On 6/30/23 08:58, Song Gao wrote:
- for (i = 0; i < LSX_LEN/BIT; i++) { \
+ \
+ len = (simd_oprsz(v) == 16) ? LSX_LEN : LASX_LEN; \
+ for (i = 0; i < len / BIT; i++) { \
Similarly, use i < oprsz / (BIT / 8) in the loop.
+ Vd->Q(0) = int128_sub(int128_make64(Vj->UD(0)),
+ int128_make64(Vk->UD(0)));
+ if (simd_oprsz(v) == 32) {
+ Vd->Q(1) = int128_sub(int128_make64(Vj->UD(2)),
+ int128_make64(Vk->UD(2)));
+ }
And loop for these. r~
