Merge commit 'a9aee08d900f686e966c64afec5d88a7d9d130a3'

* commit 'a9aee08d900f686e966c64afec5d88a7d9d130a3':
  dsputil: Split off FDCT bits into their own context

Conflicts:
	configure
	libavcodec/Makefile
	libavcodec/asvenc.c
	libavcodec/dnxhdenc.c
	libavcodec/dsputil.c
	libavcodec/mpegvideo.h
	libavcodec/mpegvideo_enc.c
	libavcodec/x86/Makefile
	libavcodec/x86/dsputilenc_mmx.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

configure

@@ -1802,6 +1802,7 @@ CONFIG_EXTRA="
     cabac
     dsputil
     exif
+    fdctdsp
     frame_thread_encoder
     gcrypt
     golomb
@@ -1995,7 +1996,7 @@ threads_if_any="$THREADS_LIST"
 
 # subsystems
 dct_select="rdft"
-dsputil_select="idctdsp"
+dsputil_select="fdctdsp idctdsp"
 error_resilience_select="dsputil"
 frame_thread_encoder_deps="encoders threads"
 intrax8_select="error_resilience"
@@ -2024,9 +2025,9 @@ amv_decoder_select="sp5x_decoder exif"
 amv_encoder_select="aandcttables mpegvideoenc"
 ape_decoder_select="bswapdsp llauddsp"
 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"
@@ -2043,9 +2044,9 @@ cscd_decoder_suggest="zlib"
 dca_decoder_select="mdct"
 dirac_decoder_select="dsputil dwt golomb videodsp"
 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_select="zlib"
 eac3_decoder_select="ac3_decoder"
 eac3_encoder_select="ac3_encoder"
@@ -2134,7 +2135,7 @@ opus_decoder_deps="swresample"
 png_decoder_select="zlib"
 png_encoder_select="huffyuvencdsp zlib"
 prores_decoder_select="blockdsp 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 audiodsp"

libavcodec/Makefile

@@ -41,9 +41,10 @@ OBJS-$(CONFIG_CRYSTALHD)               += crystalhd.o
 OBJS-$(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-$(CONFIG_EXIF)                    += exif.o tiff_common.o
+OBJS-$(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_fixed_32.o fft_init_table.o \

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;

libavcodec/asvenc.c

@@ -28,6 +28,7 @@
 
 #include "asv.h"
 #include "avcodec.h"
+#include "fdctdsp.h"
 #include "internal.h"
 #include "mathops.h"
 #include "mpeg12data.h"
@@ -164,13 +165,13 @@ 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]);
     }
 }
 
@@ -282,6 +283,7 @@ 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;
 

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"
@@ -109,7 +110,7 @@ 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;
@@ -322,6 +323,7 @@ static av_cold int dnxhd_encode_init(AVCodecContext *avctx)
     avctx->bits_per_raw_sample = ctx->cid_table->bit_depth;
 
     ff_blockdsp_init(&ctx->bdsp, 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);

libavcodec/dsputil.c

@@ -29,10 +29,8 @@
 #include "libavutil/internal.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"
 
@@ -589,7 +587,7 @@ static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,
     av_assert2(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);
 }
 
@@ -656,7 +654,7 @@ static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,
     av_assert2(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]));
@@ -973,24 +971,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
 
     ff_check_alignment();
 
-#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;

libavcodec/dsputil.h

@@ -95,10 +95,6 @@ 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);

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);
@@ -66,13 +68,14 @@ static av_cold int dvvideo_encode_init(AVCodecContext *avctx)
 
     memset(&dsp,0, sizeof(dsp));
     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[0]    = fdsp.fdct;
-    s->fdct[1]    = dsp.fdct248;
+    s->fdct[1]    = fdsp.fdct248;
 
     return ff_dvvideo_init(avctx);
 }

libavcodec/fdctdsp.c

@@ -0,0 +1,50 @@
+/*
+ * 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 "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);
+}

libavcodec/fdctdsp.h

@@ -0,0 +1,37 @@
+/*
+ * 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
+ */
+
+#ifndef AVCODEC_FDCTDSP_H
+#define AVCODEC_FDCTDSP_H
+
+#include <stdint.h>
+
+#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 */

libavcodec/libavcodec.v

@@ -28,5 +28,6 @@ LIBAVCODEC_$MAJOR {
                 ff_dnxhd_get_cid_table;
                 ff_dnxhd_cid_table;
                 ff_idctdsp_init;
+                ff_fdctdsp_init;
         local:  *;
 };

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 "h264chroma.h"
 #include "h263dsp.h"
@@ -364,6 +365,7 @@ typedef struct MpegEncContext {
 
