Merge pull request #11881 from JosJuice/aarch64-function-call

JitArm64: Add utility for calling a function with arguments

12 files changed, 411 insertions, 99 deletions

diff --git a/Source/Core/Common/Arm64Emitter.cpp b/Source/Core/Common/Arm64Emitter.cpp
index 296b9976c7..93a05fc144 100644
--- a/Source/Core/Common/Arm64Emitter.cpp
+++ b/Source/Core/Common/Arm64Emitter.cpp
@@ -1795,6 +1795,62 @@ void ARM64XEmitter::ADRP(ARM64Reg Rd, s64 imm)
   EncodeAddressInst(1, Rd, static_cast<s32>(imm >> 12));
 }
 
+// This is using a hand-rolled algorithm. The goal is zero memory allocations, not necessarily
+// the best JIT-time time complexity. (The number of moves is usually very small.)
+void ARM64XEmitter::ParallelMoves(RegisterMove* begin, RegisterMove* end,
+                                  std::array<u8, 32>* source_gpr_usages)
+{
+  // X0-X7 are used for passing arguments.
+  // X18-X31 are either callee saved or used for special purposes.
+  constexpr size_t temp_reg_begin = 8;
+  constexpr size_t temp_reg_end = 18;
+
+  while (begin != end)
+  {
+    bool removed_moves_during_this_loop_iteration = false;
+
+    RegisterMove* move = end;
+    while (move != begin)
+    {
+      RegisterMove* prev_move = move;
+      --move;
+      if ((*source_gpr_usages)[DecodeReg(move->dst)] == 0)
+      {
+        MOV(move->dst, move->src);
+        (*source_gpr_usages)[DecodeReg(move->src)]--;
+        std::move(prev_move, end, move);
+        --end;
+        removed_moves_during_this_loop_iteration = true;
+      }
+    }
+
+    if (!removed_moves_during_this_loop_iteration)
+    {
+      // We need to break a cycle using a temporary register.
+
+      size_t temp_reg = temp_reg_begin;
+      while ((*source_gpr_usages)[temp_reg] != 0)
+      {
+        ++temp_reg;
+        ASSERT_MSG(COMMON, temp_reg != temp_reg_end, "Out of registers");
+      }
+
+      const ARM64Reg src = begin->src;
+      const ARM64Reg dst =
+          (Is64Bit(src) ? EncodeRegTo64 : EncodeRegTo32)(static_cast<ARM64Reg>(temp_reg));
+
+      MOV(dst, src);
+      (*source_gpr_usages)[DecodeReg(dst)] = (*source_gpr_usages)[DecodeReg(src)];
+      (*source_gpr_usages)[DecodeReg(src)] = 0;
+
+      std::for_each(begin, end, [src, dst](RegisterMove& move) {
+        if (move.src == src)
+          move.src = dst;
+      });
+    }
+  }
+}
+
 template <typename T>
 void ARM64XEmitter::MOVI2RImpl(ARM64Reg Rd, T imm)
 {
diff --git a/Source/Core/Common/Arm64Emitter.h b/Source/Core/Common/Arm64Emitter.h
index 00691dc647..e8bee81f48 100644
--- a/Source/Core/Common/Arm64Emitter.h
+++ b/Source/Core/Common/Arm64Emitter.h
@@ -3,10 +3,12 @@
 
 #pragma once
 
+#include <array>
 #include <bit>
 #include <cstring>
 #include <functional>
 #include <optional>
+#include <type_traits>
 #include <utility>
 
 #include "Common/ArmCommon.h"
@@ -17,6 +19,7 @@
 #include "Common/Common.h"
 #include "Common/CommonTypes.h"
 #include "Common/MathUtil.h"
+#include "Common/SmallVector.h"
 
 namespace Arm64Gen
 {
@@ -599,6 +602,12 @@ class ARM64XEmitter
   friend class ARM64FloatEmitter;
 
 private:
+  struct RegisterMove
+  {
+    ARM64Reg dst;
+    ARM64Reg src;
+  };
+
   // Pointer to memory where code will be emitted to.
   u8* m_code = nullptr;
 
@@ -646,6 +655,10 @@ private:
 
   [[nodiscard]] FixupBranch WriteFixupBranch();
 
+  // This function solves the "parallel moves" problem common in compilers.
+  // The arguments are mutated!
+  void ParallelMoves(RegisterMove* begin, RegisterMove* end, std::array<u8, 32>* source_gpr_usages);
+
   template <typename T>
   void MOVI2RImpl(ARM64Reg Rd, T imm);
 
