[Bug target/82682] [8 Regression] FAIL: gcc.target/i386/pr50038.c scan-assembler-times movzbl 2 (found 3 times) since r253958

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82682

--- Comment #2 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
We now generate:
 .L3:
        movzbl  (%edx), %esi
        addl    $2, %edx
        addl    $1, %ecx
        movzbl  -1(%edx), %eax
        movl    %esi, %ebx
        imull   $38470, %eax, %eax
        movzbl  %bl, %esi
        imull   $19595, %esi, %esi
        addl    %esi, %eax
        sarl    $16, %eax
        movb    %al, -1(%ecx)
        cmpl    %edi, %edx
        jne     .L3

while from older gcc I get
.L3:
        movzbl  (%ecx,%edx,2), %eax
        movzbl  1(%ecx,%edx,2), %edi
        imull   $19595, %eax, %eax
        imull   $38470, %edi, %edi
        addl    %edi, %eax
        sarl    $16, %eax
        movb    %al, (%esi,%edx)
        addl    $1, %edx
        cmpl    %edx, %ebx
        jne     .L3
.L1:

There is clearly missed optimization on movzbl $bl, esi because it is already
extended. 

I wonder how this can be triggered by the move cost changes, perhaps regalloc
difference?
Jakub, it is easy for you to get .s files from  r253958 and just before?