https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90202
Bug ID: 90202
Summary: AVX-512 instructions not used
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: antoshkka at gmail dot com
Target Milestone: ---
Consider the following test program:
struct v {
int val[16];
};
v test(v a, v b) {
v res;
for (int i = 0; i < 16; i++)
res.val[i] = a.val[i] + b.val[i];
return res;
}
When compiled with `g++ -O3 -march=skylake-avx512` the following assembly is
produced:
test(v, v):
push rbp
mov rax, rdi
mov rbp, rsp
vmovdqu32 ymm1, YMMWORD PTR [rbp+16]
vmovdqu32 ymm2, YMMWORD PTR [rbp+48]
vpaddd ymm0, ymm1, YMMWORD PTR [rbp+80]
vmovdqu32 YMMWORD PTR [rdi], ymm0
vpaddd ymm0, ymm2, YMMWORD PTR [rbp+112]
vmovdqu32 YMMWORD PTR [rdi+32], ymm0
vzeroupper
pop rbp
ret
it seems suboptimal, as the 512 registers are available and a better assembly
is possible:
test(v, v):
vmovdqu32 zmm0, zmmword ptr [rsp + 72]
vpaddd zmm0, zmm0, zmmword ptr [rsp + 8]
vmovdqu32 zmmword ptr [rdi], zmm0
mov rax, rdi
vzeroupper
ret
