[dynarmic] enforce higher constraints

Signed-off-by: lizzie <lizzie@eden-emu.dev>

src/core/arm/dynarmic/arm_dynarmic_32.cpp

@@ -210,12 +210,9 @@ std::shared_ptr<Dynarmic::A32::Jit> ArmDynarmic32::MakeJit(Common::PageTable* pa
     config.wall_clock_cntpct = m_uses_wall_clock;
     config.enable_cycle_counting = !m_uses_wall_clock;
 
-    // Code cache size
-#ifdef ARCHITECTURE_arm64
+    // Code cache size - max in ARM is 128MiB, max in x86_64 is 2GiB
+    // Solaris doesn't support kPageSize >= 512MiB
     config.code_cache_size = std::uint32_t(128_MiB);
-#else
-    config.code_cache_size = std::uint32_t(512_MiB);
-#endif
 
     // Allow memory fault handling to work
     if (m_system.DebuggerEnabled()) {

src/core/arm/dynarmic/arm_dynarmic_64.cpp

@@ -269,12 +269,9 @@ std::shared_ptr<Dynarmic::A64::Jit> ArmDynarmic64::MakeJit(Common::PageTable* pa
     config.wall_clock_cntpct = m_uses_wall_clock;
     config.enable_cycle_counting = !m_uses_wall_clock;
 
-    // Code cache size
-#ifdef ARCHITECTURE_arm64
+    // Code cache size - max in ARM is 128MiB, max in x86_64 is 2GiB
+    // Solaris doesn't support kPageSize >= 512MiB
     config.code_cache_size = std::uint32_t(128_MiB);
-#else
-    config.code_cache_size = std::uint32_t(512_MiB);
-#endif
 
     // Allow memory fault handling to work
     if (m_system.DebuggerEnabled()) {

src/dynarmic/src/dynarmic/CMakeLists.txt

@@ -58,14 +58,11 @@ add_library(dynarmic
     common/lut_from_list.h
     common/math_util.cpp
     common/math_util.h
-    common/memory_pool.cpp
-    common/memory_pool.h
     common/safe_ops.h
     common/spin_lock.h
     common/string_util.h
     common/u128.cpp
     common/u128.h
-    common/variant_util.h
     frontend/A32/a32_types.cpp
     frontend/A32/a32_types.h
     frontend/A64/a64_types.cpp
@@ -80,7 +77,6 @@ add_library(dynarmic
     ir/basic_block.cpp
     ir/basic_block.h
     ir/cond.h
-    ir/ir_emitter.cpp
     ir/ir_emitter.h
     ir/location_descriptor.cpp
     ir/location_descriptor.h

src/dynarmic/src/dynarmic/backend/exception_handler.h

@@ -15,15 +15,15 @@
 #include <mcl/macro/architecture.hpp>
 #include "dynarmic/common/common_types.h"
 
-#if defined(MCL_ARCHITECTURE_X86_64)
+#if defined(ARCHITECTURE_x86_64)
 namespace Dynarmic::Backend::X64 {
 class BlockOfCode;
 }  // namespace Dynarmic::Backend::X64
-#elif defined(MCL_ARCHITECTURE_ARM64)
+#elif defined(ARCHITECTURE_arm64)
 namespace oaknut {
 class CodeBlock;
 }  // namespace oaknut
-#elif defined(MCL_ARCHITECTURE_RISCV)
+#elif defined(ARCHITECTURE_riscv64)
 namespace Dynarmic::Backend::RV64 {
 class CodeBlock;
 }  // namespace Dynarmic::Backend::RV64
@@ -33,16 +33,16 @@ class CodeBlock;
 
 namespace Dynarmic::Backend {
 
-#if defined(MCL_ARCHITECTURE_X86_64)
+#if defined(ARCHITECTURE_x86_64)
 struct FakeCall {
     u64 call_rip;
     u64 ret_rip;
 };
-#elif defined(MCL_ARCHITECTURE_ARM64)
+#elif defined(ARCHITECTURE_arm64)
 struct FakeCall {
     u64 call_pc;
 };
-#elif defined(MCL_ARCHITECTURE_RISCV)
+#elif defined(ARCHITECTURE_riscv64)
 struct FakeCall {
 };
 #else
@@ -54,11 +54,11 @@ public:
     ExceptionHandler();
     ~ExceptionHandler();
 
-#if defined(MCL_ARCHITECTURE_X86_64)
+#if defined(ARCHITECTURE_x86_64)
     void Register(X64::BlockOfCode& code);
-#elif defined(MCL_ARCHITECTURE_ARM64)
+#elif defined(ARCHITECTURE_arm64)
     void Register(oaknut::CodeBlock& mem, std::size_t mem_size);
-#elif defined(MCL_ARCHITECTURE_RISCV)
+#elif defined(ARCHITECTURE_riscv64)
     void Register(RV64::CodeBlock& mem, std::size_t mem_size);
 #else
 #    error "Invalid architecture"

src/dynarmic/src/dynarmic/backend/riscv64/a32_address_space.cpp

@@ -28,19 +28,7 @@ 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, {});
-    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::VerificationPass(ir_block);
-
+    Optimization::Optimize(ir_block, conf, {});
     return ir_block;
 }
 

src/dynarmic/src/dynarmic/backend/x64/a32_emit_x64.cpp

@@ -28,7 +28,6 @@
 #include "dynarmic/backend/x64/nzcv_util.h"
 #include "dynarmic/backend/x64/perf_map.h"
 #include "dynarmic/backend/x64/stack_layout.h"
-#include "dynarmic/common/variant_util.h"
 #include "dynarmic/frontend/A32/a32_location_descriptor.h"
 #include "dynarmic/frontend/A32/a32_types.h"
 #include "dynarmic/interface/A32/coprocessor.h"

src/dynarmic/src/dynarmic/backend/x64/a64_emit_x64.cpp

@@ -122,9 +122,9 @@ A64EmitX64::BlockDescriptor A64EmitX64::Emit(IR::Block& block) noexcept {
         auto const opcode = inst.GetOpcode();
         // Call the relevant Emit* member function.
         switch (opcode) {
-#define OPCODE(name, type, ...) [[likely]] case IR::Opcode::name: goto opcode_branch;
+#define OPCODE(name, type, ...) case IR::Opcode::name: goto opcode_branch;
 #define A32OPC(name, type, ...)
-#define A64OPC(name, type, ...) [[likely]] case IR::Opcode::A64##name: goto a64_branch;
+#define A64OPC(name, type, ...) case IR::Opcode::A64##name: goto a64_branch;
 #include "dynarmic/ir/opcodes.inc"
 #undef OPCODE
 #undef A32OPC
@@ -764,7 +764,7 @@ void A64EmitX64::EmitPatchMovRcx(CodePtr target_code_ptr) {
         target_code_ptr = code.GetReturnFromRunCodeAddress();
     }
     const CodePtr patch_location = code.getCurr();
-    code.mov(code.rcx, reinterpret_cast<u64>(target_code_ptr));
+    code.mov(code.rcx, u64(target_code_ptr));
     code.EnsurePatchLocationSize(patch_location, 10);
 }
 

src/dynarmic/src/dynarmic/backend/x64/a64_interface.cpp

@@ -80,16 +80,16 @@ public:
         };
 
         // TODO: Check code alignment
-
-        const CodePtr current_code_ptr = [this] {
+        const CodePtr aligned_code_ptr = CodePtr((uintptr_t(GetCurrentBlock()) + 15) & ~uintptr_t(15));
+        const CodePtr current_code_ptr = [this, aligned_code_ptr] {
             // RSB optimization
             const u32 new_rsb_ptr = (jit_state.rsb_ptr - 1) & A64JitState::RSBPtrMask;
             if (jit_state.GetUniqueHash() == jit_state.rsb_location_descriptors[new_rsb_ptr]) {
                 jit_state.rsb_ptr = new_rsb_ptr;
-                return reinterpret_cast<CodePtr>(jit_state.rsb_codeptrs[new_rsb_ptr]);
+                return CodePtr(jit_state.rsb_codeptrs[new_rsb_ptr]);
             }
-
-            return GetCurrentBlock();
+            return aligned_code_ptr;
+            //return GetCurrentBlock();
         }();
 
         const HaltReason hr = block_of_code.RunCode(&jit_state, current_code_ptr);

src/dynarmic/src/dynarmic/backend/x64/abi.cpp

@@ -10,7 +10,6 @@
 
 #include <algorithm>
 
-#include <mcl/iterator/reverse.hpp>
 #include "dynarmic/common/common_types.h"
 #include <xbyak/xbyak.h>
 