@@ -1058,6 +1071,114 @@ public:
   void ABI_PushRegisters(BitSet32 registers);
   void ABI_PopRegisters(BitSet32 registers, BitSet32 ignore_mask = BitSet32(0));
 
+  // Plain function call
+  void QuickCallFunction(ARM64Reg scratchreg, const void* func);
+  template <typename T>
+  void QuickCallFunction(ARM64Reg scratchreg, T func)
+  {
+    QuickCallFunction(scratchreg, (const void*)func);
+  }
+
+  template <typename FuncRet, typename... FuncArgs, typename... Args>
+  void ABI_CallFunction(FuncRet (*func)(FuncArgs...), Args... args)
+  {
+    static_assert(sizeof...(FuncArgs) == sizeof...(Args), "Wrong number of arguments");
+    static_assert(sizeof...(FuncArgs) <= 8, "Passing arguments on the stack is not supported");
+
+    if constexpr (!std::is_void_v<FuncRet>)
+      static_assert(sizeof(FuncRet) <= 16, "Large return types are not supported");
+
+    std::array<u8, 32> source_gpr_uses{};
+
+    auto check_argument = [&](auto& arg) {
+      using Arg = std::decay_t<decltype(arg)>;
+
+      if constexpr (std::is_same_v<Arg, ARM64Reg>)
+      {
+        ASSERT(IsGPR(arg));
+        source_gpr_uses[DecodeReg(arg)]++;
+      }
+      else
+      {
+        // To be more correct, we should be checking FuncArgs here rather than Args, but that's a
+        // lot more effort to implement. Let's just do these best-effort checks for now.
+        static_assert(!std::is_floating_point_v<Arg>, "Floating-point arguments are not supported");
+        static_assert(sizeof(Arg) <= 8, "Arguments bigger than a register are not supported");
+      }
+    };
+
+    (check_argument(args), ...);
+
+    {
+      Common::SmallVector<RegisterMove, sizeof...(Args)> pending_moves;
+
+      size_t i = 0;
+
+      auto handle_register_argument = [&](auto& arg) {
+        using Arg = std::decay_t<decltype(arg)>;
+
+        if constexpr (std::is_same_v<Arg, ARM64Reg>)
+        {
+          const ARM64Reg dst_reg =
+              (Is64Bit(arg) ? EncodeRegTo64 : EncodeRegTo32)(static_cast<ARM64Reg>(i));
+
+          if (dst_reg == arg)
+          {
+            // The value is already in the right register.
+            source_gpr_uses[DecodeReg(arg)]--;
+          }
+          else if (source_gpr_uses[i] == 0)
+          {
+            // The destination register isn't used as the source of another move.
+            // We can go ahead and do the move right away.
+            MOV(dst_reg, arg);
+            source_gpr_uses[DecodeReg(arg)]--;
+          }
+          else
+          {
+            // The destination register is used as the source of a move we haven't gotten to yet.
+            // Let's record that we need to deal with this move later.
+            pending_moves.emplace_back(dst_reg, arg);
+          }
+        }
+
+        ++i;
+      };
+
+      (handle_register_argument(args), ...);
+
+      if (!pending_moves.empty())
+      {
+        ParallelMoves(pending_moves.data(), pending_moves.data() + pending_moves.size(),
+                      &source_gpr_uses);
+      }
+    }
+
+    {
+      size_t i = 0;
+
+      auto handle_immediate_argument = [&](auto& arg) {
+        using Arg = std::decay_t<decltype(arg)>;
+
+        if constexpr (!std::is_same_v<Arg, ARM64Reg>)
+        {
+          const ARM64Reg dst_reg =
+              (sizeof(arg) == 8 ? EncodeRegTo64 : EncodeRegTo32)(static_cast<ARM64Reg>(i));
+          if constexpr (std::is_pointer_v<Arg>)
+            MOVP2R(dst_reg, arg);
+          else
+            MOVI2R(dst_reg, arg);
+        }
+
+        ++i;
+      };
+
+      (handle_immediate_argument(args), ...);
+    }
+
+    QuickCallFunction(ARM64Reg::X8, func);
+  }
+
   // Utility to generate a call to a std::function object.
   //
   // Unfortunately, calling operator() directly is undefined behavior in C++
@@ -1069,23 +1190,11 @@ public:
     return (*f)(args...);
   }
 
