Merge remote-tracking branch 'qatar/master'

[ffmpeg] / libavcodec / dsputil.c

diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 9821703d37f5e24e2a9433d5ec4bc2c598d9858c..bbcf761552df68d028ce11e97f5d347c4846940e 100644 (file)
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -131,9 +131,6 @@ void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_s
         int j;
         j = src_scantable[i];
         st->permutated[i] = permutation[j];
-#if ARCH_PPC
-        st->inverse[j] = i;
-#endif
     }
 
     end=-1;
@@ -367,18 +364,17 @@ void ff_put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels,
                              int line_size)
 {
     int i;
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 
     /* read the pixels */
     for(i=0;i<8;i++) {
-        pixels[0] = cm[block[0]];
-        pixels[1] = cm[block[1]];
-        pixels[2] = cm[block[2]];
-        pixels[3] = cm[block[3]];
-        pixels[4] = cm[block[4]];
-        pixels[5] = cm[block[5]];
-        pixels[6] = cm[block[6]];
-        pixels[7] = cm[block[7]];
+        pixels[0] = av_clip_uint8(block[0]);
+        pixels[1] = av_clip_uint8(block[1]);
+        pixels[2] = av_clip_uint8(block[2]);
+        pixels[3] = av_clip_uint8(block[3]);
+        pixels[4] = av_clip_uint8(block[4]);
+        pixels[5] = av_clip_uint8(block[5]);
+        pixels[6] = av_clip_uint8(block[6]);
+        pixels[7] = av_clip_uint8(block[7]);
 
         pixels += line_size;
         block += 8;
@@ -389,14 +385,13 @@ static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels
                                  int line_size)
 {
     int i;
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 
     /* read the pixels */
     for(i=0;i<4;i++) {
-        pixels[0] = cm[block[0]];
-        pixels[1] = cm[block[1]];
-        pixels[2] = cm[block[2]];
-        pixels[3] = cm[block[3]];
+        pixels[0] = av_clip_uint8(block[0]);
+        pixels[1] = av_clip_uint8(block[1]);
+        pixels[2] = av_clip_uint8(block[2]);
+        pixels[3] = av_clip_uint8(block[3]);
 
         pixels += line_size;
         block += 8;
@@ -407,12 +402,11 @@ static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels
                                  int line_size)
 {
     int i;
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 
     /* read the pixels */
     for(i=0;i<2;i++) {
-        pixels[0] = cm[block[0]];
-        pixels[1] = cm[block[1]];
+        pixels[0] = av_clip_uint8(block[0]);
+        pixels[1] = av_clip_uint8(block[1]);
 
         pixels += line_size;
         block += 8;
@@ -444,18 +438,17 @@ void ff_add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels,
                              int line_size)
 {
     int i;
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 
     /* read the pixels */
     for(i=0;i<8;i++) {
-        pixels[0] = cm[pixels[0] + block[0]];
-        pixels[1] = cm[pixels[1] + block[1]];
-        pixels[2] = cm[pixels[2] + block[2]];
-        pixels[3] = cm[pixels[3] + block[3]];
-        pixels[4] = cm[pixels[4] + block[4]];
-        pixels[5] = cm[pixels[5] + block[5]];
-        pixels[6] = cm[pixels[6] + block[6]];
-        pixels[7] = cm[pixels[7] + block[7]];
+        pixels[0] = av_clip_uint8(pixels[0] + block[0]);
+        pixels[1] = av_clip_uint8(pixels[1] + block[1]);
+        pixels[2] = av_clip_uint8(pixels[2] + block[2]);
+        pixels[3] = av_clip_uint8(pixels[3] + block[3]);
+        pixels[4] = av_clip_uint8(pixels[4] + block[4]);
+        pixels[5] = av_clip_uint8(pixels[5] + block[5]);
+        pixels[6] = av_clip_uint8(pixels[6] + block[6]);
+        pixels[7] = av_clip_uint8(pixels[7] + block[7]);
         pixels += line_size;
         block += 8;
     }
@@ -465,14 +458,13 @@ static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels
                           int line_size)
 {
     int i;
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 
     /* read the pixels */
     for(i=0;i<4;i++) {
-        pixels[0] = cm[pixels[0] + block[0]];
-        pixels[1] = cm[pixels[1] + block[1]];
-        pixels[2] = cm[pixels[2] + block[2]];
-        pixels[3] = cm[pixels[3] + block[3]];
+        pixels[0] = av_clip_uint8(pixels[0] + block[0]);
+        pixels[1] = av_clip_uint8(pixels[1] + block[1]);
+        pixels[2] = av_clip_uint8(pixels[2] + block[2]);
+        pixels[3] = av_clip_uint8(pixels[3] + block[3]);
         pixels += line_size;
         block += 8;
     }
@@ -482,12 +474,11 @@ static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels
                           int line_size)
 {
     int i;
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
 
