From a9aee08d900f686e966c64afec5d88a7d9d130a3 Mon Sep 17 00:00:00 2001 From: Diego Biurrun Date: Mon, 3 Feb 2014 10:09:45 -0800 Subject: [PATCH] dsputil: Split off FDCT bits into their own context --- configure | 13 ++--- 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 => fdctdsp.c} | 27 ++++++++++- 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, 238 insertions(+), 93 deletions(-) create mode 100644 libavcodec/fdctdsp.c create mode 100644 libavcodec/fdctdsp.h rename libavcodec/ppc/{fdct_altivec.c => fdctdsp.c} (96%) create mode 100644 libavcodec/ppc/fdctdsp.h create mode 100644 libavcodec/x86/fdctdsp_init.c diff --git a/configure b/configure index be97868031..c4033eae9e 100755 --- a/configure +++ b/configure @@ -1534,6 +1534,7 @@ CONFIG_EXTRA=" bswapdsp cabac dsputil + fdctdsp gcrypt golomb gplv3 @@ -1704,7 +1705,7 @@ threads_if_any="$THREADS_LIST" # subsystems dct_select="rdft" -dsputil_select="idctdsp" +dsputil_select="fdctdsp idctdsp" error_resilience_select="dsputil" intrax8_select="error_resilience" mdct_select="fft" @@ -1730,9 +1731,9 @@ amrwb_decoder_select="lsp" amv_decoder_select="sp5x_decoder" ape_decoder_select="bswapdsp" asv1_decoder_select="blockdsp bswapdsp idctdsp" -asv1_encoder_select="bswapdsp dsputil" +asv1_encoder_select="bswapdsp dsputil fdctdsp" asv2_decoder_select="blockdsp bswapdsp idctdsp" -asv2_encoder_select="bswapdsp dsputil" +asv2_encoder_select="bswapdsp dsputil fdctdsp" atrac1_decoder_select="mdct sinewin" atrac3_decoder_select="mdct" atrac3p_decoder_select="mdct sinewin" @@ -1747,9 +1748,9 @@ cscd_decoder_select="lzo" cscd_decoder_suggest="zlib" dca_decoder_select="mdct" dnxhd_decoder_select="blockdsp idctdsp" -dnxhd_encoder_select="aandcttables blockdsp dsputil idctdsp mpegvideoenc" +dnxhd_encoder_select="aandcttables blockdsp dsputil fdctdsp idctdsp mpegvideoenc" dvvideo_decoder_select="idctdsp" -dvvideo_encoder_select="dsputil" +dvvideo_encoder_select="dsputil fdctdsp" dxa_decoder_deps="zlib" eac3_decoder_select="ac3_decoder" eac3_encoder_select="ac3_encoder" @@ -1838,7 +1839,7 @@ png_decoder_deps="zlib" png_encoder_deps="zlib" png_encoder_select="huffyuvencdsp" prores_decoder_select="idctdsp" -prores_encoder_select="dsputil" +prores_encoder_select="fdctdsp" qcelp_decoder_select="lsp" qdm2_decoder_select="mdct rdft mpegaudiodsp" ra_144_encoder_select="audio_frame_queue lpc" diff --git a/libavcodec/Makefile b/libavcodec/Makefile index fcd35cd4bd..ab340d79e3 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -35,8 +35,9 @@ OBJS-$(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/fdctdsp.c similarity index 96% rename from libavcodec/ppc/fdct_altivec.c rename to libavcodec/ppc/fdctdsp.c index 8914d9bf9b..d7b5b7ab7c 100644 --- a/libavcodec/ppc/fdct_altivec.c +++ b/libavcodec/ppc/fdctdsp.c @@ -23,8 +23,13 @@ #include #endif -#include "libavutil/common.h" -#include "dsputil_altivec.h" +#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)) @@ -454,3 +459,21 @@ void ff_fdct_altivec(int16_t *block) #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; + } + } +}