From a9aee08d900f686e966c64afec5d88a7d9d130a3 Mon Sep 17 00:00:00 2001 From: Diego Biurrun Date: Mon, 3 Feb 2014 10:09:45 -0800 Subject: dsputil: Split off FDCT bits into their own context --- libavcodec/Makefile | 3 +- libavcodec/asv.h | 2 + libavcodec/asvenc.c | 6 +- libavcodec/dnxhdenc.c | 4 +- libavcodec/dsputil.c | 24 +- libavcodec/dsputil.h | 4 - libavcodec/dvenc.c | 7 +- libavcodec/fdctdsp.c | 50 ++++ libavcodec/fdctdsp.h | 37 +++ libavcodec/mpegvideo.h | 2 + libavcodec/mpegvideo_enc.c | 21 +- libavcodec/ppc/Makefile | 2 +- libavcodec/ppc/dsputil_altivec.h | 2 - libavcodec/ppc/dsputil_ppc.c | 9 - libavcodec/ppc/fdct_altivec.c | 456 ------------------------------------- libavcodec/ppc/fdctdsp.c | 479 +++++++++++++++++++++++++++++++++++++++ libavcodec/ppc/fdctdsp.h | 26 +++ libavcodec/proresenc.c | 31 ++- libavcodec/x86/Makefile | 3 +- libavcodec/x86/dsputilenc_mmx.c | 14 -- libavcodec/x86/fdctdsp_init.c | 44 ++++ 21 files changed, 685 insertions(+), 541 deletions(-) create mode 100644 libavcodec/fdctdsp.c create mode 100644 libavcodec/fdctdsp.h delete mode 100644 libavcodec/ppc/fdct_altivec.c create mode 100644 libavcodec/ppc/fdctdsp.c create mode 100644 libavcodec/ppc/fdctdsp.h create mode 100644 libavcodec/x86/fdctdsp_init.c (limited to 'libavcodec') diff --git a/libavcodec/Makefile b/libavcodec/Makefile index fcd35cd4bd..ab340d79e3 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -35,8 +35,9 @@ OBJS-$(CONFIG_CABAC) += cabac.o OBJS-$(CONFIG_DCT) += dct.o dct32_fixed.o dct32_float.o OBJS-$(CONFIG_DSPUTIL) += dsputil.o OBJS-$(CONFIG_DXVA2) += dxva2.o -OBJS-$(CONFIG_ENCODERS) += faandct.o jfdctfst.o jfdctint.o OBJS-$(CONFIG_ERROR_RESILIENCE) += error_resilience.o +OBJS-$(CONFIG_FDCTDSP) += fdctdsp.o faandct.o \ + jfdctfst.o jfdctint.o FFT-OBJS-$(CONFIG_HARDCODED_TABLES) += cos_tables.o cos_fixed_tables.o OBJS-$(CONFIG_FFT) += avfft.o fft_fixed.o fft_float.o \ $(FFT-OBJS-yes) diff --git a/libavcodec/asv.h b/libavcodec/asv.h index 3f8d56cf8b..71d2583c2c 100644 --- a/libavcodec/asv.h +++ b/libavcodec/asv.h @@ -34,6 +34,7 @@ #include "blockdsp.h" #include "bswapdsp.h" #include "dsputil.h" +#include "fdctdsp.h" #include "idctdsp.h" #include "get_bits.h" #include "put_bits.h" @@ -43,6 +44,7 @@ typedef struct ASV1Context{ BlockDSPContext bdsp; BswapDSPContext bbdsp; DSPContext dsp; + FDCTDSPContext fdsp; IDCTDSPContext idsp; PutBitContext pb; GetBitContext gb; diff --git a/libavcodec/asvenc.c b/libavcodec/asvenc.c index e8c6d00197..1a7021f5a4 100644 --- a/libavcodec/asvenc.c +++ b/libavcodec/asvenc.c @@ -28,6 +28,7 @@ #include "asv.h" #include "avcodec.h" +#include "fdctdsp.h" #include "mathops.h" #include "mpeg12data.h" @@ -163,13 +164,13 @@ static inline void dct_get(ASV1Context *a, const AVFrame *frame, a->dsp.get_pixels(block[2], ptr_y + 8*linesize , linesize); a->dsp.get_pixels(block[3], ptr_y + 8*linesize + 8, linesize); for(i=0; i<4; i++) - a->dsp.fdct(block[i]); + a->fdsp.fdct(block[i]); if(!(a->avctx->flags&CODEC_FLAG_GRAY)){ a->dsp.get_pixels(block[4], ptr_cb, frame->linesize[1]); a->dsp.get_pixels(block[5], ptr_cr, frame->linesize[2]); for(i=4; i<6; i++) - a->dsp.fdct(block[i]); + a->fdsp.fdct(block[i]); } } @@ -248,6 +249,7 @@ static av_cold int encode_init(AVCodecContext *avctx){ ff_asv_common_init(avctx); ff_dsputil_init(&a->dsp, avctx); + ff_fdctdsp_init(&a->fdsp, avctx); if(avctx->global_quality == 0) avctx->global_quality= 4*FF_QUALITY_SCALE; diff --git a/libavcodec/dnxhdenc.c b/libavcodec/dnxhdenc.c index f3c291fa7c..6ba0e12c77 100644 --- a/libavcodec/dnxhdenc.c +++ b/libavcodec/dnxhdenc.c @@ -31,6 +31,7 @@ #include "avcodec.h" #include "blockdsp.h" #include "dsputil.h" +#include "fdctdsp.h" #include "internal.h" #include "mpegvideo.h" #include "dnxhdenc.h" @@ -100,7 +101,7 @@ static int dnxhd_10bit_dct_quantize(MpegEncContext *ctx, int16_t *block, int last_non_zero = 0; int i; - ctx->dsp.fdct(block); + ctx->fdsp.fdct(block); // Divide by 4 with rounding, to compensate scaling of DCT coefficients block[0] = (block[0] + 2) >> 2; @@ -308,6 +309,7 @@ static av_cold int dnxhd_encode_init(AVCodecContext *avctx) ff_blockdsp_init(&ctx->bdsp, avctx); ff_dsputil_init(&ctx->m.dsp, avctx); + ff_fdctdsp_init(&ctx->m.fdsp, avctx); ff_idctdsp_init(&ctx->m.idsp, avctx); ff_mpegvideoencdsp_init(&ctx->m.mpvencdsp, avctx); ff_dct_common_init(&ctx->m); diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c index ec73bbab4e..0dbd91078d 100644 --- a/libavcodec/dsputil.c +++ b/libavcodec/dsputil.c @@ -28,10 +28,8 @@ #include "libavutil/attributes.h" #include "avcodec.h" #include "copy_block.h" -#include "dct.h" #include "dsputil.h" #include "simple_idct.h" -#include "faandct.h" #include "mpegvideo.h" #include "config.h" @@ -580,7 +578,7 @@ static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1, assert(h == 8); s->dsp.diff_pixels(temp, src1, src2, stride); - s->dsp.fdct(temp); + s->fdsp.fdct(temp); return s->dsp.sum_abs_dctelem(temp); } @@ -647,7 +645,7 @@ static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1, assert(h == 8); s->dsp.diff_pixels(temp, src1, src2, stride); - s->dsp.fdct(temp); + s->fdsp.fdct(temp); for (i = 0; i < 64; i++) sum = FFMAX(sum, FFABS(temp[i])); @@ -934,24 +932,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx) { const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8; -#if CONFIG_ENCODERS - if (avctx->bits_per_raw_sample == 10) { - c->fdct = ff_jpeg_fdct_islow_10; - c->fdct248 = ff_fdct248_islow_10; - } else { - if (avctx->dct_algo == FF_DCT_FASTINT) { - c->fdct = ff_fdct_ifast; - c->fdct248 = ff_fdct_ifast248; - } else if (avctx->dct_algo == FF_DCT_FAAN) { - c->fdct = ff_faandct; - c->fdct248 = ff_faandct248; - } else { - c->fdct = ff_jpeg_fdct_islow_8; // slow/accurate/default - c->fdct248 = ff_fdct248_islow_8; - } - } -#endif /* CONFIG_ENCODERS */ - c->diff_pixels = diff_pixels_c; c->sum_abs_dctelem = sum_abs_dctelem_c; diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index 5a83fe0b64..e640f3a86b 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -79,10 +79,6 @@ typedef struct DSPContext { me_cmp_func frame_skip_cmp[6]; // only width 8 used me_cmp_func pix_abs[2][4]; - - /* (I)DCT */ - void (*fdct)(int16_t *block /* align 16 */); - void (*fdct248)(int16_t *block /* align 16 */); } DSPContext; void ff_dsputil_static_init(void); diff --git a/libavcodec/dvenc.c b/libavcodec/dvenc.c index 5b013adf51..7f0320087e 100644 --- a/libavcodec/dvenc.c +++ b/libavcodec/dvenc.c @@ -29,6 +29,7 @@ #include "config.h" #include "avcodec.h" #include "dsputil.h" +#include "fdctdsp.h" #include "internal.h" #include "put_bits.h" #include "dv.h" @@ -38,6 +39,7 @@ static av_cold int dvvideo_encode_init(AVCodecContext *avctx) { DVVideoContext *s = avctx->priv_data; DSPContext dsp; + FDCTDSPContext fdsp; int ret; s->sys = avpriv_dv_codec_profile(avctx); @@ -61,13 +63,14 @@ static av_cold int dvvideo_encode_init(AVCodecContext *avctx) dv_vlc_map_tableinit(); ff_dsputil_init(&dsp, avctx); + ff_fdctdsp_init(&fdsp, avctx); ff_set_cmp(&dsp, dsp.ildct_cmp, avctx->ildct_cmp); s->get_pixels = dsp.get_pixels; s->ildct_cmp = dsp.ildct_cmp[5]; - s->fdct[0] = dsp.fdct; - s->fdct[1] = dsp.fdct248; + s->fdct[0] = fdsp.fdct; + s->fdct[1] = fdsp.fdct248; return ff_dvvideo_init(avctx); } diff --git a/libavcodec/fdctdsp.c b/libavcodec/fdctdsp.c new file mode 100644 index 0000000000..b245198b90 --- /dev/null +++ b/libavcodec/fdctdsp.c @@ -0,0 +1,50 @@ +/* + * This file is part of Libav. + * + * Libav 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. + * + * Libav 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 Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "libavutil/attributes.h" +#include "avcodec.h" +#include "dct.h" +#include "faandct.h" +#include "fdctdsp.h" +#include "config.h" + +av_cold void ff_fdctdsp_init(FDCTDSPContext *c, AVCodecContext *avctx) +{ + const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8; + + if (avctx->bits_per_raw_sample == 10) { + c->fdct = ff_jpeg_fdct_islow_10; + c->fdct248 = ff_fdct248_islow_10; + } else { + if (avctx->dct_algo == FF_DCT_FASTINT) { + c->fdct = ff_fdct_ifast; + c->fdct248 = ff_fdct_ifast248; + } else if (avctx->dct_algo == FF_DCT_FAAN) { + c->fdct = ff_faandct; + c->fdct248 = ff_faandct248; + } else { + c->fdct = ff_jpeg_fdct_islow_8; // slow/accurate/default + c->fdct248 = ff_fdct248_islow_8; + } + } + + if (ARCH_PPC) + ff_fdctdsp_init_ppc(c, avctx, high_bit_depth); + if (ARCH_X86) + ff_fdctdsp_init_x86(c, avctx, high_bit_depth); +} diff --git a/libavcodec/fdctdsp.h b/libavcodec/fdctdsp.h new file mode 100644 index 0000000000..944dc6d1bf --- /dev/null +++ b/libavcodec/fdctdsp.h @@ -0,0 +1,37 @@ +/* + * This file is part of Libav. + * + * Libav 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. + * + * Libav 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 Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVCODEC_FDCTDSP_H +#define AVCODEC_FDCTDSP_H + +#include + +#include "avcodec.h" + +typedef struct FDCTDSPContext { + void (*fdct)(int16_t *block /* align 16 */); + void (*fdct248)(int16_t *block /* align 16 */); +} FDCTDSPContext; + +void ff_fdctdsp_init(FDCTDSPContext *c, AVCodecContext *avctx); +void ff_fdctdsp_init_ppc(FDCTDSPContext *c, AVCodecContext *avctx, + unsigned high_bit_depth); +void ff_fdctdsp_init_x86(FDCTDSPContext *c, AVCodecContext *avctx, + unsigned high_bit_depth); + +#endif /* AVCODEC_FDCTDSP_H */ diff --git a/libavcodec/mpegvideo.h b/libavcodec/mpegvideo.h index ccdc0d851c..b2da21d43b 100644 --- a/libavcodec/mpegvideo.h +++ b/libavcodec/mpegvideo.h @@ -32,6 +32,7 @@ #include "blockdsp.h" #include "dsputil.h" #include "error_resilience.h" +#include "fdctdsp.h" #include "get_bits.h" #include "h263dsp.h" #include "hpeldsp.h" @@ -355,6 +356,7 @@ typedef struct MpegEncContext { BlockDSPContext bdsp; DSPContext dsp; ///< pointers for accelerated dsp functions + FDCTDSPContext fdsp; HpelDSPContext hdsp; IDCTDSPContext idsp; MpegVideoDSPContext mdsp; diff --git a/libavcodec/mpegvideo_enc.c b/libavcodec/mpegvideo_enc.c index 5ea240e84e..0cf7ed4d65 100644 --- a/libavcodec/mpegvideo_enc.c +++ b/libavcodec/mpegvideo_enc.c @@ -77,15 +77,15 @@ void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64], const uint16_t *quant_matrix, int bias, int qmin, int qmax, int intra) { - DSPContext *dsp = &s->dsp; + FDCTDSPContext *fdsp = &s->fdsp; int qscale; int shift = 0; for (qscale = qmin; qscale <= qmax; qscale++) { int i; - if (dsp->fdct == ff_jpeg_fdct_islow_8 || - dsp->fdct == ff_jpeg_fdct_islow_10 || - dsp->fdct == ff_faandct) { + if (fdsp->fdct == ff_jpeg_fdct_islow_8 || + fdsp->fdct == ff_jpeg_fdct_islow_10 || + fdsp->fdct == ff_faandct) { for (i = 0; i < 64; i++) { const int j = s->idsp.idct_permutation[i]; /* 16 <= qscale * quant_matrix[i] <= 7905 @@ -97,7 +97,7 @@ void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64], qmat[qscale][i] = (int)((UINT64_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j])); } - } else if (dsp->fdct == ff_fdct_ifast) { + } else if (fdsp->fdct == ff_fdct_ifast) { for (i = 0; i < 64; i++) { const int j = s->idsp.idct_permutation[i]; /* 16 <= qscale * quant_matrix[i] <= 7905 @@ -136,7 +136,7 @@ void ff_convert_matrix(MpegEncContext *s, int (*qmat)[64], for (i = intra; i < 64; i++) { int64_t max = 8191; - if (dsp->fdct == ff_fdct_ifast) { + if (fdsp->fdct == ff_fdct_ifast) { max = (8191LL * ff_aanscales[i]) >> 14; } while (((max * qmat[qscale][i]) >> shift) > INT_MAX) { @@ -701,6 +701,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx) if (ARCH_X86) ff_MPV_encode_init_x86(s); + ff_fdctdsp_init(&s->fdsp, avctx); ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx); ff_qpeldsp_init(&s->qdsp); @@ -3492,7 +3493,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s, uint8_t * last_length; const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6); - s->dsp.fdct (block); + s->fdsp.fdct(block); if(s->dct_error_sum) s->denoise_dct(s, block); @@ -3587,7 +3588,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s, int dct_coeff= FFABS(block[ scantable[i] ]); int best_score=256*256*256*120; - if (s->dsp.fdct == ff_fdct_ifast) + if (s->fdsp.fdct == ff_fdct_ifast) dct_coeff= (dct_coeff*ff_inv_aanscales[ scantable[i] ]) >> 12; zero_distortion= dct_coeff*dct_coeff; @@ -3919,7 +3920,7 @@ STOP_TIMER("init rem[]") STOP_TIMER("rem*w*w")} {START_TIMER #endif - s->dsp.fdct(d1); + s->fdsp.fdct(d1); #ifdef REFINE_STATS STOP_TIMER("dct")} #endif @@ -4166,7 +4167,7 @@ int ff_dct_quantize_c(MpegEncContext *s, int max=0; unsigned int threshold1, threshold2; - s->dsp.fdct (block); + s->fdsp.fdct(block); if(s->dct_error_sum) s->denoise_dct(s, block); diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile index 256c081dc1..5118b9e6c4 100644 --- a/libavcodec/ppc/Makefile +++ b/libavcodec/ppc/Makefile @@ -9,6 +9,7 @@ OBJS-$(CONFIG_H264DSP) += ppc/h264dsp.o OBJS-$(CONFIG_H264QPEL) += ppc/h264qpel.o OBJS-$(CONFIG_HPELDSP) += ppc/hpeldsp_altivec.o OBJS-$(CONFIG_HUFFYUVDSP) += ppc/huffyuvdsp_altivec.o +OBJS-$(CONFIG_FDCTDSP) += ppc/fdctdsp.o OBJS-$(CONFIG_IDCTDSP) += ppc/idctdsp.o OBJS-$(CONFIG_MPEGAUDIODSP) += ppc/mpegaudiodsp_altivec.o OBJS-$(CONFIG_MPEGVIDEO) += ppc/mpegvideo_altivec.o \ @@ -25,7 +26,6 @@ OBJS-$(CONFIG_VP7_DECODER) += ppc/vp8dsp_altivec.o OBJS-$(CONFIG_VP8_DECODER) += ppc/vp8dsp_altivec.o ALTIVEC-OBJS-$(CONFIG_DSPUTIL) += ppc/dsputil_altivec.o \ - ppc/fdct_altivec.o \ FFT-OBJS-$(HAVE_GNU_AS) += ppc/fft_altivec_s.o ALTIVEC-OBJS-$(CONFIG_FFT) += $(FFT-OBJS-yes) diff --git a/libavcodec/ppc/dsputil_altivec.h b/libavcodec/ppc/dsputil_altivec.h index be5fd58669..bbf9a9d5b7 100644 --- a/libavcodec/ppc/dsputil_altivec.h +++ b/libavcodec/ppc/dsputil_altivec.h @@ -27,8 +27,6 @@ #include "libavcodec/dsputil.h" -void ff_fdct_altivec(int16_t *block); - void ff_dsputil_init_altivec(DSPContext *c, AVCodecContext *avctx, unsigned high_bit_depth); diff --git a/libavcodec/ppc/dsputil_ppc.c b/libavcodec/ppc/dsputil_ppc.c index b54111310e..890157fc1c 100644 --- a/libavcodec/ppc/dsputil_ppc.c +++ b/libavcodec/ppc/dsputil_ppc.c @@ -34,14 +34,5 @@ av_cold void ff_dsputil_init_ppc(DSPContext *c, AVCodecContext *avctx, { if (PPC_ALTIVEC(av_get_cpu_flags())) { ff_dsputil_init_altivec(c, avctx, high_bit_depth); - - if (!high_bit_depth) { -#if CONFIG_ENCODERS - if (avctx->dct_algo == FF_DCT_AUTO || - avctx->dct_algo == FF_DCT_ALTIVEC) { - c->fdct = ff_fdct_altivec; - } -#endif //CONFIG_ENCODERS - } } } diff --git a/libavcodec/ppc/fdct_altivec.c b/libavcodec/ppc/fdct_altivec.c deleted file mode 100644 index 8914d9bf9b..0000000000 --- a/libavcodec/ppc/fdct_altivec.c +++ /dev/null @@ -1,456 +0,0 @@ -/* - * Copyright (C) 2003 James Klicman - * - * This file is part of Libav. - * - * Libav 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. - * - * Libav 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 Libav; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#include "config.h" -#if HAVE_ALTIVEC_H -#include -#endif - -#include "libavutil/common.h" -#include "dsputil_altivec.h" - -#define vs16(v) ((vector signed short) (v)) -#define vs32(v) ((vector signed int) (v)) -#define vu8(v) ((vector unsigned char) (v)) -#define vu16(v) ((vector unsigned short) (v)) -#define vu32(v) ((vector unsigned int) (v)) - -#define C1 0.98078525066375732421875000 /* cos(1 * PI / 16) */ -#define C2 0.92387950420379638671875000 /* cos(2 * PI / 16) */ -#define C3 0.83146959543228149414062500 /* cos(3 * PI / 16) */ -#define C4 0.70710676908493041992187500 /* cos(4 * PI / 16) */ -#define C5 0.55557024478912353515625000 /* cos(5 * PI / 16) */ -#define C6 0.38268342614173889160156250 /* cos(6 * PI / 16) */ -#define C7 0.19509032368659973144531250 /* cos(7 * PI / 16) */ -#define SQRT_2 1.41421353816986083984375000 /* sqrt(2) */ - -#define W0 -(2 * C2) -#define W1 (2 * C6) -#define W2 (SQRT_2 * C6) -#define W3 (SQRT_2 * C3) -#define W4 (SQRT_2 * (-C1 + C3 + C5 - C7)) -#define W5 (SQRT_2 * (C1 + C3 - C5 + C7)) -#define W6 (SQRT_2 * (C1 + C3 + C5 - C7)) -#define W7 (SQRT_2 * (C1 + C3 - C5 - C7)) -#define W8 (SQRT_2 * (C7 - C3)) -#define W9 (SQRT_2 * (-C1 - C3)) -#define WA (SQRT_2 * (-C3 - C5)) -#define WB (SQRT_2 * (C5 - C3)) - -static vector float fdctconsts[3] = { - { W0, W1, W2, W3 }, - { W4, W5, W6, W7 }, - { W8, W9, WA, WB } -}; - -#define LD_W0 vec_splat(cnsts0, 0) -#define LD_W1 vec_splat(cnsts0, 1) -#define LD_W2 vec_splat(cnsts0, 2) -#define LD_W3 vec_splat(cnsts0, 3) -#define LD_W4 vec_splat(cnsts1, 0) -#define LD_W5 vec_splat(cnsts1, 1) -#define LD_W6 vec_splat(cnsts1, 2) -#define LD_W7 vec_splat(cnsts1, 3) -#define LD_W8 vec_splat(cnsts2, 0) -#define LD_W9 vec_splat(cnsts2, 1) -#define LD_WA vec_splat(cnsts2, 2) -#define LD_WB vec_splat(cnsts2, 3) - -#define FDCTROW(b0, b1, b2, b3, b4, b5, b6, b7) /* {{{ */ \ - x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \ - x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \ - x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \ - x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \ - x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \ - x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \ - x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \ - x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \ - \ - b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \ - b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \ - b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \ - b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \ - \ - b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \ - b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \ - b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \ - cnst = LD_W2; \ - b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \ - cnst = LD_W1; \ - b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \ - cnst = LD_W0; \ - b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \ - \ - x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \ - x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \ - x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \ - x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \ - x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \ - cnst = LD_W3; \ - x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \ - \ - cnst = LD_W8; \ - x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \ - cnst = LD_W9; \ - x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \ - cnst = LD_WA; \ - x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \ - cnst = LD_WB; \ - x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \ - \ - cnst = LD_W4; \ - b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \ - cnst = LD_W5; \ - b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \ - cnst = LD_W6; \ - b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \ - cnst = LD_W7; \ - b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \ - \ - b7 = vec_add(b7, x2); /* b7 = b7 + x2; */ \ - b5 = vec_add(b5, x3); /* b5 = b5 + x3; */ \ - b3 = vec_add(b3, x2); /* b3 = b3 + x2; */ \ - b1 = vec_add(b1, x3) /* b1 = b1 + x3; */ \ - /* }}} */ - -#define FDCTCOL(b0, b1, b2, b3, b4, b5, b6, b7) /* {{{ */ \ - x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \ - x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \ - x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \ - x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \ - x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \ - x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \ - x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \ - x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \ - \ - b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \ - b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \ - b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \ - b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \ - \ - b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \ - b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \ - b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \ - cnst = LD_W2; \ - b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \ - cnst = LD_W1; \ - b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \ - cnst = LD_W0; \ - b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \ - \ - x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \ - x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \ - x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \ - x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \ - x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \ - cnst = LD_W3; \ - x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \ - \ - cnst = LD_W8; \ - x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \ - cnst = LD_W9; \ - x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \ - cnst = LD_WA; \ - x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \ - cnst = LD_WB; \ - x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \ - \ - cnst = LD_W4; \ - b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \ - cnst = LD_W5; \ - b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \ - cnst = LD_W6; \ - b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \ - cnst = LD_W7; \ - b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \ - \ - b7 = vec_add(b7, x2); /* b7 += x2; */ \ - b5 = vec_add(b5, x3); /* b5 += x3; */ \ - b3 = vec_add(b3, x2); /* b3 += x2; */ \ - b1 = vec_add(b1, x3) /* b1 += x3; */ \ - /* }}} */ - -/* two dimensional discrete cosine transform */ -void ff_fdct_altivec(int16_t *block) -{ - vector signed short *bp; - vector float *cp = fdctconsts; - vector float b00, b10, b20, b30, b40, b50, b60, b70; - vector float b01, b11, b21, b31, b41, b51, b61, b71; - vector float mzero, cnst, cnsts0, cnsts1, cnsts2; - vector float x0, x1, x2, x3, x4, x5, x6, x7, x8; - - /* setup constants {{{ */ - /* mzero = -0.0 */ - mzero = ((vector float) vec_splat_u32(-1)); - mzero = ((vector float) vec_sl(vu32(mzero), vu32(mzero))); - cnsts0 = vec_ld(0, cp); - cp++; - cnsts1 = vec_ld(0, cp); - cp++; - cnsts2 = vec_ld(0, cp); - /* }}} */ - - /* 8x8 matrix transpose (vector short[8]) {{{ */ -#define MERGE_S16(hl, a, b) vec_merge ## hl(vs16(a), vs16(b)) - - bp = (vector signed short *) block; - b00 = ((vector float) vec_ld(0, bp)); - b40 = ((vector float) vec_ld(16 * 4, bp)); - b01 = ((vector float) MERGE_S16(h, b00, b40)); - b11 = ((vector float) MERGE_S16(l, b00, b40)); - bp++; - b10 = ((vector float) vec_ld(0, bp)); - b50 = ((vector float) vec_ld(16 * 4, bp)); - b21 = ((vector float) MERGE_S16(h, b10, b50)); - b31 = ((vector float) MERGE_S16(l, b10, b50)); - bp++; - b20 = ((vector float) vec_ld(0, bp)); - b60 = ((vector float) vec_ld(16 * 4, bp)); - b41 = ((vector float) MERGE_S16(h, b20, b60)); - b51 = ((vector float) MERGE_S16(l, b20, b60)); - bp++; - b30 = ((vector float) vec_ld(0, bp)); - b70 = ((vector float) vec_ld(16 * 4, bp)); - b61 = ((vector float) MERGE_S16(h, b30, b70)); - b71 = ((vector float) MERGE_S16(l, b30, b70)); - - x0 = ((vector float) MERGE_S16(h, b01, b41)); - x1 = ((vector float) MERGE_S16(l, b01, b41)); - x2 = ((vector float) MERGE_S16(h, b11, b51)); - x3 = ((vector float) MERGE_S16(l, b11, b51)); - x4 = ((vector float) MERGE_S16(h, b21, b61)); - x5 = ((vector float) MERGE_S16(l, b21, b61)); - x6 = ((vector float) MERGE_S16(h, b31, b71)); - x7 = ((vector float) MERGE_S16(l, b31, b71)); - - b00 = ((vector float) MERGE_S16(h, x0, x4)); - b10 = ((vector float) MERGE_S16(l, x0, x4)); - b20 = ((vector float) MERGE_S16(h, x1, x5)); - b30 = ((vector float) MERGE_S16(l, x1, x5)); - b40 = ((vector float) MERGE_S16(h, x2, x6)); - b50 = ((vector float) MERGE_S16(l, x2, x6)); - b60 = ((vector float) MERGE_S16(h, x3, x7)); - b70 = ((vector float) MERGE_S16(l, x3, x7)); - -#undef MERGE_S16 - /* }}} */ - - /* Some of the initial calculations can be done as vector short - * before conversion to vector float. The following code section - * takes advantage of this. */ - - /* fdct rows {{{ */ - x0 = ((vector float) vec_add(vs16(b00), vs16(b70))); - x7 = ((vector float) vec_sub(vs16(b00), vs16(b70))); - x1 = ((vector float) vec_add(vs16(b10), vs16(b60))); - x6 = ((vector float) vec_sub(vs16(b10), vs16(b60))); - x2 = ((vector float) vec_add(vs16(b20), vs16(b50))); - x5 = ((vector float) vec_sub(vs16(b20), vs16(b50))); - x3 = ((vector float) vec_add(vs16(b30), vs16(b40))); - x4 = ((vector float) vec_sub(vs16(b30), vs16(b40))); - - b70 = ((vector float) vec_add(vs16(x0), vs16(x3))); - b10 = ((vector float) vec_add(vs16(x1), vs16(x2))); - - b00 = ((vector float) vec_add(vs16(b70), vs16(b10))); - b40 = ((vector float) vec_sub(vs16(b70), vs16(b10))); - -#define CTF0(n) \ - b ## n ## 1 = ((vector float) vec_unpackl(vs16(b ## n ## 0))); \ - b ## n ## 0 = ((vector float) vec_unpackh(vs16(b ## n ## 0))); \ - b ## n ## 1 = vec_ctf(vs32(b ## n ## 1), 0); \ - b ## n ## 0 = vec_ctf(vs32(b ## n ## 0), 0) - - CTF0(0); - CTF0(4); - - b20 = ((vector float) vec_sub(vs16(x0), vs16(x3))); - b60 = ((vector float) vec_sub(vs16(x1), vs16(x2))); - - CTF0(2); - CTF0(6); - -#undef