From patchwork Thu Jun 1 09:13:19 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: =?utf-8?b?Q2zDqW1lbnQgQsWTc2No?= X-Patchwork-Id: 3778 Delivered-To: ffmpegpatchwork@gmail.com Received: by 10.103.10.2 with SMTP id 2csp673956vsk; Thu, 1 Jun 2017 02:13:50 -0700 (PDT) X-Received: by 10.28.174.213 with SMTP id x204mr563656wme.12.1496308430365; Thu, 01 Jun 2017 02:13:50 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1496308430; cv=none; d=google.com; s=arc-20160816; b=lI8km8zQ/KPwOlKe7T+LFDURKkKt8dSuDFbk58Ud1DG7+GXJPVDRpYjzxzEJ8FzgHD yEefVH9bx246G8eDqtUicd0Mod+l3wejAXypNerIJc0O7YzHAfcIvSA2ok4hDAkczsdl 5Q5ZfVnhGum1aFY2E5j/gxytIjXaXbLH/6lntVA2YHrbcOPZHeaJle0aVUSCr6y6+Qet eT6Q0GW6OvDg5crtj8hD0LHz6XHbObKH6N+6UoYA2XuWn8x2xp/sJMCk0hPW1cBxTZvK u24ezk6W1Adkc8d8JVTuPyG3zU4VbrCoijAMifF5hR8HTlV9Ngqj4sugmanXVkmbiB/B coAA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:reply-to:list-subscribe:list-help:list-post :list-archive:list-unsubscribe:list-id:precedence:subject:user-agent :in-reply-to:mime-version:references:mail-followup-to:message-id:to :from:date:delivered-to:arc-authentication-results; bh=iUf4NqwPjpgyZ6k7ARTHcd/2LHNL8H4Bbfu6a3dzghA=; b=QgtRqoC3IfSm0ej2M0OdioNNuMn6Q163jRsLS7WIXx/HJ3Gw6aGFJpLpyABbPF9HDS NrBOAZZoqJZPlOThBrwZNK+0UVZdd+opBj6iPtq32HT/jcK5iUTmQMiHQAMNpYXz+V7s 1PiYhORZapeAGhkCmL9ZCi8a4zrvh5dHT7MAhB6nH6zAhCwr/X/w0VejaaCaShETrZ8m URGsd6B2Tav/wt80oQOMrGl8xtYh6Zyme7V6KjY++nl19yOgxpO8EvJilYazYonwBhXc SDPe0+cY4ekT7HcvflRPKpAwpoB/Dw/1e319m5R1lI4H+6qmk+Q3ieRW5AXS+4KgVb8x SWnA== ARC-Authentication-Results: i=1; mx.google.com; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id o18si22441699wmg.87.2017.06.01.02.13.49; Thu, 01 Jun 2017 02:13:50 -0700 (PDT) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id E08D568991D; Thu, 1 Jun 2017 12:13:41 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from golem.pkh.me (LStLambert-657-1-117-164.w92-154.abo.wanadoo.fr [92.154.28.164]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 9F688688314 for ; Thu, 1 Jun 2017 12:13:34 +0300 (EEST) Received: from golem.pkh.me (localhost.localdomain [127.0.0.1]) by golem.pkh.me (OpenSMTPD) with ESMTPSA id 9d6fb8b7 (TLSv1.2:ECDHE-RSA-AES256-GCM-SHA384:256:NO) for ; Thu, 1 Jun 2017 09:13:39 +0000 (UTC) Date: Thu, 1 Jun 2017 11:13:19 +0200 From: =?utf-8?B?Q2zDqW1lbnQgQsWTc2No?= To: FFmpeg development discussions and patches Message-ID: <20170601091319.GA15232@golem.pkh.me> Mail-Followup-To: FFmpeg development discussions and patches References: <20170525155052.3585-1-u@pkh.me> MIME-Version: 1.0 In-Reply-To: User-Agent: Mutt/1.8.2 (2017-04-18) Subject: Re: [FFmpeg-devel] [PATCH] lavc/aarch64: add a few SIMD function for AAC PS X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" On Thu, May 25, 2017 at 01:22:22PM -0300, James Almer wrote: [...] > > +function ff_ps_stereo_interpolate_ipdopd_neon, export=1 > > + movrel x5, ipdopd_factors > > + ld1 {v20.4S}, [x5] > > + ld1 {v0.4S,v1.4S}, [x2] > > + ld1 {v6.4S,v7.4S}, [x3] > > +1: > > + ld1 {v2.2S}, [x0] > > + ld1 {v3.2S}, [x1] > > + dup v2.2D, v2.D[0] > > + dup v3.2D, v3.D[0] > > + fadd v0.4S, v0.4S, v6.4S > > + fadd v1.4S, v1.4S, v7.4S > > + zip1 v16.4S, v0.4S, v0.4S > > + zip2 v17.4S, v0.4S, v0.4S > > + zip1 v18.4S, v1.4S, v1.4S > > + zip2 v19.4S, v1.4S, v1.4S > > + fmul v4.4S, v2.4S, v16.4S > > + fmla v4.4S, v3.4S, v17.4S > > + ext v2.16B, v2.16B, v2.16B, #4 > > + ext v3.16B, v3.16B, v3.16B, #4 > > > + fmul v5.4S, v2.4S, v18.4S > > + fmla v5.4S, v3.4S, v19.4S > > + fmla v4.4S, v5.4S, v20.4S > > You could make ipdopd_factors be 0, INT32_MIN, 0, INT32_MIN then replace > the fmla with eor + fadd. > No idea if that will actually be faster, though. I'll check that when benchmarking. Here is a new version adding ff_ps_hybrid_analysis_neon. It was very fun to write, but there are some weirdness: - filter lane is 8 but we're reading only 6 (I suppose that for performance though) - this part is strange: INT64FLOAT sum_re = (INT64FLOAT)filter[i][6][0] * in[6][0]; INT64FLOAT sum_im = (INT64FLOAT)filter[i][6][0] * in[6][1]; why isn't it using the im part of the filter for sum_im? I'll post some benchmark later. From 302346449ce53b92f2a40f6033704b72d90d47ad Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Cl=C3=A9ment=20B=C5=93sch?= Date: Thu, 25 May 2017 17:50:52 +0200 Subject: [PATCH] lavc/aarch64: add a few SIMD function for AAC PS --- For the record, this is the C equivalent for ff_ps_hybrid_analysis_neon: static void ps_hybrid_analysis_c(INTFLOAT (*out)[2], INTFLOAT (*in)[2], const INTFLOAT (*filter)[8][2], int stride, int n) { int i; const INT64FLOAT re0123pCBA9[4] = { // v16.4S in[0][0] + in[12][0], in[1][0] + in[11][0], in[2][0] + in[10][0], in[3][0] + in[ 9][0], }; const INT64FLOAT im0123pCBA9[4] = { // v17.4S in[0][1] + in[12][1], in[1][1] + in[11][1], in[2][1] + in[10][1], in[3][1] + in[ 9][1], }; const INT64FLOAT im0123mCBA9[4] = { // v18.4S in[0][1] - in[12][1], in[1][1] - in[11][1], in[2][1] - in[10][1], in[3][1] - in[ 9][1], }; const INT64FLOAT re0123mCBA9[4] = { // v19.4S in[0][0] - in[12][0], in[1][0] - in[11][0], in[2][0] - in[10][0], in[3][0] - in[ 9][0], }; const INT64FLOAT re45p87_im87m45[4] = { // v20.4S in[4][0] + in[8][0], in[5][0] + in[7][0], in[8][1] - in[4][1], in[7][1] - in[5][1], }; const INT64FLOAT im45p87_re45m87[4] = { // v21.4S in[4][1] + in[8][1], in[5][1] + in[7][1], in[4][0] - in[8][0], in[5][0] - in[7][0], }; for (i = 0; i < n; i++) { INT64FLOAT sum_re4[4]; INT64FLOAT sum_im4[4]; INT64FLOAT sum_re, sum_im; const INT64FLOAT filter_re0123[4] = { // v2.4S filter[i][0][0], filter[i][1][0], filter[i][2][0], filter[i][3][0], }; const INT64FLOAT filter_im0123[4] = { // v3.4S filter[i][0][1], filter[i][1][1], filter[i][2][1], filter[i][3][1], }; const