From patchwork Thu Mar 31 17:23:42 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35105 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp229781pzb; Thu, 31 Mar 2022 10:24:22 -0700 (PDT) X-Google-Smtp-Source: ABdhPJwh3tFWZMu79gl5vZfQXpjqfph0yb2Qq0G2bTLmIg5wcgCb3aQ1a0rLrr5ylx8iIDnDKz4O X-Received: by 2002:a17:907:7f8b:b0:6e0:612b:38fd with SMTP id qk11-20020a1709077f8b00b006e0612b38fdmr5931343ejc.552.1648747462408; Thu, 31 Mar 2022 10:24:22 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747462; cv=none; d=google.com; s=arc-20160816; b=AXqqtmGWVmkSk8wHGo0W806Qi8JmRezMWjNM2O37zp/26cO//33AxMpxls13GaOzW+ LxHkOPnnizsoJK/m3Nab/NhUrS4sFE27SFgx4G1zPSF8dcii8NUYLjprAmWMhXHiVaLD ANfySFclIN5mMuGxdP5Ir4K6D4M12zxDK1cl0/l7Oo3bjtdiLxCUNPtMGOMyx5ENBRst 6N60rNcbNlQtEDpAzx1AkQte8iHIAxjzqyO2l8oaXy0GSqnrI837a1EYxgUbpMcMaJ2H WToi5DXEbd1OMHzbbHXC9j1Y0xpsnBvDK8oL9dNtdtkZkEQEkNK5Ey6qsZjmpLe5C+yE NHlg== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=73U8LaxR5dGn4uPvCgceicAmmfI9uxhrLKTGieqR1TA=; b=liaktFSd2J9RKE/5YtFf4TWrxmfoBhPyOCYWqJhAs1GqJEo+yHza1M/4+dgNviZiXw wBeqRdfLbFAmltcKHugCFiRgO9vxxe6b0LazEo8ZUelfimM4evHSIJUP9Ef8bzI7LtWE k0W3SNCj2VetODn3DaIG2wqfEtW70ZAx+Ru9FoXyxQsNJdJri1XdBYnFSDrP2/953lWc eb7G+UnDYnWCXW3qzSG3PkdUEgel8EqKgIC9CBE+UXMiorkhDraIZCiHGfI5WtXPUnX4 PlSng33YHAM0CTaK9WpOj3I7E8lSOy4vA3/zV6iK79ekuBnQ6erPgPRw9gxWa5hqExJw ozNw== 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 k10-20020a17090646ca00b006df76385d9csi70539ejs.572.2022.03.31.10.24.22; Thu, 31 Mar 2022 10:24:22 -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 9867468B22F; Thu, 31 Mar 2022 20:24:08 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail148106.authsmtp.co.uk (outmail148106.authsmtp.co.uk [62.13.148.106]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 0710768B21F for ; Thu, 31 Mar 2022 20:24:06 +0300 (EEST) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt16.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHO5UV010903; Thu, 31 Mar 2022 18:24:05 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHO3GW062220 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:03 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:03 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:42 +0100 Message-Id: <20220331172351.550818-2-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 5d58465c-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQpUIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLRoFEAVAenxy ZQhjW3BcVAp/IUN1 RkhRRnBUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdW0Gb30N d0kGLV8ZQk8QG2x0 bhwOVSskREEeX20y KVkiLVkEVEocOEE1 OlggX11QPR4WFgBE V11AGzNCTwAI X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthSMTP-Origin: 51.9.63.237/2525 X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 01/10] checkasm: Add vc1dsp in-loop deblocking filter tests 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: 68RIL775fcmL Note that the benchmarking results for these functions are highly dependent upon the input data. Therefore, each function is benchmarked twice, corresponding to the best and worst case complexity of the reference C implementation. The performance of a real stream decode will fall somewhere between these two extremes. Signed-off-by: Ben Avison --- tests/checkasm/Makefile | 1 + tests/checkasm/checkasm.c | 3 ++ tests/checkasm/checkasm.h | 1 + tests/checkasm/vc1dsp.c | 102 ++++++++++++++++++++++++++++++++++++++ tests/fate/checkasm.mak | 1 + 5 files changed, 108 insertions(+) create mode 100644 tests/checkasm/vc1dsp.c diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile index f768b1144e..7133a6ee66 100644 --- a/tests/checkasm/Makefile +++ b/tests/checkasm/Makefile @@ -11,6 +11,7 @@ AVCODECOBJS-$(CONFIG_H264PRED) += h264pred.o AVCODECOBJS-$(CONFIG_H264QPEL) += h264qpel.o AVCODECOBJS-$(CONFIG_LLVIDDSP) += llviddsp.o AVCODECOBJS-$(CONFIG_LLVIDENCDSP) += llviddspenc.o +AVCODECOBJS-$(CONFIG_VC1DSP) += vc1dsp.o AVCODECOBJS-$(CONFIG_VP8DSP) += vp8dsp.o AVCODECOBJS-$(CONFIG_VIDEODSP) += videodsp.o diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c index 748d6a9f3a..c2efd81b6d 100644 --- a/tests/checkasm/checkasm.c +++ b/tests/checkasm/checkasm.c @@ -147,6 +147,9 @@ static const struct { #if CONFIG_V210_ENCODER { "v210enc", checkasm_check_v210enc }, #endif + #if CONFIG_VC1DSP + { "vc1dsp", checkasm_check_vc1dsp }, + #endif #if CONFIG_VP8DSP { "vp8dsp", checkasm_check_vp8dsp }, #endif diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h index c3192d8c23..52ab18a5b1 100644 --- a/tests/checkasm/checkasm.h +++ b/tests/checkasm/checkasm.h @@ -78,6 +78,7 @@ void checkasm_check_sw_scale(void); void checkasm_check_utvideodsp(void); void checkasm_check_v210dec(void); void checkasm_check_v210enc(void); +void checkasm_check_vc1dsp(void); void checkasm_check_vf_eq(void); void checkasm_check_vf_gblur(void); void checkasm_check_vf_hflip(void); diff --git a/tests/checkasm/vc1dsp.c b/tests/checkasm/vc1dsp.c new file mode 100644 index 0000000000..2fd6c74d6c --- /dev/null +++ b/tests/checkasm/vc1dsp.c @@ -0,0 +1,102 @@ +/* + * Copyright (c) 2022 Ben Avison + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 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 General Public License for more details. + * + * You should have received a copy of the GNU 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 + +#include "checkasm.h" + +#include "libavcodec/vc1dsp.h" + +#include "libavutil/common.h" +#include "libavutil/internal.h" +#include "libavutil/intreadwrite.h" +#include "libavutil/mem_internal.h" + +#define VC1DSP_TEST(func) { #func, offsetof(VC1DSPContext, func) }, + +typedef struct { + const char *name; + size_t offset; +} test; + +#define RANDOMIZE_BUFFER8_MID_WEIGHTED(name, size) \ + do { \ + uint8_t *p##0 = name##0, *p##1 = name##1; \ + int i = (size); \ + while (i-- > 0) { \ + int x = 0x80 | (rnd() & 0x7F); \ + x >>= rnd() % 9; \ + if (rnd() & 1) \ + x = -x; \ + *p##1++ = *p##0++ = 0x80 + x; \ + } \ + } while (0) + +static void check_loop_filter(void) +{ + /* Deblocking filter buffers are big enough to hold a 16x16 block, + * plus 16 columns left and 4 rows above to hold filter inputs + * (depending on whether v or h neighbouring block edge, oversized + * horizontally to maintain 16-byte alignment) plus 16 columns and + * 4 rows below to catch write overflows */ + LOCAL_ALIGNED_16(uint8_t, filter_buf0, [24 * 48]); + LOCAL_ALIGNED_16(uint8_t, filter_buf1, [24 * 48]); + + VC1DSPContext h; + + const test tests[] = { + VC1DSP_TEST(vc1_v_loop_filter4) + VC1DSP_TEST(vc1_h_loop_filter4) + VC1DSP_TEST(vc1_v_loop_filter8) + VC1DSP_TEST(vc1_h_loop_filter8) + VC1DSP_TEST(vc1_v_loop_filter16) + VC1DSP_TEST(vc1_h_loop_filter16) + }; + + ff_vc1dsp_init(&h); + + for (size_t t = 0; t < FF_ARRAY_ELEMS(tests); ++t) { + void (*func)(uint8_t *, ptrdiff_t, int) = *(void **)((intptr_t) &h + tests[t].offset); + declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *, ptrdiff_t, int); + if (check_func(func, "vc1dsp.%s", tests[t].name)) { + for (int count = 1000; count > 0; --count) { + int pq = rnd() % 31 + 1; + RANDOMIZE_BUFFER8_MID_WEIGHTED(filter_buf, 24 * 48); + call_ref(filter_buf0 + 4 * 48 + 16, 48, pq); + call_new(filter_buf1 + 4 * 48 + 16, 48, pq); + if (memcmp(filter_buf0, filter_buf1, 24 * 48)) + fail(); + } + } + for (int j = 0; j < 24; ++j) + for (int i = 0; i < 48; ++i) + filter_buf1[j * 48 + i] = 0x60 + 0x40 * (i >= 16 && j >= 4); + if (check_func(func, "vc1dsp.%s_bestcase", tests[t].name)) + bench_new(filter_buf1 + 4 * 48 + 16, 48, 1); + if (check_func(func, "vc1dsp.%s_worstcase", tests[t].name)) + bench_new(filter_buf1 + 4 * 48 + 16, 48, 31); + } +} + +void checkasm_check_vc1dsp(void) +{ + check_loop_filter(); + report("loop_filter"); +} diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak index 6db8f09d12..99e6bb13c4 100644 --- a/tests/fate/checkasm.mak +++ b/tests/fate/checkasm.mak @@ -32,6 +32,7 @@ FATE_CHECKASM = fate-checkasm-aacpsdsp \ fate-checkasm-utvideodsp \ fate-checkasm-v210dec \ fate-checkasm-v210enc \ + fate-checkasm-vc1dsp \ fate-checkasm-vf_blend \ fate-checkasm-vf_colorspace \ fate-checkasm-vf_eq \ From patchwork Thu Mar 31 17:23:43 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35106 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp229844pzb; Thu, 31 Mar 2022 10:24:33 -0700 (PDT) X-Google-Smtp-Source: ABdhPJzgc62MBrUQ2mFaKmFlBi3uri9+M9t/ntMLMKlhV9r0QgAz67gRMxc/4glk2Ul23HlIDY3f X-Received: by 2002:a17:907:a421:b0:6da:b635:fbed with SMTP id sg33-20020a170907a42100b006dab635fbedmr6089944ejc.91.1648747473495; Thu, 31 Mar 2022 10:24:33 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747473; cv=none; d=google.com; s=arc-20160816; b=bSf7GdHr8lkjJ2mbFGKiGqGWQaMDaghAQfkssisRwmz5kJLmEIYMOaSqjQKfn2c1+Y DKpuaeBlHzOUZRRyl6apdlJ39MvL06INVsYltQE5+h6b1aciMhk0gbszt8HEf9OwnEK5 U7UqA6jSXJVbEwwaLl8Dc+SuWw4r5XtMZwHwKV4MfXSVtBT33pC63VEyOZHMUrdKc9sh b2T+pya4xhxjsagqoGu56mE5+k8BObAay78uTK2uzJd9XXYlAnNgcsm1Ap+PI0m3VaYG 01ITeiDVsSU4RC/2cM9BKXjP7UdpvDpUc+2jC01hRIn3VJ6NJtkEPRmhDGQghrKX1M7/ FITQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=nQ1G7qagDAfRdmg8EJshtF++zFEB8erXRVCcun2c10A=; b=MAT/KJB62lidVrBYyv2SrOv/6HYzpEg5a79sEJ0T0yb4xKS/mOSuPzcZnD5UY5l0bf 9pTVrTBan+/FqKE55h4hmGqY7oXdVG1czZ/Y0D/UxnA16R19N+hxEn3uh4MMGoBnikZL rF08VgT/yu5jMLXYxmzFmv/RSh0JGzN2dkK34nJwHZGWKb8lw0Ingwh30sl2hIn8KWSD oHj+VLqaQZIfdk3PViYfwBf7Q+SU1HJVLnvb4AyZCVZnWA7gyawMPn98v03/vGbFetzQ jf89B0TS8t099T5qqHOZHbY7Ag93eQXq3Lfna/e2Pdw6V3rfXg3G42u3u9eGcQB91S1s awiQ== 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 g6-20020a17090670c600b006e02924244dsi140392ejk.52.2022.03.31.10.24.33; Thu, 31 Mar 2022 10:24:33 -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 AC85C68B24F; Thu, 31 Mar 2022 20:24:14 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail149075.authsmtp.net (outmail149075.authsmtp.net [62.13.149.75]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 3D08268B239 for ; Thu, 31 Mar 2022 20:24:08 +0300 (EEST) Received: from punt22.authsmtp.com (punt22.authsmtp.com [62.13.128.207]) by punt15.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHO7Zs019522 for ; Thu, 31 Mar 2022 18:24:07 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt22.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHO7DJ034590; Thu, 31 Mar 2022 18:24:07 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHO5RR062272 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:05 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:05 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:43 +0100 Message-Id: <20220331172351.550818-3-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 5e67dfc9-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQpXIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLRwQFQVAenx3 ZwhgVnZdXAp/IEB6 FhgHHHBUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdW0Fb30N d0kGLV8ZQk8QG2x0 bhwOVSskREEeX20y KQIrMEURVFoLO0Mp P1w7XBoEPgQOEWUA X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 02/10] checkasm: Add vc1dsp inverse transform tests 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: kTiZ6h2UIf3j This test deliberately doesn't exercise the full range of inputs described in the committee draft VC-1 standard. It says: input coefficients in frequency domain, D, satisfy -2048 <= D < 2047 intermediate coefficients, E, satisfy -4096 <= E < 4095 fully inverse-transformed coefficients, R, satisfy -512 <= R < 511 For one thing, the inequalities look odd. Did they mean them to go the other way round? That would make more sense because the equations generally both add and subtract coefficients multiplied by constants, including powers of 2. Requiring the most-negative values to be valid extends the number of bits to represent the intermediate values just for the sake of that one case! For another thing, the extreme values don't look to occur in real streams - both in my experience and supported by the following comment in the AArch32 decoder: tNhalf is half of the value of tN (as described in vc1_inv_trans_8x8_c). This is done because sometimes files have input that causes tN + tM to overflow. To avoid this overflow, we compute tNhalf, then compute tNhalf + tM (which doesn't overflow), and then we use vhadd to compute (tNhalf + (tNhalf + tM)) >> 1 which does not overflow because it is one instruction. My AArch64 decoder goes further than this. It calculates tNhalf and tM then does an SRA (essentially a fused halve and add) to compute (tN + tM) >> 1 without ever having to hold (tNhalf + tM) in a 16-bit element without overflowing. It only encounters difficulties if either tNhalf or tM overflow in isolation. I haven't had sight of the final standard, so it's possible that these issues were dealt with during finalisation, which could explain the lack of usage of extreme inputs in real streams. Or a preponderance of decoders that only support 16-bit intermediate values in their inverse transforms might have caused encoders to steer clear of such cases. I have effectively followed this approach in the test, and limited the scale of the coefficients sufficient that both the existing AArch32 decoder and my new AArch64 decoder both pass. Signed-off-by: Ben Avison --- tests/checkasm/vc1dsp.c | 283 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 283 insertions(+) diff --git a/tests/checkasm/vc1dsp.c b/tests/checkasm/vc1dsp.c index 2fd6c74d6c..7d4457306f 100644 --- a/tests/checkasm/vc1dsp.c +++ b/tests/checkasm/vc1dsp.c @@ -30,12 +30,208 @@ #include "libavutil/mem_internal.h" #define VC1DSP_TEST(func) { #func, offsetof(VC1DSPContext, func) }, +#define VC1DSP_SIZED_TEST(func, width, height) { #func, offsetof(VC1DSPContext, func), width, height }, typedef struct { const char *name; size_t offset; + int width; + int height; } test; +typedef struct matrix { + size_t width; + size_t height; + float d[]; +} matrix; + +static const matrix T8 = { 8, 8, { + 12, 12, 12, 12, 12, 12, 12, 12, + 16, 15, 9, 4, -4, -9, -15, -16, + 16, 6, -6, -16, -16, -6, 6, 16, + 15, -4, -16, -9, 9, 16, 4, -15, + 12, -12, -12, 12, 12, -12, -12, 12, + 9, -16, 4, 15, -15, -4, 16, -9, + 6, -16, 16, -6, -6, 16, -16, 6, + 4, -9, 15, -16, 16, -15, 9, -4 +} }; + +static const matrix T4 = { 4, 4, { + 17, 17, 17, 17, + 22, 10, -10, -22, + 17, -17, -17, 17, + 10, -22, 22, -10 +} }; + +static const matrix T8t = { 8, 8, { + 12, 16, 16, 15, 12, 9, 6, 4, + 12, 15, 6, -4, -12, -16, -16, -9, + 12, 9, -6, -16, -12, 4, 16, 15, + 12, 4, -16, -9, 12, 15, -6, -16, + 12, -4, -16, 9, 12, -15, -6, 16, + 12, -9, -6, 16, -12, -4, 16, -15, + 12, -15, 6, 4, -12, 16, -16, 9, + 12, -16, 16, -15, 12, -9, 6, -4 +} }; + +static const matrix T4t = { 4, 4, { + 17, 22, 17, 10, + 17, 10, -17, -22, + 17, -10, -17, 22, + 17, -22, 17, -10 +} }; + +static matrix *new_matrix(size_t width, size_t height) +{ + matrix *out = av_mallocz(sizeof (matrix) + height * width * sizeof (float)); + if (out == NULL) { + fprintf(stderr, "Memory allocation failure\n"); + exit(EXIT_FAILURE); + } + out->width = width; + out->height = height; + return out; +} + +static matrix *multiply(const matrix *a, const matrix *b) +{ + matrix *out; + if (a->width != b->height) { + fprintf(stderr, "Incompatible multiplication\n"); + exit(EXIT_FAILURE); + } + out = new_matrix(b->width, a->height); + for (int j = 0; j < out->height; ++j) + for (int i = 0; i < out->width; ++i) { + float sum = 0; + for (int k = 0; k < a->width; ++k) + sum += a->d[j * a->width + k] * b->d[k * b->width + i]; + out->d[j * out->width + i] = sum; + } + return out; +} + +static void normalise(matrix *a) +{ + for (int j = 0; j < a->height; ++j) + for (int i = 0; i < a->width; ++i) { + float *p = a->d + j * a->width + i; + *p *= 64; + if (a->height == 4) + *p /= (const unsigned[]) { 289, 292, 289, 292 } [j]; + else + *p /= (const unsigned[]) { 288, 289, 292, 289, 288, 289, 292, 289 } [j]; + if (a->width == 4) + *p /= (const unsigned[]) { 289, 292, 289, 292 } [i]; + else + *p /= (const unsigned[]) { 288, 289, 292, 289, 288, 289, 292, 289 } [i]; + } +} + +static void divide_and_round_nearest(matrix *a, float by) +{ + for (int j = 0; j < a->height; ++j) + for (int i = 0; i < a->width; ++i) { + float *p = a->d + j * a->width + i; + *p = rintf(*p / by); + } +} + +static void tweak(matrix *a) +{ + for (int j = 4; j < a->height; ++j) + for (int i = 0; i < a->width; ++i) { + float *p = a->d + j * a->width + i; + *p += 1; + } +} + +/* The VC-1 spec places restrictions on the values permitted at three + * different stages: + * - D: the input coefficients in frequency domain + * - E: the intermediate coefficients, inverse-transformed only horizontally + * - R: the fully inverse-transformed coefficients + * + * To fully cater for the ranges specified requires various intermediate + * values to be held to 17-bit precision; yet these conditions do not appear + * to be utilised in real-world streams. At least some assembly + * implementations have chosen to restrict these values to 16-bit precision, + * to accelerate the decoding of real-world streams at the cost of strict + * adherence to the spec. To avoid our test marking these as failures, + * reduce our random inputs. + */ +#define ATTENUATION 4 + +static matrix *generate_inverse_quantized_transform_coefficients(size_t width, size_t height) +{ + matrix *raw, *tmp, *D, *E, *R; + raw = new_matrix(width, height); + for (int i = 0; i < width * height; ++i) + raw->d[i] = (int) (rnd() % (1024/ATTENUATION)) - 512/ATTENUATION; + tmp = multiply(height == 8 ? &T8 : &T4, raw); + D = multiply(tmp, width == 8 ? &T8t : &T4t); + normalise(D); + divide_and_round_nearest(D, 1); + for (int i = 0; i < width * height; ++i) { + if (D->d[i] < -2048/ATTENUATION || D->d[i] > 2048/ATTENUATION-1) { + /* Rare, so simply try again */ + av_free(raw); + av_free(tmp); + av_free(D); + return generate_inverse_quantized_transform_coefficients(width, height); + } + } + E = multiply(D, width == 8 ? &T8 : &T4); + divide_and_round_nearest(E, 8); + for (int i = 0; i < width * height; ++i) + if (E->d[i] < -4096/ATTENUATION || E->d[i] > 4096/ATTENUATION-1) { + /* Rare, so simply try again */ + av_free(raw); + av_free(tmp); + av_free(D); + av_free(E); + return generate_inverse_quantized_transform_coefficients(width, height); + } + R = multiply(height == 8 ? &T8t : &T4t, E); + tweak(R); + divide_and_round_nearest(R, 128); + for (int i = 0; i < width * height; ++i) + if (R->d[i] < -512/ATTENUATION || R->d[i] > 512/ATTENUATION-1) { + /* Rare, so simply try again */ + av_free(raw); + av_free(tmp); + av_free(D); + av_free(E); + av_free(R); + return generate_inverse_quantized_transform_coefficients(width, height); + } + av_free(raw); + av_free(tmp); + av_free(E); + av_free(R); + return D; +} + +#define RANDOMIZE_BUFFER16(name, size) \ + do { \ + int i; \ + for (i = 0; i < size; ++i) { \ + uint16_t r = rnd(); \ + AV_WN16A(name##0 + i, r); \ + AV_WN16A(name##1 + i, r); \ + } \ + } while (0) + +#define RANDOMIZE_BUFFER8(name, size) \ + do { \ + int i; \ + for (i = 0; i < size; ++i) { \ + uint8_t r = rnd(); \ + name##0[i] = r; \ + name##1[i] = r; \ + } \ + } while (0) + #define RANDOMIZE_BUFFER8_MID_WEIGHTED(name, size) \ do { \ uint8_t *p##0 = name##0, *p##1 = name##1; \ @@ -49,6 +245,89 @@ typedef struct { } \ } while (0) +static void check_inv_trans_inplace(void) +{ + /* Inverse transform input coefficients are stored in a 16-bit buffer + * with row stride of 8 coefficients irrespective of transform size. + * vc1_inv_trans_8x8 differs from the others in two ways: coefficients + * are stored in column-major order, and the outputs are written back + * to the input buffer, so we oversize it slightly to catch overruns. */ + LOCAL_ALIGNED_16(int16_t, inv_trans_in0, [10 * 8]); + LOCAL_ALIGNED_16(int16_t, inv_trans_in1, [10 * 8]); + + VC1DSPContext h; + + ff_vc1dsp_init(&h); + + if (check_func(h.vc1_inv_trans_8x8, "vc1dsp.vc1_inv_trans_8x8")) { + matrix *coeffs; + declare_func_emms(AV_CPU_FLAG_MMX, void, int16_t *); + RANDOMIZE_BUFFER16(inv_trans_in, 10 * 8); + coeffs = generate_inverse_quantized_transform_coefficients(8, 8); + for (int j = 0; j < 8; ++j) + for (int i = 0; i < 8; ++i) { + int idx = 8 + i * 8 + j; + inv_trans_in1[idx] = inv_trans_in0[idx] = coeffs->d[j * 8 + i]; + } + call_ref(inv_trans_in0 + 8); + call_new(inv_trans_in1 + 8); + if (memcmp(inv_trans_in0, inv_trans_in1, 10 * 8 * sizeof (int16_t))) + fail(); + bench_new(inv_trans_in1 + 8); + av_free(coeffs); + } +} + +static void check_inv_trans_adding(void) +{ + /* Inverse transform input coefficients are stored in a 16-bit buffer + * with row stride of 8 coefficients irrespective of transform size. */ + LOCAL_ALIGNED_16(int16_t, inv_trans_in0, [8 * 8]); + LOCAL_ALIGNED_16(int16_t, inv_trans_in1, [8 * 8]); + + /* For all but vc1_inv_trans_8x8, the inverse transform is narrowed and + * added with saturation to an array of unsigned 8-bit values. Oversize + * this by 8 samples left and right and one row above and below. */ + LOCAL_ALIGNED_8(uint8_t, inv_trans_out0, [10 * 24]); + LOCAL_ALIGNED_8(uint8_t, inv_trans_out1, [10 * 24]); + + VC1DSPContext h; + + const test tests[] = { + VC1DSP_SIZED_TEST(vc1_inv_trans_8x4, 8, 4) + VC1DSP_SIZED_TEST(vc1_inv_trans_4x8, 4, 8) + VC1DSP_SIZED_TEST(vc1_inv_trans_4x4, 4, 4) + VC1DSP_SIZED_TEST(vc1_inv_trans_8x8_dc, 8, 8) + VC1DSP_SIZED_TEST(vc1_inv_trans_8x4_dc, 8, 4) + VC1DSP_SIZED_TEST(vc1_inv_trans_4x8_dc, 4, 8) + VC1DSP_SIZED_TEST(vc1_inv_trans_4x4_dc, 4, 4) + }; + + ff_vc1dsp_init(&h); + + for (size_t t = 0; t < FF_ARRAY_ELEMS(tests); ++t) { + void (*func)(uint8_t *, ptrdiff_t, int16_t *) = *(void **)((intptr_t) &h + tests[t].offset); + if (check_func(func, "vc1dsp.