diff options
| author | Måns Rullgård <mans@mansr.com> | 2010-02-09 16:13:52 +0000 |
|---|---|---|
| committer | Måns Rullgård <mans@mansr.com> | 2010-02-09 16:13:52 +0000 |
| commit | 41c2bd0a26545aa79b71000b75f2f2c5380361d3 (patch) | |
| tree | aa1e5566e7c1c37338dec4056dd072c824493fcf /libavcodec/arm | |
| parent | 66ec243d95e971903d26764916991fa0c36a9481 (diff) | |
| download | ffmpeg-41c2bd0a26545aa79b71000b75f2f2c5380361d3.tar.gz | |
ARMv6 optimised pix_sum
Originally committed as revision 21705 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/arm')
| -rw-r--r-- | libavcodec/arm/dsputil_armv6.S | 25 | ||||
| -rw-r--r-- | libavcodec/arm/dsputil_init_armv6.c | 2 |
2 files changed, 27 insertions, 0 deletions
diff --git a/libavcodec/arm/dsputil_armv6.S b/libavcodec/arm/dsputil_armv6.S index ffe6c42252..acdfa12fdc 100644 --- a/libavcodec/arm/dsputil_armv6.S +++ b/libavcodec/arm/dsputil_armv6.S @@ -594,3 +594,28 @@ function ff_pix_norm1_armv6, export=1 mov r0, lr pop {r4-r6, pc} .endfunc + +function ff_pix_sum_armv6, export=1 + push {r4-r7, lr} + mov r12, #16 + mov r2, #0 + mov r3, #0 + mov lr, #0 + ldr r4, [r0] +1: + subs r12, r12, #1 + ldr r5, [r0, #4] + usada8 r2, r4, lr, r2 + ldr r6, [r0, #8] + usada8 r3, r5, lr, r3 + ldr r7, [r0, #12] + usada8 r2, r6, lr, r2 + beq 2f + ldr r4, [r0, r1]! + usada8 r3, r7, lr, r3 + bgt 1b +2: + usada8 r3, r7, lr, r3 + add r0, r2, r3 + pop {r4-r7, pc} +.endfunc diff --git a/libavcodec/arm/dsputil_init_armv6.c b/libavcodec/arm/dsputil_init_armv6.c index 6892e00774..3209062e59 100644 --- a/libavcodec/arm/dsputil_init_armv6.c +++ b/libavcodec/arm/dsputil_init_armv6.c @@ -68,6 +68,7 @@ int ff_sse16_armv6(void *s, uint8_t *blk1, uint8_t *blk2, int line_size, int h); int ff_pix_norm1_armv6(uint8_t *pix, int line_size); +int ff_pix_sum_armv6(uint8_t *pix, int line_size); void av_cold ff_dsputil_init_armv6(DSPContext* c, AVCodecContext *avctx) { @@ -116,4 +117,5 @@ void av_cold ff_dsputil_init_armv6(DSPContext* c, AVCodecContext *avctx) c->sse[0] = ff_sse16_armv6; c->pix_norm1 = ff_pix_norm1_armv6; + c->pix_sum = ff_pix_sum_armv6; } |