[FFmpeg-cvslog] r16465 - trunk/libavcodec/ppc/h264_altivec.c

gpoirier subversion
Wed Jan 7 14:54:26 CET 2009


Author: gpoirier
Date: Wed Jan  7 14:54:26 2009
New Revision: 16465

Log:
Add AltiVec versions of h264_idct_add(8|16|16intra),
allowing to re-enable ff_h264_idct_add_altivec's usage.

Patch by David Conrad %lessen42 A gmail P com%

Modified:
   trunk/libavcodec/ppc/h264_altivec.c

Modified: trunk/libavcodec/ppc/h264_altivec.c
==============================================================================
--- trunk/libavcodec/ppc/h264_altivec.c	Wed Jan  7 14:34:29 2009	(r16464)
+++ trunk/libavcodec/ppc/h264_altivec.c	Wed Jan  7 14:54:26 2009	(r16465)
@@ -651,6 +651,25 @@ static void ff_h264_idct8_dc_add_altivec
     h264_idct_dc_add_internal(dst, block, stride, 8);
 }
 
+static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
+    int i;
+    for(i=0; i<16; i++){
+        int nnz = nnzc[ scan8[i] ];
+        if(nnz){
+            if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
+            else                      ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
+        }
+    }
+}
+
+static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
+    int i;
+    for(i=0; i<16; i++){
+        if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride);
+        else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride);
+    }
+}
+
 static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
     int i;
     for(i=0; i<16; i+=4){
@@ -662,6 +681,16 @@ static void ff_h264_idct8_add4_altivec(u
     }
 }
 
+static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){
+    int i;
+    for(i=16; i<16+8; i++){
+        if(nnzc[ scan8[i] ])
+            ff_h264_idct_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
+        else if(block[i*16])
+            h264_idct_dc_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride);
+    }
+}
+
 #define transpose4x16(r0, r1, r2, r3) {      \
     register vec_u8 r4;                    \
     register vec_u8 r5;                    \
@@ -1065,10 +1094,10 @@ void dsputil_h264_init_ppc(DSPContext* c
         c->put_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec;
         c->put_no_rnd_h264_chroma_pixels_tab[0] = put_no_rnd_h264_chroma_mc8_altivec;
         c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec;
-/* ff_h264_idct_add_altivec may be re-enabled once AltiVec versions of
-   h264_idct_add16, h264_idct_add16intra, h264_idct_add8 are implemented
         c->h264_idct_add = ff_h264_idct_add_altivec;
-*/
+        c->h264_idct_add8 = ff_h264_idct_add8_altivec;
+        c->h264_idct_add16 = ff_h264_idct_add16_altivec;
+        c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec;
         c->h264_idct_dc_add= h264_idct_dc_add_altivec;
         c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec;
         c->h264_idct8_add = ff_h264_idct8_add_altivec;




More information about the ffmpeg-cvslog mailing list