%s", tests[t].name)) { + matrix *coeffs; + declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *, ptrdiff_t, int16_t *); + RANDOMIZE_BUFFER16(inv_trans_in, 8 * 8); + RANDOMIZE_BUFFER8(inv_trans_out, 10 * 24); + coeffs = generate_inverse_quantized_transform_coefficients(tests[t].width, tests[t].height); + for (int j = 0; j < tests[t].height; ++j) + for (int i = 0; i < tests[t].width; ++i) { + int idx = j * 8 + i; + inv_trans_in1[idx] = inv_trans_in0[idx] = coeffs->d[j * tests[t].width + i]; + } + call_ref(inv_trans_out0 + 24 + 8, 24, inv_trans_in0); + call_new(inv_trans_out1 + 24 + 8, 24, inv_trans_in1); + if (memcmp(inv_trans_out0, inv_trans_out1, 10 * 24)) + fail(); + bench_new(inv_trans_out1 + 24 + 8, 24, inv_trans_in1 + 8); + av_free(coeffs); + } + } +} + static void check_loop_filter(void) { /* Deblocking filter buffers are big enough to hold a 16x16 block, @@ -97,6 +376,10 @@ static void check_loop_filter(void) void checkasm_check_vc1dsp(void) { + check_inv_trans_inplace(); + check_inv_trans_adding(); + report("inv_trans"); + check_loop_filter(); report("loop_filter"); } From patchwork Thu Mar 31 17:23:44 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35108 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp229982pzb; Thu, 31 Mar 2022 10:24:58 -0700 (PDT) X-Google-Smtp-Source: ABdhPJxtJLJX+65fm1KwQeZL3zEOzHu5mIL37EzXttbyoGJs8+6ASSi6gwcaTX79Agm3vzdUO5l2 X-Received: by 2002:a17:906:57c1:b0:6d6:da73:e9c0 with SMTP id u1-20020a17090657c100b006d6da73e9c0mr6059023ejr.45.1648747498748; Thu, 31 Mar 2022 10:24:58 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747498; cv=none; d=google.com; s=arc-20160816; b=LSj84P1bh7c8CDlCZsJGzbDKfnWEm6nUmDl/VSBWHZu26v9+YmbMHScbiJ3E8ilYcT YNJhlz0Cm5ZsW7Z6rhRMxKIPKoZv582Bw6mKoHYDqOr4Ho/GbMF0x5hrUrq4ib8mf/0k IaWgH02z5KIrHMPIeJroyUbJC3XoasR42kRsDYL5ZmVj6oFf/gzFAVprulxH+CgU6LBu qzk4pyFU1+ZaIF3KkG8IM4ZHz1xfd7OTRrv2AQLoICmzQ0HeSTS8UzcaiDnIFo1Q5SF6 0Aew9yJ28GCBFeGlsWO57BYSTst3yLx5Cv3y5Gru5x8zFd9AFSjU12OB5VKInpvAOcyc 2OxA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=eqbzVNticPeSRrCcnKKOJJZwZJI7v/SjGPwOsCyQXRM=; b=MnsIRU07AjMHoeH41Ev/dVHCBjL11VJIenpxvg95F7NMqoFv08MFyMMt16HLfi2lGr womSIpdN+fkzvgePlgX1sIPyEXtjGv8CA24wWZrppo/GxlyZvNSHo0cxcZRy2csDZdiZ rv4VVPKJojNL7TPaFl9uOwK/nw9KLJOiDmnvVTYlEsHGBjKRvkwdnsAuoJAOff8q9Hc5 u/iNGIUqv/ww7EaI9rKQNjTXuCtg5nDvXl6UrI738LtDA91b/8cbONFN9ARpP+wa+Pei SkKwkXUXz2PpNgYGMFB1U6FSPAXGemKJBOXGwUbvq8atwG7xuVDaW9S7zBuQU+HDybQF s9yg== 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 y13-20020a056402270d00b00418c2b5bd33si239969edd.21.2022.03.31.10.24.57; Thu, 31 Mar 2022 10:24:58 -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 8F49668B281; Thu, 31 Mar 2022 20:24:17 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail149082.authsmtp.co.uk (outmail149082.authsmtp.co.uk [62.13.149.82]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 0957568B253 for ; Thu, 31 Mar 2022 20:24:09 +0300 (EEST) Received: from punt22.authsmtp.com (punt22.authsmtp.com [62.13.128.207]) by punt17.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHO9Xf021531 for ; Thu, 31 Mar 2022 18:24:09 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt22.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHO9Fw034634; Thu, 31 Mar 2022 18:24:09 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHO7IT062317 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:07 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:07 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:44 +0100 Message-Id: <20220331172351.550818-4-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 5f75cecd-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQpZIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLR4LEwVAen1x YghjXXheXAp/I0F4 ERsHQXBUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdW0Eb30N d0kGLV8ZQk8QG2x0 bhwOVTQjFlEJXD17 JhAqbUYBAAMJM1U/ NUBkUlYRNgcfBkVC ElpRG0cA X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 03/10] checkasm: Add idctdsp add/put-pixels-clamped tests 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: 4BiSSe5W8EGO Signed-off-by: Ben Avison --- tests/checkasm/Makefile | 1 + tests/checkasm/checkasm.c | 3 ++ tests/checkasm/checkasm.h | 1 + tests/checkasm/idctdsp.c | 98 +++++++++++++++++++++++++++++++++++++++ tests/fate/checkasm.mak | 1 + 5 files changed, 104 insertions(+) create mode 100644 tests/checkasm/idctdsp.c diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile index 7133a6ee66..f6b1008855 100644 --- a/tests/checkasm/Makefile +++ b/tests/checkasm/Makefile @@ -9,6 +9,7 @@ AVCODECOBJS-$(CONFIG_G722DSP) += g722dsp.o AVCODECOBJS-$(CONFIG_H264DSP) += h264dsp.o AVCODECOBJS-$(CONFIG_H264PRED) += h264pred.o AVCODECOBJS-$(CONFIG_H264QPEL) += h264qpel.o +AVCODECOBJS-$(CONFIG_IDCTDSP) += idctdsp.o AVCODECOBJS-$(CONFIG_LLVIDDSP) += llviddsp.o AVCODECOBJS-$(CONFIG_LLVIDENCDSP) += llviddspenc.o AVCODECOBJS-$(CONFIG_VC1DSP) += vc1dsp.o diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c index c2efd81b6d..57134f96ea 100644 --- a/tests/checkasm/checkasm.c +++ b/tests/checkasm/checkasm.c @@ -123,6 +123,9 @@ static const struct { #if CONFIG_HUFFYUV_DECODER { "huffyuvdsp", checkasm_check_huffyuvdsp }, #endif + #if CONFIG_IDCTDSP + { "idctdsp", checkasm_check_idctdsp }, + #endif #if CONFIG_JPEG2000_DECODER { "jpeg2000dsp", checkasm_check_jpeg2000dsp }, #endif diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h index 52ab18a5b1..a86db140e3 100644 --- a/tests/checkasm/checkasm.h +++ b/tests/checkasm/checkasm.h @@ -64,6 +64,7 @@ void checkasm_check_hevc_idct(void); void checkasm_check_hevc_pel(void); void checkasm_check_hevc_sao(void); void checkasm_check_huffyuvdsp(void); +void checkasm_check_idctdsp(void); void checkasm_check_jpeg2000dsp(void); void checkasm_check_llviddsp(void); void checkasm_check_llviddspenc(void); diff --git a/tests/checkasm/idctdsp.c b/tests/checkasm/idctdsp.c new file mode 100644 index 0000000000..02724536a7 --- /dev/null +++ b/tests/checkasm/idctdsp.c @@ -0,0 +1,98 @@ +/* + * Copyright (c) 2022 Ben Avison + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 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 General Public License for more details. + * + * You should have received a copy of the GNU 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 + +#include "checkasm.h" + +#include "libavcodec/idctdsp.h" + +#include "libavutil/common.h" +#include "libavutil/internal.h" +#include "libavutil/intreadwrite.h" +#include "libavutil/mem_internal.h" + +#define IDCTDSP_TEST(func) { #func, offsetof(IDCTDSPContext, func) }, + +typedef struct { + const char *name; + size_t offset; +} test; + +#define RANDOMIZE_BUFFER16(name, size) \ + do { \ + int i; \ + for (i = 0; i < size; ++i) { \ + uint16_t r = rnd() % 0x201 - 0x100; \ + AV_WN16A(name##0 + i, r); \ + AV_WN16A(name##1 + i, r); \ + } \ + } while (0) + +#define RANDOMIZE_BUFFER8(name, size) \ + do { \ + int i; \ + for (i = 0; i < size; ++i) { \ + uint8_t r = rnd(); \ + name##0[i] = r; \ + name##1[i] = r; \ + } \ + } while (0) + +static void check_add_put_clamped(void) +{ + /* Source buffers are only as big as needed, since any over-read won't affect results */ + LOCAL_ALIGNED_16(int16_t, src0, [64]); + LOCAL_ALIGNED_16(int16_t, src1, [64]); + /* Destination buffers have borders of one row above/below and 8 columns left/right to catch overflows */ + LOCAL_ALIGNED_8(uint8_t, dst0, [10 * 24]); + LOCAL_ALIGNED_8(uint8_t, dst1, [10 * 24]); + + AVCodecContext avctx = { 0 }; + IDCTDSPContext h; + + const test tests[] = { + IDCTDSP_TEST(add_pixels_clamped) + IDCTDSP_TEST(put_pixels_clamped) + IDCTDSP_TEST(put_signed_pixels_clamped) + }; + + ff_idctdsp_init(&h, &avctx); + + for (size_t t = 0; t < FF_ARRAY_ELEMS(tests); ++t) { + void (*func)(const int16_t *, uint8_t * ptrdiff_t) = *(void **)((intptr_t) &h + tests[t].offset); + if (check_func(func, "idctdsp.%s", tests[t].name)) { + declare_func_emms(AV_CPU_FLAG_MMX, void, const int16_t *, uint8_t *, ptrdiff_t); + RANDOMIZE_BUFFER16(src, 64); + RANDOMIZE_BUFFER8(dst, 10 * 24); + call_ref(src0, dst0 + 24 + 8, 24); + call_new(src1, dst1 + 24 + 8, 24); + if (memcmp(dst0, dst1, 10 * 24)) + fail(); + bench_new(src1, dst1 + 24 + 8, 24); + } + } +} + +void checkasm_check_idctdsp(void) +{ + check_add_put_clamped(); + report("idctdsp"); +} diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak index 99e6bb13c4..c6273db183 100644 --- a/tests/fate/checkasm.mak +++ b/tests/fate/checkasm.mak @@ -19,6 +19,7 @@ FATE_CHECKASM = fate-checkasm-aacpsdsp \ fate-checkasm-hevc_pel \ fate-checkasm-hevc_sao \ fate-checkasm-huffyuvdsp \ + fate-checkasm-idctdsp \ fate-checkasm-jpeg2000dsp \ fate-checkasm-llviddsp \ fate-checkasm-llviddspenc \ From patchwork Thu Mar 31 17:23:45 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35110 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp230164pzb; Thu, 31 Mar 2022 10:25:22 -0700 (PDT) X-Google-Smtp-Source: ABdhPJx2xRpo7fN6a31F7OE0TT7DjGQbiwHt3l3o6W4bmLfGH4XGQPD95G4MwCAXgLG+shLJ5glF X-Received: by 2002:a17:907:7899:b0:6e0:f285:a860 with SMTP id ku25-20020a170907789900b006e0f285a860mr5817640ejc.261.1648747522749; Thu, 31 Mar 2022 10:25:22 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747522; cv=none; d=google.com; s=arc-20160816; b=ipR01UViBYIg0D9xPJfzR9dxUqP03/YxlC3JIAp3FhYOroyv0SbhPhTmvRg5GQOquK mj8fvGRWEuQ3HkMrpCYJOtyqRGFTFtMxttoYyTiTWRcuwYFC2Ne1Z7N681VsW3UCEX0+ IuAirzSCuv/8QshZ7tx51F4vEai95/SZ67FeFL5B6oAfZNxeOg3E50+hCCLlVz7wh9F2 8glRji14yEdPW/Jz6t0IejYC+atsHRv1zT36ebEFOJozb03jTx20X4VLcMftGCesJzQo upi4A2Xrn/LHVId/3TupP32vUIHJ0yIwcRE42sjbHhBR5wg2aKaj+z6pmU0BjPdEtW9U LmEQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=LUYrzAgwJZmPnTLYTgD7nju71kfIH7ifhD16nQW1Mso=; b=SmzOLR5ALWA4wH3AvVxaCRjFuieTdOMQMnHiSdOCO6WEadclYxgi+z7pUipYOam1mU wF0v7qE008z8h9Q+v3qkA+prnv4fmXpVTtYX3qj8mME+/pb/zbi+Yu1BKuXphtXemHhm wAJL4x83Uyw3aKtjMeC+8syBD3/AG5aBG2O7d+bLQGzTkruVeFKZHmjq3JwhIqfAlnc4 kSAWvdzKgfyvF1ZfJfAel1slNwM7ouv1Cy6Wt7CZruzvLIH/vmfIaK1EGWExLwakk//S EJtUq2cyTPo/zpks7+o9EIYMyI+DNPIgW1j7wnw4gicdalm1UVokDxhYRunER/O55FmO P6Vg== 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 f13-20020a170906824d00b006df76385f4csi53995ejx.1004.2022.03.31.10.25.21; Thu, 31 Mar 2022 10:25:22 -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 8EDB468B2AE; Thu, 31 Mar 2022 20:24:21 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail148106.authsmtp.co.uk (outmail148106.authsmtp.co.uk [62.13.148.106]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id D5F9168B281 for ; Thu, 31 Mar 2022 20:24:11 +0300 (EEST) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt16.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOB6i011191; Thu, 31 Mar 2022 18:24:11 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHO80e062360 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:09 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:08 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:45 +0100 Message-Id: <20220331172351.550818-5-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 607f0363-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQtRIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLRFyIgVAen12 ZQhjV3lfWgp8dUEr Qk1SFnBUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdW0Db30N d0kEM1kVTUsAWSA3 HkJKPDMzB0oJWi4+ ZxIvMUJUBE8NMg08 NkFpRFQVKBQbEgxY EAlHATNCO0MMTi9j Bx5UU0MEeAAA X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthSMTP-Origin: 51.9.63.237/2525 X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 04/10] avcodec/vc1: Introduce fast path for unescaping bitstream buffer 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: YEGCWLElPmTr Includes a checkasm test. Signed-off-by: Ben Avison --- libavcodec/vc1dec.c | 20 ++++++------ libavcodec/vc1dsp.c | 2 ++ libavcodec/vc1dsp.h | 3 ++ tests/checkasm/vc1dsp.c | 67 +++++++++++++++++++++++++++++++++++++++++ 4 files changed, 82 insertions(+), 10 deletions(-) diff --git a/libavcodec/vc1dec.c b/libavcodec/vc1dec.c index e279ffd1c1..0426e8a752 100644 --- a/libavcodec/vc1dec.c +++ b/libavcodec/vc1dec.c @@ -491,7 +491,7 @@ static av_cold int vc1_decode_init(AVCodecContext *avctx) size = next - start - 4; if (size <= 0) continue; - buf2_size = vc1_unescape_buffer(start + 4, size, buf2); + buf2_size = v->vc1dsp.vc1_unescape_buffer(start + 4, size, buf2); init_get_bits(&gb, buf2, buf2_size * 8); switch (AV_RB32(start)) { case VC1_CODE_SEQHDR: @@ -681,7 +681,7 @@ static int vc1_decode_frame(AVCodecContext *avctx, void *data, case VC1_CODE_FRAME: if (avctx->hwaccel) buf_start = start; - buf_size2 = vc1_unescape_buffer(start + 4, size, buf2); + buf_size2 = v->vc1dsp.vc1_unescape_buffer(start + 4, size, buf2); break; case VC1_CODE_FIELD: { int buf_size3; @@ -698,8 +698,8 @@ static int vc1_decode_frame(AVCodecContext *avctx, void *data, ret = AVERROR(ENOMEM); goto err; } - buf_size3 = vc1_unescape_buffer(start + 4, size, - slices[n_slices].buf); + buf_size3 = v->vc1dsp.vc1_unescape_buffer(start + 4, size, + slices[n_slices].buf); init_get_bits(&slices[n_slices].gb, slices[n_slices].buf, buf_size3 << 3); slices[n_slices].mby_start = avctx->coded_height + 31 >> 5; @@ -710,7 +710,7 @@ static int vc1_decode_frame(AVCodecContext *avctx, void *data, break; } case VC1_CODE_ENTRYPOINT: /* it should be before frame data */ - buf_size2 = vc1_unescape_buffer(start + 4, size, buf2); + buf_size2 = v->vc1dsp.vc1_unescape_buffer(start + 4, size, buf2); init_get_bits(&s->gb, buf2, buf_size2 * 8); ff_vc1_decode_entry_point(avctx, v, &s->gb); break; @@ -727,8 +727,8 @@ static int vc1_decode_frame(AVCodecContext *avctx, void *data, ret = AVERROR(ENOMEM); goto err; } - buf_size3 = vc1_unescape_buffer(start + 4, size, - slices[n_slices].buf); + buf_size3 = v->vc1dsp.vc1_unescape_buffer(start + 4, size, + slices[n_slices].buf); init_get_bits(&slices[n_slices].gb, slices[n_slices].buf, buf_size3 << 3); slices[n_slices].mby_start = get_bits(&slices[n_slices].gb, 9); @@ -762,7 +762,7 @@ static int vc1_decode_frame(AVCodecContext *avctx, void *data, ret = AVERROR(ENOMEM); goto err; } - buf_size3 = vc1_unescape_buffer(divider + 4, buf + buf_size - divider - 4, slices[n_slices].buf); + buf_size3 = v->vc1dsp.vc1_unescape_buffer(divider + 4, buf + buf_size - divider - 4, slices[n_slices].buf); init_get_bits(&slices[n_slices].gb, slices[n_slices].buf, buf_size3 << 3); slices[n_slices].mby_start = s->mb_height + 1 >> 1; @@ -771,9 +771,9 @@ static int vc1_decode_frame(AVCodecContext *avctx, void *data, n_slices1 = n_slices - 1; n_slices++; } - buf_size2 = vc1_unescape_buffer(buf, divider - buf, buf2); + buf_size2 = v->vc1dsp.vc1_unescape_buffer(buf, divider - buf, buf2); } else { - buf_size2 = vc1_unescape_buffer(buf, buf_size, buf2); + buf_size2 = v->vc1dsp.vc1_unescape_buffer(buf, buf_size, buf2); } init_get_bits(&s->gb, buf2, buf_size2*8); } else{ diff --git a/libavcodec/vc1dsp.c b/libavcodec/vc1dsp.c index f651d7d461..f1b7bb2397 100644 --- a/libavcodec/vc1dsp.c +++ b/libavcodec/vc1dsp.c @@ -34,6 +34,7 @@ #include "rnd_avg.h" #include "vc1dsp.h" #include "startcode.h" +#include "vc1_common.h" /* Apply overlap transform to horizontal edge */ static void vc1_v_overlap_c(uint8_t *src, ptrdiff_t stride) @@ -1030,6 +1031,7 @@ av_cold void ff_vc1dsp_init(VC1DSPContext *dsp) #endif /* CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER */ dsp->startcode_find_candidate = ff_startcode_find_candidate_c; + dsp->vc1_unescape_buffer = vc1_unescape_buffer; if (ARCH_AARCH64) ff_vc1dsp_init_aarch64(dsp); diff --git a/libavcodec/vc1dsp.h b/libavcodec/vc1dsp.h index fe60025a2a..7ed1776ca7 100644 --- a/libavcodec/vc1dsp.h +++ b/libavcodec/vc1dsp.h @@ -80,6 +80,9 @@ typedef struct VC1DSPContext { * one or more further zero bytes and a one byte. */ int (*startcode_find_candidate)(const uint8_t *buf, int size); + + /* Copy a buffer, removing startcode emulation escape bytes as we go */ + int (*vc1_unescape_buffer)(const uint8_t *src, int size, uint8_t *dst); } VC1DSPContext; void ff_vc1dsp_init(VC1DSPContext* c); diff --git a/tests/checkasm/vc1dsp.c b/tests/checkasm/vc1dsp.c index 7d4457306f..52628d15e4 100644 --- a/tests/checkasm/vc1dsp.c +++ b/tests/checkasm/vc1dsp.c @@ -374,6 +374,70 @@ static void check_loop_filter(void) } } +#define TEST_UNESCAPE \ + do { \ + for (int count = 100; count > 0; --count) { \ + escaped_offset = rnd() & 7; \ + unescaped_offset = rnd() & 7; \ + escaped_len = (1u << (rnd() % 8) + 3) - (rnd() & 7); \ + RANDOMIZE_BUFFER8(unescaped, UNESCAPE_BUF_SIZE); \ + len0 = call_ref(escaped0 + escaped_offset, escaped_len, unescaped0 + unescaped_offset); \ + len1 = call_new(escaped1 + escaped_offset, escaped_len, unescaped1 + unescaped_offset); \ + if (len0 != len1 || memcmp(unescaped0, unescaped1, UNESCAPE_BUF_SIZE)) \ + fail(); \ + } \ + } while (0) + +static void check_unescape(void) +{ + /* This appears to be a typical length of buffer in use */ +#define LOG2_UNESCAPE_BUF_SIZE 17 +#define UNESCAPE_BUF_SIZE (1u< X-Patchwork-Id: 35107 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp229904pzb; Thu, 31 Mar 2022 10:24:46 -0700 (PDT) X-Google-Smtp-Source: ABdhPJzsKwwVK/yZSQugCg2chf2ELVEK0arCGg+Et6LBxWMOtC5Vird6aqigaYLRE9CSet37GlsK X-Received: by 2002:a17:907:97c8:b0:6db:6c57:64ee with SMTP id js8-20020a17090797c800b006db6c5764eemr6033434ejc.320.1648747485843; Thu, 31 Mar 2022 10:24:45 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747485; cv=none; d=google.com; s=arc-20160816; b=MxekxebCxYjVOYmr+0RasylMoNSQbZArGCXCo06OYs/lAU05UxAxraDlG9XCOiROnx R3bEq1THRzUIbfGrwuu79QvTSNwVa24giPsB6/dTYQDjhsOQG9V4fZylvVuDY6NA+TM6 m3kdOZv3l288vy9hl296CCqBXWN3uPdrfqZBAdQD0L3mgEWpgU2S+NEprZDrrryZDiwX AehUa9al55IiSLmMfwmDQCNYO+M9KtOWNksDU68ol/giqbO06vxiOts0TPpd4DWg0iZy 8wtEAVdq72pLkl3G6+TVbXvIU8APsraQfWEa5yYxxTwDGoScvr7HXqf8ofqem9oMgqX9 vOBw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=FWN+Sser/VeF4x6WQBfGvidyHfM8rqROzf6x4fnQehU=; b=Oknu/pZbsgxM0o4giiv1RFKzP6tPm4h6Yh3VLx/OpFNct/oC6Ff6bAIZ8qeRKbg5Z7 H29TDnHBhfIvNbd9RspOOqfBCz4i7uA0GLdktO2f900+J+MxY3t3gWIRUv3BZjuWp/Br 0TQON0Lffx9dPF6IV2Atm+NtCYT3Ux8H/2s4UCTw1MBWDZNhYTXDDAva0P5XpKwYRkdb ycizI37aZdH/Zen2Crv6mEBefq6DadqTcuJHIWe197hZ6LvvhBPymNlmIUdiZNi1+Obx qAbMAbVGdZp2KnA9T8ciVKbIDzdQwiV+Laanz1dlnwMAxIkBj+HLKFgIhAIw94QqsWd7 Giog== 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 n22-20020aa7db56000000b00418f6de8a7csi111710edt.444.2022.03.31.10.24.45; Thu, 31 Mar 2022 10:24:45 -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 B263068B27B; Thu, 31 Mar 2022 20:24:16 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail149078.authsmtp.net (outmail149078.authsmtp.net [62.13.149.78]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 9C52568B24C for ; Thu, 31 Mar 2022 20:24:14 +0300 (EEST) Received: from punt18.authsmtp.com (punt18.authsmtp.com [62.13.128.225]) by punt15.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOD5x019712 for ; Thu, 31 Mar 2022 18:24:13 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt18.