[Atomics] use TFJ builtins for atomic add, sub, and, or, and xor

src/compiler/x64/code-generator-x64.cc

 // Copyright 2013 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/compiler/code-generator.h"
 
 #include <limits>
 
 #include "src/compilation-info.h"
 #include "src/compiler/code-generator-impl.h"
@@ skipping 692 matching lines @@
                      2);                                                      \
   } while (false)
 
 #define ASSEMBLE_IEEE754_UNOP(name)                                           \
   do {                                                                        \
     __ PrepareCallCFunction(1);                                               \
     __ CallCFunction(ExternalReference::ieee754_##name##_function(isolate()), \
                      1);                                                      \
   } while (false)
 
+#define ASSEMBLE_ATOMIC_BINOP(bin_inst, mov_inst, cmpxchg_inst) \
+  do {                                                          \
+    Label binop;                                                \
+    __ bind(&binop);                                            \
+    __ mov_inst(rax, i.MemoryOperand(1));                       \
+    __ movl(i.TempRegister(0), rax);                            \
+    __ bin_inst(i.TempRegister(0), i.InputRegister(0));         \
+    __ lock();                                                  \
+    __ cmpxchg_inst(i.MemoryOperand(1), i.TempRegister(0));     \
+    __ j(not_equal, &binop);                                    \
+  } while (false)
+
 void CodeGenerator::AssembleDeconstructFrame() {
   unwinding_info_writer_.MarkFrameDeconstructed(__ pc_offset());
   __ movq(rsp, rbp);
   __ popq(rbp);
 }
 
 void CodeGenerator::AssemblePrepareTailCall() {
   if (frame_access_state()->has_frame()) {
     __ movq(rbp, MemOperand(rbp, 0));
   }
@@ skipping 1600 matching lines @@
       __ lock();
       __ cmpxchgw(i.MemoryOperand(2), i.InputRegister(1));
       __ movzxwl(rax, rax);
       break;
     }
     case kAtomicCompareExchangeWord32: {
       __ lock();
       __ cmpxchgl(i.MemoryOperand(2), i.InputRegister(1));
       break;
     }
+#define ATOMIC_BINOP_CASE(op, inst)              \
+  case kAtomic##op##Int8:                        \
+    ASSEMBLE_ATOMIC_BINOP(inst, movb, cmpxchgb); \
+    __ movsxbl(rax, rax);                        \
+    break;                                       \
+  case kAtomic##op##Uint8:                       \
+    ASSEMBLE_ATOMIC_BINOP(inst, movb, cmpxchgb); \
+    __ movzxbl(rax, rax);                        \
+    break;                                       \
+  case kAtomic##op##Int16:                       \
+    ASSEMBLE_ATOMIC_BINOP(inst, movw, cmpxchgw); \
+    __ movsxwl(rax, rax);                        \
+    break;                                       \
+  case kAtomic##op##Uint16:                      \
+    ASSEMBLE_ATOMIC_BINOP(inst, movw, cmpxchgw); \
+    __ movzxwl(rax, rax);                        \
+    break;                                       \
+  case kAtomic##op##Word32:                      \
+    ASSEMBLE_ATOMIC_BINOP(inst, movl, cmpxchgl); \
+    break;
+      ATOMIC_BINOP_CASE(Add, addl)
+      ATOMIC_BINOP_CASE(Sub, subl)
+      ATOMIC_BINOP_CASE(And, andl)
+      ATOMIC_BINOP_CASE(Or, orl)
+      ATOMIC_BINOP_CASE(Xor, xorl)
+#undef ATOMIC_BINOP_CASE
     case kAtomicLoadInt8:
     case kAtomicLoadUint8:
     case kAtomicLoadInt16:
     case kAtomicLoadUint16:
     case kAtomicLoadWord32:
     case kAtomicStoreWord8:
     case kAtomicStoreWord16:
     case kAtomicStoreWord32:
       UNREACHABLE();  // Won't be generated by instruction selector.
       break;
@@ skipping 620 matching lines @@
     int padding_size = last_lazy_deopt_pc_ + space_needed - current_pc;
     __ Nop(padding_size);
   }
 }
 
 #undef __
 
 }  // namespace compiler
 }  // namespace internal
 }  // namespace v8