vp10: reimplement d45/4x4 to match vp8 instead of vp9.

This is more a proof of concept than anything else. The problem here
isn't so much how to code it, but rather where to place the resulting
code. All intrapred DSP code lives in vpx_dsp, so do we want the vp10
specific intra pred functions to live there, or in vp10/?

See issue 1015.

Change-Id: I675f7badcc8e18fd99a9553910ecf3ddf81f0a05
diff --git a/vp10/common/reconintra.c b/vp10/common/reconintra.c
index 9350d67..52df2fe 100644
--- a/vp10/common/reconintra.c
+++ b/vp10/common/reconintra.c
@@ -55,16 +55,24 @@
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
 static void vp10_init_intra_predictors_internal(void) {
-#define INIT_ALL_SIZES(p, type) \
-  p[TX_4X4] = vpx_##type##_predictor_4x4; \
+#define INIT_NO_4X4(p, type) \
   p[TX_8X8] = vpx_##type##_predictor_8x8; \
   p[TX_16X16] = vpx_##type##_predictor_16x16; \
   p[TX_32X32] = vpx_##type##_predictor_32x32
 
+#define INIT_ALL_SIZES(p, type) \
+  p[TX_4X4] = vpx_##type##_predictor_4x4; \
+  INIT_NO_4X4(p, type)
+
   INIT_ALL_SIZES(pred[V_PRED], v);
   INIT_ALL_SIZES(pred[H_PRED], h);
   INIT_ALL_SIZES(pred[D207_PRED], d207);
+#if CONFIG_MISC_FIXES
+  pred[D45_PRED][TX_4X4] = vpx_d45e_predictor_4x4;
+  INIT_NO_4X4(pred[D45_PRED], d45);
+#else
   INIT_ALL_SIZES(pred[D45_PRED], d45);
+#endif
   INIT_ALL_SIZES(pred[D63_PRED], d63);
   INIT_ALL_SIZES(pred[D117_PRED], d117);
   INIT_ALL_SIZES(pred[D135_PRED], d135);
@@ -80,7 +88,12 @@
   INIT_ALL_SIZES(pred_high[V_PRED], highbd_v);
   INIT_ALL_SIZES(pred_high[H_PRED], highbd_h);
   INIT_ALL_SIZES(pred_high[D207_PRED], highbd_d207);
+#if CONFIG_MISC_FIXES
+  pred_high[D45_PRED][TX_4X4] = vpx_highbd_d45e_predictor_4x4;
+  INIT_NO_4X4(pred_high[D45_PRED], highbd_d45);
+#else
   INIT_ALL_SIZES(pred_high[D45_PRED], highbd_d45);
+#endif
   INIT_ALL_SIZES(pred_high[D63_PRED], highbd_d63);
   INIT_ALL_SIZES(pred_high[D117_PRED], highbd_d117);
   INIT_ALL_SIZES(pred_high[D135_PRED], highbd_d135);
diff --git a/vpx_dsp/intrapred.c b/vpx_dsp/intrapred.c
index aafd8bf..a5e3d30 100644
--- a/vpx_dsp/intrapred.c
+++ b/vpx_dsp/intrapred.c
@@ -518,6 +518,21 @@
   }
 }
 
+static INLINE void highbd_d45e_predictor(uint16_t *dst, ptrdiff_t stride,
+                                         int bs, const uint16_t *above,
+                                         const uint16_t *left, int bd) {
+  int r, c;
+  (void) left;
+  (void) bd;
+  for (r = 0; r < bs; ++r) {
+    for (c = 0; c < bs; ++c) {
+      dst[c] = AVG3(above[r + c], above[r + c + 1],
+                    above[r + c + 1 + (r + c + 2 < 8)]);
+    }
+    dst += stride;
+  }
+}
+
 static INLINE void highbd_d117_predictor(uint16_t *dst, ptrdiff_t stride,
                                          int bs, const uint16_t *above,
                                          const uint16_t *left, int bd) {
@@ -766,4 +781,7 @@
 intra_pred_allsizes(dc_left)
 intra_pred_allsizes(dc_top)
 intra_pred_allsizes(dc)
+#if CONFIG_VP9_HIGHBITDEPTH && CONFIG_MISC_FIXES
+intra_pred_highbd_sized(d45e, 4)
+#endif
 #undef intra_pred_allsizes
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 352e1ff..7782168 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -230,6 +230,9 @@
   add_proto qw/void vpx_highbd_d45_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
   specialize qw/vpx_highbd_d45_predictor_4x4/;
 
+  add_proto qw/void vpx_highbd_d45e_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+  specialize qw/vpx_highbd_d45e_predictor_4x4/;
+
   add_proto qw/void vpx_highbd_d63_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
   specialize qw/vpx_highbd_d63_predictor_4x4/;