FFmpeg

Commit Graph

Author	SHA1	Message	Date
James Almer	fc8db12a73	x86/vp9: inital AVX2 intra_pred tos3k-vp9-b10000.webm on a Core i5-4200U @1.6GHz 1219 decicycles in ff_vp9_ipred_dc_32x32_ssse3, 131070 runs, 2 skips 439 decicycles in ff_vp9_ipred_dc_32x32_avx2, 131070 runs, 2 skips 3570 decicycles in ff_vp9_ipred_dc_top_32x32_ssse3, 4096 runs, 0 skips 2494 decicycles in ff_vp9_ipred_dc_top_32x32_avx2, 4096 runs, 0 skips 1419 decicycles in ff_vp9_ipred_dc_left_32x32_ssse3, 16384 runs, 0 skips 717 decicycles in ff_vp9_ipred_dc_left_32x32_avx2, 16384 runs, 0 skips 2737 decicycles in ff_vp9_ipred_tm_32x32_avx, 1024 runs, 0 skips 2088 decicycles in ff_vp9_ipred_tm_32x32_avx2, 1024 runs, 0 skips 3090 decicycles in ff_vp9_ipred_v_32x32_avx, 512 runs, 0 skips 2226 decicycles in ff_vp9_ipred_v_32x32_avx2, 512 runs, 0 skips 1565 decicycles in ff_vp9_ipred_h_32x32_avx, 1024 runs, 0 skips 922 decicycles in ff_vp9_ipred_h_32x32_avx2, 1024 runs, 0 skips Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	ec98f80af4	x86/dsputil: move some mmx init code inside dsputil_init_mmx() This reduces differences with the fork Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	ccff45a0d3	apedsp: move to llauddsp APE is not the sole codec using scalarproduct_and_madd_int16. Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	d5c9d055ea	avcodec/x86/dsputilenc_mmx: fix build without yasm Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	625ffa1457	x86/motion_est: sad_{x, y}2_mmxext functions are bitexact Only the xy2 functions aren't. Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Timothy Gu	108dec3055	x86: dsputilenc: convert hf_noise_mmx to yasm Signed-off-by: Timothy Gu <timothygu99@gmail.com> Several bugfixes by: Christophe Gisquet <christophe.gisquet@gmail.com> See: [FFmpeg-devel] [WIP] [PATCH 4/4] x86: dsputilenc: convert hf_noise_mmx to yasm Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	dcd2a6ca36	x86: hevc_mc: remove unneeded shift The immediate value may be 0. Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	09fc28aed1	x86: hevcdsp_init: fix macro usage The macro was not using the parameter but unconditionally using sse4. Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	e1bd40fe6b	x86/motion_est: enable sad16_sse2 on k10 CPUs The check is meant for k8 CPUs. sad16_sse2 is ~20% faster than sad16_mmxext on k10. Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	f128342df2	build: fix compilation of svq1enc_mmx.c with --disable-mmx It's needed for ff_svq1enc_init_x86() even if simd functions are disabled. Alternatively, svq1enc_init.c could be made and the relevant code moved there. Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	4ac41a52e2	x86/huffyuvdsp: fix some prototypes Remove duplicate prototypes and fix int -> intptr_t in another Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	d136fe6fd7	x86: huffyuvdsp: fewer functions for x86_64 When there are 2 functions that are <= SSE2, only one is needed for x86_64. Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Timothy Gu	154cee9292	x86: dsputilenc: convert ff_sse{8, 16}_mmx() to yasm Signed-off-by: Timothy Gu <timothygu99@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Timothy Gu	0b6292b7b8	x86: dsputilenc: move all the function prototypes together Signed-off-by: Timothy Gu <timothygu99@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	f743fa9c7f	x86: huffyuvdsp: add_hfyu_left_pred_bgr32 C MMX SSE2 Cycles: 3092 1053 578 Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	7be79c76d3	avcodec/huffyuvdsp: Change w to intptr in add_hfyu_median_pred() and add_hfyu_left_pred() This avoids potential issues with the high 32bits being random in x86-64 asm Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	884078d2df	x86: huffyuvdsp: add SSE2 median prediction From 5010c to 4566 on lagarith YUY2. Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	8c891d90ca	avcodec/x86/qpeldsp_init: Restore author attribution See: `368f50359e` See: `44eb495128`, and many others See: similarity index 83% copy from libavcodec/x86/dsputil_init.c copy to libavcodec/x86/qpeldsp_init.c index ebbf97f..8f296a1 100644 --- a/libavcodec/x86/dsputil_init.c +++ b/libavcodec/x86/qpeldsp_init.c @@ -1,6 +1,5 @@ /* - * Copyright (c) 2000, 2001 Fabrice Bellard - * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at> + * quarterpel DSP functions * * This file is part of FFmpeg. * Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	c814a6c778	avcodec/x86/svq1enc_mmx: Add author attribution See: `5900637219` Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	02a3e327f1	x86/dsputilenc: add missing guards to ff_pix_sum16_xop XOP support was added in Yasm 1.0.0 and Nasm 2.06, and we still support older versions. Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	99a319c4e7	x86: huffyuvdsp: port add_bytes to yasm C MMX SSE2 Cycles: 2972 587 302 Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	2267003981	x86: hpeldsp: better factorization Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	7b4c46050e	rename add_hfyu_left_prediction_int16 to add_hfyu_left_pred_int16 This makes the naming more consistent with the 8bit variant Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	550ae6c02f	rename add_hfyu_median_prediction_int16 to add_hfyu_median_pred_int16 This makes the naming more consistent with the 8bit variant Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	40a4ab8ba4	rename sub_hfyu_median_prediction_int16 to sub_hfyu_median_pred_int16 This makes the naming more consistent with the 8bit variant Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	05de4d3011	x86/dsputilenc: implement XOP version of pix_sum16 SSE2: 137 cycles XOP: 87 cycles Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Diego Biurrun	368f50359e	dsputil: Split off quarterpel bits into their own context	11 years ago
Diego Biurrun	054013a0fc	dsputil: Move APE-specific bits into apedsp	11 years ago
Diego Biurrun	65d5d58658	dsputil: Move SVQ1 encoding specific bits into svq1enc	11 years ago
Michael Niedermayer	b50559fc0b	libavcodec/x86/dsputilenc: drop and 0xffff that should have becomei redundant Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	561bfc85eb	x86/dsputilenc: implement SSE2 versions of pix_{sum16, norm1} Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	0810608e23	x86: hevc_mc: better register allocation The xmm reg count was incorrect, and manual loading of the gprs furthermore allows to noticeable reduce the number needed. The modified functions are used in weighted prediction, so only a few samples like WP_* exhibit a change. For this one and Win64 (some widths removed because of too few occurrences): WP_A_Toshiba_3.bit, ff_hevc_put_hevc_uni_w 16 32 before: 2194 3872 after: 2119 3767 WP_B_Toshiba_3.bit, ff_hevc_put_hevc_bi_w 16 32 64 before: 2819 4960 9396 after: 2617 4788 9150 Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Diego Biurrun	512f3ffe9b	dsputil: Split off HuffYUV encoding bits into their own context Also shorten HuffYUV context member names to avoid clutter.	11 years ago
Diego Biurrun	0d439fbede	dsputil: Split off HuffYUV decoding bits into their own context Also shorten HuffYUV context member names to avoid clutter.	11 years ago