@@ -76,7 +75,8 @@ void ABI_PopRegistersAndAdjustStack(BlockOfCode& code, const size_t frame_size,
     const FrameInfo frame_info = CalculateFrameInfo(num_gprs, num_xmms, frame_size);
 
     size_t xmm_offset = frame_info.xmm_offset + (num_xmms * XMM_SIZE);
-    for (auto const xmm : mcl::iterator::reverse(regs)) {
+    for (auto it = regs.rbegin(); it != regs.rend(); ++it) {
+        auto const xmm = *it;
         if (HostLocIsXMM(xmm)) {
             xmm_offset -= XMM_SIZE;
             if (code.HasHostFeature(HostFeature::AVX)) {
@@ -88,9 +88,11 @@ void ABI_PopRegistersAndAdjustStack(BlockOfCode& code, const size_t frame_size,
     }
     if (frame_info.stack_subtraction != 0)
         code.add(rsp, u32(frame_info.stack_subtraction));
-    for (auto const gpr : mcl::iterator::reverse(regs))
+    for (auto it = regs.rbegin(); it != regs.rend(); ++it) {
+        auto const gpr = *it;
         if (HostLocIsGPR(gpr))
             code.pop(HostLocToReg64(gpr));
+    }
 }
 
 void ABI_PushCalleeSaveRegistersAndAdjustStack(BlockOfCode& code, const std::size_t frame_size) {

src/dynarmic/src/dynarmic/backend/x64/block_of_code.cpp

@@ -364,8 +364,7 @@ void BlockOfCode::GenRunCode(std::function<void(BlockOfCode&)> rcp) {
 
     cmp(dword[ABI_JIT_PTR + jsi.offsetof_halt_reason], 0);
     jne(return_to_caller_mxcsr_already_exited, T_NEAR);
-    lock();
-    or_(dword[ABI_JIT_PTR + jsi.offsetof_halt_reason], static_cast<u32>(HaltReason::Step));
+    lock(); or_(dword[ABI_JIT_PTR + jsi.offsetof_halt_reason], static_cast<u32>(HaltReason::Step));
 
     SwitchMxcsrOnEntry();
     jmp(ABI_PARAM2);
@@ -415,7 +414,6 @@ void BlockOfCode::GenRunCode(std::function<void(BlockOfCode&)> rcp) {
     }
 
     xor_(eax, eax);
-    lock();
     xchg(dword[ABI_JIT_PTR + jsi.offsetof_halt_reason], eax);
 
     ABI_PopCalleeSaveRegistersAndAdjustStack(*this, sizeof(StackLayout));

src/dynarmic/src/dynarmic/backend/x64/emit_x64.cpp

@@ -11,6 +11,7 @@
 #include <iterator>
 
 #include "dynarmic/common/assert.h"
+#include <boost/variant/detail/apply_visitor_binary.hpp>
 #include <mcl/bit/bit_field.hpp>
 #include <mcl/scope_exit.hpp>
 #include "dynarmic/common/common_types.h"
@@ -21,7 +22,6 @@
 #include "dynarmic/backend/x64/perf_map.h"
 #include "dynarmic/backend/x64/stack_layout.h"
 #include "dynarmic/backend/x64/verbose_debugging_output.h"
-#include "dynarmic/common/variant_util.h"
 #include "dynarmic/ir/basic_block.h"
 #include "dynarmic/ir/microinstruction.h"
 #include "dynarmic/ir/opcodes.h"
@@ -347,14 +347,14 @@ EmitX64::BlockDescriptor EmitX64::RegisterBlock(const IR::LocationDescriptor& de
 }
 
 void EmitX64::EmitTerminal(IR::Terminal terminal, IR::LocationDescriptor initial_location, bool is_single_step) {
-    Common::VisitVariant<void>(terminal, [this, initial_location, is_single_step](auto x) {
+    boost::apply_visitor([this, initial_location, is_single_step](auto x) {
         using T = std::decay_t<decltype(x)>;
         if constexpr (!std::is_same_v<T, IR::Term::Invalid>) {
             this->EmitTerminalImpl(x, initial_location, is_single_step);
         } else {
             ASSERT_MSG(false, "Invalid terminal");
         }
-    });
+    }, terminal);
 }
 
 void EmitX64::Patch(const IR::LocationDescriptor& target_desc, CodePtr target_code_ptr) {

src/dynarmic/src/dynarmic/backend/x64/emit_x64_floating_point.cpp

@@ -92,13 +92,10 @@ void ForceDenormalsToZero(BlockOfCode& code, std::initializer_list<Xbyak::Xmm> t
             FpFixup::Norm_Src,
             FpFixup::Norm_Src,
             FpFixup::Norm_Src);
-
-        const Xbyak::Xmm tmp = xmm16;
+        const Xbyak::Xmm tmp = xmm0;
         FCODE(vmovap)(tmp, code.BConst<fsize>(xword, denormal_to_zero));
-
-        for (const Xbyak::Xmm& xmm : to_daz) {
+        for (const Xbyak::Xmm& xmm : to_daz)
             FCODE(vfixupimms)(xmm, xmm, tmp, u8(0));
-        }
         return;
     }
 

src/dynarmic/src/dynarmic/backend/x64/emit_x64_memory.cpp.inc

@@ -273,34 +273,31 @@ void AxxEmitX64::EmitExclusiveWriteMemory(AxxEmitContext& ctx, IR::Inst* inst) {
     auto args = ctx.reg_alloc.GetArgumentInfo(inst);
     const bool ordered = IsOrdered(args[3].GetImmediateAccType());
 
-    if constexpr (bitsize != 128) {
-        ctx.reg_alloc.HostCall(inst, {}, args[1], args[2]);
-    } else {
+    if constexpr (bitsize == 128) {
         ctx.reg_alloc.Use(args[1], ABI_PARAM2);
         ctx.reg_alloc.Use(args[2], HostLoc::XMM1);
         ctx.reg_alloc.EndOfAllocScope();
         ctx.reg_alloc.HostCall(inst);
+    } else {
+        ctx.reg_alloc.HostCall(inst, {}, args[1], args[2]);
     }
 
+    const Xbyak::Reg64 tmp = ctx.reg_alloc.ScratchGpr();
     Xbyak::Label end;
-
     code.mov(code.ABI_RETURN, u32(1));
-    code.cmp(code.byte[code.ABI_JIT_PTR + offsetof(AxxJitState, exclusive_state)], u8(0));
+    code.movzx(tmp.cvt32(), code.byte[code.ABI_JIT_PTR + offsetof(AxxJitState, exclusive_state)]);
+    code.test(tmp.cvt8(), tmp.cvt8());
     code.je(end);
-    code.mov(code.byte[code.ABI_JIT_PTR + offsetof(AxxJitState, exclusive_state)], u8(0));
-    code.mov(code.ABI_PARAM1, reinterpret_cast<u64>(&conf));
+    code.xor_(tmp.cvt32(), tmp.cvt32());
+    code.xchg(tmp.cvt8(), code.byte[code.ABI_JIT_PTR + offsetof(AxxJitState, exclusive_state)]);
+    code.mov(code.ABI_PARAM1, u64(&conf));
     if constexpr (bitsize != 128) {
         using T = mcl::unsigned_integer_of_size<bitsize>;
-
-        code.CallLambda(
-            [](AxxUserConfig& conf, Axx::VAddr vaddr, T value) -> u32 {
-                return conf.global_monitor->DoExclusiveOperation<T>(conf.processor_id, vaddr,
-                                                                    [&](T expected) -> bool {
-                                                                        return (conf.callbacks->*callback)(vaddr, value, expected);
-                                                                    })
-                         ? 0
-                         : 1;
-            });
+        code.CallLambda([](AxxUserConfig& conf, Axx::VAddr vaddr, T value) -> u32 {
+            return conf.global_monitor->DoExclusiveOperation<T>(conf.processor_id, vaddr, [&](T expected) -> bool {
+                return (conf.callbacks->*callback)(vaddr, value, expected);
+            }) ? 0 : 1;
+        });
         if (ordered) {
             code.mfence();
         }
@@ -308,15 +305,11 @@ void AxxEmitX64::EmitExclusiveWriteMemory(AxxEmitContext& ctx, IR::Inst* inst) {
         ctx.reg_alloc.AllocStackSpace(16 + ABI_SHADOW_SPACE);
         code.lea(code.ABI_PARAM3, ptr[rsp + ABI_SHADOW_SPACE]);
         code.movaps(xword[code.ABI_PARAM3], xmm1);
-        code.CallLambda(
-            [](AxxUserConfig& conf, Axx::VAddr vaddr, Vector& value) -> u32 {
-                return conf.global_monitor->DoExclusiveOperation<Vector>(conf.processor_id, vaddr,
-                                                                         [&](Vector expected) -> bool {
-                                                                             return (conf.callbacks->*callback)(vaddr, value, expected);
-                                                                         })
-                         ? 0
-                         : 1;
-            });
+        code.CallLambda([](AxxUserConfig& conf, Axx::VAddr vaddr, Vector& value) -> u32 {
+            return conf.global_monitor->DoExclusiveOperation<Vector>(conf.processor_id, vaddr, [&](Vector expected) -> bool {
+                return (conf.callbacks->*callback)(vaddr, value, expected);
+            }) ? 0 : 1;
+        });
         if (ordered) {
             code.mfence();
         }
@@ -437,10 +430,11 @@ void AxxEmitX64::EmitExclusiveWriteMemoryInline(AxxEmitContext& ctx, IR::Inst* i
 
