From 7c51f5bd3903a9eaf80af64495174db8809ef31e Mon Sep 17 00:00:00 2001 From: Michael Niedermayer Date: Sun, 24 Aug 2014 17:09:45 +0200 Subject: [PATCH] swr: aarch64 audio_convert and neon clobber test Ported from avresample Code by: Mans Rullgard, Janne Grunau, Martin Storsjo Signed-off-by: Michael Niedermayer --- libswresample/aarch64/Makefile | 5 + libswresample/aarch64/audio_convert_init.c | 67 ++++ libswresample/aarch64/audio_convert_neon.S | 363 +++++++++++++++++++++ libswresample/aarch64/neontest.c | 29 ++ libswresample/audioconvert.c | 1 + libswresample/swresample_internal.h | 4 + 6 files changed, 469 insertions(+) create mode 100644 libswresample/aarch64/Makefile create mode 100644 libswresample/aarch64/audio_convert_init.c create mode 100644 libswresample/aarch64/audio_convert_neon.S create mode 100644 libswresample/aarch64/neontest.c diff --git a/libswresample/aarch64/Makefile b/libswresample/aarch64/Makefile new file mode 100644 index 0000000000..320ed67e82 --- /dev/null +++ b/libswresample/aarch64/Makefile @@ -0,0 +1,5 @@ +OBJS += aarch64/audio_convert_init.o + +OBJS-$(CONFIG_NEON_CLOBBER_TEST) += aarch64/neontest.o + +NEON-OBJS += aarch64/audio_convert_neon.o diff --git a/libswresample/aarch64/audio_convert_init.c b/libswresample/aarch64/audio_convert_init.c new file mode 100644 index 0000000000..60e24adb1c --- /dev/null +++ b/libswresample/aarch64/audio_convert_init.c @@ -0,0 +1,67 @@ +/* + * This file is part of libswresample. + * + * libswresample 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. + * + * libswresample 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 libswresample; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include + +#include "config.h" +#include "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/aarch64/cpu.h" +#include "libavutil/samplefmt.h" +#include "libswresample/swresample_internal.h" +#include "libswresample/audioconvert.h" + +void swri_oldapi_conv_flt_to_s16_neon(int16_t *dst, const float *src, int len); +void swri_oldapi_conv_fltp_to_s16_2ch_neon(int16_t *dst, float *const *src, int len, int channels); +void swri_oldapi_conv_fltp_to_s16_nch_neon(int16_t *dst, float *const *src, int len, int channels); + +static void conv_flt_to_s16_neon(uint8_t **dst, const uint8_t **src, int len){ + swri_oldapi_conv_flt_to_s16_neon((int16_t*)*dst, (const float*)*src, len); +} + +static void conv_fltp_to_s16_2ch_neon(uint8_t **dst, const uint8_t **src, int len){ + swri_oldapi_conv_fltp_to_s16_2ch_neon((int16_t*)*dst, (float *const*)src, len, 2); +} + +static void conv_fltp_to_s16_nch_neon(uint8_t **dst, const uint8_t **src, int len){ + int