Code clean of sad4xN(_avg)_sse

Replace MMX with SSE2, reduce psadbw ops which may help Silvermont.

Change-Id: Ic7aec15245c9e5b2f3903dc7631f38e60be7c93d

test/sad_test.cc

@@ -702,18 +702,6 @@ INSTANTIATE_TEST_CASE_P(MMX, SADTest, ::testing::ValuesIn(mmx_tests));
 
 #if HAVE_SSE
 #if CONFIG_USE_X86INC
-const SadMxNParam sse_tests[] = {
-  make_tuple(4, 8, &vpx_sad4x8_sse, -1),
-  make_tuple(4, 4, &vpx_sad4x4_sse, -1),
-};
-INSTANTIATE_TEST_CASE_P(SSE, SADTest, ::testing::ValuesIn(sse_tests));
-
-const SadMxNAvgParam avg_sse_tests[] = {
-  make_tuple(4, 8, &vpx_sad4x8_avg_sse, -1),
-  make_tuple(4, 4, &vpx_sad4x4_avg_sse, -1),
-};
-INSTANTIATE_TEST_CASE_P(SSE, SADavgTest, ::testing::ValuesIn(avg_sse_tests));
-
 const SadMxNx4Param x4d_sse_tests[] = {
   make_tuple(4, 8, &vpx_sad4x8x4d_sse, -1),
   make_tuple(4, 4, &vpx_sad4x4x4d_sse, -1),
@@ -736,6 +724,8 @@ const SadMxNParam sse2_tests[] = {
   make_tuple(8, 16, &vpx_sad8x16_sse2, -1),
   make_tuple(8, 8, &vpx_sad8x8_sse2, -1),
   make_tuple(8, 4, &vpx_sad8x4_sse2, -1),
+  make_tuple(4, 8, &vpx_sad4x8_sse2, -1),
+  make_tuple(4, 4, &vpx_sad4x4_sse2, -1),
 #if CONFIG_VP9_HIGHBITDEPTH
   make_tuple(64, 64, &vpx_highbd_sad64x64_sse2, 8),
   make_tuple(64, 32, &vpx_highbd_sad64x32_sse2, 8),
@@ -786,6 +776,8 @@ const SadMxNAvgParam avg_sse2_tests[] = {
   make_tuple(8, 16, &vpx_sad8x16_avg_sse2, -1),
   make_tuple(8, 8, &vpx_sad8x8_avg_sse2, -1),
   make_tuple(8, 4, &vpx_sad8x4_avg_sse2, -1),
+  make_tuple(4, 8, &vpx_sad4x8_avg_sse2, -1),
+  make_tuple(4, 4, &vpx_sad4x4_avg_sse2, -1),
 #if CONFIG_VP9_HIGHBITDEPTH
   make_tuple(64, 64, &vpx_highbd_sad64x64_avg_sse2, 8),
   make_tuple(64, 32, &vpx_highbd_sad64x32_avg_sse2, 8),

vpx_dsp/vpx_dsp_rtcd_defs.pl

@@ -990,10 +990,10 @@ add_proto qw/unsigned int vpx_sad8x4/, "const uint8_t *src_ptr, int src_stride,
 specialize qw/vpx_sad8x4 msa/, "$sse2_x86inc";
 
 add_proto qw/unsigned int vpx_sad4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-specialize qw/vpx_sad4x8 msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x8 msa/, "$sse2_x86inc";
 
 add_proto qw/unsigned int vpx_sad4x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-specialize qw/vpx_sad4x4 mmx neon msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x4 mmx neon msa/, "$sse2_x86inc";
 
 #
 # Avg
@@ -1061,10 +1061,10 @@ add_proto qw/unsigned int vpx_sad8x4_avg/, "const uint8_t *src_ptr, int src_stri
 specialize qw/vpx_sad8x4_avg msa/, "$sse2_x86inc";
 
 add_proto qw/unsigned int vpx_sad4x8_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_sad4x8_avg msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x8_avg msa/, "$sse2_x86inc";
 
 add_proto qw/unsigned int vpx_sad4x4_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_sad4x4_avg msa/, "$sse_x86inc";
+specialize qw/vpx_sad4x4_avg msa/, "$sse2_x86inc";
 
 #
 # Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally

vpx_dsp/x86/sad_sse2.asm

@@ -222,8 +222,8 @@ SAD8XN 16, 1 ; sad8x16_avg_sse2
 SAD8XN  8, 1 ; sad8x8_avg_sse2
 SAD8XN  4, 1 ; sad8x4_avg_sse2
 
-; unsigned int vpx_sad4x{4, 8}_sse(uint8_t *src, int src_stride,
-;                                  uint8_t *ref, int ref_stride);
+; unsigned int vpx_sad4x{4, 8}_sse2(uint8_t *src, int src_stride,
+;                                   uint8_t *ref, int ref_stride);
 %macro SAD4XN 1-2 0
   SAD_FN 4, %1, 7, %2
   mov              n_rowsd, %1/4
@@ -236,10 +236,10 @@ SAD8XN  4, 1 ; sad8x4_avg_sse2
   movd                  m4, [refq+ref_stride3q]
   punpckldq             m1, m2
   punpckldq             m3, m4
+  movlhps               m1, m3
 %if %2 == 1
   pavgb                 m1, [second_predq+mmsize*0]
-  pavgb                 m3, [second_predq+mmsize*1]
-  lea         second_predq, [second_predq+mmsize*2]
+  lea         second_predq, [second_predq+mmsize*1]
 %endif
   movd                  m2, [srcq]
   movd                  m5, [srcq+src_strideq]
@@ -247,20 +247,21 @@ SAD8XN  4, 1 ; sad8x4_avg_sse2
   movd                  m6, [srcq+src_stride3q]
   punpckldq             m2, m5
   punpckldq             m4, m6
+  movlhps               m2, m4
   psadbw                m1, m2
-  psadbw                m3, m4
   lea                 refq, [refq+ref_strideq*4]
   paddd                 m0, m1
   lea                 srcq, [srcq+src_strideq*4]
-  paddd                 m0, m3
   dec              n_rowsd
   jg .loop
 
+  movhlps               m1, m0
+  paddd                 m0, m1
   movd                 eax, m0
   RET
 %endmacro
 
-INIT_MMX sse
+INIT_XMM sse2
 SAD4XN  8 ; sad4x8_sse
 SAD4XN  4 ; sad4x4_sse
 SAD4XN  8, 1 ; sad4x8_avg_sse