     SharedLabel end = GenSharedLabel();
 
-    code.mov(tmp, mcl::bit_cast<u64>(GetExclusiveMonitorAddressPointer(conf.global_monitor, conf.processor_id)));
     code.mov(status, u32(1));
-    code.cmp(code.byte[code.ABI_JIT_PTR + offsetof(AxxJitState, exclusive_state)], u8(0));
+    code.movzx(tmp.cvt32(), code.byte[code.ABI_JIT_PTR + offsetof(AxxJitState, exclusive_state)]);
+    code.test(tmp.cvt8(), tmp.cvt8());
     code.je(*end, code.T_NEAR);
+    code.mov(tmp, mcl::bit_cast<u64>(GetExclusiveMonitorAddressPointer(conf.global_monitor, conf.processor_id)));
     code.cmp(qword[tmp], vaddr);
     code.jne(*end, code.T_NEAR);
 
@@ -474,30 +468,29 @@ void AxxEmitX64::EmitExclusiveWriteMemoryInline(AxxEmitContext& ctx, IR::Inst* i
 
         const auto location = code.getCurr();
 
-        if constexpr (bitsize == 128) {
+        switch (bitsize) {
+        case 8:
+            code.lock();
+            code.cmpxchg(code.byte[dest_ptr], value.cvt8());
+            break;
+        case 16:
+            code.lock();
+            code.cmpxchg(word[dest_ptr], value.cvt16());
+            break;
+        case 32:
+            code.lock();
+            code.cmpxchg(dword[dest_ptr], value.cvt32());
+            break;
+        case 64:
+            code.lock();
+            code.cmpxchg(qword[dest_ptr], value.cvt64());
+            break;
+        case 128:
             code.lock();
             code.cmpxchg16b(ptr[dest_ptr]);
-        } else {
-            switch (bitsize) {
-            case 8:
-                code.lock();
-                code.cmpxchg(code.byte[dest_ptr], value.cvt8());
-                break;
-            case 16:
-                code.lock();
-                code.cmpxchg(word[dest_ptr], value.cvt16());
-                break;
-            case 32:
-                code.lock();
-                code.cmpxchg(dword[dest_ptr], value.cvt32());
-                break;
-            case 64:
-                code.lock();
-                code.cmpxchg(qword[dest_ptr], value.cvt64());
-                break;
-            default:
-                UNREACHABLE();
-            }
+            break;
+        default:
+            UNREACHABLE();
         }
 
         code.setnz(status.cvt8());

src/dynarmic/src/dynarmic/backend/x64/emit_x64_vector.cpp

@@ -609,8 +609,8 @@ void EmitX64::EmitVectorArithmeticVShift16(EmitContext& ctx, IR::Inst* inst) {
 
         const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
         const Xbyak::Xmm left_shift = ctx.reg_alloc.UseScratchXmm(args[1]);
-        const Xbyak::Xmm right_shift = xmm16;
-        const Xbyak::Xmm tmp = xmm17;
+        const Xbyak::Xmm right_shift = ctx.reg_alloc.ScratchXmm();
+        const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
 
         code.vmovdqa32(tmp, code.Const(xword, 0x00FF00FF00FF00FF, 0x00FF00FF00FF00FF));
         code.vpxord(right_shift, right_shift, right_shift);
@@ -674,8 +674,8 @@ void EmitX64::EmitVectorArithmeticVShift64(EmitContext& ctx, IR::Inst* inst) {
 
         const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
         const Xbyak::Xmm left_shift = ctx.reg_alloc.UseScratchXmm(args[1]);
-        const Xbyak::Xmm right_shift = xmm16;
-        const Xbyak::Xmm tmp = xmm17;
+        const Xbyak::Xmm right_shift = ctx.reg_alloc.ScratchXmm();
+        const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
 
         code.vmovdqa32(tmp, code.Const(xword, 0x00000000000000FF, 0x00000000000000FF));
         code.vpxorq(right_shift, right_shift, right_shift);
@@ -1955,8 +1955,8 @@ void EmitX64::EmitVectorLogicalVShift16(EmitContext& ctx, IR::Inst* inst) {
 
         const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
         const Xbyak::Xmm left_shift = ctx.reg_alloc.UseScratchXmm(args[1]);
-        const Xbyak::Xmm right_shift = xmm16;
-        const Xbyak::Xmm tmp = xmm17;
+        const Xbyak::Xmm right_shift = ctx.reg_alloc.ScratchXmm();
+        const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
 
         code.vmovdqa32(tmp, code.Const(xword, 0x00FF00FF00FF00FF, 0x00FF00FF00FF00FF));
         code.vpxord(right_shift, right_shift, right_shift);
@@ -2737,7 +2737,7 @@ void EmitX64::EmitVectorPairedAddSignedWiden32(EmitContext& ctx, IR::Inst* inst)
     const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
 
     if (code.HasHostFeature(HostFeature::AVX512_Ortho)) {
-        const Xbyak::Xmm c = xmm16;
+        const Xbyak::Xmm c = ctx.reg_alloc.ScratchXmm();
         code.vpsraq(c, a, 32);
         code.vpsllq(a, a, 32);
         code.vpsraq(a, a, 32);
@@ -5461,7 +5461,7 @@ void EmitX64::EmitVectorTableLookup128(EmitContext& ctx, IR::Inst* inst) {
     if (code.HasHostFeature(HostFeature::AVX512_Ortho | HostFeature::AVX512BW)) {
         const Xbyak::Xmm indicies = ctx.reg_alloc.UseXmm(args[2]);
         const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
-        const Xbyak::Xmm masked = xmm16;
+        const Xbyak::Xmm masked = ctx.reg_alloc.ScratchXmm();
 
         code.vpandd(masked, indicies, code.Const(xword_b, 0xF0F0F0F0F0F0F0F0, 0xF0F0F0F0F0F0F0F0));
 

src/dynarmic/src/dynarmic/backend/x64/reg_alloc.cpp

@@ -9,6 +9,7 @@
 #include "dynarmic/backend/x64/reg_alloc.h"
 
 #include <algorithm>
+#include <limits>
 #include <numeric>
 #include <utility>
 
@@ -118,7 +119,7 @@ void HostLocInfo::AddValue(IR::Inst* inst) noexcept {
     values.push_back(inst);
     ASSERT(size_t(total_uses) + inst->UseCount() < (std::numeric_limits<uint16_t>::max)());
     total_uses += inst->UseCount();
-    max_bit_width = std::max<uint8_t>(max_bit_width, GetBitWidth(inst->GetType()));
+    max_bit_width = std::max<uint8_t>(max_bit_width, std::countr_zero(GetBitWidth(inst->GetType())));
 }
 
 void HostLocInfo::EmitVerboseDebuggingOutput(BlockOfCode* code, size_t host_loc_index) const noexcept {
@@ -152,19 +153,19 @@ bool Argument::GetImmediateU1() const noexcept {
 
 u8 Argument::GetImmediateU8() const noexcept {
     const u64 imm = value.GetImmediateAsU64();
-    ASSERT(imm < 0x100);
+    ASSERT(imm <= u64(std::numeric_limits<u8>::max()));
     return u8(imm);
 }
 
 u16 Argument::GetImmediateU16() const noexcept {
     const u64 imm = value.GetImmediateAsU64();
-    ASSERT(imm < 0x10000);
+    ASSERT(imm <= u64(std::numeric_limits<u16>::max()));
     return u16(imm);
 }
 
 u32 Argument::GetImmediateU32() const noexcept {
     const u64 imm = value.GetImmediateAsU64();
-    ASSERT(imm < 0x100000000);
+    ASSERT(imm <= u64(std::numeric_limits<u32>::max()));
     return u32(imm);
 }
 
@@ -366,10 +367,20 @@ void RegAlloc::HostCall(IR::Inst* result_def,
     if (result_def) {
         DefineValueImpl(result_def, ABI_RETURN);
     }
-
+    for (size_t i = 0; i < args.size(); i++) {
+        if (args[i]) {
+            UseScratch(*args[i], args_hostloc[i]);
+        } else {
+            ScratchGpr(args_hostloc[i]); // TODO: Force spill
+        }
+    }
+    // Must match with with ScratchImpl
+    for (auto const gpr : other_caller_save) {
+        MoveOutOfTheWay(gpr);
+        LocInfo(gpr).WriteLock();
+    }
     for (size_t i = 0; i < args.size(); i++) {
         if (args[i] && !args[i]->get().IsVoid()) {
-            UseScratch(*args[i], args_hostloc[i]);
             // LLVM puts the burden of zero-extension of 8 and 16 bit values on the caller instead of the callee
             const Xbyak::Reg64 reg = HostLocToReg64(args_hostloc[i]);
             switch (args[i]->get().GetType()) {
@@ -389,14 +400,6 @@ void RegAlloc::HostCall(IR::Inst* result_def,
             }
         }
     }
-
-    for (size_t i = 0; i < args.size(); i++)
-        if (!args[i]) {
-            // TODO: Force spill
-            ScratchGpr(args_hostloc[i]);
-        }
-    for (auto const caller_saved : other_caller_save)
-        ScratchImpl({caller_saved});
 }
 
 void RegAlloc::AllocStackSpace(const size_t stack_space) noexcept {
@@ -559,13 +562,12 @@ void RegAlloc::SpillRegister(HostLoc loc) noexcept {
 }
 