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHODKA097028; Thu, 31 Mar 2022 18:24:13 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHOA9r062402 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:11 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:10 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:46 +0100 Message-Id: <20220331172351.550818-6-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 61df63ff-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQtTIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLWh7NQVAenpw ZwhjXnBYWUA0c0cp FEhXQzsbN2Zhb30e URULalJUJgFXdhtG bll4VXJZN2cPKBEA NjMYH3gvZn0HdWMM GjRFJEwZRkoGFXki TElQVRw1GAVbG2A5 LgBuDHM7Og4dP082 NlAiWFQXexETDhFT BQlDCTRFb0EIWyow ZQAA X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 05/10] avcodec/vc1: Arm 64-bit NEON deblocking filter fast paths 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: 8TDKeJAERr2c checkasm benchmarks on 1.5 GHz Cortex-A72 are as follows. Note that the C version can still outperform the NEON version in specific cases. The balance between different code paths is stream-dependent, but in practice the best case happens about 5% of the time, the worst case happens about 40% of the time, and the complexity of the remaining cases fall somewhere in between. Therefore, taking the average of the best and worst case timings is probably a conservative estimate of the degree by which the NEON code improves performance. vc1dsp.vc1_h_loop_filter4_bestcase_c: 10.7 vc1dsp.vc1_h_loop_filter4_bestcase_neon: 43.5 vc1dsp.vc1_h_loop_filter4_worstcase_c: 184.5 vc1dsp.vc1_h_loop_filter4_worstcase_neon: 73.7 vc1dsp.vc1_h_loop_filter8_bestcase_c: 31.2 vc1dsp.vc1_h_loop_filter8_bestcase_neon: 62.2 vc1dsp.vc1_h_loop_filter8_worstcase_c: 358.2 vc1dsp.vc1_h_loop_filter8_worstcase_neon: 88.2 vc1dsp.vc1_h_loop_filter16_bestcase_c: 51.0 vc1dsp.vc1_h_loop_filter16_bestcase_neon: 107.7 vc1dsp.vc1_h_loop_filter16_worstcase_c: 722.7 vc1dsp.vc1_h_loop_filter16_worstcase_neon: 140.5 vc1dsp.vc1_v_loop_filter4_bestcase_c: 9.7 vc1dsp.vc1_v_loop_filter4_bestcase_neon: 43.0 vc1dsp.vc1_v_loop_filter4_worstcase_c: 178.7 vc1dsp.vc1_v_loop_filter4_worstcase_neon: 69.0 vc1dsp.vc1_v_loop_filter8_bestcase_c: 30.2 vc1dsp.vc1_v_loop_filter8_bestcase_neon: 50.7 vc1dsp.vc1_v_loop_filter8_worstcase_c: 353.0 vc1dsp.vc1_v_loop_filter8_worstcase_neon: 69.2 vc1dsp.vc1_v_loop_filter16_bestcase_c: 60.0 vc1dsp.vc1_v_loop_filter16_bestcase_neon: 90.0 vc1dsp.vc1_v_loop_filter16_worstcase_c: 714.2 vc1dsp.vc1_v_loop_filter16_worstcase_neon: 97.2 Signed-off-by: Ben Avison --- libavcodec/aarch64/Makefile | 1 + libavcodec/aarch64/vc1dsp_init_aarch64.c | 14 + libavcodec/aarch64/vc1dsp_neon.S | 692 +++++++++++++++++++++++ 3 files changed, 707 insertions(+) create mode 100644 libavcodec/aarch64/vc1dsp_neon.S diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile index 954461f81d..5b25e4dfb9 100644 --- a/libavcodec/aarch64/Makefile +++ b/libavcodec/aarch64/Makefile @@ -48,6 +48,7 @@ NEON-OBJS-$(CONFIG_IDCTDSP) += aarch64/simple_idct_neon.o NEON-OBJS-$(CONFIG_MDCT) += aarch64/mdct_neon.o NEON-OBJS-$(CONFIG_MPEGAUDIODSP) += aarch64/mpegaudiodsp_neon.o NEON-OBJS-$(CONFIG_PIXBLOCKDSP) += aarch64/pixblockdsp_neon.o +NEON-OBJS-$(CONFIG_VC1DSP) += aarch64/vc1dsp_neon.o NEON-OBJS-$(CONFIG_VP8DSP) += aarch64/vp8dsp_neon.o # decoders/encoders diff --git a/libavcodec/aarch64/vc1dsp_init_aarch64.c b/libavcodec/aarch64/vc1dsp_init_aarch64.c index 13dfd74940..8f96e4802d 100644 --- a/libavcodec/aarch64/vc1dsp_init_aarch64.c +++ b/libavcodec/aarch64/vc1dsp_init_aarch64.c @@ -25,6 +25,13 @@ #include "config.h" +void ff_vc1_v_loop_filter4_neon(uint8_t *src, ptrdiff_t stride, int pq); +void ff_vc1_h_loop_filter4_neon(uint8_t *src, ptrdiff_t stride, int pq); +void ff_vc1_v_loop_filter8_neon(uint8_t *src, ptrdiff_t stride, int pq); +void ff_vc1_h_loop_filter8_neon(uint8_t *src, ptrdiff_t stride, int pq); +void ff_vc1_v_loop_filter16_neon(uint8_t *src, ptrdiff_t stride, int pq); +void ff_vc1_h_loop_filter16_neon(uint8_t *src, ptrdiff_t stride, int pq); + void ff_put_vc1_chroma_mc8_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int h, int x, int y); void ff_avg_vc1_chroma_mc8_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride, @@ -39,6 +46,13 @@ av_cold void ff_vc1dsp_init_aarch64(VC1DSPContext *dsp) int cpu_flags = av_get_cpu_flags(); if (have_neon(cpu_flags)) { + dsp->vc1_v_loop_filter4 = ff_vc1_v_loop_filter4_neon; + dsp->vc1_h_loop_filter4 = ff_vc1_h_loop_filter4_neon; + dsp->vc1_v_loop_filter8 = ff_vc1_v_loop_filter8_neon; + dsp->vc1_h_loop_filter8 = ff_vc1_h_loop_filter8_neon; + dsp->vc1_v_loop_filter16 = ff_vc1_v_loop_filter16_neon; + dsp->vc1_h_loop_filter16 = ff_vc1_h_loop_filter16_neon; + dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = ff_put_vc1_chroma_mc8_neon; dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_vc1_chroma_mc8_neon; dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = ff_put_vc1_chroma_mc4_neon; diff --git a/libavcodec/aarch64/vc1dsp_neon.S b/libavcodec/aarch64/vc1dsp_neon.S new file mode 100644 index 0000000000..1ea9fa75ff --- /dev/null +++ b/libavcodec/aarch64/vc1dsp_neon.S @@ -0,0 +1,692 @@ +/* + * VC1 AArch64 NEON optimisations + * + * Copyright (c) 2022 Ben Avison + * + * 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" + +.align 5 +.Lcoeffs: +.quad 0x00050002 + +// VC-1 in-loop deblocking filter for 4 pixel pairs at boundary of vertically-neighbouring blocks +// On entry: +// x0 -> top-left pel of lower block +// x1 = row stride, bytes +// w2 = PQUANT bitstream parameter +function ff_vc1_v_loop_filter4_neon, export=1 + sub x3, x0, w1, sxtw #2 + ldr d0, .Lcoeffs + ld1 {v1.s}[0], [x0], x1 // P5 + ld1 {v2.s}[0], [x3], x1 // P1 + ld1 {v3.s}[0], [x3], x1 // P2 + ld1 {v4.s}[0], [x0], x1 // P6 + ld1 {v5.s}[0], [x3], x1 // P3 + ld1 {v6.s}[0], [x0], x1 // P7 + ld1 {v7.s}[0], [x3] // P4 + ld1 {v16.s}[0], [x0] // P8 + ushll v17.8h, v1.8b, #1 // 2*P5 + dup v18.8h, w2 // pq + ushll v2.8h, v2.8b, #1 // 2*P1 + uxtl v3.8h, v3.8b // P2 + uxtl v4.8h, v4.8b // P6 + uxtl v19.8h, v5.8b // P3 + mls v2.4h, v3.4h, v0.h[1] // 2*P1-5*P2 + uxtl v3.8h, v6.8b // P7 + mls v17.4h, v4.4h, v0.h[1] // 2*P5-5*P6 + ushll v5.8h, v5.8b, #1 // 2*P3 + uxtl v6.8h, v7.8b // P4 + mla v17.4h, v3.4h, v0.h[1] // 2*P5-5*P6+5*P7 + uxtl v3.8h, v16.8b // P8 + mla v2.4h, v19.4h, v0.h[1] // 2*P1-5*P2+5*P3 + uxtl v1.8h, v1.8b // P5 + mls v5.4h, v6.4h, v0.h[1] // 2*P3-5*P4 + mls v17.4h, v3.4h, v0.h[0] // 2*P5-5*P6+5*P7-2*P8 + sub v3.4h, v6.4h, v1.4h // P4-P5 + mls v2.4h, v6.4h, v0.h[0] // 2*P1-5*P2+5*P3-2*P4 + mla v5.4h, v1.4h, v0.h[1] // 2*P3-5*P4+5*P5 + mls v5.4h, v4.4h, v0.h[0] // 2*P3-5*P4+5*P5-2*P6 + abs v4.4h, v3.4h + srshr v7.4h, v17.4h, #3 + srshr v2.4h, v2.4h, #3 + sshr v4.4h, v4.4h, #1 // clip + srshr v5.4h, v5.4h, #3 + abs v7.4h, v7.4h // a2 + sshr v3.4h, v3.4h, #8 // clip_sign + abs v2.4h, v2.4h // a1 + cmeq v16.4h, v4.4h, #0 // test clip == 0 + abs v17.4h, v5.4h // a0 + sshr v5.4h, v5.4h, #8 // a0_sign + cmhs v19.4h, v2.4h, v7.4h // test a1 >= a2 + cmhs v18.4h, v17.4h, v18.4h // test a0 >= pq + sub v3.4h, v3.4h, v5.4h // clip_sign - a0_sign + bsl v19.8b, v7.8b, v2.8b // a3 + orr v2.8b, v16.8b, v18.8b // test clip == 0 || a0 >= pq + uqsub v5.4h, v17.4h, v19.4h // a0 >= a3 ? a0-a3 : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + cmhs v7.4h, v19.4h, v17.4h // test a3 >= a0 + mul v0.4h, v5.4h, v0.h[1] // a0 >= a3 ? 5*(a0-a3) : 0 + orr v5.8b, v2.8b, v7.8b // test clip == 0 || a0 >= pq || a3 >= a0 + mov w0, v5.s[1] // move to gp reg + ushr v0.4h, v0.4h, #3 // a0 >= a3 ? (5*(a0-a3))>>3 : 0 + cmhs v5.4h, v0.4h, v4.4h + tbnz w0, #0, 1f // none of the 4 pixel pairs should be updated if this one is not filtered + bsl v5.8b, v4.8b, v0.8b // FFMIN(d, clip) + bic v0.8b, v5.8b, v2.8b // set each d to zero if it should not be filtered because clip == 0 || a0 >= pq (a3 > a0 case already zeroed by saturating sub) + mls v6.4h, v0.4h, v3.4h // invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P4 + mla v1.4h, v0.4h, v3.4h // invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P5 + sqxtun v0.8b, v6.8h + sqxtun v1.8b, v1.8h + st1 {v0.s}[0], [x3], x1 + st1 {v1.s}[0], [x3] +1: ret +endfunc + +// VC-1 in-loop deblocking filter for 4 pixel pairs at boundary of horizontally-neighbouring blocks +// On entry: +// x0 -> top-left pel of right block +// x1 = row stride, bytes +// w2 = PQUANT bitstream parameter +function ff_vc1_h_loop_filter4_neon, export=1 + sub x3, x0, #4 // where to start reading + ldr d0, .Lcoeffs + ld1 {v1.8b}, [x3], x1 + sub x0, x0, #1 // where to start writing + ld1 {v2.8b}, [x3], x1 + ld1 {v3.8b}, [x3], x1 + ld1 {v4.8b}, [x3] + dup v5.8h, w2 // pq + trn1 v6.8b, v1.8b, v2.8b + trn2 v1.8b, v1.8b, v2.8b + trn1 v2.8b, v3.8b, v4.8b + trn2 v3.8b, v3.8b, v4.8b + trn1 v4.4h, v6.4h, v2.4h // P1, P5 + trn1 v7.4h, v1.4h, v3.4h // P2, P6 + trn2 v2.4h, v6.4h, v2.4h // P3, P7 + trn2 v1.4h, v1.4h, v3.4h // P4, P8 + ushll v3.8h, v4.8b, #1 // 2*P1, 2*P5 + uxtl v6.8h, v7.8b // P2, P6 + uxtl v7.8h, v2.8b // P3, P7 + uxtl v1.8h, v1.8b // P4, P8 + mls v3.8h, v6.8h, v0.h[1] // 2*P1-5*P2, 2*P5-5*P6 + ushll v2.8h, v2.8b, #1 // 2*P3, 2*P7 + uxtl v4.8h, v4.8b // P1, P5 + mla v3.8h, v7.8h, v0.h[1] // 2*P1-5*P2+5*P3, 2*P5-5*P6+5*P7 + mov d6, v6.d[1] // P6 + mls v3.8h, v1.8h, v0.h[0] // 2*P1-5*P2+5*P3-2*P4, 2*P5-5*P6+5*P7-2*P8 + mov d4, v4.d[1] // P5 + mls v2.4h, v1.4h, v0.h[1] // 2*P3-5*P4 + mla v2.4h, v4.4h, v0.h[1] // 2*P3-5*P4+5*P5 + sub v7.4h, v1.4h, v4.4h // P4-P5 + mls v2.4h, v6.4h, v0.h[0] // 2*P3-5*P4+5*P5-2*P6 + srshr v3.8h, v3.8h, #3 + abs v6.4h, v7.4h + sshr v7.4h, v7.4h, #8 // clip_sign + srshr v2.4h, v2.4h, #3 + abs v3.8h, v3.8h // a1, a2 + sshr v6.4h, v6.4h, #1 // clip + mov d16, v3.d[1] // a2 + abs v17.4h, v2.4h // a0 + cmeq v18.4h, v6.4h, #0 // test clip == 0 + sshr v2.4h, v2.4h, #8 // a0_sign + cmhs v19.4h, v3.4h, v16.4h // test a1 >= a2 + cmhs v5.4h, v17.4h, v5.4h // test a0 >= pq + sub v2.4h, v7.4h, v2.4h // clip_sign - a0_sign + bsl v19.8b, v16.8b, v3.8b // a3 + orr v3.8b, v18.8b, v5.8b // test clip == 0 || a0 >= pq + uqsub v5.4h, v17.4h, v19.4h // a0 >= a3 ? a0-a3 : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + cmhs v7.4h, v19.4h, v17.4h // test a3 >= a0 + mul v0.4h, v5.4h, v0.h[1] // a0 >= a3 ? 5*(a0-a3) : 0 + orr v5.8b, v3.8b, v7.8b // test clip == 0 || a0 >= pq || a3 >= a0 + mov w2, v5.s[1] // move to gp reg + ushr v0.4h, v0.4h, #3 // a0 >= a3 ? (5*(a0-a3))>>3 : 0 + cmhs v5.4h, v0.4h, v6.4h + tbnz w2, #0, 1f // none of the 4 pixel pairs should be updated if this one is not filtered + bsl v5.8b, v6.8b, v0.8b // FFMIN(d, clip) + bic v0.8b, v5.8b, v3.8b // set each d to zero if it should not be filtered because clip == 0 || a0 >= pq (a3 > a0 case already zeroed by saturating sub) + mla v4.4h, v0.4h, v2.4h // invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P5 + mls v1.4h, v0.4h, v2.4h // invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P4 + sqxtun v3.8b, v4.8h + sqxtun v2.8b, v1.8h + st2 {v2.b, v3.b}[0], [x0], x1 + st2 {v2.b, v3.b}[1], [x0], x1 + st2 {v2.b, v3.b}[2], [x0], x1 + st2 {v2.b, v3.b}[3], [x0] +1: ret +endfunc + +// VC-1 in-loop deblocking filter for 8 pixel pairs at boundary of vertically-neighbouring blocks +// On entry: +// x0 -> top-left pel of lower block +// x1 = row stride, bytes +// w2 = PQUANT bitstream parameter +function ff_vc1_v_loop_filter8_neon, export=1 + sub x3, x0, w1, sxtw #2 + ldr d0, .Lcoeffs + ld1 {v1.8b}, [x0], x1 // P5 + movi v2.2d, #0x0000ffff00000000 + ld1 {v3.8b}, [x3], x1 // P1 + ld1 {v4.8b}, [x3], x1 // P2 + ld1 {v5.8b}, [x0], x1 // P6 + ld1 {v6.8b}, [x3], x1 // P3 + ld1 {v7.8b}, [x0], x1 // P7 + ushll v16.8h, v1.8b, #1 // 2*P5 + ushll v3.8h, v3.8b, #1 // 2*P1 + ld1 {v17.8b}, [x3] // P4 + uxtl v4.8h, v4.8b // P2 + ld1 {v18.8b}, [x0] // P8 + uxtl v5.8h, v5.8b // P6 + dup v19.8h, w2 // pq + uxtl v20.8h, v6.8b // P3 + mls v3.8h, v4.8h, v0.h[1] // 2*P1-5*P2 + uxtl v4.8h, v7.8b // P7 + ushll v6.8h, v6.8b, #1 // 2*P3 + mls v16.8h, v5.8h, v0.h[1] // 2*P5-5*P6 + uxtl v7.8h, v17.8b // P4 + uxtl v17.8h, v18.8b // P8 + mla v16.8h, v4.8h, v0.h[1] // 2*P5-5*P6+5*P7 + uxtl v1.8h, v1.8b // P5 + mla v3.8h, v20.8h, v0.h[1] // 2*P1-5*P2+5*P3 + sub v4.8h, v7.8h, v1.8h // P4-P5 + mls v6.8h, v7.8h, v0.h[1] // 2*P3-5*P4 + mls v16.8h, v17.8h, v0.h[0] // 2*P5-5*P6+5*P7-2*P8 + abs v17.8h, v4.8h + sshr v4.8h, v4.8h, #8 // clip_sign + mls v3.8h, v7.8h, v0.h[0] // 2*P1-5*P2+5*P3-2*P4 + sshr v17.8h, v17.8h, #1 // clip + mla v6.8h, v1.8h, v0.h[1] // 2*P3-5*P4+5*P5 + srshr v16.8h, v16.8h, #3 + mls v6.8h, v5.8h, v0.h[0] // 2*P3-5*P4+5*P5-2*P6 + cmeq v5.8h, v17.8h, #0 // test clip == 0 + srshr v3.8h, v3.8h, #3 + abs v16.8h, v16.8h // a2 + abs v3.8h, v3.8h // a1 + srshr v6.8h, v6.8h, #3 + cmhs v18.8h, v3.8h, v16.8h // test a1 >= a2 + abs v20.8h, v6.8h // a0 + sshr v6.8h, v6.8h, #8 // a0_sign + bsl v18.16b, v16.16b, v3.16b // a3 + cmhs v3.8h, v20.8h, v19.8h // test a0 >= pq + sub v4.8h, v4.8h, v6.8h // clip_sign - a0_sign + uqsub v6.8h, v20.8h, v18.8h // a0 >= a3 ? a0-a3 : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + cmhs v16.8h, v18.8h, v20.8h // test a3 >= a0 + orr v3.16b, v5.16b, v3.16b // test clip == 0 || a0 >= pq + mul v0.8h, v6.8h, v0.h[1] // a0 >= a3 ? 5*(a0-a3) : 0 + orr v5.16b, v3.16b, v16.16b // test clip == 0 || a0 >= pq || a3 >= a0 + cmtst v2.2d, v5.2d, v2.2d // if 2nd of each group of is not filtered, then none of the others in the group should be either + mov w0, v5.s[1] // move to gp reg + ushr v0.8h, v0.8h, #3 // a0 >= a3 ? (5*(a0-a3))>>3 : 0 + mov w2, v5.s[3] + orr v2.16b, v3.16b, v2.16b + cmhs v3.8h, v0.8h, v17.8h + and w0, w0, w2 + bsl v3.16b, v17.16b, v0.16b // FFMIN(d, clip) + tbnz w0, #0, 1f // none of the 8 pixel pairs should be updated in this case + bic v0.16b, v3.16b, v2.16b // set each d to zero if it should not be filtered + mls v7.8h, v0.8h, v4.8h // invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P4 + mla v1.8h, v0.8h, v4.8h // invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P5 + sqxtun v0.8b, v7.8h + sqxtun v1.8b, v1.8h + st1 {v0.8b}, [x3], x1 + st1 {v1.8b}, [x3] +1: ret +endfunc + +// VC-1 in-loop deblocking filter for 8 pixel pairs at boundary of horizontally-neighbouring blocks +// On entry: +// x0 -> top-left pel of right block +// x1 = row stride, bytes +// w2 = PQUANT bitstream parameter +function ff_vc1_h_loop_filter8_neon, export=1 + sub x3, x0, #4 // where to start reading + ldr d0, .Lcoeffs + ld1 {v1.8b}, [x3], x1 // P1[0], P2[0]... + sub x0, x0, #1 // where to start writing + ld1 {v2.8b}, [x3], x1 + add x4, x0, x1, lsl #2 + ld1 {v3.8b}, [x3], x1 + ld1 {v4.8b}, [x3], x1 + ld1 {v5.8b}, [x3], x1 + ld1 {v6.8b}, [x3], x1 + ld1 {v7.8b}, [x3], x1 + trn1 v16.8b, v1.8b, v2.8b // P1[0], P1[1], P3[0]... + ld1 {v17.8b}, [x3] + trn2 v1.8b, v1.8b, v2.8b // P2[0], P2[1], P4[0]... + trn1 v2.8b, v3.8b, v4.8b // P1[2], P1[3], P3[2]... + trn2 v3.8b, v3.8b, v4.8b // P2[2], P2[3], P4[2]... + dup v4.8h, w2 // pq + trn1 v18.8b, v5.8b, v6.8b // P1[4], P1[5], P3[4]... + trn2 v5.8b, v5.8b, v6.8b // P2[4], P2[5], P4[4]... + trn1 v6.4h, v16.4h, v2.4h // P1[0], P1[1], P1[2], P1[3], P5[0]... + trn1 v19.4h, v1.4h, v3.4h // P2[0], P2[1], P2[2], P2[3], P6[0]... + trn1 v20.8b, v7.8b, v17.8b // P1[6], P1[7], P3[6]... + trn2 v7.8b, v7.8b, v17.8b // P2[6], P2[7], P4[6]... + trn2 v2.4h, v16.4h, v2.4h // P3[0], P3[1], P3[2], P3[3], P7[0]... + trn2 v1.4h, v1.4h, v3.4h // P4[0], P4[1], P4[2], P4[3], P8[0]... + trn1 v3.4h, v18.4h, v20.4h // P1[4], P1[5], P1[6], P1[7], P5[4]... + trn1 v16.4h, v5.4h, v7.4h // P2[4], P2[5], P2[6], P2[7], P6[4]... + trn2 v17.4h, v18.4h, v20.4h // P3[4], P3[5], P3[6], P3[7], P7[4]... + trn2 v5.4h, v5.4h, v7.4h // P4[4], P4[5], P4[6], P4[7], P8[4]... + trn1 v7.2s, v6.2s, v3.2s // P1 + trn1 v18.2s, v19.2s, v16.2s // P2 + trn2 v3.2s, v6.2s, v3.2s // P5 + trn2 v6.2s, v19.2s, v16.2s // P6 + trn1 v16.2s, v2.2s, v17.2s // P3 + trn2 v2.2s, v2.2s, v17.2s // P7 + ushll v7.8h, v7.8b, #1 // 2*P1 + trn1 v17.2s, v1.2s, v5.2s // P4 + ushll v19.8h, v3.8b, #1 // 2*P5 + trn2 v1.2s, v1.2s, v5.2s // P8 + uxtl v5.8h, v18.8b // P2 + uxtl v6.8h, v6.8b // P6 + uxtl v18.8h, v16.8b // P3 + mls v7.8h, v5.8h, v0.h[1] // 2*P1-5*P2 + uxtl v2.8h, v2.8b // P7 + ushll v5.8h, v16.8b, #1 // 2*P3 + mls v19.8h, v6.8h, v0.h[1] // 2*P5-5*P6 + uxtl v16.8h, v17.8b // P4 + uxtl v1.8h, v1.8b // P8 + mla v19.8h, v2.8h, v0.h[1] // 2*P5-5*P6+5*P7 + uxtl v2.8h, v3.8b // P5 + mla v7.8h, v18.8h, v0.h[1] // 2*P1-5*P2+5*P3 + sub v3.8h, v16.8h, v2.8h // P4-P5 + mls v5.8h, v16.8h, v0.h[1] // 2*P3-5*P4 + mls v19.8h, v1.8h, v0.h[0] // 2*P5-5*P6+5*P7-2*P8 + abs v1.8h, v3.8h + sshr v3.8h, v3.8h, #8 // clip_sign + mls v7.8h, v16.8h, v0.h[0] // 2*P1-5*P2+5*P3-2*P4 + sshr v1.8h, v1.8h, #1 // clip + mla v5.8h, v2.8h, v0.h[1] // 2*P3-5*P4+5*P5 + srshr v17.8h, v19.8h, #3 + mls v5.8h, v6.8h, v0.h[0] // 2*P3-5*P4+5*P5-2*P6 + cmeq v6.8h, v1.8h, #0 // test clip == 0 + srshr v7.8h, v7.8h, #3 + abs v17.8h, v17.8h // a2 + abs v7.8h, v7.8h // a1 + srshr v5.8h, v5.8h, #3 + cmhs v18.8h, v7.8h, v17.8h // test a1 >= a2 + abs v19.8h, v5.8h // a0 + sshr v5.8h, v5.8h, #8 // a0_sign + bsl v18.16b, v17.16b, v7.16b // a3 + cmhs v4.8h, v19.8h, v4.8h // test a0 >= pq + sub v3.8h, v3.8h, v5.8h // clip_sign - a0_sign + uqsub v5.8h, v19.8h, v18.8h // a0 >= a3 ? a0-a3 : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + cmhs v7.8h, v18.8h, v19.8h // test a3 >= a0 + orr v4.16b, v6.16b, v4.16b // test clip == 0 || a0 >= pq + mul v0.8h, v5.8h, v0.h[1] // a0 >= a3 ? 5*(a0-a3) : 0 + orr v5.16b, v4.16b, v7.16b // test clip == 0 || a0 >= pq || a3 >= a0 + mov w2, v5.s[1] // move to gp reg + ushr v0.8h, v0.8h, #3 // a0 >= a3 ? (5*(a0-a3))>>3 : 0 + mov w3, v5.s[3] + cmhs v5.8h, v0.8h, v1.8h + and w5, w2, w3 + bsl v5.16b, v1.16b, v0.16b // FFMIN(d, clip) + tbnz w5, #0, 2f // none of the 8 pixel pairs should be updated in this case + bic v0.16b, v5.16b, v4.16b // set each d to zero if it should not be filtered because clip == 0 || a0 >= pq (a3 > a0 case already zeroed by saturating sub) + mla v2.8h, v0.8h, v3.8h // invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P5 + mls v16.8h, v0.8h, v3.8h // invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P4 + sqxtun v1.8b, v2.8h + sqxtun v0.8b, v16.8h + tbnz w2, #0, 1f // none of the first 4 pixel pairs should be updated if so + st2 {v0.b, v1.b}[0], [x0], x1 + st2 {v0.b, v1.b}[1], [x0], x1 + st2 {v0.b, v1.b}[2], [x0], x1 + st2 {v0.b, v1.b}[3], [x0] +1: tbnz w3, #0, 2f // none of the second 4 pixel pairs should be updated if so + st2 {v0.b, v1.b}[4], [x4], x1 + st2 {v0.b, v1.b}[5], [x4], x1 + st2 {v0.b, v1.b}[6], [x4], x1 + st2 {v0.b, v1.b}[7], [x4] +2: ret +endfunc + +// VC-1 in-loop deblocking filter for 16 pixel pairs at boundary of vertically-neighbouring blocks +// On entry: +// x0 -> top-left pel of lower block +// x1 = row stride, bytes +// w2 = PQUANT bitstream parameter +function ff_vc1_v_loop_filter16_neon, export=1 + sub x3, x0, w1, sxtw #2 + ldr d0, .Lcoeffs + ld1 {v1.16b}, [x0], x1 // P5 + movi v2.2d, #0x0000ffff00000000 + ld1 {v3.16b}, [x3], x1 // P1 + ld1 {v4.16b}, [x3], x1 // P2 + ld1 {v5.16b}, [x0], x1 // P6 + ld1 {v6.16b}, [x3], x1 // P3 + ld1 {v7.16b}, [x0], x1 // P7 + ushll v16.8h, v1.8b, #1 // 2*P5[0..7] + ushll v17.8h, v3.8b, #1 // 2*P1[0..7] + ld1 {v18.16b}, [x3] // P4 + uxtl v19.8h, v4.8b // P2[0..7] + ld1 {v20.16b}, [x0] // P8 + uxtl v21.8h, v5.8b // P6[0..7] + dup v22.8h, w2 // pq + ushll2 v3.8h, v3.16b, #1 // 2*P1[8..15] + mls v17.8h, v19.8h, v0.h[1] // 2*P1[0..7]-5*P2[0..7] + ushll2 v19.8h, v1.16b, #1 // 2*P5[8..15] + uxtl2 v4.8h, v4.16b // P2[8..15] + mls v16.8h, v21.8h, v0.h[1] // 2*P5[0..7]-5*P6[0..7] + uxtl2 v5.8h, v5.16b // P6[8..15] + uxtl v23.8h, v6.8b // P3[0..7] + uxtl v24.8h, v7.8b // P7[0..7] + mls v3.8h, v4.8h, v0.h[1] // 2*P1[8..15]-5*P2[8..15] + ushll v4.8h, v6.8b, #1 // 2*P3[0..7] + uxtl v25.8h, v18.8b // P4[0..7] + mls v19.8h, v5.8h, v0.h[1] // 2*P5[8..15]-5*P6[8..15] + uxtl2 v26.8h, v6.16b // P3[8..15] + mla v17.8h, v23.8h, v0.h[1] // 2*P1[0..7]-5*P2[0..7]+5*P3[0..7] + uxtl2 v7.8h, v7.16b // P7[8..15] + ushll2 v6.8h, v6.16b, #1 // 2*P3[8..15] + mla v16.8h, v24.8h, v0.h[1] // 2*P5[0..7]-5*P6[0..7]+5*P7[0..7] + uxtl2 v18.8h, v18.16b // P4[8..15] + uxtl v23.8h, v20.8b // P8[0..7] + mls v4.8h, v25.8h, v0.h[1] // 2*P3[0..7]-5*P4[0..7] + uxtl v24.8h, v1.8b // P5[0..7] + uxtl2 v20.8h, v20.16b // P8[8..15] + mla v3.8h, v26.8h, v0.h[1] // 2*P1[8..15]-5*P2[8..15]+5*P3[8..15] + uxtl2 v1.8h, v1.16b // P5[8..15] + sub v26.8h, v25.8h, v24.8h // P4[0..7]-P5[0..7] + mla v19.8h, v7.8h, v0.h[1] // 2*P5[8..15]-5*P6[8..15]+5*P7[8..15] + sub v7.8h, v18.8h, v1.8h // P4[8..15]-P5[8..15] + mls v6.8h, v18.8h, v0.h[1] // 2*P3[8..15]-5*P4[8..15] + abs v27.8h, v26.8h + sshr v26.8h, v26.8h, #8 // clip_sign[0..7] + mls v17.8h, v25.8h, v0.h[0] // 2*P1[0..7]-5*P2[0..7]+5*P3[0..7]-2*P4[0..7] + abs v28.8h, v7.8h + sshr v27.8h, v27.8h, #1 // clip[0..7] + mls v16.8h, v23.8h, v0.h[0] // 2*P5[0..7]-5*P6[0..7]+5*P7[0..7]-2*P8[0..7] + sshr v7.8h, v7.8h, #8 // clip_sign[8..15] + sshr v23.8h, v28.8h, #1 // clip[8..15] + mla v4.8h, v24.8h, v0.h[1] // 2*P3[0..7]-5*P4[0..7]+5*P5[0..7] + cmeq v28.8h, v27.8h, #0 // test clip[0..7] == 0 + srshr v17.8h, v17.8h, #3 + mls v3.8h, v18.8h, v0.h[0] // 2*P1[8..15]-5*P2[8..15]+5*P3[8..15]-2*P4[8..15] + cmeq v29.8h, v23.8h, #0 // test clip[8..15] == 0 + srshr v16.8h, v16.8h, #3 + mls v19.8h, v20.8h, v0.h[0] // 2*P5[8..15]-5*P6[8..15]+5*P7[8..15]-2*P8[8..15] + abs v17.8h, v17.8h // a1[0..7] + mla v6.8h, v1.8h, v0.h[1] // 2*P3[8..15]-5*P4[8..15]+5*P5[8..15] + srshr v3.8h, v3.8h, #3 + mls v4.8h, v21.8h, v0.h[0] // 2*P3[0..7]-5*P4[0..7]+5*P5[0..7]-2*P6[0..7] + abs v16.8h, v16.8h // a2[0..7] + srshr v19.8h, v19.8h, #3 + mls v6.8h, v5.8h, v0.h[0] // 2*P3[8..15]-5*P4[8..15]+5*P5[8..15]-2*P6[8..15] + cmhs v5.8h, v17.8h, v16.8h // test a1[0..7] >= a2[0..7] + abs v3.8h, v3.8h // a1[8..15] + srshr v4.8h, v4.8h, #3 + abs v19.8h, v19.8h // a2[8..15] + bsl v5.16b, v16.16b, v17.16b // a3[0..7] + srshr v6.8h, v6.8h, #3 + cmhs v16.8h, v3.8h, v19.8h // test a1[8..15] >= a2[8.15] + abs v17.8h, v4.8h // a0[0..7] + sshr v4.8h, v4.8h, #8 // a0_sign[0..7] + bsl v16.16b, v19.16b, v3.16b // a3[8..15] + uqsub v3.8h, v17.8h, v5.8h // a0[0..7] >= a3[0..7] ? a0[0..7]-a3[0..7] : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + abs v19.8h, v6.8h // a0[8..15] + cmhs v20.8h, v17.8h, v22.8h // test a0[0..7] >= pq + cmhs v5.8h, v5.8h, v17.8h // test a3[0..7] >= a0[0..7] + sub v4.8h, v26.8h, v4.8h // clip_sign[0..7] - a0_sign[0..7] + sshr v6.8h, v6.8h, #8 // a0_sign[8..15] + mul v3.8h, v3.8h, v0.h[1] // a0[0..7] >= a3[0..7] ? 5*(a0[0..7]-a3[0..7]) : 0 + uqsub v17.8h, v19.8h, v16.8h // a0[8..15] >= a3[8..15] ? a0[8..15]-a3[8..15] : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + orr v20.16b, v28.16b, v20.16b // test clip[0..7] == 0 || a0[0..7] >= pq + cmhs v21.8h, v19.8h, v22.8h // test a0[8..15] >= pq + cmhs v16.8h, v16.8h, v19.8h // test a3[8..15] >= a0[8..15] + mul v0.8h, v17.8h, v0.h[1] // a0[8..15] >= a3[8..15] ? 