From c91dd55eda7717d7a5f9726977138fb430cbbcfb Mon Sep 17 00:00:00 2001 From: Jian Zhou Date: Fri, 18 Dec 2015 15:25:27 -0800 Subject: [PATCH] Code clean of highbd_v_predictor_4x4 MMX replaced with SSE2, same performance. Change-Id: I2ab8f30a71e5fadbbc172fb385093dec1e11a696 --- test/vp9_intrapred_test.cc | 12 ++++++------ vpx_dsp/vpx_dsp_rtcd_defs.pl | 2 +- vpx_dsp/x86/highbd_intrapred_sse2.asm | 2 +- 3 files changed, 8 insertions(+), 8 deletions(-) diff --git a/test/vp9_intrapred_test.cc b/test/vp9_intrapred_test.cc index ad3327e2d..bc85daff8 100644 --- a/test/vp9_intrapred_test.cc +++ b/test/vp9_intrapred_test.cc @@ -147,7 +147,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest, &vpx_highbd_dc_predictor_8x8_c, 8, 8), make_tuple(&vpx_highbd_dc_predictor_16x16_sse2, &vpx_highbd_dc_predictor_16x16_c, 16, 8), - make_tuple(&vpx_highbd_v_predictor_4x4_sse, + make_tuple(&vpx_highbd_v_predictor_4x4_sse2, &vpx_highbd_v_predictor_4x4_c, 4, 8), make_tuple(&vpx_highbd_v_predictor_8x8_sse2, &vpx_highbd_v_predictor_8x8_c, 8, 8), @@ -168,7 +168,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest, &vpx_highbd_dc_predictor_8x8_c, 8, 8), make_tuple(&vpx_highbd_dc_predictor_16x16_sse2, &vpx_highbd_dc_predictor_16x16_c, 16, 8), - make_tuple(&vpx_highbd_v_predictor_4x4_sse, + make_tuple(&vpx_highbd_v_predictor_4x4_sse2, &vpx_highbd_v_predictor_4x4_c, 4, 8), make_tuple(&vpx_highbd_v_predictor_8x8_sse2, &vpx_highbd_v_predictor_8x8_c, 8, 8), @@ -201,7 +201,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest, make_tuple(&vpx_highbd_dc_predictor_16x16_sse2, &vpx_highbd_dc_predictor_16x16_c, 16, 10), - make_tuple(&vpx_highbd_v_predictor_4x4_sse, + make_tuple(&vpx_highbd_v_predictor_4x4_sse2, &vpx_highbd_v_predictor_4x4_c, 4, 10), make_tuple(&vpx_highbd_v_predictor_8x8_sse2, &vpx_highbd_v_predictor_8x8_c, 8, 10), @@ -225,7 +225,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest, make_tuple(&vpx_highbd_dc_predictor_16x16_sse2, &vpx_highbd_dc_predictor_16x16_c, 16, 10), - make_tuple(&vpx_highbd_v_predictor_4x4_sse, + make_tuple(&vpx_highbd_v_predictor_4x4_sse2, &vpx_highbd_v_predictor_4x4_c, 4, 10), make_tuple(&vpx_highbd_v_predictor_8x8_sse2, &vpx_highbd_v_predictor_8x8_c, 8, 10), @@ -258,7 +258,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest, make_tuple(&vpx_highbd_dc_predictor_16x16_sse2, &vpx_highbd_dc_predictor_16x16_c, 16, 12), - make_tuple(&vpx_highbd_v_predictor_4x4_sse, + make_tuple(&vpx_highbd_v_predictor_4x4_sse2, &vpx_highbd_v_predictor_4x4_c, 4, 12), make_tuple(&vpx_highbd_v_predictor_8x8_sse2, &vpx_highbd_v_predictor_8x8_c, 8, 12), @@ -282,7 +282,7 @@ INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest, make_tuple(&vpx_highbd_dc_predictor_16x16_sse2, &vpx_highbd_dc_predictor_16x16_c, 16, 12), - make_tuple(&vpx_highbd_v_predictor_4x4_sse, + make_tuple(&vpx_highbd_v_predictor_4x4_sse2, &vpx_highbd_v_predictor_4x4_c, 4, 12), make_tuple(&vpx_highbd_v_predictor_8x8_sse2, &vpx_highbd_v_predictor_8x8_c, 8, 12), diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl index 897913571..3f4fe1127 100644 --- a/vpx_dsp/vpx_dsp_rtcd_defs.pl +++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl @@ -288,7 +288,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") { specialize qw/vpx_highbd_d153_predictor_4x4/; add_proto qw/void vpx_highbd_v_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd"; - specialize qw/vpx_highbd_v_predictor_4x4/, "$sse_x86inc"; + specialize qw/vpx_highbd_v_predictor_4x4/, "$sse2_x86inc"; add_proto qw/void vpx_highbd_tm_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd"; specialize qw/vpx_highbd_tm_predictor_4x4/, "$sse_x86inc"; diff --git a/vpx_dsp/x86/highbd_intrapred_sse2.asm b/vpx_dsp/x86/highbd_intrapred_sse2.asm index b12d29c0a..9e6dcb50c 100644 --- a/vpx_dsp/x86/highbd_intrapred_sse2.asm +++ b/vpx_dsp/x86/highbd_intrapred_sse2.asm @@ -183,7 +183,7 @@ cglobal highbd_dc_predictor_32x32, 4, 5, 9, dst, stride, above, left, goffset REP_RET %endif -INIT_MMX sse +INIT_XMM sse2 cglobal highbd_v_predictor_4x4, 3, 3, 1, dst, stride, above movq m0, [aboveq] movq [dstq ], m0