portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-16 14:57:05 +01:00
Code Releases Activity

FP16 optional compile time

Browse Source
This commit is contained in:
paboyle
2017-04-13 11:55:24 +01:00
parent 73cdf0fffe
commit 1d502e4ed6
5 changed files with 50 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
lib/simd/Grid_avx512.h
View File

@ -343,15 +343,23 @@ namespace Optimization {
struct PrecisionChange {
static inline __m512i StoH (__m512 a,__m512 b) {
#ifdef USE_FP16
__m256i ha = _mm512_cvtps_ph(a,0);
__m256i hb = _mm512_cvtps_ph(b,0);
__m512 h = _mm512_castps256_ps512(ha);
h = _mm512_insertf256_ps(h,hb,1);
#else
assert(0);
#endif
return h;
}
static inline void HtoS (__m512i h,__m512 &sa,__m512 &sb) {
#ifdef USE_FP16
sa = _mm512_cvtph_ps(_mm512_extractf256_ps(h,0));
sb = _mm512_cvtph_ps(_mm512_extractf256_ps(h,1));
#else
assert(0);
#endif
}
static inline __m512 DtoS (__m512d a,__m512d b) {
__m256 sa = _mm512_cvtpd_ps(a);