From f0802d75fadeca7cb002e294df867d2a0e4fb608 Mon Sep 17 00:00:00 2001
From: "Dr. Chat" <arkolbed@gmail.com>
Date: Wed, 13 Jan 2016 01:45:31 -0600
Subject: [PATCH] Use atomic compare exchange in stdcx/stwcx (still under the
 global lock)

---
 src/xenia/cpu/ppc/ppc_context.h      |  4 ++--
 src/xenia/cpu/ppc/ppc_emit_memory.cc | 12 ++++++++----
 src/xenia/cpu/ppc/ppc_hir_builder.cc |  9 +++++++++
 src/xenia/cpu/ppc/ppc_hir_builder.h  |  3 +++
 4 files changed, 22 insertions(+), 6 deletions(-)

diff --git a/src/xenia/cpu/ppc/ppc_context.h b/src/xenia/cpu/ppc/ppc_context.h
index b37f4cda4..9c96daa6b 100644
--- a/src/xenia/cpu/ppc/ppc_context.h
+++ b/src/xenia/cpu/ppc/ppc_context.h
@@ -423,8 +423,8 @@ typedef struct PPCContext_s {
 
   uint8_t* physical_membase;
 
-  // Keep the struct padded out to 64b total.
-  uint8_t _padding[8];
+  // Value of last reserved load
+  uint64_t reserved_val;
 
   static std::string GetRegisterName(PPCRegister reg);
   std::string GetStringFromValue(PPCRegister reg) const;
diff --git a/src/xenia/cpu/ppc/ppc_emit_memory.cc b/src/xenia/cpu/ppc/ppc_emit_memory.cc
index 8749deb9a..517081c68 100644
--- a/src/xenia/cpu/ppc/ppc_emit_memory.cc
+++ b/src/xenia/cpu/ppc/ppc_emit_memory.cc
@@ -658,6 +658,7 @@ int InstrEmit_ldarx(PPCHIRBuilder& f, const InstrData& i) {
 
   Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
   Value* rt = f.ByteSwap(f.Load(ea, INT64_TYPE));
+  f.StoreReserved(rt);
   f.StoreGPR(i.X.RT, rt);
   return 0;
 }
@@ -682,6 +683,7 @@ int InstrEmit_lwarx(PPCHIRBuilder& f, const InstrData& i) {
 
   Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
   Value* rt = f.ZeroExtend(f.ByteSwap(f.Load(ea, INT32_TYPE)), INT64_TYPE);
+  f.StoreReserved(rt);
   f.StoreGPR(i.X.RT, rt);
   return 0;
 }
@@ -703,8 +705,9 @@ int InstrEmit_stdcx(PPCHIRBuilder& f, const InstrData& i) {
 
   Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
   Value* rt = f.ByteSwap(f.LoadGPR(i.X.RT));
-  f.Store(ea, rt);
-  f.StoreContext(offsetof(PPCContext, cr0.cr0_eq), f.LoadConstantInt8(1));
+  Value* res = f.ByteSwap(f.LoadReserved());
+  Value* v = f.AtomicCompareExchange(ea, res, rt);
+  f.StoreContext(offsetof(PPCContext, cr0.cr0_eq), v);
   f.StoreContext(offsetof(PPCContext, cr0.cr0_lt), f.LoadZeroInt8());
   f.StoreContext(offsetof(PPCContext, cr0.cr0_gt), f.LoadZeroInt8());
 
@@ -732,8 +735,9 @@ int InstrEmit_stwcx(PPCHIRBuilder& f, const InstrData& i) {
 
   Value* ea = CalculateEA_0(f, i.X.RA, i.X.RB);
   Value* rt = f.ByteSwap(f.Truncate(f.LoadGPR(i.X.RT), INT32_TYPE));
-  f.Store(ea, rt);
-  f.StoreContext(offsetof(PPCContext, cr0.cr0_eq), f.LoadConstantInt8(1));
+  Value* res = f.ByteSwap(f.Truncate(f.LoadReserved(), INT32_TYPE));
+  Value* v = f.AtomicCompareExchange(ea, res, rt);
+  f.StoreContext(offsetof(PPCContext, cr0.cr0_eq), v);
   f.StoreContext(offsetof(PPCContext, cr0.cr0_lt), f.LoadZeroInt8());
   f.StoreContext(offsetof(PPCContext, cr0.cr0_gt), f.LoadZeroInt8());
 
diff --git a/src/xenia/cpu/ppc/ppc_hir_builder.cc b/src/xenia/cpu/ppc/ppc_hir_builder.cc
index bce5d9d16..e18118d7b 100644
--- a/src/xenia/cpu/ppc/ppc_hir_builder.cc
+++ b/src/xenia/cpu/ppc/ppc_hir_builder.cc
@@ -511,6 +511,15 @@ void PPCHIRBuilder::StoreVR(uint32_t reg, Value* value) {
   trace_reg.value = value;
 }
 
+void PPCHIRBuilder::StoreReserved(Value* val) {
+  assert_true(val->type == INT64_TYPE);
+  StoreContext(offsetof(PPCContext, reserved_val), val);
+}
+
+Value* PPCHIRBuilder::LoadReserved() {
+  return LoadContext(offsetof(PPCContext, reserved_val), INT64_TYPE);
+}
+
 }  // namespace ppc
 }  // namespace cpu
 }  // namespace xe
diff --git a/src/xenia/cpu/ppc/ppc_hir_builder.h b/src/xenia/cpu/ppc/ppc_hir_builder.h
index ca9830799..8b88ae35d 100644
--- a/src/xenia/cpu/ppc/ppc_hir_builder.h
+++ b/src/xenia/cpu/ppc/ppc_hir_builder.h
@@ -78,6 +78,9 @@ class PPCHIRBuilder : public hir::HIRBuilder {
   Value* LoadVR(uint32_t reg);
   void StoreVR(uint32_t reg, Value* value);
 
+  void StoreReserved(Value* val);
+  Value* LoadReserved();
+
  private:
   void AnnotateLabel(uint32_t address, Label* label);