From 701ba7cd4368939d4b7dd2d2feb11f3a9a51fa92 Mon Sep 17 00:00:00 2001
From: JosJuice <josjuice@gmail.com>
Date: Sun, 20 Nov 2022 14:56:35 +0100
Subject: [PATCH] JitArm64: Improve pipelining of lmw/stmw

The calculation of each address in lmw/stmw currently has a dependency
on the calculation of the previous address. By removing this dependency,
the host CPU should be able to pipeline the loads/stores better. The cost
we pay for this is up to one extra register and one extra MOV instruction
per guest instruction, but often nothing.

Making EmitBackpatchRoutine support using any register as the address
register would let us get rid of the MOV, but I consider that to be too
big of a task to do in one go at the same time as this.
---
 .../PowerPC/JitArm64/JitArm64_LoadStore.cpp   | 72 +++++++++++--------
 1 file changed, 41 insertions(+), 31 deletions(-)

diff --git a/Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp b/Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp
index 6642cec959..f00a700faf 100644
--- a/Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/JitArm64_LoadStore.cpp
@@ -527,19 +527,21 @@ void JitArm64::lmw(UGeckoInstruction inst)
     gpr.Lock(ARM64Reg::W2);
 
   // MMU games make use of a >= d despite this being invalid according to the PEM.
-  // Because of this, make sure to not re-read rA after starting doing the loads.
+  // If a >= d occurs, we must make sure to not re-read rA after starting doing the loads.
   ARM64Reg addr_reg = ARM64Reg::W0;
-  if (a)
-  {
-    if (gpr.IsImm(a))
-      MOVI2R(addr_reg, gpr.GetImm(a) + offset);
-    else
-      ADDI2R(addr_reg, gpr.R(a), offset, addr_reg);
-  }
-  else
-  {
+  bool a_is_addr_base_reg = false;
+  if (!a)
     MOVI2R(addr_reg, offset);
-  }
+  else if (gpr.IsImm(a))
+    MOVI2R(addr_reg, gpr.GetImm(a) + offset);
+  else if (a < d && offset + (31 - d) * 4 < 0x1000)
+    a_is_addr_base_reg = true;
+  else
+    ADDI2R(addr_reg, gpr.R(a), offset, addr_reg);
+
+  ARM64Reg addr_base_reg = a_is_addr_base_reg ? ARM64Reg::INVALID_REG : gpr.GetReg();
+  if (!a_is_addr_base_reg)
+    MOV(addr_base_reg, addr_reg);
 
   // TODO: This doesn't handle rollback on DSI correctly
   constexpr u32 flags = BackPatchInfo::FLAG_LOAD | BackPatchInfo::FLAG_SIZE_32;
@@ -548,12 +550,16 @@ void JitArm64::lmw(UGeckoInstruction inst)
     gpr.BindToRegister(i, false, false);
     ARM64Reg dest_reg = gpr.R(i);
 
+    if (a_is_addr_base_reg)
+      ADDI2R(addr_reg, gpr.R(a), offset + (i - d) * 4);
+    else if (i != d)
+      ADDI2R(addr_reg, addr_base_reg, (i - d) * 4);
+
     BitSet32 regs_in_use = gpr.GetCallerSavedUsed();
     BitSet32 fprs_in_use = fpr.GetCallerSavedUsed();
+    regs_in_use[DecodeReg(addr_reg)] = 0;
     if (!jo.fastmem_arena)
       regs_in_use[DecodeReg(ARM64Reg::W2)] = 0;
-    if (i == 31)
-      regs_in_use[DecodeReg(addr_reg)] = 0;
     if (!jo.memcheck)
       regs_in_use[DecodeReg(dest_reg)] = 0;
 
@@ -562,14 +568,13 @@ void JitArm64::lmw(UGeckoInstruction inst)
 
     gpr.BindToRegister(i, false, true);
     ASSERT(dest_reg == gpr.R(i));
-
-    if (i != 31)
-      ADD(addr_reg, addr_reg, 4);
   }
 
   gpr.Unlock(ARM64Reg::W0, ARM64Reg::W30);
   if (!jo.fastmem_arena)
     gpr.Unlock(ARM64Reg::W2);
+  if (!a_is_addr_base_reg)
+    gpr.Unlock(addr_base_reg);
 }
 
 void JitArm64::stmw(UGeckoInstruction inst)
@@ -585,17 +590,19 @@ void JitArm64::stmw(UGeckoInstruction inst)
     gpr.Lock(ARM64Reg::W2);
 
   ARM64Reg addr_reg = ARM64Reg::W1;
-  if (a)
-  {
-    if (gpr.IsImm(a))
-      MOVI2R(addr_reg, gpr.GetImm(a) + offset);
-    else
-      ADDI2R(addr_reg, gpr.R(a), offset, addr_reg);
-  }
-  else
-  {
+  bool a_is_addr_base_reg = false;
+  if (!a)
     MOVI2R(addr_reg, offset);
-  }
+  else if (gpr.IsImm(a))
+    MOVI2R(addr_reg, gpr.GetImm(a) + offset);
+  else if (offset + (31 - s) * 4 < 0x1000)
+    a_is_addr_base_reg = true;
+  else
+    ADDI2R(addr_reg, gpr.R(a), offset, addr_reg);
+
+  ARM64Reg addr_base_reg = a_is_addr_base_reg ? ARM64Reg::INVALID_REG : gpr.GetReg();
+  if (!a_is_addr_base_reg)
+    MOV(addr_base_reg, addr_reg);
 
   // TODO: This doesn't handle rollback on DSI correctly
   constexpr u32 flags = BackPatchInfo::FLAG_STORE | BackPatchInfo::FLAG_SIZE_32;
@@ -603,24 +610,27 @@ void JitArm64::stmw(UGeckoInstruction inst)
   {
     ARM64Reg src_reg = gpr.R(i);
 
+    if (a_is_addr_base_reg)
+      ADDI2R(addr_reg, gpr.R(a), offset + (i - s) * 4);
+    else if (i != s)
+      ADDI2R(addr_reg, addr_base_reg, (i - s) * 4);
+
     BitSet32 regs_in_use = gpr.GetCallerSavedUsed();
     BitSet32 fprs_in_use = fpr.GetCallerSavedUsed();
     regs_in_use[DecodeReg(ARM64Reg::W0)] = 0;
+    regs_in_use[DecodeReg(addr_reg)] = 0;
     if (!jo.fastmem_arena)
       regs_in_use[DecodeReg(ARM64Reg::W2)] = 0;
-    if (i == 31)
-      regs_in_use[DecodeReg(addr_reg)] = 0;
 
     EmitBackpatchRoutine(flags, MemAccessMode::Auto, src_reg, EncodeRegTo64(addr_reg), regs_in_use,
                          fprs_in_use);
-
-    if (i != 31)
-      ADD(addr_reg, addr_reg, 4);
   }
 
   gpr.Unlock(ARM64Reg::W0, ARM64Reg::W1, ARM64Reg::W30);
   if (!jo.fastmem_arena)
     gpr.Unlock(ARM64Reg::W2);
+  if (!a_is_addr_base_reg)
+    gpr.Unlock(addr_base_reg);
 }
 
 void JitArm64::dcbx(UGeckoInstruction inst)