From patchwork Tue Nov 14 06:11:05 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Manojkumar Bhosale X-Patchwork-Id: 6041 Delivered-To: ffmpegpatchwork@gmail.com Received: by 10.2.161.94 with SMTP id m30csp2952833jah; Mon, 13 Nov 2017 22:11:20 -0800 (PST) X-Google-Smtp-Source: AGs4zMas1pHFbQa6VAeKj4IJ3b3xqKmGOpUWgSmxjEqEDiJm5cxz2jMgDFQdUfAuZPJ4XatI0yCg X-Received: by 10.28.157.7 with SMTP id g7mr6624174wme.89.1510639880225; Mon, 13 Nov 2017 22:11:20 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1510639880; cv=none; d=google.com; s=arc-20160816; b=Sbwy0AcxZ3scKldSH3gOQeMZ9K523t+InC6xr4/JuisHZH1jSgmEWseik/Ssl9hXD0 6/AOTC/FDGevJCXtKGwvaLH6qJYR7oAf7uF0ugOjRjYUlIs68BltltWdzJj2Wms+NYNq 0mphfE9g4FXuk9uY+wS7hhTskajMTB9ygEVtqW7xZcfI/+1eoGzmRuSaByMcb3C3qwlO GEogwXrN0OcPUjsSb/itlqOObMTh1A9Xm3LJXvsFUR7rGIk1NIPAHI/Y9p8UHs7t2Mmb NJzfKFGyitPITYcydykOtoYz9FfVVwh7x+aQG1y2o2CsB6KYzcUU/sFd5XWLV7Z5ZfJ/ NPAA== 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:content-language :accept-language:in-reply-to:references:message-id:date:thread-index :thread-topic:to:from:delivered-to:arc-authentication-results; bh=lHM7UcKpeb0L/vIZX4CHoW5gRZcI5IXhyD37IS9q1TY=; b=yU5rQDzlAA9vTO1x9NYR7YXhovjKKrLJqq0FvJD0/A76NltQhJzJmjjDT3AGcuwI9U dTvPcCvpgGheJgeb8dczVUQfHMlpMj5f3BHw0fbsMAh6ErgkjlB2Ji5SfbQVHCxHZEVc OzuBALhaxgJCNe/GUGioGQUoRP6VzzCWKzITeoTBituJkZsvskmb3MMrikzB7PHsL8Vg slLxRoJ5Y/uFDiSQOdHcpR//GNtPlKAfxLZLBnhAv+E4fJtniGnBOAeu0/1qX9Zoe1UK hAxEuO0HqAH4DwHHaruDCTIP4HXbBiK0/APnEWBBNSmSomTigjQelmD4+UGabC5ejvBp buWw== 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 f15si6960206wmg.97.2017.11.13.22.11.19; Mon, 13 Nov 2017 22:11:20 -0800 (PST) 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 8A7D7689FAB; Tue, 14 Nov 2017 08:11:03 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mailapp01.imgtec.com (mailapp01.imgtec.com [195.59.15.196]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id A4768689EAE for ; Tue, 14 Nov 2017 08:10:56 +0200 (EET) Received: from HHMAIL01.hh.imgtec.org (unknown [10.100.10.19]) by Forcepoint Email with ESMTPS id 4178AB4B5E274 for ; Tue, 14 Nov 2017 06:11:08 +0000 (GMT) Received: from HHMAIL-X.hh.imgtec.org (10.100.10.113) by HHMAIL01.hh.imgtec.org (10.100.10.19) with Microsoft SMTP Server (TLS) id 14.3.361.1; Tue, 14 Nov 2017 06:11:09 +0000 Received: from PUMAIL01.pu.imgtec.org (192.168.91.250) by HHMAIL-X.hh.imgtec.org (10.100.10.113) with Microsoft SMTP Server (TLS) id 14.3.361.1; Tue, 14 Nov 2017 06:11:09 +0000 Received: from PUMAIL01.pu.imgtec.org ([::1]) by PUMAIL01.pu.imgtec.org ([::1]) with mapi id 14.03.0266.001; Tue, 14 Nov 2017 11:41:06 +0530 From: Manojkumar Bhosale To: FFmpeg development discussions and patches Thread-Topic: [FFmpeg-devel] [PATCH] avcodec/mips: Improve hevc non-uni hz and vt mc msa functions Thread-Index: AQHTWULwjZtFBgipPUGlGSSPSmUPoqMTbCTA Date: Tue, 14 Nov 2017 06:11:05 +0000 Message-ID: <70293ACCC3BA6A4E81FFCA024C7A86E1014A74A339@PUMAIL01.pu.imgtec.org> References: <1510222189-16250-1-git-send-email-kaustubh.raste@imgtec.com> In-Reply-To: <1510222189-16250-1-git-send-email-kaustubh.raste@imgtec.com> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [192.168.91.86] MIME-Version: 1.0 Subject: Re: [FFmpeg-devel] [PATCH] avcodec/mips: Improve hevc non-uni hz and vt mc msa functions X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Kaustubh Raste Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" LGTM -----Original Message----- From: ffmpeg-devel [mailto:ffmpeg-devel-bounces@ffmpeg.org] On Behalf Of kaustubh.raste@imgtec.com Sent: Thursday, November 9, 2017 3:40 PM To: ffmpeg-devel@ffmpeg.org Cc: Kaustubh Raste Subject: [FFmpeg-devel] [PATCH] avcodec/mips: Improve hevc non-uni hz and vt mc msa functions From: Kaustubh Raste Use mask buffer. Signed-off-by: Kaustubh Raste --- libavcodec/mips/hevcdsp_msa.c | 541 ++++++++++++++++++++++++----------------- 1 file changed, 312 insertions(+), 229 deletions(-) uint32_t loop_cnt; - v16i8 src0, src1, src2, src3, src4, src5, src6, src7; + v16i8 src0, src1, src2, src3; v8i16 filt0, filt1, filt2, filt3; v16i8 mask1, mask2, mask3; v16i8 