diff options
| -rw-r--r-- | test/test_intra_pred_speed.cc | 14 | ||||
| -rw-r--r-- | vpx_dsp/vpx_dsp_rtcd_defs.pl | 2 | ||||
| -rw-r--r-- | vpx_dsp/x86/intrapred_sse2.asm | 28 | ||||
| -rw-r--r-- | vpx_dsp/x86/intrapred_ssse3.asm | 20 |
4 files changed, 36 insertions, 28 deletions
diff --git a/test/test_intra_pred_speed.cc b/test/test_intra_pred_speed.cc index c270072d7..4064ea645 100644 --- a/test/test_intra_pred_speed.cc +++ b/test/test_intra_pred_speed.cc @@ -342,23 +342,23 @@ INTRA_PRED_TEST(SSE2, TestIntraPred32, vpx_dc_predictor_32x32_sse2, vpx_dc_left_predictor_32x32_sse2, vpx_dc_top_predictor_32x32_sse2, vpx_dc_128_predictor_32x32_sse2, vpx_v_predictor_32x32_sse2, - NULL, NULL, NULL, NULL, NULL, NULL, NULL, - vpx_tm_predictor_32x32_sse2) + vpx_h_predictor_32x32_sse2, NULL, NULL, NULL, NULL, NULL, + NULL, vpx_tm_predictor_32x32_sse2) #else INTRA_PRED_TEST(SSE2, TestIntraPred32, vpx_dc_predictor_32x32_sse2, vpx_dc_left_predictor_32x32_sse2, vpx_dc_top_predictor_32x32_sse2, vpx_dc_128_predictor_32x32_sse2, vpx_v_predictor_32x32_sse2, - NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL) + vpx_h_predictor_32x32_sse2, NULL, NULL, NULL, NULL, NULL, + NULL, NULL) #endif // ARCH_X86_64 #endif // HAVE_SSE2 && CONFIG_USE_X86INC #if HAVE_SSSE3 && CONFIG_USE_X86INC INTRA_PRED_TEST(SSSE3, TestIntraPred32, NULL, NULL, NULL, NULL, NULL, - vpx_h_predictor_32x32_ssse3, vpx_d45_predictor_32x32_ssse3, - NULL, NULL, vpx_d153_predictor_32x32_ssse3, - vpx_d207_predictor_32x32_ssse3, vpx_d63_predictor_32x32_ssse3, - NULL) + NULL, vpx_d45_predictor_32x32_ssse3, NULL, NULL, + vpx_d153_predictor_32x32_ssse3, vpx_d207_predictor_32x32_ssse3, + vpx_d63_predictor_32x32_ssse3, NULL) #endif // HAVE_SSSE3 && CONFIG_USE_X86INC #if HAVE_NEON diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl index 5ce7134e7..4d36e2796 100644 --- a/vpx_dsp/vpx_dsp_rtcd_defs.pl +++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl @@ -226,7 +226,7 @@ add_proto qw/void vpx_d63e_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, specialize qw/vpx_d63e_predictor_32x32/; add_proto qw/void vpx_h_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; -specialize qw/vpx_h_predictor_32x32 neon msa/, "$ssse3_x86inc"; +specialize qw/vpx_h_predictor_32x32 neon msa/, "$sse2_x86inc"; add_proto qw/void vpx_d117_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left"; specialize qw/vpx_d117_predictor_32x32/; diff --git a/vpx_dsp/x86/intrapred_sse2.asm b/vpx_dsp/x86/intrapred_sse2.asm index edbf05e33..4aca3ec16 100644 --- a/vpx_dsp/x86/intrapred_sse2.asm +++ b/vpx_dsp/x86/intrapred_sse2.asm @@ -582,6 +582,34 @@ cglobal h_predictor_16x16, 2, 5, 3, dst, stride, line, left REP_RET INIT_XMM sse2 +cglobal h_predictor_32x32, 2, 5, 3, dst, stride, line, left + movifnidn leftq, leftmp + mov lineq, -8 + DEFINE_ARGS dst, stride, line, left, stride3 + lea stride3q, [strideq*3] +.loop: + movd m0, [leftq] + punpcklbw m0, m0 + punpcklbw m0, m0 ; l1 to l4 each repeated 4 times + pshufd m1, m0, 0x0 ; l1 repeated 16 times + pshufd m2, m0, 0x55 ; l2 repeated 16 times + mova [dstq ], m1 + mova [dstq+16 ], m1 + mova [dstq+strideq ], m2 + mova [dstq+strideq+16 ], m2 + pshufd m1, m0, 0xaa + pshufd m2, m0, 0xff + mova [dstq+strideq*2 ], m1 + mova [dstq+strideq*2+16], m1 + mova [dstq+stride3q ], m2 + mova [dstq+stride3q+16 ], m2 + inc lineq + lea leftq, [leftq+4 ] + 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 movq m0, [aboveq-1]; [63:0] tl t1 t2 t3 t4 x x x diff --git a/vpx_dsp/x86/intrapred_ssse3.asm b/vpx_dsp/x86/intrapred_ssse3.asm index f1a193b19..d061278c7 100644 --- a/vpx_dsp/x86/intrapred_ssse3.asm +++ b/vpx_dsp/x86/intrapred_ssse3.asm @@ -33,26 +33,6 @@ sh_b2333: db 2, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 SECTION .text -INIT_XMM ssse3 -cglobal h_predictor_32x32, 2, 4, 3, dst, stride, line, left - movifnidn leftq, leftmp - add leftq, 32 - mov lineq, -16 - pxor m0, m0 -.loop: - movd m1, [leftq+lineq*2 ] - movd m2, [leftq+lineq*2+1] - pshufb m1, m0 - pshufb m2, m0 - mova [dstq ], m1 - mova [dstq +16], m1 - mova [dstq+strideq ], m2 - mova [dstq+strideq+16], m2 - lea dstq, [dstq+strideq*2] - inc lineq - jnz .loop - REP_RET - INIT_MMX ssse3 cglobal d45_predictor_4x4, 3, 4, 4, dst, stride, above, goffset GET_GOT goffsetq |