From aa5b517a39ae9f311129636a2c0dc41892d3d44f Mon Sep 17 00:00:00 2001 From: Jian Zhou Date: Mon, 7 Dec 2015 18:50:37 -0800 Subject: [PATCH] Re-enable SSE2 based intra 4x4 prediction 4x4 Intra predictor implemented with MMX is replaced with SSE2. Segfault in change 315561 when decoding vp8 is taken care of. Change-Id: I083a7cb4eb8982954c20865160f91ebec777ec76 --- test/test_intra_pred_speed.cc | 11 +++-------- vpx_dsp/vpx_dsp_rtcd_defs.pl | 10 +++++----- vpx_dsp/x86/intrapred_sse2.asm | 30 ++++++++++++++++-------------- 3 files changed, 24 insertions(+), 27 deletions(-) diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc index 5f4c48f84..c270072d7 100644 --- a/test/test_intra_pred_speed.cc +++ b/test/test_intra_pred_speed.cc @@ -187,15 +187,10 @@ INTRA_PRED_TEST(C, TestIntraPred4, vpx_dc_predictor_4x4_c, vpx_d153_predictor_4x4_c, vpx_d207_predictor_4x4_c, vpx_d63_predictor_4x4_c, vpx_tm_predictor_4x4_c) -#if HAVE_SSE && CONFIG_USE_X86INC -INTRA_PRED_TEST(SSE, TestIntraPred4, vpx_dc_predictor_4x4_sse, - vpx_dc_left_predictor_4x4_sse, vpx_dc_top_predictor_4x4_sse, - vpx_dc_128_predictor_4x4_sse, vpx_v_predictor_4x4_sse, NULL, - NULL, NULL, NULL, NULL, NULL, NULL, NULL) -#endif // HAVE_SSE && CONFIG_USE_X86INC - #if HAVE_SSE2 && CONFIG_USE_X86INC -INTRA_PRED_TEST(SSE2, TestIntraPred4, NULL, NULL, NULL, NULL, NULL, +INTRA_PRED_TEST(SSE2, TestIntraPred4, vpx_dc_predictor_4x4_sse2, + vpx_dc_left_predictor_4x4_sse2, vpx_dc_top_predictor_4x4_sse2, + vpx_dc_128_predictor_4x4_sse2, vpx_v_predictor_4x4_sse2, vpx_h_predictor_4x4_sse2, NULL, NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_4x4_sse2) #endif // HAVE_SSE2 && CONFIG_USE_X86INC diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl index bb9981bbf..5ce7134e7 100644 --- a/vpx_dsp/vpx_dsp_rtcd_defs.pl +++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl @@ -91,7 +91,7 @@ add_proto qw/void vpx_d153_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, co specialize qw/vpx_d153_predictor_4x4/, "$ssse3_x86inc"; add_proto qw/void vpx_v_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; -specialize qw/vpx_v_predictor_4x4 neon msa/, "$sse_x86inc"; +specialize qw/vpx_v_predictor_4x4 neon msa/, "$sse2_x86inc"; add_proto qw/void vpx_ve_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; specialize qw/vpx_ve_predictor_4x4/; @@ -100,16 +100,16 @@ add_proto qw/void vpx_tm_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, cons specialize qw/vpx_tm_predictor_4x4 neon dspr2 msa/, "$sse2_x86inc"; add_proto qw/void vpx_dc_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; -specialize qw/vpx_dc_predictor_4x4 dspr2 msa neon/, "$sse_x86inc"; +specialize qw/vpx_dc_predictor_4x4 dspr2 msa neon/, "$sse2_x86inc"; add_proto qw/void vpx_dc_top_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; -specialize qw/vpx_dc_top_predictor_4x4 msa neon/, "$sse_x86inc"; +specialize qw/vpx_dc_top_predictor_4x4 