     BlockDSPContext bdsp;
     DSPContext dsp;             ///< pointers for accelerated dsp functions
+    FDCTDSPContext fdsp;
     H264ChromaContext h264chroma;
     HpelDSPContext hdsp;
     IDCTDSPContext idsp;

libavcodec/mpegvideo_enc.c

@@ -78,15 +78,15 @@ 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 ||
+        if (fdsp->fdct == ff_jpeg_fdct_islow_8  ||
-            dsp->fdct == ff_jpeg_fdct_islow_10 ||
+            fdsp->fdct == ff_jpeg_fdct_islow_10 ||
-            dsp->fdct == ff_faandct) {
+            fdsp->fdct == ff_faandct) {
             for (i = 0; i < 64; i++) {
                 const int j = s->idsp.idct_permutation[i];
                 /* 16 <= qscale * quant_matrix[i] <= 7905
@@ -98,7 +98,7 @@ 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) {
@@ -818,6 +818,7 @@ av_cold int ff_MPV_encode_init(AVCodecContext *avctx)
     if (ff_MPV_common_init(s) < 0)
         return -1;
 
+    ff_fdctdsp_init(&s->fdsp, avctx);
     ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
     ff_qpeldsp_init(&s->qdsp);
 
@@ -3714,7 +3715,7 @@ 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);
@@ -3809,7 +3810,7 @@ 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;
 
@@ -4141,7 +4142,7 @@ 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
@@ -4388,7 +4389,7 @@ 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);

libavcodec/ppc/Makefile

@@ -9,6 +9,7 @@ OBJS-$(CONFIG_H264DSP)                 += ppc/h264dsp.o ppc/hpeldsp_altivec.o
 OBJS-$(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-$(HAVE_GNU_AS)                += ppc/fft_altivec_s.o
 FFT-OBJS-$(HAVE_VSX)                   += ppc/fft_vsx.o

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);
 

libavcodec/ppc/dsputil_ppc.c

@@ -35,14 +35,5 @@ av_cold void ff_dsputil_init_ppc(DSPContext *c, AVCodecContext *avctx,
     int mm_flags = av_get_cpu_flags();
     if (PPC_ALTIVEC(mm_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
-        }
     }
 }

libavcodec/ppc/fdctdsp.c

@@ -23,8 +23,13 @@
 #include <altivec.h>
 #endif
 
-#include "libavutil/common.h"
+#include "libavutil/attributes.h"
-#include "dsputil_altivec.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 */
+}

libavcodec/ppc/fdctdsp.h

@@ -0,0 +1,26 @@
+/*
+ * 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
+ */
+
+#ifndef AVCODEC_PPC_FDCTDSP_H
+#define AVCODEC_PPC_FDCTDSP_H
+
+#include <stdint.h>
+
+void ff_fdct_altivec(int16_t *block);
+
+#endif /* AVCODEC_PPC_FDCTDSP_H */

libavcodec/proresenc_anatoliy.c

@@ -32,6 +32,7 @@
 #include "put_bits.h"
 #include "bytestream.h"
 #include "dsputil.h"
+#include "fdctdsp.h"
 
 #define DEFAULT_SLICE_MB_WIDTH 8
 
@@ -145,7 +146,7 @@ static const uint8_t QMAT_CHROMA[4][64] = {
 
 
 typedef struct {
-    DSPContext dsp;
+    FDCTDSPContext fdsp;
     uint8_t* fill_y;
     uint8_t* fill_u;
     uint8_t* fill_v;
@@ -274,10 +275,10 @@ static void get(uint8_t *pixels, int stride, int16_t* block)
     }
 }
 
-static void fdct_get(DSPContext *dsp, uint8_t *pixels, int stride, int16_t* block)
+static void fdct_get(FDCTDSPContext *fdsp, uint8_t *pixels, int stride, int16_t* block)
 {
     get(pixels, stride, block);
-    dsp->fdct(block);
+    fdsp->fdct(block);
 }
 
 static int encode_slice_plane(AVCodecContext *avctx, int mb_count,
@@ -285,18 +286,18 @@ static int encode_slice_plane(AVCodecContext *avctx, int mb_count,
         int *qmat, int chroma)
 {
     ProresContext* ctx = avctx->priv_data;
-    DSPContext *dsp = &ctx->dsp;
+    FDCTDSPContext *fdsp = &ctx->fdsp;
     DECLARE_ALIGNED(16, int16_t, blocks)[DEFAULT_SLICE_MB_WIDTH << 8], *block;
     int i, blocks_per_slice;
     PutBitContext pb;
 
     block = blocks;
     for (i = 0; i < mb_count; i++) {
-        fdct_get(dsp, src,                  src_stride, block + (0 << 6));
+        fdct_get(fdsp, src,                  src_stride, block + (0 << 6));
-        fdct_get(dsp, src + 8 * src_stride, src_stride, block + ((2 - chroma) << 6));
+        fdct_get(fdsp, src + 8 * src_stride, src_stride, block + ((2 - chroma) << 6));
         if (!chroma) {
-            fdct_get(dsp, src + 16,                  src_stride, block + (1 << 6));
+            fdct_get(fdsp, src + 16,                  src_stride, block + (1 << 6));
-            fdct_get(dsp, src + 16 + 8 * src_stride, src_stride, block + (3 << 6));
+            fdct_get(fdsp, src + 16 + 8 * src_stride, src_stride, block + (3 << 6));
         }
 
         block += (256 >> chroma);
@@ -576,7 +577,7 @@ static av_cold int prores_encode_init(AVCodecContext *avctx)
         return -1;
     }
 
