From patchwork Thu Jul 2 15:46:20 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jiaxun Yang X-Patchwork-Id: 20804 Delivered-To: andriy.gelman@gmail.com Received: by 2002:a25:80ca:0:0:0:0:0 with SMTP id c10csp1582501ybm; Thu, 2 Jul 2020 08:48:25 -0700 (PDT) X-Google-Smtp-Source: ABdhPJxTIdnWHsfTmoPhWGxSMdlj4rJfJdeq20QQh7zBZRln1hIsGSY9D0HR1xNCRS0FVSWee7Vx X-Received: by 2002:a1c:48c5:: with SMTP id v188mr31263467wma.58.1593704904793; Thu, 02 Jul 2020 08:48:24 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1593704904; cv=none; d=google.com; s=arc-20160816; b=gK/MGfSLwYWj7kqV07uCepSHOJW7/5efC8OxvWuj+6ToTVqcGfMTE/I+g4Ev6VHjSh QJ39VLIqY86+/cSjcVeo4fYzvMjINSbziV8wuFCu+nf/zmd9iCOY41ioItob4nfCiHU4 moFHnpnGq/6Dy/WBTd2MzvkEslXsKq9/vIjdQPm4wiT+gmRZv1dFy2dTWNkxyOw8I4rt 4HoL23aPtxcl4ldrshny621yvAicBvEGRYB7hUUTBw3aWjyRTk6nHexizQPf/cj2ce+w kuFQ/Iprg2Bz6kqouirXMmJ4PvKZM+2uMh5nVCMM238sET3SDGvlvo/BmItiBlERYXul nbNg== 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:dkim-signature:ai-spam-status :dkim-signature:delivered-to; bh=8KpuCf8yKaAJNZXFmMdF0dCdstYFQVtaDHGnAW4VJJY=; b=nkgzmQTS+1nyAXY61NzuUCN0iA+r9q+rkIo3e0S/vNU5xucNqMQZ4J4YZqR6wVPzlT YCPJRGkXHj79qQlB/QbMuub+4oNojlbEk6MxBMliTa8iqTDSqX56YhCog3Du/MN6aZUg LuDZjILN7ILvWJhjOAAG85092mmna/e5BT22ijpCtHXZyBwrWq6lAj6onFVo9Mr/cH3v Wmv49sHRYiX+c5lKUK6iZ+ESq73wv/lK0n+8kahcKW+fP6TXOj4E8pSBx9qoWNCVoSYn lmGAco4lrF1TKBbodF8lV0Y0gSgG6LkzV12IRTFBkkdNEBFDUG3MIxeFCeKq/yXec8Hx T0bw== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@mymailcheap.com header.s=default header.b=WQ3+S6jv; dkim=neutral (body hash did not verify) header.i=@flygoat.com header.s=default header.b=hTHfWUpG; 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; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=flygoat.com Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id s2si8757787wrm.172.2020.07.02.08.48.24; Thu, 02 Jul 2020 08:48:24 -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; dkim=neutral (body hash did not verify) header.i=@mymailcheap.com header.s=default header.b=WQ3+S6jv; dkim=neutral (body hash did not verify) header.i=@flygoat.com header.s=default header.b=hTHfWUpG; 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; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=flygoat.com Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 10CB368B1AE; Thu, 2 Jul 2020 18:48:04 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from relay1.mymailcheap.com (relay1.mymailcheap.com [144.217.248.102]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 05A9068B162 for ; Thu, 2 Jul 2020 18:48:01 +0300 (EEST) Received: from filter2.mymailcheap.com (filter2.mymailcheap.com [91.134.140.82]) by relay1.mymailcheap.com (Postfix) with ESMTPS id 77D723F157 for ; Thu, 2 Jul 2020 11:48:00 -0400 (EDT) Received: from localhost (localhost [127.0.0.1]) by filter2.mymailcheap.com (Postfix) with ESMTP id A59652A4F0 for ; Thu, 2 Jul 2020 17:47:59 +0200 (CEST) DKIM-Signature: v=1; a=rsa-sha256; c=simple/simple; d=mymailcheap.com; s=default; t=1593704879; bh=MM8So4rMJN04tB+vxhzJQGj0EW1O8QEMFBWB7ADDpLc=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=WQ3+S6jvxwK8FERzosr8NBxK+CCcB8g9GC89KPjfOa8D+FknmZHwH4KlCwQJZ41CO Lv/MbBSQqPPO2lfimwKzATpEcNH9nG1ydwTvoGLBYa4+JsYr0b7InvOiaZO1PQqp8o N5Q0L9A7b1+B/zRWcrMeEyRC8cskKAd4WAu3nZZE= X-Virus-Scanned: Debian amavisd-new at filter2.mymailcheap.com Received: from filter2.mymailcheap.com ([127.0.0.1]) by localhost (filter2.mymailcheap.com [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id jdTckojpURQu for ; Thu, 2 Jul 2020 17:47:53 +0200 (CEST) Received: from mail20.mymailcheap.com (mail20.mymailcheap.com [51.83.111.147]) (using TLSv1.2 with cipher ADH-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by filter2.mymailcheap.com (Postfix) with ESMTPS for ; Thu, 2 Jul 2020 17:47:52 +0200 (CEST) Received: from [148.251.23.173] (ml.mymailcheap.com [148.251.23.173]) by mail20.mymailcheap.com (Postfix) with ESMTP id DE8114085C; Thu, 2 Jul 2020 15:47:50 +0000 (UTC) Authentication-Results: mail20.mymailcheap.com; dkim=pass (1024-bit key; unprotected) header.d=flygoat.com header.i=@flygoat.com header.b="hTHfWUpG"; dkim-atps=neutral AI-Spam-Status: Not processed Received: from halation.202.net.flygoat.com (unknown [115.227.164.72]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange ECDHE (P-256) server-signature RSA-PSS (2048 bits) server-digest SHA256) (No client certificate requested) by mail20.mymailcheap.com (Postfix) with ESMTPSA id 23129403B3; Thu, 2 Jul 2020 15:46:38 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=simple/simple; d=flygoat.com; s=default; t=1593704804; bh=MM8So4rMJN04tB+vxhzJQGj0EW1O8QEMFBWB7ADDpLc=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=hTHfWUpGYVty1/+ieqSiHv0pNHI2mLD1gz5BR8/wWCczD9NQPYeN0Ij+SJe5iXsQQ 4JjCbJ4KmW1dyVKnvu7+ELArtMtjzWQfme3/PHFiYWJJHuNgc+UsoeO6U8xwatnTKT zNkYM0qDPMOLtZ4+Ks4gYAnw6spzC5C67ExuTsGM= From: Jiaxun Yang To: ffmpeg-devel@ffmpeg.org Date: Thu, 2 Jul 2020 23:46:20 +0800 Message-Id: <20200702154622.21280-5-jiaxun.yang@flygoat.com> X-Mailer: git-send-email 2.27.0 In-Reply-To: <20200702154622.21280-1-jiaxun.yang@flygoat.com> References: <20200702154622.21280-1-jiaxun.yang@flygoat.com> MIME-Version: 1.0 X-Rspamd-Queue-Id: DE8114085C X-Spamd-Result: default: False [4.90 / 