diff options
Diffstat (limited to 'ffmpeg/libavcodec/ppc')
| -rw-r--r-- | ffmpeg/libavcodec/ppc/Makefile | 22 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/dsputil_ppc.c | 5 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/fdct_altivec.c | 2 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/fft_altivec.c | 25 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/fmtconvert_altivec.c | 21 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/gmc_altivec.c | 4 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/h264_altivec.c | 748 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/h264_qpel.c | 317 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/h264_qpel_template.c | 507 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/h264chroma_init.c | 10 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/hpeldsp_altivec.c | 37 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/int_altivec.c | 8 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/mpegaudiodec_altivec.c | 130 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/mpegvideo_altivec.c | 14 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/vc1dsp_altivec.c | 9 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/vorbisdsp_altivec.c | 7 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/vp3dsp_altivec.c | 16 | ||||
| -rw-r--r-- | ffmpeg/libavcodec/ppc/vp8dsp_altivec.c | 57 |
18 files changed, 153 insertions, 1786 deletions
diff --git a/ffmpeg/libavcodec/ppc/Makefile b/ffmpeg/libavcodec/ppc/Makefile index febbb0a..71b23da 100644 --- a/ffmpeg/libavcodec/ppc/Makefile +++ b/ffmpeg/libavcodec/ppc/Makefile @@ -1,24 +1,24 @@ OBJS += ppc/dsputil_ppc.o \ + ppc/fmtconvert_altivec.o \ ppc/videodsp_ppc.o \ +OBJS-$(CONFIG_FFT) += ppc/fft_altivec.o OBJS-$(CONFIG_H264CHROMA) += ppc/h264chroma_init.o -OBJS-$(CONFIG_H264QPEL) += ppc/h264_qpel.o +OBJS-$(CONFIG_H264DSP) += ppc/h264dsp.o +OBJS-$(CONFIG_H264QPEL) += ppc/h264qpel.o OBJS-$(CONFIG_HPELDSP) += ppc/hpeldsp_altivec.o +OBJS-$(CONFIG_MPEGAUDIODSP) += ppc/mpegaudiodsp_altivec.o +OBJS-$(CONFIG_MPEGVIDEO) += ppc/mpegvideo_altivec.o +OBJS-$(CONFIG_VC1_DECODER) += ppc/vc1dsp_altivec.o OBJS-$(CONFIG_VORBIS_DECODER) += ppc/vorbisdsp_altivec.o OBJS-$(CONFIG_VP3DSP) += ppc/vp3dsp_altivec.o - -FFT-OBJS-$(HAVE_GNU_AS) += ppc/fft_altivec_s.o -ALTIVEC-OBJS-$(CONFIG_FFT) += ppc/fft_altivec.o \ - $(FFT-OBJS-yes) -ALTIVEC-OBJS-$(CONFIG_H264DSP) += ppc/h264_altivec.o -ALTIVEC-OBJS-$(CONFIG_MPEGAUDIODSP) += ppc/mpegaudiodec_altivec.o -ALTIVEC-OBJS-$(CONFIG_MPEGVIDEO) += ppc/mpegvideo_altivec.o -ALTIVEC-OBJS-$(CONFIG_VC1_DECODER) += ppc/vc1dsp_altivec.o -ALTIVEC-OBJS-$(CONFIG_VP8_DECODER) += ppc/vp8dsp_altivec.o +OBJS-$(CONFIG_VP8_DECODER) += ppc/vp8dsp_altivec.o ALTIVEC-OBJS += ppc/dsputil_altivec.o \ ppc/fdct_altivec.o \ - ppc/fmtconvert_altivec.o \ ppc/gmc_altivec.o \ ppc/idct_altivec.o \ ppc/int_altivec.o \ + +FFT-OBJS-$(HAVE_GNU_AS) += ppc/fft_altivec_s.o +ALTIVEC-OBJS-$(CONFIG_FFT) += $(FFT-OBJS-yes) diff --git a/ffmpeg/libavcodec/ppc/dsputil_ppc.c b/ffmpeg/libavcodec/ppc/dsputil_ppc.c index 6112b0c..7454ea0 100644 --- a/ffmpeg/libavcodec/ppc/dsputil_ppc.c +++ b/ffmpeg/libavcodec/ppc/dsputil_ppc.c @@ -25,6 +25,7 @@ #include "libavutil/attributes.h" #include "libavutil/cpu.h" #include "libavutil/mem.h" +#include "libavutil/ppc/cpu.h" #include "dsputil_altivec.h" /* ***** WARNING ***** WARNING ***** WARNING ***** */ @@ -156,8 +157,7 @@ av_cold void ff_dsputil_init_ppc(DSPContext *c, AVCodecContext *avctx) } } -#if HAVE_ALTIVEC - if (mm_flags & AV_CPU_FLAG_ALTIVEC) { + if (PPC_ALTIVEC(mm_flags)) { ff_dsputil_init_altivec(c, avctx); ff_int_init_altivec(c, avctx); c->gmc1 = ff_gmc1_altivec; @@ -180,5 +180,4 @@ av_cold void ff_dsputil_init_ppc(DSPContext *c, AVCodecContext *avctx) } } -#endif /* HAVE_ALTIVEC */ } diff --git a/ffmpeg/libavcodec/ppc/fdct_altivec.c b/ffmpeg/libavcodec/ppc/fdct_altivec.c index acab127..ff816e2 100644 --- a/ffmpeg/libavcodec/ppc/fdct_altivec.c +++ b/ffmpeg/libavcodec/ppc/fdct_altivec.c @@ -458,5 +458,3 @@ void ff_fdct_altivec(int16_t *block) #undef CTS /* }}} */ } - -/* vim:set foldmethod=marker foldlevel=0: */ diff --git a/ffmpeg/libavcodec/ppc/fft_altivec.c b/ffmpeg/libavcodec/ppc/fft_altivec.c index 651ee26..2357198 100644 --- a/ffmpeg/libavcodec/ppc/fft_altivec.c +++ b/ffmpeg/libavcodec/ppc/fft_altivec.c @@ -20,6 +20,8 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "config.h" +#include "libavutil/cpu.h" #include "libavutil/ppc/types_altivec.h" #include "libavutil/ppc/util_altivec.h" #include "libavcodec/fft.h" @@ -36,8 +38,8 @@ void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z); void ff_fft_calc_interleave_altivec(FFTContext *s, FFTComplex *z); -#if HAVE_GNU_AS -static void ff_imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input) +#if HAVE_GNU_AS && HAVE_ALTIVEC +static void imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input) { int j, k; int n = 1 << s->mdct_bits; @@ -117,7 +119,7 @@ static void ff_imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSam } while(k >= 0); } -static void ff_imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSample *input) +static void imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSample *input) { int k; int n = 1 << s->mdct_bits; @@ -127,7 +129,7 @@ static void ff_imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSam vec_u32 *p0 = (vec_u32*)(output+n4); vec_u32 *p1 = (vec_u32*)(output+n4*3); - ff_imdct_half_altivec(s, output+n4, input); + imdct_half_altivec(s, output + n4, input); for (k = 0; k < n16; k++) { vec_u32 a = p0[k] ^ sign; @@ -136,15 +138,18 @@ static void ff_imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSam p1[k] = vec_perm(b, b, vcprm(3,2,1,0)); } } -#endif /* HAVE_GNU_AS */ +#endif /* HAVE_GNU_AS && HAVE_ALTIVEC */ -av_cold void ff_fft_init_altivec(FFTContext *s) +av_cold void ff_fft_init_ppc(FFTContext *s) { -#if HAVE_GNU_AS +#if HAVE_GNU_AS && HAVE_ALTIVEC + if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) + return; + s->fft_calc = ff_fft_calc_interleave_altivec; if (s->mdct_bits >= 5) { - s->imdct_calc = ff_imdct_calc_altivec; - s->imdct_half = ff_imdct_half_altivec; + s->imdct_calc = imdct_calc_altivec; + s->imdct_half = imdct_half_altivec; } -#endif +#endif /* HAVE_GNU_AS && HAVE_ALTIVEC */ } diff --git a/ffmpeg/libavcodec/ppc/fmtconvert_altivec.c b/ffmpeg/libavcodec/ppc/fmtconvert_altivec.c index b29c7d4..cd32e39 100644 --- a/ffmpeg/libavcodec/ppc/fmtconvert_altivec.c +++ b/ffmpeg/libavcodec/ppc/fmtconvert_altivec.c @@ -18,14 +18,17 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ -#include "libavcodec/fmtconvert.h" - -#include "libavutil/ppc/util_altivec.h" +#include "config.h" #include "libavutil/attributes.h" +#include "libavutil/cpu.h" #include "libavutil/mem.h" +#include "libavutil/ppc/util_altivec.h" +#include "libavcodec/fmtconvert.h" #include "dsputil_altivec.h" -static void int32_to_float_fmul_scalar_altivec(float *dst, const int *src, +#if HAVE_ALTIVEC + +static void int32_to_float_fmul_scalar_altivec(float *dst, const int32_t *src, float mul, int len) { union { @@ -156,11 +159,19 @@ static void float_to_int16_interleave_altivec(int16_t *dst, const float **src, } } -av_cold void ff_fmt_convert_init_altivec(FmtConvertContext *c, AVCodecContext *avctx) +#endif /* HAVE_ALTIVEC */ + +av_cold void ff_fmt_convert_init_ppc(FmtConvertContext *c, + AVCodecContext *avctx) { +#if HAVE_ALTIVEC + if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) + return; + c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_altivec; if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { c->float_to_int16 = float_to_int16_altivec; c->float_to_int16_interleave = float_to_int16_interleave_altivec; } +#endif /* HAVE_ALTIVEC */ } diff --git a/ffmpeg/libavcodec/ppc/gmc_altivec.c b/ffmpeg/libavcodec/ppc/gmc_altivec.c index 4db761d..45d850a 100644 --- a/ffmpeg/libavcodec/ppc/gmc_altivec.c +++ b/ffmpeg/libavcodec/ppc/gmc_altivec.c @@ -66,7 +66,7 @@ void ff_gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src)); if (src_really_odd != 0x0000000F) { - // if src & 0xF == 0xF, then (src+1) is properly aligned + // if (src & 0xF) == 0xF, then (src+1) is properly aligned // on the second vector. srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src)); } else { @@ -90,7 +90,7 @@ void ff_gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src)); if (src_really_odd != 0x0000000F) { - // if src & 0xF == 0xF, then (src+1) is properly aligned + // if (src & 0xF) == 0xF, then (src+1) is properly aligned // on the second vector. srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src)); } else { diff --git a/ffmpeg/libavcodec/ppc/h264_altivec.c b/ffmpeg/libavcodec/ppc/h264_altivec.c deleted file mode 100644 index 3c2bb4d..0000000 --- a/ffmpeg/libavcodec/ppc/h264_altivec.c +++ /dev/null @@ -1,748 +0,0 @@ -/* - * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> - * - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#include "libavutil/attributes.h" -#include "libavutil/cpu.h" -#include "libavutil/intreadwrite.h" -#include "libavutil/ppc/types_altivec.h" -#include "libavutil/ppc/util_altivec.h" -#include "libavcodec/h264data.h" -#include "libavcodec/h264dsp.h" - -/**************************************************************************** - * IDCT transform: - ****************************************************************************/ - -#define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \ - /* 1st stage */ \ - vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \ - vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \ - vz2 = vec_sra(vb1,vec_splat_u16(1)); \ - vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \ - vz3 = vec_sra(vb3,vec_splat_u16(1)); \ - vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \ - /* 2nd stage: output */ \ - va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \ - va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \ - va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \ - va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */ - -#define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \ - b0 = vec_mergeh( a0, a0 ); \ - b1 = vec_mergeh( a1, a0 ); \ - b2 = vec_mergeh( a2, a0 ); \ - b3 = vec_mergeh( a3, a0 ); \ - a0 = vec_mergeh( b0, b2 ); \ - a1 = vec_mergel( b0, b2 ); \ - a2 = vec_mergeh( b1, b3 ); \ - a3 = vec_mergel( b1, b3 ); \ - b0 = vec_mergeh( a0, a2 ); \ - b1 = vec_mergel( a0, a2 ); \ - b2 = vec_mergeh( a1, a3 ); \ - b3 = vec_mergel( a1, a3 ) - -#define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \ - vdst_orig = vec_ld(0, dst); \ - vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \ - vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \ - va = vec_add(va, vdst_ss); \ - va_u8 = vec_packsu(va, zero_s16v); \ - va_u32 = vec_splat((vec_u32)va_u8, 0); \ - vec_ste(va_u32, element, (uint32_t*)dst); - -static void ff_h264_idct_add_altivec(uint8_t *dst, int16_t *block, int stride) -{ - vec_s16 va0, va1, va2, va3; - vec_s16 vz0, vz1, vz2, vz3; - vec_s16 vtmp0, vtmp1, vtmp2, vtmp3; - vec_u8 va_u8; - vec_u32 va_u32; - vec_s16 vdst_ss; - const vec_u16 v6us = vec_splat_u16(6); - vec_u8 vdst, vdst_orig; - vec_u8 vdst_mask = vec_lvsl(0, dst); - int element = ((unsigned long)dst & 0xf) >> 2; - LOAD_ZERO; - - block[0] += 32; /* add 32 as a DC-level for rounding */ - - vtmp0 = vec_ld(0,block); - 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); - VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3); - - va0 = vec_sra(va0,v6us); - va1 = vec_sra(va1,v6us); - va2 = vec_sra(va2,v6us); - va3 = vec_sra(va3,v6us); - - VEC_LOAD_U8_ADD_S16_STORE_U8(va0); - dst += stride; - VEC_LOAD_U8_ADD_S16_STORE_U8(va1); - dst += stride; - VEC_LOAD_U8_ADD_S16_STORE_U8(va2); - dst += stride; - VEC_LOAD_U8_ADD_S16_STORE_U8(va3); -} - -#define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\ - /* a0 = SRC(0) + SRC(4); */ \ - vec_s16 a0v = vec_add(s0, s4); \ - /* a2 = SRC(0) - SRC(4); */ \ - vec_s16 a2v = vec_sub(s0, s4); \ - /* a4 = (SRC(2)>>1) - SRC(6); */ \ - vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \ - /* a6 = (SRC(6)>>1) + SRC(2); */ \ - vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \ - /* b0 = a0 + a6; */ \ - vec_s16 b0v = vec_add(a0v, a6v); \ - /* b2 = a2 + a4; */ \ - vec_s16 b2v = vec_add(a2v, a4v); \ - /* b4 = a2 - a4; */ \ - vec_s16 b4v = vec_sub(a2v, a4v); \ - /* b6 = a0 - a6; */ \ - vec_s16 b6v = vec_sub(a0v, a6v); \ - /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \ - /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \ - vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \ - /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \ - /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \ - vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\ - /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \ - /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \ - vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\ - /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \ - vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\ - /* b1 = (a7>>2) + a1; */ \ - vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \ - /* b3 = a3 + (a5>>2); */ \ - vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \ - /* b5 = (a3>>2) - a5; */ \ - vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \ - /* b7 = a7 - (a1>>2); */ \ - vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \ - /* DST(0, b0 + b7); */ \ - d0 = vec_add(b0v, b7v); \ - /* DST(1, b2 + b5); */ \ - d1 = vec_add(b2v, b5v); \ - /* DST(2, b4 + b3); */ \ - d2 = vec_add(b4v, b3v); \ - /* DST(3, b6 + b1); */ \ - d3 = vec_add(b6v, b1v); \ - /* DST(4, b6 - b1); */ \ - d4 = vec_sub(b6v, b1v); \ - /* DST(5, b4 - b3); */ \ - d5 = vec_sub(b4v, b3v); \ - /* DST(6, b2 - b5); */ \ - d6 = vec_sub(b2v, b5v); \ - /* DST(7, b0 - b7); */ \ - d7 = vec_sub(b0v, b7v); \ -} - -#define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \ - /* unaligned load */ \ - vec_u8 hv = vec_ld( 0, dest ); \ - vec_u8 lv = vec_ld( 7, dest ); \ - vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \ - vec_s16 idct_sh6 = vec_sra(idctv, sixv); \ - vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \ - vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \ - vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \ - vec_u8 edgehv; \ - /* unaligned store */ \ - vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\ - vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \ - lv = vec_sel( lv, bodyv, edgelv ); \ - vec_st( lv, 7, dest ); \ - hv = vec_ld( 0, dest ); \ - edgehv = vec_perm( zero_u8v, sel, perm_stv ); \ - hv = vec_sel( hv, bodyv, edgehv ); \ - vec_st( hv, 0, dest ); \ - } - -static void ff_h264_idct8_add_altivec( uint8_t *dst, int16_t *dct, int stride ) { - vec_s16 s0, s1, s2, s3, s4, s5, s6, s7; - vec_s16 d0, d1, d2, d3, d4, d5, d6, d7; - vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7; - - vec_u8 perm_ldv = vec_lvsl(0, dst); - vec_u8 perm_stv = vec_lvsr(8, dst); - - const vec_u16 onev = vec_splat_u16(1); - const vec_u16 twov = vec_splat_u16(2); - const vec_u16 sixv = vec_splat_u16(6); - - const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1}; - LOAD_ZERO; - - dct[0] += 32; // rounding for the >>6 at the end - - s0 = vec_ld(0x00, (int16_t*)dct); - s1 = vec_ld(0x10, (int16_t*)dct); - s2 = vec_ld(0x20, (int16_t*)dct); - s3 = vec_ld(0x30, (int16_t*)dct); - s4 = vec_ld(0x40, (int16_t*)dct); - 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); - - TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 ); - - IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7, - idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7); - - ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel); - ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel); - ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel); - ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel); - ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel); - ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel); - ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel); - ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel); -} - -static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, int16_t *block, int stride, int size) -{ - vec_s16 dc16; - vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner; - LOAD_ZERO; - DECLARE_ALIGNED(16, int, dc); - int i; - - dc = (block[0] + 32) >> 6; - block[0] = 0; - dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1); - - if (size == 4) - dc16 = vec_sld(dc16, zero_s16v, 8); - dcplus = vec_packsu(dc16, zero_s16v); - dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v); - - aligner = vec_lvsr(0, dst); - dcplus = vec_perm(dcplus, dcplus, aligner); - dcminus = vec_perm(dcminus, dcminus, aligner); - - for (i = 0; i < size; i += 4) { - v0 = vec_ld(0, dst+0*stride); - v1 = vec_ld(0, dst+1*stride); - v2 = vec_ld(0, dst+2*stride); - v3 = vec_ld(0, dst+3*stride); - - v0 = vec_adds(v0, dcplus); - v1 = vec_adds(v1, dcplus); - v2 = vec_adds(v2, dcplus); - v3 = vec_adds(v3, dcplus); - - v0 = vec_subs(v0, dcminus); - v1 = vec_subs(v1, dcminus); - v2 = vec_subs(v2, dcminus); - v3 = vec_subs(v3, dcminus); - - vec_st(v0, 0, dst+0*stride); - vec_st(v1, 0, dst+1*stride); - vec_st(v2, 0, dst+2*stride); - vec_st(v3, 0, dst+3*stride); - - dst += 4*stride; - } -} - -static void h264_idct_dc_add_altivec(uint8_t *dst, int16_t *block, int stride) -{ - h264_idct_dc_add_internal(dst, block, stride, 4); -} - -static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, int16_t *block, int stride) -{ - h264_idct_dc_add_internal(dst, block, stride, 