diff options
Diffstat (limited to 'libswresample/arm')
-rw-r--r-- | libswresample/arm/Makefile | 8 | ||||
-rw-r--r-- | libswresample/arm/audio_convert_init.c | 67 | ||||
-rw-r--r-- | libswresample/arm/audio_convert_neon.S | 363 | ||||
-rw-r--r-- | libswresample/arm/neontest.c | 29 | ||||
-rw-r--r-- | libswresample/arm/resample.S | 77 | ||||
-rw-r--r-- | libswresample/arm/resample_init.c | 122 |
6 files changed, 666 insertions, 0 deletions
diff --git a/libswresample/arm/Makefile b/libswresample/arm/Makefile new file mode 100644 index 0000000000..53ab4626f4 --- /dev/null +++ b/libswresample/arm/Makefile @@ -0,0 +1,8 @@ +OBJS += arm/audio_convert_init.o \ + arm/resample_init.o + + +OBJS-$(CONFIG_NEON_CLOBBER_TEST) += arm/neontest.o + +NEON-OBJS += arm/audio_convert_neon.o \ + arm/resample.o diff --git a/libswresample/arm/audio_convert_init.c b/libswresample/arm/audio_convert_init.c new file mode 100644 index 0000000000..ec9e62ede7 --- /dev/null +++ b/libswresample/arm/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 <stdint.h> + +#include "config.h" +#include "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/arm/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; channels<SWR_CH_MAX && src[channels]; channels++) + ; + swri_oldapi_conv_fltp_to_s16_nch_neon((int16_t*)*dst, (float *const*)src, len, channels); +} + +av_cold void swri_audio_convert_init_arm(struct AudioConvert *ac, + enum AVSampleFormat out_fmt, + enum AVSampleFormat in_fmt, + int channels) +{ + int cpu_flags = av_get_cpu_flags(); + + ac->simd_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/arm/audio_convert_neon.S b/libswresample/arm/audio_convert_neon.S new file mode 100644 index 0000000000..1f88316dde --- /dev/null +++ b/libswresample/arm/audio_convert_neon.S @@ -0,0 +1,363 @@ +/* + * Copyright (c) 2008 Mans Rullgard <mans@mansr.com> + * + * 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 "config.h" +#include "libavutil/arm/asm.S" + +function swri_oldapi_conv_flt_to_s16_neon, export=1 + subs r2, r2, #8 + vld1.32 {q0}, [r1,:128]! + vcvt.s32.f32 q8, q0, #31 + vld1.32 {q1}, [r1,:128]! + vcvt.s32.f32 q9, q1, #31 + beq 3f + bics r12, r2, #15 + beq 2f +1: subs r12, r12, #16 + vqrshrn.s32 d4, q8, #16 + vld1.32 {q0}, [r1,:128]! + vcvt.s32.f32 q0, q0, #31 + vqrshrn.s32 d5, q9, #16 + vld1.32 {q1}, [r1,:128]! + vcvt.s32.f32 q1, q1, #31 + vqrshrn.s32 d6, q0, #16 + vst1.16 {q2}, [r0,:128]! + vqrshrn.s32 d7, q1, #16 + vld1.32 {q8}, [r1,:128]! + vcvt.s32.f32 q8, q8, #31 + vld1.32 {q9}, [r1,:128]! + vcvt.s32.f32 q9, q9, #31 + vst1.16 {q3}, [r0,:128]! + bne 1b + ands r2, r2, #15 + beq 3f +2: vld1.32 {q0}, [r1,:128]! + vqrshrn.s32 d4, q8, #16 + vcvt.s32.f32 q0, q0, #31 + vld1.32 {q1}, [r1,:128]! + vqrshrn.s32 d5, q9, #16 + vcvt.s32.f32 q1, q1, #31 + vqrshrn.s32 d6, q0, #16 + vst1.16 {q2}, [r0,:128]! + vqrshrn.s32 d7, q1, #16 + vst1.16 {q3}, [r0,:128]! + bx lr +3: vqrshrn.s32 d4, q8, #16 + vqrshrn.s32 d5, q9, #16 + vst1.16 {q2}, [r0,:128]! + bx lr +endfunc + +function swri_oldapi_conv_fltp_to_s16_2ch_neon, export=1 + ldm r1, {r1, r3} + subs r2, r2, #8 + vld1.32 {q0}, [r1,:128]! + vcvt.s32.f32 q8, q0, #31 + vld1.32 {q1}, [r1,:128]! + vcvt.s32.f32 q9, q1, #31 + vld1.32 {q10}, [r3,:128]! + vcvt.s32.f32 q10, q10, #31 + vld1.32 {q11}, [r3,:128]! + vcvt.s32.f32 q11, q11, #31 + beq 3f + bics r12, r2, #15 + beq 2f +1: subs r12, r12, #16 + vld1.32 {q0}, [r1,:128]! + vcvt.s32.f32 q0, q0, #31 + vsri.32 q10, q8, #16 + vld1.32 {q1}, [r1,:128]! + vcvt.s32.f32 q1, q1, #31 + vld1.32 {q12}, [r3,:128]! + vcvt.s32.f32 q12, q12, #31 + vld1.32 {q13}, [r3,:128]! + vsri.32 q11, q9, #16 + vst1.16 {q10}, [r0,:128]! + vcvt.s32.f32 q13, q13, #31 + vst1.16 {q11}, [r0,:128]! + vsri.32 q12, q0, #16 + vld1.32 {q8}, [r1,:128]! + vsri.32 q13, q1, #16 + vst1.16 {q12}, [r0,:128]! + vcvt.s32.f32 q8, q8, #31 + vld1.32 {q9}, [r1,:128]! + vcvt.s32.f32 q9, q9, #31 + vld1.32 {q10}, [r3,:128]! + vcvt.s32.f32 q10, q10, #31 + vld1.32 {q11}, [r3,:128]! + vcvt.s32.f32 q11, q11, #31 + vst1.16 {q13}, [r0,:128]! + bne 1b + ands r2, r2, #15 + beq 3f +2: vsri.32 q10, q8, #16 + vld1.32 {q0}, [r1,:128]! + vcvt.s32.f32 q0, q0, #31 + vld1.32 {q1}, [r1,:128]! + vcvt.s32.f32 q1, q1, #31 + vld1.32 {q12}, [r3,:128]! + vcvt.s32.f32 q12, q12, #31 + vsri.32 q11, q9, #16 + vld1.32 {q13}, [r3,:128]! + vcvt.s32.f32 q13, q13, #31 + vst1.16 {q10}, [r0,:128]! + vsri.32 q12, q0, #16 + vst1.16 {q11}, [r0,:128]! + vsri.32 q13, q1, #16 + vst1.16 {q12-q13},[r0,:128]! + bx lr +3: vsri.32 q10, q8, #16 + vsri.32 q11, q9, #16 + vst1.16 {q10-q11},[r0,:128]! + bx lr +endfunc + +function swri_oldapi_conv_fltp_to_s16_nch_neon, export=1 + cmp r3, #2 + itt lt + ldrlt r1, [r1] + blt X(swri_oldapi_conv_flt_to_s16_neon) + beq X(swri_oldapi_conv_fltp_to_s16_2ch_neon) + + push {r4-r8, lr} + cmp r3, #4 + lsl r12, r3, #1 + blt 4f + + @ 4 channels +5: ldm r1!, {r4-r7} + mov lr, r2 + mov r8, r0 + vld1.32 {q8}, [r4,:128]! + vcvt.s32.f32 q8, q8, #31 + vld1.32 {q9}, [r5,:128]! + vcvt.s32.f32 q9, q9, #31 + vld1.32 {q10}, [r6,:128]! + vcvt.s32.f32 q10, q10, #31 + vld1.32 {q11}, [r7,:128]! + vcvt.s32.f32 q11, q11, #31 +6: subs lr, lr, #8 + vld1.32 {q0}, [r4,:128]! + vcvt.s32.f32 q0, q0, #31 + vsri.32 q9, q8, #16 + vld1.32 {q1}, [r5,:128]! + vcvt.s32.f32 q1, q1, #31 + vsri.32 q11, q10, #16 + vld1.32 {q2}, [r6,:128]! + vcvt.s32.f32 q2, q2, #31 + vzip.32 d18, d22 + vld1.32 {q3}, [r7,:128]! + vcvt.s32.f32 