Code clean of highbd_dc_predictor_32x32

Remove the ARCH_X86_64 constraint.

Change-Id: I7d2545fc4f24eb352cf3e03082fc4d48d46fbb09

diff --git a/test/vp9_intrapred_test.cc b/test/vp9_intrapred_test.cc
index 2bebdcb..db41da5 100644 (file)
--- a/test/vp9_intrapred_test.cc
+++ b/test/vp9_intrapred_test.cc
@@ -162,6 +162,8 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
 #else
 INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
                         ::testing::Values(
+                            make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
+                                       &vpx_highbd_dc_predictor_32x32_c, 32, 8),
                             make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
                                        &vpx_highbd_dc_predictor_4x4_c, 4, 8),
                             make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
@@ -218,6 +220,9 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
 #else
 INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
                         ::testing::Values(
+                            make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
+                                       &vpx_highbd_dc_predictor_32x32_c, 32,
+                                       10),
                             make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
                                        &vpx_highbd_dc_predictor_4x4_c, 4, 10),
                             make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,
@@ -275,6 +280,9 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
 #else
 INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
                         ::testing::Values(
+                            make_tuple(&vpx_highbd_dc_predictor_32x32_sse2,
+                                       &vpx_highbd_dc_predictor_32x32_c, 32,
+                                       12),
                             make_tuple(&vpx_highbd_dc_predictor_4x4_sse2,
                                        &vpx_highbd_dc_predictor_4x4_c, 4, 12),
                             make_tuple(&vpx_highbd_dc_predictor_8x8_sse2,

diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index a2a0674..23d7a24 100644 (file)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -438,7 +438,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_highbd_tm_predictor_32x32/, "$sse2_x86_64_x86inc";
 
   add_proto qw/void vpx_highbd_dc_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
-  specialize qw/vpx_highbd_dc_predictor_32x32/, "$sse2_x86_64_x86inc";
+  specialize qw/vpx_highbd_dc_predictor_32x32/, "$sse2_x86inc";
 
   add_proto qw/void vpx_highbd_dc_top_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
   specialize qw/vpx_highbd_dc_top_predictor_32x32/;

diff --git a/vpx_dsp/x86/highbd_intrapred_sse2.asm b/vpx_dsp/x86/highbd_intrapred_sse2.asm
index 233958a..6719d86 100644 (file)
--- a/vpx_dsp/x86/highbd_intrapred_sse2.asm
+++ b/vpx_dsp/x86/highbd_intrapred_sse2.asm
@@ -118,30 +118,29 @@ cglobal highbd_dc_predictor_16x16, 4, 5, 5, dst, stride, above, left, goffset
   RESTORE_GOT
   REP_RET
 
-%if ARCH_X86_64
 INIT_XMM sse2
-cglobal highbd_dc_predictor_32x32, 4, 5, 9, dst, stride, above, left, goffset
+cglobal highbd_dc_predictor_32x32, 4, 5, 7, dst, stride, above, left, goffset
   GET_GOT     goffsetq
 
-  pxor                  m1, m1
   mova                  m0, [aboveq]
   mova                  m2, [aboveq+16]
   mova                  m3, [aboveq+32]
   mova                  m4, [aboveq+48]
-  mova                  m5, [leftq]
-  mova                  m6, [leftq+16]
-  mova                  m7, [leftq+32]
-  mova                  m8, [leftq+48]
+  paddw                 m0, m2
+  paddw                 m3, m4
+  mova                  m2, [leftq]
+  mova                  m4, [leftq+16]
+  mova                  m5, [leftq+32]
+  mova                  m6, [leftq+48]
+  paddw                 m2, m4
+  paddw                 m5, m6
+  paddw                 m0, m3
+  paddw                 m2, m5
+  pxor                  m1, m1
+  paddw                 m0, m2
   DEFINE_ARGS dst, stride, stride3, lines4
   lea             stride3q, [strideq*3]
   mov              lines4d, 8
-  paddw                 m0, m2
-  paddw                 m0, m3
-  paddw                 m0, m4
-  paddw                 m0, m5
-  paddw                 m0, m6
-  paddw                 m0, m7
-  paddw                 m0, m8
   movhlps               m2, m0
   paddw                 m0, m2
   punpcklwd             m0, m1
@@ -177,7 +176,6 @@ cglobal highbd_dc_predictor_32x32, 4, 5, 9, dst, stride, above, left, goffset
 
   RESTORE_GOT
   REP_RET
-%endif
 
 INIT_XMM sse2
 cglobal highbd_v_predictor_4x4, 3, 3, 1, dst, stride, above