vc2enc: add support for Haar wavelet transforms

This commit adds support for the (simple, allowed in the spec, but
inferior in quality) Haar wavelet transforms.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

libavcodec/vc2enc.c

@@ -1153,8 +1153,10 @@ static const AVOption vc2enc_options[] = {
     {"slice_height",  "Slice height", offsetof(VC2EncContext, slice_height), AV_OPT_TYPE_INT, {.i64 = 64}, 8, 1024, VC2ENC_FLAGS, "slice_height"},
     {"wavelet_depth", "Transform depth", offsetof(VC2EncContext, wavelet_depth), AV_OPT_TYPE_INT, {.i64 = 5}, 1, 5, VC2ENC_FLAGS, "wavelet_depth"},
     {"wavelet_type",  "Transform type",  offsetof(VC2EncContext, wavelet_idx), AV_OPT_TYPE_INT, {.i64 = VC2_TRANSFORM_9_7}, 0, VC2_TRANSFORMS_NB, VC2ENC_FLAGS, "wavelet_idx"},
-        {"9_7",       "Deslauriers-Dubuc (9,7)", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_9_7}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
-        {"5_3",       "LeGall (5,3)",            0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_5_3}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
+        {"9_7",          "Deslauriers-Dubuc (9,7)", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_9_7},    INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
+        {"5_3",          "LeGall (5,3)",            0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_5_3},    INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
+        {"haar",         "Haar (with shift)",       0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_HAAR_S}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
+        {"haar_noshift", "Haar (without shift)",    0, AV_OPT_TYPE_CONST, {.i64 = VC2_TRANSFORM_HAAR},   INT_MIN, INT_MAX, VC2ENC_FLAGS, "wavelet_idx"},
     {"qm", "Custom quantization matrix", offsetof(VC2EncContext, quant_matrix), AV_OPT_TYPE_INT, {.i64 = VC2_QM_DEF}, 0, VC2_QM_NB, VC2ENC_FLAGS, "quant_matrix"},
         {"default",   "Default from the specifications", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_QM_DEF}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "quant_matrix"},
         {"color",     "Prevents low bitrate discoloration", 0, AV_OPT_TYPE_CONST, {.i64 = VC2_QM_COL}, INT_MIN, INT_MAX, VC2ENC_FLAGS, "quant_matrix"},

libavcodec/vc2enc_dwt.c

@@ -211,10 +211,56 @@ static void vc2_subband_dwt_53(VC2TransformContext *t, dwtcoef *data,
     deinterleave(data, stride, width, height, synth);
 }
 
+static av_always_inline void dwt_haar(VC2TransformContext *t, dwtcoef *data,
+                                      ptrdiff_t stride, int width, int height,
+                                      const int s)
+{
+    int x, y;
+    dwtcoef *synth = t->buffer, *synthl = synth, *datal = data;
+    const ptrdiff_t synth_width  = width  << 1;
+    const ptrdiff_t synth_height = height << 1;
+
+    /* Horizontal synthesis. */
+    for (y = 0; y < synth_height; y++) {
+        for (x = 0; x < synth_width; x += 2) {
+            synthl[y*synth_width + x + 1] = (datal[y*stride + x + 1] << s) -
+                                            (datal[y*stride + x] << s);
+            synthl[y*synth_width + x] = (datal[y*stride + x + 0] << s) +
+                                        ((synthl[y*synth_width + x + 1] + 1) >> 1);
+        }
+    }
+
+    /* Vertical synthesis. */
+    for (x = 0; x < synth_width; x++) {
+        for (y = 0; y < synth_height; y += 2) {
+            synthl[(y + 1)*synth_width + x] = synthl[(y + 1)*synth_width + x] -
+                                              synthl[y*synth_width + x];
+            synthl[y*synth_width + x] = synthl[y*synth_width + x] +
+                                        ((synthl[(y + 1)*synth_width + x] + 1) >> 1);
+        }
+    }
+
+    deinterleave(data, stride, width, height, synth);
+}
+
+static void vc2_subband_dwt_haar(VC2TransformContext *t, dwtcoef *data,
+                                 ptrdiff_t stride, int width, int height)
+{
+    dwt_haar(t, data, stride, width, height, 0);
+}
+
+static void vc2_subband_dwt_haar_shift(VC2TransformContext *t, dwtcoef *data,
+                                       ptrdiff_t stride, int width, int height)
+{
+    dwt_haar(t, data, stride, width, height, 1);
+}
+
 av_cold int ff_vc2enc_init_transforms(VC2TransformContext *s, int p_width, int p_height)
 {
     s->vc2_subband_dwt[VC2_TRANSFORM_9_7]    = vc2_subband_dwt_97;
     s->vc2_subband_dwt[VC2_TRANSFORM_5_3]    = vc2_subband_dwt_53;
+    s->vc2_subband_dwt[VC2_TRANSFORM_HAAR]   = vc2_subband_dwt_haar;
+    s->vc2_subband_dwt[VC2_TRANSFORM_HAAR_S] = vc2_subband_dwt_haar_shift;
 
     s->buffer = av_malloc(2*p_width*p_height*sizeof(dwtcoef));
     if (!s->buffer)