[Bug tree-optimization/53355] Autovectorization of a simple loop could be improved.

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53355

--- Comment #3 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-05-15 13:19:06 UTC ---
With the patch we now emit

foo:
.LFB0:
        .cfi_startproc
        movq    %rdi, %rax
        salq    $60, %rax
        sarq    $63, %rax
        movq    %rax, %rdx
        andl    $1, %edx
        testl   %edx, %edx
        je      .L7
        movsd   .LC0(%rip), %xmm0
        movl    $99999, %r10d
        movl    $1, %r11d
        addsd   (%rdi), %xmm0
        movsd   %xmm0, (%rdi)
.L2:
        movl    $100000, %r8d
        andl    $1, %eax
        subl    %edx, %r8d
        movapd  .LC1(%rip), %xmm1
        movl    %r8d, %esi
        leaq    (%rdi,%rax,8), %rcx
        xorl    %edx, %edx
        shrl    %esi
        xorl    %eax, %eax
        leal    (%rsi,%rsi), %r9d
        .p2align 4,,10
        .p2align 3
.L6:
        movapd  (%rcx,%rax), %xmm0
        addl    $1, %edx
        addpd   %xmm1, %xmm0
        movapd  %xmm0, (%rcx,%rax)
        addq    $16, %rax
        cmpl    %esi, %edx
        jb      .L6
        subl    %r9d, %r10d
        cmpl    %r9d, %r8d
        leal    (%r11,%r9), %edx
        je      .L1
        leal    -1(%r10), %ecx
        movslq  %edx, %rdx
        leaq    0(,%rdx,8), %rax
        movsd   .LC0(%rip), %xmm1
        leaq    1(%rdx,%rcx), %rdx
        salq    $3, %rdx
        .p2align 4,,10
        .p2align 3
.L5:
        movsd   (%rdi,%rax), %xmm0
        addsd   %xmm1, %xmm0
        movsd   %xmm0, (%rdi,%rax)
        addq    $8, %rax
        cmpq    %rdx, %rax
        jne     .L5
.L1:
        rep
        ret
.L7:
        movl    $100000, %r10d
        xorl    %r11d, %r11d
        jmp     .L2