https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91796
Bug ID: 91796
Summary: Sub-optimal YMM register allocation.
Product: gcc
Version: 9.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: maxim.yegorushkin at gmail dot com
Target Milestone: ---
The following code when compiled with `g++ -O3 -mavx2 -std=c++11`
__m256d copysign2_pd(__m256d from, __m256d to) {
auto a = _mm256_castpd_si256(from);
auto avx_signbit =
_mm256_castsi256_pd(_mm256_slli_epi64(_mm256_cmpeq_epi64(a, a), 63));
return _mm256_or_pd(_mm256_and_pd(avx_signbit, from),
_mm256_andnot_pd(avx_signbit, to)); // (avx_signbit & from) | (~avx_signbit &
to)
}
Generates the following assembly:
copysign2_pd(double __vector(4), double __vector(4)):
vmovapd ymm2, ymm0
vmovapd ymm0, YMMWORD PTR .LC3[rip]
vandnpd ymm1, ymm0, ymm1
vandpd ymm0, ymm0, ymm2
vorpd ymm0, ymm0, ymm1
ret
.LC3:
.long 0
.long -2147483648
.long 0
.long -2147483648
.long 0
.long -2147483648
.long 0
.long -2147483648
In the assembly instruction `vmovapd ymm2, ymm0` is unnecessary. It can instead
load constant .LC3 directly into ymm2. The expected code is:
copysign2_pd(double __vector(4), double __vector(4)):
vmovapd ymm2, YMMWORD PTR .LC3[rip]
vandnpd ymm1, ymm2, ymm1
vandpd ymm0, ymm2, ymm0
vorpd ymm0, ymm0, ymm1
ret
