[Bug target/84328] [6/7/8 Regression] -finline-small-functions and inline keyword lead to slowdown since version 6

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84328

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Target|                            |x86_64-*-* i?86-*-*
           Priority|P3                          |P2
          Component|c++                         |target
   Target Milestone|---                         |6.5
            Summary|[6 Regression]              |[6/7/8 Regression]
                   |-finline-small-functions    |-finline-small-functions
                   |and inline keyword lead to  |and inline keyword lead to
                   |slowdown since version 6    |slowdown since version 6

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The loop is (fast)

.L3:
        movzwl  %bx, %eax
        movl    %eax, %edx
        sall    $8, %edx
        orl     %eax, %edx
        movl    %edx, %eax
        andl    $16711935, %eax
        movl    %eax, %edx
        sall    $4, %edx
        orl     %eax, %edx
        andl    $252645135, %edx
        leal    0(,%rdx,4), %eax
        orl     %edx, %eax
        andl    $858993459, %eax
        movl    %eax, %edx
        leal    (%rax,%rax), %eax
        orl     %edx, %eax
        andl    $1431655765, %eax
        orl     $74565, %eax
        xorl    %eax, %ebx
        subq    $1, %rcx
        jne     .L3

vs. (slow)

.L3:
        movl    %ebx, %eax
        andl    $1023, %eax
        movl    %eax, %edx
        sall    $16, %edx
        xorl    %eax, %edx
        movl    %edx, %eax
        andl    $-16776961, %eax
        movl    %eax, %edx
        sall    $8, %edx
        xorl    %eax, %edx
        andl    $50393103, %edx
        movl    %edx, %eax
        sall    $4, %eax
        xorl    %edx, %eax
        andl    $51130563, %eax
        movl    %eax, %edx
        leal    0(,%rax,4), %eax
        xorl    %edx, %eax
        andl    $153391689, %eax
        orl     $74565, %eax
        xorl    %eax, %ebx
        subq    $1, %rcx
        jne     .L3

I didn't try to confirm the actual slowdown.  I think it has nothing to do with
inlining but with something x86 specific.