From f1e490b1adedd90ce35db894520ccc61cacb991d Mon Sep 17 00:00:00 2001 From: Aurelien Jacobs Date: Sun, 17 Dec 2017 20:07:33 +0100 Subject: [PATCH] sbcenc: add MMX optimizations This was originally based on libsbc, and was fully integrated into ffmpeg. Rough speed test: C version: speed= 592x MMX version: speed= 785x --- libavcodec/sbcdsp.c | 3 + libavcodec/sbcdsp.h | 2 + libavcodec/x86/Makefile | 2 + libavcodec/x86/sbcdsp.asm | 168 +++++++++++++++++++++++++++++++++++ libavcodec/x86/sbcdsp_init.c | 51 +++++++++++ 5 files changed, 226 insertions(+) create mode 100644 libavcodec/x86/sbcdsp.asm create mode 100644 libavcodec/x86/sbcdsp_init.c diff --git a/libavcodec/sbcdsp.c b/libavcodec/sbcdsp.c index e155387f0d..2d0addcf28 100644 --- a/libavcodec/sbcdsp.c +++ b/libavcodec/sbcdsp.c @@ -379,4 +379,7 @@ av_cold void ff_sbcdsp_init(SBCDSPContext *s) /* Default implementation for scale factors calculation */ s->sbc_calc_scalefactors = sbc_calc_scalefactors; s->sbc_calc_scalefactors_j = sbc_calc_scalefactors_j; + + if (ARCH_X86) + ff_sbcdsp_init_x86(s); } diff --git a/libavcodec/sbcdsp.h b/libavcodec/sbcdsp.h index 66ed7d324e..127e6a8a11 100644 --- a/libavcodec/sbcdsp.h +++ b/libavcodec/sbcdsp.h @@ -80,4 +80,6 @@ struct sbc_dsp_context { */ void ff_sbcdsp_init(SBCDSPContext *s); +void ff_sbcdsp_init_x86(SBCDSPContext *s); + #endif /* AVCODEC_SBCDSP_H */ diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index a805cd37b4..2350c8bbee 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -63,6 +63,7 @@ OBJS-$(CONFIG_PNG_DECODER) += x86/pngdsp_init.o OBJS-$(CONFIG_PRORES_DECODER) += x86/proresdsp_init.o OBJS-$(CONFIG_PRORES_LGPL_DECODER) += x86/proresdsp_init.o OBJS-$(CONFIG_RV40_DECODER) += x86/rv40dsp_init.o +OBJS-$(CONFIG_SBC_ENCODER) += x86/sbcdsp_init.o OBJS-$(CONFIG_SVQ1_ENCODER) += x86/svq1enc_init.o OBJS-$(CONFIG_TAK_DECODER) += x86/takdsp_init.o OBJS-$(CONFIG_TRUEHD_DECODER) += x86/mlpdsp_init.o @@ -172,6 +173,7 @@ X86ASM-OBJS-$(CONFIG_PNG_DECODER) += x86/pngdsp.o X86ASM-OBJS-$(CONFIG_PRORES_DECODER) += x86/proresdsp.o X86ASM-OBJS-$(CONFIG_PRORES_LGPL_DECODER) += x86/proresdsp.o X86ASM-OBJS-$(CONFIG_RV40_DECODER) += x86/rv40dsp.o +X86ASM-OBJS-$(CONFIG_SBC_ENCODER) += x86/sbcdsp.o X86ASM-OBJS-$(CONFIG_SVQ1_ENCODER) += x86/svq1enc.o X86ASM-OBJS-$(CONFIG_TAK_DECODER) += x86/takdsp.o X86ASM-OBJS-$(CONFIG_TRUEHD_DECODER) += x86/mlpdsp.o diff --git a/libavcodec/x86/sbcdsp.asm b/libavcodec/x86/sbcdsp.asm new file mode 100644 index 0000000000..d68d3a9ae8 --- /dev/null +++ b/libavcodec/x86/sbcdsp.asm @@ -0,0 +1,168 @@ +;****************************************************************************** +;* SIMD optimized SBC encoder DSP functions +;* +;* Copyright (C) 2017 Aurelien Jacobs +;* Copyright (C) 2008-2010 Nokia Corporation +;* Copyright (C) 2004-2010 Marcel Holtmann +;* Copyright (C) 2004-2005 Henryk Ploetz +;* Copyright (C) 2005-2006 Brad Midgley +;* +;* 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 + +scale_mask: times 2 dd 0x8000 ; 1 << (SBC_PROTO_FIXED_SCALE - 1) + +SECTION .text + +%macro NIDN 3 +%ifnidn %2, %3 + %1 %2, %3 +%endif +%endmacro + +%macro ANALYZE_MAC 9 ; out1, out2, in1, in2, tmp1, tmp2, add1, add2, offset + NIDN movq, %5, %3 + NIDN movq, %6, %4 + pmaddwd %5, [constsq+%9] + pmaddwd %6, [constsq+%9+8] + NIDN paddd, %1, %7 + NIDN paddd, %2, %8 +%endmacro + +%macro ANALYZE_MAC_IN 7 ; out1, out2, tmp1, tmp2, add1, add2, offset + ANALYZE_MAC %1, %2, [inq+%7], [inq+%7+8], %3, %4, %5, %6, %7 +%endmacro + +%macro ANALYZE_MAC_REG 7 ; out1, out2, in, tmp1, tmp2, offset, pack +%ifidn %7, pack + psrad %3, 16 ; SBC_PROTO_FIXED_SCALE + packssdw %3, %3 +%endif + ANALYZE_MAC %1, %2, %3, %3, %4, %5, %4, %5, %6 +%endmacro + +;******************************************************************* +;void ff_sbc_analyze_4(const int16_t *in, int32_t *out, const int16_t *consts); +;******************************************************************* +INIT_MMX mmx +cglobal sbc_analyze_4, 3, 3, 4, in, out, consts + ANALYZE_MAC_IN m0, m1, m0, m1, [scale_mask], [scale_mask], 0 + ANALYZE_MAC_IN m0, m1, m2, m3, m2, m3, 16 + ANALYZE_MAC_IN m0, m1, m2, m3, m2, m3, 32 + ANALYZE_MAC_IN m0, m1, m2, m3, m2, m3, 48 + ANALYZE_MAC_IN m0, m1, m2, m3, m2, m3, 64 + + ANALYZE_MAC_REG m0, m2, m0, m0, m2, 80, pack + ANALYZE_MAC_REG m0, m2, m1, m1, m3, 96, pack + + movq [outq ], m0 + movq [outq+8], m2 + + RET + + +;******************************************************************* +;void ff_sbc_analyze_8(const int16_t *in, int32_t *out, const int16_t *consts); +;******************************************************************* +INIT_MMX mmx +cglobal sbc_analyze_8, 3, 3, 4, in, out, consts + ANALYZE_MAC_IN m0, m1, m0, m1, [scale_mask], [scale_mask], 0 + ANALYZE_MAC_IN m2, m3, m2, m3, [scale_mask], [scale_mask], 16 + ANALYZE_MAC_IN m0, m1, m4, m5, m4, m5, 32 + ANALYZE_MAC_IN m2, m3, m6, m7, m6, m7, 48 + ANALYZE_MAC_IN m0, m1, m4, m5, m4, m5, 64 + ANALYZE_MAC_IN m2, m3, m6, m7, m6, m7, 80 + ANALYZE_MAC_IN m0, m1, m4, m5, m4, m5, 96 + ANALYZE_MAC_IN m2, m3, m6, m7, m6, m7, 112 + ANALYZE_MAC_IN m0, m1, m4, m5, m4, m5, 128 + ANALYZE_MAC_IN m2, m3, m6, m7, m6, m7, 144 + + ANALYZE_MAC_REG m4, m5, m0, m4, m5, 160, pack + ANALYZE_MAC_REG m4, m5, m1, m6, m7, 192, pack + ANALYZE_MAC_REG m4, m5, m2, m6, m7, 224, pack + ANALYZE_MAC_REG m4, m5, m3, m6, m7, 256, pack + + movq [outq ], m4 + movq [outq+8], m5 + + ANALYZE_MAC_REG m0, m5, m0, m0, m5, 176, no + ANALYZE_MAC_REG m0, m5, m1, m1, m7, 208, no + ANALYZE_MAC_REG m0, m5, m2, m2, m7, 240, no + ANALYZE_MAC_REG m0, m5, m3, m3, m7, 272, no + + movq [outq+16], m0 + movq [outq+24], m5 + + RET + + +;******************************************************************* +;void ff_sbc_calc_scalefactors(int32_t sb_sample_f[16][2][8], +; uint32_t scale_factor[2][8], +; int blocks, int channels, int subbands) +;******************************************************************* +INIT_MMX mmx +cglobal sbc_calc_scalefactors, 5, 7, 4, sb_sample_f, scale_factor, blocks, channels, subbands, ptr, blk + ; subbands = 4 * subbands * channels + movq m3, [scale_mask] + shl subbandsd, 2 + cmp channelsd, 2 + jl .loop_1 + shl subbandsd, 1 + +.loop_1: + sub subbandsq, 8 + lea ptrq, [sb_sample_fq + subbandsq] + + ; blk = (blocks - 1) * 64; + lea blkq, [blocksq - 1] + shl blkd, 6 + + movq m0, m3 +.loop_2: + movq m1, [ptrq+blkq] + pxor m2, m2 + pcmpgtd m1, m2 + paddd m1, [ptrq+blkq] + pcmpgtd m2, m1 + pxor m1, m2 + + por m0, m1 + + sub blkq, 64 + jns .loop_2 + + movd blkd, m0 + psrlq m0, 32 + bsr blkd, blkd + sub blkd, 15 ; SCALE_OUT_BITS + mov [scale_factorq + subbandsq], blkd + + movd blkd, m0 + bsr blkd, blkd + sub blkd, 15 ; SCALE_OUT_BITS + mov [scale_factorq + subbandsq + 4], blkd + + cmp subbandsq, 0 + jg .loop_1 + + emms + RET diff --git a/libavcodec/x86/sbcdsp_init.c b/libavcodec/x86/sbcdsp_init.c new file mode 100644 index 0000000000..86effecfdf --- /dev/null +++ b/libavcodec/x86/sbcdsp_init.c @@ -0,0 +1,51 @@ +/* + * Bluetooth low-complexity, subband codec (SBC) + * + * Copyright (C) 2017 Aurelien Jacobs + * Copyright (C) 2008-2010 Nokia Corporation + * Copyright (C) 2004-2010 Marcel Holtmann + * Copyright (C) 2004-2005 Henryk Ploetz + * Copyright (C) 2005-2006 Brad Midgley + * + * 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 + */ + +/** + * @file + * SBC MMX optimization for some basic "building bricks" + */ + +#include "libavutil/cpu.h" +#include "libavutil/x86/cpu.h" +#include "libavcodec/sbcdsp.h" + +void ff_sbc_analyze_4_mmx(const int16_t *in, int32_t *out, const int16_t *consts); +void ff_sbc_analyze_8_mmx(const int16_t *in, int32_t *out, const int16_t *consts); +void ff_sbc_calc_scalefactors_mmx(int32_t sb_sample_f[16][2][8], + uint32_t scale_factor[2][8], + int blocks, int channels, int subbands); + +av_cold void ff_sbcdsp_init_x86(SBCDSPContext *s) +{ + int cpu_flags = av_get_cpu_flags(); + + if (EXTERNAL_MMX(cpu_flags)) { + s->sbc_analyze_4 = ff_sbc_analyze_4_mmx; + s->sbc_analyze_8 = ff_sbc_analyze_8_mmx; + s->sbc_calc_scalefactors = ff_sbc_calc_scalefactors_mmx; + } +}