Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(280)

Unified Diff: src/runtime/runtime-atomics.cc

Issue 1318713007: [Atomics] Remove support for atomic accesses on floating-point values. (Closed) Base URL: https://chromium.googlesource.com/v8/v8.git@master
Patch Set: feedback Created 5 years, 3 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « src/harmony-atomics.js ('k') | test/mjsunit/asm/atomics-compareexchange.js » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: src/runtime/runtime-atomics.cc
diff --git a/src/runtime/runtime-atomics.cc b/src/runtime/runtime-atomics.cc
index bdb7ae7e62ef219430a630066083737541e01b06..84eab2ce11321943b1e074da5798e019a6e59230 100644
--- a/src/runtime/runtime-atomics.cc
+++ b/src/runtime/runtime-atomics.cc
@@ -12,33 +12,17 @@
// Implement Atomic accesses to SharedArrayBuffers as defined in the
// SharedArrayBuffer draft spec, found here
-// https://docs.google.com/document/d/1NDGA_gZJ7M7w1Bh8S0AoDyEqwDdRh4uSoTPSNn77PFk
+// https://github.com/lars-t-hansen/ecmascript_sharedmem
namespace v8 {
namespace internal {
namespace {
-// Assume that 32-bit architectures don't have 64-bit atomic ops.
-// TODO(binji): can we do better here?
-#if V8_TARGET_ARCH_64_BIT && V8_HOST_ARCH_64_BIT
-
-#define ATOMICS_REQUIRE_LOCK_64_BIT 0
-
-inline bool AtomicIsLockFree(uint32_t size) {
- return size == 1 || size == 2 || size == 4 || size == 8;
-}
-
-#else
-
-#define ATOMICS_REQUIRE_LOCK_64_BIT 1
-
inline bool AtomicIsLockFree(uint32_t size) {
return size == 1 || size == 2 || size == 4;
}
-#endif
-
#if V8_CC_GNU
template <typename T>
@@ -90,37 +74,6 @@ inline T ExchangeSeqCst(T* p, T value) {
return __atomic_exchange_n(p, value, __ATOMIC_SEQ_CST);
}
-#if ATOMICS_REQUIRE_LOCK_64_BIT
-
-// We only need to implement the following functions, because the rest of the
-// atomic operations only work on integer types, and the only 64-bit type is
-// float64. Similarly, because the values are being bit_cast from double ->
-// uint64_t, we don't need to implement these functions for int64_t either.
-
-static base::LazyMutex atomic_mutex = LAZY_MUTEX_INITIALIZER;
-
-inline uint64_t CompareExchangeSeqCst(uint64_t* p, uint64_t oldval,
- uint64_t newval) {
- base::LockGuard<base::Mutex> lock_guard(atomic_mutex.Pointer());
- uint64_t result = *p;
- if (result == oldval) *p = newval;
- return result;
-}
-
-
-inline uint64_t LoadSeqCst(uint64_t* p) {
- base::LockGuard<base::Mutex> lock_guard(atomic_mutex.Pointer());
- return *p;
-}
-
-
-inline void StoreSeqCst(uint64_t* p, uint64_t value) {
- base::LockGuard<base::Mutex> lock_guard(atomic_mutex.Pointer());
- *p = value;
-}
-
-#endif // ATOMICS_REQUIRE_LOCK_64_BIT
-
#elif V8_CC_MSVC
#define InterlockedCompareExchange32 _InterlockedCompareExchange
@@ -133,33 +86,32 @@ inline void StoreSeqCst(uint64_t* p, uint64_t value) {
#define InterlockedCompareExchange8 _InterlockedCompareExchange8
#define InterlockedExchangeAdd8 _InterlockedExchangeAdd8
-#define ATOMIC_OPS_INTEGER(type, suffix, vctype) \
- inline type AddSeqCst(type* p, type value) { \
- return InterlockedExchangeAdd##suffix(reinterpret_cast<vctype*>(p), \
- bit_cast<vctype>(value)); \
- } \
- inline type SubSeqCst(type* p, type value) { \
- return InterlockedExchangeAdd##suffix(reinterpret_cast<vctype*>(p), \
- -bit_cast<vctype>(value)); \
- } \
- inline type AndSeqCst(type* p, type value) { \
- return InterlockedAnd##suffix(reinterpret_cast<vctype*>(p), \
- bit_cast<vctype>(value)); \
- } \
- inline type OrSeqCst(type* p, type value) { \