James Almer	5863207086	x86/dsputilenc: use HADDD in ff_sse16_sse2 Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	e64e079ece	x86/dsputilenc: implement SSE2 version of diff_pixels Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	a0c5cd3475	avcodec/x86/dsputilenc: set the count of SSE registers correctly for get_pixels Found-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	86ae0da60c	x86: hpeldsp: propagate changes across codecs Some codecs still use mmx versions, so have them use the versions with newer instruction sets. Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Michael Niedermayer	a3950a90f6	Revert "x86: dsputilenc: convert ff_sse{8, 16}_mmx() to yasm" This reverts commit `ad733089b0`. breaks with --disable-yasm revert requested by: Christophe Gisquet <christophe.gisquet@gmail.com>	11 years ago
Timothy Gu	ad733089b0	x86: dsputilenc: convert ff_sse{8, 16}_mmx() to yasm Signed-off-by: Timothy Gu <timothygu99@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	d94e255dd1	x86/dsputilenc: make the SUM_ABS_DCTELEM macro more readable Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	61eea421b2	x86/dsputilenc: port sum_abs_dctelem functions to yasm Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	81aa0f4604	x86: hpeldsp: implement SSSE3 version of _xy2 Loading pb_1 rather than pw_8192 was benchmarked to be more efficient. Loading of the 2 yields no advantage. Loading of one saves ~11 cycles. decicycles count: put8: 3223(mmx) -> 2387 avg8: 2863(mmxext) -> 2125 put16: 4356(sse2) -> 3553 avg16: 4481(sse2) -> 3513 Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	9722a6a3f3	x86: hpeldsp: implement SSE2 put_pixels16_xy2 This is obviously equivalent to the avg version, without the avg. 3223(mmx) -> 2006(sse2) Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
Christophe Gisquet	f0aca50e0b	x86: hpeldsp: implement SSE2 versions Those are mostly used in codecs older than H.264, eg MPEG-2. put16 versions: mmx mmx2 sse2 x2: 1888 1185 552 y2: 1778 1092 510 avg16 xy2: 3509(mmx2) -> 2169(sse2) Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	7538ad2248	x86/hevc_deblock: improve chroma functions register allocation Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	584327f22f	x86/dsputil: fix argument declaration in vector_clipf Should fix fate failures in msvc x86_64 Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	518cbf9b4a	x86/dsputil: fix VECTOR_CLIP_INT32 macro The inline loop was incrementing and using the value of %%i the wrong way. Disassembly of ff_vector_clip_int32_sse2 before and after this patch: movdqa (%rdx),%xmm0 \| movdqa (%rdx),%xmm0 movdqa 0x10(%rdx),%xmm1 \| movdqa 0x10(%rdx),%xmm1 movdqa 0x20(%rdx),%xmm2 \| movdqa 0x20(%rdx),%xmm2 movdqa 0x30(%rdx),%xmm3 \| movdqa 0x30(%rdx),%xmm3 [...] \| movdqa %xmm0,(%rcx) \| movdqa %xmm0,(%rcx) movdqa %xmm1,0x10(%rcx) \| movdqa %xmm1,0x10(%rcx) movdqa %xmm2,0x20(%rcx) \| movdqa %xmm2,0x20(%rcx) movdqa %xmm3,0x30(%rcx) \| movdqa %xmm3,0x30(%rcx) movdqa (%rdx),%xmm0 \| movdqa 0x40(%rdx),%xmm0 movdqa 0x20(%rdx),%xmm1 \| movdqa 0x50(%rdx),%xmm1 movdqa 0x40(%rdx),%xmm2 \| movdqa 0x60(%rdx),%xmm2 movdqa 0x60(%rdx),%xmm3 \| movdqa 0x70(%rdx),%xmm3 [...] \| movdqa %xmm0,(%rcx) \| movdqa %xmm0,0x40(%rcx) movdqa %xmm1,0x20(%rcx) \| movdqa %xmm1,0x50(%rcx) movdqa %xmm2,0x40(%rcx) \| movdqa %xmm2,0x60(%rcx) movdqa %xmm3,0x60(%rcx) \| movdqa %xmm3,0x70(%rcx) add $0x80,%rdx \| add $0x80,%rdx add $0x80,%rcx \| add $0x80,%rcx Other versions were unaffected. Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	6a4832caae	x86/diracdsp: mark all functions as yasm No inline asm dirac code remains in the tree, so replace every relevant check. This also moves all the dirac functions from dsputil_mmx.c to diracdsp_mmx.c Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago
James Almer	1d36defe94	x86/dsputil: port ff_vector_clipf_sse to yasm Signed-off-by: James Almer <jamrial@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>	11 years ago

... 2 3 4 5 6 ...

1851 Commits (4327088da355509b9f97e4bdb658456ee56de72b)