 HostLoc RegAlloc::FindFreeSpill(bool is_xmm) const noexcept {
-#if 0
     // TODO(lizzie): Ok, Windows hates XMM spills, this means less perf for windows
     // but it's fine anyways. We can find other ways to cheat it later - but which?!?!
     // we should NOT save xmm each block entering... MAYBE xbyak has a bug on start/end?
     // TODO(lizzie): This needs to be investigated further later.
     // Do not spill XMM into other XMM silly
-    if (!is_xmm) {
+    /*if (!is_xmm) {
         // TODO(lizzie): Using lower (xmm0 and such) registers results in issues/crashes - INVESTIGATE WHY
         // Intel recommends to spill GPR onto XMM registers IF POSSIBLE
         // TODO(lizzie): Issues on DBZ, theory: Scratch XMM not properly restored after a function call?
@@ -573,8 +575,9 @@ HostLoc RegAlloc::FindFreeSpill(bool is_xmm) const noexcept {
         for (size_t i = size_t(HostLoc::XMM15); i >= size_t(HostLoc::XMM3); --i)
             if (const auto loc = HostLoc(i); LocInfo(loc).IsEmpty())
                 return loc;
-    }
-#endif
+    }*/
+    // TODO: Doing this would mean saving XMM on each call... need to benchmark the benefits
+    // of spilling on XMM versus the potential cost of using XMM registers.....
     // Otherwise go to stack spilling
     for (size_t i = size_t(HostLoc::FirstSpill); i < hostloc_info.size(); ++i)
         if (const auto loc = HostLoc(i); LocInfo(loc).IsEmpty())

src/dynarmic/src/dynarmic/backend/x64/reg_alloc.h

@@ -12,6 +12,7 @@
 #include <functional>
 #include <optional>
 
+#include "boost/container/small_vector.hpp"
 #include "dynarmic/common/common_types.h"
 #include <xbyak/xbyak.h>
 #include <boost/container/static_vector.hpp>
@@ -77,13 +78,13 @@ public:
         return std::find(values.begin(), values.end(), inst) != values.end();
     }
     inline size_t GetMaxBitWidth() const noexcept {
-        return max_bit_width;
+        return 1 << max_bit_width;
     }
     void AddValue(IR::Inst* inst) noexcept;
     void EmitVerboseDebuggingOutput(BlockOfCode* code, size_t host_loc_index) const noexcept;
 private:
 //non trivial
-    std::vector<IR::Inst*> values; //24
+    boost::container::small_vector<IR::Inst*, 3> values; //24
     // Block state
     uint16_t total_uses = 0; //8
     //sometimes zeroed
@@ -93,10 +94,10 @@ private:
     uint16_t is_being_used_count = 0; //8
     uint16_t current_references = 0; //8
     // Value state
-    uint8_t max_bit_width = 0; //Valid values: 1,2,4,8,16,32,128
+    uint8_t max_bit_width : 4 = 0; //Valid values: log2(1,2,4,8,16,32,128) = (0, 1, 2, 3, 4, 5, 6)
+    uint8_t lru_counter : 2 = 0; //1
     bool is_scratch : 1 = false; //1
     bool is_set_last_use : 1 = false; //1
-    alignas(16) uint8_t lru_counter = 0; //1
     friend class RegAlloc;
 };
 static_assert(sizeof(HostLocInfo) == 64);

src/dynarmic/src/dynarmic/common/memory_pool.cpp

@@ -1,13 +0,0 @@
-/* This file is part of the dynarmic project.
- * Copyright (c) 2016 MerryMage
- * SPDX-License-Identifier: 0BSD
- */
-
-#include "dynarmic/common/memory_pool.h"
-
-#include <cstdlib>
-
-namespace Dynarmic::Common {
-
-
-}  // namespace Dynarmic::Common

src/dynarmic/src/dynarmic/common/memory_pool.h

@@ -1,61 +0,0 @@
-/* This file is part of the dynarmic project.
- * Copyright (c) 2016 MerryMage
- * SPDX-License-Identifier: 0BSD
- */
-
-#pragma once
-
-#include <cstddef>
-#include <vector>
-
-namespace Dynarmic::Common {
-
-/// @tparam object_size Byte-size of objects to construct
-/// @tparam slab_size Number of objects to have per slab
-template<size_t object_size, size_t slab_size>
-class Pool {
-public:
-    inline Pool() noexcept {
-        AllocateNewSlab();
-    }
-    inline ~Pool() noexcept {
-        std::free(current_slab);
-        for (char* slab : slabs) {
-            std::free(slab);
-        }
-    }
-
-    Pool(const Pool&) = delete;
-    Pool(Pool&&) = delete;
-
-    Pool& operator=(const Pool&) = delete;
-    Pool& operator=(Pool&&) = delete;
-
-    /// @brief Returns a pointer to an `object_size`-bytes block of memory.
-    [[nodiscard]] void* Alloc() noexcept {
-        if (remaining == 0) {
-            slabs.push_back(current_slab);
-            AllocateNewSlab();
-        }
-        void* ret = static_cast<void*>(current_ptr);
-        current_ptr += object_size;
-        remaining--;
-        return ret;
-    }
-private:
-    /// @brief Allocates a completely new memory slab.
-    /// Used when an entirely new slab is needed
-    /// due the current one running out of usable space.
-    void AllocateNewSlab() noexcept {
-        current_slab = static_cast<char*>(std::malloc(object_size * slab_size));
-        current_ptr = current_slab;
-        remaining = slab_size;
-    }
-
-    std::vector<char*> slabs;
-    char* current_slab = nullptr;
-    char* current_ptr = nullptr;
-    size_t remaining = 0;
-};
-
-}  // namespace Dynarmic::Common

src/dynarmic/src/dynarmic/common/variant_util.h

@@ -1,29 +0,0 @@
-/* This file is part of the dynarmic project.
- * Copyright (c) 2016 MerryMage
- * SPDX-License-Identifier: 0BSD
- */
-
-#pragma once
-
-#include <boost/variant.hpp>
-
-namespace Dynarmic::Common {
-namespace detail {
-
-template<typename ReturnT, typename Lambda>
-struct VariantVisitor : boost::static_visitor<ReturnT>
-        , Lambda {
-    VariantVisitor(Lambda&& lambda)
-            : Lambda(std::move(lambda)) {}
-
-    using Lambda::operator();
-};
-
-}  // namespace detail
-
-template<typename ReturnT, typename Variant, typename Lambda>
-inline ReturnT VisitVariant(Variant&& variant, Lambda&& lambda) {
-    return boost::apply_visitor(detail::VariantVisitor<ReturnT, Lambda>(std::move(lambda)), variant);
-}
-
-}  // namespace Dynarmic::Common

src/dynarmic/src/dynarmic/frontend/A32/a32_types.h

@@ -9,12 +9,9 @@
 #pragma once
 
 #include <string>
-#include <utility>
-
 #include <fmt/format.h>
 #include "dynarmic/common/assert.h"
 #include "dynarmic/common/common_types.h"
-
 #include "dynarmic/interface/A32/coprocessor_util.h"
 #include "dynarmic/ir/cond.h"
 
@@ -89,24 +86,17 @@ constexpr bool IsQuadExtReg(ExtReg reg) {
 
 inline size_t RegNumber(Reg reg) {
     ASSERT(reg != Reg::INVALID_REG);
-    return static_cast<size_t>(reg);
+    return size_t(reg);
 }
 
 inline size_t RegNumber(ExtReg reg) {
     if (IsSingleExtReg(reg)) {
-        return static_cast<size_t>(reg) - static_cast<size_t>(ExtReg::S0);
+        return size_t(reg) - size_t(ExtReg::S0);
+    } else if (IsDoubleExtReg(reg)) {
+        return size_t(reg) - size_t(ExtReg::D0);
     }
-
-    if (IsDoubleExtReg(reg)) {
-        return static_cast<size_t>(reg) - static_cast<size_t>(ExtReg::D0);
-    }
-
-    if (IsQuadExtReg(reg)) {
-        return static_cast<size_t>(reg) - static_cast<size_t>(ExtReg::Q0);
-    }
-
-    ASSERT_MSG(false, "Invalid extended register");
-    return 0;
+    ASSERT(IsQuadExtReg(reg));
+    return size_t(reg) - size_t(ExtReg::Q0);
 }
 
 inline Reg operator+(Reg reg, size_t number) {

src/dynarmic/src/dynarmic/frontend/A32/decoder/arm.h

@@ -30,13 +30,13 @@ template<typename Visitor>
 using ArmDecodeTable = std::array<std::vector<ArmMatcher<Visitor>>, 0x1000>;
 