channels; + for(channels=3; channelssimd_f= NULL; + + if (have_neon(cpu_flags)) { + if(out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLT || out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_FLTP) + ac->simd_f = conv_flt_to_s16_neon; + if(out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLTP && channels == 2) + ac->simd_f = conv_fltp_to_s16_2ch_neon; + if(out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLTP && channels > 2) + ac->simd_f = conv_fltp_to_s16_nch_neon; + if(ac->simd_f) + ac->in_simd_align_mask = ac->out_simd_align_mask = 15; + } +} diff --git a/libswresample/aarch64/audio_convert_neon.S b/libswresample/aarch64/audio_convert_neon.S new file mode 100644 index 0000000000..74feff448a --- /dev/null +++ b/libswresample/aarch64/audio_convert_neon.S @@ -0,0 +1,363 @@ +/* + * Copyright (c) 2008 Mans Rullgard + * Copyright (c) 2014 Janne Grunau + * + * 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 "config.h" +#include "libavutil/aarch64/asm.S" + +function swri_oldapi_conv_flt_to_s16_neon, export=1 + subs x2, x2, #8 + ld1 {v0.4s}, [x1], #16 + fcvtzs v4.4s, v0.4s, #31 + ld1 {v1.4s}, [x1], #16 + fcvtzs v5.4s, v1.4s, #31 + b.eq 3f + ands x12, x2, #~15 + b.eq 2f +1: subs x12, x12, #16 + sqrshrn v4.4h, v4.4s, #16 + ld1 {v2.4s}, [x1], #16 + fcvtzs v6.4s, v2.4s, #31 + sqrshrn2 v4.8h, v5.4s, #16 + ld1 {v3.4s}, [x1], #16 + fcvtzs v7.4s, v3.4s, #31 + sqrshrn v6.4h, v6.4s, #16 + st1 {v4.8h}, [x0], #16 + sqrshrn2 v6.8h, v7.4s, #16 + ld1 {v0.4s}, [x1], #16 + fcvtzs v4.4s, v0.4s, #31 + ld1 {v1.4s}, [x1], #16 + fcvtzs v5.4s, v1.4s, #31 + st1 {v6.8h}, [x0], #16 + b.ne 1b + ands x2, x2, #15 + b.eq 3f +2: ld1 {v2.4s}, [x1], #16 + sqrshrn v4.4h, v4.4s, #16 + fcvtzs v6.4s, v2.4s, #31 + ld1 {v3.4s}, [x1], #16 + sqrshrn2 v4.8h, v5.4s, #16 + fcvtzs v7.4s, v3.4s, #31 + sqrshrn v6.4h, v6.4s, #16 + st1 {v4.8h}, [x0], #16 + sqrshrn2 v6.8h, v7.4s, #16 + st1 {v6.8h}, [x0] + ret +3: sqrshrn v4.4h, v4.4s, #16 + sqrshrn2 v4.8h, v5.4s, #16 + st1 {v4.8h}, [x0] + ret +endfunc + +function swri_oldapi_conv_fltp_to_s16_2ch_neon, export=1 + ldp x4, x5, [x1] + subs x2, x2, #8 + ld1 {v0.4s}, [x4], #16 + fcvtzs v4.4s, v0.4s, #31 + ld1 {v1.4s}, [x4], #16 + fcvtzs v5.4s, v1.4s, #31 + ld1 {v2.4s}, [x5], #16 + fcvtzs v6.4s, v2.4s, #31 + ld1 {v3.4s}, [x5], #16 + fcvtzs v7.4s, v3.4s, #31 + b.eq 3f + ands x12, x2, #~15 + b.eq 2f +1: subs x12, x12, #16 + ld1 {v16.4s}, [x4], #16 + fcvtzs v20.4s, v16.4s, #31 + sri v6.4s, v4.4s, #16 + ld1 {v17.4s}, [x4], #16 + fcvtzs v21.4s, v17.4s, #31 + ld1 {v18.4s}, [x5], #16 + fcvtzs v22.4s, v18.4s, #31 + ld1 {v19.4s}, [x5], #16 + sri v7.4s, v5.4s, #16 + st1 {v6.4s}, [x0], #16 + fcvtzs