5*(a0[8..15]-a3[8..15]) : 0 + sub v6.8h, v7.8h, v6.8h // clip_sign[8..15] - a0_sign[8..15] + orr v5.16b, v20.16b, v5.16b // test clip[0..7] == 0 || a0[0..7] >= pq || a3[0..7] >= a0[0..7] + ushr v3.8h, v3.8h, #3 // a0[0..7] >= a3[0..7] ? (5*(a0[0..7]-a3[0..7]))>>3 : 0 + orr v7.16b, v29.16b, v21.16b // test clip[8..15] == 0 || a0[8..15] >= pq + cmtst v17.2d, v5.2d, v2.2d // if 2nd of each group of is not filtered, then none of the others in the group should be either + mov w0, v5.s[1] // move to gp reg + cmhs v19.8h, v3.8h, v27.8h + ushr v0.8h, v0.8h, #3 // a0[8..15] >= a3[8..15] ? (5*(a0[8..15]-a3[8..15]))>>3 : 0 + mov w2, v5.s[3] + orr v5.16b, v7.16b, v16.16b // test clip[8..15] == 0 || a0[8..15] >= pq || a3[8..15] >= a0[8..15] + orr v16.16b, v20.16b, v17.16b + bsl v19.16b, v27.16b, v3.16b // FFMIN(d[0..7], clip[0..7]) + cmtst v2.2d, v5.2d, v2.2d + cmhs v3.8h, v0.8h, v23.8h + mov w4, v5.s[1] + mov w5, v5.s[3] + and w0, w0, w2 + bic v5.16b, v19.16b, v16.16b // set each d[0..7] to zero if it should not be filtered because clip[0..7] == 0 || a0[0..7] >= pq (a3 > a0 case already zeroed by saturating sub) + orr v2.16b, v7.16b, v2.16b + bsl v3.16b, v23.16b, v0.16b // FFMIN(d[8..15], clip[8..15]) + mls v25.8h, v5.8h, v4.8h // invert d[0..7] depending on clip_sign[0..7] & a0_sign[0..7], or zero it if they match, and accumulate into P4[0..7] + and w2, w4, w5 + bic v0.16b, v3.16b, v2.16b // set each d[8..15] to zero if it should not be filtered because clip[8..15] == 0 || a0[8..15] >= pq (a3 > a0 case already zeroed by saturating sub) + mla v24.8h, v5.8h, v4.8h // invert d[0..7] depending on clip_sign[0..7] & a0_sign[0..7], or zero it if they match, and accumulate into P5[0..7] + and w0, w0, w2 + mls v18.8h, v0.8h, v6.8h // invert d[8..15] depending on clip_sign[8..15] & a0_sign[8..15], or zero it if they match, and accumulate into P4[8..15] + sqxtun v2.8b, v25.8h + tbnz w0, #0, 1f // none of the 16 pixel pairs should be updated in this case + mla v1.8h, v0.8h, v6.8h // invert d[8..15] depending on clip_sign[8..15] & a0_sign[8..15], or zero it if they match, and accumulate into P5[8..15] + sqxtun v0.8b, v24.8h + sqxtun2 v2.16b, v18.8h + sqxtun2 v0.16b, v1.8h + st1 {v2.16b}, [x3], x1 + st1 {v0.16b}, [x3] +1: ret +endfunc + +// VC-1 in-loop deblocking filter for 16 pixel pairs at boundary of horizontally-neighbouring blocks +// On entry: +// x0 -> top-left pel of right block +// x1 = row stride, bytes +// w2 = PQUANT bitstream parameter +function ff_vc1_h_loop_filter16_neon, export=1 + sub x3, x0, #4 // where to start reading + ldr d0, .Lcoeffs + ld1 {v1.8b}, [x3], x1 // P1[0], P2[0]... + sub x0, x0, #1 // where to start writing + ld1 {v2.8b}, [x3], x1 + add x4, x0, x1, lsl #3 + ld1 {v3.8b}, [x3], x1 + add x5, x0, x1, lsl #2 + ld1 {v4.8b}, [x3], x1 + add x6, x4, x1, lsl #2 + ld1 {v5.8b}, [x3], x1 + ld1 {v6.8b}, [x3], x1 + ld1 {v7.8b}, [x3], x1 + trn1 v16.8b, v1.8b, v2.8b // P1[0], P1[1], P3[0]... + ld1 {v17.8b}, [x3], x1 + trn2 v1.8b, v1.8b, v2.8b // P2[0], P2[1], P4[0]... + ld1 {v2.8b}, [x3], x1 + trn1 v18.8b, v3.8b, v4.8b // P1[2], P1[3], P3[2]... + ld1 {v19.8b}, [x3], x1 + trn2 v3.8b, v3.8b, v4.8b // P2[2], P2[3], P4[2]... + ld1 {v4.8b}, [x3], x1 + trn1 v20.8b, v5.8b, v6.8b // P1[4], P1[5], P3[4]... + ld1 {v21.8b}, [x3], x1 + trn2 v5.8b, v5.8b, v6.8b // P2[4], P2[5], P4[4]... + ld1 {v6.8b}, [x3], x1 + trn1 v22.8b, v7.8b, v17.8b // P1[6], P1[7], P3[6]... + ld1 {v23.8b}, [x3], x1 + trn2 v7.8b, v7.8b, v17.8b // P2[6], P2[7], P4[6]... + ld1 {v17.8b}, [x3], x1 + trn1 v24.8b, v2.8b, v19.8b // P1[8], P1[9], P3[8]... + ld1 {v25.8b}, [x3] + trn2 v2.8b, v2.8b, v19.8b // P2[8], P2[9], P4[8]... + trn1 v19.4h, v16.4h, v18.4h // P1[0], P1[1], P1[2], P1[3], P5[0]... + trn1 v26.8b, v4.8b, v21.8b // P1[10], P1[11], P3[10]... + trn2 v4.8b, v4.8b, v21.8b // P2[10], P2[11], P4[10]... + trn1 v21.4h, v1.4h, v3.4h // P2[0], P2[1], P2[2], P2[3], P6[0]... + trn1 v27.4h, v20.4h, v22.4h // P1[4], P1[5], P1[6], P1[7], P5[4]... + trn1 v28.8b, v6.8b, v23.8b // P1[12], P1[13], P3[12]... + trn2 v6.8b, v6.8b, v23.8b // P2[12], P2[13], P4[12]... + trn1 v23.4h, v5.4h, v7.4h // P2[4], P2[5], P2[6], P2[7], P6[4]... + trn1 v29.4h, v24.4h, v26.4h // P1[8], P1[9], P1[10], P1[11], P5[8]... + trn1 v30.8b, v17.8b, v25.8b // P1[14], P1[15], P3[14]... + trn2 v17.8b, v17.8b, v25.8b // P2[14], P2[15], P4[14]... + trn1 v25.4h, v2.4h, v4.4h // P2[8], P2[9], P2[10], P2[11], P6[8]... + trn1 v31.2s, v19.2s, v27.2s // P1[0..7] + trn2 v19.2s, v19.2s, v27.2s // P5[0..7] + trn1 v27.2s, v21.2s, v23.2s // P2[0..7] + trn2 v21.2s, v21.2s, v23.2s // P6[0..7] + trn1 v23.4h, v28.4h, v30.4h // P1[12], P1[13], P1[14], P1[15], P5[12]... + trn2 v16.4h, v16.4h, v18.4h // P3[0], P3[1], P3[2], P3[3], P7[0]... + trn1 v18.4h, v6.4h, v17.4h // P2[12], P2[13], P2[14], P2[15], P6[12]... + trn2 v20.4h, v20.4h, v22.4h // P3[4], P3[5], P3[6], P3[7], P7[4]... + trn2 v22.4h, v24.4h, v26.4h // P3[8], P3[9], P3[10], P3[11], P7[8]... + trn1 v24.2s, v29.2s, v23.2s // P1[8..15] + trn2 v23.2s, v29.2s, v23.2s // P5[8..15] + trn1 v26.2s, v25.2s, v18.2s // P2[8..15] + trn2 v18.2s, v25.2s, v18.2s // P6[8..15] + trn2 v25.4h, v28.4h, v30.4h // P3[12], P3[13], P3[14], P3[15], P7[12]... + trn2 v1.4h, v1.4h, v3.4h // P4[0], P4[1], P4[2], P4[3], P8[0]... + trn2 v3.4h, v5.4h, v7.4h // P4[4], P4[5], P4[6], P4[7], P8[4]... + trn2 v2.4h, v2.4h, v4.4h // P4[8], P4[9], P4[10], P4[11], P8[8]... + trn2 v4.4h, v6.4h, v17.4h // P4[12], P4[13], P4[14], P4[15], P8[12]... + ushll v5.8h, v31.8b, #1 // 2*P1[0..7] + ushll v6.8h, v19.8b, #1 // 2*P5[0..7] + trn1 v7.2s, v16.2s, v20.2s // P3[0..7] + uxtl v17.8h, v27.8b // P2[0..7] + trn2 v16.2s, v16.2s, v20.2s // P7[0..7] + uxtl v20.8h, v21.8b // P6[0..7] + trn1 v21.2s, v22.2s, v25.2s // P3[8..15] + ushll v24.8h, v24.8b, #1 // 2*P1[8..15] + trn2 v22.2s, v22.2s, v25.2s // P7[8..15] + ushll v25.8h, v23.8b, #1 // 2*P5[8..15] + trn1 v27.2s, v1.2s, v3.2s // P4[0..7] + uxtl v26.8h, v26.8b // P2[8..15] + mls v5.8h, v17.8h, v0.h[1] // 2*P1[0..7]-5*P2[0..7] + uxtl v17.8h, v18.8b // P6[8..15] + mls v6.8h, v20.8h, v0.h[1] // 2*P5[0..7]-5*P6[0..7] + trn1 v18.2s, v2.2s, v4.2s // P4[8..15] + uxtl v28.8h, v7.8b // P3[0..7] + mls v24.8h, v26.8h, v0.h[1] // 2*P1[8..15]-5*P2[8..15] + uxtl v16.8h, v16.8b // P7[0..7] + uxtl v26.8h, v21.8b // P3[8..15] + mls v25.8h, v17.8h, v0.h[1] // 2*P5[8..15]-5*P6[8..15] + uxtl v22.8h, v22.8b // P7[8..15] + ushll v7.8h, v7.8b, #1 // 2*P3[0..7] + uxtl v27.8h, v27.8b // P4[0..7] + trn2 v1.2s, v1.2s, v3.2s // P8[0..7] + ushll v3.8h, v21.8b, #1 // 2*P3[8..15] + trn2 v2.2s, v2.2s, v4.2s // P8[8..15] + uxtl v4.8h, v18.8b // P4[8..15] + mla v5.8h, v28.8h, v0.h[1] // 2*P1[0..7]-5*P2[0..7]+5*P3[0..7] + uxtl v1.8h, v1.8b // P8[0..7] + mla v6.8h, v16.8h, v0.h[1] // 2*P5[0..7]-5*P6[0..7]+5*P7[0..7] + uxtl v2.8h, v2.8b // P8[8..15] + uxtl v16.8h, v19.8b // P5[0..7] + mla v24.8h, v26.8h, v0.h[1] // 2*P1[8..15]-5*P2[8..15]+5*P3[8..15] + uxtl v18.8h, v23.8b // P5[8..15] + dup v19.8h, w2 // pq + mla v25.8h, v22.8h, v0.h[1] // 2*P5[8..15]-5*P6[8..15]+5*P7[8..15] + sub v21.8h, v27.8h, v16.8h // P4[0..7]-P5[0..7] + sub v22.8h, v4.8h, v18.8h // P4[8..15]-P5[8..15] + mls v7.8h, v27.8h, v0.h[1] // 2*P3[0..7]-5*P4[0..7] + abs v23.8h, v21.8h + mls v3.8h, v4.8h, v0.h[1] // 2*P3[8..15]-5*P4[8..15] + abs v26.8h, v22.8h + sshr v21.8h, v21.8h, #8 // clip_sign[0..7] + mls v5.8h, v27.8h, v0.h[0] // 2*P1[0..7]-5*P2[0..7]+5*P3[0..7]-2*P4[0..7] + sshr v23.8h, v23.8h, #1 // clip[0..7] + sshr v26.8h, v26.8h, #1 // clip[8..15] + mls v6.8h, v1.8h, v0.h[0] // 2*P5[0..7]-5*P6[0..7]+5*P7[0..7]-2*P8[0..7] + sshr v1.8h, v22.8h, #8 // clip_sign[8..15] + cmeq v22.8h, v23.8h, #0 // test clip[0..7] == 0 + mls v24.8h, v4.8h, v0.h[0] // 2*P1[8..15]-5*P2[8..15]+5*P3[8..15]-2*P4[8..15] + cmeq v28.8h, v26.8h, #0 // test clip[8..15] == 0 + srshr v5.8h, v5.8h, #3 + mls v25.8h, v2.8h, v0.h[0] // 2*P5[8..15]-5*P6[8..15]+5*P7[8..15]-2*P8[8..15] + srshr v2.8h, v6.8h, #3 + mla v7.8h, v16.8h, v0.h[1] // 2*P3[0..7]-5*P4[0..7]+5*P5[0..7] + srshr v6.8h, v24.8h, #3 + mla v3.8h, v18.8h, v0.h[1] // 2*P3[8..15]-5*P4[8..15]+5*P5[8..15] + abs v5.8h, v5.8h // a1[0..7] + srshr v24.8h, v25.8h, #3 + mls v3.8h, v17.8h, v0.h[0] // 2*P3[8..15]-5*P4[8..15]+5*P5[8..15]-2*P6[8..15] + abs v2.8h, v2.8h // a2[0..7] + abs v6.8h, v6.8h // a1[8..15] + mls v7.8h, v20.8h, v0.h[0] // 2*P3[0..7]-5*P4[0..7]+5*P5[0..7]-2*P6[0..7] + abs v17.8h, v24.8h // a2[8..15] + cmhs v20.8h, v5.8h, v2.8h // test a1[0..7] >= a2[0..7] + srshr v3.8h, v3.8h, #3 + cmhs v24.8h, v6.8h, v17.8h // test a1[8..15] >= a2[8.15] + srshr v7.8h, v7.8h, #3 + bsl v20.16b, v2.16b, v5.16b // a3[0..7] + abs v2.8h, v3.8h // a0[8..15] + sshr v3.8h, v3.8h, #8 // a0_sign[8..15] + bsl v24.16b, v17.16b, v6.16b // a3[8..15] + abs v5.8h, v7.8h // a0[0..7] + sshr v6.8h, v7.8h, #8 // a0_sign[0..7] + cmhs v7.8h, v2.8h, v19.8h // test a0[8..15] >= pq + sub v1.8h, v1.8h, v3.8h // clip_sign[8..15] - a0_sign[8..15] + uqsub v3.8h, v2.8h, v24.8h // a0[8..15] >= a3[8..15] ? a0[8..15]-a3[8..15] : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + cmhs v2.8h, v24.8h, v2.8h // test a3[8..15] >= a0[8..15] + uqsub v17.8h, v5.8h, v20.8h // a0[0..7] >= a3[0..7] ? a0[0..7]-a3[0..7] : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + cmhs v19.8h, v5.8h, v19.8h // test a0[0..7] >= pq + orr v7.16b, v28.16b, v7.16b // test clip[8..15] == 0 || a0[8..15] >= pq + sub v6.8h, v21.8h, v6.8h // clip_sign[0..7] - a0_sign[0..7] + mul v3.8h, v3.8h, v0.h[1] // a0[8..15] >= a3[8..15] ? 5*(a0[8..15]-a3[8..15]) : 0 + cmhs v5.8h, v20.8h, v5.8h // test a3[0..7] >= a0[0..7] + orr v19.16b, v22.16b, v19.16b // test clip[0..7] == 0 || a0[0..7] >= pq + mul v0.8h, v17.8h, v0.h[1] // a0[0..7] >= a3[0..7] ? 5*(a0[0..7]-a3[0..7]) : 0 + orr v2.16b, v7.16b, v2.16b // test clip[8..15] == 0 || a0[8..15] >= pq || a3[8..15] >= a0[8..15] + orr v5.16b, v19.16b, v5.16b // test clip[0..7] == 0 || a0[0..7] >= pq || a3[0..7] >= a0[0..7] + ushr v3.8h, v3.8h, #3 // a0[8..15] >= a3[8..15] ? (5*(a0[8..15]-a3[8..15]))>>3 : 0 + mov w7, v2.s[1] + mov w8, v2.s[3] + ushr v0.8h, v0.8h, #3 // a0[0..7] >= a3[0..7] ? (5*(a0[0..7]-a3[0..7]))>>3 : 0 + mov w2, v5.s[1] // move to gp reg + cmhs v2.8h, v3.8h, v26.8h + mov w3, v5.s[3] + cmhs v5.8h, v0.8h, v23.8h + bsl v2.16b, v26.16b, v3.16b // FFMIN(d[8..15], clip[8..15]) + and w9, w7, w8 + bsl v5.16b, v23.16b, v0.16b // FFMIN(d[0..7], clip[0..7]) + and w10, w2, w3 + bic v0.16b, v2.16b, v7.16b // set each d[8..15] to zero if it should not be filtered because clip[8..15] == 0 || a0[8..15] >= pq (a3 > a0 case already zeroed by saturating sub) + and w9, w10, w9 + bic v2.16b, v5.16b, v19.16b // set each d[0..7] to zero if it should not be filtered because clip[0..7] == 0 || a0[0..7] >= pq (a3 > a0 case already zeroed by saturating sub) + mls v4.8h, v0.8h, v1.8h // invert d[8..15] depending on clip_sign[8..15] & a0_sign[8..15], or zero it if they match, and accumulate into P4 + tbnz w9, #0, 4f // none of the 16 pixel pairs should be updated in this case + mls v27.8h, v2.8h, v6.8h // invert d[0..7] depending on clip_sign[0..7] & a0_sign[0..7], or zero it if they match, and accumulate into P4 + mla v16.8h, v2.8h, v6.8h // invert d[0..7] depending on clip_sign[0..7] & a0_sign[0..7], or zero it if they match, and accumulate into P5 + sqxtun v2.8b, v4.8h + mla v18.8h, v0.8h, v1.8h // invert d[8..15] depending on clip_sign[8..15] & a0_sign[8..15], or zero it if they match, and accumulate into P5 + sqxtun v0.8b, v27.8h + sqxtun v1.8b, v16.8h + sqxtun v3.8b, v18.8h + tbnz w2, #0, 1f + st2 {v0.b, v1.b}[0], [x0], x1 + st2 {v0.b, v1.b}[1], [x0], x1 + st2 {v0.b, v1.b}[2], [x0], x1 + st2 {v0.b, v1.b}[3], [x0] +1: tbnz w3, #0, 2f + st2 {v0.b, v1.b}[4], [x5], x1 + st2 {v0.b, v1.b}[5], [x5], x1 + st2 {v0.b, v1.b}[6], [x5], x1 + st2 {v0.b, v1.b}[7], [x5] +2: tbnz w7, #0, 3f + st2 {v2.b, v3.b}[0], [x4], x1 + st2 {v2.b, v3.b}[1], [x4], x1 + st2 {v2.b, v3.b}[2], [x4], x1 + st2 {v2.b, v3.b}[3], [x4] +3: tbnz w8, #0, 4f + st2 {v2.b, v3.b}[4], [x6], x1 + st2 {v2.b, v3.b}[5], [x6], x1 + st2 {v2.b, v3.b}[6], [x6], x1 + st2 {v2.b, v3.b}[7], [x6] +4: ret +endfunc From patchwork Thu Mar 31 17:23:47 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35109 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp230062pzb; Thu, 31 Mar 2022 10:25:10 -0700 (PDT) X-Google-Smtp-Source: ABdhPJwfaoJYvSLDR7DscCRcG4dLDIR4K6i506beuWyoddaG1TEyrUV9bQeXtjjPfK4vjsPf3Z2W X-Received: by 2002:a17:907:6ea4:b0:6e1:260e:a232 with SMTP id sh36-20020a1709076ea400b006e1260ea232mr6032145ejc.15.1648747509754; Thu, 31 Mar 2022 10:25:09 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747509; cv=none; d=google.com; s=arc-20160816; b=d73BQs1hk5JtfNmiT7Oxg0gBWx/iL3dZkKDl+mceUaDIvmrPO5Dyg/nU2GE+CYjutB cv4XY8+tCeqWVb8YUal2CmzgtKEdP6XrCOGaKqJSxkZQEynEk5/fhavI6wF+/sxhuHlE F4zC9Qwyr5CWxn1Q8J6l9PdilUoWihX4+yWpyhLfvyKhXqWwxe13Bf3brMTD0pFHmujy qb0KschgjpcplotRiLH/B9MSJxsi+MRTqWjtrZzg8tieAvmXN0gV/LUxrY7xiYm+GIM1 QMkLe/MoXssQl3bDNCkoitsHr0g165GlpS5xEWn9CLgDRbOJoD4ca8dteBtbMmpzOhhJ 9bpg== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=FJ+n+VAi6JPovBKVQoiBcz++o3zcMDlfIXCfQrvQgtM=; b=U1PH9CLn/8A7I5cOPIO26gYIelKmymGevJxS2eNM4c0dA6GHSuzVUf1MQ1eHxo+8g2 6loBZPKuk4JPj1IoGVZLDtZG/DPfpwG8DAfK/53bwliha7UMjl/Ls5bx/GVMo6S8bVkF dTcOma2MPawdx/6mb3h6FSaCJZfIBxoQIeGJkohc3gwowK+rY/DM2oxg3bPpO/BqfdTg 1s1SrGITEIA/gFYe/MQGJ7UxD4NXoKOstSC6Dcb5IKjqlqIiWLPhNFMPdpCdWBoyM02H 6Fk6E0fPv3D8ln31hgE5UqeJlJ1LRgNggbg6wMUe7G2RKFY18pJXCnsM/Cr7EoeOHmsv gm3A== 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 j6-20020a50e0c6000000b004194b1618f9si180615edl.62.2022.03.31.10.25.09; Thu, 31 Mar 2022 10:25:09 -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 3FE5268B2A1; Thu, 31 Mar 2022 20:24:20 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail148161.authsmtp.com (outmail148161.authsmtp.com [62.13.148.161]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id F14B968B098 for ; Thu, 31 Mar 2022 20:24:15 +0300 (EEST) Received: from punt18.authsmtp.com (punt18.authsmtp.com [62.13.128.225]) by punt17.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOF3K021849 for ; Thu, 31 Mar 2022 18:24:15 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt18.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOFd2097059; Thu, 31 Mar 2022 18:24:15 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHOC1g062500 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:13 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:12 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:47 +0100 Message-Id: <20220331172351.550818-7-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 6309bf44-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQtVIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLWpzIAVAentw ZQhrX3dfVAp8dkZ0 EBhQEXBUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdW0Bb30N d0kEM1kVTUsAWSA3 HkJKNC8qVRZfAi8y M1QAB3k6VEocOEE1 OlggX11QPR4WFgBE V09EGzMRP1AdRzFD X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 06/10] avcodec/vc1: Arm 32-bit NEON deblocking filter fast paths 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: IfPyNbp0QkrY checkasm benchmarks on 1.5 GHz Cortex-A72 are as follows. Note that the C version can still outperform the NEON version in specific cases. The balance between different code paths is stream-dependent, but in practice the best case happens about 5% of the time, the worst case happens about 40% of the time, and the complexity of the remaining cases fall somewhere in between. Therefore, taking the average of the best and worst case timings is probably a conservative estimate of the degree by which the NEON code improves performance. vc1dsp.vc1_h_loop_filter4_bestcase_c: 19.0 vc1dsp.vc1_h_loop_filter4_bestcase_neon: 48.5 vc1dsp.vc1_h_loop_filter4_worstcase_c: 144.7 vc1dsp.vc1_h_loop_filter4_worstcase_neon: 76.2 vc1dsp.vc1_h_loop_filter8_bestcase_c: 41.0 vc1dsp.vc1_h_loop_filter8_bestcase_neon: 75.0 vc1dsp.vc1_h_loop_filter8_worstcase_c: 294.0 vc1dsp.vc1_h_loop_filter8_worstcase_neon: 102.7 vc1dsp.vc1_h_loop_filter16_bestcase_c: 54.7 vc1dsp.vc1_h_loop_filter16_bestcase_neon: 130.0 vc1dsp.vc1_h_loop_filter16_worstcase_c: 569.7 vc1dsp.vc1_h_loop_filter16_worstcase_neon: 186.7 vc1dsp.vc1_v_loop_filter4_bestcase_c: 20.2 vc1dsp.vc1_v_loop_filter4_bestcase_neon: 47.2 vc1dsp.vc1_v_loop_filter4_worstcase_c: 164.2 vc1dsp.vc1_v_loop_filter4_worstcase_neon: 68.5 vc1dsp.vc1_v_loop_filter8_bestcase_c: 43.5 vc1dsp.vc1_v_loop_filter8_bestcase_neon: 55.2 vc1dsp.vc1_v_loop_filter8_worstcase_c: 316.2 vc1dsp.vc1_v_loop_filter8_worstcase_neon: 72.7 vc1dsp.vc1_v_loop_filter16_bestcase_c: 62.2 vc1dsp.vc1_v_loop_filter16_bestcase_neon: 103.7 vc1dsp.vc1_v_loop_filter16_worstcase_c: 646.5 vc1dsp.vc1_v_loop_filter16_worstcase_neon: 110.7 Signed-off-by: Ben Avison --- libavcodec/arm/vc1dsp_init_neon.c | 14 + libavcodec/arm/vc1dsp_neon.S | 643 ++++++++++++++++++++++++++++++ 2 files changed, 657 insertions(+) diff --git a/libavcodec/arm/vc1dsp_init_neon.c b/libavcodec/arm/vc1dsp_init_neon.c index 2cca784f5a..f5f5c702d7 100644 --- a/libavcodec/arm/vc1dsp_init_neon.c +++ b/libavcodec/arm/vc1dsp_init_neon.c @@ -32,6 +32,13 @@ void ff_vc1_inv_trans_4x8_dc_neon(uint8_t *dest, ptrdiff_t stride, int16_t *bloc void ff_vc1_inv_trans_8x4_dc_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); void ff_vc1_inv_trans_4x4_dc_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); +void ff_vc1_v_loop_filter4_neon(uint8_t *src, int stride, int pq); +void ff_vc1_h_loop_filter4_neon(uint8_t *src, int stride, int pq); +void ff_vc1_v_loop_filter8_neon(uint8_t *src, int stride, int pq); +void ff_vc1_h_loop_filter8_neon(uint8_t *src, int stride, int pq); +void ff_vc1_v_loop_filter16_neon(uint8_t *src, int stride, int pq); +void ff_vc1_h_loop_filter16_neon(uint8_t *src, int stride, int pq); + void ff_put_pixels8x8_neon(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int rnd); @@ -92,6 +99,13 @@ av_cold void ff_vc1dsp_init_neon(VC1DSPContext *dsp) dsp->vc1_inv_trans_8x4_dc = ff_vc1_inv_trans_8x4_dc_neon; dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_neon; + dsp->vc1_v_loop_filter4 = ff_vc1_v_loop_filter4_neon; + dsp->vc1_h_loop_filter4 = ff_vc1_h_loop_filter4_neon; + dsp->vc1_v_loop_filter8 = ff_vc1_v_loop_filter8_neon; + dsp->vc1_h_loop_filter8 = ff_vc1_h_loop_filter8_neon; + dsp->vc1_v_loop_filter16 = ff_vc1_v_loop_filter16_neon; + dsp->vc1_h_loop_filter16 = ff_vc1_h_loop_filter16_neon; + dsp->put_vc1_mspel_pixels_tab[1][ 0] = ff_put_pixels8x8_neon; FN_ASSIGN(1, 0); FN_ASSIGN(2, 0); diff --git a/libavcodec/arm/vc1dsp_neon.S b/libavcodec/arm/vc1dsp_neon.S index 93f043bf08..ba54221ef6 100644 --- a/libavcodec/arm/vc1dsp_neon.S +++ b/libavcodec/arm/vc1dsp_neon.S @@ -1161,3 +1161,646 @@ function ff_vc1_inv_trans_4x4_dc_neon, export=1 vst1.32 {d1[1]}, [r0,:32] bx lr endfunc + +@ VC-1 in-loop deblocking filter for 4 pixel pairs at boundary of vertically-neighbouring blocks +@ On entry: +@ r0 -> top-left pel of lower block +@ r1 = row stride, bytes +@ r2 = PQUANT bitstream parameter +function ff_vc1_v_loop_filter4_neon, export=1 + sub r3, r0, r1, lsl #2 + vldr d0, .Lcoeffs + vld1.32 {d1[0]}, [r0], r1 @ P5 + vld1.32 {d2[0]}, [r3], r1 @ P1 + vld1.32 {d3[0]}, [r3], r1 @ P2 + vld1.32 {d4[0]}, [r0], r1 @ P6 + vld1.32 {d5[0]}, [r3], r1 @ P3 + vld1.32 {d6[0]}, [r0], r1 @ P7 + vld1.32 {d7[0]}, [r3] @ P4 + vld1.32 {d16[0]}, [r0] @ P8 + vshll.u8 q9, d1, #1 @ 2*P5 + vdup.16 d17, r2 @ pq + vshll.u8 q10, d2, #1 @ 2*P1 + vmovl.u8 q11, d3 @ P2 + vmovl.u8 q1, d4 @ P6 + vmovl.u8 q12, d5 @ P3 + vmls.i16 d20, d22, d0[1] @ 2*P1-5*P2 + vmovl.u8 q11, d6 @ P7 + vmls.i16 d18, d2, d0[1] @ 2*P5-5*P6 + vshll.u8 q2, d5, #1 @ 2*P3 + vmovl.u8 q3, d7 @ P4 + vmla.i16 d18, d22, d0[1] @ 2*P5-5*P6+5*P7 + vmovl.u8 q11, d16 @ P8 + vmla.u16 d20, d24, d0[1] @ 2*P1-5*P2+5*P3 + vmovl.u8 q12, d1 @ P5 + vmls.u16 d4, d6, d0[1] @ 2*P3-5*P4 + vmls.u16 d18, d22, d0[0] @ 2*P5-5*P6+5*P7-2*P8 + vsub.i16 d1, d6, d24 @ P4-P5 + vmls.i16 d20, d6, d0[0] @ 2*P1-5*P2+5*P3-2*P4 + vmla.i16 d4, d24, d0[1] @ 2*P3-5*P4+5*P5 + vmls.i16 d4, d2, d0[0] @ 2*P3-5*P4+5*P5-2*P6 + vabs.s16 d2, d1 + vrshr.s16 d3, d18, #3 + vrshr.s16 d5, d20, #3 + vshr.s16 d2, d2, #1 @ clip + vrshr.s16 d4, d4, #3 + vabs.s16 d3, d3 @ a2 + vshr.s16 d1, d1, #8 @ clip_sign + vabs.s16 d5, d5 @ a1 + vceq.i16 d7, d2, #0 @ test clip == 0 + vabs.s16 d16, d4 @ a0 + vshr.s16 d4, d4, #8 @ a0_sign + vcge.s16 d18, d5, d3 @ test a1 >= a2 + vcge.s16 d17, d16, d17 @ test a0 >= pq + vbsl d18, d3, d5 @ a3 + vsub.i16 d1, d1, d4 @ clip_sign - a0_sign + vorr d3, d7, d17 @ test clip == 0 || a0 >= pq + vqsub.u16 d4, d16, d18 @ a0 >= a3 ? a0-a3 : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + vcge.s16 d5, d18, d16 @ test a3 >= a0 + vmul.i16 d0, d4, d0[1] @ a0 >= a3 ? 