Speed up h_predictor_8x8

Relocate the function from SSSE3 to SSE2, Unroll loop from 4 to 2,
and reduce mem access to left.
Speed up by >20% in ./test_intra_pred_speed.

Change-Id: Ib9f1846819783b6e05e2a310c930eb844b2b4d2e

test/test_intra_pred_speed.cc

@@ -239,15 +239,16 @@ INTRA_PRED_TEST(C, TestIntraPred8, vpx_dc_predictor_8x8_c,
 #if HAVE_SSE2 && CONFIG_USE_X86INC
 INTRA_PRED_TEST(SSE2, TestIntraPred8, vpx_dc_predictor_8x8_sse2,
                 vpx_dc_left_predictor_8x8_sse2, vpx_dc_top_predictor_8x8_sse2,
-                vpx_dc_128_predictor_8x8_sse2, vpx_v_predictor_8x8_sse2, NULL,
-                NULL, NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_8x8_sse2)
+                vpx_dc_128_predictor_8x8_sse2, vpx_v_predictor_8x8_sse2,
+                vpx_h_predictor_8x8_sse2, NULL, NULL, NULL, NULL, NULL,
+                NULL, vpx_tm_predictor_8x8_sse2)
 #endif  // HAVE_SSE2 && CONFIG_USE_X86INC
 
 #if HAVE_SSSE3 && CONFIG_USE_X86INC
 INTRA_PRED_TEST(SSSE3, TestIntraPred8, NULL, NULL, NULL, NULL, NULL,
-                vpx_h_predictor_8x8_ssse3, vpx_d45_predictor_8x8_ssse3, NULL,
-                NULL, vpx_d153_predictor_8x8_ssse3,
-                vpx_d207_predictor_8x8_ssse3, vpx_d63_predictor_8x8_ssse3, NULL)
+                NULL, vpx_d45_predictor_8x8_ssse3, NULL, NULL,
+                vpx_d153_predictor_8x8_ssse3, vpx_d207_predictor_8x8_ssse3,
+                vpx_d63_predictor_8x8_ssse3, NULL)
 #endif  // HAVE_SSSE3 && CONFIG_USE_X86INC
 
 #if HAVE_DSPR2

vpx_dsp/vpx_dsp_rtcd_defs.pl

@@ -130,7 +130,7 @@ add_proto qw/void vpx_d63e_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, co
 specialize qw/vpx_d63e_predictor_8x8/;
 
 add_proto qw/void vpx_h_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_8x8 neon dspr2 msa/, "$ssse3_x86inc";
+specialize qw/vpx_h_predictor_8x8 neon dspr2 msa/, "$sse2_x86inc";
 
 add_proto qw/void vpx_d117_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
 specialize qw/vpx_d117_predictor_8x8/;

vpx_dsp/x86/intrapred_sse2.asm

@@ -533,6 +533,29 @@ cglobal h_predictor_4x4, 2, 4, 4, dst, stride, line, left
   movd      [dstq+strideq], m3
   RET
 
+INIT_XMM sse2
+cglobal h_predictor_8x8, 2, 5, 3, dst, stride, line, left
+  movifnidn          leftq, leftmp
+  mov                lineq, -2
+  DEFINE_ARGS  dst, stride, line, left, stride3
+  lea             stride3q, [strideq*3]
+  movq                  m0, [leftq    ]
+  punpcklbw             m0, m0              ; l1 l1 l2 l2 ... l8 l8
+.loop:
+  pshuflw               m1, m0, 0x0         ; l1 l1 l1 l1 l1 l1 l1 l1
+  pshuflw               m2, m0, 0x55        ; l2 l2 l2 l2 l2 l2 l2 l2
+  movq      [dstq        ], m1
+  movq      [dstq+strideq], m2
+  pshuflw               m1, m0, 0xaa
+  pshuflw               m2, m0, 0xff
+  movq    [dstq+strideq*2], m1
+  movq    [dstq+stride3q ], m2
+  pshufd                m0, m0, 0xe         ; [63:0] l5 l5 l6 l6 l7 l7 l8 l8
+  inc                lineq
+  lea                 dstq, [dstq+strideq*4]
+  jnz .loop
+  REP_RET
+
 INIT_XMM sse2
 cglobal tm_predictor_4x4, 4, 4, 5, dst, stride, above, left
   pxor                  m1, m1

vpx_dsp/x86/intrapred_ssse3.asm

@@ -33,24 +33,6 @@ sh_b2333: db 2, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
 
 SECTION .text
 
-INIT_MMX ssse3
-cglobal h_predictor_8x8, 2, 4, 3, dst, stride, line, left
-  movifnidn          leftq, leftmp
-  add                leftq, 8
-  mov                lineq, -4
-  pxor                  m0, m0
-.loop:
-  movd                  m1, [leftq+lineq*2  ]
-  movd                  m2, [leftq+lineq*2+1]
-  pshufb                m1, m0
-  pshufb                m2, m0
-  movq      [dstq        ], m1
-  movq      [dstq+strideq], m2
-  lea                 dstq, [dstq+strideq*2]
-  inc                lineq
-  jnz .loop
-  REP_RET
-
 INIT_XMM ssse3
 cglobal h_predictor_16x16, 2, 4, 3, dst, stride, line, left
   movifnidn          leftq, leftmp