From patchwork Thu Nov 2 21:07:03 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: =?utf-8?q?R=C3=A9mi_Denis-Courmont?= X-Patchwork-Id: 44482 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:671c:b0:181:818d:5e7f with SMTP id q28csp148435pzh; Thu, 2 Nov 2023 14:07:15 -0700 (PDT) X-Google-Smtp-Source: AGHT+IFOREcAM1NFVE4Ru7647U9pZ+AxBcfPqukb9rvnara3rr8TamTv3rZ8kRXjA2TjZklJR5Dw X-Received: by 2002:a17:907:7e9e:b0:9c7:5186:de1a with SMTP id qb30-20020a1709077e9e00b009c75186de1amr5561322ejc.8.1698959234864; Thu, 02 Nov 2023 14:07:14 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1698959234; cv=none; d=google.com; s=arc-20160816; b=sJmHOO850mFdfO4lbqa533Ut12N2AAywK5Vl6ua4NIyBWWWefjxUBxtHFjd87noOfZ lkbiopc+8TmOqA0SDw4+CzQrH6d9rPNZa1Gw/9YwuoOTtNYWiOOPXFROqeXFw8TG+RxP yQYoJOY1l6HS2J4gb77mTSF/AoOwHWE2aWsZwqKNc1j6Z3dF7bwOSN7CmMBdiE6krDnp rN0920eOBKUEY5EKoROhpdX3LGSue5iIsfzCbUlIPKaMkTqGUHlwIEQorWKCdaLXhlcD hYZT48JiTaW7GQfwnkAxOWl3c4mLbJrqfZkUovN8zwb+339F9thrddpYByck/RVsNPO5 wQZA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:reply-to:list-subscribe :list-help:list-post:list-archive:list-unsubscribe:list-id :precedence:subject:mime-version:message-id:date:to:from :delivered-to; bh=Frsdfeqhxie+CVyo4hw8vAT9PPrEabPk9++/zkOuEKQ=; fh=YOA8vD9MJZuwZ71F/05pj6KdCjf6jQRmzLS+CATXUQk=; b=XLN7CSo8TalD2T4W8t0IBFgBSLYGmKlOx9q7s1doaPemJvuutXvcqRCdYkzl5UHxfL srWqQzWAXP4PO94XKqP4GlR6fWdRJ4QIpjOalw8cZeZpYYpa3Hjr6M1KPHmZPXUPYZ1Z x7Q7fLllLrCSMG4KGLjnIimYdW5dkSFbHjvXl8UbHGIUoFfdwcza7kXfBMTXrAGbeJLi vMzjWl2bbxicz++bH5q7DZPyZ1Vhd1GWnOmamB+31wMUcqrqi7hPDqa7zkV4Zia9+KzF 73O3ZX/eg7EwsY/MDdgGkzVEfMVgiw2+ESO1XJsRMBWg+Anhxyo9AUW+kWKmt6IIsz0H VSCQ== 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 23-20020a170906209700b009b94cff2899si147311ejq.451.2023.11.02.14.07.14; Thu, 02 Nov 2023 14:07:14 -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 AE72368CCBE; Thu, 2 Nov 2023 23:07:10 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from ursule.remlab.net (vps-a2bccee9.vps.ovh.net [51.75.19.47]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 6739868CAE8 for ; Thu, 2 Nov 2023 23:07:04 +0200 (EET) Received: from basile.remlab.net (localhost [IPv6:::1]) by ursule.remlab.net (Postfix) with ESMTP id B8FB2C0172 for ; Thu, 2 Nov 2023 23:07:03 +0200 (EET) From: =?utf-8?q?R=C3=A9mi_Denis-Courmont?= To: ffmpeg-devel@ffmpeg.org Date: Thu, 2 Nov 2023 23:07:03 +0200 Message-ID: <20231102210703.27337-1-remi@remlab.net> X-Mailer: git-send-email 2.42.0 MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH] lavc/opusdsp: rewrite R-V V postfilter X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 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" X-TUID: 5lG7JFJPWvsP This uses a more traditional approach allowing up processing of up to period minus two elements per iteration. This also allows the algorithm to work for all and any vector length. As the T-Head C908 device under test can load 16 elements loop, there is unsurprisingly a little performance drop when the period is minimal and the parallelism is capped at 13 elements: Before: postfilter_15_c: 21222.2 postfilter_15_rvv_f32: 22007.7 postfilter_512_c: 20189.7 postfilter_512_rvv_f32: 22004.2 postfilter_1022_c: 20189.7 postfilter_1022_rvv_f32: 22004.2 After: postfilter_15_c: 20189.5 postfilter_15_rvv_f32: 7057.2 postfilter_512_c: 20189.5 postfilter_512_rvv_f32: 5667.2 postfilter_1022_c: 20192.7 postfilter_1022_rvv_f32: 5667.2 --- libavcodec/riscv/opusdsp_init.c | 22 ++------- libavcodec/riscv/opusdsp_rvv.S | 87 +++++++++++++++------------------ 2 files changed, 42 insertions(+), 67 deletions(-) diff --git a/libavcodec/riscv/opusdsp_init.c b/libavcodec/riscv/opusdsp_init.c index 7fde9b1fa8..f5a842a326 100644 --- a/libavcodec/riscv/opusdsp_init.c +++ b/libavcodec/riscv/opusdsp_init.c @@ -25,30 +25,14 @@ #include "libavutil/riscv/cpu.h" #include "libavcodec/opusdsp.h" -void ff_opus_postfilter_rvv_128(float *data, int period, float *g, int len); -void ff_opus_postfilter_rvv_256(float *data, int period, float *g, int len); -void ff_opus_postfilter_rvv_512(float *data, int period, float *g, int len); -void ff_opus_postfilter_rvv_1024(float *data, int period, float *g, int len); +void ff_opus_postfilter_rvv(float *data, int period, float *g, int len); av_cold void ff_opus_dsp_init_riscv(OpusDSP *d) { #if HAVE_RVV int flags = av_get_cpu_flags(); - if (flags & AV_CPU_FLAG_RVV_F32) - switch (ff_get_rv_vlenb()) { - case 16: - d->postfilter = ff_opus_postfilter_rvv_128; - break; - case 32: - d->postfilter = ff_opus_postfilter_rvv_256; - break; - case 64: - d->postfilter = ff_opus_postfilter_rvv_512; - break; - case 128: - d->postfilter = ff_opus_postfilter_rvv_512; - break; - } + if ((flags & AV_CPU_FLAG_RVV_F32) && (flags & AV_CPU_FLAG_RVB_ADDR)) + d->postfilter = ff_opus_postfilter_rvv; #endif } diff --git a/libavcodec/riscv/opusdsp_rvv.S b/libavcodec/riscv/opusdsp_rvv.S index b3d23a9de5..79ae86c30e 100644 --- a/libavcodec/riscv/opusdsp_rvv.S +++ b/libavcodec/riscv/opusdsp_rvv.S @@ -20,56 +20,47 @@ #include "libavutil/riscv/asm.S" -func ff_opus_postfilter_rvv_128, zve32f - lvtypei a5, e32, m2, ta, ma - j 1f -endfunc - -func ff_opus_postfilter_rvv_512, zve32f - lvtypei a5, e32, mf2, ta, ma - j 1f -endfunc +func ff_opus_postfilter_rvv, zve32f + flw fa0, 0(a2) // g0 + slli t1, a1, 2 + flw fa1, 4(a2) // g1 + sub t0, a0, t1 + flw fa2, 8(a2) // g2 + addi t0, t0, 2 * 4 // data - (period - 2) = initial &x0 -func ff_opus_postfilter_rvv_1024, zve32f - lvtypei a5, e32, mf4, ta, ma - j 1f -endfunc - -func ff_opus_postfilter_rvv_256, zve32f - lvtypei a5, e32, m1, ta, ma + flw ft4, -16(t0) + addi t3, a1, -2 // maximum parallelism w/o stepping our tail + flw ft3, -12(t0) + flw ft2, -8(t0) + flw ft1, -4(t0) 1: - li a4, 5 - addi a1, a1, 2 - slli a1, a1, 2 - lw t1, 4(a2) - vsetivli zero, 3, e32, m1, ta, ma - vle32.v v24, (a2) - sub a1, a0, a1 // a1 = &x4 = &data[-(period + 2)] - vsetvl zero, a4, a5 - vslide1up.vx v8, v24, t1 - lw t2, 8(a2) - vle32.v v16, (a1) - vslide1up.vx v24, v8, t2 // v24 = { g[2], g[1], g[0], g[1], g[2] } -2: - vsetvl t0, a3, a5 - vle32.v v0, (a0) - sub a3, a3, t0 -3: - vsetvl zero, a4, a5 - lw t2, 20(a1) - vfmul.vv v8, v24, v16 - addi a0, a0, 4 - vslide1down.vx v16, v16, t2 - addi a1, a1, 4 - vfredusum.vs v0, v8, v0 - vsetvl zero, t0, a5 - vmv.x.s t1, v0 - addi t0, t0, -1 - vslide1down.vx v0, v0, zero - sw t1, -4(a0) - bnez t0, 3b - - bnez a3, 2b + min t1, a3, t3 + vsetvli t1, t1, e32, m4, ta, ma + vle32.v v0, (t0) // x0 + sub a3, a3, t1 + vle32.v v28, (a0) + sh2add t0, t1, t0 + vfslide1up.vf v4, v0, ft1 + addi t2, t1, -4 + vfslide1up.vf v8, v4, ft2 + vfslide1up.vf v12, v8, ft3 + vfslide1up.vf v16, v12, ft4 + vfadd.vv v20, v4, v12 + vfadd.vv v24, v0, v16 + vslidedown.vx v12, v0, t2 + vfmacc.vf v28, fa0, v8 + vslidedown.vi v4, v12, 2 + vfmacc.vf v28, fa1, v20 + vslide1down.vx v8, v12, zero + vfmacc.vf v28, fa2, v24 + vslide1down.vx v0, v4, zero + vse32.v v28, (a0) + vfmv.f.s ft4, v12 + sh2add a0, t1, a0 + vfmv.f.s ft2, v4 + vfmv.f.s ft3, v8 + vfmv.f.s ft1, v0 + bnez a3, 1b ret endfunc