-  // This function expects you to have set up the state.
-  // Overwrites X0 and X8
-  template <typename T, typename... Args>
-  ARM64Reg ABI_SetupLambda(const std::function<T(Args...)>* f)
-  {
-    auto trampoline = &ARM64XEmitter::CallLambdaTrampoline<T, Args...>;
-    MOVP2R(ARM64Reg::X8, trampoline);
-    MOVP2R(ARM64Reg::X0, const_cast<void*>((const void*)f));
-    return ARM64Reg::X8;
-  }
-
-  // Plain function call
-  void QuickCallFunction(ARM64Reg scratchreg, const void* func);
-  template <typename T>
-  void QuickCallFunction(ARM64Reg scratchreg, T func)
+  template <typename FuncRet, typename... FuncArgs, typename... Args>
+  void ABI_CallLambdaFunction(const std::function<FuncRet(FuncArgs...)>* f, Args... args)
   {
-    QuickCallFunction(scratchreg, (const void*)func);
+    auto trampoline = &ARM64XEmitter::CallLambdaTrampoline<FuncRet, FuncArgs...>;
+    ABI_CallFunction(trampoline, f, args...);
   }
 };
 
diff --git a/Source/Core/Common/SmallVector.h b/Source/Core/Common/SmallVector.h
index 09559ed21a..c7018f4741 100644
--- a/Source/Core/Common/SmallVector.h
+++ b/Source/Core/Common/SmallVector.h
@@ -29,9 +29,11 @@ public:
   T& operator[](size_t i) { return m_array[i]; }
   const T& operator[](size_t i) const { return m_array[i]; }
 
+  auto data() { return m_array.data(); }
   auto begin() { return m_array.begin(); }
   auto end() { return m_array.begin() + m_size; }
 
+  auto data() const { return m_array.data(); }
   auto begin() const { return m_array.begin(); }
   auto end() const { return m_array.begin() + m_size; }
 
diff --git a/Source/Core/Core/PowerPC/JitArm64/Jit.cpp b/Source/Core/Core/PowerPC/JitArm64/Jit.cpp
index 436229c07b..df782e4de8 100644
--- a/Source/Core/Core/PowerPC/JitArm64/Jit.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/Jit.cpp
@@ -199,10 +199,7 @@ void JitArm64::FallBackToInterpreter(UGeckoInstruction inst)
   }
 
   Interpreter::Instruction instr = Interpreter::GetInterpreterOp(inst);
-  MOVP2R(ARM64Reg::X8, instr);
-  MOVP2R(ARM64Reg::X0, &m_system.GetInterpreter());
-  MOVI2R(ARM64Reg::W1, inst.hex);
-  BLR(ARM64Reg::X8);
+  ABI_CallFunction(instr, &m_system.GetInterpreter(), inst.hex);
 
   // If the instruction wrote to any registers which were marked as discarded,
   // we must mark them as no longer discarded
@@ -250,10 +247,7 @@ void JitArm64::HLEFunction(u32 hook_index)
   gpr.Flush(FlushMode::All, ARM64Reg::INVALID_REG);
   fpr.Flush(FlushMode::All, ARM64Reg::INVALID_REG);
 
-  MOVP2R(ARM64Reg::X8, &HLE::ExecuteFromJIT);
-  MOVI2R(ARM64Reg::W0, js.compilerPC);
-  MOVI2R(ARM64Reg::W1, hook_index);
-  BLR(ARM64Reg::X8);
+  ABI_CallFunction(&HLE::ExecuteFromJIT, js.compilerPC, hook_index);
 }
 
 void JitArm64::DoNothing(UGeckoInstruction inst)
@@ -277,21 +271,15 @@ void JitArm64::Cleanup()
     SUB(ARM64Reg::X0, ARM64Reg::X0, ARM64Reg::X1);
     CMP(ARM64Reg::X0, GPFifo::GATHER_PIPE_SIZE);
     FixupBranch exit = B(CC_LT);
-    MOVP2R(ARM64Reg::X1, &GPFifo::UpdateGatherPipe);
-    MOVP2R(ARM64Reg::X0, &m_system.GetGPFifo());
-    BLR(ARM64Reg::X1);
+    ABI_CallFunction(&GPFifo::UpdateGatherPipe, &m_system.GetGPFifo());
     SetJumpTarget(exit);
   }
 
   // SPEED HACK: MMCR0/MMCR1 should be checked at run-time, not at compile time.
   if (MMCR0(m_ppc_state).Hex || MMCR1(m_ppc_state).Hex)
   {
-    MOVP2R(ARM64Reg::X8, &PowerPC::UpdatePerformanceMonitor);
-    MOVI2R(ARM64Reg::X0, js.downcountAmount);
-    MOVI2R(ARM64Reg::X1, js.numLoadStoreInst);
-    MOVI2R(ARM64Reg::X2, js.numFloatingPointInst);
-    MOVP2R(ARM64Reg::X3, &m_ppc_state);
-    BLR(ARM64Reg::X8);
+    ABI_CallFunction(&PowerPC::UpdatePerformanceMonitor, js.downcountAmount, js.numLoadStoreInst,
+                     js.numFloatingPointInst, &m_ppc_state);
   }
 }
 
