.cpp+ ir/opt_passes.hir/terminal.hir/type.cppir/type.h@@ -135,8 +129,6 @@ add_library(dynarmicinterface/A32/coprocessor.hinterface/A32/coprocessor_util.hinterface/A32/disassembler.h- ir/opt/a32_constant_memory_reads_pass.cpp- ir/opt/a32_get_set_elimination_pass.cpp# A64frontend/A64/a64_ir_emitter.cppfrontend/A64/a64_ir_emitter.h@@ -148,9 +140,6 @@ add_library(dynarmicfrontend/A64/translate/a64_translate.hinterface/A64/a64.hinterface/A64/config.h- ir/opt/a64_callback_config_pass.cpp- ir/opt/a64_get_set_elimination_pass.cpp- ir/opt/a64_merge_interpret_blocks.cpp)if ("x86_64" IN_LIST ARCHITECTURE)diff --git a/src/dynarmic/src/dynarmic/backend/arm64/a32_address_space.cpp b/src/dynarmic/src/dynarmic/backend/arm64/a32_address_space.cppindex bbdf925ff4..01af361b27 100644--- a/src/dynarmic/src/dynarmic/backend/arm64/a32_address_space.cpp+++ b/src/dynarmic/src/dynarmic/backend/arm64/a32_address_space.cpp@@ -1,3 +1,6 @@+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project+// SPDX-License-Identifier: GPL-3.0-or-later+/* This file is part of the dynarmic project.* Copyright (c) 2022 MerryMage* SPDX-License-Identifier: 0BSD@@ -16,7 +19,7 @@#include "dynarmic/frontend/A32/translate/a32_translate.h"#include "dynarmic/interface/A32/config.h"#include "dynarmic/interface/exclusive_monitor.h"-#include "dynarmic/ir/opt/passes.h"+#include "dynarmic/ir/opt_passes.h"namespace Dynarmic::Backend::Arm64 {@@ -163,21 +166,7 @@ A32AddressSpace::A32AddressSpace(const A32::UserConfig& conf)IR::Block A32AddressSpace::GenerateIR(IR::LocationDescriptor descriptor) const {IR::Block ir_block = A32::Translate(A32::LocationDescriptor{descriptor}, conf.callbacks, {conf.arch_version, conf.define_unpredictable_behaviour, conf.hook_hint_instructions});-- Optimization::PolyfillPass(ir_block, {});- Optimization::NamingPass(ir_block);- if (conf.HasOptimization(OptimizationFlag::GetSetElimination)) {- Optimization::A32GetSetElimination(ir_block, {.convert_nzc_to_nz = true});- Optimization::DeadCodeElimination(ir_block);- }- if (conf.HasOptimization(OptimizationFlag::ConstProp)) {- Optimization::A32ConstantMemoryReads(ir_block, conf.callbacks);- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);- }- Optimization::IdentityRemovalPass(ir_block);- Optimization::VerificationPass(ir_block);-+ Optimization::Optimize(ir_block, conf, {});return ir_block;}diff --git a/src/dynarmic/src/dynarmic/backend/arm64/a64_address_space.cpp b/src/dynarmic/src/dynarmic/backend/arm64/a64_address_space.cppindex d4fe9a9cb7..c4c1c42792 100644--- a/src/dynarmic/src/dynarmic/backend/arm64/a64_address_space.cpp+++ b/src/dynarmic/src/dynarmic/backend/arm64/a64_address_space.cpp@@ -1,3 +1,6 @@+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project+// SPDX-License-Identifier: GPL-3.0-or-later+/* This file is part of the dynarmic project.* Copyright (c) 2022 MerryMage* SPDX-License-Identifier: 0BSD@@ -15,7 +18,7 @@#include "dynarmic/frontend/A64/translate/a64_translate.h"#include "dynarmic/interface/A64/config.h"#include "dynarmic/interface/exclusive_monitor.h"-#include "dynarmic/ir/opt/passes.h"+#include "dynarmic/ir/opt_passes.h"namespace Dynarmic::Backend::Arm64 {@@ -331,22 +334,7 @@ IR::Block A64AddressSpace::GenerateIR(IR::LocationDescriptor descriptor) const {const auto get_code = [this](u64 vaddr) { return conf.callbacks->MemoryReadCode(vaddr); };IR::Block ir_block = A64::Translate(A64::LocationDescriptor{descriptor}, get_code,{conf.define_unpredictable_behaviour, conf.wall_clock_cntpct});-- Optimization::A64CallbackConfigPass(ir_block, conf);- Optimization::NamingPass(ir_block);- if (conf.HasOptimization(OptimizationFlag::GetSetElimination) && !conf.check_halt_on_memory_access) {- Optimization::A64GetSetElimination(ir_block);- Optimization::DeadCodeElimination(ir_block);- }- if (conf.HasOptimization(OptimizationFlag::ConstProp)) {- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);- }- if (conf.HasOptimization(OptimizationFlag::MiscIROpt)) {- Optimization::A64MergeInterpretBlocksPass(ir_block, conf.callbacks);- }- Optimization::VerificationPass(ir_block);-+ Optimization::Optimize(ir_block, conf, {});return ir_block;}diff --git a/src/dynarmic/src/dynarmic/backend/riscv64/a32_address_space.cpp b/src/dynarmic/src/dynarmic/backend/riscv64/a32_address_space.cppindex efa211618b..8a21aafcf8 100644--- a/src/dynarmic/src/dynarmic/backend/riscv64/a32_address_space.cpp+++ b/src/dynarmic/src/dynarmic/backend/riscv64/a32_address_space.cpp@@ -15,7 +15,7 @@#include "dynarmic/backend/riscv64/stack_layout.h"#include "dynarmic/frontend/A32/a32_location_descriptor.h"#include "dynarmic/frontend/A32/translate/a32_translate.h"-#include "dynarmic/ir/opt/passes.h"+#include "dynarmic/ir/opt_passes.h"namespace Dynarmic::Backend::RV64 {diff --git a/src/dynarmic/src/dynarmic/backend/x64/a32_interface.cpp b/src/dynarmic/src/dynarmic/backend/x64/a32_interface.cppindex b116ec180e..382eb70f3f 100644--- a/src/dynarmic/src/dynarmic/backend/x64/a32_interface.cpp+++ b/src/dynarmic/src/dynarmic/backend/x64/a32_interface.cpp@@ -29,7 +29,7 @@#include "dynarmic/interface/A32/a32.h"#include "dynarmic/ir/basic_block.h"#include "dynarmic/ir/location_descriptor.h"-#include "dynarmic/ir/opt/passes.h"+#include "dynarmic/ir/opt_passes.h"namespace Dynarmic::A32 {@@ -217,19 +217,7 @@ private:block_of_code.EnsureMemoryCommitted(MINIMUM_REMAINING_CODESIZE);IR::Block ir_block = A32::Translate(A32::LocationDescriptor{descriptor}, conf.callbacks, {conf.arch_version, conf.define_unpredictable_behaviour, conf.hook_hint_instructions});- Optimization::PolyfillPass(ir_block, polyfill_options);- Optimization::NamingPass(ir_block);- if (conf.HasOptimization(OptimizationFlag::GetSetElimination) && !conf.check_halt_on_memory_access) {- Optimization::A32GetSetElimination(ir_block, {.convert_nz_to_nzc = true});- Optimization::DeadCodeElimination(ir_block);- }- if (conf.HasOptimization(OptimizationFlag::ConstProp)) {- Optimization::A32ConstantMemoryReads(ir_block, conf.callbacks);- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);- }- Optimization::IdentityRemovalPass(ir_block);- Optimization::VerificationPass(ir_block);+ Optimization::Optimize(ir_block, conf, polyfill_options);return emitter.Emit(ir_block);}diff --git a/src/dynarmic/src/dynarmic/backend/x64/a64_interface.cpp b/src/dynarmic/src/dynarmic/backend/x64/a64_interface.cppindex ddd2327395..c65b582982 100644--- a/src/dynarmic/src/dynarmic/backend/x64/a64_interface.cpp+++ b/src/dynarmic/src/dynarmic/backend/x64/a64_interface.cpp@@ -25,7 +25,7 @@#include "dynarmic/frontend/A64/translate/a64_translate.h"#include "dynarmic/interface/A64/a64.h"#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opt/passes.h"+#include "dynarmic/ir/opt_passes.h"namespace Dynarmic::A64 {@@ -275,21 +275,7 @@ private:const auto get_code = [this](u64 vaddr) { return conf.callbacks->MemoryReadCode(vaddr); };IR::Block ir_block = A64::Translate(A64::LocationDescriptor{current_location}, get_code,{conf.define_unpredictable_behaviour, conf.wall_clock_cntpct});- Optimization::PolyfillPass(ir_block, polyfill_options);- Optimization::A64CallbackConfigPass(ir_block, conf);- Optimization::NamingPass(ir_block);- if (conf.HasOptimization(OptimizationFlag::GetSetElimination) && !conf.check_halt_on_memory_access) {- Optimization::A64GetSetElimination(ir_block);- Optimization::DeadCodeElimination(ir_block);- }- if (conf.HasOptimization(OptimizationFlag::ConstProp)) {- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);- }- if (conf.HasOptimization(OptimizationFlag::MiscIROpt)) {- Optimization::A64MergeInterpretBlocksPass(ir_block, conf.callbacks);- }- Optimization::VerificationPass(ir_block);+ Optimization::Optimize(ir_block, conf, polyfill_options);return emitter.Emit(ir_block).entrypoint;}diff --git a/src/dynarmic/src/dynarmic/ir/opt/a32_constant_memory_reads_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/a32_constant_memory_reads_pass.cppdeleted file mode 100644index 9699f18345..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/a32_constant_memory_reads_pass.cpp+++ /dev/null@@ -1,70 +0,0 @@-/* This file is part of the dynarmic project.- * Copyright (c) 2016 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include "dynarmic/interface/A32/config.h"-#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"--namespace Dynarmic::Optimization {--void A32ConstantMemoryReads(IR::Block& block, A32::UserCallbacks* cb) {- for (auto& inst : block) {- switch (inst.GetOpcode()) {- case IR::Opcode::A32ReadMemory8: {- if (!inst.AreAllArgsImmediates()) {- break;- }-- const u32 vaddr = inst.GetArg(1).GetU32();- if (cb->IsReadOnlyMemory(vaddr)) {- const u8 value_from_memory = cb->MemoryRead8(vaddr);- inst.ReplaceUsesWith(IR::Value{value_from_memory});- }- break;- }- case IR::Opcode::A32ReadMemory16: {- if (!inst.AreAllArgsImmediates()) {- break;- }-- const u32 vaddr = inst.GetArg(1).GetU32();- if (cb->IsReadOnlyMemory(vaddr)) {- const u16 value_from_memory = cb->MemoryRead16(vaddr);- inst.ReplaceUsesWith(IR::Value{value_from_memory});- }- break;- }- case IR::Opcode::A32ReadMemory32: {- if (!inst.AreAllArgsImmediates()) {- break;- }-- const u32 vaddr = inst.GetArg(1).GetU32();- if (cb->IsReadOnlyMemory(vaddr)) {- const u32 value_from_memory = cb->MemoryRead32(vaddr);- inst.ReplaceUsesWith(IR::Value{value_from_memory});- }- break;- }- case IR::Opcode::A32ReadMemory64: {- if (!inst.AreAllArgsImmediates()) {- break;- }-- const u32 vaddr = inst.GetArg(1).GetU32();- if (cb->IsReadOnlyMemory(vaddr)) {- const u64 value_from_memory = cb->MemoryRead64(vaddr);- inst.ReplaceUsesWith(IR::Value{value_from_memory});- }- break;- }- default:- break;- }- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/a32_get_set_elimination_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/a32_get_set_elimination_pass.cppdeleted file mode 100644index 499b38b120..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/a32_get_set_elimination_pass.cpp+++ /dev/null@@ -1,382 +0,0 @@-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project-// SPDX-License-Identifier: GPL-3.0-or-later--/* This file is part of the dynarmic project.- * Copyright (c) 2016 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include <algorithm>-#include <array>-#include <functional>--#include "dynarmic/common/assert.h"-#include "dynarmic/common/common_types.h"--#include "dynarmic/frontend/A32/a32_ir_emitter.h"-#include "dynarmic/frontend/A32/a32_types.h"-#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"-#include "dynarmic/ir/value.h"--namespace Dynarmic::Optimization {--namespace {--void FlagsPass(IR::Block& block) {- using Iterator = std::reverse_iterator<IR::Block::iterator>;-- struct FlagInfo {- bool set_not_required = false;- bool has_value_request = false;- Iterator value_request = {};- };- struct ValuelessFlagInfo {- bool set_not_required = false;- };- ValuelessFlagInfo nzcvq;- ValuelessFlagInfo nzcv;- ValuelessFlagInfo nz;- FlagInfo c_flag;- FlagInfo ge;-- auto do_set = [&](FlagInfo& info, IR::Value value, Iterator inst) {- if (info.has_value_request) {- info.value_request->ReplaceUsesWith(value);- }- info.has_value_request = false;-- if (info.set_not_required) {- inst->Invalidate();- }- info.set_not_required = true;- };-- auto do_set_valueless = [&](ValuelessFlagInfo& info, Iterator inst) {- if (info.set_not_required) {- inst->Invalidate();- }- info.set_not_required = true;- };-- auto do_get = [](FlagInfo& info, Iterator inst) {- if (info.has_value_request) {- info.value_request->ReplaceUsesWith(IR::Value{&*inst});- }- info.has_value_request = true;- info.value_request = inst;- };-- A32::IREmitter ir{block, A32::LocationDescriptor{block.Location()}, {}};-- for (auto inst = block.rbegin(); inst != block.rend(); ++inst) {- auto const opcode = inst->GetOpcode();- switch (opcode) {- case IR::Opcode::A32GetCFlag: {- do_get(c_flag, inst);- break;- }- case IR::Opcode::A32SetCpsrNZCV: {- if (c_flag.has_value_request) {- ir.SetInsertionPointBefore(inst.base()); // base is one ahead- IR::U1 c = ir.GetCFlagFromNZCV(IR::NZCV{inst->GetArg(0)});- c_flag.value_request->ReplaceUsesWith(c);- c_flag.has_value_request = false;- break; // This case will be executed again because of the above- }-- do_set_valueless(nzcv, inst);-- nz = {.set_not_required = true};- c_flag = {.set_not_required = true};- break;- }- case IR::Opcode::A32SetCpsrNZCVRaw: {- if (c_flag.has_value_request) {- nzcv.set_not_required = false;- }-- do_set_valueless(nzcv, inst);-- nzcvq = {};- nz = {.set_not_required = true};- c_flag = {.set_not_required = true};- break;- }- case IR::Opcode::A32SetCpsrNZCVQ: {- if (c_flag.has_value_request) {- nzcvq.set_not_required = false;- }-- do_set_valueless(nzcvq, inst);-- nzcv = {.set_not_required = true};- nz = {.set_not_required = true};- c_flag = {.set_not_required = true};- break;- }- case IR::Opcode::A32SetCpsrNZ: {- do_set_valueless(nz, inst);-- nzcvq = {};- nzcv = {};- break;- }- case IR::Opcode::A32SetCpsrNZC: {- if (c_flag.has_value_request) {- c_flag.value_request->ReplaceUsesWith(inst->GetArg(1));- c_flag.has_value_request = false;- }-- if (!inst->GetArg(1).IsImmediate() && inst->GetArg(1).GetInstRecursive()->GetOpcode() == IR::Opcode::A32GetCFlag) {- const auto nz_value = inst->GetArg(0);-- inst->Invalidate();-- ir.SetInsertionPointBefore(inst.base());- ir.SetCpsrNZ(IR::NZCV{nz_value});-- nzcvq = {};- nzcv = {};- nz = {.set_not_required = true};- break;- }-- if (nz.set_not_required && c_flag.set_not_required) {- inst->Invalidate();- } else