vec0, vec1, vec2, vec3; - v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; + v8i16 dst0, dst1, dst2, dst3; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); src -= 3; const_vec = __msa_ldi_h(128); @@ -587,56 +654,36 @@ static void hevc_hz_8t_16w_msa(uint8_t *src, int32_t src_stride, mask2 = mask0 + 4; mask3 = mask0 + 6; - for (loop_cnt = (height >> 2); loop_cnt--;) { - LD_SB4(src, src_stride, src0, src2, src4, src6); - LD_SB4(src + 8, src_stride, src1, src3, src5, src7); - src += (4 * src_stride); - XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7); + for (loop_cnt = (height >> 1); loop_cnt--;) { + LD_SB2(src, src_stride, src0, src2); + LD_SB2(src + 8, src_stride, src1, src3); + src += (2 * src_stride); + XORI_B4_128_SB(src0, src1, src2, src3); - VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst0 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst0, dst0, dst0, dst0); - VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst1 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst1, dst1, dst1, dst1); - VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst2 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst2, dst2, dst2, dst2); - VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst3 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst3, dst3, dst3, dst3); - VSHF_B4_SB(src4, src4, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); - dst4 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst4, dst4, dst4, dst4); - VSHF_B4_SB(src5, src5, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); - dst5 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst5, dst5, dst5, dst5); - VSHF_B4_SB(src6, src6, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); - dst6 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst6, dst6, dst6, dst6); - VSHF_B4_SB(src7, src7, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); - dst7 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst7, dst7, dst7, dst7); - - ST_SH4(dst0, dst2, dst4, dst6, dst, dst_stride); - ST_SH4(dst1, dst3, dst5, dst7, dst + 8, dst_stride); - dst += (4 * dst_stride); + VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0, vec1); + VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0, + dst1, dst2, dst3); + VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec0, vec1); + VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0, + dst1, dst2, dst3); + VSHF_B2_SB(src0, src0, src1, src1, mask2, mask2, vec0, vec1); + VSHF_B2_SB(src2, src2, src3, src3, mask2, mask2, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt2, filt2, filt2, filt2, dst0, + dst1, dst2, dst3); + VSHF_B2_SB(src0, src0, src1, src1, mask3, mask3, vec0, vec1); + VSHF_B2_SB(src2, src2, src3, src3, mask3, mask3, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt3, filt3, filt3, filt3, dst0, + dst1, dst2, dst3); + + ST_SH2(dst0, dst2, dst, dst_stride); + ST_SH2(dst1, dst3, dst + 8, dst_stride); + dst += (2 * dst_stride); } } @@ -648,10 +695,10 @@ static void hevc_hz_8t_24w_msa(uint8_t *src, int32_t src_stride, v16i8 src0, src1, src2, src3; v8i16 filt0, filt1, filt2, filt3; v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7; - v16i8 vec0, vec1, vec2, vec3; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; v8i16 dst0, dst1, dst2, dst3, dst4, dst5; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); src -= 3; filter_vec = LD_SH(filter); @@ -675,36 +722,36 @@ static void hevc_hz_8t_24w_msa(uint8_t *src, int32_t src_stride, src += src_stride; XORI_B4_128_SB(src0, src1, src2, src3); - VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst0 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst0, dst0, dst0, dst0); - VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7, - vec0, vec1, vec2, vec3); dst1 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst1, dst1, dst1, dst1); - VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst2 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst2, dst2, dst2, dst2); - VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst3 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst3, dst3, dst3, dst3); - VSHF_B4_SB(src2, src3, mask4, mask5, mask6, mask7, - vec0, vec1, vec2, vec3); dst4 