diff options
author | Diego Biurrun <diego@biurrun.de> | 2013-04-27 22:52:26 +0200 |
---|---|---|
committer | Diego Biurrun <diego@biurrun.de> | 2013-05-02 11:45:37 +0200 |
commit | 932806232108872655556100011fe369125805d3 (patch) | |
tree | c4db29d8a8eb4ab870f0ef362cf48bdaccf7e5ce | |
parent | 9c112a6158928e8d69443adeb9a37e258bf59e55 (diff) | |
download | ffmpeg-932806232108872655556100011fe369125805d3.tar.gz |
x86: dsputil: Move avg_pixels16_mmx() out of rnd_template.c
The function does not do any rounding, so there is no point in
keeping it in a round template file.
-rw-r--r-- | libavcodec/x86/dsputil_mmx.c | 2 | ||||
-rw-r--r-- | libavcodec/x86/dsputil_mmx.h | 2 | ||||
-rw-r--r-- | libavcodec/x86/fpel_mmx.c | 24 | ||||
-rw-r--r-- | libavcodec/x86/hpeldsp_init.c | 2 | ||||
-rw-r--r-- | libavcodec/x86/rnd_template.c | 23 |
5 files changed, 29 insertions, 24 deletions
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c index 0a581070a2..9dacff625f 100644 --- a/libavcodec/x86/dsputil_mmx.c +++ b/libavcodec/x86/dsputil_mmx.c @@ -1028,7 +1028,7 @@ void ff_put_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src, ptrdiff_t stride) void ff_avg_cavs_qpel16_mc00_mmx(uint8_t *dst, uint8_t *src, ptrdiff_t stride) { - avg_pixels16_mmx(dst, src, stride, 16); + ff_avg_pixels16_mmx(dst, src, stride, 16); } /* VC-1-specific */ diff --git a/libavcodec/x86/dsputil_mmx.h b/libavcodec/x86/dsputil_mmx.h index eb23377cb3..7e98190d73 100644 --- a/libavcodec/x86/dsputil_mmx.h +++ b/libavcodec/x86/dsputil_mmx.h @@ -158,6 +158,8 @@ void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, int void ff_avg_pixels8_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h); +void ff_avg_pixels16_mmx(uint8_t *block, const uint8_t *pixels, + ptrdiff_t line_size, int h); void ff_put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h); void ff_put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, diff --git a/libavcodec/x86/fpel_mmx.c b/libavcodec/x86/fpel_mmx.c index c42791426e..be64098e0b 100644 --- a/libavcodec/x86/fpel_mmx.c +++ b/libavcodec/x86/fpel_mmx.c @@ -50,6 +50,30 @@ void ff_avg_pixels8_mmx(uint8_t *block, const uint8_t *pixels, while (--h); } +void ff_avg_pixels16_mmx(uint8_t *block, const uint8_t *pixels, + ptrdiff_t line_size, int h) +{ + MOVQ_BFE(mm6); + JUMPALIGN(); + do { + __asm__ volatile( + "movq %0, %%mm0 \n\t" + "movq %1, %%mm1 \n\t" + PAVGB_MMX(%%mm0, %%mm1, %%mm2, %%mm6) + "movq %%mm2, %0 \n\t" + "movq 8%0, %%mm0 \n\t" + "movq 8%1, %%mm1 \n\t" + PAVGB_MMX(%%mm0, %%mm1, %%mm2, %%mm6) + "movq %%mm2, 8%0 \n\t" + :"+m"(*block) + :"m"(*pixels) + :"memory"); + pixels += line_size; + block += line_size; + } + while (--h); +} + void ff_put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) { diff --git a/libavcodec/x86/hpeldsp_init.c b/libavcodec/x86/hpeldsp_init.c index bc2a08f10e..d573d05c76 100644 --- a/libavcodec/x86/hpeldsp_init.c +++ b/libavcodec/x86/hpeldsp_init.c @@ -75,8 +75,10 @@ void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h); #define avg_pixels8_mmx ff_avg_pixels8_mmx +#define avg_pixels16_mmx ff_avg_pixels16_mmx #define put_pixels8_mmx ff_put_pixels8_mmx #define put_pixels16_mmx ff_put_pixels16_mmx +#define avg_no_rnd_pixels16_mmx ff_avg_pixels16_mmx #define put_no_rnd_pixels8_mmx ff_put_pixels8_mmx #define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx diff --git a/libavcodec/x86/rnd_template.c b/libavcodec/x86/rnd_template.c index 08e8593a20..4443ec5729 100644 --- a/libavcodec/x86/rnd_template.c +++ b/libavcodec/x86/rnd_template.c @@ -92,29 +92,6 @@ static void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, ptrdiff } // avg_pixels -static void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) -{ - MOVQ_BFE(mm6); - JUMPALIGN(); - do { - __asm__ volatile( - "movq %0, %%mm0 \n\t" - "movq %1, %%mm1 \n\t" - OP_AVG(%%mm0, %%mm1, %%mm2, %%mm6) - "movq %%mm2, %0 \n\t" - "movq 8%0, %%mm0 \n\t" - "movq 8%1, %%mm1 \n\t" - OP_AVG(%%mm0, %%mm1, %%mm2, %%mm6) - "movq %%mm2, 8%0 \n\t" - :"+m"(*block) - :"m"(*pixels) - :"memory"); - pixels += line_size; - block += line_size; - } - while (--h); -} - // this routine is 'slightly' suboptimal but mostly unused static void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) { |