5*(a0-a3) : 0 + vorr d4, d3, d5 @ test clip == 0 || a0 >= pq || a3 >= a0 + vmov.32 r0, d4[1] @ move to gp reg + vshr.u16 d0, d0, #3 @ a0 >= a3 ? (5*(a0-a3))>>3 : 0 + vcge.s16 d4, d0, d2 + tst r0, #1 + bne 1f @ none of the 4 pixel pairs should be updated if this one is not filtered + vbsl d4, d2, d0 @ FFMIN(d, clip) + vbic d0, d4, d3 @ set each d to zero if it should not be filtered because clip == 0 || a0 >= pq (a3 > a0 case already zeroed by saturating sub) + vmls.i16 d6, d0, d1 @ invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P4 + vmla.i16 d24, d0, d1 @ invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P5 + vqmovun.s16 d0, q3 + vqmovun.s16 d1, q12 + vst1.32 {d0[0]}, [r3], r1 + vst1.32 {d1[0]}, [r3] +1: bx lr +endfunc + +@ VC-1 in-loop deblocking filter for 4 pixel pairs at boundary of horizontally-neighbouring blocks +@ On entry: +@ r0 -> top-left pel of right block +@ r1 = row stride, bytes +@ r2 = PQUANT bitstream parameter +function ff_vc1_h_loop_filter4_neon, export=1 + sub r3, r0, #4 @ where to start reading + vldr d0, .Lcoeffs + vld1.32 {d2}, [r3], r1 + sub r0, r0, #1 @ where to start writing + vld1.32 {d4}, [r3], r1 + vld1.32 {d3}, [r3], r1 + vld1.32 {d5}, [r3] + vdup.16 d1, r2 @ pq + vtrn.8 q1, q2 + vtrn.16 d2, d3 @ P1, P5, P3, P7 + vtrn.16 d4, d5 @ P2, P6, P4, P8 + vshll.u8 q3, d2, #1 @ 2*P1, 2*P5 + vmovl.u8 q8, d4 @ P2, P6 + vmovl.u8 q9, d3 @ P3, P7 + vmovl.u8 q2, d5 @ P4, P8 + vmls.i16 q3, q8, d0[1] @ 2*P1-5*P2, 2*P5-5*P6 + vshll.u8 q10, d3, #1 @ 2*P3, 2*P7 + vmovl.u8 q1, d2 @ P1, P5 + vmla.i16 q3, q9, d0[1] @ 2*P1-5*P2+5*P3, 2*P5-5*P6+5*P7 + vmls.i16 q3, q2, d0[0] @ 2*P1-5*P2+5*P3-2*P4, 2*P5-5*P6+5*P7-2*P8 + vmov d2, d3 @ needs to be in an even-numbered vector for when we come to narrow it later + vmls.i16 d20, d4, d0[1] @ 2*P3-5*P4 + vmla.i16 d20, d3, d0[1] @ 2*P3-5*P4+5*P5 + vsub.i16 d3, d4, d2 @ P4-P5 + vmls.i16 d20, d17, d0[0] @ 2*P3-5*P4+5*P5-2*P6 + vrshr.s16 q3, q3, #3 + vabs.s16 d5, d3 + vshr.s16 d3, d3, #8 @ clip_sign + vrshr.s16 d16, d20, #3 + vabs.s16 q3, q3 @ a1, a2 + vshr.s16 d5, d5, #1 @ clip + vabs.s16 d17, d16 @ a0 + vceq.i16 d18, d5, #0 @ test clip == 0 + vshr.s16 d16, d16, #8 @ a0_sign + vcge.s16 d19, d6, d7 @ test a1 >= a2 + vcge.s16 d1, d17, d1 @ test a0 >= pq + vsub.i16 d16, d3, d16 @ clip_sign - a0_sign + vbsl d19, d7, d6 @ a3 + vorr d1, d18, d1 @ test clip == 0 || a0 >= pq + vqsub.u16 d3, d17, d19 @ a0 >= a3 ? a0-a3 : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + vcge.s16 d6, d19, d17 @ test a3 >= a0 @ + vmul.i16 d0, d3, d0[1] @ a0 >= a3 ? 5*(a0-a3) : 0 + vorr d3, d1, d6 @ test clip == 0 || a0 >= pq || a3 >= a0 + vmov.32 r2, d3[1] @ move to gp reg + vshr.u16 d0, d0, #3 @ a0 >= a3 ? (5*(a0-a3))>>3 : 0 + vcge.s16 d3, d0, d5 + tst r2, #1 + bne 1f @ none of the 4 pixel pairs should be updated if this one is not filtered + vbsl d3, d5, d0 @ FFMIN(d, clip) + vbic d0, d3, d1 @ set each d to zero if it should not be filtered because clip == 0 || a0 >= pq (a3 > a0 case already zeroed by saturating sub) + vmla.i16 d2, d0, d16 @ invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P5 + vmls.i16 d4, d0, d16 @ invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P4 + vqmovun.s16 d1, q1 + vqmovun.s16 d0, q2 + vst2.8 {d0[0], d1[0]}, [r0], r1 + vst2.8 {d0[1], d1[1]}, [r0], r1 + vst2.8 {d0[2], d1[2]}, [r0], r1 + vst2.8 {d0[3], d1[3]}, [r0] +1: bx lr +endfunc + +@ VC-1 in-loop deblocking filter for 8 pixel pairs at boundary of vertically-neighbouring blocks +@ On entry: +@ r0 -> top-left pel of lower block +@ r1 = row stride, bytes +@ r2 = PQUANT bitstream parameter +function ff_vc1_v_loop_filter8_neon, export=1 + sub r3, r0, r1, lsl #2 + vldr d0, .Lcoeffs + vld1.32 {d1}, [r0 :64], r1 @ P5 + vld1.32 {d2}, [r3 :64], r1 @ P1 + vld1.32 {d3}, [r3 :64], r1 @ P2 + vld1.32 {d4}, [r0 :64], r1 @ P6 + vld1.32 {d5}, [r3 :64], r1 @ P3 + vld1.32 {d6}, [r0 :64], r1 @ P7 + vshll.u8 q8, d1, #1 @ 2*P5 + vshll.u8 q9, d2, #1 @ 2*P1 + vld1.32 {d7}, [r3 :64] @ P4 + vmovl.u8 q1, d3 @ P2 + vld1.32 {d20}, [r0 :64] @ P8 + vmovl.u8 q11, d4 @ P6 + vdup.16 q12, r2 @ pq + vmovl.u8 q13, d5 @ P3 + vmls.i16 q9, q1, d0[1] @ 2*P1-5*P2 + vmovl.u8 q1, d6 @ P7 + vshll.u8 q2, d5, #1 @ 2*P3 + vmls.i16 q8, q11, d0[1] @ 2*P5-5*P6 + vmovl.u8 q3, d7 @ P4 + vmovl.u8 q10, d20 @ P8 + vmla.i16 q8, q1, d0[1] @ 2*P5-5*P6+5*P7 + vmovl.u8 q1, d1 @ P5 + vmla.i16 q9, q13, d0[1] @ 2*P1-5*P2+5*P3 + vsub.i16 q13, q3, q1 @ P4-P5 + vmls.i16 q2, q3, d0[1] @ 2*P3-5*P4 + vmls.i16 q8, q10, d0[0] @ 2*P5-5*P6+5*P7-2*P8 + vabs.s16 q10, q13 + vshr.s16 q13, q13, #8 @ clip_sign + vmls.i16 q9, q3, d0[0] @ 2*P1-5*P2+5*P3-2*P4 + vshr.s16 q10, q10, #1 @ clip + vmla.i16 q2, q1, d0[1] @ 2*P3-5*P4+5*P5 + vrshr.s16 q8, q8, #3 + vmls.i16 q2, q11, d0[0] @ 2*P3-5*P4+5*P5-2*P6 + vceq.i16 q11, q10, #0 @ test clip == 0 + vrshr.s16 q9, q9, #3 + vabs.s16 q8, q8 @ a2 + vabs.s16 q9, q9 @ a1 + vrshr.s16 q2, q2, #3 + vcge.s16 q14, q9, q8 @ test a1 >= a2 + vabs.s16 q15, q2 @ a0 + vshr.s16 q2, q2, #8 @ a0_sign + vbsl q14, q8, q9 @ a3 + vcge.s16 q8, q15, q12 @ test a0 >= pq + vsub.i16 q2, q13, q2 @ clip_sign - a0_sign + vqsub.u16 q9, q15, q14 @ a0 >= a3 ? a0-a3 : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + vcge.s16 q12, q14, q15 @ test a3 >= a0 + vorr q8, q11, q8 @ test clip == 0 || a0 >= pq + vmul.i16 q0, q9, d0[1] @ a0 >= a3 ? 5*(a0-a3) : 0 + vorr q9, q8, q12 @ test clip == 0 || a0 >= pq || a3 >= a0 + vshl.i64 q11, q9, #16 + vmov.32 r0, d18[1] @ move to gp reg + vshr.u16 q0, q0, #3 @ a0 >= a3 ? (5*(a0-a3))>>3 : 0 + vmov.32 r2, d19[1] + vshr.s64 q9, q11, #48 + vcge.s16 q11, q0, q10 + vorr q8, q8, q9 + and r0, r0, r2 + vbsl q11, q10, q0 @ FFMIN(d, clip) + tst r0, #1 + bne 1f @ none of the 8 pixel pairs should be updated in this case + vbic q0, q11, q8 @ set each d to zero if it should not be filtered + vmls.i16 q3, q0, q2 @ invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P4 + vmla.i16 q1, q0, q2 @ invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P5 + vqmovun.s16 d0, q3 + vqmovun.s16 d1, q1 + vst1.32 {d0}, [r3 :64], r1 + vst1.32 {d1}, [r3 :64] +1: bx lr +endfunc + +.align 5 +.Lcoeffs: +.quad 0x00050002 + +@ VC-1 in-loop deblocking filter for 8 pixel pairs at boundary of horizontally-neighbouring blocks +@ On entry: +@ r0 -> top-left pel of right block +@ r1 = row stride, bytes +@ r2 = PQUANT bitstream parameter +function ff_vc1_h_loop_filter8_neon, export=1 + push {lr} + sub r3, r0, #4 @ where to start reading + vldr d0, .Lcoeffs + vld1.32 {d2}, [r3], r1 @ P1[0], P2[0]... + sub r0, r0, #1 @ where to start writing + vld1.32 {d4}, [r3], r1 + add r12, r0, r1, lsl #2 + vld1.32 {d3}, [r3], r1 + vld1.32 {d5}, [r3], r1 + vld1.32 {d6}, [r3], r1 + vld1.32 {d16}, [r3], r1 + vld1.32 {d7}, [r3], r1 + vld1.32 {d17}, [r3] + vtrn.8 q1, q2 @ P1[0], P1[1], P3[0]... P1[2], P1[3], P3[2]... P2[0], P2[1], P4[0]... P2[2], P2[3], P4[2]... + vdup.16 q9, r2 @ pq + vtrn.16 d2, d3 @ P1[0], P1[1], P1[2], P1[3], P5[0]... P3[0], P3[1], P3[2], P3[3], P7[0]... + vtrn.16 d4, d5 @ P2[0], P2[1], P2[2], P2[3], P6[0]... P4[0], P4[1], P4[2], P4[3], P8[0]... + vtrn.8 q3, q8 @ P1[4], P1[5], P3[4]... P1[6], P1[7], P3[6]... P2[4], P2[5], P4[4]... P2[6], P2[7], P4[6]... + vtrn.16 d6, d7 @ P1[4], P1[5], P1[6], P1[7], P5[4]... P3[4], P3[5], P3[5], P3[7], P7[4]... + vtrn.16 d16, d17 @ P2[4], P2[5], P2[6], P2[7], P6[4]... P4[4], P4[5], P4[6], P4[7], P8[4]... + vtrn.32 d2, d6 @ P1, P5 + vtrn.32 d4, d16 @ P2, P6 + vtrn.32 d3, d7 @ P3, P7 + vtrn.32 d5, d17 @ P4, P8 + vshll.u8 q10, d2, #1 @ 2*P1 + vshll.u8 q11, d6, #1 @ 2*P5 + vmovl.u8 q12, d4 @ P2 + vmovl.u8 q13, d16 @ P6 + vmovl.u8 q14, d3 @ P3 + vmls.i16 q10, q12, d0[1] @ 2*P1-5*P2 + vmovl.u8 q12, d7 @ P7 + vshll.u8 q1, d3, #1 @ 2*P3 + vmls.i16 q11, q13, d0[1] @ 2*P5-5*P6 + vmovl.u8 q2, d5 @ P4 + vmovl.u8 q8, d17 @ P8 + vmla.i16 q11, q12, d0[1] @ 2*P5-5*P6+5*P7 + vmovl.u8 q3, d6 @ P5 + vmla.i16 q10, q14, d0[1] @ 2*P1-5*P2+5*P3 + vsub.i16 q12, q2, q3 @ P4-P5 + vmls.i16 q1, q2, d0[1] @ 2*P3-5*P4 + vmls.i16 q11, q8, d0[0] @ 2*P5-5*P6+5*P7-2*P8 + vabs.s16 q8, q12 + vshr.s16 q12, q12, #8 @ clip_sign + vmls.i16 q10, q2, d0[0] @ 2*P1-5*P2+5*P3-2*P4 + vshr.s16 q8, q8, #1 @ clip + vmla.i16 q1, q3, d0[1] @ 2*P3-5*P4+5*P5 + vrshr.s16 q11, q11, #3 + vmls.i16 q1, q13, d0[0] @ 2*P3-5*P4+5*P5-2*P6 + vceq.i16 q13, q8, #0 @ test clip == 0 + vrshr.s16 q10, q10, #3 + vabs.s16 q11, q11 @ a2 + vabs.s16 q10, q10 @ a1 + vrshr.s16 q1, q1, #3 + vcge.s16 q14, q10, q11 @ test a1 >= a2 + vabs.s16 q15, q1 @ a0 + vshr.s16 q1, q1, #8 @ a0_sign + vbsl q14, q11, q10 @ a3 + vcge.s16 q9, q15, q9 @ test a0 >= pq + vsub.i16 q1, q12, q1 @ clip_sign - a0_sign + vqsub.u16 q10, q15, q14 @ a0 >= a3 ? a0-a3 : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + vcge.s16 q11, q14, q15 @ test a3 >= a0 + vorr q9, q13, q9 @ test clip == 0 || a0 >= pq + vmul.i16 q0, q10, d0[1] @ a0 >= a3 ? 5*(a0-a3) : 0 + vorr q10, q9, q11 @ test clip == 0 || a0 >= pq || a3 >= a0 + vmov.32 r2, d20[1] @ move to gp reg + vshr.u16 q0, q0, #3 @ a0 >= a3 ? (5*(a0-a3))>>3 : 0 + vmov.32 r3, d21[1] + vcge.s16 q10, q0, q8 + and r14, r2, r3 + vbsl q10, q8, q0 @ FFMIN(d, clip) + tst r14, #1 + bne 2f @ none of the 8 pixel pairs should be updated in this case + vbic q0, q10, q9 @ set each d to zero if it should not be filtered because clip == 0 || a0 >= pq (a3 > a0 case already zeroed by saturating sub) + vmla.i16 q3, q0, q1 @ invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P5 + vmls.i16 q2, q0, q1 @ invert d depending on clip_sign & a0_sign, or zero it if they match, and accumulate into P4 + vqmovun.s16 d1, q3 + vqmovun.s16 d0, q2 + tst r2, #1 + bne 1f @ none of the first 4 pixel pairs should be updated if so + vst2.8 {d0[0], d1[0]}, [r0], r1 + vst2.8 {d0[1], d1[1]}, [r0], r1 + vst2.8 {d0[2], d1[2]}, [r0], r1 + vst2.8 {d0[3], d1[3]}, [r0] +1: tst r3, #1 + bne 2f @ none of the second 4 pixel pairs should be updated if so + vst2.8 {d0[4], d1[4]}, [r12], r1 + vst2.8 {d0[5], d1[5]}, [r12], r1 + vst2.8 {d0[6], d1[6]}, [r12], r1 + vst2.8 {d0[7], d1[7]}, [r12] +2: pop {pc} +endfunc + +@ VC-1 in-loop deblocking filter for 16 pixel pairs at boundary of vertically-neighbouring blocks +@ On entry: +@ r0 -> top-left pel of lower block +@ r1 = row stride, bytes +@ r2 = PQUANT bitstream parameter +function ff_vc1_v_loop_filter16_neon, export=1 + vpush {d8-d15} + sub r3, r0, r1, lsl #2 + vldr d0, .Lcoeffs + vld1.64 {q1}, [r0 :128], r1 @ P5 + vld1.64 {q2}, [r3 :128], r1 @ P1 + vld1.64 {q3}, [r3 :128], r1 @ P2 + vld1.64 {q4}, [r0 :128], r1 @ P6 + vld1.64 {q5}, [r3 :128], r1 @ P3 + vld1.64 {q6}, [r0 :128], r1 @ P7 + vshll.u8 q7, d2, #1 @ 2*P5[0..7] + vshll.u8 q8, d4, #1 @ 2*P1[0..7] + vld1.64 {q9}, [r3 :128] @ P4 + vmovl.u8 q10, d6 @ P2[0..7] + vld1.64 {q11}, [r0 :128] @ P8 + vmovl.u8 q12, d8 @ P6[0..7] + vdup.16 q13, r2 @ pq + vshll.u8 q2, d5, #1 @ 2*P1[8..15] + vmls.i16 q8, q10, d0[1] @ 2*P1[0..7]-5*P2[0..7] + vshll.u8 q10, d3, #1 @ 2*P5[8..15] + vmovl.u8 q3, d7 @ P2[8..15] + vmls.i16 q7, q12, d0[1] @ 2*P5[0..7]-5*P6[0..7] + vmovl.u8 q4, d9 @ P6[8..15] + vmovl.u8 q14, d10 @ P3[0..7] + vmovl.u8 q15, d12 @ P7[0..7] + vmls.i16 q2, q3, d0[1] @ 2*P1[8..15]-5*P2[8..15] + vshll.u8 q3, d10, #1 @ 2*P3[0..7] + vmls.i16 q10, q4, d0[1] @ 2*P5[8..15]-5*P6[8..15] + vmovl.u8 q6, d13 @ P7[8..15] + vmla.i16 q8, q14, d0[1] @ 2*P1[0..7]-5*P2[0..7]+5*P3[0..7] + vmovl.u8 q14, d18 @ P4[0..7] + vmovl.u8 q9, d19 @ P4[8..15] + vmla.i16 q7, q15, d0[1] @ 2*P5[0..7]-5*P6[0..7]+5*P7[0..7] + vmovl.u8 q15, d11 @ P3[8..15] + vshll.u8 q5, d11, #1 @ 2*P3[8..15] + vmls.i16 q3, q14, d0[1] @ 2*P3[0..7]-5*P4[0..7] + vmla.i16 q2, q15, d0[1] @ 2*P1[8..15]-5*P2[8..15]+5*P3[8..15] + vmovl.u8 q15, d22 @ P8[0..7] + vmovl.u8 q11, d23 @ P8[8..15] + vmla.i16 q10, q6, d0[1] @ 2*P5[8..15]-5*P6[8..15]+5*P7[8..15] + vmovl.u8 q6, d2 @ P5[0..7] + vmovl.u8 q1, d3 @ P5[8..15] + vmls.i16 q5, q9, d0[1] @ 2*P3[8..15]-5*P4[8..15] + vmls.i16 q8, q14, d0[0] @ 2*P1[0..7]-5*P2[0..7]+5*P3[0..7]-2*P4[0..7] + vmls.i16 q7, q15, d0[0] @ 2*P5[0..7]-5*P6[0..7]+5*P7[0..7]-2*P8[0..7] + vsub.i16 q15, q14, q6 @ P4[0..7]-P5[0..7] + vmla.i16 q3, q6, d0[1] @ 2*P3[0..7]-5*P4[0..7]+5*P5[0..7] + vrshr.s16 q8, q8, #3 + vmls.i16 q2, q9, d0[0] @ 2*P1[8..15]-5*P2[8..15]+5*P3[8..15]-2*P4[8..15] + vrshr.s16 q7, q7, #3 + vmls.i16 q10, q11, d0[0] @ 2*P5[8..15]-5*P6[8..15]+5*P7[8..15]-2*P8[8..15] + vabs.s16 q11, q15 + vabs.s16 q8, q8 @ a1[0..7] + vmla.i16 q5, q1, d0[1] @ 2*P3[8..15]-5*P4[8..15]+5*P5[8..15] + vshr.s16 q15, q15, #8 @ clip_sign[0..7] + vrshr.s16 q2, q2, #3 + vmls.i16 q3, q12, d0[0] @ 2*P3[0..7]-5*P4[0..7]+5*P5[0..7]-2*P6[0..7] + vabs.s16 q7, q7 @ a2[0..7] + vrshr.s16 q10, q10, #3 + vsub.i16 q12, q9, q1 @ P4[8..15]-P5[8..15] + vshr.s16 q11, q11, #1 @ clip[0..7] + vmls.i16 q5, q4, d0[0] @ 2*P3[8..15]-5*P4[8..15]+5*P5[8..15]-2*P6[8..15] + vcge.s16 q4, q8, q7 @ test a1[0..7] >= a2[0..7] + vabs.s16 q2, q2 @ a1[8..15] + vrshr.s16 q3, q3, #3 + vabs.s16 q10, q10 @ a2[8..15] + vbsl q4, q7, q8 @ a3[0..7] + vabs.s16 q7, q12 + vshr.s16 q8, q12, #8 @ clip_sign[8..15] + vrshr.s16 q5, q5, #3 + vcge.s16 q12, q2, q10 @ test a1[8..15] >= a2[8.15] + vshr.s16 q7, q7, #1 @ clip[8..15] + vbsl q12, q10, q2 @ a3[8..15] + vabs.s16 q2, q3 @ a0[0..7] + vceq.i16 q10, q11, #0 @ test clip[0..7] == 0 + vshr.s16 q3, q3, #8 @ a0_sign[0..7] + vsub.i16 q3, q15, q3 @ clip_sign[0..7] - a0_sign[0..7] + vcge.s16 q15, q2, q13 @ test a0[0..7] >= pq + vorr q10, q10, q15 @ test clip[0..7] == 0 || a0[0..7] >= pq + vqsub.u16 q15, q2, q4 @ a0[0..7] >= a3[0..7] ? a0[0..7]-a3[0..7] : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + vcge.s16 q2, q4, q2 @ test a3[0..7] >= a0[0..7] + vabs.s16 q4, q5 @ a0[8..15] + vshr.s16 q5, q5, #8 @ a0_sign[8..15] + vmul.i16 q15, q15, d0[1] @ a0[0..7] >= a3[0..7] ? 5*(a0[0..7]-a3[0..7]) : 0 + vcge.s16 q13, q4, q13 @ test a0[8..15] >= pq + vorr q2, q10, q2 @ test clip[0..7] == 0 || a0[0..7] >= pq || a3[0..7] >= a0[0..7] + vsub.i16 q5, q8, q5 @ clip_sign[8..15] - a0_sign[8..15] + vceq.i16 q8, q7, #0 @ test clip[8..15] == 0 + vshr.u16 q15, q15, #3 @ a0[0..7] >= a3[0..7] ? (5*(a0[0..7]-a3[0..7]))>>3 : 0 + vmov.32 r0, d4[1] @ move to gp reg + vorr q8, q8, q13 @ test clip[8..15] == 0 || a0[8..15] >= pq + vqsub.u16 q13, q4, q12 @ a0[8..15] >= a3[8..15] ? a0[8..15]-a3[8..15] : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + vmov.32 r2, d5[1] + vcge.s16 q4, q12, q4 @ test a3[8..15] >= a0[8..15] + vshl.i64 q2, q2, #16 + vcge.s16 q12, q15, q11 + vmul.i16 q0, q13, d0[1] @ a0[8..15] >= a3[8..15] ? 5*(a0[8..15]-a3[8..15]) : 0 + vorr q4, q8, q4 @ test clip[8..15] == 0 || a0[8..15] >= pq || a3[8..15] >= a0[8..15] + vshr.s64 q2, q2, #48 + and r0, r0, r2 + vbsl q12, q11, q15 @ FFMIN(d[0..7], clip[0..7]) + vshl.i64 q11, q4, #16 + vmov.32 r2, d8[1] + vshr.u16 q0, q0, #3 @ a0[8..15] >= a3[8..15] ? (5*(a0[8..15]-a3[8..15]))>>3 : 0 + vorr q2, q10, q2 + vmov.32 r12, d9[1] + vshr.s64 q4, q11, #48 + vcge.s16 q10, q0, q7 + vbic q2, q12, q2 @ set each d[0..7] to zero if it should not be filtered because clip[0..7] == 0 || a0[0..7] >= pq (a3 > a0 case already zeroed by saturating sub) + vorr q4, q8, q4 + and r2, r2, r12 + vbsl q10, q7, q0 @ FFMIN(d[8..15], clip[8..15]) + vmls.i16 q14, q2, q3 @ invert d[0..7] depending on clip_sign[0..7] & a0_sign[0..7], or zero it if they match, and accumulate into P4[0..7] + and r0, r0, r2 + vbic q0, q10, q4 @ set each d[8..15] to zero if it should not be filtered because clip[8..15] == 0 || a0[8..15] >= pq (a3 > a0 case already zeroed by saturating sub) + tst r0, #1 + bne 1f @ none of the 16 pixel pairs should be updated in this case + vmla.i16 q6, q2, q3 @ invert d[0..7] depending on clip_sign[0..7] & a0_sign[0..7], or zero it if they match, and accumulate into P5[0..7] + vmls.i16 q9, q0, q5 @ invert d[8..15] depending on clip_sign[8..15] & a0_sign[8..15], or zero it if they match, and accumulate into P4[8..15] + vqmovun.s16 d4, q14 + vmla.i16 q1, q0, q5 @ invert d[8..15] depending on clip_sign[8..15] & a0_sign[8..15], or zero it if they match, and accumulate into P5[8..15] + vqmovun.s16 d0, q6 + vqmovun.s16 d5, q9 + vqmovun.s16 d1, q1 + vst1.64 {q2}, [r3 :128], r1 + vst1.64 {q0}, [r3 :128] +1: vpop {d8-d15} + bx lr +endfunc + +@ VC-1 in-loop deblocking filter for 16 pixel pairs at boundary of horizontally-neighbouring blocks +@ On entry: +@ r0 -> top-left pel of right block +@ r1 = row stride, bytes +@ r2 = PQUANT bitstream parameter +function ff_vc1_h_loop_filter16_neon, export=1 + push {r4-r6,lr} + vpush {d8-d15} + sub r3, r0, #4 @ where to start reading + vldr d0, .Lcoeffs + vld1.32 {d2}, [r3], r1 @ P1[0], P2[0]... + sub r0, r0, #1 @ where to start writing + vld1.32 {d3}, [r3], r1 + add r4, r0, r1, lsl #2 + vld1.32 {d10}, [r3], r1 + vld1.32 {d11}, [r3], r1 + vld1.32 {d16}, [r3], r1 + vld1.32 {d4}, [r3], r1 + vld1.32 {d8}, [r3], r1 + vtrn.8 d2, d3 @ P1[0], P1[1], P3[0]... P2[0], P2[1], P4[0]... + vld1.32 {d14}, [r3], r1 + vld1.32 {d5}, [r3], r1 + vtrn.8 d10, d11 @ P1[2], P1[3], P3[2]... P2[2], P2[3], P4[2]... + vld1.32 {d6}, [r3], r1 + vld1.32 {d12}, [r3], r1 + vtrn.8 d16, d4 @ P1[4], P1[5], P3[4]... P2[4], P2[5], P4[4]... + vld1.32 {d13}, [r3], r1 + vtrn.16 d2, d10 @ P1[0], P1[1], P1[2], P1[3], P5[0]... P3[0], P3[1], P3[2], P3[3], P7[0]... + vld1.32 {d1}, [r3], r1 + vtrn.8 d8, d14 @ P1[6], P1[7], P3[6]... P2[6], P2[7], P4[6]... + vld1.32 {d7}, [r3], r1 + vtrn.16 d3, d11 @ P2[0], P2[1], P2[2], P2[3], P6[0]... P4[0], P4[1], P4[2], P4[3], P8[0]... + vld1.32 {d9}, [r3], r1 + vtrn.8 d5, d6 @ P1[8], P1[9], P3[8]... P2[8], P2[9], P4[8]... + vld1.32 {d15}, [r3] + vtrn.16 d16, d8 @ P1[4], P1[5], P1[6], P1[7], P5[4]... P3[4], P3[5], P3[6], P3[7], P7[4]... + vtrn.16 d4, d14 @ P2[4], P2[5], P2[6], P2[7], P6[4]... P4[4], P4[5], P4[6], P4[7], P8[4]... + vtrn.8 d12, d13 @ P1[10], P1[11], P3[10]... P2[10], P2[11], P4[10]... + vdup.16 q9, r2 @ pq + vtrn.8 d1, d7 @ P1[12], P1[13], P3[12]... P2[12], P2[13], P4[12]... + vtrn.32 d2, d16 @ P1[0..7], P5[0..7] + vtrn.16 d5, d12 @ P1[8], P1[7], P1[10], P1[11], P5[8]... P3[8], P3[9], P3[10], P3[11], P7[8]... + vtrn.16 d6, d13 @ P2[8], P2[7], P2[10], P2[11], P6[8]... P4[8], P4[9], P4[10], P4[11], P8[8]... + vtrn.8 d9, d15 @ P1[14], P1[15], P3[14]... P2[14], P2[15], P4[14]... + vtrn.32 d3, d4 @ P2[0..7], P6[0..7] + vshll.u8 q10, d2, #1 @ 2*P1[0..7] + vtrn.32 d10, d8 @ P3[0..7], P7[0..7] + vshll.u8 q11, d16, #1 @ 2*P5[0..7] + vtrn.32 d11, d14 @ P4[0..7], P8[0..7] + vtrn.16 d1, d9 @ P1[12], P1[13], P1[14], P1[15], P5[12]... P3[12], P3[13], P3[14], P3[15], P7[12]... + vtrn.16 d7, d15 @ P2[12], P2[13], P2[14], P2[15], P6[12]... P4[12], P4[13], P4[14], P4[15], P8[12]... + vmovl.u8 q1, d3 @ P2[0..7] + vmovl.u8 q12, d4 @ P6[0..7] + vtrn.32 d5, d1 @ P1[8..15], P5[8..15] + vtrn.32 d6, d7 @ P2[8..15], P6[8..15] + vtrn.32 d12, d9 @ P3[8..15], P7[8..15] + vtrn.32 d13, d15 @ P4[8..15], P8[8..15] + vmls.i16 q10, q1, d0[1] @ 2*P1[0..7]-5*P2[0..7] + vmovl.u8 q1, d10 @ P3[0..7] + vshll.u8 q2, d5, #1 @ 2*P1[8..15] + vshll.u8 q13, d1, #1 @ 2*P5[8..15] + vmls.i16 q11, q12, d0[1] @ 2*P5[0..7]-5*P6[0..7] + vmovl.u8 q14, d6 @ P2[8..15] + vmovl.u8 q3, d7 @ P6[8..15] + vmovl.u8 q15, d8 @ P7[0..7] + vmla.i16 q10, q1, d0[1] @ 2*P1[0..7]-5*P2[0..7]+5*P3[0..7] + vmovl.u8 q1, d12 @ P3[8..15] + vmls.i16 q2, q14, d0[1] @ 2*P1[8..15]-5*P2[8..15] + vmovl.u8 q4, d9 @ P7[8..15] + vshll.u8 q14, d10, #1 @ 2*P3[0..7] + vmls.i16 q13, q3, d0[1] @ 2*P5[8..15]-5*P6[8..15] + vmovl.u8 q5, d11 @ P4[0..7] + vmla.i16 q11, q15, d0[1] @ 2*P5[0..7]-5*P6[0..7]+5*P7[0..7] + vshll.u8 q15, d12, #1 @ 2*P3[8..15] + vmovl.u8 q6, d13 @ P4[8..15] + vmla.i16 q2, q1, d0[1] @ 2*P1[8..15]-5*P2[8..15]+5*P3[8..15] + vmovl.u8 q1, d14 @ P8[0..7] + vmovl.u8 q7, d15 @ P8[8..15] + vmla.i16 q13, q4, d0[1] @ 2*P5[8..15]-5*P6[8..15]+5*P7[8..15] + vmovl.u8 q4, d16 @ P5[0..7] + vmovl.u8 q8, d1 @ P5[8..15] + vmls.i16 q14, q5, d0[1] @ 2*P3[0..7]-5*P4[0..7] + vmls.i16 q15, q6, d0[1] @ 2*P3[8..15]-5*P4[8..15] + vmls.i16 q10, q5, d0[0] @ 2*P1[0..7]-5*P2[0..7]+5*P3[0..7]-2*P4[0..7] + vmls.i16 q11, q1, d0[0] @ 2*P5[0..7]-5*P6[0..7]+5*P7[0..7]-2*P8[0..7] + vsub.i16 q1, q5, q4 @ P4[0..7]-P5[0..7] + vmls.i16 q2, q6, d0[0] @ 2*P1[8..15]-5*P2[8..15]+5*P3[8..15]-2*P4[8..15] + vrshr.s16 q10, q10, #3 + vmls.i16 q13, q7, d0[0] @ 2*P5[8..15]-5*P6[8..15]+5*P7[8..15]-2*P8[8..15] + vsub.i16 q7, q6, q8 @ P4[8..15]-P5[8..15] + vrshr.s16 q11, q11, #3 + vmla.s16 q14, q4, d0[1] @ 2*P3[0..7]-5*P4[0..7]+5*P5[0..7] + vrshr.s16 q2, q2, #3 + vmla.i16 q15, q8, d0[1] @ 2*P3[8..15]-5*P4[8..15]+5*P5[8..15] + vabs.s16 q10, q10 @ a1[0..7] + vrshr.s16 q13, q13, #3 + vmls.i16 q15, q3, d0[0] @ 2*P3[8..15]-5*P4[8..15]+5*P5[8..15]-2*P6[8..15] + vabs.s16 q3, q11 @ a2[0..7] + vabs.s16 q2, q2 @ a1[8..15] + vmls.i16 q14, q12, d0[0] @ 2*P3[0..7]-5*P4[0..7]+5*P5[0..7]-2*P6[0..7] + vabs.s16 q11, q1 + vabs.s16 q12, q13 @ a2[8..15] + vcge.s16 q13, q10, q3 @ test a1[0..7] >= a2[0..7] + vshr.s16 q1, q1, #8 @ clip_sign[0..7] + vrshr.s16 q15, q15, #3 + vshr.s16 q11, q11, #1 @ clip[0..7] + vrshr.s16 q14, q14, #3 + vbsl q13, q3, q10 @ a3[0..7] + vcge.s16 q3, q2, q12 @ test a1[8..15] >= a2[8.15] + vabs.s16 q10, q15 @ a0[8..15] + vshr.s16 q15, q15, #8 @ a0_sign[8..15] + vbsl q3, q12, q2 @ a3[8..15] + vabs.s16 q2, q14 @ a0[0..7] + vabs.s16 q12, q7 + vshr.s16 q7, q7, #8 @ clip_sign[8..15] + vshr.s16 q14, q14, #8 @ a0_sign[0..7] + vshr.s16 q12, q12, #1 @ clip[8..15] + vsub.i16 q7, q7, q15 @ clip_sign[8..15] - a0_sign[8..15] + vqsub.u16 q15, q10, q3 @ a0[8..15] >= a3[8..15] ? a0[8..15]-a3[8..15] : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + vcge.s16 q3, q3, q10 @ test a3[8..15] >= a0[8..15] + vcge.s16 q10, q10, q9 @ test a0[8..15] >= pq + vcge.s16 q9, q2, q9 @ test a0[0..7] >= pq + vsub.i16 q1, q1, q14 @ clip_sign[0..7] - a0_sign[0..7] + vqsub.u16 q14, q2, q13 @ a0[0..7] >= a3[0..7] ? a0[0..7]-a3[0..7] : 0 (a0 > a3 in all cases where filtering is enabled, so makes more sense to subtract this way round than the opposite and then taking the abs) + vcge.s16 q2, q13, q2 @ test a3[0..7] >= a0[0..7] + vmul.i16 q13, q15, d0[1] @ a0[8..15] >= a3[8..15] ? 5*(a0[8..15]-a3[8..15]) : 0 + vceq.i16 q15, q11, #0 @ test clip[0..7] == 0 + vmul.i16 q0, q14, d0[1] @ a0[0..7] >= a3[0..7] ? 