v23.4s, v19.4s, #31 + st1 {v7.4s}, [x0], #16 + sri v22.4s, v20.4s, #16 + ld1 {v0.4s}, [x4], #16 + sri v23.4s, v21.4s, #16 + st1 {v22.4s}, [x0], #16 + fcvtzs v4.4s, v0.4s, #31 + ld1 {v1.4s}, [x4], #16 + fcvtzs v5.4s, v1.4s, #31 + ld1 {v2.4s}, [x5], #16 + fcvtzs v6.4s, v2.4s, #31 + ld1 {v3.4s}, [x5], #16 + fcvtzs v7.4s, v3.4s, #31 + st1 {v23.4s}, [x0], #16 + b.ne 1b + ands x2, x2, #15 + b.eq 3f +2: sri v6.4s, v4.4s, #16 + ld1 {v0.4s}, [x4], #16 + fcvtzs v0.4s, v0.4s, #31 + ld1 {v1.4s}, [x4], #16 + fcvtzs v1.4s, v1.4s, #31 + ld1 {v2.4s}, [x5], #16 + fcvtzs v2.4s, v2.4s, #31 + sri v7.4s, v5.4s, #16 + ld1 {v3.4s}, [x5], #16 + fcvtzs v3.4s, v3.4s, #31 + sri v2.4s, v0.4s, #16 + st1 {v6.4s,v7.4s}, [x0], #32 + sri v3.4s, v1.4s, #16 + st1 {v2.4s,v3.4s}, [x0], #32 + ret +3: sri v6.4s, v4.4s, #16 + sri v7.4s, v5.4s, #16 + st1 {v6.4s,v7.4s}, [x0] + ret +endfunc + +function swri_oldapi_conv_fltp_to_s16_nch_neon, export=1 + cmp w3, #2 + b.eq X(swri_oldapi_conv_fltp_to_s16_2ch_neon) + b.gt 1f + ldr x1, [x1] + b X(swri_oldapi_conv_flt_to_s16_neon) +1: + cmp w3, #4 + lsl x12, x3, #1 + b.lt 4f + +5: // 4 channels + ldp x4, x5, [x1], #16 + ldp x6, x7, [x1], #16 + mov w9, w2 + mov x8, x0 + ld1 {v4.4s}, [x4], #16 + fcvtzs v4.4s, v4.4s, #31 + ld1 {v5.4s}, [x5], #16 + fcvtzs v5.4s, v5.4s, #31 + ld1 {v6.4s}, [x6], #16 + fcvtzs v6.4s, v6.4s, #31 + ld1 {v7.4s}, [x7], #16 + fcvtzs v7.4s, v7.4s, #31 +6: + subs w9, w9, #8 + ld1 {v0.4s}, [x4], #16 + fcvtzs v0.4s, v0.4s, #31 + sri v5.4s, v4.4s, #16 + ld1 {v1.4s}, [x5], #16 + fcvtzs v1.4s, v1.4s, #31 + sri v7.4s, v6.4s, #16 + ld1 {v2.4s}, [x6], #16 + fcvtzs v2.4s, v2.4s, #31 + zip1 v16.4s, v5.4s, v7.4s + ld1 {v3.4s}, [x7], #16 + fcvtzs v3.4s, v3.4s, #31 + zip2 v17.4s, v5.4s, v7.4s + st1 {v16.d}[0], [x8], x12 + sri v1.4s, v0.4s, #16 + st1 {v16.d}[1], [x8], x12 + sri v3.4s, v2.4s, #16 + st1 {v17.d}[0], [x8], x12 + zip1 v18.4s, v1.4s, v3.4s + st1 {v17.d}[1], [x8], x12 + zip2 v19.4s, v1.4s, v3.4s + b.eq 7f + ld1 {v4.4s}, [x4], #16 + fcvtzs v4.4s, v4.4s, #31 + st1 {v18.d}[0], [x8], x12 + ld1 {v5.4s}, [x5], #16 + fcvtzs v5.4s, v5.4s, #31 + st1 {v18.d}[1], [x8], x12 + ld1 {v6.4s}, [x6], #16 + fcvtzs v6.4s, v6.4s, #31 + st1 {v19.d}[0], [x8], x12 + ld1 {v7.4s}, [x7], #16 + fcvtzs v7.4s, v7.4s, #31 + st1 {v19.d}[1], [x8], x12 + b 6b +7: + st1 {v18.d}[0], [x8], x12 + st1 {v18.d}[1], [x8], x12 + st1 {v19.d}[0], [x8], x12 + st1 {v19.d}[1], [x8], x12 + subs w3, w3, #4 + b.eq end + cmp w3, #4 + add x0, x0, #8 + b.ge 5b + +4: // 2 channels + cmp w3, #2 + b.lt 4f + ldp x4, x5, [x1], #16 + mov w9, w2 + mov x8, x0 + tst w9, #8 + ld1 {v4.4s}, [x4], #16 + fcvtzs v4.4s, v4.4s, #31 + ld1 {v5.4s}, [x5], #16 + fcvtzs v5.4s, v5.4s, #31 + ld1 {v6.4s}, [x4], #16 + fcvtzs v6.4s, v6.4s, #31 + ld1 {v7.4s}, [x5], #16 + fcvtzs v7.4s, v7.4s, #31 + b.eq 6f + subs w9, w9, #8 + b.eq 7f + sri v5.4s, v4.4s, #16 + ld1 {v4.4s}, [x4], #16 + fcvtzs v4.4s, v4.4s, #31 + st1 {v5.s}[0], [x8], x12 + sri v7.4s, v6.4s, #16 + st1 {v5.s}[1], [x8], x12 + ld1 {v6.4s}, [x4], #16 + fcvtzs v6.4s, v6.4s, #31 + st1 {v5.s}[2], [x8], x12 + st1 {v5.s}[3], [x8], x12 + st1 {v7.s}[0], [x8], x12 + st1 {v7.s}[1], [x8], x12 + ld1 {v5.4s}, [x5], #16 + fcvtzs v5.4s, v5.4s, #31 + st1 {v7.s}[2], [x8], x12 + st1 {v7.s}[3], [x8], x12 + ld1 {v7.4s}, [x5], #16 + fcvtzs v7.4s, v7.4s, #31 +6: + subs w9, w9, #16 + ld1 {v0.4s}, [x4], #16 + sri v5.4s, v4.4s, #16 + fcvtzs v0.4s, v0.4s, #31 + ld1 {v1.4s}, [x5], #16 + sri v7.4s, v6.4s, #16 + st1 {v5.s}[0], [x8], x12 + st1 {v5.s}[1], [x8], x12 + fcvtzs v1.4s, v1.4s, #31 + st1 {v5.s}[2], [x8], x12 + st1 {v5.s}[3], [x8], x12 + ld1 {v2.4s}, [x4], #16 + st1 {v7.s}[0], [x8], x12 + fcvtzs v2.4s, v2.4s, #31 + st1 {v7.s}[1], [x8], x12 + ld1 {v3.4s}, [x5], #16 + st1 {v7.s}[2], [x8], x12 + fcvtzs v3.4s, v3.4s, #31 + st1 {v7.s}[3], [x8], x12 + sri v1.4s, v0.4s, #16 + sri v3.4s, v2.4s, #16 + b.eq 6f + ld1 {v4.4s}, [x4], #16 + st1 {v1.s}[0], [x8], x12 + fcvtzs v4.4s, v4.4s, #31 + st1 {v1.s}[1], [x8], x12 + ld1 {v5.4s}, [x5], #16 + st1 {v1.s}[2], [x8], x12 + fcvtzs v5.4s, v5.4s, #31 + st1 {v1.s}[3], [x8], x12 + ld1 {v6.4s}, [x4], #16 + st1 {v3.s}[0], [x8], x12 + fcvtzs v6.4s, v6.4s, #31 + st1 {v3.s}[1], [x8], x12 + ld1 {v7.4s}, [x5], #16 + st1 {v3.s}[2], [x8], x12 + fcvtzs v7.4s, v7.4s, #31 + st1 {v3.s}[3], [x8], x12 + b.gt 6b +6: + st1 {v1.s}[0], [x8], x12 + st1 {v1.s}[1], [x8], x12 + st1 {v1.s}[2], [x8], x12 + st1 {v1.s}[3], [x8], x12 + st1 {v3.s}[0], [x8], x12 + st1 {v3.s}[1], [x8], x12 + st1 {v3.s}[2], [x8], x12 + st1 {v3.s}[3], [x8], x12 + b 8f +7: + sri v5.4s, v4.4s, #16 + sri v7.4s, v6.4s, #16 + st1 {v5.s}[0], [x8], x12 + st1 {v5.s}[1], [x8], x12 + st1 {v5.s}[2], [x8], x12 + st1 {v5.s}[3], [x8], x12 + st1 {v7.s}[0], [x8], x12 + st1 {v7.s}[1], [x8], x12 + st1 {v7.s}[2], [x8], x12 + st1 {v7.s}[3], [x8], x12 +8: + subs w3, w3, #2 + add x0, x0, #4 + b.eq end + +4: // 1 channel + ldr x4, [x1] + tst w2, #8 + mov w9, w2 + mov x5, x0 + ld1 {v0.4s}, [x4], #16 + fcvtzs v0.4s, v0.4s, #31 + ld1 {v1.4s}, [x4], #16 + fcvtzs v1.4s, v1.4s, #31 + b.ne 8f +6: + subs w9, w9, #16 + ld1 {v2.4s}, [x4], #16 + fcvtzs v2.4s, v2.4s, #31 + ld1 {v3.4s}, [x4], #16 + fcvtzs v3.4s, v3.4s, #31 + st1 {v0.h}[1], [x5], x12 + st1 {v0.h}[3], [x5], x12 + st1 {v0.h}[5], [x5], x12 + st1 {v0.h}[7], [x5], x12 + st1 {v1.h}[1], [x5], x12 + st1 {v1.h}[3], [x5], x12 + st1 {v1.h}[5], [x5], x12 + st1 {v1.h}[7], [x5], x12 + b.eq 7f + ld1 {v0.4s}, [x4], #16 + fcvtzs v0.4s, v0.4s, #31 + ld1 {v1.4s}, [x4], #16 + fcvtzs v1.4s, v1.4s, #31 +7: + st1 {v2.h}[1], [x5], x12 + st1 {v2.h}[3], [x5], x12 + st1 {v2.h}[5], [x5], x12 + st1 {v2.h}[7], [x5], x12 + st1 {v3.h}[1], [x5], x12 + st1 {v3.h}[3], [x5], x12 + st1 {v3.h}[5], [x5], x12 + st1 {v3.h}[7], [x5], x12 + b.gt 6b + ret +8: + subs w9, w9, #8 + st1 {v0.h}[1], [x5], x12 + st1 {v0.h}[3], [x5], x12 + st1 {v0.h}[5], [x5], x12 + st1 {v0.h}[7], [x5], x12 + st1 {v1.h}[1], [x5], x12 + st1 {v1.h}[3], [x5], x12 + st1 {v1.h}[5], [x5], x12 + st1 {v1.h}[7], [x5], x12 + b.eq end + ld1 {v0.4s}, [x4], #16 + fcvtzs v0.4s, v0.4s, #31 + ld1 {v1.4s}, [x4], #16 + fcvtzs v1.4s, v1.4s, #31 + b 6b +end: + ret +endfunc diff --git a/libswresample/aarch64/neontest.c b/libswresample/aarch64/neontest.c new file mode 100644 index 0000000000..85c71bf4c9 --- /dev/null +++ b/libswresample/aarch64/neontest.c @@ -0,0 +1,29 @@ +/* + * check NEON registers for clobbers + * Copyright (c) 2013 Martin Storsjo + * + * 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 "libswresample/swresample.h" +#include "libavutil/aarch64/neontest.h" + +wrap(swr_convert(struct SwrContext *s, uint8_t **out, int out_count, + const uint8_t **in , int in_count)) +{ + testneonclobbers(swr_convert, s, out, out_count, in, in_count); +} diff --git a/libswresample/audioconvert.c b/libswresample/audioconvert.c index 325bdf4de7..efdc9b5a89 100644 --- a/libswresample/audioconvert.c +++ b/libswresample/audioconvert.c @@ -156,6 +156,7 @@ AudioConvert *swri_audio_convert_alloc(enum AVSampleFormat out_fmt, if(HAVE_YASM && HAVE_MMX) swri_audio_convert_init_x86(ctx, out_fmt, in_fmt, channels); if(ARCH_ARM) swri_audio_convert_init_arm(ctx, out_fmt, in_fmt, channels); + if(ARCH_AARCH64) swri_audio_convert_init_aarch64(ctx, out_fmt, in_fmt, channels); return ctx; } diff --git a/libswresample/swresample_internal.h b/libswresample/swresample_internal.h index 792768afbf..3761843e3b 100644 --- a/libswresample/swresample_internal.h +++ b/libswresample/swresample_internal.h @@ -188,6 +188,10 @@ void swri_rematrix_init_x86(struct SwrContext *s); void swri_get_dither(SwrContext *s, void *dst, int len, unsigned seed, enum AVSampleFormat noise_fmt); int swri_dither_init(SwrContext *s, enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt); +void swri_audio_convert_init_aarch64(struct AudioConvert *ac, + enum AVSampleFormat out_fmt, + enum AVSampleFormat in_fmt, + int channels); void swri_audio_convert_init_arm(struct AudioConvert *ac, enum AVSampleFormat out_fmt, enum AVSampleFormat in_fmt,