 namespace detail {
-inline size_t ToFastLookupIndexArm(u32 instruction) {
+inline size_t ToFastLookupIndexArm(u32 instruction) noexcept {
     return ((instruction >> 4) & 0x00F) | ((instruction >> 16) & 0xFF0);
 }
 }  // namespace detail
 
 template<typename V>
-constexpr ArmDecodeTable<V> GetArmDecodeTable() {
+constexpr ArmDecodeTable<V> GetArmDecodeTable() noexcept {
     std::vector<ArmMatcher<V>> list = {
 #define INST(fn, name, bitstring) DYNARMIC_DECODER_GET_MATCHER(ArmMatcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)),
 #include "./arm.inc"
@@ -62,15 +62,27 @@ constexpr ArmDecodeTable<V> GetArmDecodeTable() {
 }
 
 template<typename V>
-std::optional<std::reference_wrapper<const ArmMatcher<V>>> DecodeArm(u32 instruction) {
+std::optional<std::reference_wrapper<const ArmMatcher<V>>> DecodeArm(u32 instruction) noexcept {
     alignas(64) static const auto table = GetArmDecodeTable<V>();
     const auto matches_instruction = [instruction](const auto& matcher) {
         return matcher.Matches(instruction);
     };
-
     const auto& subtable = table[detail::ToFastLookupIndexArm(instruction)];
     auto iter = std::find_if(subtable.begin(), subtable.end(), matches_instruction);
     return iter != subtable.end() ? std::optional<std::reference_wrapper<const ArmMatcher<V>>>(*iter) : std::nullopt;
 }
 
+template<typename V>
+std::optional<std::string_view> GetNameARM(u32 inst) noexcept {
+    std::vector<std::pair<std::string_view, ArmMatcher<V>>> list = {
+#define INST(fn, name, bitstring) { name, DYNARMIC_DECODER_GET_MATCHER(ArmMatcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)) },
+#include "./arm.inc"
+#undef INST
+    };
+    auto const iter = std::find_if(list.cbegin(), list.cend(), [inst](auto const& m) {
+        return m.second.Matches(inst);
+    });
+    return iter != list.cend() ? std::optional{iter->first} : std::nullopt;
+}
+
 }  // namespace Dynarmic::A32

src/dynarmic/src/dynarmic/frontend/A32/decoder/asimd.h

@@ -26,15 +26,12 @@ template<typename Visitor>
 using ASIMDMatcher = Decoder::Matcher<Visitor, u32>;
 
 template<typename V>
-std::vector<ASIMDMatcher<V>> GetASIMDDecodeTable() {
-    std::vector<ASIMDMatcher<V>> table = {
-
-#define INST(fn, name, bitstring) DYNARMIC_DECODER_GET_MATCHER(ASIMDMatcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)),
+std::vector<ASIMDMatcher<V>> GetASIMDDecodeTable() noexcept {
+    std::vector<std::pair<const char*, ASIMDMatcher<V>>> table = {
+#define INST(fn, name, bitstring) { name, DYNARMIC_DECODER_GET_MATCHER(ASIMDMatcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)) },
 #include "./asimd.inc"
 #undef INST
-
     };
-
     // Exceptions to the rule of thumb.
     const std::set<std::string> comes_first{
         "VBIC, VMOV, VMVN, VORR (immediate)",
@@ -53,29 +50,43 @@ std::vector<ASIMDMatcher<V>> GetASIMDDecodeTable() {
         "VQDMULH (scalar)",
         "VQRDMULH (scalar)",
     };
-    const auto sort_begin = std::stable_partition(table.begin(), table.end(), [&](const auto& matcher) {
-        return comes_first.count(matcher.GetName()) > 0;
+    const auto sort_begin = std::stable_partition(table.begin(), table.end(), [&](const auto& e) {
+        return comes_first.count(e.first) > 0;
     });
-    const auto sort_end = std::stable_partition(table.begin(), table.end(), [&](const auto& matcher) {
-        return comes_last.count(matcher.GetName()) == 0;
+    const auto sort_end = std::stable_partition(table.begin(), table.end(), [&](const auto& e) {
+        return comes_last.count(e.first) == 0;
     });
-
     // If a matcher has more bits in its mask it is more specific, so it should come first.
-    std::stable_sort(sort_begin, sort_end, [](const auto& matcher1, const auto& matcher2) {
-        return mcl::bit::count_ones(matcher1.GetMask()) > mcl::bit::count_ones(matcher2.GetMask());
+    std::stable_sort(sort_begin, sort_end, [](const auto& a, const auto& b) {
+        return mcl::bit::count_ones(a.second.GetMask()) > mcl::bit::count_ones(b.second.GetMask());
     });
-
-    return table;
+    std::vector<ASIMDMatcher<V>> final_table;
+    std::transform(table.cbegin(), table.cend(), std::back_inserter(final_table), [](auto const& e) {
+        return e.second;
+    });
+    return final_table;
 }
 
 template<typename V>
-std::optional<std::reference_wrapper<const ASIMDMatcher<V>>> DecodeASIMD(u32 instruction) {
-    static const auto table = GetASIMDDecodeTable<V>();
-
-    const auto matches_instruction = [instruction](const auto& matcher) { return matcher.Matches(instruction); };
-
-    auto iter = std::find_if(table.begin(), table.end(), matches_instruction);
+std::optional<std::reference_wrapper<const ASIMDMatcher<V>>> DecodeASIMD(u32 instruction) noexcept {
+    alignas(64) static const auto table = GetASIMDDecodeTable<V>();
+    auto iter = std::find_if(table.begin(), table.end(), [instruction](const auto& matcher) {
+        return matcher.Matches(instruction);
+    });
     return iter != table.end() ? std::optional<std::reference_wrapper<const ASIMDMatcher<V>>>(*iter) : std::nullopt;
 }
 
+template<typename V>
+std::optional<std::string_view> GetNameASIMD(u32 inst) noexcept {
+    std::vector<std::pair<std::string_view, ASIMDMatcher<V>>> list = {
+#define INST(fn, name, bitstring) { name, DYNARMIC_DECODER_GET_MATCHER(ASIMDMatcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)) },
+#include "./asimd.inc"
+#undef INST
+    };
+    auto const iter = std::find_if(list.cbegin(), list.cend(), [inst](auto const& m) {
+        return m.second.Matches(inst);
+    });
+    return iter != list.cend() ? std::optional{iter->first} : std::nullopt;
+}
+
 }  // namespace Dynarmic::A32

src/dynarmic/src/dynarmic/frontend/A32/decoder/thumb16.h

@@ -25,18 +25,28 @@ using Thumb16Matcher = Decoder::Matcher<Visitor, u16>;
 
 template<typename V>
 std::optional<std::reference_wrapper<const Thumb16Matcher<V>>> DecodeThumb16(u16 instruction) {
-    static const std::vector<Thumb16Matcher<V>> table = {
-
+    alignas(64) static const std::vector<Thumb16Matcher<V>> table = {
 #define INST(fn, name, bitstring) DYNARMIC_DECODER_GET_MATCHER(Thumb16Matcher, fn, name, Decoder::detail::StringToArray<16>(bitstring)),
 #include "./thumb16.inc"
 #undef INST
-
     };
-
-    const auto matches_instruction = [instruction](const auto& matcher) { return matcher.Matches(instruction); };
-
-    auto iter = std::find_if(table.begin(), table.end(), matches_instruction);
+    auto iter = std::find_if(table.begin(), table.end(), [instruction](const auto& matcher) {
+        return matcher.Matches(instruction);
+    });
     return iter != table.end() ? std::optional<std::reference_wrapper<const Thumb16Matcher<V>>>(*iter) : std::nullopt;
 }
 
+template<typename V>
+std::optional<std::string_view> GetNameThumb16(u32 inst) noexcept {
+    std::vector<std::pair<std::string_view, Thumb16Matcher<V>>> list = {
+#define INST(fn, name, bitstring) { name, DYNARMIC_DECODER_GET_MATCHER(Thumb16Matcher, fn, name, Decoder::detail::StringToArray<16>(bitstring)) },
+#include "./thumb16.inc"
+#undef INST
+    };
+    auto const iter = std::find_if(list.cbegin(), list.cend(), [inst](auto const& m) {
+        return m.second.Matches(inst);
+    });
+    return iter != list.cend() ? std::optional{iter->first} : std::nullopt;
+}
+
 }  // namespace Dynarmic::A32

src/dynarmic/src/dynarmic/frontend/A32/decoder/thumb32.h

@@ -24,18 +24,28 @@ using Thumb32Matcher = Decoder::Matcher<Visitor, u32>;
 
 template<typename V>
 std::optional<std::reference_wrapper<const Thumb32Matcher<V>>> DecodeThumb32(u32 instruction) {
-    static const std::vector<Thumb32Matcher<V>> table = {
-
+    alignas(64) static const std::vector<Thumb32Matcher<V>> table = {
 #define INST(fn, name, bitstring) DYNARMIC_DECODER_GET_MATCHER(Thumb32Matcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)),
 #include "./thumb32.inc"
 #undef INST
-
     };
-
-    const auto matches_instruction = [instruction](const auto& matcher) { return matcher.Matches(instruction); };
-
-    auto iter = std::find_if(table.begin(), table.end(), matches_instruction);
+    auto iter = std::find_if(table.begin(), table.end(), [instruction](const auto& matcher) {
+        return matcher.Matches(instruction);
+    });
     return iter != table.end() ? std::optional<std::reference_wrapper<const Thumb32Matcher<V>>>(*iter) : std::nullopt;
 }
 