@@ -333,10 +321,8 @@ void JitArm64::IntializeSpeculativeConstants()
         fail = GetCodePtr();
         MOVI2R(DISPATCHER_PC, js.blockStart);
         STR(IndexType::Unsigned, DISPATCHER_PC, PPC_REG, PPCSTATE_OFF(pc));
-        MOVP2R(ARM64Reg::X8, &JitInterface::CompileExceptionCheckFromJIT);
-        MOVP2R(ARM64Reg::X0, &m_system.GetJitInterface());
-        MOVI2R(ARM64Reg::W1, static_cast<u32>(JitInterface::ExceptionType::SpeculativeConstants));
-        BLR(ARM64Reg::X8);
+        ABI_CallFunction(&JitInterface::CompileExceptionCheckFromJIT, &m_system.GetJitInterface(),
+                         static_cast<u32>(JitInterface::ExceptionType::SpeculativeConstants));
         B(dispatcher_no_check);
         SwitchToNearCode();
       }
@@ -654,12 +640,10 @@ void JitArm64::WriteExceptionExit(ARM64Reg dest, bool only_external, bool always
   static_assert(PPCSTATE_OFF(pc) + 4 == PPCSTATE_OFF(npc));
   STP(IndexType::Signed, DISPATCHER_PC, DISPATCHER_PC, PPC_REG, PPCSTATE_OFF(pc));
 
-  MOVP2R(ARM64Reg::X0, &m_system.GetPowerPC());
-  if (only_external)
-    MOVP2R(EncodeRegTo64(DISPATCHER_PC), &PowerPC::CheckExternalExceptionsFromJIT);
-  else
-    MOVP2R(EncodeRegTo64(DISPATCHER_PC), &PowerPC::CheckExceptionsFromJIT);
-  BLR(EncodeRegTo64(DISPATCHER_PC));
+  const auto f =
+      only_external ? &PowerPC::CheckExternalExceptionsFromJIT : &PowerPC::CheckExceptionsFromJIT;
+  ABI_CallFunction(f, &m_system.GetPowerPC());
+
   EmitUpdateMembase();
 
   LDR(IndexType::Unsigned, DISPATCHER_PC, PPC_REG, PPCSTATE_OFF(npc));
@@ -1000,10 +984,8 @@ bool JitArm64::DoJit(u32 em_address, JitBlock* b, u32 nextPC)
       SetJumpTarget(fail);
       MOVI2R(DISPATCHER_PC, js.blockStart);
       STR(IndexType::Unsigned, DISPATCHER_PC, PPC_REG, PPCSTATE_OFF(pc));
-      MOVP2R(ARM64Reg::X0, &m_system.GetJitInterface());
-      MOVI2R(ARM64Reg::W1, static_cast<u32>(JitInterface::ExceptionType::PairedQuantize));
-      MOVP2R(ARM64Reg::X2, &JitInterface::CompileExceptionCheckFromJIT);
-      BLR(ARM64Reg::X2);
+      ABI_CallFunction(&JitInterface::CompileExceptionCheckFromJIT, &m_system.GetJitInterface(),
+                       static_cast<u32>(JitInterface::ExceptionType::PairedQuantize));
       B(dispatcher_no_check);
       SwitchToNearCode();
       SetJumpTarget(no_fail);
@@ -1066,9 +1048,7 @@ bool JitArm64::DoJit(u32 em_address, JitBlock* b, u32 nextPC)
 
         ABI_PushRegisters(regs_in_use);
         m_float_emit.ABI_PushRegisters(fprs_in_use, ARM64Reg::X30);
-        MOVP2R(ARM64Reg::X8, &GPFifo::FastCheckGatherPipe);
-        MOVP2R(ARM64Reg::X0, &m_system.GetGPFifo());
-        BLR(ARM64Reg::X8);
+        ABI_CallFunction(&GPFifo::FastCheckGatherPipe, &m_system.GetGPFifo());
         m_float_emit.ABI_PopRegisters(fprs_in_use, ARM64Reg::X30);
         ABI_PopRegisters(regs_in_use);
 
@@ -1184,9 +1164,7 @@ bool JitArm64::DoJit(u32 em_address, JitBlock* b, u32 nextPC)
 