5*(a0[0..7]-a3[0..7]) : 0 + vorr q9, q15, q9 @ test clip[0..7] == 0 || a0[0..7] >= pq + vceq.i16 q14, q12, #0 @ test clip[8..15] == 0 + vshr.u16 q13, q13, #3 @ a0[8..15] >= a3[8..15] ? (5*(a0[8..15]-a3[8..15]))>>3 : 0 + vorr q2, q9, q2 @ test clip[0..7] == 0 || a0[0..7] >= pq || a3[0..7] >= a0[0..7] + vshr.u16 q0, q0, #3 @ a0[0..7] >= a3[0..7] ? (5*(a0[0..7]-a3[0..7]))>>3 : 0 + vorr q10, q14, q10 @ test clip[8..15] == 0 || a0[8..15] >= pq + vcge.s16 q14, q13, q12 + vmov.32 r2, d4[1] @ move to gp reg + vorr q3, q10, q3 @ test clip[8..15] == 0 || a0[8..15] >= pq || a3[8..15] >= a0[8..15] + vmov.32 r3, d5[1] + vcge.s16 q2, q0, q11 + vbsl q14, q12, q13 @ FFMIN(d[8..15], clip[8..15]) + vbsl q2, q11, q0 @ FFMIN(d[0..7], clip[0..7]) + vmov.32 r5, d6[1] + vbic q0, q14, q10 @ set each d[8..15] to zero if it should not be filtered because clip[8..15] == 0 || a0[8..15] >= pq (a3 > a0 case already zeroed by saturating sub) + vmov.32 r6, d7[1] + and r12, r2, r3 + vbic q2, q2, q9 @ set each d[0..7] to zero if it should not be filtered because clip[0..7] == 0 || a0[0..7] >= pq (a3 > a0 case already zeroed by saturating sub) + vmls.i16 q6, q0, q7 @ invert d[8..15] depending on clip_sign[8..15] & a0_sign[8..15], or zero it if they match, and accumulate into P4 + vmls.i16 q5, q2, q1 @ invert d[0..7] depending on clip_sign[0..7] & a0_sign[0..7], or zero it if they match, and accumulate into P4 + and r14, r5, r6 + vmla.i16 q4, q2, q1 @ invert d[0..7] depending on clip_sign[0..7] & a0_sign[0..7], or zero it if they match, and accumulate into P5 + and r12, r12, r14 + vqmovun.s16 d4, q6 + vmla.i16 q8, q0, q7 @ invert d[8..15] depending on clip_sign[8..15] & a0_sign[8..15], or zero it if they match, and accumulate into P5 + tst r12, #1 + bne 4f @ none of the 16 pixel pairs should be updated in this case + vqmovun.s16 d2, q5 + vqmovun.s16 d3, q4 + vqmovun.s16 d5, q8 + tst r2, #1 + bne 1f + vst2.8 {d2[0], d3[0]}, [r0], r1 + vst2.8 {d2[1], d3[1]}, [r0], r1 + vst2.8 {d2[2], d3[2]}, [r0], r1 + vst2.8 {d2[3], d3[3]}, [r0] +1: add r0, r4, r1, lsl #2 + tst r3, #1 + bne 2f + vst2.8 {d2[4], d3[4]}, [r4], r1 + vst2.8 {d2[5], d3[5]}, [r4], r1 + vst2.8 {d2[6], d3[6]}, [r4], r1 + vst2.8 {d2[7], d3[7]}, [r4] +2: add r4, r0, r1, lsl #2 + tst r5, #1 + bne 3f + vst2.8 {d4[0], d5[0]}, [r0], r1 + vst2.8 {d4[1], d5[1]}, [r0], r1 + vst2.8 {d4[2], d5[2]}, [r0], r1 + vst2.8 {d4[3], d5[3]}, [r0] +3: tst r6, #1 + bne 4f + vst2.8 {d4[4], d5[4]}, [r4], r1 + vst2.8 {d4[5], d5[5]}, [r4], r1 + vst2.8 {d4[6], d5[6]}, [r4], r1 + vst2.8 {d4[7], d5[7]}, [r4] +4: vpop {d8-d15} + pop {r4-r6,pc} +endfunc From patchwork Thu Mar 31 17:23:48 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35111 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp230235pzb; Thu, 31 Mar 2022 10:25:34 -0700 (PDT) X-Google-Smtp-Source: ABdhPJxRWI2MPRK8zudY5vylDrCsY9BeK0YfrDM0V7FGgeJjZp/nvQilJfzGvQDlliXz+hfOlOyw X-Received: by 2002:a17:906:dc95:b0:6df:d2cf:4d93 with SMTP id cs21-20020a170906dc9500b006dfd2cf4d93mr5784025ejc.66.1648747533803; Thu, 31 Mar 2022 10:25:33 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747533; cv=none; d=google.com; s=arc-20160816; b=uP2kcIGdYuVFNOTgOXcvJFjK/zWNqiZnKVHjDAT2p7U+5THnGXfHird+W8GHaI+Bex 9CKz2mXITDrffKmO7qo0oXa4HQldzVBLAoiziXkissxK6Y2jiyzKFgXUHicUzx9idJhr U+cZlBO5GZurzoGnE+Y40r8Q7EPJP2CK7AfhnVT2HzWuaHcSB5VwFVYZtK81szBjINoB 1AK/VbY5fShyntj4iKU8dl2Geoe7ZBg7uH5Gr8vJ7ExQ7PQHWxkgHOH/D9ZI7ULjq580 ObGNPtWYsd5oGLQkPJoa11cLAfuq63F9jugO3XQQrizwvRekd3u04ywzo/wMZFb3IFvm Xjmw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=TTOKYSKoVVbgksWQLYxhfFHPjfPJaf+EPI1rqrXuyyk=; b=vA/OuEsmaCmT+SrZR+5ONPIeY7XudeUo2ar9wtw8Xk55IeJrCtMDE/2kGypDMKMmuU iuM4vXxyACHEsOnRsalG++0G62j1GwtHNs2WaTncKXsQCxD+gyMu8xjQib+OGcx4d54F 5oJv9vtCP8saiHP93KLxlnuz2iGAufzjpe0uQPzzBm0VNghPS4EAr4J6NzkaxwH4tu0M WC7fsj5wgkCY9Y30ZtM9gA0YskmCd9+RYS8j7ssRBXTnAAPtDDjgcBr8ja8syjRsI8Hp Qes0yUJblPTEfDDVS9qkKVDr8z+aPvJGqxRh4v37E18Agc4qP3GwIHA5GzrByzf4MUJn HutQ== 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 h5-20020a50e105000000b004190fd858eesi119882edl.432.2022.03.31.10.25.33; Thu, 31 Mar 2022 10:25:33 -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 EE26768B2B7; Thu, 31 Mar 2022 20:24:22 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail149055.authsmtp.co.uk (outmail149055.authsmtp.co.uk [62.13.149.55]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 6124B68B250 for ; Thu, 31 Mar 2022 20:24:18 +0300 (EEST) Received: from punt18.authsmtp.com (punt18.authsmtp.com [62.13.128.225]) by punt15.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOHkJ019862 for ; Thu, 31 Mar 2022 18:24:17 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt18.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOHJQ097097; Thu, 31 Mar 2022 18:24:17 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHOEmx062541 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:15 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:14 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:48 +0100 Message-Id: <20220331172351.550818-8-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 64503ca0-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQtXIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLWwvPwVAent0 ZAhrXHVTWAp8cUN9 QR0FFXBUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdW0Ab30N d0kEM1kVTUsAWSA3 HkJKNC8qVRNZAi8y M1QAB3k6VEcXLEgo KlZpRUgRNQQcDRdb V09EGzMRP1AdRzFD X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 07/10] avcodec/vc1: Arm 64-bit NEON inverse transform fast paths 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: HWPs0LCDudVb checkasm benchmarks on 1.5 GHz Cortex-A72 are as follows. vc1dsp.vc1_inv_trans_4x4_c: 158.2 vc1dsp.vc1_inv_trans_4x4_neon: 65.7 vc1dsp.vc1_inv_trans_4x4_dc_c: 86.5 vc1dsp.vc1_inv_trans_4x4_dc_neon: 26.5 vc1dsp.vc1_inv_trans_4x8_c: 335.2 vc1dsp.vc1_inv_trans_4x8_neon: 106.2 vc1dsp.vc1_inv_trans_4x8_dc_c: 151.2 vc1dsp.vc1_inv_trans_4x8_dc_neon: 25.5 vc1dsp.vc1_inv_trans_8x4_c: 365.7 vc1dsp.vc1_inv_trans_8x4_neon: 97.2 vc1dsp.vc1_inv_trans_8x4_dc_c: 139.7 vc1dsp.vc1_inv_trans_8x4_dc_neon: 16.5 vc1dsp.vc1_inv_trans_8x8_c: 547.7 vc1dsp.vc1_inv_trans_8x8_neon: 137.0 vc1dsp.vc1_inv_trans_8x8_dc_c: 268.2 vc1dsp.vc1_inv_trans_8x8_dc_neon: 30.5 Signed-off-by: Ben Avison --- libavcodec/aarch64/vc1dsp_init_aarch64.c | 19 + libavcodec/aarch64/vc1dsp_neon.S | 678 +++++++++++++++++++++++ 2 files changed, 697 insertions(+) diff --git a/libavcodec/aarch64/vc1dsp_init_aarch64.c b/libavcodec/aarch64/vc1dsp_init_aarch64.c index 8f96e4802d..e0eb52dd63 100644 --- a/libavcodec/aarch64/vc1dsp_init_aarch64.c +++ b/libavcodec/aarch64/vc1dsp_init_aarch64.c @@ -25,6 +25,16 @@ #include "config.h" +void ff_vc1_inv_trans_8x8_neon(int16_t *block); +void ff_vc1_inv_trans_8x4_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); +void ff_vc1_inv_trans_4x8_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); +void ff_vc1_inv_trans_4x4_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); + +void ff_vc1_inv_trans_8x8_dc_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); +void ff_vc1_inv_trans_8x4_dc_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); +void ff_vc1_inv_trans_4x8_dc_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); +void ff_vc1_inv_trans_4x4_dc_neon(uint8_t *dest, ptrdiff_t stride, int16_t *block); + void ff_vc1_v_loop_filter4_neon(uint8_t *src, ptrdiff_t stride, int pq); void ff_vc1_h_loop_filter4_neon(uint8_t *src, ptrdiff_t stride, int pq); void ff_vc1_v_loop_filter8_neon(uint8_t *src, ptrdiff_t stride, int pq); @@ -46,6 +56,15 @@ av_cold void ff_vc1dsp_init_aarch64(VC1DSPContext *dsp) int cpu_flags = av_get_cpu_flags(); if (have_neon(cpu_flags)) { + dsp->vc1_inv_trans_8x8 = ff_vc1_inv_trans_8x8_neon; + dsp->vc1_inv_trans_8x4 = ff_vc1_inv_trans_8x4_neon; + dsp->vc1_inv_trans_4x8 = ff_vc1_inv_trans_4x8_neon; + dsp->vc1_inv_trans_4x4 = ff_vc1_inv_trans_4x4_neon; + dsp->vc1_inv_trans_8x8_dc = ff_vc1_inv_trans_8x8_dc_neon; + dsp->vc1_inv_trans_8x4_dc = ff_vc1_inv_trans_8x4_dc_neon; + dsp->vc1_inv_trans_4x8_dc = ff_vc1_inv_trans_4x8_dc_neon; + dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_neon; + dsp->vc1_v_loop_filter4 = ff_vc1_v_loop_filter4_neon; dsp->vc1_h_loop_filter4 = ff_vc1_h_loop_filter4_neon; dsp->vc1_v_loop_filter8 = ff_vc1_v_loop_filter8_neon; diff --git a/libavcodec/aarch64/vc1dsp_neon.S b/libavcodec/aarch64/vc1dsp_neon.S index 1ea9fa75ff..0201db4f78 100644 --- a/libavcodec/aarch64/vc1dsp_neon.S +++ b/libavcodec/aarch64/vc1dsp_neon.S @@ -22,7 +22,685 @@ #include "libavutil/aarch64/asm.S" +// VC-1 8x8 inverse transform +// On entry: +// x0 -> array of 16-bit inverse transform coefficients, in column-major order +// On exit: +// array at x0 updated to hold transformed block; also now held in row-major order +function ff_vc1_inv_trans_8x8_neon, export=1 + ld1 {v1.16b, v2.16b}, [x0], #32 + ld1 {v3.16b, v4.16b}, [x0], #32 + ld1 {v5.16b, v6.16b}, [x0], #32 + shl v1.8h, v1.8h, #2 // 8/2 * src[0] + sub x1, x0, #3*32 + ld1 {v16.16b, v17.16b}, [x0] + shl v7.8h, v2.8h, #4 // 16 * src[8] + shl v18.8h, v2.8h, #2 // 4 * src[8] + shl v19.8h, v4.8h, #4 // 16 * src[24] + ldr d0, .Lcoeffs_it8 + shl v5.8h, v5.8h, #2 // 8/2 * src[32] + shl v20.8h, v6.8h, #4 // 16 * src[40] + shl v21.8h, v6.8h, #2 // 4 * src[40] + shl v22.8h, v17.8h, #4 // 16 * src[56] + ssra v20.8h, v19.8h, #2 // 4 * src[24] + 16 * src[40] + mul v23.8h, v3.8h, v0.h[0] // 6/2 * src[16] + sub v19.8h, v19.8h, v21.8h // 16 * src[24] - 4 * src[40] + ssra v7.8h, v22.8h, #2 // 16 * src[8] + 4 * src[56] + sub v18.8h, v22.8h, v18.8h // - 4 * src[8] + 16 * src[56] + shl v3.8h, v3.8h, #3 // 16/2 * src[16] + mls v20.8h, v2.8h, v0.h[2] // - 15 * src[8] + 4 * src[24] + 16 * src[40] + ssra v1.8h, v1.8h, #1 // 12/2 * src[0] + ssra v5.8h, v5.8h, #1 // 12/2 * src[32] + mla v7.8h, v4.8h, v0.h[2] // 16 * src[8] + 15 * src[24] + 4 * src[56] + shl v21.8h, v16.8h, #3 // 16/2 * src[48] + mls v19.8h, v2.8h, v0.h[1] // - 9 * src[8] + 16 * src[24] - 4 * src[40] + sub v2.8h, v23.8h, v21.8h // t4/2 = 6/2 * src[16] - 16/2 * src[48] + mla v18.8h, v4.8h, v0.h[1] // - 4 * src[8] + 9 * src[24] + 16 * src[56] + add v4.8h, v1.8h, v5.8h // t1/2 = 12/2 * src[0] + 12/2 * src[32] + sub v1.8h, v1.8h, v5.8h // t2/2 = 12/2 * src[0] - 12/2 * src[32] + mla v3.8h, v16.8h, v0.h[0] // t3/2 = 16/2 * src[16] + 6/2 * src[48] + mla v7.8h, v6.8h, v0.h[1] // t1 = 16 * src[8] + 15 * src[24] + 9 * src[40] + 4 * src[56] + add v5.8h, v1.8h, v2.8h // t6/2 = t2/2 + t4/2 + sub v16.8h, v1.8h, v2.8h // t7/2 = t2/2 - t4/2 + mla v20.8h, v17.8h, v0.h[1] // -t2 = - 15 * src[8] + 4 * src[24] + 16 * src[40] + 9 * src[56] + add v21.8h, v1.8h, v2.8h // t6/2 = t2/2 + t4/2 + add v22.8h, v4.8h, v3.8h // t5/2 = t1/2 + t3/2 + mls v19.8h, v17.8h, v0.h[2] // -t3 = - 9 * src[8] + 16 * src[24] - 4 * src[40] - 15 * src[56] + sub v17.8h, v4.8h, v3.8h // t8/2 = t1/2 - t3/2 + add v23.8h, v4.8h, v3.8h // t5/2 = t1/2 + t3/2 + mls v18.8h, v6.8h, v0.h[2] // -t4 = - 4 * src[8] + 9 * src[24] - 15 * src[40] + 16 * src[56] + sub v1.8h, v1.8h, v2.8h // t7/2 = t2/2 - t4/2 + sub v2.8h, v4.8h, v3.8h // t8/2 = t1/2 - t3/2 + neg v3.8h, v7.8h // -t1 + neg v4.8h, v20.8h // +t2 + neg v6.8h, v19.8h // +t3 + ssra v22.8h, v7.8h, #1 // (t5 + t1) >> 1 + ssra v1.8h, v19.8h, #1 // (t7 - t3) >> 1 + neg v7.8h, v18.8h // +t4 + ssra v5.8h, v4.8h, #1 // (t6 + t2) >> 1 + ssra v16.8h, v6.8h, #1 // (t7 + t3) >> 1 + ssra v2.8h, v18.8h, #1 // (t8 - t4) >> 1 + ssra v17.8h, v7.8h, #1 // (t8 + t4) >> 1 + ssra v21.8h, v20.8h, #1 // (t6 - t2) >> 1 + ssra v23.8h, v3.8h, #1 // (t5 - t1) >> 1 + srshr v3.8h, v22.8h, #2 // (t5 + t1 + 4) >> 3 + srshr v4.8h, v5.8h, #2 // (t6 + t2 + 4) >> 3 + srshr v5.8h, v16.8h, #2 // (t7 + t3 + 4) >> 3 + srshr v6.8h, v17.8h, #2 // (t8 + t4 + 4) >> 3 + srshr v2.8h, v2.8h, #2 // (t8 - t4 + 4) >> 3 + srshr v1.8h, v1.8h, #2 // (t7 - t3 + 4) >> 3 + srshr v7.8h, v21.8h, #2 // (t6 - t2 + 4) >> 3 + srshr v16.8h, v23.8h, #2 // (t5 - t1 + 4) >> 3 + trn2 v17.8h, v3.8h, v4.8h + trn2 v18.8h, v5.8h, v6.8h + trn2 v19.8h, v2.8h, v1.8h + trn2 v20.8h, v7.8h, v16.8h + trn1 v21.4s, v17.4s, v18.4s + trn2 v17.4s, v17.4s, v18.4s + trn1 v18.4s, v19.4s, v20.4s + trn2 v19.4s, v19.4s, v20.4s + trn1 v3.8h, v3.8h, v4.8h + trn2 v4.2d, v21.2d, v18.2d + trn1 v20.2d, v17.2d, v19.2d + trn1 v5.8h, v5.8h, v6.8h + trn1 v1.8h, v2.8h, v1.8h + trn1 v2.8h, v7.8h, v16.8h + trn1 v6.2d, v21.2d, v18.2d + trn2 v7.2d, v17.2d, v19.2d + shl v16.8h, v20.8h, #4 // 16 * src[24] + shl v17.8h, v4.8h, #4 // 16 * src[40] + trn1 v18.4s, v3.4s, v5.4s + trn1 v19.4s, v1.4s, v2.4s + shl v21.8h, v7.8h, #4 // 16 * src[56] + shl v22.8h, v6.8h, #2 // 4 * src[8] + shl v23.8h, v4.8h, #2 // 4 * src[40] + trn2 v3.4s, v3.4s, v5.4s + trn2 v1.4s, v1.4s, v2.4s + shl v2.8h, v6.8h, #4 // 16 * src[8] + sub v5.8h, v16.8h, v23.8h // 16 * src[24] - 4 * src[40] + ssra v17.8h, v16.8h, #2 // 4 * src[24] + 16 * src[40] + sub v16.8h, v21.8h, v22.8h // - 4 * src[8] + 16 * src[56] + trn1 v22.2d, v18.2d, v19.2d + trn2 v18.2d, v18.2d, v19.2d + trn1 v19.2d, v3.2d, v1.2d + ssra v2.8h, v21.8h, #2 // 16 * src[8] + 4 * src[56] + mls v17.8h, v6.8h, v0.h[2] // - 15 * src[8] + 4 * src[24] + 16 * src[40] + shl v21.8h, v22.8h, #2 // 8/2 * src[0] + shl v18.8h, v18.8h, #2 // 8/2 * src[32] + mls v5.8h, v6.8h, v0.h[1] // - 9 * src[8] + 16 * src[24] - 4 * src[40] + shl v6.8h, v19.8h, #3 // 16/2 * src[16] + trn2 v1.2d, v3.2d, v1.2d + mla v16.8h, v20.8h, v0.h[1] // - 4 * src[8] + 9 * src[24] + 16 * src[56] + ssra v21.8h, v21.8h, #1 // 12/2 * src[0] + ssra v18.8h, v18.8h, #1 // 12/2 * src[32] + mul v3.8h, v19.8h, v0.h[0] // 6/2 * src[16] + shl v19.8h, v1.8h, #3 // 16/2 * src[48] + mla v2.8h, v20.8h, v0.h[2] // 16 * src[8] + 15 * src[24] + 4 * src[56] + add v20.8h, v21.8h, v18.8h // t1/2 = 12/2 * src[0] + 12/2 * src[32] + mla v6.8h, v1.8h, v0.h[0] // t3/2 = 16/2 * src[16] + 6/2 * src[48] + sub v1.8h, v21.8h, v18.8h // t2/2 = 12/2 * src[0] - 12/2 * src[32] + sub v3.8h, v3.8h, v19.8h // t4/2 = 6/2 * src[16] - 16/2 * src[48] + mla v17.8h, v7.8h, v0.h[1] // -t2 = - 15 * src[8] + 4 * src[24] + 16 * src[40] + 9 * src[56] + mls v5.8h, v7.8h, v0.h[2] // -t3 = - 9 * src[8] + 16 * src[24] - 4 * src[40] - 15 * src[56] + add v7.8h, v1.8h, v3.8h // t6/2 = t2/2 + t4/2 + add v18.8h, v20.8h, v6.8h // t5/2 = t1/2 + t3/2 + mls v16.8h, v4.8h, v0.h[2] // -t4 = - 4 * src[8] + 9 * src[24] - 15 * src[40] + 16 * src[56] + sub v19.8h, v1.8h, v3.8h // t7/2 = t2/2 - t4/2 + neg v21.8h, v17.8h // +t2 + mla v2.8h, v4.8h, v0.h[1] // t1 = 16 * src[8] + 15 * src[24] + 9 * src[40] + 4 * src[56] + sub v0.8h, v20.8h, v6.8h // t8/2 = t1/2 - t3/2 + neg v4.8h, v5.8h // +t3 + sub v22.8h, v1.8h, v3.8h // t7/2 = t2/2 - t4/2 + sub v23.8h, v20.8h, v6.8h // t8/2 = t1/2 - t3/2 + neg v24.8h, v16.8h // +t4 + add v6.8h, v20.8h, v6.8h // t5/2 = t1/2 + t3/2 + add v1.8h, v1.8h, v3.8h // t6/2 = t2/2 + t4/2 + ssra v7.8h, v21.8h, #1 // (t6 + t2) >> 1 + neg v3.8h, v2.8h // -t1 + ssra v18.8h, v2.8h, #1 // (t5 + t1) >> 1 + ssra v19.8h, v4.8h, #1 // (t7 + t3) >> 1 + ssra v0.8h, v24.8h, #1 // (t8 + t4) >> 1 + srsra v23.8h, v16.8h, #1 // (t8 - t4 + 1) >> 1 + srsra v22.8h, v5.8h, #1 // (t7 - t3 + 1) >> 1 + srsra v1.8h, v17.8h, #1 // (t6 - t2 + 1) >> 1 + srsra v6.8h, v3.8h, #1 // (t5 - t1 + 1) >> 1 + srshr v2.8h, v18.8h, #6 // (t5 + t1 + 64) >> 7 + srshr v3.8h, v7.8h, #6 // (t6 + t2 + 64) >> 7 + srshr v4.8h, v19.8h, #6 // (t7 + t3 + 64) >> 7 + srshr v5.8h, v0.8h, #6 // (t8 + t4 + 64) >> 7 + srshr v16.8h, v23.8h, #6 // (t8 - t4 + 65) >> 7 + srshr v17.8h, v22.8h, #6 // (t7 - t3 + 65) >> 7 + st1 {v2.16b, v3.16b}, [x1], #32 + srshr v0.8h, v1.8h, #6 // (t6 - t2 + 65) >> 7 + srshr v1.8h, v6.8h, #6 // (t5 - t1 + 65) >> 7 + st1 {v4.16b, v5.16b}, [x1], #32 + st1 {v16.16b, v17.16b}, [x1], #32 + st1 {v0.16b, v1.16b}, [x1] + ret +endfunc + +// VC-1 8x4 inverse transform +// On entry: +// x0 -> array of 8-bit samples, in row-major order +// x1 = row stride for 8-bit sample array +// x2 -> array of 16-bit inverse transform coefficients, in row-major order +// On exit: +// array at x0 updated by saturated addition of (narrowed) transformed block +function ff_vc1_inv_trans_8x4_neon, export=1 + ld1 {v1.8b, v2.8b, v3.8b, v4.8b}, [x2], #32 + mov x3, x0 + ld1 {v16.8b, v17.8b, v18.8b, v19.8b}, [x2] + ldr q0, .Lcoeffs_it8 // includes 4-point coefficients in upper half of vector + ld1 {v5.8b}, [x0], x1 + trn2 v6.4h, v1.4h, v3.4h + trn2 v7.4h, v2.4h, v4.4h + trn1 v1.4h, v1.4h, v3.4h + trn1 v2.4h, v2.4h, v4.4h + trn2 v3.4h, v16.4h, v18.4h + trn2 v4.4h, v17.4h, v19.4h + trn1 v16.4h, v16.4h, v18.4h + trn1 v17.4h, v17.4h, v19.4h + ld1 {v18.8b}, [x0], x1 + trn1 v19.2s, v6.2s, v3.2s + trn2 v3.2s, v6.2s, v3.2s + trn1 v6.2s, v7.2s, v4.2s + trn2 v4.2s, v7.2s, v4.2s + trn1 v7.2s, v1.2s, v16.2s + trn1 v20.2s, v2.2s, v17.2s + shl v21.4h, v19.4h, #4 // 16 * src[1] + trn2 v1.2s, v1.2s, v16.2s + shl v16.4h, v3.4h, #4 // 16 * src[3] + trn2 v2.2s, v2.2s, v17.2s + shl v17.4h, v6.4h, #4 // 16 * src[5] + ld1 {v22.8b}, [x0], x1 + shl v23.4h, v4.4h, #4 // 16 * src[7] + mul v24.4h, v1.4h, v0.h[0] // 6/2 * src[2] + ld1 {v25.8b}, [x0] + shl v26.4h, v19.4h, #2 // 4 * src[1] + shl v27.4h, v6.4h, #2 // 4 * src[5] + ssra v21.4h, v23.4h, #2 // 16 * src[1] + 4 * src[7] + ssra v17.4h, v16.4h, #2 // 4 * src[3] + 16 * src[5] + sub v23.4h, v23.4h, v26.4h // - 4 * src[1] + 16 * src[7] + sub v16.4h, v16.4h, v27.4h // 16 * src[3] - 4 * src[5] + shl v7.4h, v7.4h, #2 // 8/2 * src[0] + shl v20.4h, v20.4h, #2 // 8/2 * src[4] + mla v21.4h, v3.4h, v0.h[2] // 16 * src[1] + 15 * src[3] + 4 * src[7] + shl v1.4h, v1.4h, #3 // 16/2 * src[2] + mls v17.4h, v19.4h, v0.h[2] // - 15 * src[1] + 4 * src[3] + 16 * src[5] + ssra v7.4h, v7.4h, #1 // 12/2 * src[0] + mls v16.4h, v19.4h, v0.h[1] // - 9 * src[1] + 16 * src[3] - 4 * src[5] + ssra v20.4h, v20.4h, #1 // 12/2 * src[4] + mla v23.4h, v3.4h, v0.h[1] // - 4 * src[1] + 9 * src[3] + 16 * src[7] + shl v3.4h, v2.4h, #3 // 16/2 * src[6] + mla v1.4h, v2.4h, v0.h[0] // t3/2 = 16/2 * src[2] + 6/2 * src[6] + mla v21.4h, v6.4h, v0.h[1] // t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7] + mla v17.4h, v4.4h, v0.h[1] // -t2 = - 15 * src[1] + 4 * src[3] + 16 * src[5] + 9 * src[7] + sub v2.4h, v24.4h, v3.4h // t4/2 = 6/2 * src[2] - 16/2 * src[6] + mls v16.4h, v4.4h, v0.h[2] // -t3 = - 9 * src[1] + 16 * src[3] - 4 * src[5] - 15 * src[7] + add v3.4h, v7.4h, v20.4h // t1/2 = 12/2 * src[0] + 12/2 * src[4] + mls v23.4h, v6.4h, v0.h[2] // -t4 = - 4 * src[1] + 9 * src[3] - 15 * src[5] + 16 * src[7] + sub v4.4h, v7.4h, v20.4h // t2/2 = 12/2 * src[0] - 12/2 * src[4] + neg v6.4h, v21.4h // -t1 + add v7.4h, v3.4h, v1.4h // t5/2 = t1/2 + t3/2 + sub v19.4h, v3.4h, v1.4h // t8/2 = t1/2 - t3/2 + add v20.4h, v4.4h, v2.4h // t6/2 = t2/2 + t4/2 + sub v24.4h, v4.4h, v2.4h // t7/2 = t2/2 - t4/2 + add v26.4h, v3.4h, v1.4h // t5/2 = t1/2 + t3/2 + add v27.4h, v4.4h, v2.4h // t6/2 = t2/2 + t4/2 + sub v2.4h, v4.4h, v2.4h // t7/2 = t2/2 - t4/2 + sub v1.4h, v3.4h, v1.4h // t8/2 = t1/2 - t3/2 + neg v3.4h, v17.4h // +t2 + neg v4.4h, v16.4h // +t3 + neg v28.4h, v23.4h // +t4 + ssra v7.4h, v21.4h, #1 // (t5 + t1) >> 1 + ssra v1.4h, v23.4h, #1 // (t8 - t4) >> 1 + ssra v20.4h, v3.4h, #1 // (t6 + t2) >> 1 + ssra v24.4h, v4.4h, #1 // (t7 + t3) >> 1 + ssra v19.4h, v28.4h, #1 // (t8 + t4) >> 1 + ssra v2.4h, v16.4h, #1 // (t7 - t3) >> 1 + ssra v27.4h, v17.4h, #1 // (t6 - t2) >> 1 + ssra v26.4h, v6.4h, #1 // (t5 - t1) >> 1 + trn1 v1.2d, v7.2d, v1.2d + trn1 v2.2d, v20.2d, v2.2d + trn1 v3.2d, v24.2d, v27.2d + trn1 v4.2d, v19.2d, v26.2d + srshr v1.8h, v1.8h, #2 // (t5 + t1 + 4) >> 3, (t8 - t4 + 4) >> 3 + srshr v2.8h, v2.8h, #2 // (t6 + t2 + 4) >> 3, (t7 - t3 + 4) >> 3 + srshr v3.8h, v3.8h, #2 // (t7 + t3 + 4) >> 3, (t6 - t2 + 4) >> 3 + srshr v4.8h, v4.8h, #2 // (t8 + t4 + 4) >> 3, (t5 - t1 + 4) >> 3 + trn2 v6.8h, v1.8h, v2.8h + trn1 v1.8h, v1.8h, v2.8h + trn2 v2.8h, v3.8h, v4.8h + trn1 v3.8h, v3.8h, v4.8h + trn2 v4.4s, v6.4s, v2.4s + trn1 v7.4s, v1.4s, v3.4s + trn2 v1.4s, v1.4s, v3.4s + mul v3.8h, v4.8h, v0.h[5] // 22/2 * src[24] + trn1 v2.4s, v6.4s, v2.4s + mul v4.8h, v4.8h, v0.h[4] // 10/2 * src[24] + mul v6.8h, v7.8h, v0.h[6] // 17 * src[0] + mul v1.8h, v1.8h, v0.h[6] // 17 * src[16] + mls v3.8h, v2.8h, v0.h[4] // t4/2 = - 10/2 * src[8] + 22/2 * src[24] + mla v4.8h, v2.8h, v0.h[5] // t3/2 = 22/2 * src[8] + 10/2 * src[24] + add v0.8h, v6.8h, v1.8h // t1 = 17 * src[0] + 17 * src[16] + sub v1.8h, v6.8h, v1.8h // t2 = 17 * src[0] - 17 * src[16] + neg v2.8h, v3.8h // -t4/2 + neg v6.8h, v4.8h // -t3/2 + ssra v4.8h, v0.8h, #1 // (t1 + t3) >> 1 + ssra v2.8h, v1.8h, #1 // (t2 - t4) >> 1 + ssra v3.8h, v1.8h, #1 // (t2 + t4) >> 1 + ssra v6.8h, v0.8h, #1 // (t1 - t3) >> 1 + srshr v0.8h, v4.8h, #6 // (t1 + t3 + 64) >> 7 + srshr v1.8h, v2.8h, #6 // (t2 - t4 + 64) >> 7 + srshr v2.8h, v3.8h, #6 // (t2 + t4 + 64) >> 7 + srshr v3.8h, v6.8h, #6 // (t1 - t3 + 64) >> 7 + uaddw v0.8h, v0.8h, v5.8b + uaddw v1.8h, v1.8h, v18.8b + uaddw v2.8h, v2.8h, v22.8b + uaddw v3.8h, v3.8h, v25.8b + sqxtun v0.8b, v0.8h + sqxtun v1.8b, v1.8h + sqxtun v2.8b, v2.8h + sqxtun v3.8b, v3.8h + st1 {v0.8b}, [x3], x1 + st1 {v1.8b}, [x3], x1 + st1 {v2.8b}, [x3], x1 + st1 {v3.8b}, [x3] + ret +endfunc + +// VC-1 4x8 inverse transform +// On entry: +// x0 -> array of 8-bit samples, in row-major order +// x1 = row stride for 8-bit sample array +// x2 -> array of 16-bit inverse transform coefficients, in row-major order (row stride is 8 coefficients) +// On exit: +// array at x0 updated by saturated addition of (narrowed) transformed block +function ff_vc1_inv_trans_4x8_neon, export=1 + mov x3, #16 + ldr q0, .Lcoeffs_it8 // includes 4-point coefficients in upper half of vector + mov x4, x0 + ld1 {v1.d}[0], [x2], x3 // 00 01 02 03 + ld1 {v2.d}[0], [x2], x3 // 