-    ff_dsputil_init(&ctx->dsp, avctx);
+    ff_fdctdsp_init(&ctx->fdsp, avctx);
 
     avctx->codec_tag = AV_RL32((const uint8_t*)profiles[avctx->profile].name);
 

libavcodec/proresenc_kostya.c

@@ -26,8 +26,7 @@
 #include "libavutil/opt.h"
 #include "libavutil/pixdesc.h"
 #include "avcodec.h"
-#include "dct.h"
+#include "fdctdsp.h"
-#include "dsputil.h"
 #include "put_bits.h"
 #include "bytestream.h"
 #include "internal.h"
@@ -195,9 +194,9 @@ typedef struct ProresContext {
     const uint8_t *quant_mat;
     const uint8_t *scantable;
 
-    void (* fdct)(DSPContext *dsp, const uint16_t *src,
+    void (*fdct)(FDCTDSPContext *fdsp, const uint16_t *src,
-                  int linesize, int16_t *block);
+                 int linesize, int16_t *block);
-    DSPContext dsp;
+    FDCTDSPContext fdsp;
 
     int mb_width, mb_height;
     int mbs_per_slice;
@@ -266,27 +265,27 @@ 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;
             }
         }
@@ -1066,7 +1065,7 @@ 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;
@@ -1077,7 +1076,7 @@ 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)
@@ -1096,7 +1095,7 @@ 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)) {

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_FLAC_DECODER)            += x86/flacdsp_init.o
 OBJS-$(CONFIG_FLAC_ENCODER)            += x86/flacdsp_init.o
@@ -59,7 +60,7 @@ OBJS-$(CONFIG_VP9_DECODER)             += x86/vp9dsp_init.o
 OBJS-$(CONFIG_WEBP_DECODER)            += x86/vp8dsp_init.o
 
 MMX-OBJS-$(CONFIG_DIRAC_DECODER)       += x86/dirac_dwt.o
-MMX-OBJS-$(CONFIG_ENCODERS)            += x86/fdct.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          \

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"
@@ -353,7 +352,6 @@ 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)
@@ -367,10 +365,6 @@ 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->vsad[4] = vsad_intra16_mmx;
 
         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
@@ -379,10 +373,6 @@ 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->vsad[4]         = vsad_intra16_mmxext;
 
         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
@@ -391,9 +381,6 @@ 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;
     }
 
 #if HAVE_SSSE3_INLINE

libavcodec/x86/fdctdsp_init.c

@@ -0,0 +1,44 @@
+/*
+ * 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/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;
+        }
+    }
+}

libavfilter/vf_spp.c

@@ -233,7 +233,7 @@ static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
                 const int y1 = y + offset[i + count - 1][1];
                 const int index = x1 + y1*linesize;
                 p->dsp.get_pixels(block, p->src + index, linesize);
-                p->dsp.fdct(block);
+                p->fdsp.fdct(block);
                 p->requantize(block2, block, qp, p->idsp.idct_permutation);
                 p->idsp.idct(block2);
                 add_block(p->temp + index, linesize, block2);
@@ -382,6 +382,7 @@ static av_cold int init(AVFilterContext *ctx)
         return AVERROR(ENOMEM);
     avpriv_dsputil_init(&spp->dsp, spp->avctx);
     ff_idctdsp_init(&spp->idsp, spp->avctx);
+    ff_fdctdsp_init(&spp->fdsp, spp->avctx);
     spp->store_slice = store_slice_c;
     switch (spp->mode) {
     case MODE_HARD: spp->requantize = hardthresh_c; break;

libavfilter/vf_spp.h

@@ -25,6 +25,7 @@
 #include "libavcodec/avcodec.h"
 #include "libavcodec/dsputil.h"
 #include "libavcodec/idctdsp.h"
+#include "libavcodec/fdctdsp.h"
 #include "avfilter.h"
 
 #define MAX_LEVEL 6 /* quality levels */
@@ -42,6 +43,7 @@ typedef struct {
     AVCodecContext *avctx;
     DSPContext dsp;
     IDCTDSPContext idsp;
+    FDCTDSPContext fdsp;
     int8_t *non_b_qp_table;
     int non_b_qp_alloc_size;
     int use_bframe_qp;