ppc Add vpx_sad32x16/32/64_vsx

Change-Id: I3c4f9d595275669580413a71b3c3c810e7ddcacd

diff --git a/test/sad_test.cc b/test/sad_test.cc
index 23991f56f7ed5fdd71bb14559d53059e92dbc483..e253f20d50c626b57369bb6be8e96308ad92c76a 100644 (file)
--- a/test/sad_test.cc
+++ b/test/sad_test.cc
@@ -924,8 +924,12 @@ INSTANTIATE_TEST_CASE_P(MSA, SADx4Test, ::testing::ValuesIn(x4d_msa_tests));
 // VSX functions
 #if HAVE_VSX
 const SadMxNParam vsx_tests[] = {
+  SadMxNParam(32, 64, &vpx_sad32x64_vsx),
+  SadMxNParam(32, 32, &vpx_sad32x32_vsx),
+  SadMxNParam(32, 16, &vpx_sad32x16_vsx),
   SadMxNParam(16, 32, &vpx_sad16x32_vsx),
-  SadMxNParam(16, 16, &vpx_sad16x16_vsx), SadMxNParam(16, 8, &vpx_sad16x8_vsx),
+  SadMxNParam(16, 16, &vpx_sad16x16_vsx),
+  SadMxNParam(16, 8, &vpx_sad16x8_vsx),
 };
 INSTANTIATE_TEST_CASE_P(VSX, SADTest, ::testing::ValuesIn(vsx_tests));
 #endif  // HAVE_VSX

diff --git a/vpx_dsp/ppc/sad_vsx.c b/vpx_dsp/ppc/sad_vsx.c
index fb6c7d5d554e606a4800336f64623f27992acdaa..c7cbd55d9cf9a0a8f98991c8f092b57cc3c492ca 100644 (file)
--- a/vpx_dsp/ppc/sad_vsx.c
+++ b/vpx_dsp/ppc/sad_vsx.c
@@ -48,6 +48,30 @@
     return sad[3] + sad[2] + sad[1] + sad[0];                             \
   }
 
+#define SAD32(height)                                                     \
+  unsigned int vpx_sad32x##height##_vsx(const uint8_t *a, int a_stride,   \
+                                        const uint8_t *b, int b_stride) { \
+    int y;                                                                \
+    unsigned int sad[4];                                                  \
+    uint8x16_t v_a, v_b;                                                  \
+    int16x8_t v_ah, v_al, v_bh, v_bl, v_absh, v_absl, v_subh, v_subl;     \
+    int32x4_t v_sad = vec_splat_s32(0);                                   \
+                                                                          \
+    for (y = 0; y < height; y++) {                                        \
+      PROCESS16(0);                                                       \
+      PROCESS16(16);                                                      \
+                                                                          \
+      a += a_stride;                                                      \
+      b += b_stride;                                                      \
+    }                                                                     \
+    vec_vsx_st((uint32x4_t)v_sad, 0, sad);                                \
+                                                                          \
+    return sad[3] + sad[2] + sad[1] + sad[0];                             \
+  }
+
 SAD16(8);
 SAD16(16);
 SAD16(32);
+SAD32(16);
+SAD32(32);
+SAD32(64);

diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index 6e34364ef9aa6e585bcb32e03b5a45ae92ad7fe3..178a1182bb53e8f54400ae7c23bbb4d3e7b633ac 100644 (file)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -702,13 +702,13 @@ add_proto qw/unsigned int vpx_sad64x32/, "const uint8_t *src_ptr, int src_stride
 specialize qw/vpx_sad64x32 avx2 msa sse2/;
 
 add_proto qw/unsigned int vpx_sad32x64/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-specialize qw/vpx_sad32x64 avx2 msa sse2/;
+specialize qw/vpx_sad32x64 avx2 msa sse2 vsx/;
 
 add_proto qw/unsigned int vpx_sad32x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-specialize qw/vpx_sad32x32 avx2 neon msa sse2/;
+specialize qw/vpx_sad32x32 avx2 neon msa sse2 vsx/;
 
 add_proto qw/unsigned int vpx_sad32x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-specialize qw/vpx_sad32x16 avx2 msa sse2/;
+specialize qw/vpx_sad32x16 avx2 msa sse2 vsx/;
 
 add_proto qw/unsigned int vpx_sad16x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
 specialize qw/vpx_sad16x32 msa sse2 vsx/;