if (nz.set_not_required) {- inst->SetArg(0, IR::Value::EmptyNZCVImmediateMarker());- }- nz.set_not_required = true;- c_flag.set_not_required = true;-- nzcv = {};- nzcvq = {};- break;- }- case IR::Opcode::A32SetGEFlags: {- do_set(ge, inst->GetArg(0), inst);- break;- }- case IR::Opcode::A32GetGEFlags: {- do_get(ge, inst);- break;- }- case IR::Opcode::A32SetGEFlagsCompressed: {- ge = {.set_not_required = true};- break;- }- case IR::Opcode::A32OrQFlag: {- break;- }- default: {- if (ReadsFromCPSR(opcode) || WritesToCPSR(opcode)) {- nzcvq = {};- nzcv = {};- nz = {};- c_flag = {};- ge = {};- }- break;- }- }- }-}--void RegisterPass(IR::Block& block) {- using Iterator = IR::Block::iterator;-- struct RegInfo {- IR::Value register_value;- std::optional<Iterator> last_set_instruction;- };- std::array<RegInfo, 15> reg_info;-- const auto do_get = [](RegInfo& info, Iterator get_inst) {- if (info.register_value.IsEmpty()) {- info.register_value = IR::Value(&*get_inst);- return;- }- get_inst->ReplaceUsesWith(info.register_value);- };-- const auto do_set = [](RegInfo& info, IR::Value value, Iterator set_inst) {- if (info.last_set_instruction) {- (*info.last_set_instruction)->Invalidate();- }- info = {- .register_value = value,- .last_set_instruction = set_inst,- };- };-- enum class ExtValueType {- Empty,- Single,- Double,- VectorDouble,- VectorQuad,- };- struct ExtRegInfo {- ExtValueType value_type = {};- IR::Value register_value;- std::optional<Iterator> last_set_instruction;- };- std::array<ExtRegInfo, 64> ext_reg_info;-- const auto do_ext_get = [](ExtValueType type, std::initializer_list<std::reference_wrapper<ExtRegInfo>> infos, Iterator get_inst) {- if (!std::all_of(infos.begin(), infos.end(), [type](const auto& info) { return info.get().value_type == type; })) {- for (auto& info : infos) {- info.get() = {- .value_type = type,- .register_value = IR::Value(&*get_inst),- .last_set_instruction = std::nullopt,- };- }- return;- }- get_inst->ReplaceUsesWith(std::data(infos)[0].get().register_value);- };-- const auto do_ext_set = [](ExtValueType type, std::initializer_list<std::reference_wrapper<ExtRegInfo>> infos, IR::Value value, Iterator set_inst) {- if (std::all_of(infos.begin(), infos.end(), [type](const auto& info) { return info.get().value_type == type; })) {- if (std::data(infos)[0].get().last_set_instruction) {- (*std::data(infos)[0].get().last_set_instruction)->Invalidate();- }- }- for (auto& info : infos) {- info.get() = {- .value_type = type,- .register_value = value,- .last_set_instruction = set_inst,- };- }- };-- // Location and version don't matter here.- A32::IREmitter ir{block, A32::LocationDescriptor{block.Location()}, {}};-- for (auto inst = block.begin(); inst != block.end(); ++inst) {- auto const opcode = inst->GetOpcode();- switch (opcode) {- case IR::Opcode::A32GetRegister: {- const A32::Reg reg = inst->GetArg(0).GetA32RegRef();- ASSERT(reg != A32::Reg::PC);- const size_t reg_index = static_cast<size_t>(reg);- do_get(reg_info[reg_index], inst);- break;- }- case IR::Opcode::A32SetRegister: {- const A32::Reg reg = inst->GetArg(0).GetA32RegRef();- if (reg == A32::Reg::PC) {- break;- }- const auto reg_index = static_cast<size_t>(reg);- do_set(reg_info[reg_index], inst->GetArg(1), inst);- break;- }- case IR::Opcode::A32GetExtendedRegister32: {- const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();- const size_t reg_index = A32::RegNumber(reg);- do_ext_get(ExtValueType::Single, {ext_reg_info[reg_index]}, inst);- break;- }- case IR::Opcode::A32SetExtendedRegister32: {- const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();- const size_t reg_index = A32::RegNumber(reg);- do_ext_set(ExtValueType::Single, {ext_reg_info[reg_index]}, inst->GetArg(1), inst);- break;- }- case IR::Opcode::A32GetExtendedRegister64: {- const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();- const size_t reg_index = A32::RegNumber(reg);- do_ext_get(ExtValueType::Double,- {- ext_reg_info[reg_index * 2 + 0],- ext_reg_info[reg_index * 2 + 1],- },- inst);- break;- }- case IR::Opcode::A32SetExtendedRegister64: {- const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();- const size_t reg_index = A32::RegNumber(reg);- do_ext_set(ExtValueType::Double,- {- ext_reg_info[reg_index * 2 + 0],- ext_reg_info[reg_index * 2 + 1],- },- inst->GetArg(1),- inst);- break;- }- case IR::Opcode::A32GetVector: {- const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();- const size_t reg_index = A32::RegNumber(reg);- if (A32::IsDoubleExtReg(reg)) {- do_ext_get(ExtValueType::VectorDouble,- {- ext_reg_info[reg_index * 2 + 0],- ext_reg_info[reg_index * 2 + 1],- },- inst);- } else {- DEBUG_ASSERT(A32::IsQuadExtReg(reg));- do_ext_get(ExtValueType::VectorQuad,- {- ext_reg_info[reg_index * 4 + 0],- ext_reg_info[reg_index * 4 + 1],- ext_reg_info[reg_index * 4 + 2],- ext_reg_info[reg_index * 4 + 3],- },- inst);- }- break;- }- case IR::Opcode::A32SetVector: {- const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();- const size_t reg_index = A32::RegNumber(reg);- if (A32::IsDoubleExtReg(reg)) {- ir.SetInsertionPointAfter(inst);- const IR::U128 stored_value = ir.VectorZeroUpper(IR::U128{inst->GetArg(1)});- do_ext_set(ExtValueType::VectorDouble,- {- ext_reg_info[reg_index * 2 + 0],- ext_reg_info[reg_index * 2 + 1],- },- stored_value,- inst);- } else {- DEBUG_ASSERT(A32::IsQuadExtReg(reg));- do_ext_set(ExtValueType::VectorQuad,- {- ext_reg_info[reg_index * 4 + 0],- ext_reg_info[reg_index * 4 + 1],- ext_reg_info[reg_index * 4 + 2],- ext_reg_info[reg_index * 4 + 3],- },- inst->GetArg(1),- inst);- }- break;- }- default: {- if (ReadsFromCoreRegister(opcode) || WritesToCoreRegister(opcode)) {- reg_info = {};- ext_reg_info = {};- }- break;- }- }- }-}--} // namespace--void A32GetSetElimination(IR::Block& block, A32GetSetEliminationOptions) {- FlagsPass(block);- RegisterPass(block);-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/a64_callback_config_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/a64_callback_config_pass.cppdeleted file mode 100644index 79d9769520..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/a64_callback_config_pass.cpp+++ /dev/null@@ -1,57 +0,0 @@-/* This file is part of the dynarmic project.- * Copyright (c) 2018 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include "dynarmic/frontend/A64/a64_ir_emitter.h"-#include "dynarmic/interface/A64/config.h"-#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/microinstruction.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"--namespace Dynarmic::Optimization {--void A64CallbackConfigPass(IR::Block& block, const A64::UserConfig& conf) {- if (conf.hook_data_cache_operations) {- return;- }-- for (auto& inst : block) {- if (inst.GetOpcode() != IR::Opcode::A64DataCacheOperationRaised) {- continue;- }-- const auto op = static_cast<A64::DataCacheOperation>(inst.GetArg(1).GetU64());- if (op == A64::DataCacheOperation::ZeroByVA) {- A64::IREmitter ir{block};- ir.current_location = A64::LocationDescriptor{IR::LocationDescriptor{inst.GetArg(0).GetU64()}};- ir.SetInsertionPointBefore(&inst);-- size_t bytes = 4 << static_cast<size_t>(conf.dczid_el0 & 0b1111);- IR::U64 addr{inst.GetArg(2)};-- const IR::U128 zero_u128 = ir.ZeroExtendToQuad(ir.Imm64(0));- while (bytes >= 16) {- ir.WriteMemory128(addr, zero_u128, IR::AccType::DCZVA);- addr = ir.Add(addr, ir.Imm64(16));- bytes -= 16;- }-- while (bytes >= 8) {- ir.WriteMemory64(addr, ir.Imm64(0), IR::AccType::DCZVA);- addr = ir.Add(addr, ir.Imm64(8));- bytes -= 8;- }-- while (bytes >= 4) {- ir.WriteMemory32(addr, ir.Imm32(0), IR::AccType::DCZVA);- addr = ir.Add(addr, ir.Imm64(4));- bytes -= 4;- }- }- inst.Invalidate();- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/a64_get_set_elimination_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/a64_get_set_elimination_pass.cppdeleted file mode 100644index 53e3b27176..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/a64_get_set_elimination_pass.cpp+++ /dev/null@@ -1,165 +0,0 @@-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project-// SPDX-License-Identifier: GPL-3.0-or-later--/* This file is part of the dynarmic project.- * Copyright (c) 2016 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include <array>--#include "dynarmic/common/common_types.h"--#include "dynarmic/frontend/A64/a64_types.h"-#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"-#include "dynarmic/ir/value.h"--namespace Dynarmic::Optimization {--void A64GetSetElimination(IR::Block& block) {- using Iterator = IR::Block::iterator;-- enum class TrackingType {- W,- X,- S,- D,- Q,- SP,- NZCV,- NZCVRaw,- };- struct RegisterInfo {- IR::Value register_value;- TrackingType tracking_type;- bool set_instruction_present = false;- Iterator last_set_instruction;- };- std::array<RegisterInfo, 31> reg_info;- std::array<RegisterInfo, 32> vec_info;- RegisterInfo sp_info;- RegisterInfo nzcv_info;-- const auto do_set = [&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;- };-- const auto do_get = [](RegisterInfo& info, Iterator get_inst, TrackingType tracking_type) {- const auto do_nothing = [&] {- info = {};- info.register_value = IR::Value(&*get_inst);- info.tracking_type = tracking_type;- };-- if (info.register_value.IsEmpty()) {- do_nothing();- return;- }-- if (info.tracking_type == tracking_type) {- get_inst->ReplaceUsesWith(info.register_value);- return;- }-- do_nothing();- };-- for (auto inst = block.begin(); inst != block.end(); ++inst) {- auto const opcode = inst->GetOpcode();- switch (opcode) {- case IR::Opcode::A64GetW: {- const size_t index = A64::RegNumber(inst->GetArg(0).GetA64RegRef());- do_get(reg_info.at(index), inst, TrackingType::W);- break;- }- case IR::Opcode::A64GetX: {- const size_t index = A64::RegNumber(inst->GetArg(0).GetA64RegRef());- do_get(reg_info.at(index), inst, TrackingType::X);- break;- }- case IR::Opcode::A64GetS: {- const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());- do_get(vec_info.at(index), inst, TrackingType::S);- break;- }- case IR::Opcode::A64GetD: {- const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());- do_get(vec_info.at(index), inst, TrackingType::D);- break;- }- case IR::Opcode::A64GetQ: {- const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());- do_get(vec_info.at(index), inst, TrackingType::Q);- break;- }- case IR::Opcode::A64GetSP: {- do_get(sp_info, inst, TrackingType::SP);- break;- }- case IR::Opcode::A64GetNZCVRaw: {- do_get(nzcv_info, inst, TrackingType::NZCVRaw);- break;- }- case IR::Opcode::A64SetW: {- const size_t index = A64::RegNumber(inst->GetArg(0).GetA64RegRef());- do_set(reg_info.at(index), inst->GetArg(1), inst, TrackingType::W);- break;- }- case IR::Opcode::A64SetX: {- const size_t index = A64::RegNumber(inst->GetArg(0).GetA64RegRef());- do_set(reg_info.at(index), inst->GetArg(1), inst, TrackingType::X);- break;- }- case IR::Opcode::A64SetS: {- const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());- do_set(vec_info.at(index), inst->GetArg(1), inst, TrackingType::S);- break;- }- case IR::Opcode::A64SetD: {- const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());- do_set(vec_info.at(index), inst->GetArg(1), inst, TrackingType::D);- break;- }- case IR::Opcode::A64SetQ: {- const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());- do_set(vec_info.at(index), inst->GetArg(1), inst, TrackingType::Q);- break;- }- case IR::Opcode::A64SetSP: {- do_set(sp_info, inst->GetArg(0), inst, TrackingType::SP);- break;- }- case IR::Opcode::A64SetNZCV: {- do_set(nzcv_info, inst->GetArg(0), inst, TrackingType::NZCV);- break;- }- case IR::Opcode::A64SetNZCVRaw: {- do_set(nzcv_info, inst->GetArg(0), inst, TrackingType::NZCVRaw);- break;- }- default: {- if (ReadsFromCPSR(opcode) || WritesToCPSR(opcode)) {- nzcv_info = {};- }- if (ReadsFromCoreRegister(opcode) || WritesToCoreRegister(opcode)) {- reg_info = {};- vec_info = {};- sp_info = {};- }- break;- }- }- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/a64_merge_interpret_blocks.cpp b/src/dynarmic/src/dynarmic/ir/opt/a64_merge_interpret_blocks.cppdeleted file mode 100644index 25b7ef0ff1..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/a64_merge_interpret_blocks.cpp+++ /dev/null@@ -1,57 +0,0 @@-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project-// SPDX-License-Identifier: GPL-3.0-or-later--/* This file is part of the dynarmic project.- * Copyright (c) 2018 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include <boost/variant/get.hpp>-#include "dynarmic/common/common_types.h"--#include "dynarmic/frontend/A64/a64_location_descriptor.h"-#include "dynarmic/frontend/A64/translate/a64_translate.h"-#include "dynarmic/interface/A64/config.h"-#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opt/passes.h"--namespace Dynarmic::Optimization {--void A64MergeInterpretBlocksPass(IR::Block& block, A64::UserCallbacks* cb) {- const auto is_interpret_instruction = [cb](A64::LocationDescriptor location) {- const auto instruction = cb->MemoryReadCode(location.PC());- if (!instruction)- return false;-- IR::Block new_block{location};- A64::TranslateSingleInstruction(new_block, location, *instruction);-- if (!new_block.Instructions().empty())- return false;-- const IR::Terminal terminal = new_block.GetTerminal();- if (auto term = boost::get<IR::Term::Interpret>(&terminal)) {- return term->next == location;- }-- return false;- };-- IR::Terminal terminal = block.GetTerminal();- auto term = boost::get<IR::Term::Interpret>(&terminal);- if (!term)- return;-- A64::LocationDescriptor location{term->next};- size_t num_instructions = 1;-- while (is_interpret_instruction(location.AdvancePC(static_cast<int>(num_instructions * 4)))) {- num_instructions++;- }-- term->num_instructions = num_instructions;- block.ReplaceTerminal(terminal);- block.CycleCount() += num_instructions - 1;-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/constant_propagation_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/constant_propagation_pass.cppdeleted file mode 100644index 86ebca87d2..