msa neon/, "$sse2_x86inc"; add_proto qw/void vpx_dc_left_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; -specialize qw/vpx_dc_left_predictor_4x4 msa neon/, "$sse_x86inc"; +specialize qw/vpx_dc_left_predictor_4x4 msa neon/, "$sse2_x86inc"; add_proto qw/void vpx_dc_128_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; -specialize qw/vpx_dc_128_predictor_4x4 msa neon/, "$sse_x86inc"; +specialize qw/vpx_dc_128_predictor_4x4 msa neon/, "$sse2_x86inc"; add_proto qw/void vpx_d207_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; specialize qw/vpx_d207_predictor_8x8/, "$ssse3_x86inc"; diff --git a/vpx_dsp/x86/intrapred_sse2.asm b/vpx_dsp/x86/intrapred_sse2.asm index 13fe1402b..edbf05e33 100644 --- a/vpx_dsp/x86/intrapred_sse2.asm +++ b/vpx_dsp/x86/intrapred_sse2.asm @@ -23,17 +23,18 @@ pw2_32: times 8 dw 16 SECTION .text -INIT_MMX sse -cglobal dc_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset +INIT_XMM sse2 +cglobal dc_predictor_4x4, 4, 5, 3, dst, stride, above, left, goffset GET_GOT goffsetq - pxor m1, m1 + movd m2, [leftq] movd m0, [aboveq] - punpckldq m0, [leftq] + pxor m1, m1 + punpckldq m0, m2 psadbw m0, m1 paddw m0, [GLOBAL(pw_4)] psraw m0, 3 - pshufw m0, m0, 0x0 + pshuflw m0, m0, 0x0 packuswb m0, m0 movd [dstq ], m0 movd [dstq+strideq], m0 @@ -44,16 +45,17 @@ cglobal dc_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset RESTORE_GOT RET -INIT_MMX sse -cglobal dc_left_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset +INIT_XMM sse2 +cglobal dc_left_predictor_4x4, 2, 5, 2, dst, stride, above, left, goffset GET_GOT goffsetq + movifnidn leftq, leftmp pxor m1, m1 movd m0, [leftq] psadbw m0, m1 paddw m0, [GLOBAL(pw2_4)] psraw m0, 2 - pshufw m0, m0, 0x0 + pshuflw m0, m0, 0x0 packuswb m0, m0 movd [dstq ], m0 movd [dstq+strideq], m0 @@ -64,8 +66,8 @@ cglobal dc_left_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset RESTORE_GOT RET -INIT_MMX sse -cglobal dc_top_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset +INIT_XMM sse2 +cglobal dc_top_predictor_4x4, 3, 5, 2, dst, stride, above, left, goffset GET_GOT goffsetq pxor m1, m1 @@ -73,7 +75,7 @@ cglobal dc_top_predictor_4x4, 4, 5, 2, dst, stride, above, left, goffset psadbw m0, m1 paddw m0, [GLOBAL(pw2_4)] psraw m0, 2 - pshufw m0, m0, 0x0 + pshuflw m0, m0, 0x0 packuswb m0, m0 movd [dstq ], m0 movd [dstq+strideq], m0 @@ -166,8 +168,8 @@ cglobal dc_left_predictor_8x8, 2, 5, 2, dst, stride, above, left, goffset RESTORE_GOT RET -INIT_MMX sse -cglobal dc_128_predictor_4x4, 4, 5, 3, dst, stride, above, left, goffset +INIT_XMM sse2 +cglobal dc_128_predictor_4x4, 2, 5, 1, dst, stride, above, left, goffset GET_GOT goffsetq DEFINE_ARGS dst, stride, stride3 @@ -453,7 +455,7 @@ cglobal dc_128_predictor_32x32, 4, 5, 3, dst, stride, above, left, goffset RESTORE_GOT RET -INIT_MMX sse +INIT_XMM sse2 cglobal v_predictor_4x4, 3, 3, 1, dst, stride, above movd m0, [aboveq] movd [dstq ], m0