+template<typename V>
+std::optional<std::string_view> GetNameThumb32(u32 inst) noexcept {
+    std::vector<std::pair<std::string_view, Thumb32Matcher<V>>> list = {
+#define INST(fn, name, bitstring) { name, DYNARMIC_DECODER_GET_MATCHER(Thumb32Matcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)) },
+#include "./thumb32.inc"
+#undef INST
+    };
+    auto const iter = std::find_if(list.cbegin(), list.cend(), [inst](auto const& m) {
+        return m.second.Matches(inst);
+    });
+    return iter != list.cend() ? std::optional{iter->first} : std::nullopt;
+}
+
 }  // namespace Dynarmic::A32

src/dynarmic/src/dynarmic/frontend/A32/decoder/vfp.h

@@ -26,36 +26,42 @@ using VFPMatcher = Decoder::Matcher<Visitor, u32>;
 template<typename V>
 std::optional<std::reference_wrapper<const VFPMatcher<V>>> DecodeVFP(u32 instruction) {
     using Table = std::vector<VFPMatcher<V>>;
-
-    static const struct Tables {
+    alignas(64) static const struct Tables {
         Table unconditional;
         Table conditional;
-    } tables = [] {
+    } tables = []() {
         Table list = {
-
 #define INST(fn, name, bitstring) DYNARMIC_DECODER_GET_MATCHER(VFPMatcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)),
 #include "./vfp.inc"
 #undef INST
-
         };
-
-        const auto division = std::stable_partition(list.begin(), list.end(), [&](const auto& matcher) {
+        auto const it = std::stable_partition(list.begin(), list.end(), [&](const auto& matcher) {
             return (matcher.GetMask() & 0xF0000000) == 0xF0000000;
         });
-
         return Tables{
-            Table{list.begin(), division},
-            Table{division, list.end()},
+            Table{list.begin(), it},
+            Table{it, list.end()},
         };
     }();
-
     const bool is_unconditional = (instruction & 0xF0000000) == 0xF0000000;
     const Table& table = is_unconditional ? tables.unconditional : tables.conditional;
-
-    const auto matches_instruction = [instruction](const auto& matcher) { return matcher.Matches(instruction); };
-
-    auto iter = std::find_if(table.begin(), table.end(), matches_instruction);
+    auto iter = std::find_if(table.begin(), table.end(), [instruction](const auto& matcher) {
+        return matcher.Matches(instruction);
+    });
     return iter != table.end() ? std::optional<std::reference_wrapper<const VFPMatcher<V>>>(*iter) : std::nullopt;
 }
 
+template<typename V>
+std::optional<std::string_view> GetNameVFP(u32 inst) noexcept {
+    std::vector<std::pair<std::string_view, VFPMatcher<V>>> list = {
+#define INST(fn, name, bitstring) { name, DYNARMIC_DECODER_GET_MATCHER(VFPMatcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)) },
+#include "./vfp.inc"
+#undef INST
+    };
+    auto const iter = std::find_if(list.cbegin(), list.cend(), [inst](auto const& m) {
+        return m.second.Matches(inst);
+    });
+    return iter != list.cend() ? std::optional{iter->first} : std::nullopt;
+}
+
 }  // namespace Dynarmic::A32

src/dynarmic/src/dynarmic/frontend/A32/translate/translate_thumb.cpp

@@ -97,7 +97,7 @@ u32 ConvertASIMDInstruction(u32 thumb_instruction) {
     return 0xF7F0A000;  // UDF
 }
 
-bool MaybeVFPOrASIMDInstruction(u32 thumb_instruction) {
+inline bool MaybeVFPOrASIMDInstruction(u32 thumb_instruction) noexcept {
     return (thumb_instruction & 0xEC000000) == 0xEC000000 || (thumb_instruction & 0xFF100000) == 0xF9000000;
 }
 

src/dynarmic/src/dynarmic/frontend/A64/decoder/a64.h

@@ -37,34 +37,31 @@ inline size_t ToFastLookupIndex(u32 instruction) {
 
 template<typename V>
 constexpr DecodeTable<V> GetDecodeTable() {
-    std::vector<Matcher<V>> list = {
-#define INST(fn, name, bitstring) DYNARMIC_DECODER_GET_MATCHER(Matcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)),
+    std::vector<std::pair<const char*, Matcher<V>>> list = {
+#define INST(fn, name, bitstring) { name, DYNARMIC_DECODER_GET_MATCHER(Matcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)) },
 #include "./a64.inc"
 #undef INST
     };
-
     // If a matcher has more bits in its mask it is more specific, so it should come first.
-    std::stable_sort(list.begin(), list.end(), [](const auto& matcher1, const auto& matcher2) {
+    std::stable_sort(list.begin(), list.end(), [](const auto& a, const auto& b) {
         // If a matcher has more bits in its mask it is more specific, so it should come first.
-        return mcl::bit::count_ones(matcher1.GetMask()) > mcl::bit::count_ones(matcher2.GetMask());
+        return mcl::bit::count_ones(a.second.GetMask()) > mcl::bit::count_ones(b.second.GetMask());
     });
-
     // Exceptions to the above rule of thumb.
-    std::stable_partition(list.begin(), list.end(), [&](const auto& matcher) {
+    std::stable_partition(list.begin(), list.end(), [&](const auto& e) {
         return std::set<std::string>{
             "MOVI, MVNI, ORR, BIC (vector, immediate)",
             "FMOV (vector, immediate)",
             "Unallocated SIMD modified immediate",
-        }.count(matcher.GetName()) > 0;
+        }.count(e.first) > 0;
     });
-
     DecodeTable<V> table{};
     for (size_t i = 0; i < table.size(); ++i) {
-        for (auto matcher : list) {
-            const auto expect = detail::ToFastLookupIndex(matcher.GetExpected());
-            const auto mask = detail::ToFastLookupIndex(matcher.GetMask());
+        for (auto const& e : list) {
+            const auto expect = detail::ToFastLookupIndex(e.second.GetExpected());
+            const auto mask = detail::ToFastLookupIndex(e.second.GetMask());
             if ((i & mask) == expect) {
-                table[i].push_back(matcher);
+                table[i].push_back(e.second);
             }
         }
     }
@@ -74,12 +71,24 @@ constexpr DecodeTable<V> GetDecodeTable() {
 template<typename V>
 std::optional<std::reference_wrapper<const Matcher<V>>> Decode(u32 instruction) {
     alignas(64) static const auto table = GetDecodeTable<V>();
-    const auto matches_instruction = [instruction](const auto& matcher) {
-        return matcher.Matches(instruction);
-    };
     const auto& subtable = table[detail::ToFastLookupIndex(instruction)];
-    auto iter = std::find_if(subtable.begin(), subtable.end(), matches_instruction);
+    auto iter = std::find_if(subtable.begin(), subtable.end(), [instruction](const auto& matcher) {
+        return matcher.Matches(instruction);
+    });
     return iter != subtable.end() ? std::optional<std::reference_wrapper<const Matcher<V>>>(*iter) : std::nullopt;
 }
 
+template<typename V>
+std::optional<std::string_view> GetName(u32 inst) noexcept {
+    std::vector<std::pair<std::string_view, Matcher<V>>> list = {
+#define INST(fn, name, bitstring) { name, DYNARMIC_DECODER_GET_MATCHER(Matcher, fn, name, Decoder::detail::StringToArray<32>(bitstring)) },
+#include "./a64.inc"
+#undef INST
+    };
+    auto const iter = std::find_if(list.cbegin(), list.cend(), [inst](auto const& m) {
+        return m.second.Matches(inst);
+    });
+    return iter != list.cend() ? std::optional{iter->first} : std::nullopt;
+}
+
 }  // namespace Dynarmic::A64

src/dynarmic/src/dynarmic/frontend/A64/translate/impl/a64_branch.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) 2018 MerryMage
  * SPDX-License-Identifier: 0BSD
@@ -20,9 +23,12 @@ bool TranslatorVisitor::B_cond(Imm<19> imm19, Cond cond) {
 bool TranslatorVisitor::B_uncond(Imm<26> imm26) {
     const s64 offset = concatenate(imm26, Imm<2>{0}).SignExtend<s64>();
     const u64 target = ir.PC() + offset;
-
-    //ir.SetTerm(IR::Term::LinkBlockFast{ir.current_location->SetPC(target)});
-    ir.SetTerm(IR::Term::LinkBlock{ir.current_location->SetPC(target)});
+    // Pattern to halt execution (B .)
+    if (target == ir.PC()) {
+        ir.SetTerm(IR::Term::LinkBlock{ir.current_location->SetPC(target)});
+        return false;
+    }
+    ir.SetTerm(IR::Term::LinkBlockFast{ir.current_location->SetPC(target)});
     return false;
 }
 

src/dynarmic/src/dynarmic/frontend/decoder/decoder_detail.h

@@ -70,11 +70,9 @@ struct detail {
         return std::make_tuple(mask, expect);
     }
 
