Adds 8-, 9- and 10-bit versions of some of the functions used by the h264 decoder.

This patch lets e.g. dsputil_init chose dsp functions with respect to
the bit depth to decode. The naming scheme of bit depth dependent
functions is <base name>_<bit depth>[_<prefix>] (i.e. the old
clear_blocks_c is now named clear_blocks_8_c).

Note: Some of the functions for high bit depth is not dependent on the
bit depth, but only on the pixel size. This leaves some room for
optimizing binary size.

Preparatory patch for high bit depth h264 decoding support.

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

1 files changed, 69 insertions, 43 deletions

diff --git a/libavcodec/h264dsp.c b/libavcodec/h264dsp.c
index 2a2072b147..04c6ea6df4 100644
--- a/libavcodec/h264dsp.c
+++ b/libavcodec/h264dsp.c
@@ -29,53 +29,79 @@
 #include "avcodec.h"
 #include "h264dsp.h"
 
+#define BIT_DEPTH 8
 #include "h264dsp_internal.h"
+#undef BIT_DEPTH
 
-void ff_h264dsp_init(H264DSPContext *c)
-{
-    c->h264_idct_add= ff_h264_idct_add_c;
-    c->h264_idct8_add= ff_h264_idct8_add_c;
-    c->h264_idct_dc_add= ff_h264_idct_dc_add_c;
-    c->h264_idct8_dc_add= ff_h264_idct8_dc_add_c;
-    c->h264_idct_add16     = ff_h264_idct_add16_c;
-    c->h264_idct8_add4     = ff_h264_idct8_add4_c;
-    c->h264_idct_add8      = ff_h264_idct_add8_c;
-    c->h264_idct_add16intra= ff_h264_idct_add16intra_c;
-    c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_c;
-    c->h264_chroma_dc_dequant_idct= ff_h264_chroma_dc_dequant_idct_c;
+#define BIT_DEPTH 9
+#include "h264dsp_internal.h"
+#undef BIT_DEPTH
+
+#define BIT_DEPTH 10
+#include "h264dsp_internal.h"
+#undef BIT_DEPTH
 
-    c->weight_h264_pixels_tab[0]= weight_h264_pixels16x16_c;
-    c->weight_h264_pixels_tab[1]= weight_h264_pixels16x8_c;
-    c->weight_h264_pixels_tab[2]= weight_h264_pixels8x16_c;
-    c->weight_h264_pixels_tab[3]= weight_h264_pixels8x8_c;
-    c->weight_h264_pixels_tab[4]= weight_h264_pixels8x4_c;
-    c->weight_h264_pixels_tab[5]= weight_h264_pixels4x8_c;
-    c->weight_h264_pixels_tab[6]= weight_h264_pixels4x4_c;
-    c->weight_h264_pixels_tab[7]= weight_h264_pixels4x2_c;
-    c->weight_h264_pixels_tab[8]= weight_h264_pixels2x4_c;
-    c->weight_h264_pixels_tab[9]= weight_h264_pixels2x2_c;
-    c->biweight_h264_pixels_tab[0]= biweight_h264_pixels16x16_c;
-    c->biweight_h264_pixels_tab[1]= biweight_h264_pixels16x8_c;
-    c->biweight_h264_pixels_tab[2]= biweight_h264_pixels8x16_c;
-    c->biweight_h264_pixels_tab[3]= biweight_h264_pixels8x8_c;
-    c->biweight_h264_pixels_tab[4]= biweight_h264_pixels8x4_c;
-    c->biweight_h264_pixels_tab[5]= biweight_h264_pixels4x8_c;
-    c->biweight_h264_pixels_tab[6]= biweight_h264_pixels4x4_c;
-    c->biweight_h264_pixels_tab[7]= biweight_h264_pixels4x2_c;
-    c->biweight_h264_pixels_tab[8]= biweight_h264_pixels2x4_c;
-    c->biweight_h264_pixels_tab[9]= biweight_h264_pixels2x2_c;
+void ff_h264dsp_init(H264DSPContext *c, const int bit_depth)
+{
+#undef FUNC
+#define FUNC(a, depth) a ## _ ## depth ## _c
 
