vp3: use hpeldsp instead of dsputil for half-pel functions.

This makes vp3 independent of dsputil.

configure

@@ -1823,7 +1823,7 @@ vc1_decoder_select="error_resilience h263_decoder h264chroma h264qpel"
 vc1image_decoder_select="vc1_decoder"
 vorbis_decoder_select="mdct"
 vorbis_encoder_select="mdct"
-vp3_decoder_select="dsputil vp3dsp videodsp"
+vp3_decoder_select="hpeldsp vp3dsp videodsp"
 vp5_decoder_select="dsputil h264chroma videodsp vp3dsp"
 vp6_decoder_select="dsputil h264chroma huffman videodsp vp3dsp"
 vp6a_decoder_select="vp6_decoder"

libavcodec/vp3.c

@@ -38,6 +38,7 @@
 #include "internal.h"
 #include "dsputil.h"
 #include "get_bits.h"
+#include "hpeldsp.h"
 #include "videodsp.h"
 #include "vp3data.h"
 #include "vp3dsp.h"
@@ -141,7 +142,7 @@ typedef struct Vp3DecodeContext {
     int keyframe;
     uint8_t idct_permutation[64];
     uint8_t idct_scantable[64];
-    DSPContext dsp;
+    HpelDSPContext hdsp;
     VideoDSPContext vdsp;
     VP3DSPContext vp3dsp;
     DECLARE_ALIGNED(16, int16_t, block)[64];
@@ -1561,7 +1562,7 @@ static void render_slice(Vp3DecodeContext *s, int slice)
                            VP3 source but this would be slower as
                            put_no_rnd_pixels_tab is better optimzed */
                         if(motion_halfpel_index != 3){
-                            s->dsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
+                            s->hdsp.put_no_rnd_pixels_tab[1][motion_halfpel_index](
                                 output_plane + first_pixel,
                                 motion_source, stride, 8);
                         }else{
@@ -1595,7 +1596,7 @@ static void render_slice(Vp3DecodeContext *s, int slice)
                 } else {
 
                     /* copy directly from the previous frame */
-                    s->dsp.put_pixels_tab[1][0](
+                    s->hdsp.put_pixels_tab[1][0](
                         output_plane + first_pixel,
                         last_plane + first_pixel,
                         stride, 8);
@@ -1694,7 +1695,7 @@ static av_cold int vp3_decode_init(AVCodecContext *avctx)
     if (avctx->codec_id != AV_CODEC_ID_THEORA)
         avctx->pix_fmt = AV_PIX_FMT_YUV420P;
     avctx->chroma_sample_location = AVCHROMA_LOC_CENTER;
-    ff_dsputil_init(&s->dsp, avctx);
+    ff_hpeldsp_init(&s->hdsp, avctx->flags | CODEC_FLAG_BITEXACT);
     ff_videodsp_init(&s->vdsp, 8);
     ff_vp3dsp_init(&s->vp3dsp, avctx->flags);
 

libavcodec/x86/dsputil_mmx.c

@@ -52,10 +52,7 @@ DECLARE_ALIGNED(8,  const uint64_t, ff_pw_255)  =   0x00ff00ff00ff00ffULL;
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_512)  = { 0x0200020002000200ULL, 0x0200020002000200ULL };
 DECLARE_ALIGNED(16, const xmm_reg,  ff_pw_1019) = { 0x03FB03FB03FB03FBULL, 0x03FB03FB03FB03FBULL };
 
-DECLARE_ALIGNED(8,  const uint64_t, ff_pb_7)    =   0x0707070707070707ULL;
-DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1F)   =   0x1F1F1F1F1F1F1F1FULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3F)   =   0x3F3F3F3F3F3F3F3FULL;
-DECLARE_ALIGNED(8,  const uint64_t, ff_pb_81)   =   0x8181818181818181ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_FC)   =   0xFCFCFCFCFCFCFCFCULL;
 
 DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };

libavcodec/x86/dsputil_mmx.h

@@ -49,10 +49,7 @@ extern const uint64_t ff_pw_255;
 
 extern const xmm_reg  ff_pb_1;
 extern const xmm_reg  ff_pb_3;
-extern const uint64_t ff_pb_7;
-extern const uint64_t ff_pb_1F;
 extern const uint64_t ff_pb_3F;
-extern const uint64_t ff_pb_81;
 extern const xmm_reg  ff_pb_F8;
 extern const uint64_t ff_pb_FC;
 

libavcodec/x86/vp3dsp.asm

@@ -33,12 +33,13 @@ vp3_idct_data: times 8 dw 64277
                times 8 dw 25080
                times 8 dw 12785
 
+pb_7: times 8 db 7
+pb_1F: times 8 db 0x1f
+pb_81: times 8 db 0x81
+
 cextern pb_1
 cextern pb_3
-cextern pb_7
-cextern pb_1F
 cextern pb_80
-cextern pb_81
 
 cextern pw_8