-    /**
-     * Generates the masks and shifts for each argument.
-     * A '-' in a bitstring indicates that we don't care about that value.
-     * An argument is specified by a continuous string of the same character.
-     */
+    /// @brief Generates the masks and shifts for each argument.
+    /// A '-' in a bitstring indicates that we don't care about that value.
+    /// An argument is specified by a continuous string of the same character.
     template<size_t N>
     static consteval auto GetArgInfo(std::array<char, opcode_bitsize> bitstring) {
         std::array<opcode_type, N> masks = {};
@@ -98,7 +96,6 @@ struct detail {
 
                 if constexpr (N > 0) {
                     const size_t bit_position = opcode_bitsize - i - 1;
-
                     if (arg_index >= N)
                         throw std::out_of_range("Unexpected field");
 
@@ -109,20 +106,16 @@ struct detail {
                 }
             }
         }
-
 #if !defined(DYNARMIC_IGNORE_ASSERTS) && !defined(__ANDROID__)
         // Avoids a MSVC ICE, and avoids Android NDK issue.
         ASSERT(std::all_of(masks.begin(), masks.end(), [](auto m) { return m != 0; }));
 #endif
-
         return std::make_tuple(masks, shifts);
     }
 
-    /**
-     * This struct's Make member function generates a lambda which decodes an instruction based on
-     * the provided arg_masks and arg_shifts. The Visitor member function to call is provided as a
-     * template argument.
-     */
+    /// @brief This struct's Make member function generates a lambda which decodes an instruction
+    /// based on the provided arg_masks and arg_shifts. The Visitor member function to call is
+    /// provided as a template argument.
     template<typename FnT>
     struct VisitorCaller;
 
@@ -130,36 +123,36 @@ struct detail {
 #    pragma warning(push)
 #    pragma warning(disable : 4800)  // forcing value to bool 'true' or 'false' (performance warning)
 #endif
-    template<typename Visitor, typename... Args, typename CallRetT>
-    struct VisitorCaller<CallRetT (Visitor::*)(Args...)> {
+    template<typename V, typename... Args, typename ReturnType>
+    struct VisitorCaller<ReturnType (V::*)(Args...)> {
         template<size_t... iota>
-        static auto Make(std::integer_sequence<size_t, iota...>,
-                         CallRetT (Visitor::*const fn)(Args...),
+        static constexpr auto Make(std::integer_sequence<size_t, iota...>,
+                         ReturnType (V::*const fn)(Args...),
                          const std::array<opcode_type, sizeof...(iota)> arg_masks,
                          const std::array<size_t, sizeof...(iota)> arg_shifts) {
-            static_assert(std::is_same_v<visitor_type, Visitor>, "Member function is not from Matcher's Visitor");
-            return [fn, arg_masks, arg_shifts](Visitor& v, opcode_type instruction) {
+            static_assert(std::is_same_v<visitor_type, V>, "Member function is not from Matcher's Visitor");
+            return [fn, arg_masks, arg_shifts](V& v, opcode_type instruction) {
                 (void)instruction;
                 (void)arg_masks;
                 (void)arg_shifts;
-                return (v.*fn)(static_cast<Args>((instruction & arg_masks[iota]) >> arg_shifts[iota])...);
+                return (v.*fn)(Args((instruction & arg_masks[iota]) >> arg_shifts[iota])...);
             };
         }
     };
 
-    template<typename Visitor, typename... Args, typename CallRetT>
-    struct VisitorCaller<CallRetT (Visitor::*)(Args...) const> {
+    template<typename V, typename... Args, typename ReturnType>
+    struct VisitorCaller<ReturnType (V::*)(Args...) const> {
         template<size_t... iota>
-        static auto Make(std::integer_sequence<size_t, iota...>,
-                         CallRetT (Visitor::*const fn)(Args...) const,
+        static constexpr auto Make(std::integer_sequence<size_t, iota...>,
+                         ReturnType (V::*const fn)(Args...) const,
                          const std::array<opcode_type, sizeof...(iota)> arg_masks,
                          const std::array<size_t, sizeof...(iota)> arg_shifts) {
-            static_assert(std::is_same_v<visitor_type, const Visitor>, "Member function is not from Matcher's Visitor");
-            return [fn, arg_masks, arg_shifts](const Visitor& v, opcode_type instruction) {
+            static_assert(std::is_same_v<visitor_type, const V>, "Member function is not from Matcher's Visitor");
+            return [fn, arg_masks, arg_shifts](const V& v, opcode_type instruction) {
                 (void)instruction;
                 (void)arg_masks;
                 (void)arg_shifts;
-                return (v.*fn)(static_cast<Args>((instruction & arg_masks[iota]) >> arg_shifts[iota])...);
+                return (v.*fn)(Args((instruction & arg_masks[iota]) >> arg_shifts[iota])...);
             };
         }
     };
@@ -167,27 +160,21 @@ struct detail {
 #    pragma warning(pop)
 #endif
 
-    /**
-     * Creates a matcher that can match and parse instructions based on bitstring.
-     * See also: GetMaskAndExpect and GetArgInfo for format of bitstring.
-     */
-    template<auto bitstring, typename FnT>
-    static auto GetMatcher(FnT fn, const char* const name) {
-        constexpr size_t args_count = mcl::parameter_count_v<FnT>;
-
+    /// @brief Creates a matcher that can match and parse instructions based on bitstring.
+    /// See also: GetMaskAndExpect and GetArgInfo for format of bitstring.
+    template<auto bitstring, typename F>
+    static constexpr auto GetMatcher(F fn) {
+        constexpr size_t args_count = mcl::parameter_count_v<F>;
         constexpr auto mask = std::get<0>(GetMaskAndExpect(bitstring));
         constexpr auto expect = std::get<1>(GetMaskAndExpect(bitstring));
         constexpr auto arg_masks = std::get<0>(GetArgInfo<args_count>(bitstring));
         constexpr auto arg_shifts = std::get<1>(GetArgInfo<args_count>(bitstring));
-
-        using Iota = std::make_index_sequence<args_count>;
-
-        const auto proxy_fn = VisitorCaller<FnT>::Make(Iota(), fn, arg_masks, arg_shifts);
-        return MatcherT(name, mask, expect, proxy_fn);
+        const auto proxy_fn = VisitorCaller<F>::Make(std::make_index_sequence<args_count>(), fn, arg_masks, arg_shifts);
+        return MatcherT(mask, expect, proxy_fn);
     }
 };
 
-#define DYNARMIC_DECODER_GET_MATCHER(MatcherT, fn, name, bitstring) Decoder::detail::detail<MatcherT<V>>::template GetMatcher<bitstring>(&V::fn, name)
+#define DYNARMIC_DECODER_GET_MATCHER(MatcherT, fn, name, bitstring) Decoder::detail::detail<MatcherT<V>>::template GetMatcher<bitstring>(&V::fn)
 
 }  // namespace detail
 }  // namespace Dynarmic::Decoder

src/dynarmic/src/dynarmic/frontend/decoder/matcher.h

@@ -14,16 +14,12 @@
 
 namespace Dynarmic::Decoder {
 
-/**
- * Generic instruction handling construct.
- *
- * @tparam Visitor An arbitrary visitor type that will be passed through
- *                 to the function being handled. This type must be the
- *                 type of the first parameter in a handler function.
- *
- * @tparam OpcodeType Type representing an opcode. This must be the
- *                    type of the second parameter in a handler function.
- */
+/// Generic instruction handling construct.
+/// @tparam Visitor An arbitrary visitor type that will be passed through
+///                 to the function being handled. This type must be the
+///                 type of the first parameter in a handler function.
+/// @tparam OpcodeType Type representing an opcode. This must be the
+///                    type of the second parameter in a handler function.
 template<typename Visitor, typename OpcodeType>
 class Matcher {
 public:
@@ -31,46 +27,35 @@ public:
     using visitor_type = Visitor;
     using handler_return_type = typename Visitor::instruction_return_type;
     using handler_function = std::function<handler_return_type(Visitor&, opcode_type)>;
-
-    Matcher(const char* const name, opcode_type mask, opcode_type expected, handler_function func)
-            : name{name}, mask{mask}, expected{expected}, fn{std::move(func)} {}
-
-    /// Gets the name of this type of instruction.
-    const char* GetName() const {
-        return name;
-    }
+    Matcher(opcode_type mask, opcode_type expected, handler_function func)
+            : mask{mask}, expected{expected}, fn{std::move(func)} {}
 