INT64FLOAT filter_re45_im45[4] = { // v4.4S filter[i][4][0], filter[i][5][0], filter[i][4][1], filter[i][5][1], }; sum_re4[0] = filter_re0123[0] * re0123pCBA9[0]; sum_re4[1] = filter_re0123[1] * re0123pCBA9[1]; sum_re4[2] = filter_re0123[2] * re0123pCBA9[2]; sum_re4[3] = filter_re0123[3] * re0123pCBA9[3]; sum_im4[0] = filter_re0123[0] * im0123pCBA9[0]; sum_im4[1] = filter_re0123[1] * im0123pCBA9[1]; sum_im4[2] = filter_re0123[2] * im0123pCBA9[2]; sum_im4[3] = filter_re0123[3] * im0123pCBA9[3]; sum_re4[0] -= filter_im0123[0] * im0123mCBA9[0]; sum_re4[1] -= filter_im0123[1] * im0123mCBA9[1]; sum_re4[2] -= filter_im0123[2] * im0123mCBA9[2]; sum_re4[3] -= filter_im0123[3] * im0123mCBA9[3]; sum_im4[0] += filter_im0123[0] * re0123mCBA9[0]; sum_im4[1] += filter_im0123[1] * re0123mCBA9[1]; sum_im4[2] += filter_im0123[2] * re0123mCBA9[2]; sum_im4[3] += filter_im0123[3] * re0123mCBA9[3]; sum_re4[0] += filter_re45_im45[0] * re45p87_im87m45[0]; sum_re4[1] += filter_re45_im45[1] * re45p87_im87m45[1]; sum_re4[2] += filter_re45_im45[2] * re45p87_im87m45[2]; sum_re4[3] += filter_re45_im45[3] * re45p87_im87m45[3]; sum_im4[0] += filter_re45_im45[0] * im45p87_re45m87[0]; sum_im4[1] += filter_re45_im45[1] * im45p87_re45m87[1]; sum_im4[2] += filter_re45_im45[2] * im45p87_re45m87[2]; sum_im4[3] += filter_re45_im45[3] * im45p87_re45m87[3]; sum_re = sum_re4[0] + sum_re4[1] + sum_re4[2] + sum_re4[3] + (INT64FLOAT)filter[i][6][0] * in[6][0]; sum_im = sum_im4[0] + sum_im4[1] + sum_im4[2] + sum_im4[3] + (INT64FLOAT)filter[i][6][0] * in[6][1]; #if USE_FIXED out[i * stride][0] = (int)((sum_re + 0x40000000) >> 31); out[i * stride][1] = (int)((sum_im + 0x40000000) >> 31); #else out[i * stride][0] = sum_re; out[i * stride][1] = sum_im; #endif /* USE_FIXED */ } } --- libavcodec/aacpsdsp.h | 1 + libavcodec/aacpsdsp_template.c | 2 + libavcodec/aarch64/Makefile | 2 + libavcodec/aarch64/aacpsdsp_init_aarch64.c | 48 ++++++++++ libavcodec/aarch64/aacpsdsp_neon.S | 145 +++++++++++++++++++++++++++++ 5 files changed, 198 insertions(+) create mode 100644 libavcodec/aarch64/aacpsdsp_init_aarch64.c create mode 100644 libavcodec/aarch64/aacpsdsp_neon.S diff --git a/libavcodec/aacpsdsp.h b/libavcodec/aacpsdsp.h index ad9bbb81bd..3714a08052 100644 --- a/libavcodec/aacpsdsp.h +++ b/libavcodec/aacpsdsp.h @@ -51,6 +51,7 @@ typedef struct PSDSPContext { void AAC_RENAME(ff_psdsp_init)(PSDSPContext *s); void ff_psdsp_init_arm(PSDSPContext *s); +void ff_psdsp_init_aarch64(PSDSPContext *s); void ff_psdsp_init_mips(PSDSPContext *s); void ff_psdsp_init_x86(PSDSPContext *s); diff --git a/libavcodec/aacpsdsp_template.c b/libavcodec/aacpsdsp_template.c index 3049ce8b79..158160b246 100644 --- a/libavcodec/aacpsdsp_template.c +++ b/libavcodec/aacpsdsp_template.c @@ -222,6 +222,8 @@ av_cold void AAC_RENAME(ff_psdsp_init)(PSDSPContext *s) #if !USE_FIXED if (ARCH_ARM) ff_psdsp_init_arm(s); + if (ARCH_AARCH64) + ff_psdsp_init_aarch64(s); if (ARCH_MIPS) ff_psdsp_init_mips(s); if (ARCH_X86) diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile index 104bc67802..d440b1b18a 100644 --- a/libavcodec/aarch64/Makefile +++ b/libavcodec/aarch64/Makefile @@ -11,6 +11,7 @@ OBJS-$(CONFIG_NEON_CLOBBER_TEST) += aarch64/neontest.o OBJS-$(CONFIG_VIDEODSP) += aarch64/videodsp_init.o # decoders/encoders +OBJS-$(CONFIG_AAC_DECODER) += aarch64/aacpsdsp_init_aarch64.o OBJS-$(CONFIG_DCA_DECODER) += aarch64/synth_filter_init.o OBJS-$(CONFIG_RV40_DECODER) += aarch64/rv40dsp_init_aarch64.o OBJS-$(CONFIG_VC1DSP) += aarch64/vc1dsp_init_aarch64.o @@ -42,6 +43,7 @@ NEON-OBJS-$(CONFIG_MDCT) += aarch64/mdct_neon.o NEON-OBJS-$(CONFIG_MPEGAUDIODSP) += aarch64/mpegaudiodsp_neon.o # decoders/encoders +NEON-OBJS-$(CONFIG_AAC_DECODER) += aarch64/aacpsdsp_neon.o NEON-OBJS-$(CONFIG_DCA_DECODER) += aarch64/synth_filter_neon.o NEON-OBJS-$(CONFIG_VORBIS_DECODER) += aarch64/vorbisdsp_neon.o NEON-OBJS-$(CONFIG_VP9_DECODER) += aarch64/vp9itxfm_16bpp_neon.o \ diff --git a/libavcodec/aarch64/aacpsdsp_init_aarch64.c b/libavcodec/aarch64/aacpsdsp_init_aarch64.c new file mode 100644 index 0000000000..9ffc4bb16f --- /dev/null +++ b/libavcodec/aarch64/aacpsdsp_init_aarch64.c @@ -0,0 +1,48 @@ +/* + * 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/cpu.h" +#include "libavcodec/aacpsdsp.h" + +void ff_ps_add_squares_neon(float *dst, const float (*src)[2], int n); +void ff_ps_mul_pair_single_neon(float (*dst)[2], float (*src0)[2], + float *src1, int n); +void ff_ps_hybrid_analysis_neon(float (*out)[2], float (*in)[2], + const float (*filter)[8][2], + int stride, int n); +void ff_ps_stereo_interpolate_neon(float (*l)[2], float (*r)[2], + float h[2][4], float h_step[2][4], + int len); +void ff_ps_stereo_interpolate_ipdopd_neon(float (*l)[2], float (*r)[2], + float h[2][4], float h_step[2][4], + int len); + +av_cold void ff_psdsp_init_aarch64(PSDSPContext *s) +{ + int cpu_flags = av_get_cpu_flags(); + + if (have_neon(cpu_flags)) { + s->add_squares = ff_ps_add_squares_neon; + s->mul_pair_single = ff_ps_mul_pair_single_neon; + s->hybrid_analysis = ff_ps_hybrid_analysis_neon; + s->stereo_interpolate[0] = ff_ps_stereo_interpolate_neon; + s->stereo_interpolate[1] = ff_ps_stereo_interpolate_ipdopd_neon; + } +} diff --git a/libavcodec/aarch64/aacpsdsp_neon.S b/libavcodec/aarch64/aacpsdsp_neon.S new file mode 100644 index 0000000000..acfc9c0f28 --- /dev/null +++ b/libavcodec/aarch64/aacpsdsp_neon.S @@ -0,0 +1,145 @@ +/* + * 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/aarch64/asm.S" + +function ff_ps_add_squares_neon, export=1 +1: ld1 {v0.4S,v1.4S}, [x1], #32 + fmul v0.4S, v0.4S, v0.4S + fmul v1.4S, v1.4S, v1.4S + faddp v2.4S, v0.4S, v1.4S + ld1 {v3.4S}, [x0] + fadd