arm: memset: zero out upper bytes in r1

Commit Message

Andrey Ryabinin May 8, 2014, 2:40 p.m. UTC

On 05/08/14 11:59, Andrey Ryabinin wrote:
> On 05/08/14 03:42, Afzal Mohammed wrote:
>>
>> int is to be converted to unsigned char in memset, would having above
>> change immediately upon entry to memset rather than at a place where it
>> won't always execute make intention clearer ? (although it doesn't make
>> difference)
>>
> 
> I think it's better to keep it near other manipulations with r1.
> Plus this will save us from executing extra instruction on 'memset -> 6 -> 5 -> return'
> path (memset for size <= 3).
> 
> 
>> ubfx r1, r1, #0, #8 would have given the needed typecasting, but seems
>> it is available only on ARMv6T2 & above.
>>
> 
> Indeed. It might be wrapped it with #if #else, but it will be a bit ugly,
> probably not worth to do.
> 
> 

Err, there is no reason to use ubfx, it just the same as and r1,r1, #0xff.
On ARMv6T2 & above one instruction could be saved by using BFI instead of ORR, like this:

Patch

diff --git a/arch/arm/lib/memset.S b/arch/arm/lib/memset.S
index 94b0650..9a7f714 100644
--- a/arch/arm/lib/memset.S
+++ b/arch/arm/lib/memset.S
@@ -22,8 +22,8 @@  ENTRY(memset)
 /*
  * we know that the pointer in ip is aligned to a word boundary.
  */
-1:     orr     r1, r1, r1, lsl #8
-       orr     r1, r1, r1, lsl #16
+1:     bfi     r1, r1, #8, #8
+       bfi     r1, r1, #16, #16
        mov     r3, r1
        cmp     r2, #16
        blt     4f