q3, q3, #31 + vzip.32 d19, d23 + vst1.16 {d18}, [r8], r12 + vsri.32 q1, q0, #16 + vst1.16 {d22}, [r8], r12 + vsri.32 q3, q2, #16 + vst1.16 {d19}, [r8], r12 + vzip.32 d2, d6 + vst1.16 {d23}, [r8], r12 + vzip.32 d3, d7 + beq 7f + vld1.32 {q8}, [r4,:128]! + vcvt.s32.f32 q8, q8, #31 + vst1.16 {d2}, [r8], r12 + vld1.32 {q9}, [r5,:128]! + vcvt.s32.f32 q9, q9, #31 + vst1.16 {d6}, [r8], r12 + vld1.32 {q10}, [r6,:128]! + vcvt.s32.f32 q10, q10, #31 + vst1.16 {d3}, [r8], r12 + vld1.32 {q11}, [r7,:128]! + vcvt.s32.f32 q11, q11, #31 + vst1.16 {d7}, [r8], r12 + b 6b +7: vst1.16 {d2}, [r8], r12 + vst1.16 {d6}, [r8], r12 + vst1.16 {d3}, [r8], r12 + vst1.16 {d7}, [r8], r12 + subs r3, r3, #4 + it eq + popeq {r4-r8, pc} + cmp r3, #4 + add r0, r0, #8 + bge 5b + + @ 2 channels +4: cmp r3, #2 + blt 4f + ldm r1!, {r4-r5} + mov lr, r2 + mov r8, r0 + tst lr, #8 + vld1.32 {q8}, [r4,:128]! + vcvt.s32.f32 q8, q8, #31 + vld1.32 {q9}, [r5,:128]! + vcvt.s32.f32 q9, q9, #31 + vld1.32 {q10}, [r4,:128]! + vcvt.s32.f32 q10, q10, #31 + vld1.32 {q11}, [r5,:128]! + vcvt.s32.f32 q11, q11, #31 + beq 6f + subs lr, lr, #8 + beq 7f + vsri.32 d18, d16, #16 + vsri.32 d19, d17, #16 + vld1.32 {q8}, [r4,:128]! + vcvt.s32.f32 q8, q8, #31 + vst1.32 {d18[0]}, [r8], r12 + vsri.32 d22, d20, #16 + vst1.32 {d18[1]}, [r8], r12 + vsri.32 d23, d21, #16 + vst1.32 {d19[0]}, [r8], r12 + vst1.32 {d19[1]}, [r8], r12 + vld1.32 {q9}, [r5,:128]! + vcvt.s32.f32 q9, q9, #31 + vst1.32 {d22[0]}, [r8], r12 + vst1.32 {d22[1]}, [r8], r12 + vld1.32 {q10}, [r4,:128]! + vcvt.s32.f32 q10, q10, #31 + vst1.32 {d23[0]}, [r8], r12 + vst1.32 {d23[1]}, [r8], r12 + vld1.32 {q11}, [r5,:128]! + vcvt.s32.f32 q11, q11, #31 +6: subs lr, lr, #16 + vld1.32 {q0}, [r4,:128]! + vcvt.s32.f32 q0, q0, #31 + vsri.32 d18, d16, #16 + vld1.32 {q1}, [r5,:128]! + vcvt.s32.f32 q1, q1, #31 + vsri.32 d19, d17, #16 + vld1.32 {q2}, [r4,:128]! + vcvt.s32.f32 q2, q2, #31 + vld1.32 {q3}, [r5,:128]! + vcvt.s32.f32 q3, q3, #31 + vst1.32 {d18[0]}, [r8], r12 + vsri.32 d22, d20, #16 + vst1.32 {d18[1]}, [r8], r12 + vsri.32 d23, d21, #16 + vst1.32 {d19[0]}, [r8], r12 + vsri.32 d2, d0, #16 + vst1.32 {d19[1]}, [r8], r12 + vsri.32 d3, d1, #16 + vst1.32 {d22[0]}, [r8], r12 + vsri.32 d6, d4, #16 + vst1.32 {d22[1]}, [r8], r12 + vsri.32 d7, d5, #16 + vst1.32 {d23[0]}, [r8], r12 + vst1.32 {d23[1]}, [r8], r12 + beq 6f + vld1.32 {q8}, [r4,:128]! + vcvt.s32.f32 q8, q8, #31 + vst1.32 {d2[0]}, [r8], r12 + vst1.32 {d2[1]}, [r8], r12 + vld1.32 {q9}, [r5,:128]! + vcvt.s32.f32 q9, q9, #31 + vst1.32 {d3[0]}, [r8], r12 + vst1.32 {d3[1]}, [r8], r12 + vld1.32 {q10}, [r4,:128]! + vcvt.s32.f32 q10, q10, #31 + vst1.32 {d6[0]}, [r8], r12 + vst1.32 {d6[1]}, [r8], r12 + vld1.32 {q11}, [r5,:128]! + vcvt.s32.f32 q11, q11, #31 + vst1.32 {d7[0]}, [r8], r12 + vst1.32 {d7[1]}, [r8], r12 + bgt 6b +6: vst1.32 {d2[0]}, [r8], r12 + vst1.32 {d2[1]}, [r8], r12 + vst1.32 {d3[0]}, [r8], r12 + vst1.32 {d3[1]}, [r8], r12 + vst1.32 {d6[0]}, [r8], r12 + vst1.32 {d6[1]}, [r8], r12 + vst1.32 {d7[0]}, [r8], r12 + vst1.32 {d7[1]}, [r8], r12 + b 8f +7: vsri.32 d18, d16, #16 + vsri.32 d19, d17, #16 + vst1.32 {d18[0]}, [r8], r12 + vsri.32 d22, d20, #16 + vst1.32 {d18[1]}, [r8], r12 + vsri.32 d23, d21, #16 + vst1.32 {d19[0]}, [r8], r12 + vst1.32 {d19[1]}, [r8], r12 + vst1.32 {d22[0]}, [r8], r12 + vst1.32 {d22[1]}, [r8], r12 + vst1.32 {d23[0]}, [r8], r12 + vst1.32 {d23[1]}, [r8], r12 +8: subs r3, r3, #2 + add r0, r0, #4 + it eq + popeq {r4-r8, pc} + + @ 1 channel +4: ldr r4, [r1] + tst r2, #8 + mov lr, r2 + mov r5, r0 + vld1.32 {q0}, [r4,:128]! + vcvt.s32.f32 q0, q0, #31 + vld1.32 {q1}, [r4,:128]! + vcvt.s32.f32 q1, q1, #31 + bne 8f +6: subs lr, lr, #16 + vld1.32 {q2}, [r4,:128]! + vcvt.s32.f32 q2, q2, #31 + vld1.32 {q3}, [r4,:128]! + vcvt.s32.f32 q3, q3, #31 + vst1.16 {d0[1]}, [r5,:16], r12 + vst1.16 {d0[3]}, [r5,:16], r12 + vst1.16 {d1[1]}, [r5,:16], r12 + vst1.16 {d1[3]}, [r5,:16], r12 + vst1.16 {d2[1]}, [r5,:16], r12 + vst1.16 {d2[3]}, [r5,:16], r12 + vst1.16 {d3[1]}, [r5,:16], r12 + vst1.16 {d3[3]}, [r5,:16], r12 + beq 7f + vld1.32 {q0}, [r4,:128]! + vcvt.s32.f32 q0, q0, #31 + vld1.32 {q1}, [r4,:128]! + vcvt.s32.f32 q1, q1, #31 +7: vst1.16 {d4[1]}, [r5,:16], r12 + vst1.16 {d4[3]}, [r5,:16], r12 + vst1.16 {d5[1]}, [r5,:16], r12 + vst1.16 {d5[3]}, [r5,:16], r12 + vst1.16 {d6[1]}, [r5,:16], r12 + vst1.16 {d6[3]}, [r5,:16], r12 + vst1.16 {d7[1]}, [r5,:16], r12 + vst1.16 {d7[3]}, [r5,:16], r12 + bgt 6b + pop {r4-r8, pc} +8: subs lr, lr, #8 + vst1.16 {d0[1]}, [r5,:16], r12 + vst1.16 {d0[3]}, [r5,:16], r12 + vst1.16 {d1[1]}, [r5,:16], r12 + vst1.16 {d1[3]}, [r5,:16], r12 + vst1.16 {d2[1]}, [r5,:16], r12 + vst1.16 {d2[3]}, [r5,:16], r12 + vst1.16 {d3[1]}, [r5,:16], r12 + vst1.16 {d3[3]}, [r5,:16], r12 + it eq + popeq {r4-r8, pc} + vld1.32 {q0}, [r4,:128]! + vcvt.s32.f32 q0, q0, #31 + vld1.32 {q1}, [r4,:128]! + vcvt.s32.f32 q1, q1, #31 + b 6b +endfunc diff --git a/libswresample/arm/neontest.c b/libswresample/arm/neontest.c new file mode 100644 index 0000000000..2abbbc2367 --- /dev/null +++ b/libswresample/arm/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/arm/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/arm/resample.S b/libswresample/arm/resample.S new file mode 100644 index 0000000000..c231301b2b --- /dev/null +++ b/libswresample/arm/resample.S @@ -0,0 +1,77 @@ +/* + * Copyright (c) 2016 Matthieu Bouron <matthieu.bouron stupeflix.com> + * + * 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/arm/asm.S" + +function ff_resample_common_apply_filter_x4_float_neon, export=1 + vmov.f32 q0, #0.0 @ accumulator +1: vld1.32 {q1}, [r1]! @ src + vld1.32 {q2}, [r2]! @ filter + vmla.f32 q0, q1, q2 @ accumulator += src + {0..3} * filter + {0..3} + subs r3, #4 @ filter_length -= 4 + bgt 1b @ loop until filter_length + vpadd.f32 d0, d0, d1 @ pair adding of the 4x32-bit accumulated values + vpadd.f32 d0, d0, d0 @ pair adding of the 4x32-bit accumulator values + vst1.32 {d0[0]}, [r0] @ write accumulator + mov pc, lr +endfunc + +function ff_resample_common_apply_filter_x8_float_neon, export=1 + vmov.f32 q0, #0.0 @ accumulator +1: vld1.32 {q1}, [r1]! @ src + vld1.32 {q2}, [r2]! @ filter + vld1.32 {q8}, [r1]! @ src + vld1.32 {q9}, [r2]! @ filter + vmla.f32 q0, q1, q2 @ accumulator += src + {0..3} * filter + {0..3} + vmla.f32 q0, q8, q9 @ accumulator += src + {4..7} * filter + {4..7} + subs r3, #8 @ filter_length -= 8 + bgt 1b @ loop until filter_length + vpadd.f32 d0, d0, d1 @ pair adding of the 4x32-bit accumulated values + vpadd.f32 d0, d0, d0 @ pair adding of the 4x32-bit accumulator values + vst1.32 {d0[0]}, [r0] @ write accumulator + mov pc, lr +endfunc + +function ff_resample_common_apply_filter_x4_s16_neon, export=1 + vmov.s32 q0, #0 @ accumulator +1: vld1.16 {d2}, [r1]! @ src + vld1.16 {d4}, [r2]! @ filter + vmlal.s16 q0, d2, d4 @ accumulator += src + {0..3} * filter + {0..3} + subs r3, #4 @ filter_length -= 4 + bgt 1b @ loop until filter_length + vpadd.s32 d0, d0, d1 @ pair adding of the 4x32-bit accumulated values + vpadd.s32 d0, d0, d0 @ pair adding of the 4x32-bit accumulator values + vst1.32 {d0[0]}, [r0] @ write accumulator + mov pc, lr +endfunc + +function ff_resample_common_apply_filter_x8_s16_neon, export=1 + vmov.s32 q0, #0 @ accumulator +1: vld1.16 {q1}, [r1]! @ src + vld1.16 {q2}, [r2]! @ filter + vmlal.s16 q0, d2, d4 @ accumulator += src + {0..3} * filter + {0..3} + vmlal.s16 q0, d3, d5 @ accumulator += src + {4..7} * filter + {4..7} + subs r3, #8 @ filter_length -= 8 + bgt 1b @ loop until filter_length + vpadd.s32 d0, d0, d1 @ pair adding of the 4x32-bit accumulated values + vpadd.s32 d0, d0, d0 @ pair adding of the 4x32-bit accumulator values + vst1.32 {d0[0]}, [r0] @ write accumulator + mov pc, lr +endfunc diff --git a/libswresample/arm/resample_init.c b/libswresample/arm/resample_init.c new file mode 100644 index 0000000000..003fafd29b --- /dev/null +++ b/libswresample/arm/resample_init.c @@ -0,0 +1,122 @@ +/* + * Audio resampling + * + * Copyright (c) 2004-2012 Michael Niedermayer <michaelni@gmx.at> + * + * 