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst4, dst4, dst4, dst4); - VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst5 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst5, dst5, dst5, dst5); + VSHF_B2_SB(src0, src0, src0, src1, mask0, mask4, vec0, vec1); + VSHF_B2_SB(src1, src1, src2, src2, mask0, mask0, vec2, vec3); + VSHF_B2_SB(src2, src3, src3, src3, mask4, mask0, vec4, vec5); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0, + dst1, dst2, dst3); + DPADD_SB2_SH(vec4, vec5, filt0, filt0, dst4, dst5); + VSHF_B2_SB(src0, src0, src0, src1, mask1, mask5, vec0, vec1); + VSHF_B2_SB(src1, src1, src2, src2, mask1, mask1, vec2, vec3); + VSHF_B2_SB(src2, src3, src3, src3, mask5, mask1, vec4, vec5); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0, + dst1, dst2, dst3); + DPADD_SB2_SH(vec4, vec5, filt1, filt1, dst4, dst5); + VSHF_B2_SB(src0, src0, src0, src1, mask2, mask6, vec0, vec1); + VSHF_B2_SB(src1, src1, src2, src2, mask2, mask2, vec2, vec3); + VSHF_B2_SB(src2, src3, src3, src3, mask6, mask2, vec4, vec5); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt2, filt2, filt2, filt2, dst0, + dst1, dst2, dst3); + DPADD_SB2_SH(vec4, vec5, filt2, filt2, dst4, dst5); + VSHF_B2_SB(src0, src0, src0, src1, mask3, mask7, vec0, vec1); + VSHF_B2_SB(src1, src1, src2, src2, mask3, mask3, vec2, vec3); + VSHF_B2_SB(src2, src3, src3, src3, mask7, mask3, vec4, vec5); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt3, filt3, filt3, filt3, dst0, + dst1, dst2, dst3); + DPADD_SB2_SH(vec4, vec5, filt3, filt3, dst4, dst5); ST_SH2(dst0, dst1, dst, 8); ST_SH(dst2, dst + 16); @@ -726,7 +773,7 @@ static void hevc_hz_8t_32w_msa(uint8_t *src, int32_t src_stride, v16i8 vec0, vec1, vec2, vec3; v8i16 dst0, dst1, dst2, dst3; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); src -= 3; filter_vec = LD_SH(filter); @@ -783,10 +830,10 @@ static void hevc_hz_8t_48w_msa(uint8_t *src, int32_t src_stride, v16i8 src0, src1, src2, src3; v8i16 filt0, filt1, filt2, filt3; v16i8 mask1, mask2, mask3, mask4, mask5, mask6, mask7; - v16i8 vec0, vec1, vec2, vec3; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; v8i16 dst0, dst1, dst2, dst3, dst4, dst5; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); src -= 3; filter_vec = LD_SH(filter); @@ -809,38 +856,39 @@ static void hevc_hz_8t_48w_msa(uint8_t *src, int32_t src_stride, src += src_stride; XORI_B4_128_SB(src0, src1, src2, src3); - VSHF_B4_SB(src0, src0, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst0 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst0, dst0, dst0, dst0); - VSHF_B4_SB(src0, src1, mask4, mask5, mask6, mask7, - vec0, vec1, vec2, vec3); dst1 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst1, dst1, dst1, dst1); - VSHF_B4_SB(src1, src1, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst2 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst2, dst2, dst2, dst2); - VSHF_B4_SB(src1, src2, mask4, mask5, mask6, mask7, - vec0, vec1, vec2, vec3); dst3 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst3, dst3, dst3, dst3); - VSHF_B4_SB(src2, src2, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst4 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst4, dst4, dst4, dst4); - VSHF_B4_SB(src3, src3, mask0, mask1, mask2, mask3, - vec0, vec1, vec2, vec3); dst5 = const_vec; - DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt1, filt2, filt3, - dst5, dst5, dst5, dst5); + VSHF_B2_SB(src0, src0, src0, src1, mask0, mask4, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src2, mask0, mask4, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0, + dst1, dst2, dst3); + VSHF_B2_SB(src0, src0, src0, src1, mask1, mask5, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src2, mask1, mask5, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0, + dst1, dst2, dst3); + VSHF_B2_SB(src0, src0, src0, src1, mask2, mask6, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src2, mask2, mask6, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt2, filt2, filt2, filt2, dst0, + dst1, dst2, dst3); + VSHF_B2_SB(src0, src0, src0, src1, mask3, mask7, vec0, vec1); + VSHF_B2_SB(src1, src1, src1, src2, mask3, mask7, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt3, filt3, filt3, filt3, dst0, + dst1, dst2, dst3); + ST_SH4(dst0, dst1, dst2, dst3, dst, 