10 11 12 13 + ld1 {v3.d}[0], [x2], x3 // 20 21 22 23 + ld1 {v4.d}[0], [x2], x3 // 30 31 32 33 + ld1 {v1.d}[1], [x2], x3 // 40 41 42 43 + ld1 {v2.d}[1], [x2], x3 // 50 51 52 53 + ld1 {v3.d}[1], [x2], x3 // 60 61 62 63 + ld1 {v4.d}[1], [x2] // 70 71 72 73 + ld1 {v5.s}[0], [x0], x1 + ld1 {v6.s}[0], [x0], x1 + ld1 {v7.s}[0], [x0], x1 + trn2 v16.8h, v1.8h, v2.8h // 01 11 03 13 41 51 43 53 + trn1 v1.8h, v1.8h, v2.8h // 00 10 02 12 40 50 42 52 + trn2 v2.8h, v3.8h, v4.8h // 21 31 23 33 61 71 63 73 + trn1 v3.8h, v3.8h, v4.8h // 20 30 22 32 60 70 62 72 + ld1 {v4.s}[0], [x0], x1 + trn2 v17.4s, v16.4s, v2.4s // 03 13 23 33 43 53 63 73 + trn1 v18.4s, v1.4s, v3.4s // 00 10 20 30 40 50 60 70 + trn1 v2.4s, v16.4s, v2.4s // 01 11 21 31 41 51 61 71 + mul v16.8h, v17.8h, v0.h[4] // 10/2 * src[3] + ld1 {v5.s}[1], [x0], x1 + mul v17.8h, v17.8h, v0.h[5] // 22/2 * src[3] + ld1 {v6.s}[1], [x0], x1 + trn2 v1.4s, v1.4s, v3.4s // 02 12 22 32 42 52 62 72 + mul v3.8h, v18.8h, v0.h[6] // 17 * src[0] + ld1 {v7.s}[1], [x0], x1 + mul v1.8h, v1.8h, v0.h[6] // 17 * src[2] + ld1 {v4.s}[1], [x0] + mla v16.8h, v2.8h, v0.h[5] // t3/2 = 22/2 * src[1] + 10/2 * src[3] + mls v17.8h, v2.8h, v0.h[4] // t4/2 = - 10/2 * src[1] + 22/2 * src[3] + add v2.8h, v3.8h, v1.8h // t1 = 17 * src[0] + 17 * src[2] + sub v1.8h, v3.8h, v1.8h // t2 = 17 * src[0] - 17 * src[2] + neg v3.8h, v16.8h // -t3/2 + ssra v16.8h, v2.8h, #1 // (t1 + t3) >> 1 + neg v18.8h, v17.8h // -t4/2 + ssra v17.8h, v1.8h, #1 // (t2 + t4) >> 1 + ssra v3.8h, v2.8h, #1 // (t1 - t3) >> 1 + ssra v18.8h, v1.8h, #1 // (t2 - t4) >> 1 + srshr v1.8h, v16.8h, #2 // (t1 + t3 + 64) >> 3 + srshr v2.8h, v17.8h, #2 // (t2 + t4 + 64) >> 3 + srshr v3.8h, v3.8h, #2 // (t1 - t3 + 64) >> 3 + srshr v16.8h, v18.8h, #2 // (t2 - t4 + 64) >> 3 + trn2 v17.8h, v2.8h, v3.8h // 12 13 32 33 52 53 72 73 + trn2 v18.8h, v1.8h, v16.8h // 10 11 30 31 50 51 70 71 + trn1 v1.8h, v1.8h, v16.8h // 00 01 20 21 40 41 60 61 + trn1 v2.8h, v2.8h, v3.8h // 02 03 22 23 42 43 62 63 + trn1 v3.4s, v18.4s, v17.4s // 10 11 12 13 50 51 52 53 + trn2 v16.4s, v18.4s, v17.4s // 30 31 32 33 70 71 72 73 + trn1 v17.4s, v1.4s, v2.4s // 00 01 02 03 40 41 42 43 + mov d18, v3.d[1] // 50 51 52 53 + shl v19.4h, v3.4h, #4 // 16 * src[8] + mov d20, v16.d[1] // 70 71 72 73 + shl v21.4h, v16.4h, #4 // 16 * src[24] + mov d22, v17.d[1] // 40 41 42 43 + shl v23.4h, v3.4h, #2 // 4 * src[8] + shl v24.4h, v18.4h, #4 // 16 * src[40] + shl v25.4h, v20.4h, #4 // 16 * src[56] + shl v26.4h, v18.4h, #2 // 4 * src[40] + trn2 v1.4s, v1.4s, v2.4s // 20 21 22 23 60 61 62 63 + ssra v24.4h, v21.4h, #2 // 4 * src[24] + 16 * src[40] + sub v2.4h, v25.4h, v23.4h // - 4 * src[8] + 16 * src[56] + shl v17.4h, v17.4h, #2 // 8/2 * src[0] + sub v21.4h, v21.4h, v26.4h // 16 * src[24] - 4 * src[40] + shl v22.4h, v22.4h, #2 // 8/2 * src[32] + mov d23, v1.d[1] // 60 61 62 63 + ssra v19.4h, v25.4h, #2 // 16 * src[8] + 4 * src[56] + mul v25.4h, v1.4h, v0.h[0] // 6/2 * src[16] + shl v1.4h, v1.4h, #3 // 16/2 * src[16] + mls v24.4h, v3.4h, v0.h[2] // - 15 * src[8] + 4 * src[24] + 16 * src[40] + ssra v17.4h, v17.4h, #1 // 12/2 * src[0] + mls v21.4h, v3.4h, v0.h[1] // - 9 * src[8] + 16 * src[24] - 4 * src[40] + ssra v22.4h, v22.4h, #1 // 12/2 * src[32] + mla v2.4h, v16.4h, v0.h[1] // - 4 * src[8] + 9 * src[24] + 16 * src[56] + shl v3.4h, v23.4h, #3 // 16/2 * src[48] + mla v19.4h, v16.4h, v0.h[2] // 16 * src[8] + 15 * src[24] + 4 * src[56] + mla v1.4h, v23.4h, v0.h[0] // t3/2 = 16/2 * src[16] + 6/2 * src[48] + mla v24.4h, v20.4h, v0.h[1] // -t2 = - 15 * src[8] + 4 * src[24] + 16 * src[40] + 9 * src[56] + add v16.4h, v17.4h, v22.4h // t1/2 = 12/2 * src[0] + 12/2 * src[32] + sub v3.4h, v25.4h, v3.4h // t4/2 = 6/2 * src[16] - 16/2 * src[48] + sub v17.4h, v17.4h, v22.4h // t2/2 = 12/2 * src[0] - 12/2 * src[32] + mls v21.4h, v20.4h, v0.h[2] // -t3 = - 9 * src[8] + 16 * src[24] - 4 * src[40] - 15 * src[56] + mla v19.4h, v18.4h, v0.h[1] // t1 = 16 * src[8] + 15 * src[24] + 9 * src[40] + 4 * src[56] + add v20.4h, v16.4h, v1.4h // t5/2 = t1/2 + t3/2 + mls v2.4h, v18.4h, v0.h[2] // -t4 = - 4 * src[8] + 9 * src[24] - 15 * src[40] + 16 * src[56] + sub v0.4h, v16.4h, v1.4h // t8/2 = t1/2 - t3/2 + add v18.4h, v17.4h, v3.4h // t6/2 = t2/2 + t4/2 + sub v22.4h, v17.4h, v3.4h // t7/2 = t2/2 - t4/2 + neg v23.4h, v24.4h // +t2 + sub v25.4h, v17.4h, v3.4h // t7/2 = t2/2 - t4/2 + add v3.4h, v17.4h, v3.4h // t6/2 = t2/2 + t4/2 + neg v17.4h, v21.4h // +t3 + sub v26.4h, v16.4h, v1.4h // t8/2 = t1/2 - t3/2 + add v1.4h, v16.4h, v1.4h // t5/2 = t1/2 + t3/2 + neg v16.4h, v19.4h // -t1 + neg v27.4h, v2.4h // +t4 + ssra v20.4h, v19.4h, #1 // (t5 + t1) >> 1 + srsra v0.4h, v2.4h, #1 // (t8 - t4 + 1) >> 1 + ssra v18.4h, v23.4h, #1 // (t6 + t2) >> 1 + srsra v22.4h, v21.4h, #1 // (t7 - t3 + 1) >> 1 + ssra v25.4h, v17.4h, #1 // (t7 + t3) >> 1 + srsra v3.4h, v24.4h, #1 // (t6 - t2 + 1) >> 1 + ssra v26.4h, v27.4h, #1 // (t8 + t4) >> 1 + srsra v1.4h, v16.4h, #1 // (t5 - t1 + 1) >> 1 + trn1 v0.2d, v20.2d, v0.2d + trn1 v2.2d, v18.2d, v22.2d + trn1 v3.2d, v25.2d, v3.2d + trn1 v1.2d, v26.2d, v1.2d + srshr v0.8h, v0.8h, #6 // (t5 + t1 + 64) >> 7, (t8 - t4 + 65) >> 7 + srshr v2.8h, v2.8h, #6 // (t6 + t2 + 64) >> 7, (t7 - t3 + 65) >> 7 + srshr v3.8h, v3.8h, #6 // (t7 + t3 + 64) >> 7, (t6 - t2 + 65) >> 7 + srshr v1.8h, v1.8h, #6 // (t8 + t4 + 64) >> 7, (t5 - t1 + 65) >> 7 + uaddw v0.8h, v0.8h, v5.8b + uaddw v2.8h, v2.8h, v6.8b + uaddw v3.8h, v3.8h, v7.8b + uaddw v1.8h, v1.8h, v4.8b + sqxtun v0.8b, v0.8h + sqxtun v2.8b, v2.8h + sqxtun v3.8b, v3.8h + sqxtun v1.8b, v1.8h + st1 {v0.s}[0], [x4], x1 + st1 {v2.s}[0], [x4], x1 + st1 {v3.s}[0], [x4], x1 + st1 {v1.s}[0], [x4], x1 + st1 {v0.s}[1], [x4], x1 + st1 {v2.s}[1], [x4], x1 + st1 {v3.s}[1], [x4], x1 + st1 {v1.s}[1], [x4] + ret +endfunc + +// VC-1 4x4 inverse transform +// On entry: +// x0 -> array of 8-bit samples, in row-major order +// x1 = row stride for 8-bit sample array +// x2 -> array of 16-bit inverse transform coefficients, in row-major order (row stride is 8 coefficients) +// On exit: +// array at x0 updated by saturated addition of (narrowed) transformed block +function ff_vc1_inv_trans_4x4_neon, export=1 + mov x3, #16 + ldr d0, .Lcoeffs_it4 + mov x4, x0 + ld1 {v1.d}[0], [x2], x3 // 00 01 02 03 + ld1 {v2.d}[0], [x2], x3 // 10 11 12 13 + ld1 {v3.d}[0], [x2], x3 // 20 21 22 23 + ld1 {v4.d}[0], [x2] // 30 31 32 33 + ld1 {v5.s}[0], [x0], x1 + ld1 {v5.s}[1], [x0], x1 + ld1 {v6.s}[0], [x0], x1 + trn2 v7.4h, v1.4h, v2.4h // 01 11 03 13 + trn1 v1.4h, v1.4h, v2.4h // 00 10 02 12 + ld1 {v6.s}[1], [x0] + trn2 v2.4h, v3.4h, v4.4h // 21 31 23 33 + trn1 v3.4h, v3.4h, v4.4h // 20 30 22 32 + trn2 v4.2s, v7.2s, v2.2s // 03 13 23 33 + trn1 v16.2s, v1.2s, v3.2s // 00 10 20 30 + trn1 v2.2s, v7.2s, v2.2s // 01 11 21 31 + trn2 v1.2s, v1.2s, v3.2s // 02 12 22 32 + mul v3.4h, v4.4h, v0.h[0] // 10/2 * src[3] + mul v4.4h, v4.4h, v0.h[1] // 22/2 * src[3] + mul v7.4h, v16.4h, v0.h[2] // 17 * src[0] + mul v1.4h, v1.4h, v0.h[2] // 17 * src[2] + mla v3.4h, v2.4h, v0.h[1] // t3/2 = 22/2 * src[1] + 10/2 * src[3] + mls v4.4h, v2.4h, v0.h[0] // t4/2 = - 10/2 * src[1] + 22/2 * src[3] + add v2.4h, v7.4h, v1.4h // t1 = 17 * src[0] + 17 * src[2] + sub v1.4h, v7.4h, v1.4h // t2 = 17 * src[0] - 17 * src[2] + neg v7.4h, v3.4h // -t3/2 + neg v16.4h, v4.4h // -t4/2 + ssra v3.4h, v2.4h, #1 // (t1 + t3) >> 1 + ssra v4.4h, v1.4h, #1 // (t2 + t4) >> 1 + ssra v16.4h, v1.4h, #1 // (t2 - t4) >> 1 + ssra v7.4h, v2.4h, #1 // (t1 - t3) >> 1 + srshr v1.4h, v3.4h, #2 // (t1 + t3 + 64) >> 3 + srshr v2.4h, v4.4h, #2 // (t2 + t4 + 64) >> 3 + srshr v3.4h, v16.4h, #2 // (t2 - t4 + 64) >> 3 + srshr v4.4h, v7.4h, #2 // (t1 - t3 + 64) >> 3 + trn2 v7.4h, v1.4h, v3.4h // 10 11 30 31 + trn1 v1.4h, v1.4h, v3.4h // 00 01 20 21 + trn2 v3.4h, v2.4h, v4.4h // 12 13 32 33 + trn1 v2.4h, v2.4h, v4.4h // 02 03 22 23 + trn2 v4.2s, v7.2s, v3.2s // 30 31 32 33 + trn1 v16.2s, v1.2s, v2.2s // 00 01 02 03 + trn1 v3.2s, v7.2s, v3.2s // 10 11 12 13 + trn2 v1.2s, v1.2s, v2.2s // 20 21 22 23 + mul v2.4h, v4.4h, v0.h[1] // 22/2 * src[24] + mul v4.4h, v4.4h, v0.h[0] // 10/2 * src[24] + mul v7.4h, v16.4h, v0.h[2] // 17 * src[0] + mul v1.4h, v1.4h, v0.h[2] // 17 * src[16] + mls v2.4h, v3.4h, v0.h[0] // t4/2 = - 10/2 * src[8] + 22/2 * src[24] + mla v4.4h, v3.4h, v0.h[1] // t3/2 = 22/2 * src[8] + 10/2 * src[24] + add v0.4h, v7.4h, v1.4h // t1 = 17 * src[0] + 17 * src[16] + sub v1.4h, v7.4h, v1.4h // t2 = 17 * src[0] - 17 * src[16] + neg v3.4h, v2.4h // -t4/2 + neg v7.4h, v4.4h // -t3/2 + ssra v4.4h, v0.4h, #1 // (t1 + t3) >> 1 + ssra v3.4h, v1.4h, #1 // (t2 - t4) >> 1 + ssra v2.4h, v1.4h, #1 // (t2 + t4) >> 1 + ssra v7.4h, v0.4h, #1 // (t1 - t3) >> 1 + trn1 v0.2d, v4.2d, v3.2d + trn1 v1.2d, v2.2d, v7.2d + srshr v0.8h, v0.8h, #6 // (t1 + t3 + 64) >> 7, (t2 - t4 + 64) >> 7 + srshr v1.8h, v1.8h, #6 // (t2 + t4 + 64) >> 7, (t1 - t3 + 64) >> 7 + uaddw v0.8h, v0.8h, v5.8b + uaddw v1.8h, v1.8h, v6.8b + sqxtun v0.8b, v0.8h + sqxtun v1.8b, v1.8h + st1 {v0.s}[0], [x4], x1 + st1 {v0.s}[1], [x4], x1 + st1 {v1.s}[0], [x4], x1 + st1 {v1.s}[1], [x4] + ret +endfunc + +// VC-1 8x8 inverse transform, DC case +// On entry: +// x0 -> array of 8-bit samples, in row-major order +// x1 = row stride for 8-bit sample array +// x2 -> 16-bit inverse transform DC coefficient +// On exit: +// array at x0 updated by saturated addition of (narrowed) transformed block +function ff_vc1_inv_trans_8x8_dc_neon, export=1 + ldrsh w2, [x2] + mov x3, x0 + ld1 {v0.8b}, [x0], x1 + ld1 {v1.8b}, [x0], x1 + ld1 {v2.8b}, [x0], x1 + add w2, w2, w2, lsl #1 + ld1 {v3.8b}, [x0], x1 + ld1 {v4.8b}, [x0], x1 + add w2, w2, #1 + ld1 {v5.8b}, [x0], x1 + asr w2, w2, #1 + ld1 {v6.8b}, [x0], x1 + add w2, w2, w2, lsl #1 + ld1 {v7.8b}, [x0] + add w0, w2, #16 + asr w0, w0, #5 + dup v16.8h, w0 + uaddw v0.8h, v16.8h, v0.8b + uaddw v1.8h, v16.8h, v1.8b + uaddw v2.8h, v16.8h, v2.8b + uaddw v3.8h, v16.8h, v3.8b + uaddw v4.8h, v16.8h, v4.8b + uaddw v5.8h, v16.8h, v5.8b + sqxtun v0.8b, v0.8h + uaddw v6.8h, v16.8h, v6.8b + sqxtun v1.8b, v1.8h + uaddw v7.8h, v16.8h, v7.8b + sqxtun v2.8b, v2.8h + sqxtun v3.8b, v3.8h + sqxtun v4.8b, v4.8h + st1 {v0.8b}, [x3], x1 + sqxtun v0.8b, v5.8h + st1 {v1.8b}, [x3], x1 + sqxtun v1.8b, v6.8h + st1 {v2.8b}, [x3], x1 + sqxtun v2.8b, v7.8h + st1 {v3.8b}, [x3], x1 + st1 {v4.8b}, [x3], x1 + st1 {v0.8b}, [x3], x1 + st1 {v1.8b}, [x3], x1 + st1 {v2.8b}, [x3] + ret +endfunc + +// VC-1 8x4 inverse transform, DC case +// On entry: +// x0 -> array of 8-bit samples, in row-major order +// x1 = row stride for 8-bit sample array +// x2 -> 16-bit inverse transform DC coefficient +// On exit: +// array at x0 updated by saturated addition of (narrowed) transformed block +function ff_vc1_inv_trans_8x4_dc_neon, export=1 + ldrsh w2, [x2] + mov x3, x0 + ld1 {v0.8b}, [x0], x1 + ld1 {v1.8b}, [x0], x1 + ld1 {v2.8b}, [x0], x1 + add w2, w2, w2, lsl #1 + ld1 {v3.8b}, [x0] + add w0, w2, #1 + asr w0, w0, #1 + add w0, w0, w0, lsl #4 + add w0, w0, #64 + asr w0, w0, #7 + dup v4.8h, w0 + uaddw v0.8h, v4.8h, v0.8b + uaddw v1.8h, v4.8h, v1.8b + uaddw v2.8h, v4.8h, v2.8b + uaddw v3.8h, v4.8h, v3.8b + sqxtun v0.8b, v0.8h + sqxtun v1.8b, v1.8h + sqxtun v2.8b, v2.8h + sqxtun v3.8b, v3.8h + st1 {v0.8b}, [x3], x1 + st1 {v1.8b}, [x3], x1 + st1 {v2.8b}, [x3], x1 + st1 {v3.8b}, [x3] + ret +endfunc + +// VC-1 4x8 inverse transform, DC case +// On entry: +// x0 -> array of 8-bit samples, in row-major order +// x1 = row stride for 8-bit sample array +// x2 -> 16-bit inverse transform DC coefficient +// On exit: +// array at x0 updated by saturated addition of (narrowed) transformed block +function ff_vc1_inv_trans_4x8_dc_neon, export=1 + ldrsh w2, [x2] + mov x3, x0 + ld1 {v0.s}[0], [x0], x1 + ld1 {v1.s}[0], [x0], x1 + ld1 {v2.s}[0], [x0], x1 + add w2, w2, w2, lsl #4 + ld1 {v3.s}[0], [x0], x1 + add w2, w2, #4 + asr w2, w2, #3 + add w2, w2, w2, lsl #1 + ld1 {v0.s}[1], [x0], x1 + add w2, w2, #16 + asr w2, w2, #5 + dup v4.8h, w2 + ld1 {v1.s}[1], [x0], x1 + ld1 {v2.s}[1], [x0], x1 + ld1 {v3.s}[1], [x0] + uaddw v0.8h, v4.8h, v0.8b + uaddw v1.8h, v4.8h, v1.8b + uaddw v2.8h, v4.8h, v2.8b + uaddw v3.8h, v4.8h, v3.8b + sqxtun v0.8b, v0.8h + sqxtun v1.8b, v1.8h + sqxtun v2.8b, v2.8h + sqxtun v3.8b, v3.8h + st1 {v0.s}[0], [x3], x1 + st1 {v1.s}[0], [x3], x1 + st1 {v2.s}[0], [x3], x1 + st1 {v3.s}[0], [x3], x1 + st1 {v0.s}[1], [x3], x1 + st1 {v1.s}[1], [x3], x1 + st1 {v2.s}[1], [x3], x1 + st1 {v3.s}[1], [x3] + ret +endfunc + +// VC-1 4x4 inverse transform, DC case +// On entry: +// x0 -> array of 8-bit samples, in row-major order +// x1 = row stride for 8-bit sample array +// x2 -> 16-bit inverse transform DC coefficient +// On exit: +// array at x0 updated by saturated addition of (narrowed) transformed block +function ff_vc1_inv_trans_4x4_dc_neon, export=1 + ldrsh w2, [x2] + mov x3, x0 + ld1 {v0.s}[0], [x0], x1 + ld1 {v1.s}[0], [x0], x1 + ld1 {v0.s}[1], [x0], x1 + add w2, w2, w2, lsl #4 + ld1 {v1.s}[1], [x0] + add w0, w2, #4 + asr w0, w0, #3 + add w0, w0, w0, lsl #4 + add w0, w0, #64 + asr w0, w0, #7 + dup v2.8h, w0 + uaddw v0.8h, v2.8h, v0.8b + uaddw v1.8h, v2.8h, v1.8b + sqxtun v0.8b, v0.8h + sqxtun v1.8b, v1.8h + st1 {v0.s}[0], [x3], x1 + st1 {v1.s}[0], [x3], x1 + st1 {v0.s}[1], [x3], x1 + st1 {v1.s}[1], [x3] + ret +endfunc + .align 5 +.Lcoeffs_it8: +.quad 0x000F00090003 +.Lcoeffs_it4: +.quad 0x0011000B0005 .Lcoeffs: .quad 0x00050002 From patchwork Thu Mar 31 17:23:49 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35112 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp230347pzb; Thu, 31 Mar 2022 10:25:45 -0700 (PDT) X-Google-Smtp-Source: ABdhPJw+aqyIRaq9OQvUSktjGnRLhJcXix8EopIAAg9S+3ln1774eSZo40+tqJ5s0H0uEO6825xT X-Received: by 2002:a17:906:68c2:b0:6b4:9f26:c099 with SMTP id y2-20020a17090668c200b006b49f26c099mr5953850ejr.41.1648747545236; Thu, 31 Mar 2022 10:25:45 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747545; cv=none; d=google.com; s=arc-20160816; b=rZkGpVlMKUaZqC4/xLQCzh4mWGZMm+NJDpwVdnqAQXvqrg889f7X6076JVYxbYafNw nSsBqxqM5UVdKHGW5twNLpfBilqHzghcrtLvjD+FFLL0rgR74PozTpCfOCCF1qrYb0Fo 9tU92lHPVdPQ/6DVn4emBydFYPwAKEKLu7T+niPz2pBYIZXUj36jLhZ8p8wHVVjEY7+S uu3EoS5ZSk+vQ02jJC5SyrcpzoapL2+MuJiBBoLoUgchUybcXU2Yxgjap4Fk06zhjcqt VbKXga5gVLn+qL1lpP2tuPuqr6f7pPbv1KM4CIO53tGxcX3UQPvNf5poPwlwpOgLAXzM +tlw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=6lHoLfur9CL72SuQ41w7zzPQ7EXhI+bJqteivdNZ/ig=; b=Tw9UKKuBihSABeUiSm2AgYJ4xFgQJxf45XFERStc3hgVXN0kavUKwMPI3diQUiQ+m8 meIl/ESP8goSPsRBid2xuCvbTVubClDG85uh2lNfSDdFj7ewzAjA86Dv4/ApTzI7Nzpp haVHtdnlutuNmiVWR6qhPPtzKaQM/SuqjAbVzMdmZLUnqd126S1OneXVrJD9wwz3rmov 9X6cx9vgyKpopuUhGB3rN5EZ9hHqZegcwbHOuNLl8ECEv+CBt9kwPMKm2G8SqYT88Vxi vru83anK9xfZWMXrkjmPdmPOxWcUodEPVEy1MHpuEYWBzOBWo3QEH9yJhRgmr6+Y8tZz yqUg== 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 e18-20020a170906845200b006e02fed8798si77226ejy.637.2022.03.31.10.25.44; Thu, 31 Mar 2022 10:25:45 -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 55C1668B2B5; Thu, 31 Mar 2022 20:24:24 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail149112.authsmtp.co.uk (outmail149112.authsmtp.co.uk [62.13.149.112]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 689CA68B098 for ; Thu, 31 Mar 2022 20:24:20 +0300 (EEST) Received: from punt18.authsmtp.com (punt18.authsmtp.com [62.13.128.225]) by punt17.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOJVq022024 for ; Thu, 31 Mar 2022 18:24:19 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt18.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOJdS097127; Thu, 31 Mar 2022 18:24:19 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHOHER062592 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:17 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:16 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:49 +0100 Message-Id: <20220331172351.550818-9-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 655bdf02-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQtZIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLWEHFQVAent5 ZwhjVnhSVAp8cEMv FhhUF3BUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdW0Pb30N d0kEM1kVTUsAWT8w TAwOBi19VWQfQm1t c1ksK0JUOms2FA04 NVwqWhoRPxNaAwtS V0pJCSpBb1cIXDZj FQpGXVV2 X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 08/10] avcodec/idctdsp: Arm 64-bit NEON block add and clamp fast paths 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: J9+EvHtj2Cyy checkasm benchmarks on 1.5 GHz Cortex-A72 are as follows. idctdsp.add_pixels_clamped_c: 313.3 idctdsp.add_pixels_clamped_neon: 24.3 idctdsp.put_pixels_clamped_c: 220.3 idctdsp.put_pixels_clamped_neon: 15.5 idctdsp.put_signed_pixels_clamped_c: 210.5 idctdsp.put_signed_pixels_clamped_neon: 19.5 Signed-off-by: Ben Avison --- libavcodec/aarch64/Makefile | 3 +- libavcodec/aarch64/idctdsp_init_aarch64.c | 26 +++-- libavcodec/aarch64/idctdsp_neon.S | 130 ++++++++++++++++++++++ 3 files changed, 150 insertions(+), 9 deletions(-) create mode 100644 libavcodec/aarch64/idctdsp_neon.S diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile index 5b25e4dfb9..c8935f205e 100644 --- a/libavcodec/aarch64/Makefile +++ b/libavcodec/aarch64/Makefile @@ -44,7 +44,8 @@ NEON-OBJS-$(CONFIG_H264PRED) += aarch64/h264pred_neon.o NEON-OBJS-$(CONFIG_H264QPEL) += aarch64/h264qpel_neon.o \ aarch64/hpeldsp_neon.o NEON-OBJS-$(CONFIG_HPELDSP) += aarch64/hpeldsp_neon.o -NEON-OBJS-$(CONFIG_IDCTDSP) += aarch64/simple_idct_neon.o +NEON-OBJS-$(CONFIG_IDCTDSP) += aarch64/idctdsp_neon.o \ + aarch64/simple_idct_neon.o NEON-OBJS-$(CONFIG_MDCT) += aarch64/mdct_neon.o NEON-OBJS-$(CONFIG_MPEGAUDIODSP) += aarch64/mpegaudiodsp_neon.o NEON-OBJS-$(CONFIG_PIXBLOCKDSP) += aarch64/pixblockdsp_neon.o diff --git a/libavcodec/aarch64/idctdsp_init_aarch64.c b/libavcodec/aarch64/idctdsp_init_aarch64.c index 742a3372e3..eec21aa5a2 100644 --- a/libavcodec/aarch64/idctdsp_init_aarch64.c +++ b/libavcodec/aarch64/idctdsp_init_aarch64.c @@ -27,19 +27,29 @@ #include "libavcodec/idctdsp.h" #include "idct.h" +void ff_put_pixels_clamped_neon(const int16_t *, uint8_t *, ptrdiff_t); +void ff_put_signed_pixels_clamped_neon(const int16_t *, uint8_t *, ptrdiff_t); +void ff_add_pixels_clamped_neon(const int16_t *, uint8_t *, ptrdiff_t); + av_cold void ff_idctdsp_init_aarch64(IDCTDSPContext *c, AVCodecContext *avctx, unsigned high_bit_depth) { int cpu_flags = av_get_cpu_flags(); - if (have_neon(cpu_flags) && !avctx->lowres && !high_bit_depth) { - if (avctx->idct_algo == FF_IDCT_AUTO || - avctx->idct_algo == FF_IDCT_SIMPLEAUTO || - avctx->idct_algo == FF_IDCT_SIMPLENEON) { - c->idct_put = ff_simple_idct_put_neon; - c->idct_add = ff_simple_idct_add_neon; - c->idct = ff_simple_idct_neon; - c->perm_type = FF_IDCT_PERM_PARTTRANS; + if (have_neon(cpu_flags)) { + if (!avctx->lowres && !high_bit_depth) { + if (avctx->idct_algo == FF_IDCT_AUTO || + avctx->idct_algo == FF_IDCT_SIMPLEAUTO || + avctx->idct_algo == FF_IDCT_SIMPLENEON) { + c->idct_put = ff_simple_idct_put_neon; + c->idct_add = ff_simple_idct_add_neon; + c->idct = ff_simple_idct_neon; + c->perm_type = FF_IDCT_PERM_PARTTRANS; + } } + + c->add_pixels_clamped = ff_add_pixels_clamped_neon; + c->put_pixels_clamped = ff_put_pixels_clamped_neon; + c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_neon; } } diff --git a/libavcodec/aarch64/idctdsp_neon.S b/libavcodec/aarch64/idctdsp_neon.S new file mode 100644 index 0000000000..7f47611206 --- /dev/null +++ b/libavcodec/aarch64/idctdsp_neon.S @@ -0,0 +1,130 @@ +/* + * IDCT AArch64 NEON optimisations + * + * Copyright (c) 2022 Ben Avison + * + * 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" + +// Clamp 16-bit signed block coefficients to unsigned 8-bit +// On entry: +// x0 -> array of 64x 16-bit coefficients +// x1 -> 8-bit results +// x2 = row stride for results, bytes +function ff_put_pixels_clamped_neon, export=1 + ld1 {v0.16b, v1.16b, v2.16b, v3.16b}, [x0], #64 + ld1 {v4.16b, v5.16b, v6.16b, v7.16b}, [x0] + sqxtun v0.8b, v0.8h + sqxtun v1.8b, v1.8h + sqxtun v2.8b, v2.8h + sqxtun v3.8b, v3.8h + sqxtun v4.8b, v4.8h + st1 {v0.8b}, [x1], x2 + sqxtun v0.8b, v5.8h + st1 {v1.8b}, [x1], x2 + sqxtun v1.8b, v6.8h + st1 {v2.8b}, [x1], x2 + sqxtun v2.8b, v7.8h + st1 {v3.8b}, [x1], x2 + st1 {v4.8b}, [x1], x2 + st1 {v0.8b}, [x1], x2 + st1 {v1.8b}, [x1], x2 + st1 {v2.8b}, [x1] + ret +endfunc + +// Clamp 16-bit signed block coefficients to signed 8-bit (biased by 128) +// On entry: +// x0 -> array of 64x 16-bit coefficients +// x1 -> 8-bit results +// x2 = row stride for results, bytes +function ff_put_signed_pixels_clamped_neon, export=1 + ld1 {v0.16b, v1.16b, v2.16b, v3.16b}, [x0], #64 + movi v4.8b, #128 + ld1 {v16.16b, v17.16b, v18.16b, v19.16b}, [x0] + sqxtn v0.8b, v0.8h + sqxtn v1.8b, v1.8h + sqxtn v2.8b, v2.8h + sqxtn v3.8b, v3.8h + sqxtn v5.8b, v16.8h + add v0.8b, v0.8b, v4.8b + sqxtn v6.8b, v17.8h + add v1.8b, v1.8b, v4.8b + sqxtn v7.8b, v18.8h + add v2.8b, v2.8b, v4.8b + sqxtn v16.8b, v19.8h + add v3.8b, v3.8b, v4.8b + st1 {v0.8b}, [x1], x2 + add v0.8b, v5.8b, v4.8b + st1 {v1.8b}, [x1], x2 + add v1.8b, v6.8b, v4.8b + st1 {v2.8b}, [x1], x2 + add v2.8b, v7.8b, v4.8b + st1 {v3.8b}, [x1], x2 + add v3.8b, v16.8b, v4.8b + st1 {v0.8b}, [x1], x2 + st1 {v1.8b}, [x1], x2 + st1 {v2.8b}, [x1], x2 + st1 {v3.8b}, [x1] + ret +endfunc + +// Add 16-bit signed block coefficients to unsigned 8-bit +// On entry: +// x0 -> array of 64x 16-bit coefficients +// x1 -> 8-bit input and results +// x2 = row stride for 8-bit input and results, bytes +function ff_add_pixels_clamped_neon, export=1 + ld1 {v0.16b, v1.16b, v2.16b, v3.16b}, [x0], #64 + mov x3, x1 + ld1 {v4.8b}, [x1], x2 + ld1 {v5.8b}, [x1], x2 + ld1 {v6.8b}, [x1], x2 + ld1 {v7.8b}, [x1], x2 + ld1 {v16.16b, v17.16b, v18.16b, v19.16b}, [x0] + uaddw v0.8h, v0.8h, v4.8b + uaddw v1.8h, v1.8h, v5.8b + uaddw v2.8h, v2.8h, v6.8b + ld1 {v4.8b}, [x1], x2 + uaddw v3.8h, v3.8h, v7.8b + ld1 {v5.8b}, [x1], x2 + sqxtun v0.8b, v0.8h + ld1 {v6.8b}, [x1], x2 + sqxtun v1.8b, v1.8h + ld1 {v7.8b}, [x1] + sqxtun v2.8b, v2.8h + sqxtun v3.8b, v3.8h + uaddw v4.8h, v16.8h, v4.8b + st1 {v0.8b}, [x3], x2 + uaddw v0.8h, v17.8h, v5.8b + st1 {v1.8b}, [x3], x2 + uaddw v1.8h, v18.8h, v6.8b + st1 {v2.8b}, [x3], x2 + uaddw v2.8h, v19.8h, v7.8b + sqxtun v4.8b, v4.8h + sqxtun v0.8b, v0.8h + st1 {v3.8b}, [x3], x2 + sqxtun v1.8b, v1.8h + sqxtun v2.8b, v2.8h + st1 {v4.8b}, [x3], x2 + st1 {v0.8b}, [x3], x2 + st1 {v1.8b}, [x3], x2 + st1 {v2.8b}, [x3] + ret +endfunc From patchwork Thu Mar 31 17:23:50 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35113 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp230414pzb; Thu, 31 Mar 2022 10:25:56 -0700 (PDT) X-Google-Smtp-Source: ABdhPJwYBQmO338ysucTpHDo4tpoEhFQr88fVSXLp+o6eDxoSGoif7et1dP3+dF/2JCa3cqx4GxP X-Received: by 2002:a17:906:4fd2:b0:6e0:5ce7:d7c7 with SMTP id i18-20020a1709064fd200b006e05ce7d7c7mr5909608ejw.113.1648747556621; Thu, 31 Mar 2022 10:25:56 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747556; cv=none; d=google.com; s=arc-20160816; b=Acys+q41sXFckfiFs5oMJhgGOVcNZ8q3E8uHUSG3aDXJpk1QjUNTy4uG+WpFgNTktt zqqGFsEuO5d1gvNKRQrXnrF5PPX05ch0s2lHzaTyPO+lhYV4cGFHv6G3Ij2kSwtJ/V7n NqYXb5rs75AHW1ZtWhJKQNlienkIzoEsW1w7fSrYANH/KLlUnzCHfEbgjoapziVytP4y 4XFVsJyMz8/Vbds+OOIsRXt5RYVxV/FMU4kmgLTb95S1DZ3zk0mJLDBh1pNIF7ktOJv8 lbO95itTZozHxON2/f0cTBXYtNRY0aNpoMaeNWEnLXrSAAru/ip+PPzZfjFCqxSbCSya YkLw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=kD5jIeQ2dkTD6nWVGxIkFSHW8SCcc7f8CYfpdAql4Lc=; b=A8NZ67FC1vjrFDwBFMU2XACgfidOdxiR9AMf1U/ykvmK6EAlH5Gpz77dSEL83i8qZv PWN1P1qZU1dg5rAUsc5chssKItPWLGxok00EyVSiynd7ZCYB6hus8BotIrSYd3HClLrT uSvVjtues7usABnUgOu31RQUyPSioVtTmlCIoLy916IQFmSJnXv1dwpU9WPx3Ac9uBd+ gllYHH5w1kCBJp24cyfywX0GW1eNXt4PvsoF/+h9hI9xhS/HIYFpF8WAgjalgD70tra7 JUaWneI57NBTUQ7bElvoG8lvV0jQmfDioAQlt8yuzW31blxge7mXWj0BJderzVAp4Zv/ YmCQ== 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 s10-20020a170906284a00b006df76385c39si114799ejc.217.2022.03.31.10.25.55; Thu, 31 Mar 2022 10:25:56 -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 4C5BC68B2C4; Thu, 31 Mar 2022 20:24:25 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail149078.authsmtp.net (outmail149078.authsmtp.net [62.13.149.78]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 093D768B239 for ; Thu, 31 Mar 2022 20:24:20 +0300 (EEST) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt15.