- return InterlockedOr##suffix(reinterpret_cast<vctype*>(p), \
- bit_cast<vctype>(value)); \
- } \
- inline type XorSeqCst(type* p, type value) { \
- return InterlockedXor##suffix(reinterpret_cast<vctype*>(p), \
- bit_cast<vctype>(value)); \
- } \
- inline type ExchangeSeqCst(type* p, type value) { \
- return InterlockedExchange##suffix(reinterpret_cast<vctype*>(p), \
- bit_cast<vctype>(value)); \
- }
-
-#define ATOMIC_OPS_FLOAT(type, suffix, vctype) \
+#define ATOMIC_OPS(type, suffix, vctype) \
+ inline type AddSeqCst(type* p, type value) { \
+ return InterlockedExchangeAdd##suffix(reinterpret_cast<vctype*>(p), \
+ bit_cast<vctype>(value)); \
+ } \
+ inline type SubSeqCst(type* p, type value) { \
+ return InterlockedExchangeAdd##suffix(reinterpret_cast<vctype*>(p), \
+ -bit_cast<vctype>(value)); \
+ } \
+ inline type AndSeqCst(type* p, type value) { \
+ return InterlockedAnd##suffix(reinterpret_cast<vctype*>(p), \
+ bit_cast<vctype>(value)); \
+ } \
+ inline type OrSeqCst(type* p, type value) { \
+ return InterlockedOr##suffix(reinterpret_cast<vctype*>(p), \
+ bit_cast<vctype>(value)); \
+ } \
+ inline type XorSeqCst(type* p, type value) { \
+ return InterlockedXor##suffix(reinterpret_cast<vctype*>(p), \
+ bit_cast<vctype>(value)); \
+ } \
+ inline type ExchangeSeqCst(type* p, type value) { \
+ return InterlockedExchange##suffix(reinterpret_cast<vctype*>(p), \
+ bit_cast<vctype>(value)); \
+ } \
+ \
inline type CompareExchangeSeqCst(type* p, type oldval, type newval) { \
return InterlockedCompareExchange##suffix(reinterpret_cast<vctype*>(p), \
bit_cast<vctype>(newval), \
@@ -171,20 +123,14 @@ inline void StoreSeqCst(uint64_t* p, uint64_t value) {
bit_cast<vctype>(value)); \
}
-#define ATOMIC_OPS(type, suffix, vctype) \
- ATOMIC_OPS_INTEGER(type, suffix, vctype) \
- ATOMIC_OPS_FLOAT(type, suffix, vctype)
-
ATOMIC_OPS(int8_t, 8, char)
ATOMIC_OPS(uint8_t, 8, char)
ATOMIC_OPS(int16_t, 16, short) /* NOLINT(runtime/int) */
ATOMIC_OPS(uint16_t, 16, short) /* NOLINT(runtime/int) */
ATOMIC_OPS(int32_t, 32, long) /* NOLINT(runtime/int) */
ATOMIC_OPS(uint32_t, 32, long) /* NOLINT(runtime/int) */
-ATOMIC_OPS_FLOAT(uint64_t, 64, LONGLONG)
#undef ATOMIC_OPS_INTEGER
-#undef ATOMIC_OPS_FLOAT
#undef ATOMIC_OPS
#undef InterlockedCompareExchange32
@@ -216,46 +162,16 @@ inline int32_t FromObject<int32_t>(Handle<Object> number) {
return NumberToInt32(*number);
}
-template <>
-inline float FromObject<float>(Handle<Object> number) {
- return static_cast<float>(number->Number());
-}
-
-template <>
-inline double FromObject<double>(Handle<Object> number) {
- return number->Number();
-}
-
template <typename T, typename F>
inline T ToAtomic(F from) {
return static_cast<T>(from);
}
-template <>
-inline uint32_t ToAtomic<uint32_t, float>(float from) {
- return bit_cast<uint32_t, float>(from);
-}
-
-template <>
-inline uint64_t ToAtomic<uint64_t, double>(double from) {
- return bit_cast<uint64_t, double>(from);
-}
-
template <typename T, typename F>
inline T FromAtomic(F from) {
return static_cast<T>(from);
}
-template <>
-inline float FromAtomic<float, uint32_t>(uint32_t from) {
- return bit_cast<float, uint32_t>(from);
-}
-
-template <>
-inline double FromAtomic<double, uint64_t>(uint64_t from) {
- return bit_cast<double, uint64_t>(from);
-}
-
template <typename T>
inline Object* ToObject(Isolate* isolate, T t);
@@ -289,16 +205,6 @@ inline Object* ToObject<uint32_t>(Isolate* isolate, uint32_t t) {
return *isolate->factory()->NewNumber(t);
}
