From: Luca Barbato <lu_zero@gentoo.org>
Date: Sat, 8 Apr 2017 00:26:54 +0000 (+0000)
Subject: ppc: dc_128 predictor 16x16
X-Git-Tag: v1.7.0~547
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=f9d20e6df21296899143d832f1a8d7ed7985e395;p=libvpx

ppc: dc_128 predictor 16x16

20x faster.

Change-Id: I05f0deb2d38ae7966eae6b71fbc0aa51880e5709
---

diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc
index 5411e2390..61b3dbeeb 100644
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -316,9 +316,10 @@ INTRA_PRED_TEST(MSA, TestIntraPred32, vpx_dc_predictor_32x32_msa,
 INTRA_PRED_TEST(VSX, TestIntraPred8, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
                 NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_8x8_vsx)
 
-INTRA_PRED_TEST(VSX, TestIntraPred16, NULL, NULL, NULL, NULL,
-                vpx_v_predictor_16x16_vsx, vpx_h_predictor_16x16_vsx, NULL,
-                NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_16x16_vsx)
+INTRA_PRED_TEST(VSX, TestIntraPred16, NULL, NULL, NULL,
+                vpx_dc_128_predictor_16x16_vsx, vpx_v_predictor_16x16_vsx,
+                vpx_h_predictor_16x16_vsx, NULL, NULL, NULL, NULL, NULL, NULL,
+                vpx_tm_predictor_16x16_vsx)
 
 INTRA_PRED_TEST(VSX, TestIntraPred32, NULL, NULL, NULL, NULL,
                 vpx_v_predictor_32x32_vsx, vpx_h_predictor_32x32_vsx, NULL,
diff --git a/vpx_dsp/ppc/intrapred_vsx.c b/vpx_dsp/ppc/intrapred_vsx.c
index 96491924e..155cc722f 100644
--- a/vpx_dsp/ppc/intrapred_vsx.c
+++ b/vpx_dsp/ppc/intrapred_vsx.c
@@ -361,3 +361,21 @@ void vpx_tm_predictor_32x32_vsx(uint8_t *dst, ptrdiff_t stride,
 
   tm_predictor_32x8(dst, stride, unpack_to_s16_l(l1), a0, a1, tl);
 }
+
+static INLINE void dc_fill_predictor_16x16(uint8_t *dst, const ptrdiff_t stride,
+                                           const uint8x16_t val) {
+  int i;
+
+  for (i = 0; i < 16; i++, dst += stride) {
+    vec_vsx_st(val, 0, dst);
+  }
+}
+
+void vpx_dc_128_predictor_16x16_vsx(uint8_t *dst, ptrdiff_t stride,
+                                    const uint8_t *above, const uint8_t *left) {
+  const uint8x16_t v128 = vec_sl(vec_splat_u8(1), vec_splat_u8(7));
+  (void)above;
+  (void)left;
+
+  dc_fill_predictor_16x16(dst, stride, v128);
+}
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 680e278e5..3450c15d7 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -145,7 +145,7 @@ add_proto qw/void vpx_dc_left_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_strid
 specialize qw/vpx_dc_left_predictor_16x16 neon msa sse2/;
 
 add_proto qw/void vpx_dc_128_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_dc_128_predictor_16x16 neon msa sse2/;
+specialize qw/vpx_dc_128_predictor_16x16 neon msa sse2 vsx/;
 
 add_proto qw/void vpx_d207_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
 specialize qw/vpx_d207_predictor_32x32 ssse3/;