https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126065
Bug ID: 126065
Summary: Optimizer fails to optimize kxnor for AVX512
Product: gcc
Version: 16.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: postmaster at raasu dot org
Target Milestone: ---
With simple code:
> #include <x86intrin.h>
>
> int main(void) {
> __mmask16 target_mask = _kxnor_mask16((__mmask16)0, (__mmask16)0);
> __mmask16 mask = _kshiftri_mask16(target_mask, 9);
> return _mm512_mask2int(mask);
> }
GCC 16.1.0 generates using "-mavx512bw -mavx512vl -mavx512dq -O2" following
assembly code:
> "main":
> mov eax, 127
> kxorw k0, k0, k0
> kxnorw k0, k0, k0
> kshiftrw k0, k0, 9
> ret
"mov eax, 127" should be the only line generated with optimizations
enabled, the following three lines are redundant although missing the last
"kmovw eax, k0" that actually got optimized and reordered as the first line.
"kxorw k0, k0, k0" is completely unnecessary as "kxnorw k0, k0, k0" is
already fused XOR+NOT.
Seems to be similar to bug 88465.