         MOVI2R(DISPATCHER_PC, op.address);
         STP(IndexType::Signed, DISPATCHER_PC, DISPATCHER_PC, PPC_REG, PPCSTATE_OFF(pc));
-        MOVP2R(ARM64Reg::X0, &m_system.GetPowerPC());
-        MOVP2R(ARM64Reg::X1, &PowerPC::CheckBreakPointsFromJIT);
-        BLR(ARM64Reg::X1);
+        ABI_CallFunction(&PowerPC::CheckBreakPointsFromJIT, &m_system.GetPowerPC());
 
         LDR(IndexType::Unsigned, ARM64Reg::W0, ARM64Reg::X0,
             MOVPage2R(ARM64Reg::X0, cpu.GetStatePtr()));
diff --git a/Source/Core/Core/PowerPC/JitArm64/JitArm64_BackPatch.cpp b/Source/Core/Core/PowerPC/JitArm64/JitArm64_BackPatch.cpp
index cb8326439f..4e217fa068 100644
--- a/Source/Core/Core/PowerPC/JitArm64/JitArm64_BackPatch.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/JitArm64_BackPatch.cpp
@@ -211,55 +211,45 @@ void JitArm64::EmitBackpatchRoutine(u32 flags, MemAccessMode mode, ARM64Reg RS,
         src_reg = dst_reg;
       }
 
-      if (dst_reg != src_reg)
-        MOV(dst_reg, src_reg);
-
       const bool reverse = (flags & BackPatchInfo::FLAG_REVERSE) != 0;
 
-      MOVP2R(ARM64Reg::X2, &m_mmu);
-
       if (access_size == 64)
       {
-        MOVP2R(ARM64Reg::X8,
-               reverse ? &PowerPC::WriteU64SwapFromJitArm64 : &PowerPC::WriteU64FromJitArm64);
+        ABI_CallFunction(reverse ? &PowerPC::WriteU64SwapFromJitArm64 :
+                                   &PowerPC::WriteU64FromJitArm64,
+                         src_reg, ARM64Reg::W1, &m_mmu);
       }
       else if (access_size == 32)
       {
-        MOVP2R(ARM64Reg::X8,
-               reverse ? &PowerPC::WriteU32SwapFromJitArm64 : &PowerPC::WriteU32FromJitArm64);
+        ABI_CallFunction(reverse ? &PowerPC::WriteU32SwapFromJitArm64 :
+                                   &PowerPC::WriteU32FromJitArm64,
+                         src_reg, ARM64Reg::W1, &m_mmu);
       }
       else if (access_size == 16)
       {
-        MOVP2R(ARM64Reg::X8,
-               reverse ? &PowerPC::WriteU16SwapFromJitArm64 : &PowerPC::WriteU16FromJitArm64);
+        ABI_CallFunction(reverse ? &PowerPC::WriteU16SwapFromJitArm64 :
+                                   &PowerPC::WriteU16FromJitArm64,
+                         src_reg, ARM64Reg::W1, &m_mmu);
       }
       else
       {
-        MOVP2R(ARM64Reg::X8, &PowerPC::WriteU8FromJitArm64);
+        ABI_CallFunction(&PowerPC::WriteU8FromJitArm64, src_reg, ARM64Reg::W1, &m_mmu);
       }
-
-      BLR(ARM64Reg::X8);
     }
     else if (flags & BackPatchInfo::FLAG_ZERO_256)
     {
-      MOVP2R(ARM64Reg::X1, &m_mmu);
-      MOVP2R(ARM64Reg::X8, &PowerPC::ClearDCacheLineFromJitArm64);
-      BLR(ARM64Reg::X8);
+      ABI_CallFunction(&PowerPC::ClearDCacheLineFromJitArm64, ARM64Reg::W0, &m_mmu);
     }
     else
     {
-      MOVP2R(ARM64Reg::X1, &m_mmu);
-
       if (access_size == 64)
-        MOVP2R(ARM64Reg::X8, &PowerPC::ReadU64FromJitArm64);
+        ABI_CallFunction(&PowerPC::ReadU64FromJitArm64, ARM64Reg::W0, &m_mmu);
       else if (access_size == 32)
-        MOVP2R(ARM64Reg::X8, &PowerPC::ReadU32FromJitArm64);
+        ABI_CallFunction(&PowerPC::ReadU32FromJitArm64, ARM64Reg::W0, &m_mmu);
       else if (access_size == 16)
-        MOVP2R(ARM64Reg::X8, &PowerPC::ReadU16FromJitArm64);
+        ABI_CallFunction(&PowerPC::ReadU16FromJitArm64, ARM64Reg::W0, &m_mmu);
       else
-        MOVP2R(ARM64Reg::X8, &PowerPC::ReadU8FromJitArm64);
-
-      BLR(ARM64Reg::X8);
+        ABI_CallFunction(&PowerPC::ReadU8FromJitArm64, ARM64Reg::W0, &m_mmu);
     }
 
