MpegEncContext.(i)dct_* -> DspContext.(i)dct_*

bitexact cleanup

Originally committed as revision 1617 to svn://svn.ffmpeg.org/ffmpeg/trunk

4 files changed, 82 insertions, 98 deletions

diff --git a/libavcodec/i386/dsputil_mmx.c b/libavcodec/i386/dsputil_mmx.c
index 8804c1cd21..906a5effa1 100644
--- a/libavcodec/i386/dsputil_mmx.c
+++ b/libavcodec/i386/dsputil_mmx.c
@@ -20,6 +20,7 @@
  */
 
 #include "../dsputil.h"
+#include "../simple_idct.h"
 
 int mm_flags; /* multimedia extension flags */
 
@@ -1408,8 +1409,35 @@ static void just_return() { return; }
     c->put_ ## postfix1 = put_ ## postfix2;\
     c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\
     c->avg_ ## postfix1 = avg_ ## postfix2;
+
+/* external functions, from idct_mmx.c */
+void ff_mmx_idct(DCTELEM *block);
+void ff_mmxext_idct(DCTELEM *block);
+
+/* XXX: those functions should be suppressed ASAP when all IDCTs are
+   converted */
+static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_mmx_idct (block);
+    put_pixels_clamped_mmx(block, dest, line_size);
+}
+static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_mmx_idct (block);
+    add_pixels_clamped_mmx(block, dest, line_size);
+}
+static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_mmxext_idct (block);
+    put_pixels_clamped_mmx(block, dest, line_size);
+}
+static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_mmxext_idct (block);
+    add_pixels_clamped_mmx(block, dest, line_size);
+}
     
-void dsputil_init_mmx(DSPContext* c, unsigned mask)
+void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 {
     mm_flags = mm_support();
 #if 0
@@ -1428,6 +1456,27 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
 #endif
 
     if (mm_flags & MM_MMX) {
+        const int dct_algo = avctx->dct_algo;
+        const int idct_algo= avctx->idct_algo;
+
+        if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX)
+            c->fdct = ff_fdct_mmx;
+
+        if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
+            c->idct_put= ff_simple_idct_put_mmx;
+            c->idct_add= ff_simple_idct_add_mmx;
+            c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
+        }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
+            if(mm_flags & MM_MMXEXT){
+                c->idct_put= ff_libmpeg2mmx2_idct_put;
+                c->idct_add= ff_libmpeg2mmx2_idct_add;
+            }else{
+                c->idct_put= ff_libmpeg2mmx_idct_put;
+                c->idct_add= ff_libmpeg2mmx_idct_add;
+            }
+            c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
+        }
+        
         c->get_pixels = get_pixels_mmx;
         c->diff_pixels = diff_pixels_mmx;
         c->put_pixels_clamped = put_pixels_clamped_mmx;
@@ -1487,23 +1536,26 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
         if (mm_flags & MM_MMXEXT) {
             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
-            c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
-            c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
 
             c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
             c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
             c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
-            c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
 
             c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
             c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
-            c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
-            c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
 
             c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
-            c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
+
+            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
+                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
+                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
+                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
+                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
+                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
+                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
+            }
 
 #if 1
             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)
@@ -1542,23 +1594,26 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
         } else if (mm_flags & MM_3DNOW) {
             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
-            c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
-            c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
 
             c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
             c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
             c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
-            c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
 
             c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
             c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
-            c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
-            c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
 
             c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
-            c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
+
+            if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
+                c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
+                c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
+                c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
+                c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
+                c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
+                c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
+            }
 
             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)
             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)
@@ -1594,7 +1649,8 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
             SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
         }
     }
-    dsputil_init_pix_mmx(c, mask);
+        
+    dsputil_init_pix_mmx(c, avctx);
 #if 0
     // for speed testing
     get_pixels = just_return;
@@ -1630,20 +1686,3 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
     //ff_idct = just_return;
 #endif
 }
-
-/* remove any non bit exact operation (testing purpose). NOTE that
-   this function should be kept as small as possible because it is
-   always difficult to test automatically non bit exact cases. */
-void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask)
-{
-    if (mm_flags & MM_MMX) {
-        /* MMX2 & 3DNOW */
-        c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;
-        c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;
-        c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;
-        c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;
-        c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;
-        c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;
-    }
-    dsputil_set_bit_exact_pix_mmx(c, mask);
-}
diff --git a/libavcodec/i386/motion_est_mmx.c b/libavcodec/i386/motion_est_mmx.c
index 1e6b9f73d2..5c4b32dcd3 100644
--- a/libavcodec/i386/motion_est_mmx.c
+++ b/libavcodec/i386/motion_est_mmx.c
@@ -386,7 +386,7 @@ static int pix_abs16x16_xy2_ ## suf(uint8_t *blk2, uint8_t *blk1, int stride)\
 PIX_SAD(mmx)
 PIX_SAD(mmx2)
 
