Merge "Wide loopfilter 16 pix at a time"

vp9/common/vp9_loopfilter.c

@@ -186,14 +186,22 @@ static void filter_selectively_horiz(uint8_t *s, int pitch,
                                      int only_4x4_1,
                                      const struct loop_filter_info *lfi) {
   unsigned int mask;
+  int count;
 
   for (mask = mask_16x16 | mask_8x8 | mask_4x4 | mask_4x4_int;
-       mask; mask >>= 1) {
+       mask; mask >>= count) {
+    count =1;
     if (mask & 1) {
       if (!only_4x4_1) {
         if (mask_16x16 & 1) {
-          vp9_mb_lpf_horizontal_edge_w(s, pitch, lfi->mblim, lfi->lim,
-                                       lfi->hev_thr);
+          if ((mask_16x16 & 3) == 3) {
+            vp9_mb_lpf_horizontal_edge_w(s, pitch, lfi->mblim, lfi->lim,
+                                         lfi->hev_thr, 2);
+            count = 2;
+          } else {
+            vp9_mb_lpf_horizontal_edge_w(s, pitch, lfi->mblim, lfi->lim,
+                                         lfi->hev_thr, 1);
+          }
           assert(!(mask_8x8 & 1));
           assert(!(mask_4x4 & 1));
           assert(!(mask_4x4_int & 1));
@@ -214,12 +222,12 @@ static void filter_selectively_horiz(uint8_t *s, int pitch,
         vp9_loop_filter_horizontal_edge(s + 4 * pitch, pitch, lfi->mblim,
                                         lfi->lim, lfi->hev_thr, 1);
     }
-    s += 8;
-    lfi++;
-    mask_16x16 >>= 1;
-    mask_8x8 >>= 1;
-    mask_4x4 >>= 1;
-    mask_4x4_int >>= 1;
+    s += 8 * count;
+    lfi += count;
+    mask_16x16 >>= count;
+    mask_8x8 >>= count;
+    mask_4x4 >>= count;
+    mask_4x4_int >>= count;
   }
 }
 

vp9/common/vp9_loopfilter_filters.c

@@ -258,12 +258,13 @@ static INLINE void wide_mbfilter(int8_t mask, uint8_t hev,
 void vp9_mb_lpf_horizontal_edge_w_c(uint8_t *s, int p,
                                     const uint8_t *blimit,
                                     const uint8_t *limit,
-                                    const uint8_t *thresh) {
+                                    const uint8_t *thresh,
+                                    int count) {
   int i;
 
   // loop filter designed to work using chars so that we can make maximum use
   // of 8 bit simd instructions.
-  for (i = 0; i < 8; ++i) {
+  for (i = 0; i < 8 * count; ++i) {
     const uint8_t p3 = s[-4 * p], p2 = s[-3 * p], p1 = s[-2 * p], p0 = s[-p];
     const uint8_t q0 = s[0 * p], q1 = s[1 * p], q2 = s[2 * p], q3 = s[3 * p];
     const int8_t mask = filter_mask(*limit, *blimit,

vp9/common/vp9_rtcd_defs.sh

@@ -222,7 +222,7 @@ specialize vp9_mbloop_filter_vertical_edge sse2 neon
 prototype void vp9_loop_filter_vertical_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
 specialize vp9_loop_filter_vertical_edge mmx neon
 
-prototype void vp9_mb_lpf_horizontal_edge_w "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh"
+prototype void vp9_mb_lpf_horizontal_edge_w "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"
 specialize vp9_mb_lpf_horizontal_edge_w sse2
 
 prototype void vp9_mbloop_filter_horizontal_edge "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count"