     m_float_emit.ABI_PopRegisters(fprs_to_push, ARM64Reg::X30);
diff --git a/Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp b/Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp
index 288f532757..75c6fe884b 100644
--- a/Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp
@@ -770,13 +770,17 @@ void JitArm64::dcbx(UGeckoInstruction inst)
   ABI_PushRegisters(gprs_to_push);
   m_float_emit.ABI_PushRegisters(fprs_to_push, WA);
 
-  MOVP2R(ARM64Reg::X0, &m_system.GetJitInterface());
-  // effective_address and loop_counter are already in W1 and W2 respectively
+  // For efficiency, effective_addr and loop_counter are already in W1 and W2 respectively
   if (make_loop)
-    MOVP2R(ARM64Reg::X8, &JitInterface::InvalidateICacheLinesFromJIT);
+  {
+    ABI_CallFunction(&JitInterface::InvalidateICacheLinesFromJIT, &m_system.GetJitInterface(),
+                     effective_addr, loop_counter);
+  }
   else
-    MOVP2R(ARM64Reg::X8, &JitInterface::InvalidateICacheLineFromJIT);
-  BLR(ARM64Reg::X8);
+  {
+    ABI_CallFunction(&JitInterface::InvalidateICacheLineFromJIT, &m_system.GetJitInterface(),
+                     effective_addr);
+  }
 
   m_float_emit.ABI_PopRegisters(fprs_to_push, WA);
   ABI_PopRegisters(gprs_to_push);
diff --git a/Source/Core/Core/PowerPC/JitArm64/JitArm64_SystemRegisters.cpp b/Source/Core/Core/PowerPC/JitArm64/JitArm64_SystemRegisters.cpp
index e7cfeb47a5..626d3025d9 100644
--- a/Source/Core/Core/PowerPC/JitArm64/JitArm64_SystemRegisters.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/JitArm64_SystemRegisters.cpp
@@ -81,9 +81,7 @@ void JitArm64::UpdateRoundingMode()
 
   ABI_PushRegisters(gprs_to_save);
   m_float_emit.ABI_PushRegisters(fprs_to_save, ARM64Reg::X8);
-  MOVP2R(ARM64Reg::X0, &m_ppc_state);
-  MOVP2R(ARM64Reg::X8, &PowerPC::RoundingModeUpdated);
-  BLR(ARM64Reg::X8);
+  ABI_CallFunction(&PowerPC::RoundingModeUpdated, &m_ppc_state);
   m_float_emit.ABI_PopRegisters(fprs_to_save, ARM64Reg::X8);
   ABI_PopRegisters(gprs_to_save);
 }
diff --git a/Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp b/Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp
index 77d64de300..9fc224dc08 100644
--- a/Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp
@@ -169,10 +169,7 @@ void JitArm64::GenerateAsm()
 
   // Call JIT
   ResetStack();
-  MOVP2R(ARM64Reg::X0, this);
-  MOV(ARM64Reg::W1, DISPATCHER_PC);
-  MOVP2R(ARM64Reg::X8, reinterpret_cast<void*>(&JitTrampoline));
-  BLR(ARM64Reg::X8);
+  ABI_CallFunction(&JitTrampoline, this, DISPATCHER_PC);
   LDR(IndexType::Unsigned, DISPATCHER_PC, PPC_REG, PPCSTATE_OFF(pc));
   B(dispatcher_no_check);
 
@@ -189,8 +186,7 @@ void JitArm64::GenerateAsm()
   FixupBranch exit = CBNZ(ARM64Reg::W0);
 
   SetJumpTarget(to_start_of_timing_slice);
-  MOVP2R(ARM64Reg::X8, &CoreTiming::GlobalAdvance);
-  BLR(ARM64Reg::X8);
+  ABI_CallFunction(&CoreTiming::GlobalAdvance);
 
