From 07b4b0ca62a1c8cd0b7b1116bffb4b7c0fe640c8 Mon Sep 17 00:00:00 2001 From: James Almer Date: Tue, 11 Feb 2014 22:28:54 -0300 Subject: [PATCH] tta/x86: add ff_ttafilter_process_dec_{ssse3, sse4} Results are from a Win64 build running on an AMD FX 6300 1121 decicycles in ttafilter_process_dec_c, 16777112 runs, 104 skips 522 decicycles in ff_ttafilter_process_dec_ssse3, 16777149 runs, 67 skips 477 decicycles in ff_ttafilter_process_dec_sse4, 16777156 runs, 60 skips Signed-off-by: James Almer Signed-off-by: Michael Niedermayer --- libavcodec/ttadsp.c | 3 + libavcodec/ttadsp.h | 1 + libavcodec/x86/Makefile | 2 + libavcodec/x86/ttadsp.asm | 119 +++++++++++++++++++++++++++++++++++ libavcodec/x86/ttadsp_init.c | 42 +++++++++++++ 5 files changed, 167 insertions(+) create mode 100644 libavcodec/x86/ttadsp.asm create mode 100644 libavcodec/x86/ttadsp_init.c diff --git a/libavcodec/ttadsp.c b/libavcodec/ttadsp.c index 691824b0d2..30b7ab9eb7 100644 --- a/libavcodec/ttadsp.c +++ b/libavcodec/ttadsp.c @@ -51,4 +51,7 @@ static void ttafilter_process_dec_c(int32_t *qm, int32_t *dx, int32_t *dl, av_cold void ff_ttadsp_init(TTADSPContext *c) { c->ttafilter_process_dec = ttafilter_process_dec_c; + + if (ARCH_X86) + ff_ttadsp_init_x86(c); } diff --git a/libavcodec/ttadsp.h b/libavcodec/ttadsp.h index 5e88e47816..56930f1c85 100644 --- a/libavcodec/ttadsp.h +++ b/libavcodec/ttadsp.h @@ -29,5 +29,6 @@ typedef struct TTADSPContext { } TTADSPContext; void ff_ttadsp_init(TTADSPContext *c); +void ff_ttadsp_init_x86(TTADSPContext *c); #endif /* AVCODEC_TTADSP_H */ diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index 0c9c13cf5a..0d3594f7c4 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -34,6 +34,7 @@ OBJS-$(CONFIG_RV30_DECODER) += x86/rv34dsp_init.o OBJS-$(CONFIG_RV40_DECODER) += x86/rv34dsp_init.o \ x86/rv40dsp_init.o OBJS-$(CONFIG_V210_DECODER) += x86/v210-init.o +OBJS-$(CONFIG_TTA_DECODER) += x86/ttadsp_init.o OBJS-$(CONFIG_TRUEHD_DECODER) += x86/mlpdsp.o OBJS-$(CONFIG_VC1_DECODER) += x86/vc1dsp_init.o OBJS-$(CONFIG_VIDEODSP) += x86/videodsp_init.o @@ -100,6 +101,7 @@ YASM-OBJS-$(CONFIG_PRORES_LGPL_DECODER) += x86/proresdsp.o YASM-OBJS-$(CONFIG_RV30_DECODER) += x86/rv34dsp.o YASM-OBJS-$(CONFIG_RV40_DECODER) += x86/rv34dsp.o \ x86/rv40dsp.o +YASM-OBJS-$(CONFIG_TTA_DECODER) += x86/ttadsp.o YASM-OBJS-$(CONFIG_V210_DECODER) += x86/v210.o YASM-OBJS-$(CONFIG_VC1_DECODER) += x86/vc1dsp.o YASM-OBJS-$(CONFIG_VIDEODSP) += x86/videodsp.o diff --git a/libavcodec/x86/ttadsp.asm b/libavcodec/x86/ttadsp.asm new file mode 100644 index 0000000000..8346cabac2 --- /dev/null +++ b/libavcodec/x86/ttadsp.asm @@ -0,0 +1,119 @@ +;****************************************************************************** +;* TTA DSP SIMD optimizations +;* +;* Copyright (C) 2014 James Almer +;* +;* This file is part of FFmpeg. +;* +;* FFmpeg is free software; you can redistribute it and/or +;* modify it under the terms of the GNU Lesser General Public +;* License as published by the Free Software Foundation; either +;* version 2.1 of the License, or (at your option) any later version. +;* +;* FFmpeg is distributed in the hope that it will be useful, +;* but WITHOUT ANY WARRANTY; without even the implied warranty of +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +;* Lesser General Public License for more details. +;* +;* You should have received a copy of the GNU Lesser General Public +;* License along with FFmpeg; if not, write to the Free Software +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA +;****************************************************************************** + +%include "libavutil/x86/x86util.asm" + +SECTION_RODATA + +pd_n0113: dd ~0, ~1, ~1, ~3 +pd_1224: dd 1, 2, 2, 4 + +SECTION .text + +%macro TTA_FILTER 2 +INIT_XMM %1 +cglobal ttafilter_process_dec, 5,5,%2, qm, dx, dl, error, in, shift, round + mova m2, [qmq ] + mova m3, [qmq + 0x10] + mova m4, [dxq ] + mova m5, [dxq + 0x10] + + movd m6, [errorq] ; if (filter->error < 0) { + SPLATD m6 ; for (int i = 0; i < 8; i++) + psignd m0, m4, m6 ; filter->qm[i] -= filter->dx[i]; + psignd m1, m5, m6 ; } else if (filter->error > 0) { + paddd m2, m0 ; for (int i = 0; i < 8; i++) + paddd m3, m1 ; filter->qm[i] += filter->dx[i]; + mova [qmq ], m2 ; } + mova [qmq + 0x10], m3 ; + + mova m0, [dlq ] + mova m1, [dlq + 0x10] + +%if cpuflag(sse4) + pmulld m2, m0 + pmulld m3, m1 +%else + pshufd m6, m0, 0xb1 + pshufd m7, m2, 0xb1 + pmuludq m6, m7 + pshufd m6, m6, 0xd8 + pmuludq m2, m0 + pshufd m2, m2, 0xd8 + punpckldq m2, m6 + + pshufd m6, m1, 0xb1 + pshufd m7, m3, 0xb1 + pmuludq m6, m7 + pshufd m6, m6, 0xd8 + pmuludq m3, m1 + pshufd m3, m3, 0xd8 + punpckldq m3, m6 +%endif + ; Using horizontal add (phaddd) seems to be slower than shuffling stuff around + paddd m2, m3 ; int sum = filter->round + + ; filter->dl[0] * filter->qm[0] + + punpckhqdq m3, m2, m2 ; filter->dl[1] * filter->qm[1] + + paddd m2, m3 ; filter->dl[2] * filter->qm[2] + + ; filter->dl[3] * filter->qm[3] + + movd m6, roundm ; filter->dl[4] * filter->qm[4] + + paddd m6, m2 ; filter->dl[5] * filter->qm[5] + + pshufd m2, m2, 0x1 ; filter->dl[6] * filter->qm[6] + + paddd m6, m2 ; filter->dl[7] * filter->qm[7]; + + palignr m5, m4, 4 ; filter->dx[0] = filter->dx[1]; filter->dx[1] = filter->dx[2]; + ; filter->dx[2] = filter->dx[3]; filter->dx[3] = filter->dx[4]; + + palignr m2, m1, m0, 4 ; filter->dl[0] = filter->dl[1]; filter->dl[1] = filter->dl[2]; + ; filter->dl[2] = filter->dl[3]; filter->dl[3] = filter->dl[4]; + + psrad m4, m1, 30 ; filter->dx[4] = ((filter->dl[4] >> 30) | 1); + por m4, [pd_1224 ] ; filter->dx[5] = ((filter->dl[5] >> 30) | 2) & ~1; + pand m4, [pd_n0113] ; filter->dx[6] = ((filter->dl[6] >> 30) | 2) & ~1; + ; filter->dx[7] = ((filter->dl[7] >> 30) | 4) & ~3; + + mova [dlq ], m2 + mova [dxq ], m5 + mova [dxq + 0x10], m4 + movd m0, [inq] ; filter->error = *in; + movd [errorq], m0 ; + + movd m2, shiftm ; *in += (sum >> filter->shift); + psrad m6, m2 ; + paddd m0, m6 ; + movd [inq], m0 ; + + psrldq m1, 4 ; + pslldq m0, 12 ; filter->dl[4] = -filter->dl[5]; + pshufd m0, m0, 0xf0 ; filter->dl[5] = -filter->dl[6]; + psubd m0, m1 ; filter->dl[6] = *in - filter->dl[7]; + psrldq m1, m0, 4 ; filter->dl[7] = *in; + pshufd m1, m1, 0xf4 ; filter->dl[5] += filter->dl[6]; + paddd m0, m1 ; filter->dl[4] += filter->dl[5]; + psrldq m1, 4 ; + paddd m0, m1 ; + mova [dlq + 0x10], m0 ; + RET +%endmacro + +TTA_FILTER ssse3, 8 +TTA_FILTER sse4, 7 diff --git a/libavcodec/x86/ttadsp_init.c b/libavcodec/x86/ttadsp_init.c new file mode 100644 index 0000000000..47dc87f6af --- /dev/null +++ b/libavcodec/x86/ttadsp_init.c @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2014 James Almer + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "libavcodec/ttadsp.h" +#include "libavutil/x86/cpu.h" +#include "config.h" + +void ff_ttafilter_process_dec_ssse3(int32_t *qm, int32_t *dx, int32_t *dl, + int32_t *error, int32_t *in, int32_t shift, + int32_t round); +void ff_ttafilter_process_dec_sse4(int32_t *qm, int32_t *dx, int32_t *dl, + int32_t *error, int32_t *in, int32_t shift, + int32_t round); + +av_cold void ff_ttadsp_init_x86(TTADSPContext *c) +{ +#if HAVE_YASM + int cpu_flags = av_get_cpu_flags(); + + if (EXTERNAL_SSSE3(cpu_flags)) + c->ttafilter_process_dec = ff_ttafilter_process_dec_ssse3; + if (EXTERNAL_SSE4(cpu_flags)) + c->ttafilter_process_dec = ff_ttafilter_process_dec_sse4; +#endif +}