Use non-temporal access for streaming data (#341)

* Use non-temporal access for streaming data

* Revert to ulong2 after fixing compiling issue
Šī revīzija ir iekļauta:
Wenkai Du
2021-04-07 17:34:35 -07:00
revīziju iesūtīja GitHub
vecāks caba0a63d2
revīzija 9dfc2c183e
5 mainīti faili ar 23 papildinājumiem un 24 dzēšanām
+5 -4
Parādīt failu
@@ -267,16 +267,17 @@ struct MULTI128 {
inline __device__ void Fetch128(Pack128& v, const Pack128* p) {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
v.x = p->x;
v.y = p->y;
v.x = __builtin_nontemporal_load(&p->x);
v.y = __builtin_nontemporal_load(&p->y);
#else
asm volatile("ld.volatile.global.v2.u64 {%0,%1}, [%2];" : "=l"(v.x), "=l"(v.y) : "l"(p) : "memory");
#endif
}
inline __device__ void Store128(Pack128* p, Pack128& v) {
#if defined(__HIP_PLATFORM_HCC__) || defined(__HCC__) || defined(__HIPCC__)
p->x = v.x;
p->y = v.y;
__builtin_nontemporal_store(v.x, &p->x);
__builtin_nontemporal_store(v.y, &p->y);
#else
asm volatile("st.volatile.global.v2.u64 [%0], {%1,%2};" :: "l"(p), "l"(v.x), "l"(v.y) : "memory");
#endif