8); - ST_SH6(dst0, dst1, dst2, dst3, dst4, dst5, dst, 8); + VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec4, vec5); + DPADD_SB2_SH(vec4, vec5, filt0, filt0, dst4, dst5); + VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec4, vec5); + DPADD_SB2_SH(vec4, vec5, filt1, filt1, dst4, dst5); + VSHF_B2_SB(src2, src2, src3, src3, mask2, mask2, vec4, vec5); + DPADD_SB2_SH(vec4, vec5, filt2, filt2, dst4, dst5); + VSHF_B2_SB(src2, src2, src3, src3, mask3, mask3, vec4, vec5); + DPADD_SB2_SH(vec4, vec5, filt3, filt3, dst4, dst5); + ST_SH2(dst4, dst5, (dst + 32), 8); dst += dst_stride; } } @@ -856,7 +904,7 @@ static void hevc_hz_8t_64w_msa(uint8_t *src, int32_t src_stride, v16i8 vec0, vec1, vec2, vec3; v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); src -= 3; @@ -1805,7 +1853,7 @@ static void hevc_hz_4t_4x2_msa(uint8_t *src, v16i8 mask1, vec0, vec1; v8i16 dst0; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16); src -= 1; @@ -1838,7 +1886,7 @@ static void hevc_hz_4t_4x4_msa(uint8_t *src, v16i8 mask1, vec0, vec1; v8i16 dst0, dst1; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16); src -= 1; @@ -1877,7 +1925,7 @@ static void hevc_hz_4t_4x8multiple_msa(uint8_t *src, v16i8 mask1, vec0, vec1; v8i16 dst0, dst1, dst2, dst3; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16); src -= 1; @@ -1942,7 +1990,7 @@ static void hevc_hz_4t_6w_msa(uint8_t *src, uint32_t dst_val_int0, dst_val_int1, dst_val_int2, dst_val_int3; v8i16 filt0, filt1, dst0, dst1, dst2, dst3; v16i8 src0, src1, src2, src3; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); v16i8 mask1; v16i8 vec0, vec1; v8i16 filter_vec, const_vec; @@ -1957,7 +2005,7 @@ static void hevc_hz_4t_6w_msa(uint8_t *src, const_vec = __msa_ldi_h(128); const_vec <<= 6; - for (loop_cnt = (height >> 2); loop_cnt--;) { + for (loop_cnt = 2; loop_cnt--;) { LD_SB4(src, src_stride, src0, src1, src2, src3); src += (4 * src_stride); @@ -2011,7 +2059,7 @@ static void hevc_hz_4t_8x2multiple_msa(uint8_t *src, uint32_t loop_cnt; v8i16 filt0, filt1, dst0, dst1; v16i8 src0, src1; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); v16i8 mask1; v16i8 vec0, vec1; v8i16 filter_vec, const_vec; @@ -2055,7 +2103,7 @@ static void hevc_hz_4t_8x4multiple_msa(uint8_t *src, uint32_t loop_cnt; v8i16 filt0, filt1; v16i8 src0, src1, src2, src3; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); v16i8 mask1; v16i8 vec0, vec1; v8i16 dst0, dst1, dst2, dst3; @@ -2129,7 +2177,7 @@ static void hevc_hz_4t_12w_msa(uint8_t *src, v8i16 dst0, dst1, dst2, dst3, dst4, dst5; v8i16 filter_vec, const_vec; v16i8 mask3; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); v16i8 mask2 = { 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28 }; @@ -2186,7 +2234,7 @@ static void hevc_hz_4t_16w_msa(uint8_t *src, v16i8 src0, src1, src2, src3; v16i8 src4, src5, src6, src7; v8i16 filt0, filt1; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); v16i8 mask1; v8i16 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; v16i8 vec0, vec1; @@ -2258,7 +2306,7 @@ static void hevc_hz_4t_24w_msa(uint8_t *src, int16_t *dst_tmp = dst + 16; v16i8 src0, src1, src2, src3, src4, src5, src6, src7; v8i16 filt0, filt1; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); v16i8 mask1, mask00, mask11; v16i8 vec0, vec1; v8i16 dst0, dst1, dst2, dst3; @@ -2358,10 +2406,10 @@ static void hevc_hz_4t_32w_msa(uint8_t *src, uint32_t loop_cnt; v16i8 src0, src1, src2; v8i16 filt0, filt1; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); v16i8 mask1, mask2, mask3; v8i16 dst0, dst1, dst2, dst3; - v16i8 vec0, vec1; + v16i8 vec0, vec1, vec2, vec3; v8i16 filter_vec, const_vec; src -= 1; @@ -2376,54 +2424,25 @@ static void hevc_hz_4t_32w_msa(uint8_t *src, mask2 = mask0 + 8; mask3 = mask0 + 10; - for (loop_cnt = (height >> 1); loop_cnt--;) { + for (loop_cnt = height; loop_cnt--;) { LD_SB2(src, 16, src0, src1); src2 = LD_SB(src + 24); src += src_stride; XORI_B3_128_SB(src0, src1, src2); - VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); dst0 = const_vec; - DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); - - VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1); dst1 = const_vec; - DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); - - VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); dst2 = const_vec; - DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); - - VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); dst3 = const_vec; - DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); - - ST_SH4(dst0, dst1, dst2, dst3, dst, 8); - dst += dst_stride; - - LD_SB2(src, 16, src0, src1); - src2 = LD_SB(src + 24); - src += src_stride; - - XORI_B3_128_SB(src0, src1, src2); - - VSHF_B2_SB(src0, src0, src0, src0, mask0, mask1, vec0, vec1); - dst0 = const_vec; - DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst0, dst0); - - VSHF_B2_SB(src0, src1, src0, src1, mask2, mask3, vec0, vec1); - dst1 = const_vec; - DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst1, dst1); - - VSHF_B2_SB(src1, src1, src1, src1, mask0, mask1, vec0, vec1); - dst2 = const_vec; - DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst2, dst2); - - VSHF_B2_SB(src2, src2, src2, src2, mask0, mask1, vec0, vec1); - dst3 = const_vec; - DPADD_SB2_SH(vec0, vec1, filt0, filt1, dst3, dst3); - + VSHF_B2_SB(src0, src0, src0, src1, mask0, mask2, vec0, vec1); + VSHF_B2_SB(src1, src1, src2, src2, mask0, mask0, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0, + dst1, dst2, dst3); + VSHF_B2_SB(src0, src0, src0, src1, mask1, mask3, vec0, vec1); + VSHF_B2_SB(src1, src1, src2, src2, mask1, mask1, vec2, vec3); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0, + dst1, dst2, dst3); ST_SH4(dst0, dst1, dst2, dst3, dst, 8); dst += dst_stride; } @@ -2499,18 +2518,17 @@ static void hevc_vt_4t_4x4_msa(uint8_t *src, ST8x4_UB(dst10, dst32, dst, 2 * dst_stride); } -static void hevc_vt_4t_4x8multiple_msa(uint8_t *src, - int32_t src_stride, - int16_t *dst, - int32_t dst_stride, - const int8_t *filter, - int32_t height) +static void hevc_vt_4t_4x8_msa(uint8_t *src, + int32_t src_stride, + int16_t *dst, + int32_t dst_stride, + const int8_t *filter, + int32_t height) { - int32_t loop_cnt; - v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, + src10; v16i8 src10_r, src32_r, src54_r, src76_r, src98_r; v16i8 src21_r, src43_r, src65_r, src87_r, src109_r; - v16i8 src2110, src4332, src6554, src8776; + v16i8 src2110, src4332, src6554, src8776, src10998; v8i16 dst10, dst32, dst54, dst76; v8i16 filt0, filt1; v8i16 filter_vec, const_vec; @@ -2529,35 +2547,96 @@ static void hevc_vt_4t_4x8multiple_msa(uint8_t *src, src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r); src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128); - for (loop_cnt = (height >> 3); loop_cnt--;) { - LD_SB6(src, src_stride, src3, src4, src5, src6, src7, src8); - src += (6 * src_stride); + LD_SB8(src, src_stride, src3, src4, src5, src6, src7, src8, src9, src10); + src += (8 * src_stride); + ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5, + src32_r, src43_r, src54_r, src65_r); + ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, + src76_r, src87_r, src98_r, src109_r); + ILVR_D4_SB(src43_r, src32_r, src65_r, src54_r, src87_r, src76_r, src109_r, + src98_r, src4332, src6554, src8776, src10998); + XORI_B4_128_SB(src4332, src6554, src8776, src10998); + dst10 = const_vec; + dst32 = const_vec; + dst54 = const_vec; + dst76 = const_vec; + DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10); + DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst32, dst32); + DPADD_SB2_SH(src6554, src8776, filt0, filt1, dst54, dst54); + DPADD_SB2_SH(src8776, src10998, filt0, filt1, dst76, dst76); + ST8x8_UB(dst10, dst32, dst54, dst76, dst, 2 * dst_stride); + dst += (8 * dst_stride); +} - ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5, - src32_r, src43_r, src54_r, src65_r); - ILVR_B2_SB(src7, src6, src8, src7, src76_r, src87_r); - ILVR_D3_SB(src43_r, src32_r, src65_r, src54_r, src87_r, src76_r, - src4332, src6554, src8776); - XORI_B3_128_SB(src4332, src6554, src8776); +static void