[Bug target/120201] G++ generates a misaligned vector operation for std::memcpy

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120201

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2025-05-09

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Partition 0: size 384 align 16
        transforms

That looks correct.

But this looks wrong:
;; [t.cc:17:20] _34 = [t.cc:17:20] MEM <vector(64) unsigned char> [(char *
{ref-all})&transforms];

(insn 98 97 0 (set (reg:V64QI 111 [ _34 ])
        (mem/c:V64QI (plus:DI (reg:DI 899)
                (const_int -448 [0xfffffffffffffe40])) [0 [t.cc:17:20] MEM
<vector(64) unsigned char> [(char * {ref-all})&transforms]+0 S64 A512]))
"t.cc":17:20 -1
     (nil))

Notice the 512 bit alignment.

But the alignment is correct in gimple:
  _34 = __MEM <unsigned char [[gnu::vector_size(64)]], 8> ((char *
{ref-all})&transforms);

Note the 8 which is 128 bit aligned.

Note this needs -march=znver4 as -march=skylake-avx512 does not unroll the loop
(I have not looked why though).