github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

Use intrinsic to get correct movq everywhere

Browse Source 
The real trick here is that _mm_set_epi64x() (note the 'x') takes a 64-bit
value - not a ptr to a 128-bit value like the non-x - so compilers don't
twist themselves in knots with alignment or whatever confuses them.
This commit is contained in:
Matthew Barr 2017-03-07 09:58:24 +11:00
parent 96fdca6f54
commit d2416736cb
1 changed files with 1 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/util/simd_utils.h
View File

@ -180,9 +180,7 @@ static really_inline u64a movq(const m128 in) {
/* another form of movq */
static really_inline
m128 load_m128_from_u64a(const u64a *p) {
m128 out;
__asm__ ("vmovq\t%1,%0" : "=x"(out) :"m"(*p));
return out;
return _mm_set_epi64x(0LL, *p);
}
#define rshiftbyte_m128(a, count_immed) _mm_srli_si128(a, count_immed)