v3.4S, v3.4S, v2.4S + st1 {v3.4S}, [x0], #16 + subs w2, w2, #4 + b.gt 1b + ret +endfunc + +function ff_ps_mul_pair_single_neon, export=1 +1: ld1 {v0.4S,v1.4S}, [x1], #32 + ld1 {v2.4S}, [x2], #16 + zip1 v3.4S, v2.4S, v2.4S + zip2 v4.4S, v2.4S, v2.4S + fmul v0.4S, v0.4S, v3.4S + fmul v1.4S, v1.4S, v4.4S + st1 {v0.4S,v1.4S}, [x0], #32 + subs w3, w3, #4 + b.gt 1b + ret +endfunc + +function ff_ps_stereo_interpolate_neon, export=1 + ld1 {v0.4S}, [x2] + ld1 {v1.4S}, [x3] +1: ld1 {v2.2S}, [x0] + ld1 {v3.2S}, [x1] + dup v2.2D, v2.D[0] + dup v3.2D, v3.D[0] + fadd v0.4S, v0.4S, v1.4S + zip1 v4.4S, v0.4S, v0.4S + zip2 v5.4S, v0.4S, v0.4S + fmul v2.4S, v2.4S, v4.4S + fmla v2.4S, v3.4S, v5.4S + st1 {v2.D}[0], [x0], #8 + st1 {v2.D}[1], [x1], #8 + subs w4, w4, #1 + b.gt 1b + ret +endfunc + +const ipdopd_factors, align=4 + .float -1.0, 1.0, -1.0, 1.0 +endconst + +function ff_ps_stereo_interpolate_ipdopd_neon, export=1 + movrel x5, ipdopd_factors + ld1 {v20.4S}, [x5] + ld1 {v0.4S,v1.4S}, [x2] + ld1 {v6.4S,v7.4S}, [x3] +1: ld1 {v2.2S}, [x0] + ld1 {v3.2S}, [x1] + dup v2.2D, v2.D[0] + dup v3.2D, v3.D[0] + fadd v0.4S, v0.4S, v6.4S + fadd v1.4S, v1.4S, v7.4S + zip1 v16.4S, v0.4S, v0.4S + zip2 v17.4S, v0.4S, v0.4S + zip1 v18.4S, v1.4S, v1.4S + zip2 v19.4S, v1.4S, v1.4S + fmul v4.4S, v2.4S, v16.4S + fmla v4.4S, v3.4S, v17.4S + ext v2.16B, v2.16B, v2.16B, #4 + ext v3.16B, v3.16B, v3.16B, #4 + fmul v5.4S, v2.4S, v18.4S + fmla v5.4S, v3.4S, v19.4S + fmla v4.4S, v5.4S, v20.4S + st1 {v4.D}[0], [x0], #8 + st1 {v4.D}[1], [x1], #8 + subs w4, w4, #1 + b.gt 1b + ret +endfunc + +function ff_ps_hybrid_analysis_neon, export=1 + sxtw x3, w3 + lsl x3, x3, #3 + ld2 {v0.4S,v1.4S}, [x1], #32 + ld2 {v2.2S,v3.2S}, [x1], #16 + ld1 {v8.2S}, [x1], #8 + ld2 {v4.2S,v5.2S}, [x1], #16 + ld2 {v6.4S,v7.4S}, [x1] + rev64 v6.4S, v6.4S + rev64 v7.4S, v7.4S + ext v6.16B, v6.16B, v6.16B, #8 + ext v7.16B, v7.16B, v7.16B, #8 + rev64 v4.2S, v4.2S + rev64 v5.2S, v5.2S + mov v2.D[1], v3.D[0] + mov v4.D[1], v5.D[0] + mov v5.D[1], v2.D[0] + mov v3.D[1], v4.D[0] + fadd v16.4S, v0.4S, v6.4S + fadd v17.4S, v1.4S, v7.4S + fsub v18.4S, v1.4S, v7.4S + fsub v19.4S, v0.4S, v6.4S + fadd v22.4S, v2.4S, v4.4S + fsub v23.4S, v5.4S, v3.4S + trn1 v20.2D, v22.2D, v23.2D // {re4+re8, re5+re7, im8-im4, im7-im5} + trn2 v21.2D, v22.2D, v23.2D // {im4+im8, im5+im7, re4-re8, re5-re7} +1: ld2 {v2.4S,v3.4S}, [x2], #32 + ld2 {v4.2S,v5.2S}, [x2], #16 + ld1 {v6.2S}, [x2], #8 + add x2, x2, #8 + mov v4.D[1], v5.D[0] + mov v6.S[1], v6.S[0] + fmul v6.2S, v6.2S, v8.2S + fmul v0.4S, v2.4S, v16.4S + fmul v1.4S, v2.4S, v17.4S + fmls v0.4S, v3.4S, v18.4S + fmla v1.4S, v3.4S, v19.4S + fmla v0.4S, v4.4S, v20.4S + fmla v1.4S, v4.4S, v21.4S + faddp v0.4S, v0.4S, v1.4S + faddp v0.4S, v0.4S, v0.4S + fadd v0.2S, v0.2S, v6.2S + st1 {v0.2S}, [x0], x3 + subs w4, w4, #1 + b.gt 1b + ret +endfunc -- 2.13.0