hevc_vt_4t_4x16_msa(uint8_t *src, int32_t src_stride, + int16_t *dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; + v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r; + v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776; + v16i8 src10998; + v8i16 dst10, dst32, dst54, dst76, filt0, filt1, filter_vec, +const_vec; - dst10 = const_vec; - DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10); - dst32 = const_vec; - DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst32, dst32); - dst54 = const_vec; - DPADD_SB2_SH(src6554, src8776, filt0, filt1, dst54, dst54); + src -= src_stride; + const_vec = __msa_ldi_h(128); + const_vec <<= 6; - LD_SB2(src, src_stride, src9, src2); - src += (2 * src_stride); - ILVR_B2_SB(src9, src8, src2, src9, src98_r, src109_r); - src2110 = (v16i8) __msa_ilvr_d((v2i64) src109_r, (v2i64) src98_r); - src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128); - dst76 = const_vec; - DPADD_SB2_SH(src8776, src2110, filt0, filt1, dst76, dst76); + filter_vec = LD_SH(filter); + SPLATI_H2_SH(filter_vec, 0, 1, filt0, filt1); - ST8x8_UB(dst10, dst32, dst54, dst76, dst, 2 * dst_stride); - dst += (8 * dst_stride); - } + LD_SB3(src, src_stride, src0, src1, src2); + src += (3 * src_stride); + + ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); + src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_r, (v2i64) src10_r); + src2110 = (v16i8) __msa_xori_b((v16u8) src2110, 128); + + LD_SB8(src, src_stride, src3, src4, src5, src6, src7, src8, src9, src10); + src += (8 * src_stride); + ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5, src32_r, src43_r, + src54_r, src65_r); + ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r, + src87_r, src98_r, src109_r); + ILVR_D4_SB(src43_r, src32_r, src65_r, src54_r, src87_r, src76_r, src109_r, + src98_r, src4332, src6554, src8776, src10998); + XORI_B4_128_SB(src4332, src6554, src8776, src10998); + + dst10 = const_vec; + dst32 = const_vec; + dst54 = const_vec; + dst76 = const_vec; + DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10); + DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst32, dst32); + DPADD_SB2_SH(src6554, src8776, filt0, filt1, dst54, dst54); + DPADD_SB2_SH(src8776, src10998, filt0, filt1, dst76, dst76); + ST8x8_UB(dst10, dst32, dst54, dst76, dst, 2 * dst_stride); + dst += (8 * dst_stride); + + src2 = src10; + src2110 = src10998; + + LD_SB8(src, src_stride, src3, src4, src5, src6, src7, src8, src9, src10); + src += (8 * src_stride); + + ILVR_B4_SB(src3, src2, src4, src3, src5, src4, src6, src5, src32_r, src43_r, + src54_r, src65_r); + ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r, + src87_r, src98_r, src109_r); + ILVR_D4_SB(src43_r, src32_r, src65_r, src54_r, src87_r, src76_r, src109_r, + src98_r, src4332, src6554, src8776, src10998); + XORI_B4_128_SB(src4332, src6554, src8776, src10998); + + dst10 = const_vec; + dst32 = const_vec; + dst54 = const_vec; + dst76 = const_vec; + DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst10, dst10); + DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst32, dst32); + DPADD_SB2_SH(src6554, src8776, filt0, filt1, dst54, dst54); + DPADD_SB2_SH(src8776, src10998, filt0, filt1, dst76, dst76); + ST8x8_UB(dst10, dst32, dst54, dst76, dst, 2 * dst_stride); + dst += (8 * dst_stride); } static void hevc_vt_4t_4w_msa(uint8_t *src, @@ -2571,9 +2650,10 @@ static void hevc_vt_4t_4w_msa(uint8_t *src, hevc_vt_4t_4x2_msa(src, src_stride, dst, dst_stride, filter); } else if (4 == height) { hevc_vt_4t_4x4_msa(src, src_stride, dst, dst_stride, filter, height); - } else if (0 == (height % 8)) { - hevc_vt_4t_4x8multiple_msa(src, src_stride, dst, dst_stride, - filter, height); + } else if (8 == height) { + hevc_vt_4t_4x8_msa(src, src_stride, dst, dst_stride, filter, height); + } else if (16 == height) { + hevc_vt_4t_4x16_msa(src, src_stride, dst, dst_stride, filter, + height); } } @@ -2756,9 +2836,9 @@ static void hevc_vt_4t_8x4multiple_msa(uint8_t *src, int32_t height) { int32_t loop_cnt; - v16i8 src0, src1, src2, src3, src4, src5; - v16i8 src10_r, src32_r, src21_r, src43_r; - v8i16 dst0_r, dst1_r; + v16i8 