CTF0 - - x0 = vec_add(b60, b20); - x1 = vec_add(b61, b21); - - cnst = LD_W2; - x0 = vec_madd(cnst, x0, mzero); - x1 = vec_madd(cnst, x1, mzero); - cnst = LD_W1; - b20 = vec_madd(cnst, b20, x0); - b21 = vec_madd(cnst, b21, x1); - cnst = LD_W0; - b60 = vec_madd(cnst, b60, x0); - b61 = vec_madd(cnst, b61, x1); - -#define CTFX(x, b) \ - b ## 0 = ((vector float) vec_unpackh(vs16(x))); \ - b ## 1 = ((vector float) vec_unpackl(vs16(x))); \ - b ## 0 = vec_ctf(vs32(b ## 0), 0); \ - b ## 1 = vec_ctf(vs32(b ## 1), 0) - - CTFX(x4, b7); - CTFX(x5, b5); - CTFX(x6, b3); - CTFX(x7, b1); - -#undef CTFX - - x0 = vec_add(b70, b10); - x1 = vec_add(b50, b30); - x2 = vec_add(b70, b30); - x3 = vec_add(b50, b10); - x8 = vec_add(x2, x3); - cnst = LD_W3; - x8 = vec_madd(cnst, x8, mzero); - - cnst = LD_W8; - x0 = vec_madd(cnst, x0, mzero); - cnst = LD_W9; - x1 = vec_madd(cnst, x1, mzero); - cnst = LD_WA; - x2 = vec_madd(cnst, x2, x8); - cnst = LD_WB; - x3 = vec_madd(cnst, x3, x8); - - cnst = LD_W4; - b70 = vec_madd(cnst, b70, x0); - cnst = LD_W5; - b50 = vec_madd(cnst, b50, x1); - cnst = LD_W6; - b30 = vec_madd(cnst, b30, x1); - cnst = LD_W7; - b10 = vec_madd(cnst, b10, x0); - - b70 = vec_add(b70, x2); - b50 = vec_add(b50, x3); - b30 = vec_add(b30, x2); - b10 = vec_add(b10, x3); - - x0 = vec_add(b71, b11); - x1 = vec_add(b51, b31); - x2 = vec_add(b71, b31); - x3 = vec_add(b51, b11); - x8 = vec_add(x2, x3); - cnst = LD_W3; - x8 = vec_madd(cnst, x8, mzero); - - cnst = LD_W8; - x0 = vec_madd(cnst, x0, mzero); - cnst = LD_W9; - x1 = vec_madd(cnst, x1, mzero); - cnst = LD_WA; - x2 = vec_madd(cnst, x2, x8); - cnst = LD_WB; - x3 = vec_madd(cnst, x3, x8); - - cnst = LD_W4; - b71 = vec_madd(cnst, b71, x0); - cnst = LD_W5; - b51 = vec_madd(cnst, b51, x1); - cnst = LD_W6; - b31 = vec_madd(cnst, b31, x1); - cnst = LD_W7; - b11 = vec_madd(cnst, b11, x0); - - b71 = vec_add(b71, x2); - b51 = vec_add(b51, x3); - b31 = vec_add(b31, x2); - b11 = vec_add(b11, x3); - /* }}} */ - - /* 8x8 matrix transpose (vector float[8][2]) {{{ */ - x0 = vec_mergel(b00, b20); - x1 = vec_mergeh(b00, b20); - x2 = vec_mergel(b10, b30); - x3 = vec_mergeh(b10, b30); - - b00 = vec_mergeh(x1, x3); - b10 = vec_mergel(x1, x3); - b20 = vec_mergeh(x0, x2); - b30 = vec_mergel(x0, x2); - - x4 = vec_mergel(b41, b61); - x5 = vec_mergeh(b41, b61); - x6 = vec_mergel(b51, b71); - x7 = vec_mergeh(b51, b71); - - b41 = vec_mergeh(x5, x7); - b51 = vec_mergel(x5, x7); - b61 = vec_mergeh(x4, x6); - b71 = vec_mergel(x4, x6); - - x0 = vec_mergel(b01, b21); - x1 = vec_mergeh(b01, b21); - x2 = vec_mergel(b11, b31); - x3 = vec_mergeh(b11, b31); - - x4 = vec_mergel(b40, b60); - x5 = vec_mergeh(b40, b60); - x6 = vec_mergel(b50, b70); - x7 = vec_mergeh(b50, b70); - - b40 = vec_mergeh(x1, x3); - b50 = vec_mergel(x1, x3); - b60 = vec_mergeh(x0, x2); - b70 = vec_mergel(x0, x2); - - b01 = vec_mergeh(x5, x7); - b11 = vec_mergel(x5, x7); - b21 = vec_mergeh(x4, x6); - b31 = vec_mergel(x4, x6); - /* }}} */ - - FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70); - FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71); - - /* round, convert back to short {{{ */ -#define CTS(n) \ - b ## n ## 0 = vec_round(b ## n ## 0); \ - b ## n ## 1 = vec_round(b ## n ## 1); \ - b ## n ## 0 = ((vector float) vec_cts(b ## n ## 0, 0)); \ - b ## n ## 1 = ((vector float) vec_cts(b ## n ## 1, 0)); \ - b ## n ## 0 = ((vector float) vec_pack(vs32(b ## n ## 0), \ - vs32(b ## n ## 1))); \ - vec_st(vs16(b ## n ## 0), 0, bp) - - bp = (vector signed short *) block; - CTS(0); - bp++; - CTS(1); - bp++; - CTS(2); - bp++; - CTS(3); - bp++; - CTS(4); - bp++; - CTS(5); - bp++; - CTS(6); - bp++; - CTS(7); - -#undef CTS - /* }}} */ -} diff --git a/libavcodec/ppc/fdctdsp.c b/libavcodec/ppc/fdctdsp.c new file mode 100644 index 0000000000..d7b5b7ab7c --- /dev/null +++ b/libavcodec/ppc/fdctdsp.c @@ -0,0 +1,479 @@ +/* + * Copyright (C) 2003 James Klicman + * + * This file is part of Libav. + * + * Libav 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. + * + * Libav 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 Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config.h" +#if HAVE_ALTIVEC_H +#include +#endif + +#include "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/ppc/cpu.h" +#include "libavcodec/fdctdsp.h" +#include "fdctdsp.h" + +#if HAVE_ALTIVEC + +#define vs16(v) ((vector signed short) (v)) +#define vs32(v) ((vector signed int) (v)) +#define vu8(v) ((vector unsigned char) (v)) +#define vu16(v) ((vector unsigned short) (v)) +#define vu32(v) ((vector unsigned int) (v)) + +#define C1 0.98078525066375732421875000 /* cos(1 * PI / 16) */ +#define C2 0.92387950420379638671875000 /* cos(2 * PI / 16) */ +#define C3 0.83146959543228149414062500 /* cos(3 * PI / 16) */ +#define C4 0.70710676908493041992187500 /* cos(4 * PI / 16) */ +#define C5 0.55557024478912353515625000 /* cos(5 * PI / 16) */ +#define C6 0.38268342614173889160156250 /* cos(6 * PI / 16) */ +#define C7 0.19509032368659973144531250 /* cos(7 * PI / 16) */ +#define SQRT_2 1.41421353816986083984375000 /* sqrt(2) */ + +#define W0 -(2 * C2) +#define W1 (2 * C6) +#define W2 (SQRT_2 * C6) +#define W3 (SQRT_2 * C3) +#define W4 (SQRT_2 * (-C1 + C3 + C5 - C7)) +#define W5 (SQRT_2 * (C1 + C3 - C5 + C7)) +#define W6 (SQRT_2 * (C1 + C3 + C5 - C7)) +#define W7 (SQRT_2 * (C1 + C3 - C5 - C7)) +#define W8 (SQRT_2 * (C7 - C3)) +#define W9 (SQRT_2 * (-C1 - C3)) +#define WA (SQRT_2 * (-C3 - C5)) +#define WB (SQRT_2 * (C5 - C3)) + +static vector float fdctconsts[3] = { + { W0, W1, W2, W3 }, + { W4, W5, W6, W7 }, + { W8, W9, WA, WB } +}; + +#define LD_W0 vec_splat(cnsts0, 0) +#define LD_W1 vec_splat(cnsts0, 1) +#define LD_W2 vec_splat(cnsts0, 2) +#define LD_W3 vec_splat(cnsts0, 3) +#define LD_W4 vec_splat(cnsts1, 0) +#define LD_W5 vec_splat(cnsts1, 1) +#define LD_W6 vec_splat(cnsts1, 2) +#define LD_W7 vec_splat(cnsts1, 3) +#define LD_W8 vec_splat(cnsts2, 0) +#define LD_W9 vec_splat(cnsts2, 1) +#define LD_WA vec_splat(cnsts2, 2) +#define LD_WB vec_splat(cnsts2, 3) + +#define FDCTROW(b0, b1, b2, b3, b4, b5, b6, b7) /* {{{ */ \ + x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \ + x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \ + x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \ + x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \ + x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \ + x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \ + x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \ + x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \ + \ + b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \ + b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \ + b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \ + b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \ + \ + b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \ + b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \ + b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \ + cnst = LD_W2; \ + b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \ + cnst = LD_W1; \ + b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \ + cnst = LD_W0; \ + b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \ + \ + x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \ + x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \ + x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \ + x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \ + x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \ + cnst = LD_W3; \ + x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \ + \ + cnst = LD_W8; \ + x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \ + cnst = LD_W9; \ + x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \ + cnst = LD_WA; \ + x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \ + cnst = LD_WB; \ + x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \ + \ + cnst = LD_W4; \ + b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \ + cnst = LD_W5; \ + b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \ + cnst = LD_W6; \ + b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \ + cnst = LD_W7; \ + b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \ + \ + b7 = vec_add(b7, x2); /* b7 = b7 + x2; */ \ + b5 = vec_add(b5, x3); /* b5 = b5 + x3; */ \ + b3 = vec_add(b3, x2); /* b3 = b3 + x2; */ \ + b1 = vec_add(b1, x3) /* b1 = b1 + x3; */ \ + /* }}} */ + +#define FDCTCOL(b0, b1, b2, b3, b4, b5, b6, b7) /* {{{ */ \ + x0 = vec_add(b0, b7); /* x0 = b0 + b7; */ \ + x7 = vec_sub(b0, b7); /* x7 = b0 - b7; */ \ + x1 = vec_add(b1, b6); /* x1 = b1 + b6; */ \ + x6 = vec_sub(b1, b6); /* x6 = b1 - b6; */ \ + x2 = vec_add(b2, b5); /* x2 = b2 + b5; */ \ + x5 = vec_sub(b2, b5); /* x5 = b2 - b5; */ \ + x3 = vec_add(b3, b4); /* x3 = b3 + b4; */ \ + x4 = vec_sub(b3, b4); /* x4 = b3 - b4; */ \ + \ + b7 = vec_add(x0, x3); /* b7 = x0 + x3; */ \ + b1 = vec_add(x1, x2); /* b1 = x1 + x2; */ \ + b0 = vec_add(b7, b1); /* b0 = b7 + b1; */ \ + b4 = vec_sub(b7, b1); /* b4 = b7 - b1; */ \ + \ + b2 = vec_sub(x0, x3); /* b2 = x0 - x3; */ \ + b6 = vec_sub(x1, x2); /* b6 = x1 - x2; */ \ + b5 = vec_add(b6, b2); /* b5 = b6 + b2; */ \ + cnst = LD_W2; \ + b5 = vec_madd(cnst, b5, mzero); /* b5 = b5 * W2; */ \ + cnst = LD_W1; \ + b2 = vec_madd(cnst, b2, b5); /* b2 = b5 + b2 * W1; */ \ + cnst = LD_W0; \ + b6 = vec_madd(cnst, b6, b5); /* b6 = b5 + b6 * W0; */ \ + \ + x0 = vec_add(x4, x7); /* x0 = x4 + x7; */ \ + x1 = vec_add(x5, x6); /* x1 = x5 + x6; */ \ + x2 = vec_add(x4, x6); /* x2 = x4 + x6; */ \ + x3 = vec_add(x5, x7); /* x3 = x5 + x7; */ \ + x8 = vec_add(x2, x3); /* x8 = x2 + x3; */ \ + cnst = LD_W3; \ + x8 = vec_madd(cnst, x8, mzero); /* x8 = x8 * W3; */ \ + \ + cnst = LD_W8; \ + x0 = vec_madd(cnst, x0, mzero); /* x0 *= W8; */ \ + cnst = LD_W9; \ + x1 = vec_madd(cnst, x1, mzero); /* x1 *= W9; */ \ + cnst = LD_WA; \ + x2 = vec_madd(cnst, x2, x8); /* x2 = x2 * WA + x8; */ \ + cnst = LD_WB; \ + x3 = vec_madd(cnst, x3, x8); /* x3 = x3 * WB + x8; */ \ + \ + cnst = LD_W4; \ + b7 = vec_madd(cnst, x4, x0); /* b7 = x4 * W4 + x0; */ \ + cnst = LD_W5; \ + b5 = vec_madd(cnst, x5, x1); /* b5 = x5 * W5 + x1; */ \ + cnst = LD_W6; \ + b3 = vec_madd(cnst, x6, x1); /* b3 = x6 * W6 + x1; */ \ + cnst = LD_W7; \ + b1 = vec_madd(cnst, x7, x0); /* b1 = x7 * W7 + x0; */ \ + \ + b7 = vec_add(b7, x2); /* b7 += x2; */ \ + b5 = vec_add(b5, x3); /* b5 += x3; */ \ + b3 = vec_add(b3, x2); /* b3 += x2; */ \ + b1 = vec_add(b1, x3) /* b1 += x3; */ \ + /* }}} */ + +/* two dimensional discrete cosine transform */ +void ff_fdct_altivec(int16_t *block) +{ + vector signed short *bp; + vector float *cp = fdctconsts; + vector float b00, b10, b20, b30, b40, b50, b60, b70; + vector float b01, b11, b21, b31, b41, b51, b61, b71; + vector float mzero, cnst, cnsts0, cnsts1, cnsts2; + vector float x0, x1, x2, x3, x4, x5, x6, x7, x8; + + /* setup constants {{{ */ + /* mzero = -0.0 */ + mzero = ((vector float) vec_splat_u32(-1)); + mzero = ((vector float) vec_sl(vu32(mzero), vu32(mzero))); + cnsts0 = vec_ld(0, cp); + cp++; + cnsts1 = vec_ld(0, cp); + cp++; + cnsts2 = vec_ld(0, cp); + /* }}} */ + + /* 8x8 matrix transpose (vector short[8]) {{{ */ +#define MERGE_S16(hl, a, b) vec_merge ## hl(vs16(a), vs16(b)) + + bp = (vector signed short *) block; + b00 = ((vector float) vec_ld(0, bp)); + b40 = ((vector float) vec_ld(16 * 4, bp)); + b01 = ((vector float) MERGE_S16(h, b00, b40)); + b11 = ((vector float) MERGE_S16(l, b00, b40)); + bp++; + b10 = ((vector float) vec_ld(0, bp)); + b50 = ((vector float) vec_ld(16 * 4, bp)); + b21 = ((vector float) MERGE_S16(h, b10, b50)); + b31 = ((vector float) MERGE_S16(l, b10, b50)); + bp++; + b20 = ((vector float) vec_ld(0, bp)); + b60 = ((vector float) vec_ld(16 * 4, bp)); + b41 = ((vector float) MERGE_S16(h, b20, b60)); + b51 = ((vector float) MERGE_S16(l, b20, b60)); + bp++; + b30 = ((vector float) vec_ld(0, bp)); + b70 = ((vector float) vec_ld(16 * 4, bp)); + b61 = ((vector float) MERGE_S16(h, b30, b70)); + b71 = ((vector float) MERGE_S16(l, b30, b70)); + + x0 = ((vector float) MERGE_S16(h, b01, b41)); + x1 = ((vector float) MERGE_S16(l, b01, b41)); + x2 = ((vector float) MERGE_S16(h, b11, b51)); + x3 = ((vector float) MERGE_S16(l, b11, b51)); + x4 = ((vector float) MERGE_S16(h, b21, b61)); + x5 = ((vector float) MERGE_S16(l, b21, b61)); + x6 = ((vector float) MERGE_S16(h, b31, b71)); + x7 = ((vector float) MERGE_S16(l, b31, b71)); + + b00 = ((vector float) MERGE_S16(h, x0, x4)); + b10 = ((vector