ppc: Add convolve_avg

Change-Id: Ib203c444c708f42072e38301ee3db97b5b53d014

diff --git a/test/convolve_test.cc b/test/convolve_test.cc
index 950a28d2b7ed65dcb0cc1456b66e2991e700215a..bb009740aa3ab3401a615578f68460fb0006ac7a 100644 (file)
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -1224,7 +1224,7 @@ INSTANTIATE_TEST_CASE_P(MSA, ConvolveTest,
 
 #if HAVE_VSX
 const ConvolveFunctions convolve8_vsx(
-    vpx_convolve_copy_vsx, vpx_convolve_avg_c, vpx_convolve8_horiz_c,
+    vpx_convolve_copy_vsx, vpx_convolve_avg_vsx, vpx_convolve8_horiz_c,
     vpx_convolve8_avg_horiz_c, vpx_convolve8_vert_c, vpx_convolve8_avg_vert_c,
     vpx_convolve8_c, vpx_convolve8_avg_c, vpx_scaled_horiz_c,
     vpx_scaled_avg_horiz_c, vpx_scaled_vert_c, vpx_scaled_avg_vert_c,

diff --git a/vpx_dsp/ppc/vpx_convolve_vsx.c b/vpx_dsp/ppc/vpx_convolve_vsx.c
index 32cdbb4d0758e5e18ad899538810c445c9c7b224..670de8118db60a4fcd42efc1d704c6bdd470eb7c 100644 (file)
--- a/vpx_dsp/ppc/vpx_convolve_vsx.c
+++ b/vpx_dsp/ppc/vpx_convolve_vsx.c
@@ -83,3 +83,78 @@ void vpx_convolve_copy_vsx(const uint8_t *src, ptrdiff_t src_stride,
     }
   }
 }
+
+static inline void avg_w16(const uint8_t *src, ptrdiff_t src_stride,
+                           uint8_t *dst, ptrdiff_t dst_stride, int32_t h) {
+  int i;
+
+  for (i = h; i--;) {
+    const uint8x16_t v = vec_avg(vec_vsx_ld(0, src), vec_vsx_ld(0, dst));
+    vec_vsx_st(v, 0, dst);
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+static inline void avg_w32(const uint8_t *src, ptrdiff_t src_stride,
+                           uint8_t *dst, ptrdiff_t dst_stride, int32_t h) {
+  int i;
+
+  for (i = h; i--;) {
+    const uint8x16_t v0 = vec_avg(vec_vsx_ld(0, src), vec_vsx_ld(0, dst));
+    const uint8x16_t v1 = vec_avg(vec_vsx_ld(16, src), vec_vsx_ld(16, dst));
+    vec_vsx_st(v0, 0, dst);
+    vec_vsx_st(v1, 16, dst);
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+static inline void avg_w64(const uint8_t *src, ptrdiff_t src_stride,
+                           uint8_t *dst, ptrdiff_t dst_stride, int32_t h) {
+  int i;
+
+  for (i = h; i--;) {
+    const uint8x16_t v0 = vec_avg(vec_vsx_ld(0, src), vec_vsx_ld(0, dst));
+    const uint8x16_t v1 = vec_avg(vec_vsx_ld(16, src), vec_vsx_ld(16, dst));
+    const uint8x16_t v2 = vec_avg(vec_vsx_ld(32, src), vec_vsx_ld(32, dst));
+    const uint8x16_t v3 = vec_avg(vec_vsx_ld(48, src), vec_vsx_ld(48, dst));
+    vec_vsx_st(v0, 0, dst);
+    vec_vsx_st(v1, 16, dst);
+    vec_vsx_st(v2, 32, dst);
+    vec_vsx_st(v3, 48, dst);
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
+void vpx_convolve_avg_vsx(const uint8_t *src, ptrdiff_t src_stride,
+                          uint8_t *dst, ptrdiff_t dst_stride,
+                          const int16_t *filter_x, int32_t filter_x_stride,
+                          const int16_t *filter_y, int32_t filter_y_stride,
+                          int32_t w, int32_t h) {
+  (void)filter_x;
+  (void)filter_y;
+  (void)filter_x_stride;
+  (void)filter_y_stride;
+
+  switch (w) {
+    case 16: {
+      avg_w16(src, src_stride, dst, dst_stride, h);
+      break;
+    }
+    case 32: {
+      avg_w32(src, src_stride, dst, dst_stride, h);
+      break;
+    }
+    case 64: {
+      avg_w64(src, src_stride, dst, dst_stride, h);
+      break;
+    }
+    default: {
+      vpx_convolve_avg_c(src, src_stride, dst, dst_stride, filter_x,
+                         filter_x_stride, filter_y, filter_y_stride, w, h);
+      break;
+    }
+  }
+}

diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 0a9327ba63655726befede537ef2c12694437008..a93e95ebea9964abd56cb8263be8bb44cac6e140 100644 (file)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -335,7 +335,7 @@ add_proto qw/void vpx_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride,
 specialize qw/vpx_convolve_copy neon dspr2 msa sse2 vsx/;
 
 add_proto qw/void vpx_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vpx_convolve_avg neon dspr2 msa sse2/;
+specialize qw/vpx_convolve_avg neon dspr2 msa sse2 vsx/;
 
 add_proto qw/void vpx_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
 specialize qw/vpx_convolve8 sse2 ssse3 avx2 neon dspr2 msa/;