   // When we've just entered the jit we need to update the membase
   // GlobalAdvance also checks exceptions after which we need to
diff --git a/Source/Core/Core/PowerPC/JitArm64/Jit_Util.cpp b/Source/Core/Core/PowerPC/JitArm64/Jit_Util.cpp
index 0a82eb0bb8..4beb74ff1b 100644
--- a/Source/Core/Core/PowerPC/JitArm64/Jit_Util.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/Jit_Util.cpp
@@ -78,10 +78,8 @@ private:
 
     m_emit->ABI_PushRegisters(m_gprs_in_use);
     float_emit.ABI_PushRegisters(m_fprs_in_use, ARM64Reg::X1);
-    m_emit->MOVP2R(ARM64Reg::X1, m_system);
-    m_emit->MOVI2R(ARM64Reg::W2, m_address);
-    m_emit->MOV(ARM64Reg::W3, m_src_reg);
-    m_emit->BLR(m_emit->ABI_SetupLambda(lambda));
+
+    m_emit->ABI_CallLambdaFunction(lambda, m_system, m_address, m_src_reg);
 
     float_emit.ABI_PopRegisters(m_fprs_in_use, ARM64Reg::X1);
     m_emit->ABI_PopRegisters(m_gprs_in_use);
@@ -176,9 +174,9 @@ private:
 