10.00]; ARC_NA(0.00)[]; RCVD_VIA_SMTP_AUTH(0.00)[]; R_DKIM_ALLOW(0.00)[flygoat.com:s=default]; FROM_HAS_DN(0.00)[]; TO_DN_SOME(0.00)[]; R_MISSING_CHARSET(2.50)[]; TO_MATCH_ENVRCPT_ALL(0.00)[]; MIME_GOOD(-0.10)[text/plain]; BROKEN_CONTENT_TYPE(1.50)[]; R_SPF_SOFTFAIL(0.00)[~all:c]; ML_SERVERS(-3.10)[148.251.23.173]; DKIM_TRACE(0.00)[flygoat.com:+]; RCPT_COUNT_TWO(0.00)[2]; MID_CONTAINS_FROM(1.00)[]; DMARC_POLICY_ALLOW(0.00)[flygoat.com,none]; DMARC_POLICY_ALLOW_WITH_FAILURES(0.00)[]; RCVD_NO_TLS_LAST(0.10)[]; FROM_EQ_ENVFROM(0.00)[]; MIME_TRACE(0.00)[0:+]; ASN(0.00)[asn:24940, ipnet:148.251.0.0/16, country:DE]; RCVD_COUNT_TWO(0.00)[2]; HFILTER_HELO_BAREIP(3.00)[148.251.23.173,1] X-Rspamd-Server: mail20.mymailcheap.com X-Spam: Yes Subject: [FFmpeg-devel] [PATCH v5 4/6] libavcodec: Enable runtime detection for MIPS MMI & MSA 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: Jiaxun Yang Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: spnyphfnkA6S Content-Length: 257999 Apply optimized functions according to cpuflags. MSA is usually put after MMI as it's generally faster than MMI. Signed-off-by: Jiaxun Yang --- v5: Refine if, also make h264chroma as a exception that MMI faster than MSA --- libavcodec/mips/blockdsp_init_mips.c | 40 +- libavcodec/mips/cabac.h | 2 +- libavcodec/mips/h263dsp_init_mips.c | 18 +- libavcodec/mips/h264chroma_init_mips.c | 55 +- libavcodec/mips/h264dsp_init_mips.c | 225 +++-- libavcodec/mips/h264pred_init_mips.c | 207 ++-- libavcodec/mips/h264qpel_init_mips.c | 412 ++++---- libavcodec/mips/hevcdsp_init_mips.c | 992 ++++++++++---------- libavcodec/mips/hevcpred_init_mips.c | 40 +- libavcodec/mips/hpeldsp_init_mips.c | 180 ++-- libavcodec/mips/idctdsp_init_mips.c | 74 +- libavcodec/mips/me_cmp_init_mips.c | 50 +- libavcodec/mips/mpegvideo_init_mips.c | 48 +- libavcodec/mips/mpegvideoencdsp_init_mips.c | 21 +- libavcodec/mips/pixblockdsp_init_mips.c | 63 +- libavcodec/mips/qpeldsp_init_mips.c | 270 +++--- libavcodec/mips/vc1dsp_init_mips.c | 186 ++-- libavcodec/mips/videodsp_init.c | 10 +- libavcodec/mips/vp3dsp_init_mips.c | 44 +- libavcodec/mips/vp8dsp_init_mips.c | 240 +++-- libavcodec/mips/vp9dsp_init_mips.c | 16 +- libavcodec/mips/wmv2dsp_init_mips.c | 18 +- libavcodec/mips/wmv2dsp_mips.h | 4 +- libavcodec/mips/wmv2dsp_mmi.c | 4 +- libavcodec/mips/xvididct_init_mips.c | 31 +- 25 files changed, 1543 insertions(+), 1707 deletions(-) diff --git a/libavcodec/mips/blockdsp_init_mips.c b/libavcodec/mips/blockdsp_init_mips.c index 55ac1c3e99..c6964fa74e 100644 --- a/libavcodec/mips/blockdsp_init_mips.c +++ b/libavcodec/mips/blockdsp_init_mips.c @@ -19,36 +19,26 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "blockdsp_mips.h" -#if HAVE_MSA -static av_cold void blockdsp_init_msa(BlockDSPContext *c) +void ff_blockdsp_init_mips(BlockDSPContext *c) { - c->clear_block = ff_clear_block_msa; - c->clear_blocks = ff_clear_blocks_msa; + int cpu_flags = av_get_cpu_flags(); - c->fill_block_tab[0] = ff_fill_block16_msa; - c->fill_block_tab[1] = ff_fill_block8_msa; -} -#endif // #if HAVE_MSA + if (have_mmi(cpu_flags)) { + c->clear_block = ff_clear_block_mmi; + c->clear_blocks = ff_clear_blocks_mmi; -#if HAVE_MMI -static av_cold void blockdsp_init_mmi(BlockDSPContext *c) -{ - c->clear_block = ff_clear_block_mmi; - c->clear_blocks = ff_clear_blocks_mmi; + c->fill_block_tab[0] = ff_fill_block16_mmi; + c->fill_block_tab[1] = ff_fill_block8_mmi; + } - c->fill_block_tab[0] = ff_fill_block16_mmi; - c->fill_block_tab[1] = ff_fill_block8_mmi; -} -#endif /* HAVE_MMI */ + if (have_msa(cpu_flags)) { + c->clear_block = ff_clear_block_msa; + c->clear_blocks = ff_clear_blocks_msa; -void ff_blockdsp_init_mips(BlockDSPContext *c) -{ -#if HAVE_MMI - blockdsp_init_mmi(c); -#endif /* HAVE_MMI */ -#if HAVE_MSA - blockdsp_init_msa(c); -#endif // #if HAVE_MSA + c->fill_block_tab[0] = ff_fill_block16_msa; + c->fill_block_tab[1] = ff_fill_block8_msa; + } } diff --git a/libavcodec/mips/cabac.h b/libavcodec/mips/cabac.h index 03b5010edc..c595915eda 100644 --- a/libavcodec/mips/cabac.h +++ b/libavcodec/mips/cabac.h @@ -25,7 +25,7 @@ #define AVCODEC_MIPS_CABAC_H #include "libavcodec/cabac.h" -#include "libavutil/mips/mmiutils.h" +#include "libavutil/mips/asmdefs.h" #include "config.h" #define get_cabac_inline get_cabac_inline_mips diff --git a/libavcodec/mips/h263dsp_init_mips.c b/libavcodec/mips/h263dsp_init_mips.c index 09bd93707d..a73eb12d87 100644 --- a/libavcodec/mips/h263dsp_init_mips.c +++ b/libavcodec/mips/h263dsp_init_mips.c @@ -18,19 +18,15 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "h263dsp_mips.h" -#if HAVE_MSA -static av_cold void h263dsp_init_msa(H263DSPContext *c) -{ - c->h263_h_loop_filter = ff_h263_h_loop_filter_msa; - c->h263_v_loop_filter = ff_h263_v_loop_filter_msa; -} -#endif // #if HAVE_MSA - av_cold void ff_h263dsp_init_mips(H263DSPContext *c) { -#if HAVE_MSA - h263dsp_init_msa(c); -#endif // #if HAVE_MSA + int cpu_flags = av_get_cpu_flags(); + + if (have_msa(cpu_flags)){ + c->h263_h_loop_filter = ff_h263_h_loop_filter_msa; + c->h263_v_loop_filter = ff_h263_v_loop_filter_msa; + } } diff --git a/libavcodec/mips/h264chroma_init_mips.c b/libavcodec/mips/h264chroma_init_mips.c index ae817e47ae..6bb19d3ddd 100644 --- a/libavcodec/mips/h264chroma_init_mips.c +++ b/libavcodec/mips/h264chroma_init_mips.c @@ -19,45 +19,34 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "h264chroma_mips.h" -#if HAVE_MSA -static av_cold void h264chroma_init_msa(H264ChromaContext *c, int bit_depth) -{ - const int high_bit_depth = bit_depth > 8; - - if (!high_bit_depth) { - c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_msa; - c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_msa; - c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_msa; - - c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_msa; - c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_msa; - c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_msa; - } -} -#endif // #if HAVE_MSA -#if HAVE_MMI -static av_cold void h264chroma_init_mmi(H264ChromaContext *c, int bit_depth) +av_cold void ff_h264chroma_init_mips(H264ChromaContext *c, int bit_depth) { + int cpu_flags = av_get_cpu_flags(); int high_bit_depth = bit_depth > 8; - if (!high_bit_depth) { - c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_mmi; - c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_mmi; - c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_mmi; - c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_mmi; + /* MMI apears to be faster than MSA here */ + if (have_msa(cpu_flags)) { + if (!high_bit_depth) { + c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_msa; + c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_msa; + c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_msa; + + c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_msa; + c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_msa; + c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_msa; + } } -} -#endif /* HAVE_MMI */ -av_cold void ff_h264chroma_init_mips(H264ChromaContext *c, int bit_depth) -{ -#if HAVE_MMI - h264chroma_init_mmi(c, bit_depth); -#endif /* HAVE_MMI */ -#if HAVE_MSA - h264chroma_init_msa(c, bit_depth); -#endif // #if HAVE_MSA + if (have_mmi(cpu_flags)) { + if (!high_bit_depth) { + c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_mmi; + c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_mmi; + c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_mmi; + c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_mmi; + } + } } diff --git a/libavcodec/mips/h264dsp_init_mips.c b/libavcodec/mips/h264dsp_init_mips.c index dc08a25800..9cd05e0f2f 100644 --- a/libavcodec/mips/h264dsp_init_mips.c +++ b/libavcodec/mips/h264dsp_init_mips.c @@ -19,129 +19,116 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "h264dsp_mips.h" -#if HAVE_MSA -static av_cold void h264dsp_init_msa(H264DSPContext *c, - const int bit_depth, - const int chroma_format_idc) -{ - if (8 == bit_depth) { - c->h264_v_loop_filter_luma = ff_h264_v_lpf_luma_inter_msa; - c->h264_h_loop_filter_luma = ff_h264_h_lpf_luma_inter_msa; - c->h264_h_loop_filter_luma_mbaff = - ff_h264_h_loop_filter_luma_mbaff_msa; - c->h264_v_loop_filter_luma_intra = ff_h264_v_lpf_luma_intra_msa; - c->h264_h_loop_filter_luma_intra = ff_h264_h_lpf_luma_intra_msa; - c->h264_h_loop_filter_luma_mbaff_intra = - ff_h264_h_loop_filter_luma_mbaff_intra_msa; - c->h264_v_loop_filter_chroma = ff_h264_v_lpf_chroma_inter_msa; - - if (chroma_format_idc <= 1) - c->h264_h_loop_filter_chroma = ff_h264_h_lpf_chroma_inter_msa; - else - c->h264_h_loop_filter_chroma = - ff_h264_h_loop_filter_chroma422_msa; - - if (chroma_format_idc > 1) - c->h264_h_loop_filter_chroma_mbaff = - ff_h264_h_loop_filter_chroma422_mbaff_msa; - - c->h264_v_loop_filter_chroma_intra = - ff_h264_v_lpf_chroma_intra_msa; - - if (chroma_format_idc <= 1) - c->h264_h_loop_filter_chroma_intra = - ff_h264_h_lpf_chroma_intra_msa; - - /* Weighted MC */ - c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16_8_msa; - c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels8_8_msa; - c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels4_8_msa; - - c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16_8_msa; - c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_8_msa; - c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels4_8_msa; - - c->h264_idct_add = ff_h264_idct_add_msa; - c->h264_idct8_add = ff_h264_idct8_addblk_msa; - c->h264_idct_dc_add = ff_h264_idct4x4_addblk_dc_msa; - c->h264_idct8_dc_add = ff_h264_idct8_dc_addblk_msa; - c->h264_idct_add16 = ff_h264_idct_add16_msa; - c->h264_idct8_add4 = ff_h264_idct8_add4_msa; - - if (chroma_format_idc <= 1) - c->h264_idct_add8 = ff_h264_idct_add8_msa; - else - c->h264_idct_add8 = ff_h264_idct_add8_422_msa; - - c->h264_idct_add16intra = ff_h264_idct_add16_intra_msa; - c->h264_luma_dc_dequant_idct = ff_h264_deq_idct_luma_dc_msa; - } // if (8 == bit_depth) -} -#endif // #if HAVE_MSA - -#if HAVE_MMI -static av_cold void h264dsp_init_mmi(H264DSPContext * c, const int bit_depth, - const int chroma_format_idc) +av_cold void ff_h264dsp_init_mips(H264DSPContext *c, const int bit_depth, + const int chroma_format_idc) { - if (bit_depth == 8) { - c->h264_add_pixels4_clear = ff_h264_add_pixels4_8_mmi; - c->h264_idct_add = ff_h264_idct_add_8_mmi; - c->h264_idct8_add = ff_h264_idct8_add_8_mmi; - c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmi; - c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmi; - c->h264_idct_add16 = ff_h264_idct_add16_8_mmi; - c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmi; - c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmi; - - if (chroma_format_idc <= 1) - c->h264_idct_add8 = ff_h264_idct_add8_8_mmi; - else - c->h264_idct_add8 = ff_h264_idct_add8_422_8_mmi; - - c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_8_mmi; - - if (chroma_format_idc <= 1) - c->h264_chroma_dc_dequant_idct = - ff_h264_chroma_dc_dequant_idct_8_mmi; - else - c->h264_chroma_dc_dequant_idct = - ff_h264_chroma422_dc_dequant_idct_8_mmi; - - c->weight_h264_pixels_tab[0] = ff_h264_weight_pixels16_8_mmi; - c->weight_h264_pixels_tab[1] = ff_h264_weight_pixels8_8_mmi; - c->weight_h264_pixels_tab[2] = ff_h264_weight_pixels4_8_mmi; - - c->biweight_h264_pixels_tab[0] = ff_h264_biweight_pixels16_8_mmi; - c->biweight_h264_pixels_tab[1] = ff_h264_biweight_pixels8_8_mmi; - c->biweight_h264_pixels_tab[2] = ff_h264_biweight_pixels4_8_mmi; - - c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_8_mmi; - c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmi; - - if (chroma_format_idc <= 1) { - c->h264_h_loop_filter_chroma = - ff_deblock_h_chroma_8_mmi; - c->h264_h_loop_filter_chroma_intra = - ff_deblock_h_chroma_intra_8_mmi; + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + if (bit_depth == 8) { + c->h264_add_pixels4_clear = ff_h264_add_pixels4_8_mmi; + c->h264_idct_add = ff_h264_idct_add_8_mmi; + c->h264_idct8_add = ff_h264_idct8_add_8_mmi; + c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmi; + c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmi; + c->h264_idct_add16 = ff_h264_idct_add16_8_mmi; + c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmi; + c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmi; + + if (chroma_format_idc <= 1) + c->h264_idct_add8 = ff_h264_idct_add8_8_mmi; + else + c->h264_idct_add8 = ff_h264_idct_add8_422_8_mmi; + + c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_8_mmi; + + if (chroma_format_idc <= 1) + c->h264_chroma_dc_dequant_idct = + ff_h264_chroma_dc_dequant_idct_8_mmi; + else + c->h264_chroma_dc_dequant_idct = + ff_h264_chroma422_dc_dequant_idct_8_mmi; + + c->weight_h264_pixels_tab[0] = ff_h264_weight_pixels16_8_mmi; + c->weight_h264_pixels_tab[1] = ff_h264_weight_pixels8_8_mmi; + c->weight_h264_pixels_tab[2] = ff_h264_weight_pixels4_8_mmi; + + c->biweight_h264_pixels_tab[0] = ff_h264_biweight_pixels16_8_mmi; + c->biweight_h264_pixels_tab[1] = ff_h264_biweight_pixels8_8_mmi; + c->biweight_h264_pixels_tab[2] = ff_h264_biweight_pixels4_8_mmi; + + c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_8_mmi; + c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmi; + + if (chroma_format_idc <= 1) { + c->h264_h_loop_filter_chroma = + ff_deblock_h_chroma_8_mmi; + c->h264_h_loop_filter_chroma_intra = + ff_deblock_h_chroma_intra_8_mmi; + } + + c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_mmi; + c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmi; + c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_mmi; + c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmi; } - - c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_mmi; - c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmi; - c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_mmi; - c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmi; } -} -#endif /* HAVE_MMI */ -av_cold void ff_h264dsp_init_mips(H264DSPContext *c, const int bit_depth, - const int chroma_format_idc) -{ -#if HAVE_MMI - h264dsp_init_mmi(c, bit_depth, chroma_format_idc); -#endif /* HAVE_MMI */ -#if HAVE_MSA - h264dsp_init_msa(c, bit_depth, chroma_format_idc); -#endif // #if HAVE_MSA + if (have_msa(cpu_flags)) { + if (bit_depth == 8) { + c->h264_v_loop_filter_luma = ff_h264_v_lpf_luma_inter_msa; + c->h264_h_loop_filter_luma = ff_h264_h_lpf_luma_inter_msa; + c->h264_h_loop_filter_luma_mbaff = + ff_h264_h_loop_filter_luma_mbaff_msa; + c->h264_v_loop_filter_luma_intra = ff_h264_v_lpf_luma_intra_msa; + c->h264_h_loop_filter_luma_intra = ff_h264_h_lpf_luma_intra_msa; + c->h264_h_loop_filter_luma_mbaff_intra = + ff_h264_h_loop_filter_luma_mbaff_intra_msa; + c->h264_v_loop_filter_chroma = ff_h264_v_lpf_chroma_inter_msa; + + if (chroma_format_idc <= 1) + c->h264_h_loop_filter_chroma = ff_h264_h_lpf_chroma_inter_msa; + else + c->h264_h_loop_filter_chroma = + ff_h264_h_loop_filter_chroma422_msa; + + if (chroma_format_idc > 1) + c->h264_h_loop_filter_chroma_mbaff = + ff_h264_h_loop_filter_chroma422_mbaff_msa; + + c->h264_v_loop_filter_chroma_intra = + ff_h264_v_lpf_chroma_intra_msa; + + if (chroma_format_idc <= 1) + c->h264_h_loop_filter_chroma_intra = + ff_h264_h_lpf_chroma_intra_msa; + + /* Weighted MC */ + c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16_8_msa; + c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels8_8_msa; + c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels4_8_msa; + + c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16_8_msa; + c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_8_msa; + c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels4_8_msa; + + c->h264_idct_add = ff_h264_idct_add_msa; + c->h264_idct8_add = ff_h264_idct8_addblk_msa; + c->h264_idct_dc_add = ff_h264_idct4x4_addblk_dc_msa; + c->h264_idct8_dc_add = ff_h264_idct8_dc_addblk_msa; + c->h264_idct_add16 = ff_h264_idct_add16_msa; + c->h264_idct8_add4 = ff_h264_idct8_add4_msa; + + if (chroma_format_idc <= 1) + c->h264_idct_add8 = ff_h264_idct_add8_msa; + else + c->h264_idct_add8 = ff_h264_idct_add8_422_msa; + + c->h264_idct_add16intra = ff_h264_idct_add16_intra_msa; + c->h264_luma_dc_dequant_idct = ff_h264_deq_idct_luma_dc_msa; + } + } } diff --git a/libavcodec/mips/h264pred_init_mips.c b/libavcodec/mips/h264pred_init_mips.c index e537ad8bd4..0fd9bb737a 100644 --- a/libavcodec/mips/h264pred_init_mips.c +++ b/libavcodec/mips/h264pred_init_mips.c @@ -19,134 +19,121 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "config.h" #include "h264dsp_mips.h" #include "h264pred_mips.h" -#if HAVE_MSA -static av_cold void h264_pred_init_msa(H264PredContext *h, int codec_id, - const int bit_depth, - const int chroma_format_idc) +av_cold void ff_h264_pred_init_mips(H264PredContext *h, int codec_id, + int bit_depth, + const int chroma_format_idc) { - if (8 == bit_depth) { - if (chroma_format_idc == 1) { - h->pred8x8[VERT_PRED8x8] = ff_h264_intra_pred_vert_8x8_msa; - h->pred8x8[HOR_PRED8x8] = ff_h264_intra_pred_horiz_8x8_msa; - } + int cpu_flags = av_get_cpu_flags(); - if (codec_id != AV_CODEC_ID_VP7 && codec_id != AV_CODEC_ID_VP8) { + if (have_mmi(cpu_flags)) { + if (bit_depth == 8) { if (chroma_format_idc == 1) { - h->pred8x8[PLANE_PRED8x8] = ff_h264_intra_predict_plane_8x8_msa; - } - } - if (codec_id != AV_CODEC_ID_RV40 && codec_id != AV_CODEC_ID_VP7 - && codec_id != AV_CODEC_ID_VP8) { - if (chroma_format_idc == 1) { - h->pred8x8[DC_PRED8x8] = ff_h264_intra_predict_dc_4blk_8x8_msa; - h->pred8x8[LEFT_DC_PRED8x8] = - ff_h264_intra_predict_hor_dc_8x8_msa; - h->pred8x8[TOP_DC_PRED8x8] = - ff_h264_intra_predict_vert_dc_8x8_msa; - h->pred8x8[ALZHEIMER_DC_L0T_PRED8x8] = - ff_h264_intra_predict_mad_cow_dc_l0t_8x8_msa; - h->pred8x8[ALZHEIMER_DC_0LT_PRED8x8] = - ff_h264_intra_predict_mad_cow_dc_0lt_8x8_msa; - h->pred8x8[ALZHEIMER_DC_L00_PRED8x8] = - ff_h264_intra_predict_mad_cow_dc_l00_8x8_msa; - h->pred8x8[ALZHEIMER_DC_0L0_PRED8x8] = - ff_h264_intra_predict_mad_cow_dc_0l0_8x8_msa; - } - } else { - if (codec_id == AV_CODEC_ID_VP7 || codec_id == AV_CODEC_ID_VP8) { - h->pred8x8[7] = ff_vp8_pred8x8_127_dc_8_msa; - h->pred8x8[8] = ff_vp8_pred8x8_129_dc_8_msa; + h->pred8x8 [VERT_PRED8x8 ] = ff_pred8x8_vertical_8_mmi; + h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_8_mmi; + } else { + h->pred8x8 [VERT_PRED8x8 ] = ff_pred8x16_vertical_8_mmi; + h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x16_horizontal_8_mmi; } - } - if (chroma_format_idc == 1) { - h->pred8x8[DC_128_PRED8x8] = ff_h264_intra_pred_dc_128_8x8_msa; - } + h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_8_mmi; + h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vertical_8_mmi; + h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_8_mmi; + h->pred8x8l [TOP_DC_PRED ] = ff_pred8x8l_top_dc_8_mmi; + h->pred8x8l [DC_PRED ] = ff_pred8x8l_dc_8_mmi; - h->pred16x16[DC_PRED8x8] = ff_h264_intra_pred_dc_16x16_msa; - h->pred16x16[VERT_PRED8x8] = ff_h264_intra_pred_vert_16x16_msa; - h->pred16x16[HOR_PRED8x8] = ff_h264_intra_pred_horiz_16x16_msa; + #if ARCH_MIPS64 + switch (codec_id) { + case AV_CODEC_ID_SVQ3: + h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_svq3_8_mmi; + break; + case AV_CODEC_ID_RV40: + h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_rv40_8_mmi; + break; + case AV_CODEC_ID_VP7: + case AV_CODEC_ID_VP8: + break; + default: + h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_h264_8_mmi; + break; + } + #endif - switch (codec_id) { - case AV_CODEC_ID_SVQ3: - case AV_CODEC_ID_RV40: - break; - case AV_CODEC_ID_VP7: - case AV_CODEC_ID_VP8: - h->pred16x16[7] = ff_vp8_pred16x16_127_dc_8_msa; - h->pred16x16[8] = ff_vp8_pred16x16_129_dc_8_msa; - break; - default: - h->pred16x16[PLANE_PRED8x8] = - ff_h264_intra_predict_plane_16x16_msa; - break; + if (codec_id == AV_CODEC_ID_SVQ3 || codec_id == AV_CODEC_ID_H264) { + if (chroma_format_idc == 1) { + h->pred8x8[TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_8_mmi; + h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_8_mmi; + } + } } - - h->pred16x16[LEFT_DC_PRED8x8] = ff_h264_intra_pred_dc_left_16x16_msa; - h->pred16x16[TOP_DC_PRED8x8] = ff_h264_intra_pred_dc_top_16x16_msa; - h->pred16x16[DC_128_PRED8x8] = ff_h264_intra_pred_dc_128_16x16_msa; } -} -#endif // #if HAVE_MSA - -#if HAVE_MMI -static av_cold void h264_pred_init_mmi(H264PredContext *h, int codec_id, - const int bit_depth, const int chroma_format_idc) -{ - if (bit_depth == 8) { - if (chroma_format_idc == 1) { - h->pred8x8 [VERT_PRED8x8 ] = ff_pred8x8_vertical_8_mmi; - h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_8_mmi; - } else { - h->pred8x8 [VERT_PRED8x8 ] = ff_pred8x16_vertical_8_mmi; - h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x16_horizontal_8_mmi; - } - h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_8_mmi; - h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vertical_8_mmi; - h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_8_mmi; - h->pred8x8l [TOP_DC_PRED ] = ff_pred8x8l_top_dc_8_mmi; - h->pred8x8l [DC_PRED ] = ff_pred8x8l_dc_8_mmi; + if (have_msa(cpu_flags)) { + if (8 == bit_depth) { + if (chroma_format_idc == 1) { + h->pred8x8[VERT_PRED8x8] = ff_h264_intra_pred_vert_8x8_msa; + h->pred8x8[HOR_PRED8x8] = ff_h264_intra_pred_horiz_8x8_msa; + } -#if