generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge "SSE2 speed up of h_predictor_4x4"

Browse Source
This commit is contained in:
Jian Zhou 2015-12-02 18:49:00 +00:00 committed by Gerrit Code Review
commit 623e988add
4 changed files with 23 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
test/test_intra_pred_speed.cc
View File

@ -195,15 +195,16 @@ INTRA_PRED_TEST(SSE, TestIntraPred4, vpx_dc_predictor_4x4_sse,
#endif // HAVE_SSE && CONFIG_USE_X86INC
#if HAVE_SSE2 && CONFIG_USE_X86INC
INTRA_PRED_TEST(SSE2, TestIntraPred4, NULL, NULL, NULL, NULL, NULL, NULL,
NULL, NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_4x4_sse2)
INTRA_PRED_TEST(SSE2, TestIntraPred4, NULL, NULL, NULL, NULL, NULL,
vpx_h_predictor_4x4_sse2, NULL, NULL, NULL, NULL, NULL, NULL,
vpx_tm_predictor_4x4_sse2)
#endif // HAVE_SSE2 && CONFIG_USE_X86INC
#if HAVE_SSSE3 && CONFIG_USE_X86INC
INTRA_PRED_TEST(SSSE3, TestIntraPred4, NULL, NULL, NULL, NULL, NULL,
vpx_h_predictor_4x4_ssse3, vpx_d45_predictor_4x4_ssse3, NULL,
NULL, vpx_d153_predictor_4x4_ssse3,
vpx_d207_predictor_4x4_ssse3, vpx_d63_predictor_4x4_ssse3, NULL)
NULL, vpx_d45_predictor_4x4_ssse3, NULL, NULL,
vpx_d153_predictor_4x4_ssse3, vpx_d207_predictor_4x4_ssse3,
vpx_d63_predictor_4x4_ssse3, NULL)
#endif // HAVE_SSSE3 && CONFIG_USE_X86INC
#if HAVE_DSPR2

2
vpx_dsp/vpx_dsp_rtcd_defs.pl
View File

@ -76,7 +76,7 @@ add_proto qw/void vpx_d63f_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, co
specialize qw/vpx_d63f_predictor_4x4/;
add_proto qw/void vpx_h_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_h_predictor_4x4 neon dspr2 msa/, "$ssse3_x86inc";
specialize qw/vpx_h_predictor_4x4 neon dspr2 msa/, "$sse2_x86inc";
add_proto qw/void vpx_he_predictor_4x4/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
specialize qw/vpx_he_predictor_4x4/;

16
vpx_dsp/x86/intrapred_sse2.asm
View File

@ -515,6 +515,22 @@ cglobal v_predictor_32x32, 3, 4, 2, dst, stride, above
jnz .loop
REP_RET
INIT_XMM sse2
cglobal h_predictor_4x4, 2, 4, 4, dst, stride, line, left
movifnidn leftq, leftmp
movd m0, [leftq]
punpcklbw m0, m0
punpcklbw m0, m0
pshufd m1, m0, 0x1
movd [dstq ], m0
movd [dstq+strideq], m1
pshufd m2, m0, 0x2
lea dstq, [dstq+strideq*2]
pshufd m3, m0, 0x3
movd [dstq ], m2
movd [dstq+strideq], m3
RET
INIT_XMM sse2
cglobal tm_predictor_4x4, 4, 4, 5, dst, stride, above, left
pxor m1, m1

18
vpx_dsp/x86/intrapred_ssse3.asm
View File

@ -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_4x4, 2, 4, 3, dst, stride, line, left
movifnidn leftq, leftmp
add leftq, 4
mov lineq, -2
pxor m0, m0
.loop:
movd m1, [leftq+lineq*2 ]
movd m2, [leftq+lineq*2+1]
pshufb m1, m0
pshufb m2, m0
movd [dstq ], m1
movd [dstq+strideq], m2
lea dstq, [dstq+strideq*2]
inc lineq
jnz .loop
REP_RET
INIT_MMX ssse3
cglobal h_predictor_8x8, 2, 4, 3, dst, stride, line, left
movifnidn leftq, leftmp