Merge commit 'a5ba798c16d0614d982a76755fdd72b37d437170'

[ffmpeg] / libavcodec / vp3dsp.c

diff --git a/libavcodec/vp3dsp.c b/libavcodec/vp3dsp.c
index 6502a158e6beaea20f96a5a6f19b416714cec0de..f3f9bd231d58f0a36c1c4ecf08e8f0deda90e676 100644 (file)
--- a/libavcodec/vp3dsp.c
+++ b/libavcodec/vp3dsp.c
@@ -213,18 +213,23 @@ static av_always_inline void idct(uint8_t *dst, int stride, int16_t *input, int
     }
 }
 
-static void vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/){
+static void vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size,
+                           int16_t *block/*align 16*/)
+{
     idct(dest, line_size, block, 1);
     memset(block, 0, sizeof(*block) * 64);
 }
 
-static void vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/){
+static void vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size,
+                           int16_t *block/*align 16*/)
+{
     idct(dest, line_size, block, 2);
     memset(block, 0, sizeof(*block) * 64);
 }
 
 static void vp3_idct_dc_add_c(uint8_t *dest/*align 8*/, int line_size,
-                              DCTELEM *block/*align 16*/){
+                              int16_t *block/*align 16*/)
+{
     int i, dc = (block[0] + 15) >> 5;
 
     for(i = 0; i < 8; i++){
@@ -274,8 +279,27 @@ static void vp3_h_loop_filter_c(uint8_t *first_pixel, int stride,
     }
 }
 
+static void put_no_rnd_pixels_l2(uint8_t *dst, const uint8_t *src1,
+                                 const uint8_t *src2, ptrdiff_t stride, int h)
+{
+    int i;
+
+    for (i = 0; i < h; i++) {
+        uint32_t a, b;
+
+        a = AV_RN32(&src1[i * stride]);
+        b = AV_RN32(&src2[i * stride]);
+        AV_WN32A(&dst[i * stride], no_rnd_avg32(a, b));
+        a = AV_RN32(&src1[i * stride + 4]);
+        b = AV_RN32(&src2[i * stride + 4]);
+        AV_WN32A(&dst[i * stride + 4], no_rnd_avg32(a, b));
+    }
+}
+
 av_cold void ff_vp3dsp_init(VP3DSPContext *c, int flags)
 {
+    c->put_no_rnd_pixels_l2 = put_no_rnd_pixels_l2;
+
     c->idct_put      = vp3_idct_put_c;
     c->idct_add      = vp3_idct_add_c;
     c->idct_dc_add   = vp3_idct_dc_add_c;
@@ -286,6 +310,8 @@ av_cold void ff_vp3dsp_init(VP3DSPContext *c, int flags)
 
     if (ARCH_ARM)
         ff_vp3dsp_init_arm(c, flags);
+    if (ARCH_BFIN)
+        ff_vp3dsp_init_bfin(c, flags);
     if (ARCH_PPC)
         ff_vp3dsp_init_ppc(c, flags);
     if (ARCH_X86)