aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Replace inline assembly in task system with equivalent gcc intrinsics.

Browse Source 
gcc/icc build only: the Windows build still uses the Win32 calls for
these.
This commit is contained in:
Matt Pharr
2013-07-19 10:08:28 -07:00
parent abf43ad01d
commit b007bba59f
1 changed files with 4 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
examples/tasksys.cpp
View File

@@ -331,21 +331,12 @@ TaskGroupBase::AllocMemory(int64_t size, int32_t alignment) {
/////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////
// Atomics and the like // Atomics and the like
#if (__SIZEOF_POINTER__ == 4) || defined(__i386__) || defined(_WIN32)
#define ISPC_POINTER_BYTES 4
#elif (__SIZEOF_POINTER__ == 8) || defined(__x86_64__) || defined(__amd64__) || defined(_WIN64)
#define ISPC_POINTER_BYTES 8
#else
#error "Pointer size unknown!"
#endif // __SIZEOF_POINTER__
static inline void static inline void
lMemFence() { lMemFence() {
// Windows atomic functions already contain the fence // Windows atomic functions already contain the fence
// KNC doesn't need the memory barrier // KNC doesn't need the memory barrier
#if !defined ISPC_IS_KNC && !defined ISPC_IS_WINDOWS #if !defined ISPC_IS_KNC && !defined ISPC_IS_WINDOWS
__asm__ __volatile__("mfence":::"memory"); __sync_synchronize();
#endif #endif
} }
@@ -354,18 +345,7 @@ lAtomicCompareAndSwapPointer(void **v, void *newValue, void *oldValue) {
#ifdef ISPC_IS_WINDOWS #ifdef ISPC_IS_WINDOWS
return InterlockedCompareExchangePointer(v, newValue, oldValue); return InterlockedCompareExchangePointer(v, newValue, oldValue);
#else #else
void *result; void *result = __sync_val_compare_and_swap(v, oldValue, newValue);
#if (ISPC_POINTER_BYTES == 4)
__asm__ __volatile__("lock\ncmpxchgl %2,%1"
: "=a"(result), "=m"(*v)
: "q"(newValue), "0"(oldValue)
: "memory");
#else
__asm__ __volatile__("lock\ncmpxchgq %2,%1"
: "=a"(result), "=m"(*v)
: "q"(newValue), "0"(oldValue)
: "memory");
#endif // ISPC_POINTER_BYTES
lMemFence(); lMemFence();
return result; return result;
#endif // ISPC_IS_WINDOWS #endif // ISPC_IS_WINDOWS
@@ -376,11 +356,7 @@ lAtomicCompareAndSwap32(volatile int32_t *v, int32_t newValue, int32_t oldValue)
#ifdef ISPC_IS_WINDOWS #ifdef ISPC_IS_WINDOWS
return InterlockedCompareExchange((volatile LONG *)v, newValue, oldValue); return InterlockedCompareExchange((volatile LONG *)v, newValue, oldValue);
#else #else
int32_t result; int32_t result = __sync_val_compare_and_swap(v, oldValue, newValue);
__asm__ __volatile__("lock\ncmpxchgl %2,%1"
: "=a"(result), "=m"(*v)
: "q"(newValue), "0"(oldValue)
: "memory");
lMemFence(); lMemFence();
return result; return result;
#endif // ISPC_IS_WINDOWS #endif // ISPC_IS_WINDOWS
@@ -391,12 +367,7 @@ lAtomicAdd(volatile int32_t *v, int32_t delta) {
#ifdef ISPC_IS_WINDOWS #ifdef ISPC_IS_WINDOWS
return InterlockedAdd((volatile LONG *)v, delta); return InterlockedAdd((volatile LONG *)v, delta);
#else #else
int32_t origValue; return __sync_fetch_and_add(v, delta);
__asm__ __volatile__("lock\n"
"xaddl %0,%1"
: "=r"(origValue), "=m"(*v) : "0"(delta)
: "memory");
return origValue;
#endif #endif
} }