     /* read the pixels */
     for(i=0;i<2;i++) {
-        pixels[0] = cm[pixels[0] + block[0]];
-        pixels[1] = cm[pixels[1] + block[1]];
+        pixels[0] = av_clip_uint8(pixels[0] + block[0]);
+        pixels[1] = av_clip_uint8(pixels[1] + block[1]);
         pixels += line_size;
         block += 8;
     }
@@ -1927,7 +1918,7 @@ static void add_bytes_c(uint8_t *dst, uint8_t *src, int w){
         dst[i+0] += src[i+0];
 }
 
-static void diff_bytes_c(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
+static void diff_bytes_c(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w){
     long i;
 #if !HAVE_FAST_UNALIGNED
     if((long)src2 & (sizeof(long)-1)){
@@ -2611,12 +2602,12 @@ static void vector_clipf_c(float *dst, const float *src, float min, float max, i
     }
 }
 
-static int32_t scalarproduct_int16_c(const int16_t * v1, const int16_t * v2, int order, int shift)
+static int32_t scalarproduct_int16_c(const int16_t * v1, const int16_t * v2, int order)
 {
     int res = 0;
 
     while (order--)
-        res += (*v1++ * *v2++) >> shift;
+        res += *v1++ * *v2++;
 
     return res;
 }
@@ -2779,15 +2770,11 @@ static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block)
 
 static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block)
 {
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
-
-    dest[0] = cm[(block[0] + 4)>>3];
+    dest[0] = av_clip_uint8((block[0] + 4)>>3);
 }
 static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block)
 {
-    uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
-
-    dest[0] = cm[dest[0] + ((block[0] + 4)>>3)];
+    dest[0] = av_clip_uint8(dest[0] + ((block[0] + 4)>>3));
 }
 
 static void just_return(void *mem av_unused, int stride av_unused, int h av_unused) { return; }
@@ -2832,7 +2819,7 @@ int ff_check_alignment(void){
 
 av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx)
 {
-    int i;
+    int i, j;
 
     ff_check_alignment();
 
@@ -3191,11 +3178,15 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx)
     if (ARCH_SH4)        ff_dsputil_init_sh4   (c, avctx);
     if (ARCH_BFIN)       ff_dsputil_init_bfin  (c, avctx);
 
-    for(i=0; i<64; i++){
-        if(!c->put_2tap_qpel_pixels_tab[0][i])
-            c->put_2tap_qpel_pixels_tab[0][i]= c->put_h264_qpel_pixels_tab[0][i];
-        if(!c->avg_2tap_qpel_pixels_tab[0][i])
-            c->avg_2tap_qpel_pixels_tab[0][i]= c->avg_h264_qpel_pixels_tab[0][i];
+    for (i = 0; i < 4; i++) {
+        for (j = 0; j < 16; j++) {
+            if(!c->put_2tap_qpel_pixels_tab[i][j])
+                c->put_2tap_qpel_pixels_tab[i][j] =
+                    c->put_h264_qpel_pixels_tab[i][j];
+            if(!c->avg_2tap_qpel_pixels_tab[i][j])
+                c->avg_2tap_qpel_pixels_tab[i][j] =
+                    c->avg_h264_qpel_pixels_tab[i][j];
+        }
     }
 
     ff_init_scantable_permutation(c->idct_permutation,