ARCH_MIPS64 - switch (codec_id) { - case AV_CODEC_ID_SVQ3: - h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_svq3_8_mmi; - break; - case AV_CODEC_ID_RV40: - h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_rv40_8_mmi; - break; - case AV_CODEC_ID_VP7: - case AV_CODEC_ID_VP8: - break; - default: - h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_h264_8_mmi; - break; - } -#endif + if (codec_id != AV_CODEC_ID_VP7 && codec_id != AV_CODEC_ID_VP8) { + if (chroma_format_idc == 1) { + h->pred8x8[PLANE_PRED8x8] = ff_h264_intra_predict_plane_8x8_msa; + } + } + if (codec_id != AV_CODEC_ID_RV40 && codec_id != AV_CODEC_ID_VP7 + && codec_id != AV_CODEC_ID_VP8) { + if (chroma_format_idc == 1) { + h->pred8x8[DC_PRED8x8] = ff_h264_intra_predict_dc_4blk_8x8_msa; + h->pred8x8[LEFT_DC_PRED8x8] = + ff_h264_intra_predict_hor_dc_8x8_msa; + h->pred8x8[TOP_DC_PRED8x8] = + ff_h264_intra_predict_vert_dc_8x8_msa; + h->pred8x8[ALZHEIMER_DC_L0T_PRED8x8] = + ff_h264_intra_predict_mad_cow_dc_l0t_8x8_msa; + h->pred8x8[ALZHEIMER_DC_0LT_PRED8x8] = + ff_h264_intra_predict_mad_cow_dc_0lt_8x8_msa; + h->pred8x8[ALZHEIMER_DC_L00_PRED8x8] = + ff_h264_intra_predict_mad_cow_dc_l00_8x8_msa; + h->pred8x8[ALZHEIMER_DC_0L0_PRED8x8] = + ff_h264_intra_predict_mad_cow_dc_0l0_8x8_msa; + } + } else { + if (codec_id == AV_CODEC_ID_VP7 || codec_id == AV_CODEC_ID_VP8) { + h->pred8x8[7] = ff_vp8_pred8x8_127_dc_8_msa; + h->pred8x8[8] = ff_vp8_pred8x8_129_dc_8_msa; + } + } - if (codec_id == AV_CODEC_ID_SVQ3 || codec_id == AV_CODEC_ID_H264) { if (chroma_format_idc == 1) { - h->pred8x8[TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_8_mmi; - h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_8_mmi; + h->pred8x8[DC_128_PRED8x8] = ff_h264_intra_pred_dc_128_8x8_msa; + } + + h->pred16x16[DC_PRED8x8] = ff_h264_intra_pred_dc_16x16_msa; + h->pred16x16[VERT_PRED8x8] = ff_h264_intra_pred_vert_16x16_msa; + h->pred16x16[HOR_PRED8x8] = ff_h264_intra_pred_horiz_16x16_msa; + + switch (codec_id) { + case AV_CODEC_ID_SVQ3: + case AV_CODEC_ID_RV40: + break; + case AV_CODEC_ID_VP7: + case AV_CODEC_ID_VP8: + h->pred16x16[7] = ff_vp8_pred16x16_127_dc_8_msa; + h->pred16x16[8] = ff_vp8_pred16x16_129_dc_8_msa; + break; + default: + h->pred16x16[PLANE_PRED8x8] = + ff_h264_intra_predict_plane_16x16_msa; + break; } + + h->pred16x16[LEFT_DC_PRED8x8] = ff_h264_intra_pred_dc_left_16x16_msa; + h->pred16x16[TOP_DC_PRED8x8] = ff_h264_intra_pred_dc_top_16x16_msa; + h->pred16x16[DC_128_PRED8x8] = ff_h264_intra_pred_dc_128_16x16_msa; } } } -#endif /* HAVE_MMI */ - -av_cold void ff_h264_pred_init_mips(H264PredContext *h, int codec_id, - int bit_depth, - const int chroma_format_idc) -{ -#if HAVE_MMI - h264_pred_init_mmi(h, codec_id, bit_depth, chroma_format_idc); -#endif /* HAVE_MMI */ -#if HAVE_MSA - h264_pred_init_msa(h, codec_id, bit_depth, chroma_format_idc); -#endif // #if HAVE_MSA -} diff --git a/libavcodec/mips/h264qpel_init_mips.c b/libavcodec/mips/h264qpel_init_mips.c index 33bae3093a..d896a4841b 100644 --- a/libavcodec/mips/h264qpel_init_mips.c +++ b/libavcodec/mips/h264qpel_init_mips.c @@ -19,231 +19,221 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "h264dsp_mips.h" -#if HAVE_MSA -static av_cold void h264qpel_init_msa(H264QpelContext *c, int bit_depth) +av_cold void ff_h264qpel_init_mips(H264QpelContext *c, int bit_depth) { - if (8 == bit_depth) { - c->put_h264_qpel_pixels_tab[0][0] = ff_put_h264_qpel16_mc00_msa; - c->put_h264_qpel_pixels_tab[0][1] = ff_put_h264_qpel16_mc10_msa; - c->put_h264_qpel_pixels_tab[0][2] = ff_put_h264_qpel16_mc20_msa; - c->put_h264_qpel_pixels_tab[0][3] = ff_put_h264_qpel16_mc30_msa; - c->put_h264_qpel_pixels_tab[0][4] = ff_put_h264_qpel16_mc01_msa; - c->put_h264_qpel_pixels_tab[0][5] = ff_put_h264_qpel16_mc11_msa; - c->put_h264_qpel_pixels_tab[0][6] = ff_put_h264_qpel16_mc21_msa; - c->put_h264_qpel_pixels_tab[0][7] = ff_put_h264_qpel16_mc31_msa; - c->put_h264_qpel_pixels_tab[0][8] = ff_put_h264_qpel16_mc02_msa; - c->put_h264_qpel_pixels_tab[0][9] = ff_put_h264_qpel16_mc12_msa; - c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_msa; - c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_msa; - c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_msa; - c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_msa; - c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_msa; - c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_msa; + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + if (bit_depth == 