     m_emit->ABI_PushRegisters(m_gprs_in_use);
     float_emit.ABI_PushRegisters(m_fprs_in_use, ARM64Reg::X1);
-    m_emit->MOVP2R(ARM64Reg::X1, m_system);
-    m_emit->MOVI2R(ARM64Reg::W2, m_address);
-    m_emit->BLR(m_emit->ABI_SetupLambda(lambda));
+
+    m_emit->ABI_CallLambdaFunction(lambda, m_system, m_address);
+
     if (m_sign_extend)
       m_emit->SBFM(m_dst_reg, ARM64Reg::W0, 0, sbits - 1);
     else
diff --git a/Source/UnitTests/Common/Arm64EmitterTest.cpp b/Source/UnitTests/Common/Arm64EmitterTest.cpp
new file mode 100644
index 0000000000..61be07fef6
--- /dev/null
+++ b/Source/UnitTests/Common/Arm64EmitterTest.cpp
@@ -0,0 +1,178 @@
+// Copyright 2023 Dolphin Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "Common/Arm64Emitter.h"
+#include "Common/BitSet.h"
+#include "Common/BitUtils.h"
+
+#include <gtest/gtest.h>
+
+using namespace Arm64Gen;
+
+namespace
+{
+u32 ZeroParameterFunction()
+{
+  return 123;
+}
+
+u32 OneParameterFunction(u64 a)
+{
+  return a + 23;
+}
+
+u32 TwoParameterFunction(u64 a, u64 b)
+{
+  return a * 10 + b + 3;
+}
+
+u32 ThreeParameterFunction(u64 a, u64 b, u64 c)
+{
+  return a * 10 + b + c / 10;
+}
+
+u32 EightParameterFunction(u64 a, u64 b, u64 c, u64 d, u64 e, u64 f, u64 g, u64 h)
+{
+  return a / 20 + b / 8 + c / 10 + d / 2 + e / 5 - f + g + h / 3;
+}
+
+class TestCallFunction : public ARM64CodeBlock
+{
+public:
+  TestCallFunction() { AllocCodeSpace(4096); }
+
+  template <typename F>
+  void Emit(F f)
+  {
+    ResetCodePtr();
+
+    m_code_pointer = GetCodePtr();
+    {
+      const Common::ScopedJITPageWriteAndNoExecute enable_jit_page_writes;
+
+      constexpr BitSet32 link_register{DecodeReg(ARM64Reg::X30)};
+      ABI_PushRegisters(link_register);
+      f();
+      ABI_PopRegisters(link_register);
+      RET();
+    }
+
+    FlushIcacheSection(const_cast<u8*>(m_code_pointer), const_cast<u8*>(GetCodePtr()));
+  }
+
+  void Run()
+  {
+    const u64 actual = Common::BitCast<u64 (*)()>(m_code_pointer)();
+    constexpr u64 expected = 123;
+    EXPECT_EQ(expected, actual);
+  }
+
+private:
+  const u8* m_code_pointer = nullptr;
+};
+
+}  // namespace
+
+TEST(Arm64Emitter, CallFunction_ZeroParameters)
+{
+  TestCallFunction test;
+  test.Emit([&] { test.ABI_CallFunction(&ZeroParameterFunction); });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_OneConstantParameter)
+{
+  TestCallFunction test;
+  test.Emit([&] { test.ABI_CallFunction(&OneParameterFunction, 100); });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_OneRegisterParameterNoMov)
+{
+  TestCallFunction test;
+  test.Emit([&] {
+    test.MOVI2R(ARM64Reg::X0, 100);
+    test.ABI_CallFunction(&OneParameterFunction, ARM64Reg::X0);
+  });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_OneRegisterParameterMov)
+{
+  TestCallFunction test;
+  test.Emit([&] {
+    test.MOVI2R(ARM64Reg::X8, 100);
+    test.ABI_CallFunction(&OneParameterFunction, ARM64Reg::X8);
+  });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_TwoRegistersMixed)
+{
+  TestCallFunction test;
+  test.Emit([&] {
+    test.MOVI2R(ARM64Reg::X0, 20);
+    test.ABI_CallFunction(&TwoParameterFunction, 10, ARM64Reg::X0);
+  });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_TwoRegistersCycle)
+{
+  TestCallFunction test;
+  test.Emit([&] {
+    test.MOVI2R(ARM64Reg::X0, 20);
+    test.MOVI2R(ARM64Reg::X1, 10);
+    test.ABI_CallFunction(&TwoParameterFunction, ARM64Reg::X1, ARM64Reg::X0);
+  });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_ThreeRegistersMixed)
+{
+  TestCallFunction test;
+  test.Emit([&] {
+    test.MOVI2R(ARM64Reg::X1, 10);
+    test.MOVI2R(ARM64Reg::X2, 20);
+    test.ABI_CallFunction(&ThreeParameterFunction, ARM64Reg::X1, ARM64Reg::X2, 30);
+  });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_ThreeRegistersCycle1)
+{
+  TestCallFunction test;
+  test.Emit([&] {
+    test.MOVI2R(ARM64Reg::X0, 30);
+    test.MOVI2R(ARM64Reg::X1, 10);
+    test.MOVI2R(ARM64Reg::X2, 20);
+    test.ABI_CallFunction(&ThreeParameterFunction, ARM64Reg::X1, ARM64Reg::X2, ARM64Reg::X0);
+  });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_ThreeRegistersCycle2)
+{
+  TestCallFunction test;
+  test.Emit([&] {
+    test.MOVI2R(ARM64Reg::X0, 20);
+    test.MOVI2R(ARM64Reg::X1, 30);
+    test.MOVI2R(ARM64Reg::X2, 10);
+    test.ABI_CallFunction(&ThreeParameterFunction, ARM64Reg::X2, ARM64Reg::X0, ARM64Reg::X1);
+  });
+  test.Run();
+}
+
+TEST(Arm64Emitter, CallFunction_EightRegistersMixed)
+{
+  TestCallFunction test;
+  test.Emit([&] {
+    test.MOVI2R(ARM64Reg::X3, 12);
+    test.MOVI2R(ARM64Reg::X4, 23);
+    test.MOVI2R(ARM64Reg::X5, 24);
+    test.MOVI2R(ARM64Reg::X30, 2000);
+    test.ABI_CallFunction(&EightParameterFunction, ARM64Reg::X30, 40, ARM64Reg::X4, ARM64Reg::X5,
+                          ARM64Reg::X4, ARM64Reg::X3, 5, ARM64Reg::X4);
+  });
+  test.Run();
+}
diff --git a/Source/UnitTests/Common/CMakeLists.txt b/Source/UnitTests/Common/CMakeLists.txt
index bbea892fdf..a4c9a67ade 100644
--- a/Source/UnitTests/Common/CMakeLists.txt
+++ b/Source/UnitTests/Common/CMakeLists.txt
@@ -21,4 +21,6 @@ add_dolphin_test(SwapTest SwapTest.cpp)
 if (_M_X86)
   add_dolphin_test(x64EmitterTest x64EmitterTest.cpp)
   target_link_libraries(x64EmitterTest PRIVATE bdisasm)
+elseif (_M_ARM_64)
+  add_dolphin_test(Arm64EmitterTest Arm64EmitterTest.cpp)
 endif()
diff --git a/Source/UnitTests/UnitTests.vcxproj b/Source/UnitTests/UnitTests.vcxproj
index cb630585f1..0608cbe307 100644
--- a/Source/UnitTests/UnitTests.vcxproj
+++ b/Source/UnitTests/UnitTests.vcxproj
@@ -80,6 +80,7 @@
     <ClCompile Include="Core\PowerPC\Jit64Common\Frsqrte.cpp" />
   </ItemGroup>
   <ItemGroup Condition="'$(Platform)'=='ARM64'">
+    <ClCompile Include="Common\Arm64EmitterTest.cpp" />
     <ClCompile Include="Core\PowerPC\JitArm64\ConvertSingleDouble.cpp" />
     <ClCompile Include="Core\PowerPC\JitArm64\FPRF.cpp" />
     <ClCompile Include="Core\PowerPC\JitArm64\Fres.cpp" />