llvm dev - Mar 2015 - [LLVMdev] suggested x86 peephole optimization: use bit rotate to save one instruction during mask creation

This kind of pattern is very common when dealing with bit arrays:

unsigned long clear_mask(int bit)
{
    return ~(1UL << bit);
}

clang 3.6 at -O3 compiles this in the straightforward way, moving 1 into
register, shifting it, then flipping all the bits:

movl $1, %eax
movb %dil, %cl
shlq %cl, %rax
notq %rax

gcc 4.8.2, however, is more clever. It moves -2, i.e. ~1, into a register
then rotates that into place, saving one instruction:

movl %edi, %ecx
movq $-2, %rax
rolq %cl, %rax
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150315/9130dcb4/attachment.html>

A fix is out for review here: http://reviews.llvm.org/D8350

On Sun, Mar 15, 2015 at 12:23 PM, Mat Hostetter <mjhostetter at gmail.com>
wrote:
> This kind of pattern is very common when dealing with bit arrays:
>
> unsigned long clear_mask(int bit)
> {
>     return ~(1UL << bit);
> }
>
> clang 3.6 at -O3 compiles this in the straightforward way, moving 1 into
> register, shifting it, then flipping all the bits:
>
> movl $1, %eax
> movb %dil, %cl
> shlq %cl, %rax
> notq %rax
>
> gcc 4.8.2, however, is more clever. It moves -2, i.e. ~1, into a register
> then rotates that into place, saving one instruction:
>
> movl %edi, %ecx
> movq $-2, %rax
> rolq %cl, %rax
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20150315/762c5d96/attachment.html>