fzorb/REDACTED-rig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

RandomX fixes

Browse source 
Intel JCC erratum fix and various other improvements, see more here: https://www.phoronix.com/scan.php?page=article&item=intel-jcc-microcode&num=1
This commit is contained in:
SChernykh 2019-12-01 08:46:35 +01:00
parent 8791261220
commit 84d7eb05f3
12 changed files with 320 additions and 40 deletions
Download patch file Download diff file Expand all files Collapse all files

7
src/crypto/randomx/intrin_portable.h
View file

@ -102,6 +102,7 @@ typedef __m128d rx_vec_f128;
#define rx_aligned_alloc(a, b) _mm_malloc(a,b)
#define rx_aligned_free(a) _mm_free(a)
#define rx_prefetch_nta(x) _mm_prefetch((const char *)(x), _MM_HINT_NTA)
#define rx_prefetch_t0(x) _mm_prefetch((const char *)(x), _MM_HINT_T0)
#define rx_load_vec_f128 _mm_load_pd
#define rx_store_vec_f128 _mm_store_pd
@ -201,6 +202,7 @@ typedef union{
#define rx_aligned_alloc(a, b) malloc(a)
#define rx_aligned_free(a) free(a)
#define rx_prefetch_nta(x)
#define rx_prefetch_t0(x)
/* Splat 64-bit long long to 2 64-bit long longs */
FORCE_INLINE __m128i vec_splat2sd (int64_t scalar)
@ -399,6 +401,10 @@ inline void rx_prefetch_nta(void* ptr) {
asm volatile ("prfm pldl1strm, [%0]\n" : : "r" (ptr));
}
inline void rx_prefetch_t0(const void* ptr) {
asm volatile ("prfm pldl1strm, [%0]\n" : : "r" (ptr));
}
FORCE_INLINE rx_vec_f128 rx_load_vec_f128(const double* pd) {
return vld1q_f64((const float64_t*)pd);
}
@ -532,6 +538,7 @@ typedef union {
#define rx_aligned_alloc(a, b) malloc(a)
#define rx_aligned_free(a) free(a)
#define rx_prefetch_nta(x)
#define rx_prefetch_t0(x)
FORCE_INLINE rx_vec_f128 rx_load_vec_f128(const double* pd) {
rx_vec_f128 x;