[Bug middle-end/42233] [4.3/4.4/4.5 regression] c++ builtin_expect code generation regression

[jakub at gcc dot gnu dot org]

------- Comment #7 from jakub at gcc dot gnu dot org  2010-02-18 12:30 -------
On the 4.4 branch actually cc1 behaves differently from cc1plus.
In any case, I believe the two setne's plus testb code is actually the worst
of the options, and e.g. with -Os it is even much larger.
Just compile the testcase on 4.5 with -Os and -Os -D'__builtin_expect(a,b)=(a)'
In the former case (i.e. with __builtin_expect) if generates:
        cmpq    %rcx, %rdx
        sete    %dl
        xorl    %eax, %eax
        cmpq    %rsi, %rdi
        movzbl  %dl, %edx
        sete    %al
        testl   %eax, %edx
        je      .L2
        xorl    %eax, %eax
        jmp     likely
.L2:
        xorl    %eax, %eax
        jmp     unlikely
while in the latter case generates:
        cmpq    %rcx, %rdx
        jne     .L2
        cmpq    %rsi, %rdi
        jne     .L2
        xorl    %eax, %eax
        jmp     likely
.L2:
        xorl    %eax, %eax
        jmp     unlikely
(what would be actually the best code for -O2 when __builtin_expect is used and
tells that likely () is likely).


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bonzini at gnu dot org
      Known to work|4.1.2 4.5.0                 |4.1.2
            Summary|[4.3/4.4 regression] c++    |[4.3/4.4/4.5 regression] c++
                   |builtin_expect code         |builtin_expect code
                   |generation regression       |generation regression


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42233