src0, src1, src2, src3, src4, src5, src6; + v16i8 src10_r, src32_r, src21_r, src43_r, src54_r, src65_r; + v8i16 dst0_r, dst1_r, dst2_r, dst3_r; v8i16 filt0, filt1; v8i16 filter_vec, const_vec; @@ -2775,29 +2855,25 @@ static void hevc_vt_4t_8x4multiple_msa(uint8_t *src, ILVR_B2_SB(src1, src0, src2, src1, src10_r, src21_r); for (loop_cnt = (height >> 2); loop_cnt--;) { - LD_SB2(src, src_stride, src3, src4); - src += (2 * src_stride); - XORI_B2_128_SB(src3, src4); + LD_SB4(src, src_stride, src3, src4, src5, src6); + src += (4 * src_stride); + XORI_B4_128_SB(src3, src4, src5, src6); ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); + ILVR_B2_SB(src5, src4, src6, src5, src54_r, src65_r); dst0_r = const_vec; - DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); dst1_r = const_vec; + dst2_r = const_vec; + dst3_r = const_vec; + DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); + DPADD_SB2_SH(src32_r, src54_r, filt0, filt1, dst2_r, dst2_r); + DPADD_SB2_SH(src43_r, src65_r, filt0, filt1, dst3_r, dst3_r); + ST_SH4(dst0_r, dst1_r, dst2_r, dst3_r, dst, dst_stride); + dst += (4 * dst_stride); - ST_SH2(dst0_r, dst1_r, dst, dst_stride); - dst += (2 * dst_stride); - - LD_SB2(src, src_stride, src5, src2); - src += (2 * src_stride); - XORI_B2_128_SB(src5, src2); - ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r); - dst0_r = const_vec; - DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst0_r, dst0_r); - dst1_r = const_vec; - DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst1_r, dst1_r); - - ST_SH2(dst0_r, dst1_r, dst, dst_stride); - dst += (2 * dst_stride); + src2 = src6; + src10_r = src54_r; + src21_r = src65_r; } } @@ -2826,11 +2902,12 @@ static void hevc_vt_4t_12w_msa(uint8_t *src, int32_t height) { int32_t loop_cnt; - v16i8 src0, src1, src2, src3, src4, src5; + v16i8 src0, src1, src2, src3, src4, src5, src6; v16i8 src10_r, src32_r, src21_r, src43_r; v8i16 dst0_r, dst1_r, dst2_r, dst3_r; v16i8 src10_l, src32_l, src54_l, src21_l, src43_l, src65_l; v16i8 src2110, src4332; + v16i8 src54_r, src65_r, src6554; v8i16 dst0_l, dst1_l; v8i16 filt0, filt1; v8i16 filter_vec, const_vec; @@ -2849,36 +2926,42 @@ static void hevc_vt_4t_12w_msa(uint8_t *src, ILVL_B2_SB(src1, src0, src2, src1, src10_l, src21_l); src2110 = (v16i8) __msa_ilvr_d((v2i64) src21_l, (v2i64) src10_l); - for (loop_cnt = (height >> 2); loop_cnt--;) { + for (loop_cnt = 4; loop_cnt--;) { LD_SB2(src, src_stride, src3, src4); src += (2 * src_stride); + LD_SB2(src, src_stride, src5, src6); + src += (2 * src_stride); XORI_B2_128_SB(src3, src4); + XORI_B2_128_SB(src5, src6); + ILVR_B2_SB(src3, src2, src4, src3, src32_r, src43_r); ILVL_B2_SB(src3, src2, src4, src3, src32_l, src43_l); src4332 = (v16i8) __msa_ilvr_d((v2i64) src43_l, (v2i64) src32_l); + ILVR_B2_SB(src5, src4, src6, src5, src54_r, src65_r); + ILVL_B2_SB(src5, src4, src6, src5, src54_l, src65_l); + src6554 = (v16i8) __msa_ilvr_d((v2i64) src65_l, (v2i64) + src54_l); + dst0_r = const_vec; DPADD_SB2_SH(src10_r, src32_r, filt0, filt1, dst0_r, dst0_r); dst1_r = const_vec; DPADD_SB2_SH(src21_r, src43_r, filt0, filt1, dst1_r, dst1_r); - dst0_l = const_vec; - DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst0_l, dst0_l); - - LD_SB2(src, src_stride, src5, src2); - src += (2 * src_stride); - XORI_B2_128_SB(src5, src2); - ILVR_B2_SB(src5, src4, src2, src5, src10_r, src21_r); - ILVL_B2_SB(src5, src4, src2, src5, src54_l, src65_l); - src2110 = (v16i8) __msa_ilvr_d((v2i64) src65_l, (v2i64) src54_l); dst2_r = const_vec; - DPADD_SB2_SH(src32_r, src10_r, filt0, filt1, dst2_r, dst2_r); + DPADD_SB2_SH(src32_r, src54_r, filt0, filt1, dst2_r, dst2_r); dst3_r = const_vec; - DPADD_SB2_SH(src43_r, src21_r, filt0, filt1, dst3_r, dst3_r); + DPADD_SB2_SH(src43_r, src65_r, filt0, filt1, dst3_r, dst3_r); + dst0_l = const_vec; + DPADD_SB2_SH(src2110, src4332, filt0, filt1, dst0_l, dst0_l); dst1_l = const_vec; - DPADD_SB2_SH(src4332, src2110, filt0, filt1, dst1_l, dst1_l); + DPADD_SB2_SH(src4332, src6554, filt0, filt1, dst1_l, dst1_l); ST_SH4(dst0_r, dst1_r, dst2_r, dst3_r, dst, dst_stride); ST8x4_UB(dst0_l, dst1_l, dst + 8, (2 * dst_stride)); dst += (4 * dst_stride); + + src2 = src6; + src10_r = src54_r; + src21_r = src65_r; + src2110 = src6554; } } -- 1.7.9.5 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel diff --git a/libavcodec/mips/hevcdsp_msa.c b/libavcodec/mips/hevcdsp_msa.c index b17127c..81db62b 100644 --- a/libavcodec/mips/hevcdsp_msa.c +++ b/libavcodec/mips/hevcdsp_msa.c @@ -456,7 +456,7 @@ static void hevc_hz_8t_4w_msa(uint8_t *src, int32_t src_stride, v16i8 vec0, vec1, vec2, vec3; v8i16 dst0, dst1, dst2, dst3; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr + 16); src -= 3; const_vec = __msa_ldi_h(128); @@ -511,7 +511,7 @@ static void hevc_hz_8t_8w_msa(uint8_t *src, int32_t src_stride, v16i8 vec0, vec1, vec2, vec3; v8i16 dst0, dst1, dst2, dst3; v8i16 filter_vec, const_vec; - v16i8 mask0 = { 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8 }; + v16i8 mask0 = LD_SB(ff_hevc_mask_arr); src -= 3; const_vec = __msa_ldi_h(128); @@ -559,8 +559,75 @@ static void hevc_hz_8t_12w_msa(uint8_t *src, int32_t src_stride, int16_t *dst, int32_t dst_stride, const int8_t *filter, int32_t height) { - hevc_hz_8t_8w_msa(src, src_stride, dst, dst_stride, filter, height); - hevc_hz_8t_4w_msa(src + 8, src_stride, dst + 8, dst_stride, filter, height); + uint32_t loop_cnt; + int64_t res0, res1, res2, res3; + v16i8 src0, src1, src2, src3, src4, src5, src6, src7; + v16i8 mask0, mask1, mask2, mask3, mask4, mask5, mask6, mask7; + v16i8 vec0, vec1, vec2, vec3, vec4, vec5; + v8i16 filt0, filt1, filt2, filt3, dst0, dst1, dst2, dst3, dst4, dst5; + v8i16 filter_vec, const_vec; + + src -= 3; + const_vec = __msa_ldi_h(128); + const_vec <<= 6; + + filter_vec = LD_SH(filter); + SPLATI_H4_SH(filter_vec, 0, 1, 2, 3, filt0, filt1, filt2, filt3); + + mask0 = LD_SB(ff_hevc_mask_arr); + mask1 = mask0 + 2; + mask2 = mask0 + 4; + mask3 = mask0 + 6; + mask4 = LD_SB(ff_hevc_mask_arr + 16); + mask5 = mask4 + 2; + mask6 = mask4 + 4; + mask7 = mask4 + 6; + + for (loop_cnt = 4; loop_cnt--;) { + LD_SB4(src, src_stride, src0, src1, src2, src3); + LD_SB4(src + 8, src_stride, src4, src5, src6, src7); + src += (4 * src_stride); + XORI_B8_128_SB(src0, src1, src2, src3, src4, src5, src6, src7); + + dst0 = const_vec; + dst1 = const_vec; + dst2 = const_vec; + dst3 = const_vec; + dst4 = const_vec; + dst5 = const_vec; + VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0, vec1); + VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2, vec3); + VSHF_B2_SB(src4, src5, src6, src7, mask4, mask4, vec4, vec5); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, dst0, + dst1, dst2, dst3); + DPADD_SB2_SH(vec4, vec5, filt0, filt0, dst4, dst5); + VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec0, vec1); + VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec2, vec3); + VSHF_B2_SB(src4, src5, src6, src7, mask5, mask5, vec4, vec5); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt1, filt1, filt1, filt1, dst0, + dst1, dst2, dst3); + DPADD_SB2_SH(vec4, vec5, filt1, filt1, dst4, dst5); + VSHF_B2_SB(src0, src0, src1, src1, mask2, mask2, vec0, vec1); + VSHF_B2_SB(src2, src2, src3, src3, mask2, mask2, vec2, vec3); + VSHF_B2_SB(src4, src5, src6, src7, mask6, mask6, vec4, vec5); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt2, filt2, filt2, filt2, dst0, + dst1, dst2, dst3); + DPADD_SB2_SH(vec4, vec5, filt2, filt2, dst4, dst5); + VSHF_B2_SB(src0, src0, src1, src1, mask3, mask3, vec0, vec1); + VSHF_B2_SB(src2, src2, src3, src3, mask3, mask3, vec2, vec3); + VSHF_B2_SB(src4, src5, src6, src7, mask7, mask7, vec4, vec5); + DPADD_SB4_SH(vec0, vec1, vec2, vec3, filt3, filt3, filt3, filt3, dst0, + dst1, dst2, dst3); + DPADD_SB2_SH(vec4, vec5, filt3, filt3, dst4, dst5); + + res0 = __msa_copy_s_d((v2i64) dst4, 0); + res1 = __msa_copy_s_d((v2i64) dst4, 1); + res2 = __msa_copy_s_d((v2i64) dst5, 0); + res3 = __msa_copy_s_d((v2i64) dst5, 1); + ST_SH4(dst0, dst1, dst2, dst3, dst, dst_stride); + SD4(res0, res1, res2, res3, (dst + 8), dst_stride); + dst += (4 * dst_stride); + } } static void hevc_hz_8t_16w_msa(uint8_t *src, int32_t src_stride, @@ -568,13 +635,13 @@ static void hevc_hz_8t_16w_msa(uint8_t *src, int32_t src_stride, const int8_t *filter, int32_t height) {