https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57952
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Known to work| |10.1.0
Resolution|--- |FIXED
Status|UNCONFIRMED |RESOLVED
--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
I can now see this vectorized with at least GCC 10 and up. Note we're
vectorizing the _outer_ loop here but we also manage to vectorize the
inner loop only if I comment out the outer one, it just looks less efficient.
.L2:
vmovdqa %ymm6, %ymm2
movl $10000000, %eax
.p2align 4,,10
.p2align 3
.L3:
vmovdqa %ymm2, %ymm0
vpaddd %ymm6, %ymm2, %ymm2
vcvtdq2ps %ymm0, %ymm0
vaddps %ymm5, %ymm0, %ymm0
vmulps %ymm11, %ymm0, %ymm0
vmovaps %ymm0, %ymm1
vfmadd132ps %ymm10, %ymm9, %ymm1
vfmadd132ps %ymm0, %ymm8, %ymm1
vfmadd132ps %ymm0, %ymm7, %ymm1
vfmadd132ps %ymm0, %ymm5, %ymm1
vfmadd132ps %ymm0, %ymm4, %ymm1
vfmadd132ps %ymm1, %ymm4, %ymm0
vaddps %ymm0, %ymm3, %ymm3
decl %eax
jne .L3
incl %edx
cmpl $12, %edx
jne .L2