8); -} - -static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, int16_t *block, int stride, const uint8_t nnzc[15*8]){ - int i; - for(i=0; i<16; i++){ - int nnz = nnzc[ scan8[i] ]; - if(nnz){ - if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride); - else ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride); - } - } -} - -static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, int16_t *block, int stride, const uint8_t nnzc[15*8]){ - int i; - for(i=0; i<16; i++){ - if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride); - else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride); - } -} - -static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, int16_t *block, int stride, const uint8_t nnzc[15*8]){ - int i; - for(i=0; i<16; i+=4){ - int nnz = nnzc[ scan8[i] ]; - if(nnz){ - if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride); - else ff_h264_idct8_add_altivec (dst + block_offset[i], block + i*16, stride); - } - } -} - -static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, int16_t *block, int stride, const uint8_t nnzc[15*8]){ - int i, j; - for (j = 1; j < 3; j++) { - for(i = j * 16; i < j * 16 + 4; i++){ - if(nnzc[ scan8[i] ]) - ff_h264_idct_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride); - else if(block[i*16]) - h264_idct_dc_add_altivec(dest[j-1] + block_offset[i], block + i*16, stride); - } - } -} - -#define transpose4x16(r0, r1, r2, r3) { \ - register vec_u8 r4; \ - register vec_u8 r5; \ - register vec_u8 r6; \ - register vec_u8 r7; \ - \ - r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \ - r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \ - r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \ - r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \ - \ - r0 = vec_mergeh(r4, r6); /*all set 0*/ \ - r1 = vec_mergel(r4, r6); /*all set 1*/ \ - r2 = vec_mergeh(r5, r7); /*all set 2*/ \ - r3 = vec_mergel(r5, r7); /*all set 3*/ \ -} - -static inline void write16x4(uint8_t *dst, int dst_stride, - register vec_u8 r0, register vec_u8 r1, - register vec_u8 r2, register vec_u8 r3) { - DECLARE_ALIGNED(16, unsigned char, result)[64]; - uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst; - int int_dst_stride = dst_stride/4; - - vec_st(r0, 0, result); - vec_st(r1, 16, result); - vec_st(r2, 32, result); - vec_st(r3, 48, result); - /* FIXME: there has to be a better way!!!! */ - *dst_int = *src_int; - *(dst_int+ int_dst_stride) = *(src_int + 1); - *(dst_int+ 2*int_dst_stride) = *(src_int + 2); - *(dst_int+ 3*int_dst_stride) = *(src_int + 3); - *(dst_int+ 4*int_dst_stride) = *(src_int + 4); - *(dst_int+ 5*int_dst_stride) = *(src_int + 5); - *(dst_int+ 6*int_dst_stride) = *(src_int + 6); - *(dst_int+ 7*int_dst_stride) = *(src_int + 7); - *(dst_int+ 8*int_dst_stride) = *(src_int + 8); - *(dst_int+ 9*int_dst_stride) = *(src_int + 9); - *(dst_int+10*int_dst_stride) = *(src_int + 10); - *(dst_int+11*int_dst_stride) = *(src_int + 11); - *(dst_int+12*int_dst_stride) = *(src_int + 12); - *(dst_int+13*int_dst_stride) = *(src_int + 13); - *(dst_int+14*int_dst_stride) = *(src_int + 14); - *(dst_int+15*int_dst_stride) = *(src_int + 15); -} - -/** @brief performs a 6x16 transpose of data in src, and stores it to dst - @todo FIXME: see if we can't spare some vec_lvsl() by them factorizing - out of unaligned_load() */ -#define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\ - register vec_u8 r0 = unaligned_load(0, src); \ - register vec_u8 r1 = unaligned_load( src_stride, src); \ - register vec_u8 r2 = unaligned_load(2* src_stride, src); \ - register vec_u8 r3 = unaligned_load(3* src_stride, src); \ - register vec_u8 r4 = unaligned_load(4* src_stride, src); \ - register vec_u8 r5 = unaligned_load(5* src_stride, src); \ - register vec_u8 r6 = unaligned_load(6* src_stride, src); \ - register vec_u8 r7 = unaligned_load(7* src_stride, src); \ - register vec_u8 r14 = unaligned_load(14*src_stride, src); \ - register vec_u8 r15 = unaligned_load(15*src_stride, src); \ - \ - r8 = unaligned_load( 8*src_stride, src); \ - r9 = unaligned_load( 9*src_stride, src); \ - r10 = unaligned_load(10*src_stride, src); \ - r11 = unaligned_load(11*src_stride, src); \ - r12 = unaligned_load(12*src_stride, src); \ - r13 = unaligned_load(13*src_stride, src); \ - \ - /*Merge first pairs*/ \ - r0 = vec_mergeh(r0, r8); /*0, 8*/ \ - r1 = vec_mergeh(r1, r9); /*1, 9*/ \ - r2 = vec_mergeh(r2, r10); /*2,10*/ \ - r3 = vec_mergeh(r3, r11); /*3,11*/ \ - r4 = vec_mergeh(r4, r12); /*4,12*/ \ - r5 = vec_mergeh(r5, r13); /*5,13*/ \ - r6 = vec_mergeh(r6, r14); /*6,14*/ \ - r7 = vec_mergeh(r7, r15); /*7,15*/ \ - \ - /*Merge second pairs*/ \ - r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \ - r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \ - r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \ - r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \ - r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \ - r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \ - r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \ - r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \ - \ - /*Third merge*/ \ - r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \ - r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \ - r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \ - r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \ - r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \ - r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \ - /* Don't need to compute 3 and 7*/ \ - \ - /*Final merge*/ \ - r8 = vec_mergeh(r0, r4); /*all set 0*/ \ - r9 = vec_mergel(r0, r4); /*all set 1*/ \ - r10 = vec_mergeh(r1, r5); /*all set 2*/ \ - r11 = vec_mergel(r1, r5); /*all set 3*/ \ - r12 = vec_mergeh(r2, r6); /*all set 4*/ \ - r13 = vec_mergel(r2, r6); /*all set 5*/ \ - /* Don't need to compute 14 and 15*/ \ - \ -} - -// out: o = |x-y| < a -static inline vec_u8 diff_lt_altivec ( register vec_u8 x, - register vec_u8 y, - register vec_u8 a) { - - register vec_u8 diff = vec_subs(x, y); - register vec_u8 diffneg = vec_subs(y, x); - register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */ - o = (vec_u8)vec_cmplt(o, a); - return o; -} - -static inline vec_u8 h264_deblock_mask ( register vec_u8 p0, - register vec_u8 p1, - register vec_u8 q0, - register vec_u8 q1, - register vec_u8 alpha, - register vec_u8 beta) { - - register vec_u8 mask; - register vec_u8 tempmask; - - mask = diff_lt_altivec(p0, q0, alpha); - tempmask = diff_lt_altivec(p1, p0, beta); - mask = vec_and(mask, tempmask); - tempmask = diff_lt_altivec(q1, q0, beta); - mask = vec_and(mask, tempmask); - - return mask; -} - -// out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0) -static inline vec_u8 h264_deblock_q1(register vec_u8 p0, - register vec_u8 p1, - register vec_u8 p2, - register vec_u8 q0, - register vec_u8 tc0) { - - register vec_u8 average = vec_avg(p0, q0); - register vec_u8 temp; - register vec_u8 uncliped; - register vec_u8 ones; - register vec_u8 max; - register vec_u8 min; - register vec_u8 newp1; - - temp = vec_xor(average, p2); - average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */ - ones = vec_splat_u8(1); - temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */ - uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */ - max = vec_adds(p1, tc0); - min = vec_subs(p1, tc0); - newp1 = vec_max(min, uncliped); - newp1 = vec_min(max, newp1); - return newp1; -} - -#define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \ - \ - const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \ - \ - register vec_u8 pq0bit = vec_xor(p0,q0); \ - register vec_u8 q1minus; \ - register vec_u8 p0minus; \ - register vec_u8 stage1; \ - register vec_u8 stage2; \ - register vec_u8 vec160; \ - register vec_u8 delta; \ - register vec_u8 deltaneg; \ - \ - q1minus = vec_nor(q1, q1); /* 255 - q1 */ \ - stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \ - stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \ - p0minus = vec_nor(p0, p0); /* 255 - p0 */ \ - stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \ - pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \ - stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \ - stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \ - vec160 = vec_ld(0, &A0v); \ - deltaneg = vec_subs(vec160, stage2); /* -d */ \ - delta = vec_subs(stage2, vec160); /* d */ \ - deltaneg = vec_min(tc0masked, deltaneg); \ - delta = vec_min(tc0masked, delta); \ - p0 = vec_subs(p0, deltaneg); \ - q0 = vec_subs(q0, delta); \ - p0 = vec_adds(p0, delta); \ - q0 = vec_adds(q0, deltaneg); \ -} - -#define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \ - DECLARE_ALIGNED(16, unsigned char, temp)[16]; \ - register vec_u8 alphavec; \ - register vec_u8 betavec; \ - register vec_u8 