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/constant_propagation_pass.cpp+++ /dev/null@@ -1,559 +0,0 @@-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project-// SPDX-License-Identifier: GPL-3.0-or-later--/* This file is part of the dynarmic project.- * Copyright (c) 2016 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include <optional>--#include "dynarmic/common/assert.h"-#include <mcl/bit/rotate.hpp>-#include <mcl/bit/swap.hpp>-#include "dynarmic/common/common_types.h"--#include "dynarmic/common/safe_ops.h"-#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/ir_emitter.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"--namespace Dynarmic::Optimization {--using Op = Dynarmic::IR::Opcode;--namespace {--// Tiny helper to avoid the need to store based off the opcode-// bit size all over the place within folding functions.-void ReplaceUsesWith(IR::Inst& inst, bool is_32_bit, u64 value) {- if (is_32_bit) {- inst.ReplaceUsesWith(IR::Value{static_cast<u32>(value)});- } else {- inst.ReplaceUsesWith(IR::Value{value});- }-}--IR::Value Value(bool is_32_bit, u64 value) {- return is_32_bit ? IR::Value{static_cast<u32>(value)} : IR::Value{value};-}--template<typename ImmFn>-bool FoldCommutative(IR::Inst& inst, bool is_32_bit, ImmFn imm_fn) {- const auto lhs = inst.GetArg(0);- const auto rhs = inst.GetArg(1);-- const bool is_lhs_immediate = lhs.IsImmediate();- const bool is_rhs_immediate = rhs.IsImmediate();-- if (is_lhs_immediate && is_rhs_immediate) {- const u64 result = imm_fn(lhs.GetImmediateAsU64(), rhs.GetImmediateAsU64());- ReplaceUsesWith(inst, is_32_bit, result);- return false;- }-- if (is_lhs_immediate && !is_rhs_immediate) {- const IR::Inst* rhs_inst = rhs.GetInstRecursive();- if (rhs_inst->GetOpcode() == inst.GetOpcode() && rhs_inst->GetArg(1).IsImmediate()) {- const u64 combined = imm_fn(lhs.GetImmediateAsU64(), rhs_inst->GetArg(1).GetImmediateAsU64());- inst.SetArg(0, rhs_inst->GetArg(0));- inst.SetArg(1, Value(is_32_bit, combined));- } else {- // Normalize- inst.SetArg(0, rhs);- inst.SetArg(1, lhs);- }- }-- if (!is_lhs_immediate && is_rhs_immediate) {- const IR::Inst* lhs_inst = lhs.GetInstRecursive();- if (lhs_inst->GetOpcode() == inst.GetOpcode() && lhs_inst->GetArg(1).IsImmediate()) {- const u64 combined = imm_fn(rhs.GetImmediateAsU64(), lhs_inst->GetArg(1).GetImmediateAsU64());- inst.SetArg(0, lhs_inst->GetArg(0));- inst.SetArg(1, Value(is_32_bit, combined));- }- }-- return true;-}--void FoldAdd(IR::Inst& inst, bool is_32_bit) {- const auto lhs = inst.GetArg(0);- const auto rhs = inst.GetArg(1);- const auto carry = inst.GetArg(2);-- if (lhs.IsImmediate() && !rhs.IsImmediate()) {- // Normalize- inst.SetArg(0, rhs);- inst.SetArg(1, lhs);- FoldAdd(inst, is_32_bit);- return;- }-- if (inst.HasAssociatedPseudoOperation()) {- return;- }-- if (!lhs.IsImmediate() && rhs.IsImmediate()) {- const IR::Inst* lhs_inst = lhs.GetInstRecursive();- if (lhs_inst->GetOpcode() == inst.GetOpcode() && lhs_inst->GetArg(1).IsImmediate() && lhs_inst->GetArg(2).IsImmediate()) {- const u64 combined = rhs.GetImmediateAsU64() + lhs_inst->GetArg(1).GetImmediateAsU64() + lhs_inst->GetArg(2).GetU1();- if (combined == 0) {- inst.ReplaceUsesWith(lhs_inst->GetArg(0));- return;- }- inst.SetArg(0, lhs_inst->GetArg(0));- inst.SetArg(1, Value(is_32_bit, combined));- return;- }- if (rhs.IsZero() && carry.IsZero()) {- inst.ReplaceUsesWith(lhs);- return;- }- }-- if (inst.AreAllArgsImmediates()) {- const u64 result = lhs.GetImmediateAsU64() + rhs.GetImmediateAsU64() + carry.GetU1();- ReplaceUsesWith(inst, is_32_bit, result);- return;- }-}--/// Folds AND operations based on the following:-///-/// 1. imm_x & imm_y -> result-/// 2. x & 0 -> 0-/// 3. 0 & y -> 0-/// 4. x & y -> y (where x has all bits set to 1)-/// 5. x & y -> x (where y has all bits set to 1)-///-void FoldAND(IR::Inst& inst, bool is_32_bit) {- if (FoldCommutative(inst, is_32_bit, [](u64 a, u64 b) { return a & b; })) {- const auto rhs = inst.GetArg(1);- if (rhs.IsZero()) {- ReplaceUsesWith(inst, is_32_bit, 0);- } else if (rhs.HasAllBitsSet()) {- inst.ReplaceUsesWith(inst.GetArg(0));- }- }-}--/// Folds byte reversal opcodes based on the following:-///-/// 1. imm -> swap(imm)-///-void FoldByteReverse(IR::Inst& inst, Op op) {- const auto operand = inst.GetArg(0);-- if (!operand.IsImmediate()) {- return;- }-- if (op == Op::ByteReverseWord) {- const u32 result = mcl::bit::swap_bytes_32(static_cast<u32>(operand.GetImmediateAsU64()));- inst.ReplaceUsesWith(IR::Value{result});- } else if (op == Op::ByteReverseHalf) {- const u16 result = mcl::bit::swap_bytes_16(static_cast<u16>(operand.GetImmediateAsU64()));- inst.ReplaceUsesWith(IR::Value{result});- } else {- const u64 result = mcl::bit::swap_bytes_64(operand.GetImmediateAsU64());- inst.ReplaceUsesWith(IR::Value{result});- }-}--/// Folds division operations based on the following:-///-/// 1. x / 0 -> 0 (NOTE: This is an ARM-specific behavior defined in the architecture reference manual)-/// 2. imm_x / imm_y -> result-/// 3. x / 1 -> x-///-void FoldDivide(IR::Inst& inst, bool is_32_bit, bool is_signed) {- const auto rhs = inst.GetArg(1);-- if (rhs.IsZero()) {- ReplaceUsesWith(inst, is_32_bit, 0);- return;- }-- const auto lhs = inst.GetArg(0);- if (lhs.IsImmediate() && rhs.IsImmediate()) {- if (is_signed) {- const s64 result = lhs.GetImmediateAsS64() / rhs.GetImmediateAsS64();- ReplaceUsesWith(inst, is_32_bit, static_cast<u64>(result));- } else {- const u64 result = lhs.GetImmediateAsU64() / rhs.GetImmediateAsU64();- ReplaceUsesWith(inst, is_32_bit, result);- }- } else if (rhs.IsUnsignedImmediate(1)) {- inst.ReplaceUsesWith(IR::Value{lhs});- }-}--// Folds EOR operations based on the following:-//-// 1. imm_x ^ imm_y -> result-// 2. x ^ 0 -> x-// 3. 0 ^ y -> y-//-void FoldEOR(IR::Inst& inst, bool is_32_bit) {- if (FoldCommutative(inst, is_32_bit, [](u64 a, u64 b) { return a ^ b; })) {- const auto rhs = inst.GetArg(1);- if (rhs.IsZero()) {- inst.ReplaceUsesWith(inst.GetArg(0));- }- }-}--void FoldLeastSignificantByte(IR::Inst& inst) {- if (!inst.AreAllArgsImmediates()) {- return;- }-- const auto operand = inst.GetArg(0);- inst.ReplaceUsesWith(IR::Value{static_cast<u8>(operand.GetImmediateAsU64())});-}--void FoldLeastSignificantHalf(IR::Inst& inst) {- if (!inst.AreAllArgsImmediates()) {- return;- }-- const auto operand = inst.GetArg(0);- inst.ReplaceUsesWith(IR::Value{static_cast<u16>(operand.GetImmediateAsU64())});-}--void FoldLeastSignificantWord(IR::Inst& inst) {- if (!inst.AreAllArgsImmediates()) {- return;- }-- const auto operand = inst.GetArg(0);- inst.ReplaceUsesWith(IR::Value{static_cast<u32>(operand.GetImmediateAsU64())});-}--void FoldMostSignificantBit(IR::Inst& inst) {- if (!inst.AreAllArgsImmediates()) {- return;- }-- const auto operand = inst.GetArg(0);- inst.ReplaceUsesWith(IR::Value{(operand.GetImmediateAsU64() >> 31) != 0});-}--void FoldMostSignificantWord(IR::Inst& inst) {- IR::Inst* carry_inst = inst.GetAssociatedPseudoOperation(Op::GetCarryFromOp);-- if (!inst.AreAllArgsImmediates()) {- return;- }-- const auto operand = inst.GetArg(0);- if (carry_inst) {- carry_inst->ReplaceUsesWith(IR::Value{mcl::bit::get_bit<31>(operand.GetImmediateAsU64())});- }- inst.ReplaceUsesWith(IR::Value{static_cast<u32>(operand.GetImmediateAsU64() >> 32)});-}--// Folds multiplication operations based on the following:-//-// 1. imm_x * imm_y -> result-// 2. x * 0 -> 0-// 3. 0 * y -> 0-// 4. x * 1 -> x-// 5. 1 * y -> y-//-void FoldMultiply(IR::Inst& inst, bool is_32_bit) {- if (FoldCommutative(inst, is_32_bit, [](u64 a, u64 b) { return a * b; })) {- const auto rhs = inst.GetArg(1);- if (rhs.IsZero()) {- ReplaceUsesWith(inst, is_32_bit, 0);- } else if (rhs.IsUnsignedImmediate(1)) {- inst.ReplaceUsesWith(inst.GetArg(0));- }- }-}--// Folds NOT operations if the contained value is an immediate.-void FoldNOT(IR::Inst& inst, bool is_32_bit) {- const auto operand = inst.GetArg(0);-- if (!operand.IsImmediate()) {- return;- }-- const u64 result = ~operand.GetImmediateAsU64();- ReplaceUsesWith(inst, is_32_bit, result);-}--// Folds OR operations based on the following:-//-// 1. imm_x | imm_y -> result-// 2. x | 0 -> x-// 3. 0 | y -> y-//-void FoldOR(IR::Inst& inst, bool is_32_bit) {- if (FoldCommutative(inst, is_32_bit, [](u64 a, u64 b) { return a | b; })) {- const auto rhs = inst.GetArg(1);- if (rhs.IsZero()) {- inst.ReplaceUsesWith(inst.GetArg(0));- }- }-}--bool FoldShifts(IR::Inst& inst) {- IR::Inst* carry_inst = inst.GetAssociatedPseudoOperation(Op::GetCarryFromOp);-- // The 32-bit variants can contain 3 arguments, while the- // 64-bit variants only contain 2.- if (inst.NumArgs() == 3 && !carry_inst) {- inst.SetArg(2, IR::Value(false));- }-- const auto shift_amount = inst.GetArg(1);-- if (shift_amount.IsZero()) {- if (carry_inst) {- carry_inst->ReplaceUsesWith(inst.GetArg(2));- }- inst.ReplaceUsesWith(inst.GetArg(0));- return false;- }-- if (inst.NumArgs() == 3 && shift_amount.IsImmediate() && !shift_amount.IsZero()) {- inst.SetArg(2, IR::Value(false));- }-- if (!inst.AreAllArgsImmediates() || carry_inst) {- return false;- }-- return true;-}--void FoldSignExtendXToWord(IR::Inst& inst) {- if (!inst.AreAllArgsImmediates()) {- return;- }-- const s64 value = inst.GetArg(0).GetImmediateAsS64();- inst.ReplaceUsesWith(IR::Value{static_cast<u32>(value)});-}--void FoldSignExtendXToLong(IR::Inst& inst) {- if (!inst.AreAllArgsImmediates()) {- return;- }-- const s64 value = inst.GetArg(0).GetImmediateAsS64();- inst.ReplaceUsesWith(IR::Value{static_cast<u64>(value)});-}--void FoldSub(IR::Inst& inst, bool is_32_bit) {- if (!inst.AreAllArgsImmediates() || inst.HasAssociatedPseudoOperation()) {- return;- }-- const auto lhs = inst.GetArg(0);- const auto rhs = inst.GetArg(1);- const auto carry = inst.GetArg(2);-- const u64 result = lhs.GetImmediateAsU64() + (~rhs.GetImmediateAsU64()) + carry.GetU1();- ReplaceUsesWith(inst, is_32_bit, result);-}--void FoldZeroExtendXToWord(IR::Inst& inst) {- if (!inst.AreAllArgsImmediates()) {- return;- }-- const u64 value = inst.GetArg(0).GetImmediateAsU64();- inst.ReplaceUsesWith(IR::Value{static_cast<u32>(value)});-}--void FoldZeroExtendXToLong(IR::Inst& inst) {- if (!inst.AreAllArgsImmediates()) {- return;- }-- const u64 value = inst.GetArg(0).GetImmediateAsU64();- inst.ReplaceUsesWith(IR::Value{value});-}-} // Anonymous namespace--void ConstantPropagation(IR::Block& block) {- for (auto& inst : block) {- const auto opcode = inst.GetOpcode();-- switch (opcode) {- case Op::LeastSignificantWord:- FoldLeastSignificantWord(inst);- break;- case Op::MostSignificantWord:- FoldMostSignificantWord(inst);- break;- case Op::LeastSignificantHalf:- FoldLeastSignificantHalf(inst);- break;- case Op::LeastSignificantByte:- FoldLeastSignificantByte(inst);- break;- case Op::MostSignificantBit:- FoldMostSignificantBit(inst);- break;- case Op::IsZero32:- if (inst.AreAllArgsImmediates()) {- inst.ReplaceUsesWith(IR::Value{inst.GetArg(0).GetU32() == 0});- }- break;- case Op::IsZero64:- if (inst.AreAllArgsImmediates()) {- inst.ReplaceUsesWith(IR::Value{inst.GetArg(0).GetU64() == 0});- }- break;- case Op::LogicalShiftLeft32:- if (FoldShifts(inst)) {- ReplaceUsesWith(inst, true, Safe::LogicalShiftLeft<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU8()));- }- break;- case Op::LogicalShiftLeft64:- if (FoldShifts(inst)) {- ReplaceUsesWith(inst, false, Safe::LogicalShiftLeft<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU8()));- }- break;- case Op::LogicalShiftRight32:- if (FoldShifts(inst)) {- ReplaceUsesWith(inst, true, Safe::LogicalShiftRight<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU8()));- }- break;- case Op::LogicalShiftRight64:- if (FoldShifts(inst)) {- ReplaceUsesWith(inst, false, Safe::LogicalShiftRight<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU8()));- }- break;- case Op::ArithmeticShiftRight32:- if (FoldShifts(inst)) {- ReplaceUsesWith(inst, true, Safe::ArithmeticShiftRight<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU8()));- }- break;- case Op::ArithmeticShiftRight64:- if (FoldShifts(inst)) {- ReplaceUsesWith(inst, false, Safe::ArithmeticShiftRight<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU8()));- }- break;- case Op::RotateRight32:- if (FoldShifts(inst)) {- ReplaceUsesWith(inst, true, mcl::bit::rotate_right<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU8()));- }- break;- case Op::RotateRight64:- if (FoldShifts(inst)) {- ReplaceUsesWith(inst, false, mcl::bit::rotate_right<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU8()));- }- break;- case Op::LogicalShiftLeftMasked32:- if (inst.AreAllArgsImmediates()) {- ReplaceUsesWith(inst, true, inst.GetArg(0).GetU32() << (inst.GetArg(1).GetU32() & 0x1f));- }- break;- case Op::LogicalShiftLeftMasked64:- if (inst.AreAllArgsImmediates()) {- ReplaceUsesWith(inst, false, inst.GetArg(0).GetU64() << (inst.GetArg(1).GetU64() & 0x3f));- }- break;- case Op::LogicalShiftRightMasked32:- if (inst.AreAllArgsImmediates()) {- ReplaceUsesWith(inst, true, inst.GetArg(0).GetU32() >> (inst.GetArg(1).GetU32() & 0x1f));- }- break;- case Op::LogicalShiftRightMasked64:- if (inst.AreAllArgsImmediates()) {- ReplaceUsesWith(inst, false, inst.GetArg(0).GetU64() >> (inst.GetArg(1).GetU64() & 0x3f));- }- break;- case Op::ArithmeticShiftRightMasked32:- if (inst.AreAllArgsImmediates()) {- ReplaceUsesWith(inst, true, static_cast<s32>(inst.GetArg(0).GetU32()) >> (inst.GetArg(1).GetU32() & 0x1f));- }- break;- case Op::ArithmeticShiftRightMasked64:- if (inst.AreAllArgsImmediates()) {- ReplaceUsesWith(inst, false, static_cast<s64>(inst.GetArg(0).GetU64()) >> (inst.GetArg(1).GetU64() & 0x3f));- }- break;- case Op::RotateRightMasked32:- if (inst.AreAllArgsImmediates()) {- ReplaceUsesWith(inst, true, mcl::bit::rotate_right<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU32()));- }- break;- case Op::RotateRightMasked64:- if (inst.AreAllArgsImmediates()) {- ReplaceUsesWith(inst, false, mcl::bit::rotate_right<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU64()));- }- break;- case Op::Add32:- case Op::Add64:- FoldAdd(inst, opcode == Op::Add32);- break;- case Op::Sub32:- case Op::Sub64:- FoldSub(inst, opcode == Op::Sub32);- break;- case Op::Mul32:- case Op::Mul64:- FoldMultiply(inst, opcode == Op::Mul32);- break;- case Op::SignedDiv32:- case Op::SignedDiv64:- FoldDivide(inst, opcode == Op::SignedDiv32, true);- break;- case Op::UnsignedDiv32:- case Op::UnsignedDiv64:- FoldDivide(inst, opcode == Op::UnsignedDiv32, false);- break;- case Op::And32:- case Op::And64:- FoldAND(inst, opcode == Op::And32);- break;- case Op::Eor32:- case Op::Eor64:- FoldEOR(inst, opcode == Op::Eor32);- break;- case Op::Or32:- case Op::Or64:- FoldOR(inst, opcode == Op::Or32);- break;- case Op::Not32:- case Op::Not64:- FoldNOT(inst, opcode == Op::Not32);- break;- case Op::SignExtendByteToWord:- case Op::SignExtendHalfToWord:- FoldSignExtendXToWord(inst);- break;- case Op::SignExtendByteToLong:- case Op::SignExtendHalfToLong:- case Op::SignExtendWordToLong:- FoldSignExtendXToLong(inst);- break;- case Op::ZeroExtendByteToWord:- case Op::ZeroExtendHalfToWord:- FoldZeroExtendXToWord(inst);- break;- case Op::ZeroExtendByteToLong:- case Op::ZeroExtendHalfToLong:- case Op::ZeroExtendWordToLong:- FoldZeroExtendXToLong(inst);- break;- case Op::ByteReverseWord:- case Op::ByteReverseHalf:- case Op::ByteReverseDual:- FoldByteReverse(inst, opcode);- break;- default:- break;- }- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/dead_code_elimination_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/dead_code_elimination_pass.cppdeleted file mode 100644index bda9f6efd1..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/dead_code_elimination_pass.cpp+++ /dev/null@@ -1,23 +0,0 @@-/* This file is part of the dynarmic project.- * Copyright (c) 2016 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include <mcl/iterator/reverse.hpp>--#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opt/passes.h"--namespace Dynarmic::Optimization {--void DeadCodeElimination(IR::Block& block) {- // We iterate over the instructions in reverse order.- // This is because removing an instruction reduces the number of uses for earlier instructions.- for (auto& inst : mcl::iterator::reverse(block)) {- if (!inst.HasUses() && !MayHaveSideEffects(inst.GetOpcode())) {- inst.Invalidate();- }- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/identity_removal_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/identity_removal_pass.cppdeleted file mode 100644index e87fcc335b..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/identity_removal_pass.cpp+++ /dev/null@@ -1,44 +0,0 @@-/* This file is part of the dynarmic project.- * Copyright (c) 2020 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include <vector>--#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"--namespace Dynarmic::Optimization {--void IdentityRemovalPass(IR::Block& block) {- std::vector<IR::Inst*> to_invalidate;-- auto iter = block.begin();- while (iter != block.end()) {- IR::Inst& inst = *iter;-- const size_t num_args = inst.NumArgs();- for (size_t i = 0; i < num_args; i++) {- while (true) {- IR::Value arg = inst.GetArg(i);- if (!arg.IsIdentity())- break;- inst.SetArg(i, arg.GetInst()->GetArg(0));- }- }-- if (inst.GetOpcode() == IR::Opcode::Identity || inst.GetOpcode() == IR::Opcode::Void) {- iter = block.Instructions().erase(inst);- to_invalidate.push_back(&inst);- } else {- ++iter;- }- }-- for (IR::Inst* inst : to_invalidate) {- inst->Invalidate();- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/ir_matcher.h b/src/dynarmic/src/dynarmic/ir/opt/ir_matcher.hdeleted file mode 100644index 5eb1a55100..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/ir_matcher.h+++ /dev/null@@ -1,127 +0,0 @@-/* This file is part of the dynarmic project.- * Copyright (c) 2020 MerryMage- * SPDX-License-Identifier: 0BSD- */--#pragma once--#include <optional>-#include <tuple>--#include <mp/metafunction/apply.h>-#include <mp/typelist/concat.h>-#include <mp/typelist/drop.h>-#include <mp/typelist/get.h>-#include <mp/typelist/head.h>-#include <mp/typelist/list.h>-#include <mp/typelist/prepend.h>--#include "dynarmic/ir/microinstruction.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/value.h"--namespace Dynarmic::Optimization::IRMatcher {--struct CaptureValue {- using ReturnType = std::tuple<IR::Value>;-- static std::optional<ReturnType> Match(IR::Value value) {- return std::tuple(value);- }-};--struct CaptureInst {- using ReturnType = std::tuple<IR::Inst*>;-- static std::optional<ReturnType> Match(IR::Value value) {- if (value.IsImmediate())- return std::nullopt;- return std::tuple(value.GetInstRecursive());- }-};--struct CaptureUImm {- using ReturnType = std::tuple<u64>;-- static std::optional<ReturnType> Match(IR::Value value) {- return std::tuple(value.GetImmediateAsU64());- }-};--struct CaptureSImm {- using ReturnType = std::tuple<s64>;-- static std::optional<ReturnType> Match(IR::Value value) {- return std::tuple(value.GetImmediateAsS64());- }-};--template<u64 Value>-struct UImm {- using ReturnType = std::tuple<>;-- static std::optional<std::tuple<>> Match(IR::Value value) {- if (value.GetImmediateAsU64() == Value)- return std::tuple();- return std::nullopt;- }-};--template<s64 Value>-struct SImm {- using ReturnType = std::tuple<>;-- static std::optional<std::tuple<>> Match(IR::Value value) {- if (value.GetImmediateAsS64() == Value)- return std::tuple();- return std::nullopt;- }-};--template<IR::Opcode Opcode, typename... Args>-struct Inst {-public:- using ReturnType = mp::concat<std::tuple<>, typename Args::ReturnType...>;-- static std::optional<ReturnType> Match(const IR::Inst& inst) {- if (inst.GetOpcode() != Opcode)- return std::nullopt;- if (inst.HasAssociatedPseudoOperation())- return std::nullopt;- return MatchArgs<0>(inst);- }-- static std::optional<ReturnType> Match(IR::Value value) {- if (value.IsImmediate())- return std::nullopt;- return Match(*value.GetInstRecursive());- }--private:- template<size_t I>- static auto MatchArgs(const IR::Inst& inst) -> std::optional<mp::apply<mp::concat, mp::prepend<mp::drop<I, mp::list<typename Args::ReturnType...>>, std::tuple<>>>> {- if constexpr (I >= sizeof...(Args)) {- return std::tuple();- } else {- using Arg = mp::get<I, mp::list<Args...>>;-- if (const auto arg = Arg::Match(inst.GetArg(I))) {- if (const auto rest = MatchArgs<I + 1>(inst)) {- return std::tuple_cat(*arg, *rest);- }- }-- return std::nullopt;- }- }-};--inline bool IsSameInst(std::tuple<IR::Inst*, IR::Inst*> t) {- return std::get<0>(t) == std::get<1>(t);-}--inline bool IsSameInst(std::tuple<IR::Inst*, IR::Inst*, IR::Inst*> t) {- return std::get<0>(t) == std::get<1>(t) && std::get<0>(t) == std::get<2>(t);-}--} // namespace Dynarmic::Optimization::IRMatcherdiff --git a/src/dynarmic/src/dynarmic/ir/opt/naming_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/naming_pass.cppdeleted file mode 100644index a766bdc83f..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/naming_pass.cpp+++ /dev/null@@ -1,18 +0,0 @@-/* This file is part of the dynarmic project.- * Copyright (c) 2023 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/microinstruction.h"--namespace Dynarmic::Optimization {--void NamingPass(IR::Block& block) {- unsigned name = 1;- for (auto& inst : block) {- inst.SetName(name++);- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/passes.h b/src/dynarmic/src/dynarmic/ir/opt/passes.hdeleted file mode 100644index 703145b556..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/passes.h+++ /dev/null@@ -1,47 +0,0 @@-/* This file is part of the dynarmic project.- * Copyright (c) 2016 MerryMage- * SPDX-License-Identifier: 0BSD- */--#pragma once--namespace Dynarmic::A32 {-struct UserCallbacks;-}--namespace Dynarmic::A64 {-struct UserCallbacks;-struct UserConfig;-} // namespace Dynarmic::A64--namespace Dynarmic::IR {-class Block;-}--namespace Dynarmic::Optimization {--struct PolyfillOptions {- bool sha256 = false;- bool vector_multiply_widen = false;-- bool operator==(const PolyfillOptions&) const = default;-};--struct A32GetSetEliminationOptions {- bool convert_nzc_to_nz = false;- bool convert_nz_to_nzc = false;-};--void PolyfillPass(IR::Block& block, const PolyfillOptions& opt);-void A32ConstantMemoryReads(IR::Block& block, A32::UserCallbacks* cb);-void A32GetSetElimination(IR::Block& block, A32GetSetEliminationOptions opt);-void A64CallbackConfigPass(IR::Block& block, const A64::UserConfig& conf);-void A64GetSetElimination(IR::Block& block);-void A64MergeInterpretBlocksPass(IR::Block& block, A64::UserCallbacks* cb);-void ConstantPropagation(IR::Block& block);-void DeadCodeElimination(IR::Block& block);-void IdentityRemovalPass(IR::Block& block);-void VerificationPass(const IR::Block& block);-void NamingPass(IR::Block& block);--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/polyfill_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/polyfill_pass.cppdeleted file mode 100644index 1aa3aea91e..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/polyfill_pass.cpp+++ /dev/null@@ -1,218 +0,0 @@-/* This file is part of the dynarmic project.- * Copyright (c) 2022 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/ir_emitter.h"-#include "dynarmic/ir/microinstruction.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"--namespace Dynarmic::Optimization {--namespace {--void PolyfillSHA256MessageSchedule0(IR::IREmitter& ir, IR::Inst& inst) {- const IR::U128 x = (IR::U128)inst.GetArg(0);- const IR::U128 y = (IR::U128)inst.GetArg(1);-- const IR::U128 t = ir.VectorExtract(x, y, 32);-- IR::U128 result = ir.ZeroVector();- for (size_t i = 0; i < 4; i++) {- const IR::U32 modified_element = [&] {- const IR::U32 element = ir.VectorGetElement(32, t, i);- const IR::U32 tmp1 = ir.RotateRight(element, ir.Imm8(7));- const IR::U32 tmp2 = ir.RotateRight(element, ir.Imm8(18));- const IR::U32 tmp3 = ir.LogicalShiftRight(element, ir.Imm8(3));-- return ir.Eor(tmp1, ir.Eor(tmp2, tmp3));- }();-- result = ir.VectorSetElement(32, result, i, modified_element);- }- result = ir.VectorAdd(32, result, x);-- inst.ReplaceUsesWith(result);-}--void PolyfillSHA256MessageSchedule1(IR::IREmitter& ir, IR::Inst& inst) {- const IR::U128 x = (IR::U128)inst.GetArg(0);- const IR::U128 y = (IR::U128)inst.GetArg(1);- const IR::U128 z = (IR::U128)inst.GetArg(2);-- const IR::U128 T0 = ir.VectorExtract(y, z, 32);-- const IR::U128 lower_half = [&] {- const IR::U128 T = ir.VectorRotateWholeVectorRight(z, 64);- const IR::U128 tmp1 = ir.VectorRotateRight(32, T, 17);- const IR::U128 tmp2 = ir.VectorRotateRight(32, T, 19);- const IR::U128 tmp3 = ir.VectorLogicalShiftRight(32, T, 10);- const IR::U128 tmp4 = ir.VectorEor(tmp1, ir.VectorEor(tmp2, tmp3));- const IR::U128 tmp5 = ir.VectorAdd(32, tmp4, ir.VectorAdd(32, x, T0));- return ir.VectorZeroUpper(tmp5);- }();-- const IR::U64 upper_half = [&] {- const IR::U128 tmp1 = ir.VectorRotateRight(32, lower_half, 17);- const IR::U128 tmp2 = ir.VectorRotateRight(32, lower_half, 19);- const IR::U128 tmp3 = ir.VectorLogicalShiftRight(32, lower_half, 10);- const IR::U128 tmp4 = ir.VectorEor(tmp1, ir.VectorEor(tmp2, tmp3));-- // Shuffle the top two 32-bit elements downwards [3, 2, 1, 0] -> [1, 0, 3, 2]- const IR::U128 shuffled_d = ir.VectorRotateWholeVectorRight(x, 64);- const IR::U128 shuffled_T0 = ir.VectorRotateWholeVectorRight(T0, 64);-- const IR::U128 tmp5 = ir.VectorAdd(32, tmp4, ir.VectorAdd(32, shuffled_d, shuffled_T0));- return ir.VectorGetElement(64, tmp5, 0);- }();-- const IR::U128 result = ir.VectorSetElement(64, lower_half, 1, upper_half);-- inst.ReplaceUsesWith(result);-}--IR::U32 SHAchoose(IR::IREmitter& ir, IR::U32 x, IR::U32 y, IR::U32 z) {- return ir.Eor(ir.And(ir.Eor(y, z), x), z);-}--IR::U32 SHAmajority(IR::IREmitter& ir, IR::U32 x, IR::U32 y, IR::U32 z) {- return ir.Or(ir.And(x, y), ir.And(ir.Or(x, y), z));-}--IR::U32 SHAhashSIGMA0(IR::IREmitter& ir, IR::U32 x) {- const IR::U32 tmp1 = ir.RotateRight(x, ir.Imm8(2));- const IR::U32 tmp2 = ir.RotateRight(x, ir.Imm8(13));- const IR::U32 tmp3 = ir.RotateRight(x, ir.Imm8(22));-- return ir.Eor(tmp1, ir.Eor(tmp2, tmp3));-}--IR::U32 SHAhashSIGMA1(IR::IREmitter& ir, IR::U32 x) {- const IR::U32 tmp1 = ir.RotateRight(x, ir.Imm8(6));- const IR::U32 tmp2 = ir.RotateRight(x, ir.Imm8(11));- const IR::U32 tmp3 = ir.RotateRight(x, ir.Imm8(25));-- return ir.Eor(tmp1, ir.Eor(tmp2, tmp3));-}--void PolyfillSHA256Hash(IR::IREmitter& ir, IR::Inst& inst) {- IR::U128 