https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71361
--- Comment #5 from amker at gcc dot gnu.org ---
(In reply to Jakub Jelinek from comment #4)
> So shall we defer this PR to GCC 8 then (i.e. [8 Regression] and Target
> Milestone: 8.0? Richard, are you ok with that?
With ivopt rewriting, we now generate below code:
.L5:
movl (%esi,%ecx,4), %eax
movl 40(%esp), %edx
movl 44(%esp), %ebx
imull (%edi,%ecx,4), %ebx
imull %eax, %edx
imull 44(%esp), %eax
addl %ebx, %edx
movl 40(%esp), %ebx
imull (%edi,%ecx,4), %ebx
addl %ebx, %eax
movl 36(%esp), %ebx
movl %edx, (%ebx,%ecx,4)
movl 32(%esp), %ebx
movl %edx, (%edi,%ecx,4)
movl %eax, (%ebx,%ecx,4)
movl %eax, (%esi,%ecx,4)
addl 28(%esp), %ecx
cmpl $511, %ecx
jle .L5
Which I think is optimal. Shall we consider this fixed?
