ppc: d63 predictor 8x8

10x faster.

Change-Id: I7cedbf4df2ce7df5b6f1108b11815d088fdb9ba8

1 files changed, 23 insertions, 0 deletions

diff --git a/vpx_dsp/ppc/intrapred_vsx.c b/vpx_dsp/ppc/intrapred_vsx.c
index 8216e3e9e..0ff0f935d 100644
--- a/vpx_dsp/ppc/intrapred_vsx.c
+++ b/vpx_dsp/ppc/intrapred_vsx.c
@@ -642,6 +642,29 @@ void vpx_d45_predictor_32x32_vsx(uint8_t *dst, ptrdiff_t stride,
   }
 }
 
+void vpx_d63_predictor_8x8_vsx(uint8_t *dst, ptrdiff_t stride,
+                               const uint8_t *above, const uint8_t *left) {
+  const uint8x16_t af = vec_vsx_ld(0, above);
+  const uint8x16_t above_right = vec_splat(af, 9);
+  const uint8x16_t a = xxpermdi(af, above_right, 1);
+  const uint8x16_t b = vec_perm(a, above_right, sl1);
+  const uint8x16_t c = vec_perm(b, above_right, sl1);
+  uint8x16_t row0 = vec_avg(a, b);
+  uint8x16_t row1 = avg3(a, b, c);
+  int i;
+  (void)left;
+
+  for (i = 0; i < 4; i++) {
+    const uint8x16_t d0 = vec_vsx_ld(0, dst);
+    const uint8x16_t d1 = vec_vsx_ld(0, dst + stride);
+    vec_vsx_st(xxpermdi(row0, d0, 1), 0, dst);
+    vec_vsx_st(xxpermdi(row1, d1, 1), 0, dst + stride);
+    dst += stride * 2;
+    row0 = vec_perm(row0, above_right, sl1);
+    row1 = vec_perm(row1, above_right, sl1);
+  }
+}
+
 void vpx_d63_predictor_16x16_vsx(uint8_t *dst, ptrdiff_t stride,
                                  const uint8_t *above, const uint8_t *left) {
   const uint8x16_t a0 = vec_vsx_ld(0, above);