ppc: h predictor 4x4

2x faster.

Change-Id: I0583dec353299c6797401b646099f18db4e0420d

diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc
index 4e9b1fa76b6521654661c8319632ffea81ca645b..2881cf8dc6235e4bdf3c847745d0d44249a7c5c6 100644 (file)
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -313,6 +313,10 @@ INTRA_PRED_TEST(MSA, TestIntraPred32, vpx_dc_predictor_32x32_msa,
 #endif  // HAVE_MSA
 
 #if HAVE_VSX
+INTRA_PRED_TEST(VSX, TestIntraPred4, NULL, NULL, NULL, NULL, NULL,
+                vpx_h_predictor_4x4_vsx, NULL, NULL, NULL, NULL, NULL, NULL,
+                NULL)
+
 INTRA_PRED_TEST(VSX, TestIntraPred8, vpx_dc_predictor_8x8_vsx, NULL, NULL, NULL,
                 NULL, NULL, vpx_d45_predictor_8x8_vsx, NULL, NULL, NULL, NULL,
                 NULL, vpx_tm_predictor_8x8_vsx)

diff --git a/vpx_dsp/ppc/intrapred_vsx.c b/vpx_dsp/ppc/intrapred_vsx.c
index b3ff5a70e896d4501f17c69da49a0a6bb50196a7..947c787608c58f5a418ab4aa01029c40e3796b84 100644 (file)
--- a/vpx_dsp/ppc/intrapred_vsx.c
+++ b/vpx_dsp/ppc/intrapred_vsx.c
@@ -35,6 +35,27 @@ void vpx_v_predictor_32x32_vsx(uint8_t *dst, ptrdiff_t stride,
   }
 }
 
+static const uint32x4_t mask4 = { 0, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF };
+
+void vpx_h_predictor_4x4_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);
+
+  (void)above;
+
+  vec_vsx_st(vec_sel(v0, vec_vsx_ld(0, dst), (uint8x16_t)mask4), 0, dst);
+  dst += stride;
+  vec_vsx_st(vec_sel(v1, vec_vsx_ld(0, dst), (uint8x16_t)mask4), 0, dst);
+  dst += stride;
+  vec_vsx_st(vec_sel(v2, vec_vsx_ld(0, dst), (uint8x16_t)mask4), 0, dst);
+  dst += stride;
+  vec_vsx_st(vec_sel(v3, vec_vsx_ld(0, dst), (uint8x16_t)mask4), 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 e96ff2a27c6427575182218e92879b213665399f..cbd1097160a32de8034ef36cdf9410a9790c2445 100644 (file)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -39,7 +39,7 @@ specialize qw/vpx_d63_predictor_4x4 ssse3/;
 add_proto qw/void vpx_d63e_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
 
 add_proto qw/void vpx_h_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_4x4 neon dspr2 msa sse2/;
+specialize qw/vpx_h_predictor_4x4 neon dspr2 msa sse2 vsx/;
 
 add_proto qw/void vpx_he_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";