x = (IR::U128)inst.GetArg(0);- IR::U128 y = (IR::U128)inst.GetArg(1);- const IR::U128 w = (IR::U128)inst.GetArg(2);- const bool part1 = inst.GetArg(3).GetU1();-- for (size_t i = 0; i < 4; i++) {- const IR::U32 low_x = ir.VectorGetElement(32, x, 0);- const IR::U32 after_low_x = ir.VectorGetElement(32, x, 1);- const IR::U32 before_high_x = ir.VectorGetElement(32, x, 2);- const IR::U32 high_x = ir.VectorGetElement(32, x, 3);-- const IR::U32 low_y = ir.VectorGetElement(32, y, 0);- const IR::U32 after_low_y = ir.VectorGetElement(32, y, 1);- const IR::U32 before_high_y = ir.VectorGetElement(32, y, 2);- const IR::U32 high_y = ir.VectorGetElement(32, y, 3);-- const IR::U32 choice = SHAchoose(ir, low_y, after_low_y, before_high_y);- const IR::U32 majority = SHAmajority(ir, low_x, after_low_x, before_high_x);-- const IR::U32 t = [&] {- const IR::U32 w_element = ir.VectorGetElement(32, w, i);- const IR::U32 sig = SHAhashSIGMA1(ir, low_y);-- return ir.Add(high_y, ir.Add(sig, ir.Add(choice, w_element)));- }();-- const IR::U32 new_low_x = ir.Add(t, ir.Add(SHAhashSIGMA0(ir, low_x), majority));- const IR::U32 new_low_y = ir.Add(t, high_x);-- // Shuffle all words left by 1 element: [3, 2, 1, 0] -> [2, 1, 0, 3]- const IR::U128 shuffled_x = ir.VectorRotateWholeVectorRight(x, 96);- const IR::U128 shuffled_y = ir.VectorRotateWholeVectorRight(y, 96);-- x = ir.VectorSetElement(32, shuffled_x, 0, new_low_x);- y = ir.VectorSetElement(32, shuffled_y, 0, new_low_y);- }-- inst.ReplaceUsesWith(part1 ? x : y);-}--template<size_t esize, bool is_signed>-void PolyfillVectorMultiplyWiden(IR::IREmitter& ir, IR::Inst& inst) {- IR::U128 n = (IR::U128)inst.GetArg(0);- IR::U128 m = (IR::U128)inst.GetArg(1);-- const IR::U128 wide_n = is_signed ? ir.VectorSignExtend(esize, n) : ir.VectorZeroExtend(esize, n);- const IR::U128 wide_m = is_signed ? ir.VectorSignExtend(esize, m) : ir.VectorZeroExtend(esize, m);-- const IR::U128 result = ir.VectorMultiply(esize * 2, wide_n, wide_m);-- inst.ReplaceUsesWith(result);-}--} // namespace--void PolyfillPass(IR::Block& block, const PolyfillOptions& polyfill) {- if (polyfill == PolyfillOptions{}) {- return;- }-- IR::IREmitter ir{block};-- for (auto& inst : block) {- ir.SetInsertionPointBefore(&inst);-- switch (inst.GetOpcode()) {- case IR::Opcode::SHA256MessageSchedule0:- if (polyfill.sha256) {- PolyfillSHA256MessageSchedule0(ir, inst);- }- break;- case IR::Opcode::SHA256MessageSchedule1:- if (polyfill.sha256) {- PolyfillSHA256MessageSchedule1(ir, inst);- }- break;- case IR::Opcode::SHA256Hash:- if (polyfill.sha256) {- PolyfillSHA256Hash(ir, inst);- }- break;- case IR::Opcode::VectorMultiplySignedWiden8:- if (polyfill.vector_multiply_widen) {- PolyfillVectorMultiplyWiden<8, true>(ir, inst);- }- break;- case IR::Opcode::VectorMultiplySignedWiden16:- if (polyfill.vector_multiply_widen) {- PolyfillVectorMultiplyWiden<16, true>(ir, inst);- }- break;- case IR::Opcode::VectorMultiplySignedWiden32:- if (polyfill.vector_multiply_widen) {- PolyfillVectorMultiplyWiden<32, true>(ir, inst);- }- break;- case IR::Opcode::VectorMultiplyUnsignedWiden8:- if (polyfill.vector_multiply_widen) {- PolyfillVectorMultiplyWiden<8, false>(ir, inst);- }- break;- case IR::Opcode::VectorMultiplyUnsignedWiden16:- if (polyfill.vector_multiply_widen) {- PolyfillVectorMultiplyWiden<16, false>(ir, inst);- }- break;- case IR::Opcode::VectorMultiplyUnsignedWiden32:- if (polyfill.vector_multiply_widen) {- PolyfillVectorMultiplyWiden<32, false>(ir, inst);- }- break;- default:- break;- }- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt/verification_pass.cpp b/src/dynarmic/src/dynarmic/ir/opt/verification_pass.cppdeleted file mode 100644index c6c2cff231..0000000000--- a/src/dynarmic/src/dynarmic/ir/opt/verification_pass.cpp+++ /dev/null@@ -1,51 +0,0 @@-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project-// SPDX-License-Identifier: GPL-3.0-or-later--/* This file is part of the dynarmic project.- * Copyright (c) 2016 MerryMage- * SPDX-License-Identifier: 0BSD- */--#include <cstdio>-#include <map>--#include "dynarmic/common/assert.h"-#include "dynarmic/common/common_types.h"-#include <ankerl/unordered_dense.h>--#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/microinstruction.h"-#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"-#include "dynarmic/ir/type.h"--namespace Dynarmic::Optimization {--void VerificationPass(const IR::Block& block) {- for (const auto& inst : block) {- for (size_t i = 0; i < inst.NumArgs(); i++) {- const IR::Type t1 = inst.GetArg(i).GetType();- const IR::Type t2 = IR::GetArgTypeOf(inst.GetOpcode(), i);- if (!IR::AreTypesCompatible(t1, t2)) {- std::puts(IR::DumpBlock(block).c_str());- ASSERT_FALSE("above block failed validation");- }- }- }-- ankerl::unordered_dense::map<IR::Inst*, size_t> actual_uses;- for (const auto& inst : block) {- for (size_t i = 0; i < inst.NumArgs(); i++) {- const auto arg = inst.GetArg(i);- if (!arg.IsImmediate()) {- actual_uses[arg.GetInst()]++;- }- }- }-- for (const auto& pair : actual_uses) {- ASSERT(pair.first->UseCount() == pair.second);- }-}--} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt_passes.cpp b/src/dynarmic/src/dynarmic/ir/opt_passes.cppnew file mode 100644index 0000000000..844e29023c--- /dev/null+++ b/src/dynarmic/src/dynarmic/ir/opt_passes.cpp@@ -0,0 +1,1519 @@+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project+// SPDX-License-Identifier: GPL-3.0-or-later++/* This file is part of the dynarmic project.+ * Copyright (c) 2016 MerryMage+ * SPDX-License-Identifier: 0BSD+ */++#include <cstdio>+#include <map>++#include <ankerl/unordered_dense.h>+#include "boost/container/small_vector.hpp"+#include "dynarmic/frontend/A32/a32_ir_emitter.h"+#include "dynarmic/frontend/A32/a32_location_descriptor.h"+#include "dynarmic/frontend/A32/a32_types.h"+#include "dynarmic/frontend/A64/a64_ir_emitter.h"+#include "dynarmic/frontend/A64/a64_location_descriptor.h"+#include "dynarmic/frontend/A64/translate/a64_translate.h"+#include "dynarmic/interface/A32/config.h"+#include "dynarmic/interface/A64/config.h"+#include "dynarmic/interface/optimization_flags.h"+#include "dynarmic/common/safe_ops.h"+#include "dynarmic/ir/basic_block.h"+#include "dynarmic/ir/microinstruction.h"+#include "dynarmic/ir/opcodes.h"+#include "dynarmic/ir/opt_passes.h"+#include "dynarmic/ir/type.h"+#include "mcl/bit/swap.hpp"+#include "mcl/bit/rotate.hpp"++namespace Dynarmic::Optimization {++static void ConstantMemoryReads(IR::Block& block, A32::UserCallbacks* cb) {+ for (auto& inst : block) {+ switch (inst.GetOpcode()) {+ case IR::Opcode::A32ReadMemory8:+ case IR::Opcode::A64ReadMemory8: {+ if (inst.AreAllArgsImmediates()) {+ const u32 vaddr = inst.GetArg(1).GetU32();+ if (cb->IsReadOnlyMemory(vaddr)) {+ const u8 value_from_memory = cb->MemoryRead8(vaddr);+ inst.ReplaceUsesWith(IR::Value{value_from_memory});+ }+ }+ break;+ }+ case IR::Opcode::A32ReadMemory16:+ case IR::Opcode::A64ReadMemory16: {+ if (inst.AreAllArgsImmediates()) {+ const u32 vaddr = inst.GetArg(1).GetU32();+ if (cb->IsReadOnlyMemory(vaddr)) {+ const u16 value_from_memory = cb->MemoryRead16(vaddr);+ inst.ReplaceUsesWith(IR::Value{value_from_memory});+ }+ }+ break;+ }+ case IR::Opcode::A32ReadMemory32:+ case IR::Opcode::A64ReadMemory32: {+ if (inst.AreAllArgsImmediates()) {+ const u32 vaddr = inst.GetArg(1).GetU32();+ if (cb->IsReadOnlyMemory(vaddr)) {+ const u32 value_from_memory = cb->MemoryRead32(vaddr);+ inst.ReplaceUsesWith(IR::Value{value_from_memory});+ }+ }+ break;+ }+ case IR::Opcode::A32ReadMemory64:+ case IR::Opcode::A64ReadMemory64: {+ if (inst.AreAllArgsImmediates()) {+ const u32 vaddr = inst.GetArg(1).GetU32();+ if (cb->IsReadOnlyMemory(vaddr)) {+ const u64 value_from_memory = cb->MemoryRead64(vaddr);+ inst.ReplaceUsesWith(IR::Value{value_from_memory});+ }+ }+ break;+ }+ default:+ break;+ }+ }+}++static void FlagsPass(IR::Block& block) {+ using Iterator = std::reverse_iterator<IR::Block::iterator>;++ struct FlagInfo {+ bool set_not_required = false;+ bool has_value_request = false;+ Iterator value_request = {};+ };+ struct ValuelessFlagInfo {+ bool set_not_required = false;+ };+ ValuelessFlagInfo nzcvq;+ ValuelessFlagInfo nzcv;+ ValuelessFlagInfo nz;+ FlagInfo c_flag;+ FlagInfo ge;++ auto do_set = [&](FlagInfo& info, IR::Value value, Iterator inst) {+ if (info.has_value_request) {+ info.value_request->ReplaceUsesWith(value);+ }+ info.has_value_request = false;++ if (info.set_not_required) {+ inst->Invalidate();+ }+ info.set_not_required = true;+ };++ auto do_set_valueless = [&](ValuelessFlagInfo& info, Iterator inst) {+ if (info.set_not_required) {+ inst->Invalidate();+ }+ info.set_not_required = true;+ };++ auto do_get = [](FlagInfo& info, Iterator inst) {+ if (info.has_value_request) {+ info.value_request->ReplaceUsesWith(IR::Value{&*inst});+ }+ info.has_value_request = true;+ info.value_request = inst;+ };++ A32::IREmitter ir{block, A32::LocationDescriptor{block.Location()}, {}};++ for (auto inst = block.rbegin(); inst != block.rend(); ++inst) {+ auto const opcode = inst->GetOpcode();+ switch (opcode) {+ case IR::Opcode::A32GetCFlag: {+ do_get(c_flag, inst);+ break;+ }+ case IR::Opcode::A32SetCpsrNZCV: {+ if (c_flag.has_value_request) {+ ir.SetInsertionPointBefore(inst.base()); // base is one ahead+ IR::U1 c = ir.GetCFlagFromNZCV(IR::NZCV{inst->GetArg(0)});+ c_flag.value_request->ReplaceUsesWith(c);+ c_flag.has_value_request = false;+ break; // This case will be executed again because of the above+ }++ do_set_valueless(nzcv, inst);++ nz = {.set_not_required = true};+ c_flag = {.set_not_required = true};+ break;+ }+ case IR::Opcode::A32SetCpsrNZCVRaw: {+ if (c_flag.has_value_request) {+ nzcv.set_not_required = false;+ }++ do_set_valueless(nzcv, inst);++ nzcvq = {};+ nz = {.set_not_required = true};+ c_flag = {.set_not_required = true};+ break;+ }+ case IR::Opcode::A32SetCpsrNZCVQ: {+ if (c_flag.has_value_request) {+ nzcvq.set_not_required = false;+ }++ do_set_valueless(nzcvq, inst);++ nzcv = {.set_not_required = true};+ nz = {.set_not_required = true};+ c_flag = {.set_not_required = true};+ break;+ }+ case IR::Opcode::A32SetCpsrNZ: {+ do_set_valueless(nz, inst);++ nzcvq = {};+ nzcv = {};+ break;+ }+ case IR::Opcode::A32SetCpsrNZC: {+ if (c_flag.has_value_request) {+ c_flag.value_request->ReplaceUsesWith(inst->GetArg(1));+ c_flag.has_value_request = false;+ }++ if (!inst->GetArg(1).IsImmediate() && inst->GetArg(1).GetInstRecursive()->GetOpcode() == IR::Opcode::A32GetCFlag) {+ const auto nz_value = inst->GetArg(0);++ inst->Invalidate();++ ir.SetInsertionPointBefore(inst.base());+ ir.SetCpsrNZ(IR::NZCV{nz_value});++ nzcvq = {};+ nzcv = {};+ nz = {.set_not_required = true};+ break;+ }++ if (nz.set_not_required && c_flag.set_not_required) {+ inst->Invalidate();+ } else if (nz.set_not_required) {+ inst->SetArg(0, IR::Value::EmptyNZCVImmediateMarker());+ }+ nz.set_not_required = true;+ c_flag.set_not_required = true;++ nzcv = {};+ nzcvq = {};+ break;+ }+ case IR::Opcode::A32SetGEFlags: {+ do_set(ge, inst->GetArg(0), inst);+ break;+ }+ case IR::Opcode::A32GetGEFlags: {+ do_get(ge, inst);+ break;+ }+ case IR::Opcode::A32SetGEFlagsCompressed: {+ ge = {.set_not_required = true};+ break;+ }+ case IR::Opcode::A32OrQFlag: {+ break;+ }+ default: {+ if (ReadsFromCPSR(opcode) || WritesToCPSR(opcode)) {+ nzcvq = {};+ nzcv = {};+ nz = {};+ c_flag = {};+ ge = {};+ }+ break;+ }+ }+ }+}++static void RegisterPass(IR::Block& block) {+ using Iterator = IR::Block::iterator;++ struct RegInfo {+ IR::Value register_value;+ std::optional<Iterator> last_set_instruction;+ };+ std::array<RegInfo, 15> reg_info;++ const auto do_get = [](RegInfo& info, Iterator get_inst) {+ if (info.register_value.IsEmpty()) {+ info.register_value = IR::Value(&*get_inst);+ return;+ }+ get_inst->ReplaceUsesWith(info.register_value);+ };++ const auto do_set = [](RegInfo& info, IR::Value value, Iterator set_inst) {+ if (info.last_set_instruction) {+ (*info.last_set_instruction)->Invalidate();+ }+ info = {+ .register_value = value,+ .last_set_instruction = set_inst,+ };+ };++ enum class ExtValueType {+ Empty,+ Single,+ Double,+ VectorDouble,+ VectorQuad,+ };+ struct ExtRegInfo {+ ExtValueType value_type = {};+ IR::Value register_value;+ std::optional<Iterator> last_set_instruction;+ };+ std::array<ExtRegInfo, 64> ext_reg_info;++ const auto do_ext_get = [](ExtValueType type, std::initializer_list<std::reference_wrapper<ExtRegInfo>> infos, Iterator get_inst) {+ if (!std::all_of(infos.begin(), infos.end(), [type](const auto& info) { return info.get().value_type == type; })) {+ for (auto& info : infos) {+ info.get() = {+ .value_type = type,+ .register_value = IR::Value(&*get_inst),+ .last_set_instruction = std::nullopt,+ };+ }+ return;+ }+ get_inst->ReplaceUsesWith(std::data(infos)[0].get().register_value);+ };++ const auto do_ext_set = [](ExtValueType type, std::initializer_list<std::reference_wrapper<ExtRegInfo>> infos, IR::Value value, Iterator set_inst) {+ if (std::all_of(infos.begin(), infos.end(), [type](const auto& info) { return info.get().value_type == type; })) {+ if (std::data(infos)[0].get().last_set_instruction) {+ (*std::data(infos)[0].get().last_set_instruction)->Invalidate();+ }+ }+ for (auto& info : infos) {+ info.get() = {+ .value_type = type,+ .register_value = value,+ .last_set_instruction = set_inst,+ };+ }+ };++ // Location and version don't matter here.