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOKT2019939; Thu, 31 Mar 2022 18:24:20 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHOIZ0062625 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:19 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:18 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:50 +0100 Message-Id: <20220331172351.550818-10-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 666a4768-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQhRIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLWAYdwVAenhy YAhgWnlcWAp8c0As RklSHXBUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdW0Ob30N d0kEM1kVTUsAWSA3 HkJKNC8qVRNZAi8y M1QAB3k6VFsXP145 OEMsEVwRKANaEgRC HykA X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthSMTP-Origin: 51.9.63.237/2525 X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 09/10] avcodec/vc1: Arm 64-bit NEON unescape fast path 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: ovcq/7C80sWi checkasm benchmarks on 1.5 GHz Cortex-A72 are as follows. vc1dsp.vc1_unescape_buffer_c: 655617.7 vc1dsp.vc1_unescape_buffer_neon: 118237.0 Signed-off-by: Ben Avison --- libavcodec/aarch64/vc1dsp_init_aarch64.c | 61 ++++++++ libavcodec/aarch64/vc1dsp_neon.S | 176 +++++++++++++++++++++++ 2 files changed, 237 insertions(+) diff --git a/libavcodec/aarch64/vc1dsp_init_aarch64.c b/libavcodec/aarch64/vc1dsp_init_aarch64.c index e0eb52dd63..a7976fd596 100644 --- a/libavcodec/aarch64/vc1dsp_init_aarch64.c +++ b/libavcodec/aarch64/vc1dsp_init_aarch64.c @@ -21,6 +21,7 @@ #include "libavutil/attributes.h" #include "libavutil/cpu.h" #include "libavutil/aarch64/cpu.h" +#include "libavutil/intreadwrite.h" #include "libavcodec/vc1dsp.h" #include "config.h" @@ -51,6 +52,64 @@ void ff_put_vc1_chroma_mc4_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride, void ff_avg_vc1_chroma_mc4_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int h, int x, int y); +int ff_vc1_unescape_buffer_helper_neon(const uint8_t *src, int size, uint8_t *dst); + +static int vc1_unescape_buffer_neon(const uint8_t *src, int size, uint8_t *dst) +{ + /* Dealing with starting and stopping, and removing escape bytes, are + * comparatively less time-sensitive, so are more clearly expressed using + * a C wrapper around the assembly inner loop. Note that we assume a + * little-endian machine that supports unaligned loads. */ + int dsize = 0; + while (size >= 4) + { + int found = 0; + while (!found && (((uintptr_t) dst) & 7) && size >= 4) + { + found = (AV_RL32(src) &~ 0x03000000) == 0x00030000; + if (!found) + { + *dst++ = *src++; + --size; + ++dsize; + } + } + if (!found) + { + int skip = size - ff_vc1_unescape_buffer_helper_neon(src, size, dst); + dst += skip; + src += skip; + size -= skip; + dsize += skip; + while (!found && size >= 4) + { + found = (AV_RL32(src) &~ 0x03000000) == 0x00030000; + if (!found) + { + *dst++ = *src++; + --size; + ++dsize; + } + } + } + if (found) + { + *dst++ = *src++; + *dst++ = *src++; + ++src; + size -= 3; + dsize += 2; + } + } + while (size > 0) + { + *dst++ = *src++; + --size; + ++dsize; + } + return dsize; +} + av_cold void ff_vc1dsp_init_aarch64(VC1DSPContext *dsp) { int cpu_flags = av_get_cpu_flags(); @@ -76,5 +135,7 @@ av_cold void ff_vc1dsp_init_aarch64(VC1DSPContext *dsp) dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_vc1_chroma_mc8_neon; dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = ff_put_vc1_chroma_mc4_neon; dsp->avg_no_rnd_vc1_chroma_pixels_tab[1] = ff_avg_vc1_chroma_mc4_neon; + + dsp->vc1_unescape_buffer = vc1_unescape_buffer_neon; } } diff --git a/libavcodec/aarch64/vc1dsp_neon.S b/libavcodec/aarch64/vc1dsp_neon.S index 0201db4f78..9a96c2523c 100644 --- a/libavcodec/aarch64/vc1dsp_neon.S +++ b/libavcodec/aarch64/vc1dsp_neon.S @@ -1368,3 +1368,179 @@ function ff_vc1_h_loop_filter16_neon, export=1 st2 {v2.b, v3.b}[7], [x6] 4: ret endfunc + +// Copy at most the specified number of bytes from source to destination buffer, +// stopping at a multiple of 32 bytes, none of which are the start of an escape sequence +// On entry: +// x0 -> source buffer +// w1 = max number of bytes to copy +// x2 -> destination buffer, optimally 8-byte aligned +// On exit: +// w0 = number of bytes not copied +function ff_vc1_unescape_buffer_helper_neon, export=1 + // Offset by 80 to screen out cases that are too short for us to handle, + // and also make it easy to test for loop termination, or to determine + // whether we need an odd number of half-iterations of the loop. + subs w1, w1, #80 + b.mi 90f + + // Set up useful constants + movi v20.4s, #3, lsl #24 + movi v21.4s, #3, lsl #16 + + tst w1, #32 + b.ne 1f + + ld1 {v0.16b, v1.16b, v2.16b}, [x0], #48 + ext v25.16b, v0.16b, v1.16b, #1 + ext v26.16b, v0.16b, v1.16b, #2 + ext v27.16b, v0.16b, v1.16b, #3 + ext v29.16b, v1.16b, v2.16b, #1 + ext v30.16b, v1.16b, v2.16b, #2 + ext v31.16b, v1.16b, v2.16b, #3 + bic v24.16b, v0.16b, v20.16b + bic v25.16b, v25.16b, v20.16b + bic v26.16b, v26.16b, v20.16b + bic v27.16b, v27.16b, v20.16b + bic v28.16b, v1.16b, v20.16b + bic v29.16b, v29.16b, v20.16b + bic v30.16b, v30.16b, v20.16b + bic v31.16b, v31.16b, v20.16b + eor v24.16b, v24.16b, v21.16b + eor v25.16b, v25.16b, v21.16b + eor v26.16b, v26.16b, v21.16b + eor v27.16b, v27.16b, v21.16b + eor v28.16b, v28.16b, v21.16b + eor v29.16b, v29.16b, v21.16b + eor v30.16b, v30.16b, v21.16b + eor v31.16b, v31.16b, v21.16b + cmeq v24.4s, v24.4s, #0 + cmeq v25.4s, v25.4s, #0 + cmeq v26.4s, v26.4s, #0 + cmeq v27.4s, v27.4s, #0 + add w1, w1, #32 + b 3f + +1: ld1 {v3.16b, v4.16b, v5.16b}, [x0], #48 + ext v25.16b, v3.16b, v4.16b, #1 + ext v26.16b, v3.16b, v4.16b, #2 + ext v27.16b, v3.16b, v4.16b, #3 + ext v29.16b, v4.16b, v5.16b, #1 + ext v30.16b, v4.16b, v5.16b, #2 + ext v31.16b, v4.16b, v5.16b, #3 + bic v24.16b, v3.16b, v20.16b + bic v25.16b, v25.16b, v20.16b + bic v26.16b, v26.16b, v20.16b + bic v27.16b, v27.16b, v20.16b + bic v28.16b, v4.16b, v20.16b + bic v29.16b, v29.16b, v20.16b + bic v30.16b, v30.16b, v20.16b + bic v31.16b, v31.16b, v20.16b + eor v24.16b, v24.16b, v21.16b + eor v25.16b, v25.16b, v21.16b + eor v26.16b, v26.16b, v21.16b + eor v27.16b, v27.16b, v21.16b + eor v28.16b, v28.16b, v21.16b + eor v29.16b, v29.16b, v21.16b + eor v30.16b, v30.16b, v21.16b + eor v31.16b, v31.16b, v21.16b + cmeq v24.4s, v24.4s, #0 + cmeq v25.4s, v25.4s, #0 + cmeq v26.4s, v26.4s, #0 + cmeq v27.4s, v27.4s, #0 + // Drop through... +2: mov v0.16b, v5.16b + ld1 {v1.16b, v2.16b}, [x0], #32 + cmeq v28.4s, v28.4s, #0 + cmeq v29.4s, v29.4s, #0 + cmeq v30.4s, v30.4s, #0 + cmeq v31.4s, v31.4s, #0 + orr v24.16b, v24.16b, v25.16b + orr v26.16b, v26.16b, v27.16b + orr v28.16b, v28.16b, v29.16b + orr v30.16b, v30.16b, v31.16b + ext v25.16b, v0.16b, v1.16b, #1 + orr v22.16b, v24.16b, v26.16b + ext v26.16b, v0.16b, v1.16b, #2 + ext v27.16b, v0.16b, v1.16b, #3 + ext v29.16b, v1.16b, v2.16b, #1 + orr v23.16b, v28.16b, v30.16b + ext v30.16b, v1.16b, v2.16b, #2 + ext v31.16b, v1.16b, v2.16b, #3 + bic v24.16b, v0.16b, v20.16b + bic v25.16b, v25.16b, v20.16b + bic v26.16b, v26.16b, v20.16b + orr v22.16b, v22.16b, v23.16b + bic v27.16b, v27.16b, v20.16b + bic v28.16b, v1.16b, v20.16b + bic v29.16b, v29.16b, v20.16b + bic v30.16b, v30.16b, v20.16b + bic v31.16b, v31.16b, v20.16b + addv s22, v22.4s + eor v24.16b, v24.16b, v21.16b + eor v25.16b, v25.16b, v21.16b + eor v26.16b, v26.16b, v21.16b + eor v27.16b, v27.16b, v21.16b + eor v28.16b, v28.16b, v21.16b + mov w3, v22.s[0] + eor v29.16b, v29.16b, v21.16b + eor v30.16b, v30.16b, v21.16b + eor v31.16b, v31.16b, v21.16b + cmeq v24.4s, v24.4s, #0 + cmeq v25.4s, v25.4s, #0 + cmeq v26.4s, v26.4s, #0 + cmeq v27.4s, v27.4s, #0 + cbnz w3, 90f + st1 {v3.16b, v4.16b}, [x2], #32 +3: mov v3.16b, v2.16b + ld1 {v4.16b, v5.16b}, [x0], #32 + cmeq v28.4s, v28.4s, #0 + cmeq v29.4s, v29.4s, #0 + cmeq v30.4s, v30.4s, #0 + cmeq v31.4s, v31.4s, #0 + orr v24.16b, v24.16b, v25.16b + orr v26.16b, v26.16b, v27.16b + orr v28.16b, v28.16b, v29.16b + orr v30.16b, v30.16b, v31.16b + ext v25.16b, v3.16b, v4.16b, #1 + orr v22.16b, v24.16b, v26.16b + ext v26.16b, v3.16b, v4.16b, #2 + ext v27.16b, v3.16b, v4.16b, #3 + ext v29.16b, v4.16b, v5.16b, #1 + orr v23.16b, v28.16b, v30.16b + ext v30.16b, v4.16b, v5.16b, #2 + ext v31.16b, v4.16b, v5.16b, #3 + bic v24.16b, v3.16b, v20.16b + bic v25.16b, v25.16b, v20.16b + bic v26.16b, v26.16b, v20.16b + orr v22.16b, v22.16b, v23.16b + bic v27.16b, v27.16b, v20.16b + bic v28.16b, v4.16b, v20.16b + bic v29.16b, v29.16b, v20.16b + bic v30.16b, v30.16b, v20.16b + bic v31.16b, v31.16b, v20.16b + addv s22, v22.4s + eor v24.16b, v24.16b, v21.16b + eor v25.16b, v25.16b, v21.16b + eor v26.16b, v26.16b, v21.16b + eor v27.16b, v27.16b, v21.16b + eor v28.16b, v28.16b, v21.16b + mov w3, v22.s[0] + eor v29.16b, v29.16b, v21.16b + eor v30.16b, v30.16b, v21.16b + eor v31.16b, v31.16b, v21.16b + cmeq v24.4s, v24.4s, #0 + cmeq v25.4s, v25.4s, #0 + cmeq v26.4s, v26.4s, #0 + cmeq v27.4s, v27.4s, #0 + cbnz w3, 91f + st1 {v0.16b, v1.16b}, [x2], #32 + subs w1, w1, #64 + b.pl 2b + +90: add w0, w1, #80 + ret + +91: sub w1, w1, #32 + b 90b +endfunc From patchwork Thu Mar 31 17:23:51 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ben Avison X-Patchwork-Id: 35114 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a05:6a20:c05:b0:7a:e998:b410 with SMTP id bw5csp230476pzb; Thu, 31 Mar 2022 10:26:06 -0700 (PDT) X-Google-Smtp-Source: ABdhPJwjXVXRYXQVdwSFyGJQJS6bKEuxgWj1R09Wjn5p+pe8JD+WGzaobAj/T2fm3eN07w6+QYDW X-Received: by 2002:a05:6402:51d4:b0:410:a328:3c86 with SMTP id r20-20020a05640251d400b00410a3283c86mr10721760edd.55.1648747565985; Thu, 31 Mar 2022 10:26:05 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1648747565; cv=none; d=google.com; s=arc-20160816; b=pO61kBtAir/PNDCe943Fq5zGQ8LOa2xrQ+kAE28GwkRf2efm+fUzQdSd9iUMtPNmOi tNkEHVOL3ckuLA2zTxp7nCj9wHXEpqwqjJam8+nnZmDRWNFftpbxy6yb44kFCoSt4UaY lDGJC5vgfNH1IfPYUH1VYclVeRed34OfHBG4c4bQyHrNcLVJoPuz++zjjrSu42vkdQMt J7TcwbLBUss1p5k5ocSnnhGRwy7KCYPj2CDMAk5prVeRnG+2NXuM/vLQmabkzn/4NLvA mLtGlmA8grt4juGIUjBOtCFLyuzfuppvj/Uh2ilIvOJy2162pCXpslrYol6HLI8v3rwf 1Ztw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=rdAxwoq7sLQyjAaqW/iCbSFVUQ5u0yYCiEATgpbWufM=; b=ki51wu9BoIguNBWBpwVzGF6Pv5XlCNI9aabQOshSiL9bvjG6dVIi3GizYgJQ364vdB Wv+sloq80UoCPFOIBwUZ1rRBtpjWQTmHTJe/Yz+rWpo9G290vxW2zqTUz8Eih+rUxuX9 1aKXl+lEdUjfwq5XJxRcgoGyusVe8H6M2myBtn20S64zI848khzDDTFIRTyABz6jwn4Z hPBkXFtbBvfH0X3CReZZ/Vwf/laGDU0YLBuNc1yC0F/pK+G0egICFoynR7sSeuwI7pl3 YmnEFAGVMZqTghof9xt3Bt85TIx1gp5E6OZx46KaprkS91ky9RgbF6P8cHIL47uFQDn0 UUzg== 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 c4-20020a170906528400b006df76385b8fsi134436ejm.47.2022.03.31.10.26.05; Thu, 31 Mar 2022 10:26:05 -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 3A7DD68B2CC; Thu, 31 Mar 2022 20:24:26 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from outmail149056.authsmtp.com (outmail149056.authsmtp.com [62.13.149.56]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 9920A68B2B8 for ; Thu, 31 Mar 2022 20:24:23 +0300 (EEST) Received: from mail-c233.authsmtp.com (mail-c233.authsmtp.com [62.13.128.233]) by punt17.authsmtp.com. (8.15.2/8.15.2) with ESMTP id 22VHOMY1022168; Thu, 31 Mar 2022 18:24:22 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: from rpi2021 (237.63.9.51.dyn.plus.net [51.9.63.237]) (authenticated bits=0) by mail.authsmtp.com (8.15.2/8.15.2) with ESMTPSA id 22VHOKxk062679 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NO); Thu, 31 Mar 2022 18:24:20 +0100 (BST) (envelope-from bavison@riscosopen.org) Received: by rpi2021 (sSMTP sendmail emulation); Thu, 31 Mar 2022 18:24:20 +0100 From: Ben Avison To: ffmpeg-devel@ffmpeg.org Date: Thu, 31 Mar 2022 18:23:51 +0100 Message-Id: <20220331172351.550818-11-bavison@riscosopen.org> X-Mailer: git-send-email 2.25.1 In-Reply-To: <20220331172351.550818-1-bavison@riscosopen.org> References: <20220331172351.550818-1-bavison@riscosopen.org> MIME-Version: 1.0 X-Server-Quench: 6779f8c5-b117-11ec-a0f2-84349711df28 X-AuthReport-Spam: If SPAM / abuse - report it at: http://www.authsmtp.com/abuse X-AuthRoute: OCd1YggXA1ZfRRob ESQCJDVBUg4iPRpU DBlFKhFVNl8UURhQ KkJXbgASJgZFAnRQ QXkJW1ZWQFx5U2Fx YQhTIwBcfENQWQZ0 UktOXVBXFgB3AFID BHhmLWI6LAVAenh3 bAhjWXFaVAp8ckF0 FEYHEHBUZGZndWEe BRNFJgMCch5CehxB Y1d+VSdbY21JDRoR IyQTdy5qdWwHb30N d0kEM1kVTUsAWSA3 HkJKNC8qVRZfAi8y M1QAB3k6VFsXP145 OEMsEVwRKANaEgRC HykA X-Authentic-SMTP: 61633632303230.1021:7600 X-AuthFastPath: 0 (Was 255) X-AuthSMTP-Origin: 51.9.63.237/2525 X-AuthVirus-Status: No virus detected - but ensure you scan with your own anti-virus system. Subject: [FFmpeg-devel] [PATCH v3 10/10] avcodec/vc1: Arm 32-bit NEON unescape fast path 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 Cc: Ben Avison Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: vIffIdWZIMXy checkasm benchmarks on 1.5 GHz Cortex-A72 are as follows. vc1dsp.vc1_unescape_buffer_c: 918624.7 vc1dsp.vc1_unescape_buffer_neon: 142958.0 Signed-off-by: Ben Avison --- libavcodec/arm/vc1dsp_init_neon.c | 61 +++++++++++++++ libavcodec/arm/vc1dsp_neon.S | 118 ++++++++++++++++++++++++++++++ 2 files changed, 179 insertions(+) diff --git a/libavcodec/arm/vc1dsp_init_neon.c b/libavcodec/arm/vc1dsp_init_neon.c index f5f5c702d7..48cb816b70 100644 --- a/libavcodec/arm/vc1dsp_init_neon.c +++ b/libavcodec/arm/vc1dsp_init_neon.c @@ -19,6 +19,7 @@ #include #include "libavutil/attributes.h" +#include "libavutil/intreadwrite.h" #include "libavcodec/vc1dsp.h" #include "vc1dsp.h" @@ -84,6 +85,64 @@ void ff_put_vc1_chroma_mc4_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride, void ff_avg_vc1_chroma_mc4_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride, int h, int x, int y); +int ff_vc1_unescape_buffer_helper_neon(const uint8_t *src, int size, uint8_t *dst); + +static int vc1_unescape_buffer_neon(const uint8_t *src, int size, uint8_t *dst) +{ + /* Dealing with starting and stopping, and removing escape bytes, are + * comparatively less time-sensitive, so are more clearly expressed using + * a C wrapper around the assembly inner loop. Note that we assume a + * little-endian machine that supports unaligned loads. */ + int dsize = 0; + while (size >= 4) + { + int found = 0; + while (!found && (((uintptr_t) dst) & 7) && size >= 4) + { + found = (AV_RL32(src) &~ 0x03000000) == 0x00030000; + if (!found) + { + *dst++ = *src++; + --size; + ++dsize; + } + } + if (!found) + { + int skip = size - ff_vc1_unescape_buffer_helper_neon(src, size, dst); + dst += skip; + src += skip; + size -= skip; + dsize += skip; + while (!found && size >= 4) + { + found = (AV_RL32(src) &~ 0x03000000) == 0x00030000; + if (!found) + { + *dst++ = *src++; + --size; + ++dsize; + } + } + } + if (found) + { + *dst++ = *src++; + *dst++ = *src++; + ++src; + size -= 3; + dsize += 2; + } + } + while (size > 0) + { + *dst++ = *src++; + --size; + ++dsize; + } + return dsize; +} + #define FN_ASSIGN(X, Y) \ dsp->put_vc1_mspel_pixels_tab[0][X+4*Y] = ff_put_vc1_mspel_mc##X##Y##_16_neon; \ dsp->put_vc1_mspel_pixels_tab[1][X+4*Y] = ff_put_vc1_mspel_mc##X##Y##_neon @@ -130,4 +189,6 @@ av_cold void ff_vc1dsp_init_neon(VC1DSPContext *dsp) dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_vc1_chroma_mc8_neon; dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = ff_put_vc1_chroma_mc4_neon; dsp->avg_no_rnd_vc1_chroma_pixels_tab[1] = ff_avg_vc1_chroma_mc4_neon; + + dsp->vc1_unescape_buffer = vc1_unescape_buffer_neon; } diff --git a/libavcodec/arm/vc1dsp_neon.S b/libavcodec/arm/vc1dsp_neon.S index ba54221ef6..96014fbebc 100644 --- a/libavcodec/arm/vc1dsp_neon.S +++ b/libavcodec/arm/vc1dsp_neon.S @@ -1804,3 +1804,121 @@ function ff_vc1_h_loop_filter16_neon, export=1 4: vpop {d8-d15} pop {r4-r6,pc} endfunc + +@ Copy at most the specified number of bytes from source to destination buffer, +@ stopping at a multiple of 16 bytes, none of which are the start of an escape sequence +@ On entry: +@ r0 -> source buffer +@ r1 = max number of bytes to copy +@ r2 -> destination buffer, optimally 8-byte aligned +@ On exit: +@ r0 = number of bytes not copied +function ff_vc1_unescape_buffer_helper_neon, export=1 + @ Offset by 48 to screen out cases that are too short for us to handle, + @ and also make it easy to test for loop termination, or to determine + @ whether we need an odd number of half-iterations of the loop. + subs r1, r1, #48 + bmi 90f + + @ Set up useful constants + vmov.i32 q0, #0x3000000 + vmov.i32 q1, #0x30000 + + tst r1, #16 + bne 1f + + vld1.8 {q8, q9}, [r0]! + vbic q12, q8, q0 + vext.8 q13, q8, q9, #1 + vext.8 q14, q8, q9, #2 + vext.8 q15, q8, q9, #3 + veor q12, q12, q1 + vbic q13, q13, q0 + vbic q14, q14, q0 + vbic q15, q15, q0 + vceq.i32 q12, q12, #0 + veor q13, q13, q1 + veor q14, q14, q1 + veor q15, q15, q1 + vceq.i32 q13, q13, #0 + vceq.i32 q14, q14, #0 + vceq.i32 q15, q15, #0 + add r1, r1, #16 + b 3f + +1: vld1.8 {q10, q11}, [r0]! + vbic q12, q10, q0 + vext.8 q13, q10, q11, #1 + vext.8 q14, q10, q11, #2 + vext.8 q15, q10, q11, #3 + veor q12, q12, q1 + vbic q13, q13, q0 + vbic q14, q14, q0 + vbic q15, q15, q0 + vceq.i32 q12, q12, #0 + veor q13, q13, q1 + veor q14, q14, q1 + veor q15, q15, q1 + vceq.i32 q13, q13, #0 + vceq.i32 q14, q14, #0 + vceq.i32 q15, q15, #0 + @ Drop through... +2: vmov q8, q11 + vld1.8 {q9}, [r0]! + vorr q13, q12, q13 + vorr q15, q14, q15 + vbic q12, q8, q0 + vorr q3, q13, q15 + vext.8 q13, q8, q9, #1 + vext.8 q14, q8, q9, #2 + vext.8 q15, q8, q9, #3 + veor q12, q12, q1 + vorr d6, d6, d7 + vbic q13, q13, q0 + vbic q14, q14, q0 + vbic q15, q15, q0 + vceq.i32 q12, q12, #0 + vmov r3, r12, d6 + veor q13, q13, q1 + veor q14, q14, q1 + veor q15, q15, q1 + vceq.i32 q13, q13, #0 + vceq.i32 q14, q14, #0 + vceq.i32 q15, q15, #0 + orrs r3, r3, r12 + bne 90f + vst1.64 {q10}, [r2]! +3: vmov q10, q9 + vld1.8 {q11}, [r0]! + vorr q13, q12, q13 + vorr q15, q14, q15 + vbic q12, q10, q0 + vorr q3, q13, q15 + vext.8 q13, q10, q11, #1 + vext.8 q14, q10, q11, #2 + vext.8 q15, q10, q11, #3 + veor q12, q12, q1 + vorr d6, d6, d7 + vbic q13, q13, q0 + vbic q14, q14, q0 + vbic q15, q15, q0 + vceq.i32 q12, q12, #0 + vmov r3, r12, d6 + veor q13, q13, q1 + veor q14, q14, q1 + veor q15, q15, q1 + vceq.i32 q13, q13, #0 + vceq.i32 q14, q14, #0 + vceq.i32 q15, q15, #0 + orrs r3, r3, r12 + bne 91f + vst1.64 {q8}, [r2]! + subs r1, r1, #32 + bpl 2b + +90: add r0, r1, #48 + bx lr + +91: sub r1, r1, #16 + b 90b +endfunc