[dynarmic] improve A32 translate loop (#3780)

Signed-off-by: lizzie <lizzie@eden-emu.dev>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/3780
Reviewed-by: Maufeat <sahyno1996@gmail.com>
Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>
Co-authored-by: lizzie <lizzie@eden-emu.dev>
Co-committed-by: lizzie <lizzie@eden-emu.dev>

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

@@ -121,20 +121,34 @@ A32EmitX64::BlockDescriptor A32EmitX64::Emit(IR::Block& block) {
     code.lea(rbp, code.ptr[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, abi_base_pointer) - 8]);
 
     EmitCondPrelude(ctx);
+    typedef void (EmitX64::*EmitHandlerFn)(EmitContext& context, IR::Inst* inst);
+    constexpr EmitHandlerFn opcode_handlers[] = {
+#define OPCODE(name, type, ...) &EmitX64::Emit##name,
+#define A32OPC(name, type, ...)
+#define A64OPC(name, type, ...)
+#include "dynarmic/ir/opcodes.inc"
+#undef OPCODE
+#undef A32OPC
+#undef A64OPC
+    };
+    typedef void (A32EmitX64::*A32EmitHandlerFn)(A32EmitContext& context, IR::Inst* inst);
+    constexpr A32EmitHandlerFn a32_handlers[] = {
+#define OPCODE(...)
+#define A32OPC(name, type, ...) &A32EmitX64::EmitA32##name,
+#define A64OPC(...)
+#include "dynarmic/ir/opcodes.inc"
+#undef OPCODE
+#undef A32OPC
+#undef A64OPC
+    };
 
-    for (auto iter = block.instructions.begin(); iter != block.instructions.end(); ++iter) [[likely]] {
-        auto* inst = &*iter;
+    for (auto& inst : block.instructions) {
+        auto const opcode = inst.GetOpcode();
         // Call the relevant Emit* member function.
-        switch (inst->GetOpcode()) {
-#define OPCODE(name, type, ...)                     \
-        case IR::Opcode::name:                  \
-            A32EmitX64::Emit##name(ctx, inst);  \
-            break;
-#define A32OPC(name, type, ...)                     \
-        case IR::Opcode::A32##name:             \
-            A32EmitX64::EmitA32##name(ctx, inst);\
-            break;
-#define A64OPC(...)
+        switch (opcode) {
+#define OPCODE(name, type, ...) case IR::Opcode::name: goto opcode_branch;
+#define A32OPC(name, type, ...) case IR::Opcode::A32##name: goto a32_branch;
+#define A64OPC(name, type, ...)
 #include "dynarmic/ir/opcodes.inc"
 #undef OPCODE
 #undef A32OPC
@@ -142,7 +156,14 @@ A32EmitX64::BlockDescriptor A32EmitX64::Emit(IR::Block& block) {
         default:
             UNREACHABLE();
         }
-        reg_alloc.EndOfAllocScope();
+opcode_branch:
+        (this->*opcode_handlers[size_t(opcode)])(ctx, &inst);
+        goto finish_this_inst;
+a32_branch:
+        // Update with FIRST A32 instruction
+        (this->*a32_handlers[size_t(opcode) - size_t(IR::Opcode::A32SetCheckBit)])(ctx, &inst);
+finish_this_inst:
+        ctx.reg_alloc.EndOfAllocScope();
 #ifndef NDEBUG
         if (conf.very_verbose_debugging_output)
             EmitVerboseDebuggingOutput(reg_alloc);

src/dynarmic/src/dynarmic/ir/opcodes.inc

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // First we list common shared opcodes
 // Since we give priority to A64 performance, we include them first, this is so we
 // can discard all A32 opcodes instead of having a "hole" in our checks
@@ -710,6 +713,8 @@ A64OPC(ExclusiveWriteMemory32,                              U32,            U64,
 A64OPC(ExclusiveWriteMemory64,                              U32,            U64,            U64,            U64,            AccType         )
 A64OPC(ExclusiveWriteMemory128,                             U32,            U64,            U64,            U128,           AccType         )
 
+// Remember to update:
+// - a32_emit_x64.cpp
 
 // A32 Context getters/setters
 A32OPC(SetCheckBit,                                         Void,           U1                                                              )