ppc: h predictor 8x8

Slightly faster with the current compiler.

Change-Id: Iae225fac08395eb430c97a2abec69c60f5cf5c47

diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc
index 539f7d4fbf968cac5fd57902b40506c4c43a9d77..23fce335af9b6a10017d06f7af8484023a29717c 100644 (file)
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -318,8 +318,9 @@ INTRA_PRED_TEST(VSX, TestIntraPred4, NULL, NULL, NULL, NULL, NULL,
                 vpx_tm_predictor_4x4_vsx)
 
 INTRA_PRED_TEST(VSX, TestIntraPred8, vpx_dc_predictor_8x8_vsx, NULL, NULL, NULL,
-                NULL, NULL, vpx_d45_predictor_8x8_vsx, NULL, NULL, NULL, NULL,
-                vpx_d63_predictor_8x8_vsx, vpx_tm_predictor_8x8_vsx)
+                NULL, vpx_h_predictor_8x8_vsx, vpx_d45_predictor_8x8_vsx, NULL,
+                NULL, NULL, NULL, vpx_d63_predictor_8x8_vsx,
+                vpx_tm_predictor_8x8_vsx)
 
 INTRA_PRED_TEST(VSX, TestIntraPred16, vpx_dc_predictor_16x16_vsx,
                 vpx_dc_left_predictor_16x16_vsx, vpx_dc_top_predictor_16x16_vsx,

diff --git a/vpx_dsp/ppc/intrapred_vsx.c b/vpx_dsp/ppc/intrapred_vsx.c
index 0ff0f935d17cf5e1be6f43151c99c7e1e7044450..6273460f1901b582a4fdc12e468a117ea067f0bb 100644 (file)
--- a/vpx_dsp/ppc/intrapred_vsx.c
+++ b/vpx_dsp/ppc/intrapred_vsx.c
@@ -56,6 +56,38 @@ void vpx_h_predictor_4x4_vsx(uint8_t *dst, ptrdiff_t stride,
   vec_vsx_st(vec_sel(v3, vec_vsx_ld(0, dst), (uint8x16_t)mask4), 0, dst);
 }
 
+void vpx_h_predictor_8x8_vsx(uint8_t *dst, ptrdiff_t stride,
+                             const uint8_t *above, const uint8_t *left) {
+  const uint8x16_t d = vec_vsx_ld(0, left);
+  const uint8x16_t v0 = vec_splat(d, 0);
+  const uint8x16_t v1 = vec_splat(d, 1);
+  const uint8x16_t v2 = vec_splat(d, 2);
+  const uint8x16_t v3 = vec_splat(d, 3);
+
+  const uint8x16_t v4 = vec_splat(d, 4);
+  const uint8x16_t v5 = vec_splat(d, 5);
+  const uint8x16_t v6 = vec_splat(d, 6);
+  const uint8x16_t v7 = vec_splat(d, 7);
+
+  (void)above;
+
+  vec_vsx_st(xxpermdi(v0, vec_vsx_ld(0, dst), 1), 0, dst);
+  dst += stride;
+  vec_vsx_st(xxpermdi(v1, vec_vsx_ld(0, dst), 1), 0, dst);
+  dst += stride;
+  vec_vsx_st(xxpermdi(v2, vec_vsx_ld(0, dst), 1), 0, dst);
+  dst += stride;
+  vec_vsx_st(xxpermdi(v3, vec_vsx_ld(0, dst), 1), 0, dst);
+  dst += stride;
+  vec_vsx_st(xxpermdi(v4, vec_vsx_ld(0, dst), 1), 0, dst);
+  dst += stride;
+  vec_vsx_st(xxpermdi(v5, vec_vsx_ld(0, dst), 1), 0, dst);
+  dst += stride;
+  vec_vsx_st(xxpermdi(v6, vec_vsx_ld(0, dst), 1), 0, dst);
+  dst += stride;
+  vec_vsx_st(xxpermdi(v7, vec_vsx_ld(0, dst), 1), 0, dst);
+}
+
 void vpx_h_predictor_16x16_vsx(uint8_t *dst, ptrdiff_t stride,
                                const uint8_t *above, const uint8_t *left) {
   const uint8x16_t d = vec_vsx_ld(0, left);

diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index ef4ce24e98c65c6ef9304aa2254a9503c40ce7eb..ed0339cbe1352c4b78774620fddb8e2238b43d71 100644 (file)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -81,7 +81,7 @@ add_proto qw/void vpx_d63_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, con
 specialize qw/vpx_d63_predictor_8x8 ssse3 vsx/;
 
 add_proto qw/void vpx_h_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_8x8 neon dspr2 msa sse2/;
+specialize qw/vpx_h_predictor_8x8 neon dspr2 msa sse2 vsx/;
 
 add_proto qw/void vpx_d117_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";