On Thu, Jun 27, 2019 at 8:17 AM Jakub Jelinek wrote:
>
> Hi!
>
> In the last two alternatives of avx_vec_concat, we can allow memory
> source, which optimizes the following testcases from weird
> vmovaps (%rdi), %xmm0
> vmovaps %xmm0, %xmm0
> and similar to just the first instructi
Hi!
In the last two alternatives of avx_vec_concat, we can allow memory
source, which optimizes the following testcases from weird
vmovaps (%rdi), %xmm0
vmovaps %xmm0, %xmm0
and similar to just the first instruction. I went through all the
gen_avx_vec_concat* users and all of them
Hi Jakub,
On 23 May 19:26, Jakub Jelinek wrote:
> Hi!
>
> Not sure how to easily test these.
> In any case, for the vinsert* case, we don't have vinserti128 nor
> vinsertf128 in evex, so need to use vinsert[if]{64x4,32x4} or
> for DQ {64x2,32x8}. For the case with zero in the other half,
> we nee
Hi!
Not sure how to easily test these.
In any case, for the vinsert* case, we don't have vinserti128 nor
vinsertf128 in evex, so need to use vinsert[if]{64x4,32x4} or
for DQ {64x2,32x8}. For the case with zero in the other half,
we need AVX512VL and it isn't guaranteed for the output operand,
bec
