Merge "mips msa vp9 convolve8 avg vert optimization"

test/convolve_test.cc

@@ -1819,7 +1819,7 @@ INSTANTIATE_TEST_CASE_P(DSPR2, ConvolveTest, ::testing::Values(
 const ConvolveFunctions convolve8_msa(
     vp9_convolve_copy_msa, vp9_convolve_avg_msa,
     vp9_convolve8_horiz_msa, vp9_convolve8_avg_horiz_c,
-    vp9_convolve8_vert_msa, vp9_convolve8_avg_vert_c,
+    vp9_convolve8_vert_msa, vp9_convolve8_avg_vert_msa,
     vp9_convolve8_msa, vp9_convolve8_avg_c, 0);
 
 INSTANTIATE_TEST_CASE_P(MSA, ConvolveTest, ::testing::Values(

vp9/common/mips/msa/vp9_macros_msa.h

@@ -870,6 +870,13 @@
 #define ILVR_W2_UB(...) ILVR_W2(v16u8, __VA_ARGS__)
 #define ILVR_W2_SH(...) ILVR_W2(v8i16, __VA_ARGS__)
 
+#define ILVR_W4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7,  \
+                out0, out1, out2, out3) {                       \
+  ILVR_W2(RTYPE, in0, in1, in2, in3, out0, out1);               \
+  ILVR_W2(RTYPE, in4, in5, in6, in7, out2, out3);               \
+}
+#define ILVR_W4_UB(...) ILVR_W4(v16u8, __VA_ARGS__)
+
 /* Description : Interleave right half of double word elements from vectors
    Arguments   : Inputs  - in0, in1, in2, in3, in4, in5, in6, in7
                  Outputs - out0, out1, out2, out3
@@ -1446,6 +1453,23 @@
   out_m;                                                 \
 })
 
+/* Description : Converts inputs to unsigned bytes, interleave, average & store
+                 as 8x4 unsigned byte block
+   Arguments   : Inputs  - in0, in1, in2, in3, dst0, dst1, dst2, dst3,
+                           pdst, stride
+*/
+#define CONVERT_UB_AVG_ST8x4_UB(in0, in1, in2, in3,                      \
+                                dst0, dst1, dst2, dst3, pdst, stride) {  \
+  v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m;                                  \
+  uint8_t *pdst_m = (uint8_t *)(pdst);                                   \
+                                                                         \
+  tmp0_m = PCKEV_XORI128_UB(in0, in1);                                   \
+  tmp1_m = PCKEV_XORI128_UB(in2, in3);                                   \
+  ILVR_D2_UB(dst1, dst0, dst3, dst2, tmp2_m, tmp3_m);                    \
+  AVER_UB2_UB(tmp0_m, tmp2_m, tmp1_m, tmp3_m, tmp0_m, tmp1_m);           \
+  ST8x4_UB(tmp0_m, tmp1_m, pdst_m, stride);                              \
+}
+
 /* Description : Pack even byte elements and store byte vector in destination
                  memory
    Arguments   : Inputs  - in0, in1, pdst

vp9/common/vp9_rtcd_defs.pl

@@ -307,7 +307,7 @@ add_proto qw/void vp9_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_s
 specialize qw/vp9_convolve8_avg_horiz sse2 ssse3 neon dspr2/;
 
 add_proto qw/void vp9_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/vp9_convolve8_avg_vert sse2 ssse3 neon dspr2/;
+specialize qw/vp9_convolve8_avg_vert sse2 ssse3 neon dspr2 msa/;
 
 #
 # dct

vp9/vp9_common.mk

@@ -132,6 +132,7 @@ VP9_COMMON_SRCS-$(HAVE_DSPR2)  += common/mips/dspr2/vp9_mblpf_vert_loopfilter_ds
 
 # common (msa)
 VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_macros_msa.h
+VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_avg_vert_msa.c
 VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_horiz_msa.c
 VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_msa.c
 VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_vert_msa.c