diff options
| author | Ronald S. Bultje <rsbultje@gmail.com> | 2013-02-18 21:03:02 -0800 |
|---|---|---|
| committer | Martin Storsjö <martin@martin.st> | 2013-04-10 11:03:06 +0300 |
| commit | 62844c3fd66940c7747e9b2bb7804e265319f43f (patch) | |
| tree | b0e5a05644457aa5d7598d1fefa1a41f83550753 /libavcodec/ppc/h264_altivec.c | |
| parent | e8cafd2773bc56455c8816593cbd9368f2d69a80 (diff) | |
h264: Integrate clear_blocks calls with IDCT
The non-intra-pcm branch in hl_decode_mb (simple, 8bpp) goes from 700
to 672 cycles, and the complete loop of decode_mb_cabac and hl_decode_mb
(in the decode_slice loop) goes from 1759 to 1733 cycles on the clip
tested (cathedral), i.e. almost 30 cycles per mb faster.
Signed-off-by: Martin Storsjö <martin@martin.st>
Diffstat (limited to 'libavcodec/ppc/h264_altivec.c')
| -rw-r--r-- | libavcodec/ppc/h264_altivec.c | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/libavcodec/ppc/h264_altivec.c b/libavcodec/ppc/h264_altivec.c index 59642fdc36..20549c5b6d 100644 --- a/libavcodec/ppc/h264_altivec.c +++ b/libavcodec/ppc/h264_altivec.c @@ -87,6 +87,7 @@ static void ff_h264_idct_add_altivec(uint8_t *dst, int16_t *block, int stride) vtmp1 = vec_sld(vtmp0, vtmp0, 8); vtmp2 = vec_ld(16,block); vtmp3 = vec_sld(vtmp2, vtmp2, 8); + memset(block, 0, 16 * sizeof(int16_t)); VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3); VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3); @@ -206,6 +207,7 @@ static void ff_h264_idct8_add_altivec( uint8_t *dst, int16_t *dct, int stride ) s5 = vec_ld(0x50, (int16_t*)dct); s6 = vec_ld(0x60, (int16_t*)dct); s7 = vec_ld(0x70, (int16_t*)dct); + memset(dct, 0, 64 * sizeof(int16_t)); IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7); @@ -234,6 +236,7 @@ static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *bl int i; dc = (block[0] + 32) >> 6; + block[0] = 0; dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1); if (size == 4) |