float) MERGE_S16(l, x0, x4)); + b20 = ((vector float) MERGE_S16(h, x1, x5)); + b30 = ((vector float) MERGE_S16(l, x1, x5)); + b40 = ((vector float) MERGE_S16(h, x2, x6)); + b50 = ((vector float) MERGE_S16(l, x2, x6)); + b60 = ((vector float) MERGE_S16(h, x3, x7)); + b70 = ((vector float) MERGE_S16(l, x3, x7)); + +#undef MERGE_S16 + /* }}} */ + + /* Some of the initial calculations can be done as vector short + * before conversion to vector float. The following code section + * takes advantage of this. */ + + /* fdct rows {{{ */ + x0 = ((vector float) vec_add(vs16(b00), vs16(b70))); + x7 = ((vector float) vec_sub(vs16(b00), vs16(b70))); + x1 = ((vector float) vec_add(vs16(b10), vs16(b60))); + x6 = ((vector float) vec_sub(vs16(b10), vs16(b60))); + x2 = ((vector float) vec_add(vs16(b20), vs16(b50))); + x5 = ((vector float) vec_sub(vs16(b20), vs16(b50))); + x3 = ((vector float) vec_add(vs16(b30), vs16(b40))); + x4 = ((vector float) vec_sub(vs16(b30), vs16(b40))); + + b70 = ((vector float) vec_add(vs16(x0), vs16(x3))); + b10 = ((vector float) vec_add(vs16(x1), vs16(x2))); + + b00 = ((vector float) vec_add(vs16(b70), vs16(b10))); + b40 = ((vector float) vec_sub(vs16(b70), vs16(b10))); + +#define CTF0(n) \ + b ## n ## 1 = ((vector float) vec_unpackl(vs16(b ## n ## 0))); \ + b ## n ## 0 = ((vector float) vec_unpackh(vs16(b ## n ## 0))); \ + b ## n ## 1 = vec_ctf(vs32(b ## n ## 1), 0); \ + b ## n ## 0 = vec_ctf(vs32(b ## n ## 0), 0) + + CTF0(0); + CTF0(4); + + b20 = ((vector float) vec_sub(vs16(x0), vs16(x3))); + b60 = ((vector float) vec_sub(vs16(x1), vs16(x2))); + + CTF0(2); + CTF0(6); + +#undef CTF0 + + x0 = vec_add(b60, b20); + x1 = vec_add(b61, b21); + + cnst = LD_W2; + x0 = vec_madd(cnst, x0, mzero); + x1 = vec_madd(cnst, x1, mzero); + cnst = LD_W1; + b20 = vec_madd(cnst, b20, x0); + b21 = vec_madd(cnst, b21, x1); + cnst = LD_W0; + b60 = vec_madd(cnst, b60, x0); + b61 = vec_madd(cnst, b61, x1); + +#define CTFX(x, b) \ + b ## 0 = ((vector float) vec_unpackh(vs16(x))); \ + b ## 1 = ((vector float) vec_unpackl(vs16(x))); \ + b ## 0 = vec_ctf(vs32(b ## 0), 0); \ + b ## 1 = vec_ctf(vs32(b ## 1), 0) + + CTFX(x4, b7); + CTFX(x5, b5); + CTFX(x6, b3); + CTFX(x7, b1); + +#undef CTFX + + x0 = vec_add(b70, b10); + x1 = vec_add(b50, b30); + x2 = vec_add(b70, b30); + x3 = vec_add(b50, b10); + x8 = vec_add(x2, x3); + cnst = LD_W3; + x8 = vec_madd(cnst, x8, mzero); + + cnst = LD_W8; + x0 = vec_madd(cnst, x0, mzero); + cnst = LD_W9; + x1 = vec_madd(cnst, x1, mzero); + cnst = LD_WA; + x2 = vec_madd(cnst, x2, x8); + cnst = LD_WB; + x3 = vec_madd(cnst, x3, x8); + + cnst = LD_W4; + b70 = vec_madd(cnst, b70, x0); + cnst = LD_W5; + b50 = vec_madd(cnst, b50, x1); + cnst = LD_W6; + b30 = vec_madd(cnst, b30, x1); + cnst = LD_W7; + b10 = vec_madd(cnst, b10, x0); + + b70 = vec_add(b70, x2); + b50 = vec_add(b50, x3); + b30 = vec_add(b30, x2); + b10 = vec_add(b10, x3); + + x0 = vec_add(b71, b11); + x1 = vec_add(b51, b31); + x2 = vec_add(b71, b31); + x3 = vec_add(b51, b11); + x8 = vec_add(x2, x3); + cnst = LD_W3; + x8 = vec_madd(cnst, x8, mzero); + + cnst = LD_W8; + x0 = vec_madd(cnst, x0, mzero); + cnst = LD_W9; + x1 = vec_madd(cnst, x1, mzero); + cnst = LD_WA; + x2 = vec_madd(cnst, x2, x8); + cnst = LD_WB; + x3 = vec_madd(cnst, x3, x8); + + cnst = LD_W4; + b71 = vec_madd(cnst, b71, x0); + cnst = LD_W5; + b51 = vec_madd(cnst, b51, x1); + cnst = LD_W6; + b31 = vec_madd(cnst, b31, x1); + cnst = LD_W7; + b11 = vec_madd(cnst, b11, x0); + + b71 = vec_add(b71, x2); + b51 = vec_add(b51, x3); + b31 = vec_add(b31, x2); + b11 = vec_add(b11, x3); + /* }}} */ + + /* 8x8 matrix transpose (vector float[8][2]) {{{ */ + x0 = vec_mergel(b00, b20); + x1 = vec_mergeh(b00, b20); + x2 = vec_mergel(b10, b30); + x3 = vec_mergeh(b10, b30); + + b00 = vec_mergeh(x1, x3); + b10 = vec_mergel(x1, x3); + b20 = vec_mergeh(x0, x2); + b30 = vec_mergel(x0, x2); + + x4 = vec_mergel(b41, b61); + x5 = vec_mergeh(b41, b61); + x6 = vec_mergel(b51, b71); + x7 = vec_mergeh(b51, b71); + + b41 = vec_mergeh(x5, x7); + b51 = vec_mergel(x5, x7); + b61 = vec_mergeh(x4, x6); + b71 = vec_mergel(x4, x6); + + x0 = vec_mergel(b01, b21); + x1 = vec_mergeh(b01, b21); + x2 = vec_mergel(b11, b31); + x3 = vec_mergeh(b11, b31); + + x4 = vec_mergel(b40, b60); + x5 = vec_mergeh(b40, b60); + x6 = vec_mergel(b50, b70); + x7 = vec_mergeh(b50, b70); + + b40 = vec_mergeh(x1, x3); + b50 = vec_mergel(x1, x3); + b60 = vec_mergeh(x0, x2); + b70 = vec_mergel(x0, x2); + + b01 = vec_mergeh(x5, x7); + b11 = vec_mergel(x5, x7); + b21 = vec_mergeh(x4, x6); + b31 = vec_mergel(x4, x6); + /* }}} */ + + FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70); + FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71); + + /* round, convert back to short {{{ */ +#define CTS(n) \ + b ## n ## 0 = vec_round(b ## n ## 0); \ + b ## n ## 1 = vec_round(b ## n ## 1); \ + b ## n ## 0 = ((vector float) vec_cts(b ## n ## 0, 0)); \ + b ## n ## 1 = ((vector float) vec_cts(b ## n ## 1, 0)); \ + b ## n ## 0 = ((vector float) vec_pack(vs32(b ## n ## 0), \ + vs32(b ## n ## 1))); \ + vec_st(vs16(b ## n ## 0), 0, bp) + + bp = (vector signed short *) block; + CTS(0); + bp++; + CTS(1); + bp++; + CTS(2); + bp++; + CTS(3); + bp++; + CTS(4); + bp++; + CTS(5); + bp++; + CTS(6); + bp++; + CTS(7); + +#undef CTS + /* }}} */ +} + +#endif /* HAVE_ALTIVEC */ + +av_cold void ff_fdctdsp_init_ppc(FDCTDSPContext *c, AVCodecContext *avctx, + unsigned high_bit_depth) +{ +#if HAVE_ALTIVEC + if (!PPC_ALTIVEC(av_get_cpu_flags())) + return; + + if (!high_bit_depth) { + if (avctx->dct_algo == FF_DCT_AUTO || + avctx->dct_algo == FF_DCT_ALTIVEC) { + c->fdct = ff_fdct_altivec; + } + } +#endif /* HAVE_ALTIVEC */ +} diff --git a/libavcodec/ppc/fdctdsp.h b/libavcodec/ppc/fdctdsp.h new file mode 100644 index 0000000000..845b4fe554 --- /dev/null +++ b/libavcodec/ppc/fdctdsp.h @@ -0,0 +1,26 @@ +/* + * This file is part of Libav. + * + * Libav 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. + * + * Libav 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 Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVCODEC_PPC_FDCTDSP_H +#define AVCODEC_PPC_FDCTDSP_H + +#include + +void ff_fdct_altivec(int16_t *block); + +#endif /* AVCODEC_PPC_FDCTDSP_H */ diff --git