Re: [ARM] architecture specific subdirectories & optimised memchr [V5]

[David Gilbert <david.gilbert@linaro.org>]

On 5 December 2011 22:11, Joseph S. Myers <joseph@codesourcery.com> wrote:
> On Fri, 2 Dec 2011, Dr. David Alan Gilbert wrote:
>
>> +     @ Work up to an aligned point
>> +5:
>> +     ldrb    r3, [r0],#1
>> +     subs    r2, r2, #1
>> +     cmp     r3, r1
>> +     beq     50f             @ If it matches exit found
>> +     tst     r0, #7
>> +     cbz     r2, 40f         @ If we run off the end, exit not found
>
> This loop is started with at least 16 bytes available and needs to skip at
> most 7 bytes to get to an aligned point, so I don't see a need for this
> test for running off the end in this loop.  That is, as far as I can see
> that cbz instruction could safely be removed.  Otherwise the patch looks
> OK (if the cbz instruction is indeed unneeded, test and post a version
> without that instruction and I'll commmit it).

Hmm interesting - I think you're right; I'd tweeked the alignments and
minimum lengths at various times and this had stayed in.

I'll just run a test through.

Dave