mask; \ - register vec_u8 p1mask; \ - register vec_u8 q1mask; \ - register vector signed char tc0vec; \ - register vec_u8 finaltc0; \ - register vec_u8 tc0masked; \ - register vec_u8 newp1; \ - register vec_u8 newq1; \ - \ - temp[0] = alpha; \ - temp[1] = beta; \ - alphavec = vec_ld(0, temp); \ - betavec = vec_splat(alphavec, 0x1); \ - alphavec = vec_splat(alphavec, 0x0); \ - mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \ - \ - AV_COPY32(temp, tc0); \ - tc0vec = vec_ld(0, (signed char*)temp); \ - tc0vec = vec_mergeh(tc0vec, tc0vec); \ - tc0vec = vec_mergeh(tc0vec, tc0vec); \ - mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \ - finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \ - \ - p1mask = diff_lt_altivec(p2, p0, betavec); \ - p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \ - tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \ - finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \ - newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \ - /*end if*/ \ - \ - q1mask = diff_lt_altivec(q2, q0, betavec); \ - q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\ - tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \ - finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \ - newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \ - /*end if*/ \ - \ - h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \ - p1 = newp1; \ - q1 = newq1; \ -} - -static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) { - - if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) { - register vec_u8 p2 = vec_ld(-3*stride, pix); - register vec_u8 p1 = vec_ld(-2*stride, pix); - register vec_u8 p0 = vec_ld(-1*stride, pix); - register vec_u8 q0 = vec_ld(0, pix); - register vec_u8 q1 = vec_ld(stride, pix); - register vec_u8 q2 = vec_ld(2*stride, pix); - h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0); - vec_st(p1, -2*stride, pix); - vec_st(p0, -1*stride, pix); - vec_st(q0, 0, pix); - vec_st(q1, stride, pix); - } -} - -static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) { - - register vec_u8 line0, line1, line2, line3, line4, line5; - if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0) - return; - readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5); - h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0); - transpose4x16(line1, line2, line3, line4); - write16x4(pix-2, stride, line1, line2, line3, line4); -} - -static av_always_inline -void weight_h264_W_altivec(uint8_t *block, int stride, int height, - int log2_denom, int weight, int offset, int w) -{ - int y, aligned; - vec_u8 vblock; - vec_s16 vtemp, vweight, voffset, v0, v1; - vec_u16 vlog2_denom; - DECLARE_ALIGNED(16, int32_t, temp)[4]; - LOAD_ZERO; - - offset <<= log2_denom; - if(log2_denom) offset += 1<<(log2_denom-1); - temp[0] = log2_denom; - temp[1] = weight; - temp[2] = offset; - - vtemp = (vec_s16)vec_ld(0, temp); - vlog2_denom = (vec_u16)vec_splat(vtemp, 1); - vweight = vec_splat(vtemp, 3); - voffset = vec_splat(vtemp, 5); - aligned = !((unsigned long)block & 0xf); - - for (y = 0; y < height; y++) { - vblock = vec_ld(0, block); - - v0 = (vec_s16)vec_mergeh(zero_u8v, vblock); - v1 = (vec_s16)vec_mergel(zero_u8v, vblock); - - if (w == 16 || aligned) { - v0 = vec_mladd(v0, vweight, zero_s16v); - v0 = vec_adds(v0, voffset); - v0 = vec_sra(v0, vlog2_denom); - } - if (w == 16 || !aligned) { - v1 = vec_mladd(v1, vweight, zero_s16v); - v1 = vec_adds(v1, voffset); - v1 = vec_sra(v1, vlog2_denom); - } - vblock = vec_packsu(v0, v1); - vec_st(vblock, 0, block); - - block += stride; - } -} - -static av_always_inline -void biweight_h264_W_altivec(uint8_t *dst, uint8_t *src, int stride, int height, - int log2_denom, int weightd, int weights, int offset, int w) -{ - int y, dst_aligned, src_aligned; - vec_u8 vsrc, vdst; - vec_s16 vtemp, vweights, vweightd, voffset, v0, v1, v2, v3; - vec_u16 vlog2_denom; - DECLARE_ALIGNED(16, int32_t, temp)[4]; - LOAD_ZERO; - - offset = ((offset + 1) | 1) << log2_denom; - temp[0] = log2_denom+1; - temp[1] = weights; - temp[2] = weightd; - temp[3] = offset; - - vtemp = (vec_s16)vec_ld(0, temp); - vlog2_denom = (vec_u16)vec_splat(vtemp, 1); - vweights = vec_splat(vtemp, 3); - vweightd = vec_splat(vtemp, 5); - voffset = vec_splat(vtemp, 7); - dst_aligned = !((unsigned long)dst & 0xf); - src_aligned = !((unsigned long)src & 0xf); - - for (y = 0; y < height; y++) { - vdst = vec_ld(0, dst); - vsrc = vec_ld(0, src); - - v0 = (vec_s16)vec_mergeh(zero_u8v, vdst); - v1 = (vec_s16)vec_mergel(zero_u8v, vdst); - v2 = (vec_s16)vec_mergeh(zero_u8v, vsrc); - v3 = (vec_s16)vec_mergel(zero_u8v, vsrc); - - if (w == 8) { - if (src_aligned) - v3 = v2; - else - v2 = v3; - } - - if (w == 16 || dst_aligned) { - v0 = vec_mladd(v0, vweightd, zero_s16v); - v2 = vec_mladd(v2, vweights, zero_s16v); - - v0 = vec_adds(v0, voffset); - v0 = vec_adds(v0, v2); - v0 = vec_sra(v0, vlog2_denom); - } - if (w == 16 || !dst_aligned) { - v1 = vec_mladd(v1, vweightd, zero_s16v); - v3 = vec_mladd(v3, vweights, zero_s16v); - - v1 = vec_adds(v1, voffset); - v1 = vec_adds(v1, v3); - v1 = vec_sra(v1, vlog2_denom); - } - vdst = vec_packsu(v0, v1); - vec_st(vdst, 0, dst); - - dst += stride; - src += stride; - } -} - -#define H264_WEIGHT(W) \ -static void ff_weight_h264_pixels ## W ## _altivec(uint8_t *block, int stride, int height, \ - int log2_denom, int weight, int offset){ \ - weight_h264_W_altivec(block, stride, height, log2_denom, weight, offset, W); \ -}\ -static void ff_biweight_h264_pixels ## W ## _altivec(uint8_t *dst, uint8_t *src, int stride, int height, \ - int log2_denom, int weightd, int weights, int offset){ \ - biweight_h264_W_altivec(dst, src, stride, height, log2_denom, weightd, weights, offset, W); \ -} - -H264_WEIGHT(16) -H264_WEIGHT( 8) - -av_cold void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth, - const int chroma_format_idc) -{ - if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) { - if (bit_depth == 8) { - c->h264_idct_add = ff_h264_idct_add_altivec; - if (chroma_format_idc == 1) - c->h264_idct_add8 = ff_h264_idct_add8_altivec; - c->h264_idct_add16 = ff_h264_idct_add16_altivec; - c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec; - c->h264_idct_dc_add= h264_idct_dc_add_altivec; - c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec; - c->h264_idct8_add = ff_h264_idct8_add_altivec; - c->h264_idct8_add4 = ff_h264_idct8_add4_altivec; - c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec; - c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec; - - c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16_altivec; - c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels8_altivec; - c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16_altivec; - c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_altivec; - } - } -} diff --git a/ffmpeg/libavcodec/ppc/h264_qpel.c b/ffmpeg/libavcodec/ppc/h264_qpel.c deleted file mode 100644 index 429ae42..0000000 --- a/ffmpeg/libavcodec/ppc/h264_qpel.c +++ /dev/null @@ -1,317 +0,0 @@ -/* - * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> - * - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#include "config.h" -#include "libavutil/attributes.h" -#include "libavcodec/h264qpel.h" - -#if HAVE_ALTIVEC -#include "libavutil/cpu.h" -#include "libavutil/intreadwrite.h" -#include "libavutil/ppc/types_altivec.h" -#include "libavutil/ppc/util_altivec.h" -#include "dsputil_altivec.h" - -#define PUT_OP_U8_ALTIVEC(d, s, dst) d = s -#define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s) - -#define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC -#define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec -#define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num -#define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec -#define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num -#define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec -#define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num -#include "h264_qpel_template.c" -#undef OP_U8_ALTIVEC -#undef PREFIX_h264_qpel16_h_lowpass_altivec -#undef PREFIX_h264_qpel16_h_lowpass_num -#undef PREFIX_h264_qpel16_v_lowpass_altivec -#undef PREFIX_h264_qpel16_v_lowpass_num -#undef PREFIX_h264_qpel16_hv_lowpass_altivec -#undef PREFIX_h264_qpel16_hv_lowpass_num - -#define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC -#define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec -#define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num -#define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec -#define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num -#define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec -#define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num -#include "h264_qpel_template.c" -#undef OP_U8_ALTIVEC -#undef PREFIX_h264_qpel16_h_lowpass_altivec -#undef PREFIX_h264_qpel16_h_lowpass_num -#undef PREFIX_h264_qpel16_v_lowpass_altivec -#undef PREFIX_h264_qpel16_v_lowpass_num -#undef PREFIX_h264_qpel16_hv_lowpass_altivec -#undef PREFIX_h264_qpel16_hv_lowpass_num - -#define H264_MC(OPNAME, SIZE, CODETYPE) \ -static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - ff_ ## OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{ \ - DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ - put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ - put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ - put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ - put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ - put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ - put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ - put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ - put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ - put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ - put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ - put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ - put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ - OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ - put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ - put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ - put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ - put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ - put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ - put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ -}\ -\ -static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\ -{\ - DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ - DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ - put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ - put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ - OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ -}\ - -static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, - const uint8_t * src2, int dst_stride, - int src_stride1, int h) -{ - int i; - vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; - - mask_ = vec_lvsl(0, src2); - - for (i = 0; i < h; i++) { - - tmp1 = vec_ld(i * src_stride1, src1); - mask = vec_lvsl(i * src_stride1, src1); - tmp2 = vec_ld(i * src_stride1 + 15, src1); - - a = vec_perm(tmp1, tmp2, mask); - - tmp1 = vec_ld(i * 16, src2); - tmp2 = vec_ld(i * 16 + 15, src2); - - b = vec_perm(tmp1, tmp2, mask_); - - tmp1 = vec_ld(0, dst); - mask = vec_lvsl(0, dst); - tmp2 = vec_ld(15, dst); - - d = vec_avg(a, b); - - edges = vec_perm(tmp2, tmp1, mask); - - align = vec_lvsr(0, dst); - - tmp2 = vec_perm(d, edges, align); - tmp1 = vec_perm(edges, d, align); - - vec_st(tmp2, 15, dst); - vec_st(tmp1, 0 , dst); - - dst += dst_stride; - } -} - -static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, - const uint8_t * src2, int dst_stride, - int src_stride1, int h) -{ - int i; - vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; - - mask_ = vec_lvsl(0, src2); - - for (i = 0; i < h; i++) { - - tmp1 = vec_ld(i * src_stride1, src1); - mask = vec_lvsl(i * src_stride1, src1); - tmp2 = vec_ld(i * src_stride1 + 15, src1); - - a = vec_perm(tmp1, tmp2, mask); - - tmp1 = vec_ld(i * 16, src2); - tmp2 = vec_ld(i * 16 + 15, src2); - - b = vec_perm(tmp1, tmp2, mask_); - - tmp1 = vec_ld(0, dst); - mask = vec_lvsl(0, dst); - tmp2 = vec_ld(15, dst); - - d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b)); - - edges = vec_perm(tmp2, tmp1, mask); - - align = vec_lvsr(0, dst); - - tmp2 = vec_perm(d, edges, align); - tmp1 = vec_perm(edges, d, align); - - vec_st(tmp2, 15, dst); - vec_st(tmp1, 0 , dst); - - dst += dst_stride; - } -} - -/* Implemented but could be faster -#define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h) -#define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h) - */ - -H264_MC(put_, 16, altivec) -H264_MC(avg_, 16, altivec) -#endif /* HAVE_ALTIVEC */ - -av_cold void ff_h264qpel_init_ppc(H264QpelContext *c, int bit_depth) -{ -#if HAVE_ALTIVEC - const int high_bit_depth = bit_depth > 8; - - if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) { - if (!high_bit_depth) { -#define dspfunc(PFX, IDX, NUM) \ - c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \ - c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \ - c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \ - c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \ - c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \ - c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \ - c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \ - c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \ - c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \ - c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \ - c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \ - c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \ - c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \ - c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \ - c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \ - c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec - - dspfunc(put_h264_qpel, 0, 16); - dspfunc(avg_h264_qpel, 0, 16); -#undef dspfunc - } - } -#endif /* HAVE_ALTIVEC */ -} diff --git a/ffmpeg/libavcodec/ppc/h264_qpel_template.c b/ffmpeg/libavcodec/ppc/h264_qpel_template.c deleted file mode 100644 index cfc4560..0000000 --- a/ffmpeg/libavcodec/ppc/h264_qpel_template.c +++ /dev/null @@ -1,507 +0,0 @@ -/* - * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org> - * - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#include "libavutil/mem.h" - -#ifdef DEBUG -#define ASSERT_ALIGNED(ptr) assert(((unsigned long)ptr&0x0000000F)); -#else -#define ASSERT_ALIGNED(ptr) ; -#endif - -/* this code assume stride % 16 == 0 */ -#ifdef PREFIX_h264_qpel16_h_lowpass_altivec -static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) { - register int i; - - LOAD_ZERO; - const vec_u8 permM2 = vec_lvsl(-2, src); - const vec_u8 permM1 = vec_lvsl(-1, src); - const vec_u8 permP0 = vec_lvsl(+0, src); - const vec_u8 permP1 = vec_lvsl(+1, src); - const vec_u8 permP2 = vec_lvsl(+2, src); - const vec_u8 permP3 = vec_lvsl(+3, src); - const vec_s16 v5ss = vec_splat_s16(5); - const vec_u16 v5us = vec_splat_u16(5); - const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); - const vec_s16 v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4)); - - vec_u8 srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; - - register int align = ((((unsigned long)src) - 2) % 16); - - vec_s16 srcP0A, srcP0B, srcP1A, srcP1B, - srcP2A, srcP2B, srcP3A, srcP3B, - srcM1A, srcM1B, srcM2A, srcM2B, - sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, - pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, - psumA, psumB, sumA, sumB; - - vec_u8 sum, fsum; - - for (i = 0 ; i < 16 ; i ++) { - vec_u8 srcR1 = vec_ld(-2, src); - vec_u8 srcR2 = vec_ld(14, src); - - switch (align) { - default: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = vec_perm(srcR1, srcR2, permP3); - } break; - case 11: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = srcR2; - } break; - case 12: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = srcR2; - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 13: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = srcR2; - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 14: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = srcR2; - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 15: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = srcR2; - srcP0 = vec_perm(srcR2, srcR3, permP0); - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - } - - srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0); - srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0); - srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1); - srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1); - - srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2); - srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2); - srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3); - srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3); - - srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1); - srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1); - srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2); - srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2); - - sum1A = vec_adds(srcP0A, srcP1A); - sum1B = vec_adds(srcP0B, srcP1B); - sum2A = vec_adds(srcM1A, srcP2A); - sum2B = vec_adds(srcM1B, srcP2B); - sum3A = vec_adds(srcM2A, srcP3A); - sum3B = vec_adds(srcM2B, srcP3B); - - pp1A = vec_mladd(sum1A, v20ss, v16ss); - pp1B = vec_mladd(sum1B, v20ss, v16ss); - - pp2A = vec_mladd(sum2A, v5ss, zero_s16v); - pp2B = vec_mladd(sum2B, v5ss, zero_s16v); - - pp3A = vec_add(sum3A, pp1A); - pp3B = vec_add(sum3B, pp1B); - - psumA = vec_sub(pp3A, pp2A); - psumB = vec_sub(pp3B, pp2B); - - sumA = vec_sra(psumA, v5us); - sumB = vec_sra(psumB, v5us); - - sum = vec_packsu(sumA, sumB); - - ASSERT_ALIGNED(dst); - - OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst)); - - vec_st(fsum, 0, dst); - - src += srcStride; - dst += dstStride; - } -} -#endif - -/* this code assume stride % 16 == 0 */ -#ifdef PREFIX_h264_qpel16_v_lowpass_altivec -static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) { - register int i; - - LOAD_ZERO; - const vec_u8 perm = vec_lvsl(0, src); - const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); - const vec_u16 v5us = vec_splat_u16(5); - const vec_s16 v5ss = vec_splat_s16(5); - const vec_s16 v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4)); - - uint8_t *srcbis = src - (srcStride * 2); - - const vec_u8 srcM2a = vec_ld(0, srcbis); - const vec_u8 srcM2b = vec_ld(16, srcbis); - const vec_u8 srcM2 = vec_perm(srcM2a, srcM2b, perm); - //srcbis += srcStride; - const vec_u8 srcM1a = vec_ld(0, srcbis += srcStride); - const vec_u8 srcM1b = vec_ld(16, srcbis); - const vec_u8 srcM1 = vec_perm(srcM1a, srcM1b, perm); - //srcbis += srcStride; - const vec_u8 srcP0a = vec_ld(0, srcbis += srcStride); - const vec_u8 srcP0b = vec_ld(16, srcbis); - const vec_u8 srcP0 = vec_perm(srcP0a, srcP0b, perm); - //srcbis += srcStride; - const vec_u8 srcP1a = vec_ld(0, srcbis += srcStride); - const vec_u8 srcP1b = vec_ld(16, srcbis); - const vec_u8 srcP1 = vec_perm(srcP1a, srcP1b, perm); - //srcbis += srcStride; - const vec_u8 srcP2a = vec_ld(0, srcbis += srcStride); - const vec_u8 srcP2b = vec_ld(16, srcbis); - const vec_u8 srcP2 = vec_perm(srcP2a, srcP2b, perm); - //srcbis += srcStride; - - vec_s16 srcM2ssA = (vec_s16) vec_mergeh(zero_u8v, srcM2); - vec_s16 srcM2ssB = (vec_s16) vec_mergel(zero_u8v, srcM2); - vec_s16 srcM1ssA = (vec_s16) vec_mergeh(zero_u8v, srcM1); - vec_s16 srcM1ssB = (vec_s16) vec_mergel(zero_u8v, srcM1); - vec_s16 srcP0ssA = (vec_s16) vec_mergeh(zero_u8v, srcP0); - vec_s16 srcP0ssB = (vec_s16) vec_mergel(zero_u8v, srcP0); - vec_s16 srcP1ssA = (vec_s16) vec_mergeh(zero_u8v, srcP1); - vec_s16 srcP1ssB = (vec_s16) vec_mergel(zero_u8v, srcP1); - vec_s16 srcP2ssA = (vec_s16) vec_mergeh(zero_u8v, srcP2); - vec_s16 srcP2ssB = (vec_s16) vec_mergel(zero_u8v, srcP2); - - vec_s16 pp1A, pp1B, pp2A, pp2B, pp3A, pp3B, - psumA, psumB, sumA, sumB, - srcP3ssA, srcP3ssB, - sum1A, sum1B, sum2A, sum2B, sum3A, sum3B; - - vec_u8 sum, fsum, srcP3a, srcP3b, srcP3; - - for (i = 0 ; i < 16 ; i++) { - srcP3a = vec_ld(0, srcbis += srcStride); - srcP3b = vec_ld(16, srcbis); - srcP3 = vec_perm(srcP3a, srcP3b, perm); - srcP3ssA = (vec_s16) vec_mergeh(zero_u8v, srcP3); - srcP3ssB = (vec_s16) vec_mergel(zero_u8v, srcP3); - //srcbis += srcStride; - - sum1A = vec_adds(srcP0ssA, srcP1ssA); - sum1B = vec_adds(srcP0ssB, srcP1ssB); - sum2A = vec_adds(srcM1ssA, srcP2ssA); - sum2B = vec_adds(srcM1ssB, srcP2ssB); - sum3A = vec_adds(srcM2ssA, srcP3ssA); - sum3B = vec_adds(srcM2ssB, srcP3ssB); - - srcM2ssA = srcM1ssA; - srcM2ssB = srcM1ssB; - srcM1ssA = srcP0ssA; - srcM1ssB = srcP0ssB; - srcP0ssA = srcP1ssA; - srcP0ssB = srcP1ssB; - srcP1ssA = srcP2ssA; - srcP1ssB = srcP2ssB; - srcP2ssA = srcP3ssA; - srcP2ssB = srcP3ssB; - - pp1A = vec_mladd(sum1A, v20ss, v16ss); - pp1B = vec_mladd(sum1B, v20ss, v16ss); - - pp2A = vec_mladd(sum2A, v5ss, zero_s16v); - pp2B = vec_mladd(sum2B, v5ss, zero_s16v); - - pp3A = vec_add(sum3A, pp1A); - pp3B = vec_add(sum3B, pp1B); - - psumA = vec_sub(pp3A, pp2A); - psumB = vec_sub(pp3B, pp2B); - - sumA = vec_sra(psumA, v5us); - sumB = vec_sra(psumB, v5us); - - sum = vec_packsu(sumA, sumB); - - ASSERT_ALIGNED(dst); - - OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst)); - - vec_st(fsum, 0, dst); - - dst += dstStride; - } -} -#endif - -/* this code assume stride % 16 == 0 *and* tmp is properly aligned */ -#ifdef PREFIX_h264_qpel16_hv_lowpass_altivec -static void PREFIX_h264_qpel16_hv_lowpass_altivec(uint8_t * dst, int16_t * tmp, uint8_t * src, int dstStride, int tmpStride, int srcStride) { - register int i; - LOAD_ZERO; - const vec_u8 permM2 = vec_lvsl(-2, src); - const vec_u8 permM1 = vec_lvsl(-1, src); - const vec_u8 permP0 = vec_lvsl(+0, src); - const vec_u8 permP1 = vec_lvsl(+1, src); - const vec_u8 permP2 = vec_lvsl(+2, src); - const vec_u8 permP3 = vec_lvsl(+3, src); - const vec_s16 v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2)); - const vec_u32 v10ui = vec_splat_u32(10); - const vec_s16 v5ss = vec_splat_s16(5); - const vec_s16 v1ss = vec_splat_s16(1); - const vec_s32 v512si = vec_sl(vec_splat_s32(1),vec_splat_u32(9)); - const vec_u32 v16ui = vec_sl(vec_splat_u32(1),vec_splat_u32(4)); - - register int align = ((((unsigned long)src) - 2) % 16); - - vec_s16 srcP0A, srcP0B, srcP1A, srcP1B, - srcP2A, srcP2B, srcP3A, srcP3B, - srcM1A, srcM1B, srcM2A, srcM2B, - sum1A, sum1B, sum2A, sum2B, sum3A, sum3B, - pp1A, pp1B, pp2A, pp2B, psumA, psumB; - - const vec_u8 mperm = (const vec_u8) - {0x00, 0x08, 0x01, 0x09, 0x02, 0x0A, 0x03, 0x0B, - 0x04, 0x0C, 0x05, 0x0D, 0x06, 0x0E, 0x07, 0x0F}; - int16_t *tmpbis = tmp; - - vec_s16 tmpM1ssA, tmpM1ssB, tmpM2ssA, tmpM2ssB, - tmpP0ssA, tmpP0ssB, tmpP1ssA, tmpP1ssB, - tmpP2ssA, tmpP2ssB; - - vec_s32 pp1Ae, pp1Ao, pp1Be, pp1Bo, pp2Ae, pp2Ao, pp2Be, pp2Bo, - pp3Ae, pp3Ao, pp3Be, pp3Bo, pp1cAe, pp1cAo, pp1cBe, pp1cBo, - pp32Ae, pp32Ao, pp32Be, pp32Bo, sumAe, sumAo, sumBe, sumBo, - ssumAe, ssumAo, ssumBe, ssumBo; - vec_u8 fsum, sumv, sum; - vec_s16 ssume, ssumo; - - src -= (2 * srcStride); - for (i = 0 ; i < 21 ; i ++) { - vec_u8 srcM2, srcM1, srcP0, srcP1, srcP2, srcP3; - vec_u8 srcR1 = vec_ld(-2, src); - vec_u8 srcR2 = vec_ld(14, src); - - switch (align) { - default: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = vec_perm(srcR1, srcR2, permP3); - } break; - case 11: { - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = vec_perm(srcR1, srcR2, permP2); - srcP3 = srcR2; - } break; - case 12: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = vec_perm(srcR1, srcR2, permP1); - srcP2 = srcR2; - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 13: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = vec_perm(srcR1, srcR2, permP0); - srcP1 = srcR2; - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 14: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = vec_perm(srcR1, srcR2, permM1); - srcP0 = srcR2; - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - case 15: { - vec_u8 srcR3 = vec_ld(30, src); - srcM2 = vec_perm(srcR1, srcR2, permM2); - srcM1 = srcR2; - srcP0 = vec_perm(srcR2, srcR3, permP0); - srcP1 = vec_perm(srcR2, srcR3, permP1); - srcP2 = vec_perm(srcR2, srcR3, permP2); - srcP3 = vec_perm(srcR2, srcR3, permP3); - } break; - } - - srcP0A = (vec_s16) vec_mergeh(zero_u8v, srcP0); - srcP0B = (vec_s16) vec_mergel(zero_u8v, srcP0); - srcP1A = (vec_s16) vec_mergeh(zero_u8v, srcP1); - srcP1B = (vec_s16) vec_mergel(zero_u8v, srcP1); - - srcP2A = (vec_s16) vec_mergeh(zero_u8v, srcP2); - srcP2B = (vec_s16) vec_mergel(zero_u8v, srcP2); - srcP3A = (vec_s16) vec_mergeh(zero_u8v, srcP3); - srcP3B = (vec_s16) vec_mergel(zero_u8v, srcP3); - - srcM1A = (vec_s16) vec_mergeh(zero_u8v, srcM1); - srcM1B = (vec_s16) vec_mergel(zero_u8v, srcM1); - srcM2A = (vec_s16) vec_mergeh(zero_u8v, srcM2); - srcM2B = (vec_s16) vec_mergel(zero_u8v, srcM2); - - sum1A = vec_adds(srcP0A, srcP1A); - sum1B = vec_adds(srcP0B, srcP1B); - sum2A = vec_adds(srcM1A, srcP2A); - sum2B = vec_adds(srcM1B, srcP2B); - sum3A = vec_adds(srcM2A, srcP3A); - sum3B = vec_adds(srcM2B, srcP3B); - - pp1A = vec_mladd(sum1A, v20ss, sum3A); - pp1B = vec_mladd(sum1B, v20ss, sum3B); - - pp2A = vec_mladd(sum2A, v5ss, zero_s16v); - pp2B = vec_mladd(sum2B, v5ss, zero_s16v); - - psumA = vec_sub(pp1A, pp2A); - psumB = vec_sub(pp1B, pp2B); - - vec_st(psumA, 0, tmp); - vec_st(psumB, 16, tmp); - - src += srcStride; - tmp += tmpStride; /* int16_t*, and stride is 16, so it's OK here */ - } - - tmpM2ssA = vec_ld(0, tmpbis); - tmpM2ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - tmpM1ssA = vec_ld(0, tmpbis); - tmpM1ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - tmpP0ssA = vec_ld(0, tmpbis); - tmpP0ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - tmpP1ssA = vec_ld(0, tmpbis); - tmpP1ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - tmpP2ssA = vec_ld(0, tmpbis); - tmpP2ssB = vec_ld(16, tmpbis); - tmpbis += tmpStride; - - for (i = 0 ; i < 16 ; i++) { - const vec_s16 tmpP3ssA = vec_ld(0, tmpbis); - const vec_s16 tmpP3ssB = vec_ld(16, tmpbis); - - const vec_s16 sum1A = vec_adds(tmpP0ssA, tmpP1ssA); - const vec_s16 sum1B = vec_adds(tmpP0ssB, tmpP1ssB); - const vec_s16 sum2A = vec_adds(tmpM1ssA, tmpP2ssA); - const vec_s16 sum2B = vec_adds(tmpM1ssB, tmpP2ssB); - const vec_s16 sum3A = vec_adds(tmpM2ssA, tmpP3ssA); - const vec_s16 sum3B = vec_adds(tmpM2ssB, tmpP3ssB); - - tmpbis += tmpStride; - - tmpM2ssA = tmpM1ssA; - tmpM2ssB = tmpM1ssB; - tmpM1ssA = tmpP0ssA; - tmpM1ssB = tmpP0ssB; - tmpP0ssA = tmpP1ssA; - tmpP0ssB = tmpP1ssB; - tmpP1ssA = tmpP2ssA; - tmpP1ssB = tmpP2ssB; - tmpP2ssA = tmpP3ssA; - tmpP2ssB = tmpP3ssB; - - pp1Ae = vec_mule(sum1A, v20ss); - pp1Ao = vec_mulo(sum1A, v20ss); - pp1Be = vec_mule(sum1B, v20ss); - pp1Bo = vec_mulo(sum1B, v20ss); - - pp2Ae = vec_mule(sum2A, v5ss); - pp2Ao = vec_mulo(sum2A, v5ss); - pp2Be = vec_mule(sum2B, v5ss); - pp2Bo = vec_mulo(sum2B, v5ss); - - pp3Ae = vec_sra((vec_s32)sum3A, v16ui); - pp3Ao = vec_mulo(sum3A, v1ss); - pp3Be = vec_sra((vec_s32)sum3B, v16ui); - pp3Bo = vec_mulo(sum3B, v1ss); - - pp1cAe = vec_add(pp1Ae, v512si); - pp1cAo = vec_add(pp1Ao, v512si); - pp1cBe = vec_add(pp1Be, v512si); - pp1cBo = vec_add(pp1Bo, v512si); - - pp32Ae = vec_sub(pp3Ae, pp2Ae); - pp32Ao = vec_sub(pp3Ao, pp2Ao); - pp32Be = vec_sub(pp3Be, pp2Be); - pp32Bo = vec_sub(pp3Bo, pp2Bo); - - sumAe = vec_add(pp1cAe, pp32Ae); - sumAo = vec_add(pp1cAo, pp32Ao); - sumBe = vec_add(pp1cBe, pp32Be); - sumBo = vec_add(pp1cBo, pp32Bo); - - ssumAe = vec_sra(sumAe, v10ui); - ssumAo = vec_sra(sumAo, v10ui); - ssumBe = vec_sra(sumBe, v10ui); - ssumBo = vec_sra(sumBo, v10ui); - - ssume = vec_packs(ssumAe, ssumBe); - ssumo = vec_packs(ssumAo, ssumBo); - - sumv = vec_packsu(ssume, ssumo); - sum = vec_perm(sumv, sumv, mperm); - - ASSERT_ALIGNED(dst); - - OP_U8_ALTIVEC(fsum, sum, vec_ld(0, dst)); - - vec_st(fsum, 0, dst); - - dst += dstStride; - } -} -#endif diff --git a/ffmpeg/libavcodec/ppc/h264chroma_init.c b/ffmpeg/libavcodec/ppc/h264chroma_init.c index f9e2a76..921f2de 100644 --- a/ffmpeg/libavcodec/ppc/h264chroma_init.c +++ b/ffmpeg/libavcodec/ppc/h264chroma_init.c @@ -20,15 +20,14 @@ #include "config.h" #include "libavutil/attributes.h" -#include "libavcodec/h264chroma.h" - -#if HAVE_ALTIVEC #include "libavutil/cpu.h" #include "libavutil/intreadwrite.h" #include "libavutil/ppc/types_altivec.h" #include "libavutil/ppc/util_altivec.h" +#include "libavcodec/h264chroma.h" #include "dsputil_altivec.h" +#if HAVE_ALTIVEC #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s) @@ -54,11 +53,12 @@ av_cold void ff_h264chroma_init_ppc(H264ChromaContext *c, int bit_depth) #if HAVE_ALTIVEC const int high_bit_depth = bit_depth > 8; - if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) { + if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) + return; + if (!high_bit_depth) { c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec; c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec; } - } #endif /* HAVE_ALTIVEC */ } diff --git a/ffmpeg/libavcodec/ppc/hpeldsp_altivec.c b/ffmpeg/libavcodec/ppc/hpeldsp_altivec.c index 4309d39..345ec39 100644 --- a/ffmpeg/libavcodec/ppc/hpeldsp_altivec.c +++ b/ffmpeg/libavcodec/ppc/hpeldsp_altivec.c @@ -21,17 +21,19 @@ */ #include "config.h" -#include "libavutil/cpu.h" -#include "libavcodec/hpeldsp.h" -#if HAVE_ALTIVEC #if HAVE_ALTIVEC_H #include <altivec.h> #endif + +#include "libavutil/attributes.h" +#include "libavutil/cpu.h" #include "libavutil/ppc/types_altivec.h" #include "libavutil/ppc/util_altivec.h" +#include "libavcodec/hpeldsp.h" #include "dsputil_altivec.h" +#if HAVE_ALTIVEC /* next one assumes that ((line_size % 16) == 0) */ void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h) { @@ -444,21 +446,22 @@ static void avg_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdi } #endif /* HAVE_ALTIVEC */ -void ff_hpeldsp_init_ppc(HpelDSPContext* c, int flags) +av_cold void ff_hpeldsp_init_ppc(HpelDSPContext *c, int flags) { #if HAVE_ALTIVEC - int mm_flags = av_get_cpu_flags(); - - if (mm_flags & AV_CPU_FLAG_ALTIVEC) { - c->put_pixels_tab[0][0] = ff_put_pixels16_altivec; - c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_altivec; - c->avg_pixels_tab[0][0] = ff_avg_pixels16_altivec; - c->avg_pixels_tab[1][0] = avg_pixels8_altivec; - c->avg_pixels_tab[1][3] = avg_pixels8_xy2_altivec; - c->put_pixels_tab[1][3] = put_pixels8_xy2_altivec; - c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_altivec; - c->put_pixels_tab[0][3] = put_pixels16_xy2_altivec; - c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_altivec; - } + if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) + return; + + c->avg_pixels_tab[0][0] = ff_avg_pixels16_altivec; + c->avg_pixels_tab[1][0] = avg_pixels8_altivec; + c->avg_pixels_tab[1][3] = avg_pixels8_xy2_altivec; + + c->put_pixels_tab[0][0] = ff_put_pixels16_altivec; + c->put_pixels_tab[1][3] = put_pixels8_xy2_altivec; + c->put_pixels_tab[0][3] = put_pixels16_xy2_altivec; + + c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_altivec; + c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_altivec; + c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_altivec; #endif /* HAVE_ALTIVEC */ } diff --git a/ffmpeg/libavcodec/ppc/int_altivec.c b/ffmpeg/libavcodec/ppc/int_altivec.c index 4386b13..d4e0c85 100644 --- a/ffmpeg/libavcodec/ppc/int_altivec.c +++ b/ffmpeg/libavcodec/ppc/int_altivec.c @@ -84,14 +84,12 @@ static int32_t scalarproduct_int16_altivec(const int16_t *v1, const int16_t *v2, { int i; LOAD_ZERO; - const vec_s16 *pv; register vec_s16 vec1; register vec_s32 res = vec_splat_s32(0), t; int32_t ires; for(i = 0; i < order; i += 8){ - pv = (const vec_s16*)v1; - vec1 = vec_perm(pv[0], pv[1], vec_lvsl(0, v1)); + vec1 = vec_unaligned_load(v1); t = vec_msum(vec1, vec_ld(0, v2), zero_s32v); res = vec_sums(t, res); v1 += 8; @@ -129,8 +127,8 @@ static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1, const int16_t * pv1[0] = vec_mladd(t0, muls, i0); pv1[1] = vec_mladd(t1, muls, i1); pv1 += 2; - v2 += 8; - v3 += 8; + v2 += 16; + v3 += 16; } while(--order); res = vec_splat(vec_sums(res, zero_s32v), 3); vec_ste(res, 0, &ires); diff --git a/ffmpeg/libavcodec/ppc/mpegaudiodec_altivec.c b/ffmpeg/libavcodec/ppc/mpegaudiodec_altivec.c deleted file mode 100644 index 1152fd7..0000000 --- a/ffmpeg/libavcodec/ppc/mpegaudiodec_altivec.c +++ /dev/null @@ -1,130 +0,0 @@ -/* - * Altivec optimized MP3 decoding functions - * Copyright (c) 2010 Vitor Sessak - * - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#include "dsputil_altivec.h" -#include "libavutil/attributes.h" -#include "libavutil/internal.h" -#include "libavutil/ppc/util_altivec.h" -#include "libavcodec/mpegaudiodsp.h" - -#define MACS(rt, ra, rb) rt+=(ra)*(rb) -#define MLSS(rt, ra, rb) rt-=(ra)*(rb) - -#define SUM8(op, sum, w, p) \ -{ \ - op(sum, (w)[0 * 64], (p)[0 * 64]); \ - op(sum, (w)[1 * 64], (p)[1 * 64]); \ - op(sum, (w)[2 * 64], (p)[2 * 64]); \ - op(sum, (w)[3 * 64], (p)[3 * 64]); \ - op(sum, (w)[4 * 64], (p)[4 * 64]); \ - op(sum, (w)[5 * 64], (p)[5 * 64]); \ - op(sum, (w)[6 * 64], (p)[6 * 64]); \ - op(sum, (w)[7 * 64], (p)[7 * 64]); \ -} - -static void apply_window(const float *buf, const float *win1, - const float *win2, float *sum1, float *sum2, int len) -{ - const vector float *win1a = (const vector float *) win1; - const vector float *win2a = (const vector float *) win2; - const vector float *bufa = (const vector float *) buf; - vector float *sum1a = (vector float *) sum1; - vector float *sum2a = (vector float *) sum2; - vector float av_uninit(v0), av_uninit(v4); - vector float v1, v2, v3; - - len = len >> 2; - -#define MULT(a, b) \ - { \ - v1 = vec_ld(a, win1a); \ - v2 = vec_ld(b, win2a); \ - v3 = vec_ld(a, bufa); \ - v0 = vec_madd(v3, v1, v0); \ - v4 = vec_madd(v2, v3, v4); \ - } - - while (len--) { - v0 = vec_xor(v0, v0); - v4 = vec_xor(v4, v4); - - MULT( 0, 0); - MULT( 256, 64); - MULT( 512, 128); - MULT( 768, 192); - MULT(1024, 256); - MULT(1280, 320); - MULT(1536, 384); - MULT(1792, 448); - - vec_st(v0, 0, sum1a); - vec_st(v4, 0, sum2a); - sum1a++; - sum2a++; - win1a++; - win2a++; - bufa++; - } -} - -static void apply_window_mp3(float *in, float *win, int *unused, float *out, - int incr) -{ - LOCAL_ALIGNED_16(float, suma, [17]); - LOCAL_ALIGNED_16(float, sumb, [17]); - LOCAL_ALIGNED_16(float, sumc, [17]); - LOCAL_ALIGNED_16(float, sumd, [17]); - - float sum; - int j; - float *out2 = out + 32 * incr; - - /* copy to avoid wrap */ - memcpy(in + 512, in, 32 * sizeof(*in)); - - apply_window(in + 16, win , win + 512, suma, sumc, 16); - apply_window(in + 32, win + 48, win + 640, sumb, sumd, 16); - - SUM8(MLSS, suma[0], win + 32, in + 48); - - sumc[ 0] = 0; - sumb[16] = 0; - sumd[16] = 0; - - out[0 ] = suma[ 0]; - out += incr; - out2 -= incr; - for(j=1;j<16;j++) { - *out = suma[ j] - sumd[16-j]; - *out2 = -sumb[16-j] - sumc[ j]; - out += incr; - out2 -= incr; - } - - sum = 0; - SUM8(MLSS, sum, win + 16 + 32, in + 32); - *out = sum; -} - -av_cold