8) { + c->put_h264_qpel_pixels_tab[0][0] = ff_put_h264_qpel16_mc00_mmi; + c->put_h264_qpel_pixels_tab[0][1] = ff_put_h264_qpel16_mc10_mmi; + c->put_h264_qpel_pixels_tab[0][2] = ff_put_h264_qpel16_mc20_mmi; + c->put_h264_qpel_pixels_tab[0][3] = ff_put_h264_qpel16_mc30_mmi; + c->put_h264_qpel_pixels_tab[0][4] = ff_put_h264_qpel16_mc01_mmi; + c->put_h264_qpel_pixels_tab[0][5] = ff_put_h264_qpel16_mc11_mmi; + c->put_h264_qpel_pixels_tab[0][6] = ff_put_h264_qpel16_mc21_mmi; + c->put_h264_qpel_pixels_tab[0][7] = ff_put_h264_qpel16_mc31_mmi; + c->put_h264_qpel_pixels_tab[0][8] = ff_put_h264_qpel16_mc02_mmi; + c->put_h264_qpel_pixels_tab[0][9] = ff_put_h264_qpel16_mc12_mmi; + c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_mmi; + c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_mmi; + c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_mmi; + c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_mmi; + c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_mmi; + c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_mmi; - c->put_h264_qpel_pixels_tab[1][0] = ff_put_h264_qpel8_mc00_msa; - c->put_h264_qpel_pixels_tab[1][1] = ff_put_h264_qpel8_mc10_msa; - c->put_h264_qpel_pixels_tab[1][2] = ff_put_h264_qpel8_mc20_msa; - c->put_h264_qpel_pixels_tab[1][3] = ff_put_h264_qpel8_mc30_msa; - c->put_h264_qpel_pixels_tab[1][4] = ff_put_h264_qpel8_mc01_msa; - c->put_h264_qpel_pixels_tab[1][5] = ff_put_h264_qpel8_mc11_msa; - c->put_h264_qpel_pixels_tab[1][6] = ff_put_h264_qpel8_mc21_msa; - c->put_h264_qpel_pixels_tab[1][7] = ff_put_h264_qpel8_mc31_msa; - c->put_h264_qpel_pixels_tab[1][8] = ff_put_h264_qpel8_mc02_msa; - c->put_h264_qpel_pixels_tab[1][9] = ff_put_h264_qpel8_mc12_msa; - c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_msa; - c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_msa; - c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_msa; - c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_msa; - c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_msa; - c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_msa; + c->put_h264_qpel_pixels_tab[1][0] = ff_put_h264_qpel8_mc00_mmi; + c->put_h264_qpel_pixels_tab[1][1] = ff_put_h264_qpel8_mc10_mmi; + c->put_h264_qpel_pixels_tab[1][2] = ff_put_h264_qpel8_mc20_mmi; + c->put_h264_qpel_pixels_tab[1][3] = ff_put_h264_qpel8_mc30_mmi; + c->put_h264_qpel_pixels_tab[1][4] = ff_put_h264_qpel8_mc01_mmi; + c->put_h264_qpel_pixels_tab[1][5] = ff_put_h264_qpel8_mc11_mmi; + c->put_h264_qpel_pixels_tab[1][6] = ff_put_h264_qpel8_mc21_mmi; + c->put_h264_qpel_pixels_tab[1][7] = ff_put_h264_qpel8_mc31_mmi; + c->put_h264_qpel_pixels_tab[1][8] = ff_put_h264_qpel8_mc02_mmi; + c->put_h264_qpel_pixels_tab[1][9] = ff_put_h264_qpel8_mc12_mmi; + c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_mmi; + c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_mmi; + c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_mmi; + c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_mmi; + c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_mmi; + c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_mmi; - c->put_h264_qpel_pixels_tab[2][1] = ff_put_h264_qpel4_mc10_msa; - c->put_h264_qpel_pixels_tab[2][2] = ff_put_h264_qpel4_mc20_msa; - c->put_h264_qpel_pixels_tab[2][3] = ff_put_h264_qpel4_mc30_msa; - c->put_h264_qpel_pixels_tab[2][4] = ff_put_h264_qpel4_mc01_msa; - c->put_h264_qpel_pixels_tab[2][5] = ff_put_h264_qpel4_mc11_msa; - c->put_h264_qpel_pixels_tab[2][6] = ff_put_h264_qpel4_mc21_msa; - c->put_h264_qpel_pixels_tab[2][7] = ff_put_h264_qpel4_mc31_msa; - c->put_h264_qpel_pixels_tab[2][8] = ff_put_h264_qpel4_mc02_msa; - c->put_h264_qpel_pixels_tab[2][9] = ff_put_h264_qpel4_mc12_msa; - c->put_h264_qpel_pixels_tab[2][10] = ff_put_h264_qpel4_mc22_msa; - c->put_h264_qpel_pixels_tab[2][11] = ff_put_h264_qpel4_mc32_msa; - c->put_h264_qpel_pixels_tab[2][12] = ff_put_h264_qpel4_mc03_msa; - c->put_h264_qpel_pixels_tab[2][13] = ff_put_h264_qpel4_mc13_msa; - c->put_h264_qpel_pixels_tab[2][14] = ff_put_h264_qpel4_mc23_msa; - c->put_h264_qpel_pixels_tab[2][15] = ff_put_h264_qpel4_mc33_msa; + c->put_h264_qpel_pixels_tab[2][0] = ff_put_h264_qpel4_mc00_mmi; + c->put_h264_qpel_pixels_tab[2][1] = ff_put_h264_qpel4_mc10_mmi; + c->put_h264_qpel_pixels_tab[2][2] = ff_put_h264_qpel4_mc20_mmi; + c->put_h264_qpel_pixels_tab[2][3] = ff_put_h264_qpel4_mc30_mmi; + c->put_h264_qpel_pixels_tab[2][4] = ff_put_h264_qpel4_mc01_mmi; + c->put_h264_qpel_pixels_tab[2][5] = ff_put_h264_qpel4_mc11_mmi; + c->put_h264_qpel_pixels_tab[2][6] = ff_put_h264_qpel4_mc21_mmi; + c->put_h264_qpel_pixels_tab[2][7] = ff_put_h264_qpel4_mc31_mmi; + c->put_h264_qpel_pixels_tab[2][8] = ff_put_h264_qpel4_mc02_mmi; + c->put_h264_qpel_pixels_tab[2][9] = ff_put_h264_qpel4_mc12_mmi; + c->put_h264_qpel_pixels_tab[2][10] = ff_put_h264_qpel4_mc22_mmi; + c->put_h264_qpel_pixels_tab[2][11] = ff_put_h264_qpel4_mc32_mmi; + c->put_h264_qpel_pixels_tab[2][12] = ff_put_h264_qpel4_mc03_mmi; + c->put_h264_qpel_pixels_tab[2][13] = ff_put_h264_qpel4_mc13_mmi; + c->put_h264_qpel_pixels_tab[2][14] = ff_put_h264_qpel4_mc23_mmi; + c->put_h264_qpel_pixels_tab[2][15] = ff_put_h264_qpel4_mc33_mmi; - c->avg_h264_qpel_pixels_tab[0][0] = ff_avg_h264_qpel16_mc00_msa; - c->avg_h264_qpel_pixels_tab[0][1] = ff_avg_h264_qpel16_mc10_msa; - c->avg_h264_qpel_pixels_tab[0][2] = ff_avg_h264_qpel16_mc20_msa; - c->avg_h264_qpel_pixels_tab[0][3] = ff_avg_h264_qpel16_mc30_msa; - c->avg_h264_qpel_pixels_tab[0][4] = ff_avg_h264_qpel16_mc01_msa; - c->avg_h264_qpel_pixels_tab[0][5] = ff_avg_h264_qpel16_mc11_msa; - c->avg_h264_qpel_pixels_tab[0][6] = ff_avg_h264_qpel16_mc21_msa; - c->avg_h264_qpel_pixels_tab[0][7] = ff_avg_h264_qpel16_mc31_msa; - c->avg_h264_qpel_pixels_tab[0][8] = ff_avg_h264_qpel16_mc02_msa; - c->avg_h264_qpel_pixels_tab[0][9] = ff_avg_h264_qpel16_mc12_msa; - c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_msa; - c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_msa; - c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_msa; - c