-    c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_c;
-    c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_c;
-    c->h264_v_loop_filter_luma_intra= h264_v_loop_filter_luma_intra_c;
-    c->h264_h_loop_filter_luma_intra= h264_h_loop_filter_luma_intra_c;
-    c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_c;
-    c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_c;
-    c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_c;
-    c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_c;
+#define H264_DSP(depth) \
+    c->h264_idct_add                   = FUNC(ff_h264_idct_add               , depth);\
+    c->h264_idct8_add                  = FUNC(ff_h264_idct8_add              , depth);\
+    c->h264_idct_dc_add                = FUNC(ff_h264_idct_dc_add            , depth);\
+    c->h264_idct8_dc_add               = FUNC(ff_h264_idct8_dc_add           , depth);\
+    c->h264_idct_add16                 = FUNC(ff_h264_idct_add16             , depth);\
+    c->h264_idct8_add4                 = FUNC(ff_h264_idct8_add4             , depth);\
+    c->h264_idct_add8                  = FUNC(ff_h264_idct_add8              , depth);\
+    c->h264_idct_add16intra            = FUNC(ff_h264_idct_add16intra        , depth);\
+    c->h264_luma_dc_dequant_idct       = FUNC(ff_h264_luma_dc_dequant_idct   , depth);\
+    c->h264_chroma_dc_dequant_idct     = FUNC(ff_h264_chroma_dc_dequant_idct , depth);\
+\
+    c->weight_h264_pixels_tab[0]       = FUNC(  weight_h264_pixels16x16      , depth);\
+    c->weight_h264_pixels_tab[1]       = FUNC(  weight_h264_pixels16x8       , depth);\
+    c->weight_h264_pixels_tab[2]       = FUNC(  weight_h264_pixels8x16       , depth);\
+    c->weight_h264_pixels_tab[3]       = FUNC(  weight_h264_pixels8x8        , depth);\
+    c->weight_h264_pixels_tab[4]       = FUNC(  weight_h264_pixels8x4        , depth);\
+    c->weight_h264_pixels_tab[5]       = FUNC(  weight_h264_pixels4x8        , depth);\
+    c->weight_h264_pixels_tab[6]       = FUNC(  weight_h264_pixels4x4        , depth);\
+    c->weight_h264_pixels_tab[7]       = FUNC(  weight_h264_pixels4x2        , depth);\
+    c->weight_h264_pixels_tab[8]       = FUNC(  weight_h264_pixels2x4        , depth);\
+    c->weight_h264_pixels_tab[9]       = FUNC(  weight_h264_pixels2x2        , depth);\
+    c->biweight_h264_pixels_tab[0]     = FUNC(biweight_h264_pixels16x16      , depth);\
+    c->biweight_h264_pixels_tab[1]     = FUNC(biweight_h264_pixels16x8       , depth);\
+    c->biweight_h264_pixels_tab[2]     = FUNC(biweight_h264_pixels8x16       , depth);\
+    c->biweight_h264_pixels_tab[3]     = FUNC(biweight_h264_pixels8x8        , depth);\
+    c->biweight_h264_pixels_tab[4]     = FUNC(biweight_h264_pixels8x4        , depth);\
+    c->biweight_h264_pixels_tab[5]     = FUNC(biweight_h264_pixels4x8        , depth);\
+    c->biweight_h264_pixels_tab[6]     = FUNC(biweight_h264_pixels4x4        , depth);\
+    c->biweight_h264_pixels_tab[7]     = FUNC(biweight_h264_pixels4x2        , depth);\
+    c->biweight_h264_pixels_tab[8]     = FUNC(biweight_h264_pixels2x4        , depth);\
+    c->biweight_h264_pixels_tab[9]     = FUNC(biweight_h264_pixels2x2        , depth);\
+\
+    c->h264_v_loop_filter_luma         = FUNC(h264_v_loop_filter_luma        , depth);\
+    c->h264_h_loop_filter_luma         = FUNC(h264_h_loop_filter_luma        , depth);\
+    c->h264_v_loop_filter_luma_intra   = FUNC(h264_v_loop_filter_luma_intra  , depth);\
+    c->h264_h_loop_filter_luma_intra   = FUNC(h264_h_loop_filter_luma_intra  , depth);\
+    c->h264_v_loop_filter_chroma       = FUNC(h264_v_loop_filter_chroma      , depth);\
+    c->h264_h_loop_filter_chroma       = FUNC(h264_h_loop_filter_chroma      , depth);\
+    c->h264_v_loop_filter_chroma_intra = FUNC(h264_v_loop_filter_chroma_intra, depth);\
+    c->h264_h_loop_filter_chroma_intra = FUNC(h264_h_loop_filter_chroma_intra, depth);\
     c->h264_loop_filter_strength= NULL;
 
-    if (ARCH_ARM) ff_h264dsp_init_arm(c);
-    if (HAVE_ALTIVEC) ff_h264dsp_init_ppc(c);
-    if (HAVE_MMX) ff_h264dsp_init_x86(c);
+    switch (bit_depth) {
+        case 9:
+            H264_DSP(9);
+            break;
+        case 10:
+            H264_DSP(10);
+            break;
+        default:
+            H264_DSP(8);
+            break;
+    }
+
+    if (ARCH_ARM) ff_h264dsp_init_arm(c, bit_depth);
+    if (HAVE_ALTIVEC) ff_h264dsp_init_ppc(c, bit_depth);
+    if (HAVE_MMX) ff_h264dsp_init_x86(c, bit_depth);
 }