https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97770
Bug ID: 97770
Summary: Missing vectorization for vpopcnt
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: crazylht at gmail dot com
Target Milestone: ---
cat test.c
---
void
foo(int* __restrict dest, int* src, int n)
{
for (int i = 0; i != 8; i++)
dest[i] = __builtin_popcount (src[i]);
}
---
gcc -O3 -march=icelake-server -S -fopt-info-all
Inlined 0 calls, eliminated 0 functions
test.c:4:3: missed: couldn't vectorize loop
test.c:5:15: missed: not vectorized: relevant stmt not supported: _7 =
__builtin_popcount (_5);
test.c:2:1: note: vectorized 0 loops in function.
test.c:4:3: note: ***** Analysis failed with vector mode VOID
test.c:4:3: note: ***** Analysis failed with vector mode V8SI
test.c:4:3: note: ***** Skipping vector mode V32QI, which would repeat the
analysis for V8SI
test.c:6:1: note: ***** Analysis failed with vector mode VOID
This loop could be vectorized by ICC and Clang:
foo(int*, int*, int):
vpopcntd ymm0, YMMWORD PTR [rsi] #5.15
vmovdqu YMMWORD PTR [rdi], ymm0 #5.5
vzeroupper #6.1
ret
