[Bug target/90073] New: Very slow code for AVX2

[rcc.dark at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90073

            Bug ID: 90073
           Summary: Very slow code for AVX2
           Product: gcc
           Version: 9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rcc.dark at gmail dot com
  Target Milestone: ---

Created attachment 46155
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46155&action=edit
Source code and assembly output

Hi all, the following code generates very poor assembly code for AVX2 targets
(GCC 8.2, 8.3 and trunk; compiler flags -O3 -mavx2).

---
#include <x86intrin.h>

using data = long long __attribute__((vector_size(64)));
void f(data& a, const data& x1, const data& x2) {
    a ^= x1 ^ x2;
}
---

GCC generates 128-loads and stores, which leads to STLF stalls. Possibly a
duplicate of https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80689

ICC generates the expected code. I attach GCC and ICC outputs, but you can
check them here https://godbolt.org/z/bwtGUE