void ff_mpadsp_init_altivec(MPADSPContext *s) -{ - s->apply_window_float = apply_window_mp3; -} diff --git a/ffmpeg/libavcodec/ppc/mpegvideo_altivec.c b/ffmpeg/libavcodec/ppc/mpegvideo_altivec.c index bf490b0..cedc1c8 100644 --- a/ffmpeg/libavcodec/ppc/mpegvideo_altivec.c +++ b/ffmpeg/libavcodec/ppc/mpegvideo_altivec.c @@ -24,14 +24,16 @@ #include <stdlib.h> #include <stdio.h> +#include "config.h" #include "libavutil/attributes.h" #include "libavutil/cpu.h" #include "libavutil/ppc/types_altivec.h" #include "libavutil/ppc/util_altivec.h" #include "libavcodec/mpegvideo.h" - #include "dsputil_altivec.h" +#if HAVE_ALTIVEC + /* AltiVec version of dct_unquantize_h263 this code assumes `block' is 16 bytes-aligned */ static void dct_unquantize_h263_altivec(MpegEncContext *s, @@ -111,14 +113,18 @@ static void dct_unquantize_h263_altivec(MpegEncContext *s, } } +#endif /* HAVE_ALTIVEC */ -av_cold void ff_MPV_common_init_altivec(MpegEncContext *s) +av_cold void ff_MPV_common_init_ppc(MpegEncContext *s) { - if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) return; +#if HAVE_ALTIVEC + if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) + return; if ((s->avctx->dct_algo == FF_DCT_AUTO) || - (s->avctx->dct_algo == FF_DCT_ALTIVEC)) { + (s->avctx->dct_algo == FF_DCT_ALTIVEC)) { s->dct_unquantize_h263_intra = dct_unquantize_h263_altivec; s->dct_unquantize_h263_inter = dct_unquantize_h263_altivec; } +#endif /* HAVE_ALTIVEC */ } diff --git a/ffmpeg/libavcodec/ppc/vc1dsp_altivec.c b/ffmpeg/libavcodec/ppc/vc1dsp_altivec.c index 9c2ad70..1b73dd0 100644 --- a/ffmpeg/libavcodec/ppc/vc1dsp_altivec.c +++ b/ffmpeg/libavcodec/ppc/vc1dsp_altivec.c @@ -19,11 +19,14 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "config.h" #include "libavutil/attributes.h" #include "libavutil/ppc/types_altivec.h" #include "libavutil/ppc/util_altivec.h" #include "libavcodec/vc1dsp.h" +#if HAVE_ALTIVEC + // main steps of 8x8 transform #define STEP8(s0, s1, s2, s3, s4, s5, s6, s7, vec_rnd) \ do { \ @@ -335,8 +338,11 @@ static void vc1_inv_trans_8x4_altivec(uint8_t *dest, int stride, int16_t *block) #undef OP_U8_ALTIVEC #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec -av_cold void ff_vc1dsp_init_altivec(VC1DSPContext *dsp) +#endif /* HAVE_ALTIVEC */ + +av_cold void ff_vc1dsp_init_ppc(VC1DSPContext *dsp) { +#if HAVE_ALTIVEC if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) return; @@ -344,4 +350,5 @@ av_cold void ff_vc1dsp_init_altivec(VC1DSPContext *dsp) dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_altivec; dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_altivec; dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_altivec; +#endif /* HAVE_ALTIVEC */ } diff --git a/ffmpeg/libavcodec/ppc/vorbisdsp_altivec.c b/ffmpeg/libavcodec/ppc/vorbisdsp_altivec.c index 08a2b26..d243bf6 100644 --- a/ffmpeg/libavcodec/ppc/vorbisdsp_altivec.c +++ b/ffmpeg/libavcodec/ppc/vorbisdsp_altivec.c @@ -54,8 +54,9 @@ static void vorbis_inverse_coupling_altivec(float *mag, float *ang, av_cold void ff_vorbisdsp_init_ppc(VorbisDSPContext *c) { #if HAVE_ALTIVEC - if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) { - c->vorbis_inverse_coupling = vorbis_inverse_coupling_altivec; - } + if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) + return; + + c->vorbis_inverse_coupling = vorbis_inverse_coupling_altivec; #endif /* HAVE_ALTIVEC */ } diff --git a/ffmpeg/libavcodec/ppc/vp3dsp_altivec.c b/ffmpeg/libavcodec/ppc/vp3dsp_altivec.c index cc587b0..56c2d0b 100644 --- a/ffmpeg/libavcodec/ppc/vp3dsp_altivec.c +++ b/ffmpeg/libavcodec/ppc/vp3dsp_altivec.c @@ -23,14 +23,13 @@ #include "config.h" #include "libavutil/attributes.h" #include "libavutil/cpu.h" -#include "libavcodec/vp3dsp.h" - -#if HAVE_ALTIVEC - #include "libavutil/ppc/types_altivec.h" #include "libavutil/ppc/util_altivec.h" +#include "libavcodec/vp3dsp.h" #include "dsputil_altivec.h" +#if HAVE_ALTIVEC + static const vec_s16 constants = {0, 64277, 60547, 54491, 46341, 36410, 25080, 12785}; static const vec_u8 interleave_high = @@ -181,9 +180,10 @@ static void vp3_idct_add_altivec(uint8_t *dst, int stride, int16_t block[64]) av_cold void ff_vp3dsp_init_ppc(VP3DSPContext *c, int flags) { #if HAVE_ALTIVEC - if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) { - c->idct_put = vp3_idct_put_altivec; - c->idct_add = vp3_idct_add_altivec; - } + if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) + return; + + c->idct_put = vp3_idct_put_altivec; + c->idct_add = vp3_idct_add_altivec; #endif } diff --git a/ffmpeg/libavcodec/ppc/vp8dsp_altivec.c b/ffmpeg/libavcodec/ppc/vp8dsp_altivec.c index 14d8784..c858d8a 100644 --- a/ffmpeg/libavcodec/ppc/vp8dsp_altivec.c +++ b/ffmpeg/libavcodec/ppc/vp8dsp_altivec.c @@ -20,6 +20,7 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "config.h" #include "libavutil/cpu.h" #include "libavutil/mem.h" #include "libavutil/ppc/types_altivec.h" @@ -27,6 +28,7 @@ #include "libavcodec/vp8dsp.h" #include "dsputil_altivec.h" +#if HAVE_ALTIVEC #define REPT4(...) { __VA_ARGS__, __VA_ARGS__, __VA_ARGS__, __VA_ARGS__ } // h subpel filter uses msum to multiply+add 4 pixel taps at once @@ -239,15 +241,15 @@ void put_vp8_epel ## WIDTH ## _v ## TAPS ## _altivec(uint8_t *dst, ptrdiff_t dst } #define EPEL_HV(WIDTH, HTAPS, VTAPS) \ -static void put_vp8_epel ## WIDTH ## _h ## HTAPS ## v ## VTAPS ## _altivec(uint8_t *dst, ptrdiff_t stride, uint8_t *src, ptrdiff_t s, int h, int mx, int my) \ +static void put_vp8_epel ## WIDTH ## _h ## HTAPS ## v ## VTAPS ## _altivec(uint8_t *dst, ptrdiff_t dstride, uint8_t *src, ptrdiff_t sstride, int h, int mx, int my) \ { \ DECLARE_ALIGNED(16, uint8_t, tmp)[(2*WIDTH+5)*16]; \ if (VTAPS == 6) { \ - put_vp8_epel ## WIDTH ## _h ## HTAPS ## _altivec(tmp, 16, src-2*stride, stride, h+5, mx, my); \ - put_vp8_epel ## WIDTH ## _v ## VTAPS ## _altivec(dst, stride, tmp+2*16, 16, h, mx, my); \ + put_vp8_epel ## WIDTH ## _h ## HTAPS ## _altivec(tmp, 16, src-2*sstride, sstride, h+5, mx, my); \ + put_vp8_epel ## WIDTH ## _v ## VTAPS ## _altivec(dst, dstride, tmp+2*16, 16, h, mx, my); \ } else { \ - put_vp8_epel ## WIDTH ## _h ## HTAPS ## _altivec(tmp, 16, src-stride, stride, h+4, mx, my); \ - put_vp8_epel ## WIDTH ## _v ## VTAPS ## _altivec(dst, stride, tmp+16, 16, h, mx, my); \ + put_vp8_epel ## WIDTH ## _h ## HTAPS ## _altivec(tmp, 16, src-sstride, sstride, h+4, mx, my); \ + put_vp8_epel ## WIDTH ## _v ## VTAPS ## _altivec(dst, dstride, tmp+16, 16, h, mx, my); \ } \ } @@ -267,13 +269,51 @@ EPEL_HV(4, 4,6) EPEL_HV(4, 6,4) EPEL_HV(4, 4,4) -static void put_vp8_pixels16_altivec(uint8_t *dst, ptrdiff_t stride, uint8_t *src, ptrdiff_t s, int h, int mx, int my) +static void put_vp8_pixels16_altivec(uint8_t *dst, ptrdiff_t dstride, uint8_t *src, ptrdiff_t sstride, int h, int mx, int my) { - ff_put_pixels16_altivec(dst, src, stride, h); + register vector unsigned char pixelsv1, pixelsv2; + register vector unsigned char pixelsv1B, pixelsv2B; + register vector unsigned char pixelsv1C, pixelsv2C; + register vector unsigned char pixelsv1D, pixelsv2D; + + register vector unsigned char perm = vec_lvsl(0, src); + int i; + register ptrdiff_t dstride2 = dstride << 1, sstride2 = sstride << 1; + register ptrdiff_t dstride3 = dstride2 + dstride, sstride3 = sstride + sstride2; + register ptrdiff_t dstride4 = dstride << 2, sstride4 = sstride << 2; + +// hand-unrolling the loop by 4 gains about 15% +// mininum execution time goes from 74 to 60 cycles +// it's faster than -funroll-loops, but using +// -funroll-loops w/ this is bad - 74 cycles again. +// all this is on a 7450, tuning for the 7450 + for (i = 0; i < h; i += 4) { + pixelsv1 = vec_ld( 0, src); + pixelsv2 = vec_ld(15, src); + pixelsv1B = vec_ld(sstride, src); + pixelsv2B = vec_ld(15 + sstride, src); + pixelsv1C = vec_ld(sstride2, src); + pixelsv2C = vec_ld(15 + sstride2, src); + pixelsv1D = vec_ld(sstride3, src); + pixelsv2D = vec_ld(15 + sstride3, src); + vec_st(vec_perm(pixelsv1, pixelsv2, perm), + 0, (unsigned char*)dst); + vec_st(vec_perm(pixelsv1B, pixelsv2B, perm), + dstride, (unsigned char*)dst); + vec_st(vec_perm(pixelsv1C, pixelsv2C, perm), + dstride2, (unsigned char*)dst); + vec_st(vec_perm(pixelsv1D, pixelsv2D, perm), + dstride3, (unsigned char*)dst); + src += sstride4; + dst += dstride4; + } } -av_cold void ff_vp8dsp_init_altivec(VP8DSPContext *c) +#endif /* HAVE_ALTIVEC */ + +av_cold void ff_vp8dsp_init_ppc(VP8DSPContext *c) { +#if HAVE_ALTIVEC if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) return; @@ -301,4 +341,5 @@ av_cold void ff_vp8dsp_init_altivec(VP8DSPContext *c) c->put_vp8_epel_pixels_tab[2][1][1] = put_vp8_epel4_h4v4_altivec; c->put_vp8_epel_pixels_tab[2][1][2] = put_vp8_epel4_h6v4_altivec; c->put_vp8_epel_pixels_tab[2][2][1] = put_vp8_epel4_h4v6_altivec; +#endif /* HAVE_ALTIVEC */ } |