lowres2 support.

The new lowres support is limited to decoders where lowres decoding
is possible in high quality.
I was not able to measure any speed difference, but if one is found
the 2-3 lines that might affect speed can be made compile time conditional

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

1 files changed, 116 insertions, 6 deletions

diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 553fb8c1cd..a13737d77e 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -381,6 +381,38 @@ void ff_put_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels,
     }
 }
 
+static void put_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels,
+                                 int line_size)
+{
+    int i;
+
+    /* read the pixels */
+    for(i=0;i<4;i++) {
+        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;
+    }
+}
+
+static void put_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels,
+                                 int line_size)
+{
+    int i;
+
+    /* read the pixels */
+    for(i=0;i<2;i++) {
+        pixels[0] = av_clip_uint8(block[0]);
+        pixels[1] = av_clip_uint8(block[1]);
+
+        pixels += line_size;
+        block += 8;
+    }
+}
+
 void ff_put_signed_pixels_clamped_c(const DCTELEM *block,
                                     uint8_t *restrict pixels,
                                     int line_size)
@@ -422,6 +454,36 @@ void ff_add_pixels_clamped_c(const DCTELEM *block, uint8_t *restrict pixels,
     }
 }
 
+static void add_pixels_clamped4_c(const DCTELEM *block, uint8_t *restrict pixels,
+                          int line_size)
+{
+    int i;
+
+    /* read the pixels */
+    for(i=0;i<4;i++) {
+        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;
+    }
+}
+
+static void add_pixels_clamped2_c(const DCTELEM *block, uint8_t *restrict pixels,
+                          int line_size)
+{
+    int i;
+
+    /* read the pixels */
+    for(i=0;i<2;i++) {
+        pixels[0] = av_clip_uint8(pixels[0] + block[0]);
+        pixels[1] = av_clip_uint8(pixels[1] + block[1]);
+        pixels += line_size;
+        block += 8;
+    }
+}
+
 static int sum_abs_dctelem_c(DCTELEM *block)
 {
     int sum=0, i;
@@ -2684,6 +2746,37 @@ static void ff_jref_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
     ff_add_pixels_clamped_c(block, dest, line_size);
 }
 
+static void ff_jref_idct4_put(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_j_rev_dct4 (block);
+    put_pixels_clamped4_c(block, dest, line_size);
+}
+static void ff_jref_idct4_add(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_j_rev_dct4 (block);
+    add_pixels_clamped4_c(block, dest, line_size);
+}
+
+static void ff_jref_idct2_put(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_j_rev_dct2 (block);
+    put_pixels_clamped2_c(block, dest, line_size);
+}
+static void ff_jref_idct2_add(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_j_rev_dct2 (block);
+    add_pixels_clamped2_c(block, dest, line_size);
+}
+
+static void ff_jref_idct1_put(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    dest[0] = av_clip_uint8((block[0] + 4)>>3);
+}
+static void ff_jref_idct1_add(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    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; }
 
 /* init static data */
@@ -2750,12 +2843,28 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx)
     }
 #endif //CONFIG_ENCODERS
 
-    if (avctx->bits_per_raw_sample == 10) {
-        c->idct_put              = ff_simple_idct_put_10;
-        c->idct_add              = ff_simple_idct_add_10;
-        c->idct                  = ff_simple_idct_10;
-        c->idct_permutation_type = FF_NO_IDCT_PERM;
-    } else {
+    if(avctx->lowres==1){
+        c->idct_put= ff_jref_idct4_put;
+        c->idct_add= ff_jref_idct4_add;
+        c->idct    = ff_j_rev_dct4;
+        c->idct_permutation_type= FF_NO_IDCT_PERM;
+    }else if(avctx->lowres==2){
+        c->idct_put= ff_jref_idct2_put;
+        c->idct_add= ff_jref_idct2_add;
+        c->idct    = ff_j_rev_dct2;
+        c->idct_permutation_type= FF_NO_IDCT_PERM;
+    }else if(avctx->lowres==3){
+        c->idct_put= ff_jref_idct1_put;
+        c->idct_add= ff_jref_idct1_add;
+        c->idct    = ff_j_rev_dct1;
+        c->idct_permutation_type= FF_NO_IDCT_PERM;
+    }else{
+        if (avctx->bits_per_raw_sample == 10) {
+            c->idct_put              = ff_simple_idct_put_10;
+            c->idct_add              = ff_simple_idct_add_10;
+            c->idct                  = ff_simple_idct_10;
+            c->idct_permutation_type = FF_NO_IDCT_PERM;
+        } else {
         if(avctx->idct_algo==FF_IDCT_INT){
             c->idct_put= ff_jref_idct_put;
             c->idct_add= ff_jref_idct_add;
@@ -2786,6 +2895,7 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx)
             c->idct     = ff_simple_idct_8;
             c->idct_permutation_type= FF_NO_IDCT_PERM;
         }
+        }
     }
 
     c->diff_pixels = diff_pixels_c;