shithub: libvpx

Info  •  Files  •  Log  •  Branches

ppc: Add vpx_sad64/32/16x64/32/16_avg_vsx

Change-Id: Ic9639b1331d8c5cbc207c2a036891ff0137fc56f

--- a/test/sad_test.cc

+++ b/test/sad_test.cc

@@ -934,5 +934,17 @@

   SadMxNParam(16, 8, &vpx_sad16x8_vsx),

 };

 INSTANTIATE_TEST_CASE_P(VSX, SADTest, ::testing::ValuesIn(vsx_tests));

+

+const SadMxNAvgParam avg_vsx_tests[] = {

+  SadMxNAvgParam(64, 64, &vpx_sad64x64_avg_vsx),

+  SadMxNAvgParam(64, 32, &vpx_sad64x32_avg_vsx),

+  SadMxNAvgParam(32, 64, &vpx_sad32x64_avg_vsx),

+  SadMxNAvgParam(32, 32, &vpx_sad32x32_avg_vsx),

+  SadMxNAvgParam(32, 16, &vpx_sad32x16_avg_vsx),

+  SadMxNAvgParam(16, 32, &vpx_sad16x32_avg_vsx),

+  SadMxNAvgParam(16, 16, &vpx_sad16x16_avg_vsx),

+  SadMxNAvgParam(16, 8, &vpx_sad16x8_avg_vsx),

+};

+INSTANTIATE_TEST_CASE_P(VSX, SADavgTest, ::testing::ValuesIn(avg_vsx_tests));

 #endif  // HAVE_VSX

 }  // namespace

--- a/vpx_dsp/ppc/sad_vsx.c

+++ b/vpx_dsp/ppc/sad_vsx.c

@@ -10,9 +10,12 @@

 #include <stdlib.h>

+#include "./vpx_dsp_rtcd.h"

+

 #include "vpx_dsp/ppc/types_vsx.h"

 #include "vpx/vpx_integer.h"

+#include "vpx_ports/mem.h"

 #define PROCESS16(offset)           \

   v_a = vec_vsx_ld(offset, a);      \

@@ -100,3 +103,45 @@

 SAD32(64);

 SAD64(32);

 SAD64(64);

+

+#define SAD16AVG(height)                                                      \

+  unsigned int vpx_sad16x##height##_avg_vsx(                                  \

+      const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \

+      const uint8_t *second_pred) {                                           \

+    DECLARE_ALIGNED(16, uint8_t, comp_pred[16 * height]);                     \

+    vpx_comp_avg_pred_vsx(comp_pred, second_pred, 16, height, ref,            \

+                          ref_stride);                                        \

+                                                                              \

+    return vpx_sad16x##height##_vsx(src, src_stride, comp_pred, 16);          \

+  }

+

+#define SAD32AVG(height)                                                      \

+  unsigned int vpx_sad32x##height##_avg_vsx(                                  \

+      const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \

+      const uint8_t *second_pred) {                                           \

+    DECLARE_ALIGNED(32, uint8_t, comp_pred[32 * height]);                     \

+    vpx_comp_avg_pred_vsx(comp_pred, second_pred, 32, height, ref,            \

+                          ref_stride);                                        \

+                                                                              \

+    return vpx_sad32x##height##_vsx(src, src_stride, comp_pred, 32);          \

+  }

+

+#define SAD64AVG(height)                                                      \

+  unsigned int vpx_sad64x##height##_avg_vsx(                                  \

+      const uint8_t *src, int src_stride, const uint8_t *ref, int ref_stride, \

+      const uint8_t *second_pred) {                                           \

+    DECLARE_ALIGNED(64, uint8_t, comp_pred[64 * height]);                     \

+    vpx_comp_avg_pred_vsx(comp_pred, second_pred, 64, height, ref,            \

+                          ref_stride);                                        \

+                                                                              \

+    return vpx_sad64x##height##_vsx(src, src_stride, comp_pred, 64);          \

+  }

+

+SAD16AVG(8);

+SAD16AVG(16);

+SAD16AVG(32);

+SAD32AVG(16);

+SAD32AVG(32);

+SAD32AVG(64);

+SAD64AVG(32);

+SAD64AVG(64);

--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl

+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl

@@ -778,28 +778,28 @@

 }  # CONFIG_VP9_ENCODER

 add_proto qw/unsigned int vpx_sad64x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

-specialize qw/vpx_sad64x64_avg avx2 msa sse2/;

+specialize qw/vpx_sad64x64_avg avx2 msa sse2 vsx/;

 add_proto qw/unsigned int vpx_sad64x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

-specialize qw/vpx_sad64x32_avg avx2 msa sse2/;

+specialize qw/vpx_sad64x32_avg avx2 msa sse2 vsx/;

 add_proto qw/unsigned int vpx_sad32x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

-specialize qw/vpx_sad32x64_avg avx2 msa sse2/;

+specialize qw/vpx_sad32x64_avg avx2 msa sse2 vsx/;

 add_proto qw/unsigned int vpx_sad32x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

-specialize qw/vpx_sad32x32_avg avx2 msa sse2/;

+specialize qw/vpx_sad32x32_avg avx2 msa sse2 vsx/;

 add_proto qw/unsigned int vpx_sad32x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

-specialize qw/vpx_sad32x16_avg avx2 msa sse2/;

+specialize qw/vpx_sad32x16_avg avx2 msa sse2 vsx/;

 add_proto qw/unsigned int vpx_sad16x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

-specialize qw/vpx_sad16x32_avg msa sse2/;

+specialize qw/vpx_sad16x32_avg msa sse2 vsx/;

 add_proto qw/unsigned int vpx_sad16x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

-specialize qw/vpx_sad16x16_avg msa sse2/;

+specialize qw/vpx_sad16x16_avg msa sse2 vsx/;

 add_proto qw/unsigned int vpx_sad16x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

-specialize qw/vpx_sad16x8_avg msa sse2/;

+specialize qw/vpx_sad16x8_avg msa sse2 vsx/;

 add_proto qw/unsigned int vpx_sad8x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";

 specialize qw/vpx_sad8x16_avg msa sse2/;