a/libavcodec/proresenc.c b/libavcodec/proresenc.c index 51e3783e50..bdb826cf07 100644 --- a/libavcodec/proresenc.c +++ b/libavcodec/proresenc.c @@ -23,8 +23,7 @@ #include "libavutil/opt.h" #include "libavutil/pixdesc.h" #include "avcodec.h" -#include "dct.h" -#include "dsputil.h" +#include "fdctdsp.h" #include "put_bits.h" #include "bytestream.h" #include "internal.h" @@ -192,9 +191,9 @@ typedef struct ProresContext { const uint8_t *quant_mat; const uint8_t *scantable; - void (* fdct)(DSPContext *dsp, const uint16_t *src, - int linesize, int16_t *block); - DSPContext dsp; + void (*fdct)(FDCTDSPContext *fdsp, const uint16_t *src, + int linesize, int16_t *block); + FDCTDSPContext fdsp; int mb_width, mb_height; int mbs_per_slice; @@ -263,27 +262,27 @@ static void get_slice_data(ProresContext *ctx, const uint16_t *src, mb_width * sizeof(*emu_buf)); } if (!is_chroma) { - ctx->fdct(&ctx->dsp, esrc, elinesize, blocks); + ctx->fdct(&ctx->fdsp, esrc, elinesize, blocks); blocks += 64; if (blocks_per_mb > 2) { - ctx->fdct(&ctx->dsp, esrc + 8, elinesize, blocks); + ctx->fdct(&ctx->fdsp, esrc + 8, elinesize, blocks); blocks += 64; } - ctx->fdct(&ctx->dsp, esrc + elinesize * 4, elinesize, blocks); + ctx->fdct(&ctx->fdsp, esrc + elinesize * 4, elinesize, blocks); blocks += 64; if (blocks_per_mb > 2) { - ctx->fdct(&ctx->dsp, esrc + elinesize * 4 + 8, elinesize, blocks); + ctx->fdct(&ctx->fdsp, esrc + elinesize * 4 + 8, elinesize, blocks); blocks += 64; } } else { - ctx->fdct(&ctx->dsp, esrc, elinesize, blocks); + ctx->fdct(&ctx->fdsp, esrc, elinesize, blocks); blocks += 64; - ctx->fdct(&ctx->dsp, esrc + elinesize * 4, elinesize, blocks); + ctx->fdct(&ctx->fdsp, esrc + elinesize * 4, elinesize, blocks); blocks += 64; if (blocks_per_mb > 2) { - ctx->fdct(&ctx->dsp, esrc + 8, elinesize, blocks); + ctx->fdct(&ctx->fdsp, esrc + 8, elinesize, blocks); blocks += 64; - ctx->fdct(&ctx->dsp, esrc + elinesize * 4 + 8, elinesize, blocks); + ctx->fdct(&ctx->fdsp, esrc + elinesize * 4 + 8, elinesize, blocks); blocks += 64; } } @@ -1068,7 +1067,7 @@ static av_cold int encode_close(AVCodecContext *avctx) return 0; } -static void prores_fdct(DSPContext *dsp, const uint16_t *src, +static void prores_fdct(FDCTDSPContext *fdsp, const uint16_t *src, int linesize, int16_t *block) { int x, y; @@ -1079,7 +1078,7 @@ static void prores_fdct(DSPContext *dsp, const uint16_t *src, block[y * 8 + x] = tsrc[x]; tsrc += linesize >> 1; } - dsp->fdct(block); + fdsp->fdct(block); } static av_cold int encode_init(AVCodecContext *avctx) @@ -1098,7 +1097,7 @@ static av_cold int encode_init(AVCodecContext *avctx) ctx->fdct = prores_fdct; ctx->scantable = interlaced ? ff_prores_interlaced_scan : ff_prores_progressive_scan; - ff_dsputil_init(&ctx->dsp, avctx); + ff_fdctdsp_init(&ctx->fdsp, avctx); mps = ctx->mbs_per_slice; if (mps & (mps - 1)) { diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index 58b27b5d5b..afb26d07cb 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -9,6 +9,7 @@ OBJS-$(CONFIG_DCT) += x86/dct_init.o OBJS-$(CONFIG_DSPUTIL) += x86/dsputil_init.o OBJS-$(CONFIG_ENCODERS) += x86/dsputilenc_mmx.o \ x86/motion_est.o +OBJS-$(CONFIG_FDCTDSP) += x86/fdctdsp_init.o OBJS-$(CONFIG_FFT) += x86/fft_init.o OBJS-$(CONFIG_H263DSP) += x86/h263dsp_init.o OBJS-$(CONFIG_H264CHROMA) += x86/h264chroma_init.o @@ -51,9 +52,9 @@ OBJS-$(CONFIG_VP8_DECODER) += x86/vp8dsp_init.o OBJS-$(CONFIG_VP9_DECODER) += x86/vp9dsp_init.o MMX-OBJS-$(CONFIG_AUDIODSP) += x86/audiodsp_mmx.o -MMX-OBJS-$(CONFIG_ENCODERS) += x86/fdct.o MMX-OBJS-$(CONFIG_HPELDSP) += x86/fpel_mmx.o \ x86/hpeldsp_mmx.o +MMX-OBJS-$(CONFIG_FDCTDSP) += x86/fdct.o MMX-OBJS-$(CONFIG_IDCTDSP) += x86/idctdsp_mmx.o \ x86/idct_mmx_xvid.o \ x86/idct_sse2_xvid.o \ diff --git a/libavcodec/x86/dsputilenc_mmx.c b/libavcodec/x86/dsputilenc_mmx.c index 563543ee7d..fb649e454e 100644 --- a/libavcodec/x86/dsputilenc_mmx.c +++ b/libavcodec/x86/dsputilenc_mmx.c @@ -26,7 +26,6 @@ #include "libavutil/cpu.h" #include "libavutil/x86/asm.h" #include "libavutil/x86/cpu.h" -#include "libavcodec/dct.h" #include "libavcodec/dsputil.h" #include "libavcodec/mpegvideo.h" #include "dsputil_x86.h" @@ -823,7 +822,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx, unsigned high_bit_depth) { int cpu_flags = av_get_cpu_flags(); - const int dct_algo = avctx->dct_algo; if (EXTERNAL_MMX(cpu_flags)) { if (!high_bit_depth) @@ -837,10 +835,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx, #if HAVE_INLINE_ASM if (INLINE_MMX(cpu_flags)) { - if (!high_bit_depth && - (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX)) - c->fdct = ff_fdct_mmx; - c->sum_abs_dctelem = sum_abs_dctelem_mmx; c->sse[0] = sse16_mmx; @@ -855,10 +849,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx, } if (INLINE_MMXEXT(cpu_flags)) { - if (!high_bit_depth && - (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX)) - c->fdct = ff_fdct_mmxext; - c->sum_abs_dctelem = sum_abs_dctelem_mmxext; c->vsad[4] = vsad_intra16_mmxext; @@ -868,10 +858,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx, } if (INLINE_SSE2(cpu_flags)) { - if (!high_bit_depth && - (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX)) - c->fdct = ff_fdct_sse2; - c->sum_abs_dctelem = sum_abs_dctelem_sse2; } diff --git a/libavcodec/x86/fdctdsp_init.c b/libavcodec/x86/fdctdsp_init.c new file mode 100644 index 0000000000..93fdd6567d --- /dev/null +++ b/libavcodec/x86/fdctdsp_init.c @@ -0,0 +1,44 @@ +/* + * This file is part of Libav. + * + * Libav 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. + * + * Libav 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 Libav; 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/x86/cpu.h" +#include "libavcodec/avcodec.h" +#include "libavcodec/dct.h" +#include "libavcodec/fdctdsp.h" + +av_cold void ff_fdctdsp_init_x86(FDCTDSPContext *c, AVCodecContext *avctx, + unsigned high_bit_depth) +{ + int cpu_flags = av_get_cpu_flags(); + const int dct_algo = avctx->dct_algo; + + if (!high_bit_depth) { + if ((dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX)) { + if (INLINE_MMX(cpu_flags)) + c->fdct = ff_fdct_mmx; + + if (INLINE_MMXEXT(cpu_flags)) + c->fdct = ff_fdct_mmxext; + + if (INLINE_SSE2(cpu_flags)) + c->fdct = ff_fdct_sse2; + } + } +} -- cgit v1.2.3