-template <>
-inline Object* ToObject<float>(Isolate* isolate, float t) {
- return *isolate->factory()->NewNumber(t);
-}
-
-template <>
-inline Object* ToObject<double>(Isolate* isolate, double t) {
- return *isolate->factory()->NewNumber(t);
-}
-
template <typename T>
struct FromObjectTraits {};
@@ -338,18 +244,6 @@ struct FromObjectTraits<uint32_t> {
typedef uint32_t atomic_type;
};
-template <>
-struct FromObjectTraits<float> {
- typedef float convert_type;
- typedef uint32_t atomic_type;
-};
-
-template <>
-struct FromObjectTraits<double> {
- typedef double convert_type;
- typedef uint64_t atomic_type;
-};
-
template <typename T>
inline Object* DoCompareExchange(Isolate* isolate, void* buffer, size_t index,
@@ -560,12 +454,6 @@ RUNTIME_FUNCTION(Runtime_AtomicsCompareExchange) {
INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
#undef TYPED_ARRAY_CASE
- case kExternalFloat32Array:
- return DoCompareExchange<float>(isolate, buffer, index, oldobj, newobj);
-
- case kExternalFloat64Array:
- return DoCompareExchange<double>(isolate, buffer, index, oldobj, newobj);
-
case kExternalUint8ClampedArray:
return DoCompareExchangeUint8Clamped(isolate, buffer, index, oldobj,
newobj);
@@ -594,9 +482,12 @@ RUNTIME_FUNCTION(Runtime_AtomicsLoad) {
case kExternal##Type##Array: \
return DoLoad<ctype>(isolate, buffer, index);
- TYPED_ARRAYS(TYPED_ARRAY_CASE)
+ INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
#undef TYPED_ARRAY_CASE
+ case kExternalUint8ClampedArray:
+ return DoLoad<uint8_t>(isolate, buffer, index);
+
default:
break;
}
@@ -625,12 +516,6 @@ RUNTIME_FUNCTION(Runtime_AtomicsStore) {
INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
#undef TYPED_ARRAY_CASE
- case kExternalFloat32Array:
- return DoStore<float>(isolate, buffer, index, value);
-
- case kExternalFloat64Array:
- return DoStore<double>(isolate, buffer, index, value);
-
case kExternalUint8ClampedArray:
return DoStoreUint8Clamped(isolate, buffer, index, value);
@@ -665,8 +550,6 @@ RUNTIME_FUNCTION(Runtime_AtomicsAdd) {
case kExternalUint8ClampedArray:
return DoAddUint8Clamped(isolate, buffer, index, value);
- case kExternalFloat32Array:
- case kExternalFloat64Array:
default:
break;
}
@@ -698,8 +581,6 @@ RUNTIME_FUNCTION(Runtime_AtomicsSub) {
case kExternalUint8ClampedArray:
return DoSubUint8Clamped(isolate, buffer, index, value);
- case kExternalFloat32Array:
- case kExternalFloat64Array:
default:
break;
}
@@ -731,8 +612,6 @@ RUNTIME_FUNCTION(Runtime_AtomicsAnd) {
case kExternalUint8ClampedArray:
return DoAndUint8Clamped(isolate, buffer, index, value);
- case kExternalFloat32Array:
- case kExternalFloat64Array:
default:
break;
}
@@ -764,8 +643,6 @@ RUNTIME_FUNCTION(Runtime_AtomicsOr) {
case kExternalUint8ClampedArray:
return DoOrUint8Clamped(isolate, buffer, index, value);
- case kExternalFloat32Array:
- case kExternalFloat64Array:
default:
break;
}
@@ -797,8 +674,6 @@ RUNTIME_FUNCTION(Runtime_AtomicsXor) {
case kExternalUint8ClampedArray:
return DoXorUint8Clamped(isolate, buffer, index, value);
- case kExternalFloat32Array:
- case kExternalFloat64Array:
default:
break;
}
@@ -830,8 +705,6 @@ RUNTIME_FUNCTION(Runtime_AtomicsExchange) {
case kExternalUint8ClampedArray:
return DoExchangeUint8Clamped(isolate, buffer, index, value);
- case kExternalFloat32Array:
- case kExternalFloat64Array:
default:
break;
}
« no previous file with comments | « src/harmony-atomics.js ('k') | test/mjsunit/asm/atomics-compareexchange.js » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698