     /// Gets the mask for this instruction.
-    opcode_type GetMask() const {
+    inline opcode_type GetMask() const noexcept {
         return mask;
     }
 
     /// Gets the expected value after masking for this instruction.
-    opcode_type GetExpected() const {
+    inline opcode_type GetExpected() const noexcept {
         return expected;
     }
 
-    /**
-     * Tests to see if the given instruction is the instruction this matcher represents.
-     * @param instruction The instruction to test
-     * @returns true if the given instruction matches.
-     */
-    bool Matches(opcode_type instruction) const {
+    /// Tests to see if the given instruction is the instruction this matcher represents.
+    /// @param instruction The instruction to test
+    /// @returns true if the given instruction matches.
+    inline bool Matches(opcode_type instruction) const noexcept {
         return (instruction & mask) == expected;
     }
 
-    /**
-     * Calls the corresponding instruction handler on visitor for this type of instruction.
-     * @param v The visitor to use
-     * @param instruction The instruction to decode.
-     */
-    handler_return_type call(Visitor& v, opcode_type instruction) const {
+    /// Calls the corresponding instruction handler on visitor for this type of instruction.
+    /// @param v The visitor to use
+    /// @param instruction The instruction to decode.
+    inline handler_return_type call(Visitor& v, opcode_type instruction) const noexcept {
         ASSERT(Matches(instruction));
         return fn(v, instruction);
     }
 
 private:
-    const char* name;
     opcode_type mask;
     opcode_type expected;
     handler_function fn;

src/dynarmic/src/dynarmic/ir/basic_block.cpp

@@ -15,8 +15,6 @@
 
 #include <fmt/format.h>
 #include "dynarmic/common/assert.h"
-
-#include "dynarmic/common/memory_pool.h"
 #include "dynarmic/frontend/A32/a32_types.h"
 #include "dynarmic/frontend/A64/a64_types.h"
 #include "dynarmic/ir/cond.h"
@@ -27,8 +25,7 @@ namespace Dynarmic::IR {
 Block::Block(const LocationDescriptor& location)
     : location{location},
     end_location{location},
-    cond{Cond::AL},
-    instruction_alloc_pool{std::make_unique<std::remove_reference_t<decltype(*instruction_alloc_pool)>>()}
+    cond{Cond::AL}
 {
 
 }
@@ -40,7 +37,21 @@ Block::Block(const LocationDescriptor& location)
 /// @param args            A sequence of Value instances used as arguments for the instruction.
 /// @returns Iterator to the newly created instruction.
 Block::iterator Block::PrependNewInst(iterator insertion_point, Opcode opcode, std::initializer_list<Value> args) noexcept {
-    IR::Inst* inst = new (instruction_alloc_pool->Alloc()) IR::Inst(opcode);
+    // First try using the "inline" buffer, otherwise fallback to a slower slab-like allocation scheme
+    // purpouse is to avoid many calls to new/delete which invoke malloc which invokes mmap
+    // just pool it!!! - reason why there is an inline buffer is because many small blocks are created
+    // with few instructions due to subpar optimisations on other passes... plus branch-heavy code will
+    // hugely benefit from the coherency of faster allocations...
+    IR::Inst* inst;
+    if (inlined_inst.size() < inlined_inst.max_size()) {
+        inst = &inlined_inst[inlined_inst.size()];
+        inlined_inst.emplace_back(opcode);
+    } else {
+        if (pooled_inst.empty() || pooled_inst.back().size() == pooled_inst.back().max_size())
+            pooled_inst.emplace_back();
+        inst = &pooled_inst.back()[pooled_inst.back().size()];
+        pooled_inst.back().emplace_back(opcode);
+    }
     DEBUG_ASSERT(args.size() == inst->NumArgs());
     std::for_each(args.begin(), args.end(), [&inst, index = size_t(0)](const auto& arg) mutable {
         inst->SetArg(index, arg);

src/dynarmic/src/dynarmic/ir/basic_block.h

@@ -13,6 +13,9 @@
 #include <optional>
 #include <string>
 
+#include <boost/container/container_fwd.hpp>
+#include <boost/container/static_vector.hpp>
+#include <boost/container/stable_vector.hpp>
 #include <mcl/container/intrusive_list.hpp>
 #include "dynarmic/common/common_types.h"
 
@@ -21,7 +24,6 @@
 #include "dynarmic/ir/terminal.h"
 #include "dynarmic/ir/value.h"
 #include "dynarmic/ir/dense_list.h"
-#include "dynarmic/common/memory_pool.h"
 
 namespace Dynarmic::IR {
 
@@ -164,8 +166,12 @@ public:
         return cycle_count;
     }
 private:
+    /// "Hot cache" for small blocks so we don't call global allocator
+    boost::container::static_vector<Inst, 14> inlined_inst;
     /// List of instructions in this block.
     instruction_list_type instructions;
+    /// "Long/far" memory pool
+    boost::container::stable_vector<boost::container::static_vector<Inst, 32>> pooled_inst;
     /// Block to execute next if `cond` did not pass.
     std::optional<LocationDescriptor> cond_failed = {};
     /// Description of the starting location of this block
@@ -174,8 +180,6 @@ private:
     LocationDescriptor end_location;
     /// Conditional to pass in order to execute this block
     Cond cond;
-    /// Memory pool for instruction list
-    std::unique_ptr<Common::Pool<sizeof(Inst), 2097152UL / sizeof(Inst)>> instruction_alloc_pool;
     /// Terminal instruction of this block.
     Terminal terminal = Term::Invalid{};
     /// Number of cycles this block takes to execute if the conditional fails.
@@ -183,6 +187,7 @@ private:
     /// Number of cycles this block takes to execute.
     size_t cycle_count = 0;
 };
+static_assert(sizeof(Block) == 2048);
 
 /// Returns a string representation of the contents of block. Intended for debugging.
 std::string DumpBlock(const IR::Block& block) noexcept;

src/dynarmic/src/dynarmic/ir/ir_emitter.cpp

@@ -1,21 +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 "dynarmic/ir/ir_emitter.h"
-
-#include <vector>
-
-#include "dynarmic/common/assert.h"
-#include <mcl/bit_cast.hpp>
-
-#include "dynarmic/ir/opcodes.h"
-
-namespace Dynarmic::IR {
-
-
-}  // namespace Dynarmic::IR

src/dynarmic/tests/decoder_tests.cpp

@@ -57,7 +57,7 @@ TEST_CASE("ASIMD Decoder: Ensure table order correctness", "[decode][a32][.]") {
             INFO("Instruction: " << std::hex << std::setfill('0') << std::setw(8) << instruction);
             INFO("Expect:      " << std::hex << std::setfill('0') << std::setw(8) << expect);
             INFO("Fill:        " << std::hex << std::setfill('0') << std::setw(8) << x);
-            //INFO("Name:        " << *A32::GetNameASIMD<A32::TranslatorVisitor>(instruction));
+            INFO("Name:        " << *A32::GetNameASIMD<A32::TranslatorVisitor>(instruction));
             INFO("iserr:       " << iserr);
             //INFO("alternative: " << alternative->GetName());
             INFO("altiserr:    " << altiserr);

src/dynarmic/tests/print_info.cpp

@@ -40,18 +40,18 @@
 using namespace Dynarmic;
 
 std::string_view GetNameOfA32Instruction(u32 instruction) {
-    //if (auto const vfp_decoder = A32::DecodeVFP<A32::TranslatorVisitor>(instruction))
-    //    return *A32::GetNameVFP<A32::TranslatorVisitor>(instruction);
-    //else if (auto const asimd_decoder = A32::DecodeASIMD<A32::TranslatorVisitor>(instruction))
-    //    return *A32::GetNameASIMD<A32::TranslatorVisitor>(instruction);
-    //else if (auto const decoder = A32::DecodeArm<A32::TranslatorVisitor>(instruction))
-    //    return *A32::GetNameARM<A32::TranslatorVisitor>(instruction);
+    if (auto const vfp_decoder = A32::DecodeVFP<A32::TranslatorVisitor>(instruction))
+        return *A32::GetNameVFP<A32::TranslatorVisitor>(instruction);
+    else if (auto const asimd_decoder = A32::DecodeASIMD<A32::TranslatorVisitor>(instruction))
+        return *A32::GetNameASIMD<A32::TranslatorVisitor>(instruction);
+    else if (auto const decoder = A32::DecodeArm<A32::TranslatorVisitor>(instruction))
+        return *A32::GetNameARM<A32::TranslatorVisitor>(instruction);
     return "<null>";
 }
 
 std::string_view GetNameOfA64Instruction(u32 instruction) {
-    //if (auto const decoder = A64::Decode<A64::TranslatorVisitor>(instruction))
-    //    return *A64::GetName<A64::TranslatorVisitor>(instruction);
+    if (auto const decoder = A64::Decode<A64::TranslatorVisitor>(instruction))
+        return *A64::GetName<A64::TranslatorVisitor>(instruction);
     return "<null>";
 }