+ A32::IREmitter ir{block, A32::LocationDescriptor{block.Location()}, {}};++ for (auto inst = block.begin(); inst != block.end(); ++inst) {+ auto const opcode = inst->GetOpcode();+ switch (opcode) {+ case IR::Opcode::A32GetRegister: {+ const A32::Reg reg = inst->GetArg(0).GetA32RegRef();+ ASSERT(reg != A32::Reg::PC);+ const size_t reg_index = size_t(reg);+ do_get(reg_info[reg_index], inst);+ break;+ }+ case IR::Opcode::A32SetRegister: {+ const A32::Reg reg = inst->GetArg(0).GetA32RegRef();+ if (reg == A32::Reg::PC) {+ break;+ }+ const auto reg_index = size_t(reg);+ do_set(reg_info[reg_index], inst->GetArg(1), inst);+ break;+ }+ case IR::Opcode::A32GetExtendedRegister32: {+ const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();+ const size_t reg_index = A32::RegNumber(reg);+ do_ext_get(ExtValueType::Single, {ext_reg_info[reg_index]}, inst);+ break;+ }+ case IR::Opcode::A32SetExtendedRegister32: {+ const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();+ const size_t reg_index = A32::RegNumber(reg);+ do_ext_set(ExtValueType::Single, {ext_reg_info[reg_index]}, inst->GetArg(1), inst);+ break;+ }+ case IR::Opcode::A32GetExtendedRegister64: {+ const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();+ const size_t reg_index = A32::RegNumber(reg);+ do_ext_get(ExtValueType::Double,+ {+ ext_reg_info[reg_index * 2 + 0],+ ext_reg_info[reg_index * 2 + 1],+ },+ inst);+ break;+ }+ case IR::Opcode::A32SetExtendedRegister64: {+ const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();+ const size_t reg_index = A32::RegNumber(reg);+ do_ext_set(ExtValueType::Double,+ {+ ext_reg_info[reg_index * 2 + 0],+ ext_reg_info[reg_index * 2 + 1],+ },+ inst->GetArg(1),+ inst);+ break;+ }+ case IR::Opcode::A32GetVector: {+ const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();+ const size_t reg_index = A32::RegNumber(reg);+ if (A32::IsDoubleExtReg(reg)) {+ do_ext_get(ExtValueType::VectorDouble,+ {+ ext_reg_info[reg_index * 2 + 0],+ ext_reg_info[reg_index * 2 + 1],+ },+ inst);+ } else {+ DEBUG_ASSERT(A32::IsQuadExtReg(reg));+ do_ext_get(ExtValueType::VectorQuad,+ {+ ext_reg_info[reg_index * 4 + 0],+ ext_reg_info[reg_index * 4 + 1],+ ext_reg_info[reg_index * 4 + 2],+ ext_reg_info[reg_index * 4 + 3],+ },+ inst);+ }+ break;+ }+ case IR::Opcode::A32SetVector: {+ const A32::ExtReg reg = inst->GetArg(0).GetA32ExtRegRef();+ const size_t reg_index = A32::RegNumber(reg);+ if (A32::IsDoubleExtReg(reg)) {+ ir.SetInsertionPointAfter(inst);+ const IR::U128 stored_value = ir.VectorZeroUpper(IR::U128{inst->GetArg(1)});+ do_ext_set(ExtValueType::VectorDouble,+ {+ ext_reg_info[reg_index * 2 + 0],+ ext_reg_info[reg_index * 2 + 1],+ },+ stored_value,+ inst);+ } else {+ DEBUG_ASSERT(A32::IsQuadExtReg(reg));+ do_ext_set(ExtValueType::VectorQuad,+ {+ ext_reg_info[reg_index * 4 + 0],+ ext_reg_info[reg_index * 4 + 1],+ ext_reg_info[reg_index * 4 + 2],+ ext_reg_info[reg_index * 4 + 3],+ },+ inst->GetArg(1),+ inst);+ }+ break;+ }+ default: {+ if (ReadsFromCoreRegister(opcode) || WritesToCoreRegister(opcode)) {+ reg_info = {};+ ext_reg_info = {};+ }+ break;+ }+ }+ }+}++struct A32GetSetEliminationOptions {+ bool convert_nzc_to_nz = false;+ bool convert_nz_to_nzc = false;+};++static void A32GetSetElimination(IR::Block& block, A32GetSetEliminationOptions) {+ FlagsPass(block);+ RegisterPass(block);+}++static void A64CallbackConfigPass(IR::Block& block, const A64::UserConfig& conf) {+ if (conf.hook_data_cache_operations) {+ return;+ }++ for (auto& inst : block) {+ if (inst.GetOpcode() != IR::Opcode::A64DataCacheOperationRaised) {+ continue;+ }++ const auto op = static_cast<A64::DataCacheOperation>(inst.GetArg(1).GetU64());+ if (op == A64::DataCacheOperation::ZeroByVA) {+ A64::IREmitter ir{block};+ ir.current_location = A64::LocationDescriptor{IR::LocationDescriptor{inst.GetArg(0).GetU64()}};+ ir.SetInsertionPointBefore(&inst);++ size_t bytes = 4 << static_cast<size_t>(conf.dczid_el0 & 0b1111);+ IR::U64 addr{inst.GetArg(2)};++ const IR::U128 zero_u128 = ir.ZeroExtendToQuad(ir.Imm64(0));+ while (bytes >= 16) {+ ir.WriteMemory128(addr, zero_u128, IR::AccType::DCZVA);+ addr = ir.Add(addr, ir.Imm64(16));+ bytes -= 16;+ }++ while (bytes >= 8) {+ ir.WriteMemory64(addr, ir.Imm64(0), IR::AccType::DCZVA);+ addr = ir.Add(addr, ir.Imm64(8));+ bytes -= 8;+ }++ while (bytes >= 4) {+ ir.WriteMemory32(addr, ir.Imm32(0), IR::AccType::DCZVA);+ addr = ir.Add(addr, ir.Imm64(4));+ bytes -= 4;+ }+ }+ inst.Invalidate();+ }+}++static void A64GetSetElimination(IR::Block& block) {+ using Iterator = IR::Block::iterator;++ enum class TrackingType {+ W,+ X,+ S,+ D,+ Q,+ SP,+ NZCV,+ NZCVRaw,+ };+ struct RegisterInfo {+ IR::Value register_value;+ TrackingType tracking_type;+ bool set_instruction_present = false;+ Iterator last_set_instruction;+ };+ std::array<RegisterInfo, 31> reg_info;+ std::array<RegisterInfo, 32> vec_info;+ RegisterInfo sp_info;+ RegisterInfo nzcv_info;++ const auto do_set = [&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;+ };++ const auto do_get = [](RegisterInfo& info, Iterator get_inst, TrackingType tracking_type) {+ const auto do_nothing = [&] {+ info = {};+ info.register_value = IR::Value(&*get_inst);+ info.tracking_type = tracking_type;+ };++ if (info.register_value.IsEmpty()) {+ do_nothing();+ return;+ }++ if (info.tracking_type == tracking_type) {+ get_inst->ReplaceUsesWith(info.register_value);+ return;+ }++ do_nothing();+ };++ for (auto inst = block.begin(); inst != block.end(); ++inst) {+ auto const opcode = inst->GetOpcode();+ switch (opcode) {+ case IR::Opcode::A64GetW: {+ const size_t index = A64::RegNumber(inst->GetArg(0).GetA64RegRef());+ do_get(reg_info.at(index), inst, TrackingType::W);+ break;+ }+ case IR::Opcode::A64GetX: {+ const size_t index = A64::RegNumber(inst->GetArg(0).GetA64RegRef());+ do_get(reg_info.at(index), inst, TrackingType::X);+ break;+ }+ case IR::Opcode::A64GetS: {+ const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());+ do_get(vec_info.at(index), inst, TrackingType::S);+ break;+ }+ case IR::Opcode::A64GetD: {+ const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());+ do_get(vec_info.at(index), inst, TrackingType::D);+ break;+ }+ case IR::Opcode::A64GetQ: {+ const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());+ do_get(vec_info.at(index), inst, TrackingType::Q);+ break;+ }+ case IR::Opcode::A64GetSP: {+ do_get(sp_info, inst, TrackingType::SP);+ break;+ }+ case IR::Opcode::A64GetNZCVRaw: {+ do_get(nzcv_info, inst, TrackingType::NZCVRaw);+ break;+ }+ case IR::Opcode::A64SetW: {+ const size_t index = A64::RegNumber(inst->GetArg(0).GetA64RegRef());+ do_set(reg_info.at(index), inst->GetArg(1), inst, TrackingType::W);+ break;+ }+ case IR::Opcode::A64SetX: {+ const size_t index = A64::RegNumber(inst->GetArg(0).GetA64RegRef());+ do_set(reg_info.at(index), inst->GetArg(1), inst, TrackingType::X);+ break;+ }+ case IR::Opcode::A64SetS: {+ const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());+ do_set(vec_info.at(index), inst->GetArg(1), inst, TrackingType::S);+ break;+ }+ case IR::Opcode::A64SetD: {+ const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());+ do_set(vec_info.at(index), inst->GetArg(1), inst, TrackingType::D);+ break;+ }+ case IR::Opcode::A64SetQ: {+ const size_t index = A64::VecNumber(inst->GetArg(0).GetA64VecRef());+ do_set(vec_info.at(index), inst->GetArg(1), inst, TrackingType::Q);+ break;+ }+ case IR::Opcode::A64SetSP: {+ do_set(sp_info, inst->GetArg(0), inst, TrackingType::SP);+ break;+ }+ case IR::Opcode::A64SetNZCV: {+ do_set(nzcv_info, inst->GetArg(0), inst, TrackingType::NZCV);+ break;+ }+ case IR::Opcode::A64SetNZCVRaw: {+ do_set(nzcv_info, inst->GetArg(0), inst, TrackingType::NZCVRaw);+ break;+ }+ default: {+ if (ReadsFromCPSR(opcode) || WritesToCPSR(opcode)) {+ nzcv_info = {};+ }+ if (ReadsFromCoreRegister(opcode) || WritesToCoreRegister(opcode)) {+ reg_info = {};+ vec_info = {};+ sp_info = {};+ }+ break;+ }+ }+ }+}++static void A64MergeInterpretBlocksPass(IR::Block& block, A64::UserCallbacks* cb) {+ const auto is_interpret_instruction = [cb](A64::LocationDescriptor location) {+ const auto instruction = cb->MemoryReadCode(location.PC());+ if (!instruction)+ return false;++ IR::Block new_block{location};+ A64::TranslateSingleInstruction(new_block, location, *instruction);++ if (!new_block.Instructions().empty())+ return false;++ const IR::Terminal terminal = new_block.GetTerminal();+ if (auto term = boost::get<IR::Term::Interpret>(&terminal)) {+ return term->next == location;+ }++ return false;+ };++ IR::Terminal terminal = block.GetTerminal();+ auto term = boost::get<IR::Term::Interpret>(&terminal);+ if (!term)+ return;++ A64::LocationDescriptor location{term->next};+ size_t num_instructions = 1;++ while (is_interpret_instruction(location.AdvancePC(static_cast<int>(num_instructions * 4)))) {+ num_instructions++;+ }++ term->num_instructions = num_instructions;+ block.ReplaceTerminal(terminal);+ block.CycleCount() += num_instructions - 1;+}++using Op = Dynarmic::IR::Opcode;++// Tiny helper to avoid the need to store based off the opcode+// bit size all over the place within folding functions.+static void ReplaceUsesWith(IR::Inst& inst, bool is_32_bit, u64 value) {+ if (is_32_bit) {+ inst.ReplaceUsesWith(IR::Value{u32(value)});+ } else {+ inst.ReplaceUsesWith(IR::Value{value});+ }+}++static IR::Value Value(bool is_32_bit, u64 value) {+ return is_32_bit ? IR::Value{u32(value)} : IR::Value{value};+}++template<typename ImmFn>+static bool FoldCommutative(IR::Inst& inst, bool is_32_bit, ImmFn imm_fn) {+ const auto lhs = inst.GetArg(0);+ const auto rhs = inst.GetArg(1);++ const bool is_lhs_immediate = lhs.IsImmediate();+ const bool is_rhs_immediate = rhs.IsImmediate();++ if (is_lhs_immediate && is_rhs_immediate) {+ const u64 result = imm_fn(lhs.GetImmediateAsU64(), rhs.GetImmediateAsU64());+ ReplaceUsesWith(inst, is_32_bit, result);+ return false;+ }++ if (is_lhs_immediate && !is_rhs_immediate) {+ const IR::Inst* rhs_inst = rhs.GetInstRecursive();+ if (rhs_inst->GetOpcode() == inst.GetOpcode() && rhs_inst->GetArg(1).IsImmediate()) {+ const u64 combined = imm_fn(lhs.GetImmediateAsU64(), rhs_inst->GetArg(1).GetImmediateAsU64());+ inst.SetArg(0, rhs_inst->GetArg(0));+ inst.SetArg(1, Value(is_32_bit, combined));+ } else {+ // Normalize+ inst.SetArg(0, rhs);+ inst.SetArg(1, lhs);+ }+ }++ if (!is_lhs_immediate && is_rhs_immediate) {+ const IR::Inst* lhs_inst = lhs.GetInstRecursive();+ if (lhs_inst->GetOpcode() == inst.GetOpcode() && lhs_inst->GetArg(1).IsImmediate()) {+ const u64 combined = imm_fn(rhs.GetImmediateAsU64(), lhs_inst->GetArg(1).GetImmediateAsU64());+ inst.SetArg(0, lhs_inst->GetArg(0));+ inst.SetArg(1, Value(is_32_bit, combined));+ }+ }++ return true;+}++static void FoldAdd(IR::Inst& inst, bool is_32_bit) {+ const auto lhs = inst.GetArg(0);+ const auto rhs = inst.GetArg(1);+ const auto carry = inst.GetArg(2);++ if (lhs.IsImmediate() && !rhs.IsImmediate()) {+ // Normalize+ inst.SetArg(0, rhs);+ inst.SetArg(1, lhs);+ FoldAdd(inst, is_32_bit);+ return;+ }++ if (inst.HasAssociatedPseudoOperation()) {+ return;+ }++ if (!lhs.IsImmediate() && rhs.IsImmediate()) {+ const IR::Inst* lhs_inst = lhs.GetInstRecursive();+ if (lhs_inst->GetOpcode() == inst.GetOpcode() && lhs_inst->GetArg(1).IsImmediate() && lhs_inst->GetArg(2).IsImmediate()) {+ const u64 combined = rhs.GetImmediateAsU64() + lhs_inst->GetArg(1).GetImmediateAsU64() + lhs_inst->GetArg(2).GetU1();+ if (combined == 0) {+ inst.ReplaceUsesWith(lhs_inst->GetArg(0));+ return;+ }+ inst.SetArg(0, lhs_inst->GetArg(0));+ inst.SetArg(1, Value(is_32_bit, combined));+ return;+ }+ if (rhs.IsZero() && carry.IsZero()) {+ inst.ReplaceUsesWith(lhs);+ return;+ }+ }++ if (inst.AreAllArgsImmediates()) {+ const u64 result = lhs.GetImmediateAsU64() + rhs.GetImmediateAsU64() + carry.GetU1();+ ReplaceUsesWith(inst, is_32_bit, result);+ return;+ }+}++/// Folds AND operations based on the following:+///+/// 1. imm_x & imm_y -> result+/// 2. x & 0 -> 0+/// 3. 0 & y -> 0+/// 4. x & y -> y (where x has all bits set to 1)+/// 5. x & y -> x (where y has all bits set to 1)+///+static void FoldAND(IR::Inst& inst, bool is_32_bit) {+ if (FoldCommutative(inst, is_32_bit, [](u64 a, u64 b) { return a & b; })) {+ const auto rhs = inst.GetArg(1);+ if (rhs.IsZero()) {+ ReplaceUsesWith(inst, is_32_bit, 0);+ } else if (rhs.HasAllBitsSet()) {+ inst.ReplaceUsesWith(inst.GetArg(0));+ }+ }+}++/// Folds byte reversal opcodes based on the following:+///+/// 1. imm -> swap(imm)+///+static void FoldByteReverse(IR::Inst& inst, Op op) {+ const auto operand = inst.GetArg(0);++ if (!operand.IsImmediate()) {+ return;+ }++ if (op == Op::ByteReverseWord) {+ const u32 result = mcl::bit::swap_bytes_32(u32(operand.GetImmediateAsU64()));+ inst.ReplaceUsesWith(IR::Value{result});+ } else if (op == Op::ByteReverseHalf) {+ const u16 result = mcl::bit::swap_bytes_16(u16(operand.GetImmediateAsU64()));+ inst.ReplaceUsesWith(IR::Value{result});+ } else {+ const u64 result = mcl::bit::swap_bytes_64(operand.GetImmediateAsU64());+ inst.ReplaceUsesWith(IR::Value{result});+ }+}++/// Folds leading zero population count+///+/// 1. imm -> countl_zero(imm)+///+static void FoldCountLeadingZeros(IR::Inst& inst, bool is_32_bit) {+ const auto operand = inst.GetArg(0);+ if (operand.IsImmediate()) {+ if (is_32_bit) {+ const u32 result = std::countl_zero(u32(operand.GetImmediateAsU64()));+ inst.ReplaceUsesWith(IR::Value{result});+ } else {+ const u64 result = std::countl_zero(operand.GetImmediateAsU64());+ inst.ReplaceUsesWith(IR::Value{result});+ }+ }+}++/// Folds division operations based on the following:+///+/// 1. x / 0 -> 0 (NOTE: This is an ARM-specific behavior defined in the architecture reference manual)+/// 2. imm_x / imm_y -> result+/// 3. x / 1 -> x+///+static void FoldDivide(IR::Inst& inst, bool is_32_bit, bool is_signed) {+ const auto rhs = inst.GetArg(1);++ if (rhs.IsZero()) {+ ReplaceUsesWith(inst, is_32_bit, 0);+ return;+ }++ const auto lhs = inst.GetArg(0);+ if (lhs.IsImmediate() && rhs.IsImmediate()) {+ if (is_signed) {+ const s64 result = lhs.GetImmediateAsS64() / rhs.GetImmediateAsS64();+ ReplaceUsesWith(inst, is_32_bit, static_cast<u64>(result));+ } else {+ const u64 result = lhs.GetImmediateAsU64() / rhs.GetImmediateAsU64();+ ReplaceUsesWith(inst, is_32_bit, result);+ }+ } else if (rhs.IsUnsignedImmediate(1)) {+ inst.ReplaceUsesWith(IR::Value{lhs});+ }+}++// Folds EOR operations based on the following:+//+// 1. imm_x ^ imm_y -> result+// 2. x ^ 0 -> x+// 3. 