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/cpu.h" +#include "libavutil/avassert.h" + +#include "libavutil/arm/cpu.h" +#include "libswresample/resample.h" + +#define DECLARE_RESAMPLE_COMMON_TEMPLATE(TYPE, DELEM, FELEM, FELEM2, OUT) \ + \ +void ff_resample_common_apply_filter_x4_##TYPE##_neon(FELEM2 *acc, const DELEM *src, \ + const FELEM *filter, int length); \ + \ +void ff_resample_common_apply_filter_x8_##TYPE##_neon(FELEM2 *acc, const DELEM *src, \ + const FELEM *filter, int length); \ + \ +static int ff_resample_common_##TYPE##_neon(ResampleContext *c, void *dest, const void *source, \ + int n, int update_ctx) \ +{ \ + DELEM *dst = dest; \ + const DELEM *src = source; \ + int dst_index; \ + int index= c->index; \ + int frac= c->frac; \ + int sample_index = 0; \ + int x4_aligned_filter_length = c->filter_length & ~3; \ + int x8_aligned_filter_length = c->filter_length & ~7; \ + \ + while (index >= c->phase_count) { \ + sample_index++; \ + index -= c->phase_count; \ + } \ + \ + for (dst_index = 0; dst_index < n; dst_index++) { \ + FELEM *filter = ((FELEM *) c->filter_bank) + c->filter_alloc * index; \ + \ + FELEM2 val=0; \ + int i = 0; \ + if (x8_aligned_filter_length >= 8) { \ + ff_resample_common_apply_filter_x8_##TYPE##_neon(&val, &src[sample_index], \ + filter, x8_aligned_filter_length); \ + i += x8_aligned_filter_length; \ + \ + } else if (x4_aligned_filter_length >= 4) { \ + ff_resample_common_apply_filter_x4_##TYPE##_neon(&val, &src[sample_index], \ + filter, x4_aligned_filter_length); \ + i += x4_aligned_filter_length; \ + } \ + for (; i < c->filter_length; i++) { \ + val += src[sample_index + i] * (FELEM2)filter[i]; \ + } \ + OUT(dst[dst_index], val); \ + \ + frac += c->dst_incr_mod; \ + index += c->dst_incr_div; \ + if (frac >= c->src_incr) { \ + frac -= c->src_incr; \ + index++; \ + } \ + \ + while (index >= c->phase_count) { \ + sample_index++; \ + index -= c->phase_count; \ + } \ + } \ + \ + if(update_ctx){ \ + c->frac= frac; \ + c->index= index; \ + } \ + \ + return sample_index; \ +} \ + +#define OUT(d, v) d = v +DECLARE_RESAMPLE_COMMON_TEMPLATE(float, float, float, float, OUT) +#undef OUT + +#define OUT(d, v) (v) = ((v) + (1<<(14)))>>15; (d) = av_clip_int16(v) +DECLARE_RESAMPLE_COMMON_TEMPLATE(s16, int16_t, int16_t, int32_t, OUT) +#undef OUT + +av_cold void swri_resample_dsp_arm_init(ResampleContext *c) +{ + int cpu_flags = av_get_cpu_flags(); + + if (!have_neon(cpu_flags)) + return; + + switch(c->format) { + case AV_SAMPLE_FMT_FLTP: + if (!c->linear) + c->dsp.resample = ff_resample_common_float_neon; + break; + case AV_SAMPLE_FMT_S16P: + if (!c->linear) + c->dsp.resample = ff_resample_common_s16_neon; + break; + } +} |