From e6ca81ee6706e3c151cdc9b5b33ea695f35e062d Mon Sep 17 00:00:00 2001
From: Luca Barbato <lu_zero@gentoo.org>
Date: Mon, 1 May 2017 15:53:44 +0000
Subject: [PATCH] ppc: Add convolve8_avg_vert_vsx

Only the generic one again, speedups for 8x8 and larger blocks to
come later.

Change-Id: I90d481d3a602d1e277ead8f3934eca126b86b72d
---
 test/convolve_test.cc          |  2 +-
 vpx_dsp/ppc/vpx_convolve_vsx.c | 36 ++++++++++++++++++++++++++++++++++
 vpx_dsp/vpx_dsp_rtcd_defs.pl   |  2 +-
 3 files changed, 38 insertions(+), 2 deletions(-)

diff --git a/test/convolve_test.cc b/test/convolve_test.cc
index 085d2a27c..a5b4b3c23 100644
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -1226,7 +1226,7 @@ INSTANTIATE_TEST_CASE_P(MSA, ConvolveTest,
 const ConvolveFunctions convolve8_vsx(
     vpx_convolve_copy_vsx, vpx_convolve_avg_vsx, vpx_convolve8_horiz_vsx,
     vpx_convolve8_avg_horiz_vsx, vpx_convolve8_vert_vsx,
-    vpx_convolve8_avg_vert_c, vpx_convolve8_c, vpx_convolve8_avg_c,
+    vpx_convolve8_avg_vert_vsx, 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, vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
 const ConvolveParam kArrayConvolve_vsx[] = { ALL_SIZES(convolve8_vsx) };
diff --git a/vpx_dsp/ppc/vpx_convolve_vsx.c b/vpx_dsp/ppc/vpx_convolve_vsx.c
index 5a789c768..f0de3eb3b 100644
--- a/vpx_dsp/ppc/vpx_convolve_vsx.c
+++ b/vpx_dsp/ppc/vpx_convolve_vsx.c
@@ -275,6 +275,27 @@ static inline void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static inline void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
+                                     uint8_t *dst, ptrdiff_t dst_stride,
+                                     const InterpKernel *y_filters, int y0_q4,
+                                     int y_step_q4, int w, int h) {
+  int x, y;
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+
+  for (x = 0; x < w; ++x) {
+    int y_q4 = y0_q4;
+    for (y = 0; y < h; ++y) {
+      uint8_t v;
+      convolve_line_v(&v, &src[(y_q4 >> SUBPEL_BITS) * src_stride], src_stride,
+                      y_filters[y_q4 & SUBPEL_MASK]);
+      dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] + v, 1);
+      y_q4 += y_step_q4;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
 void vpx_convolve8_horiz_vsx(const uint8_t *src, ptrdiff_t src_stride,
                              uint8_t *dst, ptrdiff_t dst_stride,
                              const int16_t *filter_x, int x_step_q4,
@@ -319,3 +340,18 @@ void vpx_convolve8_vert_vsx(const uint8_t *src, ptrdiff_t src_stride,
   convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4, y_step_q4,
                 w, h);
 }
+
+void vpx_convolve8_avg_vert_vsx(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) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+  const int y0_q4 = get_filter_offset(filter_y, filters_y);
+
+  (void)filter_x;
+  (void)x_step_q4;
+
+  convolve_avg_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
+                    y_step_q4, w, h);
+}
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 7282671c1..3488e5eb3 100644
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -353,7 +353,7 @@ add_proto qw/void vpx_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_s
 specialize qw/vpx_convolve8_avg_horiz sse2 ssse3 neon dspr2 msa vsx/;
 
 add_proto qw/void vpx_convolve8_avg_vert/, "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_avg_vert sse2 ssse3 neon dspr2 msa/;
+specialize qw/vpx_convolve8_avg_vert sse2 ssse3 neon dspr2 msa vsx/;
 
 add_proto qw/void vpx_scaled_2d/, "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_scaled_2d ssse3/;