0 ^ y -> y+//+static void FoldEOR(IR::Inst& inst, bool is_32_bit) {+ if (FoldCommutative(inst, is_32_bit, [](u64 a, u64 b) { return a ^ b; })) {+ const auto rhs = inst.GetArg(1);+ if (rhs.IsZero()) {+ inst.ReplaceUsesWith(inst.GetArg(0));+ }+ }+}++static void FoldLeastSignificantByte(IR::Inst& inst) {+ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const auto operand = inst.GetArg(0);+ inst.ReplaceUsesWith(IR::Value{static_cast<u8>(operand.GetImmediateAsU64())});+}++static void FoldLeastSignificantHalf(IR::Inst& inst) {+ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const auto operand = inst.GetArg(0);+ inst.ReplaceUsesWith(IR::Value{static_cast<u16>(operand.GetImmediateAsU64())});+}++static void FoldLeastSignificantWord(IR::Inst& inst) {+ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const auto operand = inst.GetArg(0);+ inst.ReplaceUsesWith(IR::Value{static_cast<u32>(operand.GetImmediateAsU64())});+}++static void FoldMostSignificantBit(IR::Inst& inst) {+ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const auto operand = inst.GetArg(0);+ inst.ReplaceUsesWith(IR::Value{(operand.GetImmediateAsU64() >> 31) != 0});+}++static void FoldMostSignificantWord(IR::Inst& inst) {+ IR::Inst* carry_inst = inst.GetAssociatedPseudoOperation(Op::GetCarryFromOp);++ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const auto operand = inst.GetArg(0);+ if (carry_inst) {+ carry_inst->ReplaceUsesWith(IR::Value{mcl::bit::get_bit<31>(operand.GetImmediateAsU64())});+ }+ inst.ReplaceUsesWith(IR::Value{static_cast<u32>(operand.GetImmediateAsU64() >> 32)});+}++// Folds multiplication operations based on the following:+//+// 1. imm_x * imm_y -> result+// 2. x * 0 -> 0+// 3. 0 * y -> 0+// 4. x * 1 -> x+// 5. 1 * y -> y+//+static void FoldMultiply(IR::Inst& inst, bool is_32_bit) {+ if (FoldCommutative(inst, is_32_bit, [](u64 a, u64 b) { return a * b; })) {+ const auto rhs = inst.GetArg(1);+ if (rhs.IsZero()) {+ ReplaceUsesWith(inst, is_32_bit, 0);+ } else if (rhs.IsUnsignedImmediate(1)) {+ inst.ReplaceUsesWith(inst.GetArg(0));+ }+ }+}++// Folds NOT operations if the contained value is an immediate.+static void FoldNOT(IR::Inst& inst, bool is_32_bit) {+ const auto operand = inst.GetArg(0);++ if (!operand.IsImmediate()) {+ return;+ }++ const u64 result = ~operand.GetImmediateAsU64();+ ReplaceUsesWith(inst, is_32_bit, result);+}++// Folds OR operations based on the following:+//+// 1. imm_x | imm_y -> result+// 2. x | 0 -> x+// 3. 0 | y -> y+//+static void FoldOR(IR::Inst& inst, bool is_32_bit) {+ if (FoldCommutative(inst, is_32_bit, [](u64 a, u64 b) { return a | b; })) {+ const auto rhs = inst.GetArg(1);+ if (rhs.IsZero()) {+ inst.ReplaceUsesWith(inst.GetArg(0));+ }+ }+}++static bool FoldShifts(IR::Inst& inst) {+ IR::Inst* carry_inst = inst.GetAssociatedPseudoOperation(Op::GetCarryFromOp);++ // The 32-bit variants can contain 3 arguments, while the+ // 64-bit variants only contain 2.+ if (inst.NumArgs() == 3 && !carry_inst) {+ inst.SetArg(2, IR::Value(false));+ }++ const auto shift_amount = inst.GetArg(1);++ if (shift_amount.IsZero()) {+ if (carry_inst) {+ carry_inst->ReplaceUsesWith(inst.GetArg(2));+ }+ inst.ReplaceUsesWith(inst.GetArg(0));+ return false;+ }++ if (inst.NumArgs() == 3 && shift_amount.IsImmediate() && !shift_amount.IsZero()) {+ inst.SetArg(2, IR::Value(false));+ }++ if (!inst.AreAllArgsImmediates() || carry_inst) {+ return false;+ }++ return true;+}++static void FoldSignExtendXToWord(IR::Inst& inst) {+ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const s64 value = inst.GetArg(0).GetImmediateAsS64();+ inst.ReplaceUsesWith(IR::Value{static_cast<u32>(value)});+}++static void FoldSignExtendXToLong(IR::Inst& inst) {+ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const s64 value = inst.GetArg(0).GetImmediateAsS64();+ inst.ReplaceUsesWith(IR::Value{static_cast<u64>(value)});+}++static void FoldSub(IR::Inst& inst, bool is_32_bit) {+ if (!inst.AreAllArgsImmediates() || inst.HasAssociatedPseudoOperation()) {+ return;+ }++ const auto lhs = inst.GetArg(0);+ const auto rhs = inst.GetArg(1);+ const auto carry = inst.GetArg(2);++ const u64 result = lhs.GetImmediateAsU64() + (~rhs.GetImmediateAsU64()) + carry.GetU1();+ ReplaceUsesWith(inst, is_32_bit, result);+}++static void FoldZeroExtendXToWord(IR::Inst& inst) {+ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const u64 value = inst.GetArg(0).GetImmediateAsU64();+ inst.ReplaceUsesWith(IR::Value{static_cast<u32>(value)});+}++static void FoldZeroExtendXToLong(IR::Inst& inst) {+ if (!inst.AreAllArgsImmediates()) {+ return;+ }++ const u64 value = inst.GetArg(0).GetImmediateAsU64();+ inst.ReplaceUsesWith(IR::Value{value});+}++static void ConstantPropagation(IR::Block& block) {+ for (auto& inst : block) {+ auto const opcode = inst.GetOpcode();+ switch (opcode) {+ case Op::LeastSignificantWord:+ FoldLeastSignificantWord(inst);+ break;+ case Op::MostSignificantWord:+ FoldMostSignificantWord(inst);+ break;+ case Op::LeastSignificantHalf:+ FoldLeastSignificantHalf(inst);+ break;+ case Op::LeastSignificantByte:+ FoldLeastSignificantByte(inst);+ break;+ case Op::MostSignificantBit:+ FoldMostSignificantBit(inst);+ break;+ case Op::IsZero32:+ if (inst.AreAllArgsImmediates()) {+ inst.ReplaceUsesWith(IR::Value{inst.GetArg(0).GetU32() == 0});+ }+ break;+ case Op::IsZero64:+ if (inst.AreAllArgsImmediates()) {+ inst.ReplaceUsesWith(IR::Value{inst.GetArg(0).GetU64() == 0});+ }+ break;+ case Op::LogicalShiftLeft32:+ if (FoldShifts(inst)) {+ ReplaceUsesWith(inst, true, Safe::LogicalShiftLeft<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU8()));+ }+ break;+ case Op::LogicalShiftLeft64:+ if (FoldShifts(inst)) {+ ReplaceUsesWith(inst, false, Safe::LogicalShiftLeft<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU8()));+ }+ break;+ case Op::LogicalShiftRight32:+ if (FoldShifts(inst)) {+ ReplaceUsesWith(inst, true, Safe::LogicalShiftRight<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU8()));+ }+ break;+ case Op::LogicalShiftRight64:+ if (FoldShifts(inst)) {+ ReplaceUsesWith(inst, false, Safe::LogicalShiftRight<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU8()));+ }+ break;+ case Op::ArithmeticShiftRight32:+ if (FoldShifts(inst)) {+ ReplaceUsesWith(inst, true, Safe::ArithmeticShiftRight<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU8()));+ }+ break;+ case Op::ArithmeticShiftRight64:+ if (FoldShifts(inst)) {+ ReplaceUsesWith(inst, false, Safe::ArithmeticShiftRight<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU8()));+ }+ break;+ case Op::RotateRight32:+ if (FoldShifts(inst)) {+ ReplaceUsesWith(inst, true, mcl::bit::rotate_right<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU8()));+ }+ break;+ case Op::RotateRight64:+ if (FoldShifts(inst)) {+ ReplaceUsesWith(inst, false, mcl::bit::rotate_right<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU8()));+ }+ break;+ case Op::LogicalShiftLeftMasked32:+ if (inst.AreAllArgsImmediates()) {+ ReplaceUsesWith(inst, true, inst.GetArg(0).GetU32() << (inst.GetArg(1).GetU32() & 0x1f));+ }+ break;+ case Op::LogicalShiftLeftMasked64:+ if (inst.AreAllArgsImmediates()) {+ ReplaceUsesWith(inst, false, inst.GetArg(0).GetU64() << (inst.GetArg(1).GetU64() & 0x3f));+ }+ break;+ case Op::LogicalShiftRightMasked32:+ if (inst.AreAllArgsImmediates()) {+ ReplaceUsesWith(inst, true, inst.GetArg(0).GetU32() >> (inst.GetArg(1).GetU32() & 0x1f));+ }+ break;+ case Op::LogicalShiftRightMasked64:+ if (inst.AreAllArgsImmediates()) {+ ReplaceUsesWith(inst, false, inst.GetArg(0).GetU64() >> (inst.GetArg(1).GetU64() & 0x3f));+ }+ break;+ case Op::ArithmeticShiftRightMasked32:+ if (inst.AreAllArgsImmediates()) {+ ReplaceUsesWith(inst, true, s32(inst.GetArg(0).GetU32()) >> (inst.GetArg(1).GetU32() & 0x1f));+ }+ break;+ case Op::ArithmeticShiftRightMasked64:+ if (inst.AreAllArgsImmediates()) {+ ReplaceUsesWith(inst, false, s64(inst.GetArg(0).GetU64()) >> (inst.GetArg(1).GetU64() & 0x3f));+ }+ break;+ case Op::RotateRightMasked32:+ if (inst.AreAllArgsImmediates()) {+ ReplaceUsesWith(inst, true, mcl::bit::rotate_right<u32>(inst.GetArg(0).GetU32(), inst.GetArg(1).GetU32()));+ }+ break;+ case Op::RotateRightMasked64:+ if (inst.AreAllArgsImmediates()) {+ ReplaceUsesWith(inst, false, mcl::bit::rotate_right<u64>(inst.GetArg(0).GetU64(), inst.GetArg(1).GetU64()));+ }+ break;+ case Op::Add32:+ case Op::Add64:+ FoldAdd(inst, opcode == Op::Add32);+ break;+ case Op::Sub32:+ case Op::Sub64:+ FoldSub(inst, opcode == Op::Sub32);+ break;+ case Op::Mul32:+ case Op::Mul64:+ FoldMultiply(inst, opcode == Op::Mul32);+ break;+ case Op::SignedDiv32:+ case Op::SignedDiv64:+ FoldDivide(inst, opcode == Op::SignedDiv32, true);+ break;+ case Op::UnsignedDiv32:+ case Op::UnsignedDiv64:+ FoldDivide(inst, opcode == Op::UnsignedDiv32, false);+ break;+ case Op::And32:+ case Op::And64:+ FoldAND(inst, opcode == Op::And32);+ break;+ case Op::Eor32:+ case Op::Eor64:+ FoldEOR(inst, opcode == Op::Eor32);+ break;+ case Op::Or32:+ case Op::Or64:+ FoldOR(inst, opcode == Op::Or32);+ break;+ case Op::Not32:+ case Op::Not64:+ FoldNOT(inst, opcode == Op::Not32);+ break;+ case Op::SignExtendByteToWord:+ case Op::SignExtendHalfToWord:+ FoldSignExtendXToWord(inst);+ break;+ case Op::SignExtendByteToLong:+ case Op::SignExtendHalfToLong:+ case Op::SignExtendWordToLong:+ FoldSignExtendXToLong(inst);+ break;+ case Op::ZeroExtendByteToWord:+ case Op::ZeroExtendHalfToWord:+ FoldZeroExtendXToWord(inst);+ break;+ case Op::ZeroExtendByteToLong:+ case Op::ZeroExtendHalfToLong:+ case Op::ZeroExtendWordToLong:+ FoldZeroExtendXToLong(inst);+ break;+ case Op::ByteReverseWord:+ case Op::ByteReverseHalf:+ case Op::ByteReverseDual:+ FoldByteReverse(inst, opcode);+ break;+ case Op::CountLeadingZeros32:+ case Op::CountLeadingZeros64:+ FoldCountLeadingZeros(inst, opcode == Op::CountLeadingZeros32);+ break;+ default:+ break;+ }+ }+}++static void DeadCodeElimination(IR::Block& block) {+ // We iterate over the instructions in reverse order.+ // This is because removing an instruction reduces the number of uses for earlier instructions.+ for (auto it = block.rbegin(); it != block.rend(); ++it)+ if (!it->HasUses() && !MayHaveSideEffects(it->GetOpcode()))+ it->Invalidate();+}++static void IdentityRemovalPass(IR::Block& block) {+ boost::container::small_vector<IR::Inst*, 128> to_invalidate;++ auto iter = block.begin();+ while (iter != block.end()) {+ IR::Inst& inst = *iter;++ const size_t num_args = inst.NumArgs();+ for (size_t i = 0; i < num_args; i++) {+ while (true) {+ IR::Value arg = inst.GetArg(i);+ if (!arg.IsIdentity())+ break;+ inst.SetArg(i, arg.GetInst()->GetArg(0));+ }+ }++ if (inst.GetOpcode() == IR::Opcode::Identity || inst.GetOpcode() == IR::Opcode::Void) {+ iter = block.Instructions().erase(inst);+ to_invalidate.push_back(&inst);+ } else {+ ++iter;+ }+ }+ for (IR::Inst* inst : to_invalidate) {+ inst->Invalidate();+ }+}++static void NamingPass(IR::Block& block) {+ u32 name = 1;+ for (auto& inst : block)+ inst.SetName(name++);+}++static void PolyfillSHA256MessageSchedule0(IR::IREmitter& ir, IR::Inst& inst) {+ const IR::U128 x = (IR::U128)inst.GetArg(0);+ const IR::U128 y = (IR::U128)inst.GetArg(1);++ const IR::U128 t = ir.VectorExtract(x, y, 32);++ IR::U128 result = ir.ZeroVector();+ for (size_t i = 0; i < 4; i++) {+ const IR::U32 modified_element = [&] {+ const IR::U32 element = ir.VectorGetElement(32, t, i);+ const IR::U32 tmp1 = ir.RotateRight(element, ir.Imm8(7));+ const IR::U32 tmp2 = ir.RotateRight(element, ir.Imm8(18));+ const IR::U32 tmp3 = ir.LogicalShiftRight(element, ir.Imm8(3));++ return ir.Eor(tmp1, ir.Eor(tmp2, tmp3));+ }();++ result = ir.VectorSetElement(32, result, i, modified_element);+ }+ result = ir.VectorAdd(32, result, x);++ inst.ReplaceUsesWith(result);+}++static void PolyfillSHA256MessageSchedule1(IR::IREmitter& ir, IR::Inst& inst) {+ const IR::U128 x = (IR::U128)inst.GetArg(0);+ const IR::U128 y = (IR::U128)inst.GetArg(1);+ const IR::U128 z = (IR::U128)inst.GetArg(2);++ const IR::U128 T0 = ir.VectorExtract(y, z, 32);++ const IR::U128 lower_half = [&] {+ const IR::U128 T = ir.VectorRotateWholeVectorRight(z, 64);+ const IR::U128 tmp1 = ir.VectorRotateRight(32, T, 17);+ const IR::U128 tmp2 = ir.VectorRotateRight(32, T, 19);+ const IR::U128 tmp3 = ir.VectorLogicalShiftRight(32, T, 10);+ const IR::U128 tmp4 = ir.VectorEor(tmp1, ir.VectorEor(tmp2, tmp3));+ const IR::U128 tmp5 = ir.VectorAdd(32, tmp4, ir.VectorAdd(32, x, T0));+ return ir.VectorZeroUpper(tmp5);+ }();++ const IR::U64 upper_half = [&] {+ const IR::U128 tmp1 = ir.VectorRotateRight(32, lower_half, 17);+ const IR::U128 tmp2 = ir.VectorRotateRight(32, lower_half, 19);+ const IR::U128 tmp3 = ir.VectorLogicalShiftRight(32, lower_half, 10);+ const IR::U128 tmp4 = ir.VectorEor(tmp1, ir.VectorEor(tmp2, tmp3));++ // Shuffle the top two 32-bit elements downwards [3, 2, 1, 0] -> [1, 0, 3, 2]+ const IR::U128 shuffled_d = ir.VectorRotateWholeVectorRight(x, 64);+ const IR::U128 shuffled_T0 = ir.VectorRotateWholeVectorRight(T0, 64);++ const IR::U128 tmp5 = ir.VectorAdd(32, tmp4, ir.VectorAdd(32, shuffled_d, shuffled_T0));+ return ir.VectorGetElement(64, tmp5, 0);+ }();++ const IR::U128 result = ir.VectorSetElement(64, lower_half, 1, upper_half);++ inst.ReplaceUsesWith(result);+}++static IR::U32 SHAchoose(IR::IREmitter& ir, IR::U32 x, IR::U32 y, IR::U32 z) {+ return ir.Eor(ir.And(ir.Eor(y, z), x), z);+}++static IR::U32 SHAmajority(IR::IREmitter& ir, IR::U32 x, IR::U32 y, IR::U32 z) {+ return ir.Or(ir.And(x, y), ir.And(ir.Or(x, y), z));+}++static IR::U32 SHAhashSIGMA0(IR::IREmitter& ir, IR::U32 x) {+ const IR::U32 tmp1 = ir.RotateRight(x, ir.Imm8(2));+ const IR::U32 tmp2 = ir.RotateRight(x, ir.Imm8(13));+ const IR::U32 tmp3 = ir.RotateRight(x, ir.Imm8(22));++ return ir.Eor(tmp1, ir.Eor(tmp2, tmp3));+}++static IR::U32 SHAhashSIGMA1(IR::IREmitter& ir, IR::U32 x) {+ const IR::U32 tmp1 = ir.RotateRight(x, ir.Imm8(6));+ const IR::U32 tmp2 = ir.RotateRight(x, ir.Imm8(11));+ const IR::U32 tmp3 = ir.RotateRight(x, ir.Imm8(25));++ return ir.Eor(tmp1, ir.Eor(tmp2, tmp3));+}++static void PolyfillSHA256Hash(IR::IREmitter& ir, IR::Inst& inst) {+ IR::U128 x = (IR::U128)inst.GetArg(0);+ IR::U128 y = (IR::U128)inst.GetArg(1);+ const IR::U128 w = (IR::U128)inst.GetArg(2);+ const bool part1 = inst.GetArg(3).GetU1();++ for (size_t i = 0; i < 4; i++) {+ const IR::U32 low_x = ir.VectorGetElement(32, x, 0);+ const IR::U32 after_low_x = ir.VectorGetElement(32, x, 1);+ const IR::U32 before_high_x = ir.VectorGetElement(32, x, 2);+ const IR::U32 high_x = ir.VectorGetElement(32, x, 3);++ const IR::U32 low_y = ir.VectorGetElement(32, y, 0);+ const IR::U32 after_low_y = ir.VectorGetElement(32, y, 1);+ const IR::U32 before_high_y = ir.VectorGetElement(32, y, 2);+ const IR::U32 high_y = ir.VectorGetElement(32, y, 3);++ const IR::U32 choice = SHAchoose(ir, low_y, after_low_y, before_high_y);+ const IR::U32 majority = SHAmajority(ir, low_x, after_low_x, before_high_x);++ const IR::U32 t = [&] {+ const IR::U32 w_element = ir.VectorGetElement(32, w, i);+ const IR::U32 sig = SHAhashSIGMA1(ir, low_y);++ return ir.Add(high_y, ir.Add(sig, ir.Add(choice, w_element)));+ }();++ const IR::U32 new_low_x = ir.Add(t, ir.Add(SHAhashSIGMA0(ir, low_x), majority));+ const IR::U32 new_low_y = ir.Add(t, high_x);++ // Shuffle all words left by 1 element: [3, 2, 1, 0] -> [2, 1, 0, 3]+ const IR::U128 shuffled_x = ir.VectorRotateWholeVectorRight(x, 96);+ const IR::U128 shuffled_y = ir.VectorRotateWholeVectorRight(y, 96);++ x = ir.VectorSetElement(32, shuffled_x, 0, new_low_x);+ y = ir.VectorSetElement(32, shuffled_y, 0, new_low_y);+ }++ inst.ReplaceUsesWith(part1 ? x : y);+}++template<size_t esize, bool is_signed>+static void PolyfillVectorMultiplyWiden(IR::IREmitter& ir, IR::Inst& inst) {+ IR::U128 n = (IR::U128)inst.GetArg(0);+ IR::U128 m = (IR::U128)inst.GetArg(1);++ const IR::U128 wide_n = is_signed ? ir.VectorSignExtend(esize, n) : ir.VectorZeroExtend(esize, n);+ const IR::U128 wide_m = is_signed ? ir.VectorSignExtend(esize, m) : ir.VectorZeroExtend(esize, m);++ const IR::U128 result = ir.VectorMultiply(esize * 2, wide_n, wide_m);++ inst.ReplaceUsesWith(result);+}++static void PolyfillPass(IR::Block& block, const PolyfillOptions& polyfill) {+ if (polyfill == PolyfillOptions{}) {+ return;+ }++ IR::IREmitter ir{block};++ for (auto& inst : block) {+ ir.SetInsertionPointBefore(&inst);++ switch (inst.GetOpcode()) {+ case IR::Opcode::SHA256MessageSchedule0:+ if (polyfill.sha256) {+ PolyfillSHA256MessageSchedule0(ir, inst);+ }+ break;+ case IR::Opcode::SHA256MessageSchedule1:+ if (polyfill.sha256) {+ PolyfillSHA256MessageSchedule1(ir, inst);+ }+ break;+ case IR::Opcode::SHA256Hash:+ if (polyfill.sha256) {+ PolyfillSHA256Hash(ir, inst);+ }+ break;+ case IR::Opcode::VectorMultiplySignedWiden8:+ if (polyfill.vector_multiply_widen) {+ PolyfillVectorMultiplyWiden<8, true>(ir, inst);+ }+ break;+ case IR::Opcode::VectorMultiplySignedWiden16:+ if (polyfill.vector_multiply_widen) {+ PolyfillVectorMultiplyWiden<16, true>(ir, inst);+ }+ break;+ case IR::Opcode::VectorMultiplySignedWiden32:+ if (polyfill.vector_multiply_widen) {+ PolyfillVectorMultiplyWiden<32, true>(ir, inst);+ }+ break;+ case IR::Opcode::VectorMultiplyUnsignedWiden8:+ if (polyfill.vector_multiply_widen) {+ PolyfillVectorMultiplyWiden<8, false>(ir, inst);+ }+ break;+ case IR::Opcode::VectorMultiplyUnsignedWiden16:+ if (polyfill.vector_multiply_widen) {+ PolyfillVectorMultiplyWiden<16, false>(ir, inst);+ }+ break;+ case IR::Opcode::VectorMultiplyUnsignedWiden32:+ if (polyfill.vector_multiply_widen) {+ PolyfillVectorMultiplyWiden<32, false>(ir, inst);+ }+ break;+ default:+ break;+ }+ }+}++static void VerificationPass(const IR::Block& block) {+ for (auto const& inst : block) {+ for (size_t i = 0; i < inst.NumArgs(); i++) {+ const IR::Type t1 = inst.GetArg(i).GetType();+ const IR::Type t2 = IR::GetArgTypeOf(inst.GetOpcode(), i);+ ASSERT_MSG(IR::AreTypesCompatible(t1, t2), "Block failed:\n{}", IR::DumpBlock(block));+ }+ }+ ankerl::unordered_dense::map<IR::Inst*, size_t> actual_uses;+ for (auto const& inst : block) {+ for (size_t i = 0; i < inst.NumArgs(); i++)+ if (IR::Value const arg = inst.GetArg(i); !arg.IsImmediate())+ actual_uses[arg.GetInst()]++;+ }+ for (auto const& pair : actual_uses)+ ASSERT(pair.first->UseCount() == pair.second);+}++void Optimize(IR::Block& block, const A32::UserConfig& conf, const Optimization::PolyfillOptions& polyfill_options) {+ Optimization::PolyfillPass(block, polyfill_options);+ Optimization::NamingPass(block);+ if (conf.HasOptimization(OptimizationFlag::GetSetElimination)) [[likely]] {+ Optimization::A32GetSetElimination(block, {.convert_nzc_to_nz = true});+ Optimization::DeadCodeElimination(block);+ }+ if (conf.HasOptimization(OptimizationFlag::ConstProp)) [[likely]] {+ Optimization::ConstantMemoryReads(block, conf.callbacks);+ Optimization::ConstantPropagation(block);+ Optimization::DeadCodeElimination(block);+ }+ Optimization::IdentityRemovalPass(block);+ //if (!conf.HasOptimization(OptimizationFlag::DisableVerification)) {+ Optimization::VerificationPass(block);+ //}+}++void Optimize(IR::Block& block, const A64::UserConfig& conf, const Optimization::PolyfillOptions& polyfill_options) {+ Optimization::PolyfillPass(block, polyfill_options);+ Optimization::A64CallbackConfigPass(block, conf);+ Optimization::NamingPass(block);+ if (conf.HasOptimization(OptimizationFlag::GetSetElimination) && !conf.check_halt_on_memory_access) [[likely]] {+ Optimization::A64GetSetElimination(block);+ Optimization::DeadCodeElimination(block);+ }+ if (conf.HasOptimization(OptimizationFlag::ConstProp)) [[likely]] {+ Optimization::ConstantPropagation(block);+ Optimization::DeadCodeElimination(block);+ }+ if (conf.HasOptimization(OptimizationFlag::MiscIROpt)) [[likely]] {+ Optimization::A64MergeInterpretBlocksPass(block, conf.callbacks);+ }+ //if (!conf.HasOptimization(OptimizationFlag::DisableVerification)) {+ Optimization::VerificationPass(block);+ //}+}++} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/src/dynarmic/ir/opt_passes.h b/src/dynarmic/src/dynarmic/ir/opt_passes.hnew file mode 100644index 0000000000..88b8020031--- /dev/null+++ b/src/dynarmic/src/dynarmic/ir/opt_passes.h@@ -0,0 +1,37 @@+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project+// SPDX-License-Identifier: GPL-3.0-or-later++/* This file is part of the dynarmic project.+ * Copyright (c) 2016 MerryMage+ * SPDX-License-Identifier: 0BSD+ */++#pragma once++namespace Dynarmic::A32 {+struct UserCallbacks;+struct UserConfig;+}++namespace Dynarmic::A64 {+struct UserCallbacks;+struct UserConfig;+}++namespace Dynarmic::IR {+class Block;+}++namespace Dynarmic::Optimization {++struct PolyfillOptions {+ bool sha256 = false;+ bool vector_multiply_widen = false;++ bool operator==(const PolyfillOptions&) const = default;+};++void Optimize(IR::Block& block, const A32::UserConfig& conf, const Optimization::PolyfillOptions& polyfill_options);+void Optimize(IR::Block& block, const A64::UserConfig& conf, const Optimization::PolyfillOptions& polyfill_options);++} // namespace Dynarmic::Optimizationdiff --git a/src/dynarmic/tests/A32/fuzz_thumb.cpp b/src/dynarmic/tests/A32/fuzz_thumb.cppindex 4d14141bbf..ad01e5718b 100644--- a/src/dynarmic/tests/A32/fuzz_thumb.cpp+++ b/src/dynarmic/tests/A32/fuzz_thumb.cpp@@ -29,7 +29,7 @@#include "dynarmic/frontend/A32/translate/a32_translate.h"#include "dynarmic/interface/A32/a32.h"#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opt/passes.h"+#include "dynarmic/ir/opt_passes.h"using namespace Dynarmic;@@ -179,13 +179,7 @@ static void RunInstance(size_t run_number, ThumbTestEnv& test_env, A32Unicorn<Thwhile (num_insts < instructions_to_execute_count) {A32::LocationDescriptor descriptor = {u32(num_insts * 4), cpsr, A32::FPSCR{}};IR::Block ir_block = A32::Translate(descriptor, &test_env, {});- Optimization::NamingPass(ir_block);- Optimization::A32GetSetElimination(ir_block, {.convert_nz_to_nzc = true});- Optimization::DeadCodeElimination(ir_block);- Optimization::A32ConstantMemoryReads(ir_block, &test_env);- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);- Optimization::VerificationPass(ir_block);+ Optimization::Optimize(ir_block, &test_env, {});printf("\n\nIR:\n%s", IR::DumpBlock(ir_block).c_str());printf("\n\nx86_64:\n");jit.DumpDisassembly();diff --git a/src/dynarmic/tests/A64/fuzz_with_unicorn.cpp b/src/dynarmic/tests/A64/fuzz_with_unicorn.cppindex 885bf3c0e7..1c3531ebc5 100644--- a/src/dynarmic/tests/A64/fuzz_with_unicorn.cpp+++ b/src/dynarmic/tests/A64/fuzz_with_unicorn.cpp@@ -28,7 +28,7 @@#include "dynarmic/frontend/A64/translate/a64_translate.h"#include "dynarmic/ir/basic_block.h"#include "dynarmic/ir/opcodes.h"-#include "dynarmic/ir/opt/passes.h"+#include "dynarmic/ir/opt_passes.h"// Must be declared last for all necessary operator<< to be declared prior to this.#include <fmt/format.h>@@ -271,17 +271,9 @@ static void RunTestInstance(Dynarmic::A64::Jit& jit, A64Unicorn& uni, A64TestEnvconst auto get_code = [&jit_env](u64 vaddr) { return jit_env.MemoryReadCode(vaddr); };IR::Block ir_block = A64::Translate({instructions_start, FP::FPCR{fpcr}}, get_code, {});- Optimization::A64CallbackConfigPass(ir_block, GetUserConfig(jit_env));- Optimization::NamingPass(ir_block);-fmt::print("IR:\n");fmt::print("{}\n", IR::DumpBlock(ir_block));-- Optimization::A64GetSetElimination(ir_block);- Optimization::DeadCodeElimination(ir_block);- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);-+ Optimization::Optimize(ir_block, conf, {});fmt::print("Optimized IR:\n");fmt::print("{}\n", IR::DumpBlock(ir_block));diff --git a/src/dynarmic/tests/print_info.cpp b/src/dynarmic/tests/print_info.cppindex ef8b87bbd1..4b6c951c03 100644--- a/src/dynarmic/tests/print_info.cpp+++ b/src/dynarmic/tests/print_info.cpp@@ -34,7 +34,7 @@#include "dynarmic/interface/A32/a32.h"#include "dynarmic/interface/A32/disassembler.h"#include "dynarmic/ir/basic_block.h"-#include "dynarmic/ir/opt/passes.h"+#include "dynarmic/ir/opt_passes.h"using namespace Dynarmic;@@ -64,18 +64,9 @@ void PrintA32Instruction(u32 instruction) {IR::Block ir_block{location};const bool should_continue = A32::TranslateSingleInstruction(ir_block, location, instruction);fmt::print("should_continue: {}\n\n", should_continue);-- Optimization::NamingPass(ir_block);-fmt::print("IR:\n");fmt::print("{}\n", IR::DumpBlock(ir_block));-- Optimization::A32GetSetElimination(ir_block, {});- Optimization::DeadCodeElimination(ir_block);- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);- Optimization::IdentityRemovalPass(ir_block);-+ Optimization::Optimize(ir_block, conf, {});fmt::print("Optimized IR:\n");fmt::print("{}\n", IR::DumpBlock(ir_block));}@@ -88,18 +79,9 @@ void PrintA64Instruction(u32 instruction) {IR::Block ir_block{location};const bool should_continue = A64::TranslateSingleInstruction(ir_block, location, instruction);fmt::print("should_continue: {}\n\n", should_continue);-- Optimization::NamingPass(ir_block);-fmt::print("IR:\n");fmt::print("{}\n", IR::DumpBlock(ir_block));-- Optimization::A64GetSetElimination(ir_block);- Optimization::DeadCodeElimination(ir_block);- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);- Optimization::IdentityRemovalPass(ir_block);-+ Optimization::Optimize(ir_block, conf, {});fmt::print("Optimized IR:\n");fmt::print("{}\n", IR::DumpBlock(ir_block));}@@ -115,18 +97,9 @@ void PrintThumbInstruction(u32 instruction) {IR::Block ir_block{location};const bool should_continue = A32::TranslateSingleInstruction(ir_block, location, instruction);fmt::print("should_continue: {}\n\n", should_continue);-- Optimization::NamingPass(ir_block);-fmt::print("IR:\n");fmt::print("{}\n", IR::DumpBlock(ir_block));-- Optimization::A32GetSetElimination(ir_block, {});- Optimization::DeadCodeElimination(ir_block);- Optimization::ConstantPropagation(ir_block);- Optimization::DeadCodeElimination(ir_block);- Optimization::IdentityRemovalPass(ir_block);-+ Optimization::Optimize(ir_block, conf, {});fmt::print("Optimized IR:\n");fmt::print("{}\n", IR::DumpBlock(ir_block));}