-void dsputil_init_pix_mmx(DSPContext* c, unsigned mask)
+void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx)
 {
     if (mm_flags & MM_MMX) {
         c->pix_abs16x16     = pix_abs16x16_mmx;
@@ -403,27 +403,18 @@ void dsputil_init_pix_mmx(DSPContext* c, unsigned mask)
     }
     if (mm_flags & MM_MMXEXT) {
 	c->pix_abs16x16     = pix_abs16x16_mmx2;
-	c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx2;
-	c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx2;
-	c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx2;
 	c->pix_abs8x8     = pix_abs8x8_mmx2;
-	c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx2;
-	c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx2;
-	c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx2;
 
 	c->sad[0]= sad16x16_mmx2;
 	c->sad[1]= sad8x8_mmx2;
-    }
-}
-
-void dsputil_set_bit_exact_pix_mmx(DSPContext* c, unsigned mask)
-{
-    if (mm_flags & MM_MMXEXT) {
-	c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx;
-	c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx;
-	c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx;
-	c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx;
-	c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx;
-	c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx;
+        
+        if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
+            c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx2;
+            c->pix_abs16x16_y2  = pix_abs16x16_y2_mmx2;
+            c->pix_abs16x16_xy2 = pix_abs16x16_xy2_mmx2;
+            c->pix_abs8x8_x2  = pix_abs8x8_x2_mmx2;
+            c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx2;
+            c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx2;
+        }
     }
 }
diff --git a/libavcodec/i386/mpegvideo_mmx.c b/libavcodec/i386/mpegvideo_mmx.c
index 23f6831071..8e452b4991 100644
--- a/libavcodec/i386/mpegvideo_mmx.c
+++ b/libavcodec/i386/mpegvideo_mmx.c
@@ -23,7 +23,6 @@
 #include "../dsputil.h"
 #include "../mpegvideo.h"
 #include "../avcodec.h"
-#include "../simple_idct.h"
 
 extern uint8_t zigzag_direct_noperm[64];
 extern uint16_t inv_zigzag_direct16[64];
@@ -499,38 +498,10 @@ static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, int w)
 #define RENAME(a) a ## _MMX2
 #include "mpegvideo_mmx_template.c"
 
-/* external functions, from idct_mmx.c */
-void ff_mmx_idct(DCTELEM *block);
-void ff_mmxext_idct(DCTELEM *block);
-
-/* XXX: those functions should be suppressed ASAP when all IDCTs are
-   converted */
-static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
-{
-    ff_mmx_idct (block);
-    put_pixels_clamped_mmx(block, dest, line_size);
-}
-static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
-{
-    ff_mmx_idct (block);
-    add_pixels_clamped_mmx(block, dest, line_size);
-}
-static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
-{
-    ff_mmxext_idct (block);
-    put_pixels_clamped_mmx(block, dest, line_size);
-}
-static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
-{
-    ff_mmxext_idct (block);
-    add_pixels_clamped_mmx(block, dest, line_size);
-}
-
 void MPV_common_init_mmx(MpegEncContext *s)
 {
     if (mm_flags & MM_MMX) {
         const int dct_algo = s->avctx->dct_algo;
-        const int idct_algo= s->avctx->idct_algo;
         
         s->dct_unquantize_h263 = dct_unquantize_h263_mmx;
         s->dct_unquantize_mpeg1 = dct_unquantize_mpeg1_mmx;
@@ -539,28 +510,11 @@ void MPV_common_init_mmx(MpegEncContext *s)
         draw_edges = draw_edges_mmx;
 
         if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
-            s->fdct = ff_fdct_mmx;
-
             if(mm_flags & MM_MMXEXT){
                 s->dct_quantize= dct_quantize_MMX2;
             } else {
                 s->dct_quantize= dct_quantize_MMX;
             }
         }
-
-        if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
-            s->idct_put= ff_simple_idct_put_mmx;
-            s->idct_add= ff_simple_idct_add_mmx;
-            s->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
-        }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
-            if(mm_flags & MM_MMXEXT){
-                s->idct_put= ff_libmpeg2mmx2_idct_put;
-                s->idct_add= ff_libmpeg2mmx2_idct_add;
-            }else{
-                s->idct_put= ff_libmpeg2mmx_idct_put;
-                s->idct_add= ff_libmpeg2mmx_idct_add;
-            }
-            s->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
-        }
     }
 }
diff --git a/libavcodec/i386/mpegvideo_mmx_template.c b/libavcodec/i386/mpegvideo_mmx_template.c
index 730b912eb8..fa1ab579d9 100644
--- a/libavcodec/i386/mpegvideo_mmx_template.c
+++ b/libavcodec/i386/mpegvideo_mmx_template.c
@@ -191,7 +191,7 @@ static int RENAME(dct_quantize)(MpegEncContext *s,
     if(s->mb_intra) block[0]= level;
     else            block[0]= temp_block[0];
 
-    if(s->idct_permutation[1]==8){
+    if(s->dsp.idct_permutation_type == FF_SIMPLE_IDCT_PERM){
         if(last_non_zero_p1 <= 1) goto end;
         block[0x08] = temp_block[0x01]; block[0x10] = temp_block[0x08]; 
         block[0x20] = temp_block[0x10]; 
@@ -235,7 +235,7 @@ static int RENAME(dct_quantize)(MpegEncContext *s,
         block[0x3E] = temp_block[0x3D]; block[0x27] = temp_block[0x36]; 
         block[0x3D] = temp_block[0x2F]; block[0x2F] = temp_block[0x37]; 
         block[0x37] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
-    }else if(s->idct_permutation[1]==4){
+    }else if(s->dsp.idct_permutation_type == FF_LIBMPEG2_IDCT_PERM){
         if(last_non_zero_p1 <= 1) goto end;
         block[0x04] = temp_block[0x01]; 
         block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10];