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_msa; - c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_msa; - c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_msa; + c->avg_h264_qpel_pixels_tab[0][0] = ff_avg_h264_qpel16_mc00_mmi; + c->avg_h264_qpel_pixels_tab[0][1] = ff_avg_h264_qpel16_mc10_mmi; + c->avg_h264_qpel_pixels_tab[0][2] = ff_avg_h264_qpel16_mc20_mmi; + c->avg_h264_qpel_pixels_tab[0][3] = ff_avg_h264_qpel16_mc30_mmi; + c->avg_h264_qpel_pixels_tab[0][4] = ff_avg_h264_qpel16_mc01_mmi; + c->avg_h264_qpel_pixels_tab[0][5] = ff_avg_h264_qpel16_mc11_mmi; + c->avg_h264_qpel_pixels_tab[0][6] = ff_avg_h264_qpel16_mc21_mmi; + c->avg_h264_qpel_pixels_tab[0][7] = ff_avg_h264_qpel16_mc31_mmi; + c->avg_h264_qpel_pixels_tab[0][8] = ff_avg_h264_qpel16_mc02_mmi; + c->avg_h264_qpel_pixels_tab[0][9] = ff_avg_h264_qpel16_mc12_mmi; + c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_mmi; + c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_mmi; + c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_mmi; + c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_mmi; + c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_mmi; + c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_mmi; - c->avg_h264_qpel_pixels_tab[1][0] = ff_avg_h264_qpel8_mc00_msa; - c->avg_h264_qpel_pixels_tab[1][1] = ff_avg_h264_qpel8_mc10_msa; - c->avg_h264_qpel_pixels_tab[1][2] = ff_avg_h264_qpel8_mc20_msa; - c->avg_h264_qpel_pixels_tab[1][3] = ff_avg_h264_qpel8_mc30_msa; - c->avg_h264_qpel_pixels_tab[1][4] = ff_avg_h264_qpel8_mc01_msa; - c->avg_h264_qpel_pixels_tab[1][5] = ff_avg_h264_qpel8_mc11_msa; - c->avg_h264_qpel_pixels_tab[1][6] = ff_avg_h264_qpel8_mc21_msa; - c->avg_h264_qpel_pixels_tab[1][7] = ff_avg_h264_qpel8_mc31_msa; - c->avg_h264_qpel_pixels_tab[1][8] = ff_avg_h264_qpel8_mc02_msa; - c->avg_h264_qpel_pixels_tab[1][9] = ff_avg_h264_qpel8_mc12_msa; - c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_msa; - c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_msa; - c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_msa; - c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_msa; - c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_msa; - c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_msa; + c->avg_h264_qpel_pixels_tab[1][0] = ff_avg_h264_qpel8_mc00_mmi; + c->avg_h264_qpel_pixels_tab[1][1] = ff_avg_h264_qpel8_mc10_mmi; + c->avg_h264_qpel_pixels_tab[1][2] = ff_avg_h264_qpel8_mc20_mmi; + c->avg_h264_qpel_pixels_tab[1][3] = ff_avg_h264_qpel8_mc30_mmi; + c->avg_h264_qpel_pixels_tab[1][4] = ff_avg_h264_qpel8_mc01_mmi; + c->avg_h264_qpel_pixels_tab[1][5] = ff_avg_h264_qpel8_mc11_mmi; + c->avg_h264_qpel_pixels_tab[1][6] = ff_avg_h264_qpel8_mc21_mmi; + c->avg_h264_qpel_pixels_tab[1][7] = ff_avg_h264_qpel8_mc31_mmi; + c->avg_h264_qpel_pixels_tab[1][8] = ff_avg_h264_qpel8_mc02_mmi; + c->avg_h264_qpel_pixels_tab[1][9] = ff_avg_h264_qpel8_mc12_mmi; + c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_mmi; + c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_mmi; + c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_mmi; + c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_mmi; + c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_mmi; + c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_mmi; - c->avg_h264_qpel_pixels_tab[2][0] = ff_avg_h264_qpel4_mc00_msa; - c->avg_h264_qpel_pixels_tab[2][1] = ff_avg_h264_qpel4_mc10_msa; - c->avg_h264_qpel_pixels_tab[2][2] = ff_avg_h264_qpel4_mc20_msa; - c->avg_h264_qpel_pixels_tab[2][3] = ff_avg_h264_qpel4_mc30_msa; - c->avg_h264_qpel_pixels_tab[2][4] = ff_avg_h264_qpel4_mc01_msa; - c->avg_h264_qpel_pixels_tab[2][5] = ff_avg_h264_qpel4_mc11_msa; - c->avg_h264_qpel_pixels_tab[2][6] = ff_avg_h264_qpel4_mc21_msa; - c->avg_h264_qpel_pixels_tab[2][7] = ff_avg_h264_qpel4_mc31_msa; - c->avg_h264_qpel_pixels_tab[2][8] = ff_avg_h264_qpel4_mc02_msa; - c->avg_h264_qpel_pixels_tab[2][9] = ff_avg_h264_qpel4_mc12_msa; - c->avg_h264_qpel_pixels_tab[2][10] = ff_avg_h264_qpel4_mc22_msa; - c->avg_h264_qpel_pixels_tab[2][11] = ff_avg_h264_qpel4_mc32_msa; - c->avg_h264_qpel_pixels_tab[2][12] = ff_avg_h264_qpel4_mc03_msa; - c->avg_h264_qpel_pixels_tab[2][13] = ff_avg_h264_qpel4_mc13_msa; - c->avg_h264_qpel_pixels_tab[2][14] = ff_avg_h264_qpel4_mc23_msa; - c->avg_h264_qpel_pixels_tab[2][15] = ff_avg_h264_qpel4_mc33_msa; + c->avg_h264_qpel_pixels_tab[2][0] = ff_avg_h264_qpel4_mc00_mmi; + c->avg_h264_qpel_pixels_tab[2][1] = ff_avg_h264_qpel4_mc10_mmi; + c->avg_h264_qpel_pixels_tab[2][2] = ff_avg_h264_qpel4_mc20_mmi; + c->avg_h264_qpel_pixels_tab[2][3] = ff_avg_h264_qpel4_mc30_mmi; + c->avg_h264_qpel_pixels_tab[2][4] = ff_avg_h264_qpel4_mc01_mmi; + c->avg_h264_qpel_pixels_tab[2][5] = ff_avg_h264_qpel4_mc11_mmi; + c->avg_h264_qpel_pixels_tab[2][6] = ff_avg_h264_qpel4_mc21_mmi; + c->avg_h264_qpel_pixels_tab[2][7] = ff_avg_h264_qpel4_mc31_mmi; + c->avg_h264_qpel_pixels_tab[2][8] = ff_avg_h264_qpel4_mc02_mmi; + c->avg_h264_qpel_pixels_tab[2][9] = ff_avg_h264_qpel4_mc12_mmi; + c->avg_h264_qpel_pixels_tab[2][10] = ff_avg_h264_qpel4_mc22_mmi; + c->avg_h264_qpel_pixels_tab[2][11] = ff_avg_h264_qpel4_mc32_mmi; + c->avg_h264_qpel_pixels_tab[2][12] = ff_avg_h264_qpel4_mc03_mmi; + c->avg_h264_qpel_pixels_tab[2][13] = ff_avg_h264_qpel4_mc13_mmi; + c->avg_h264_qpel_pixels_tab[2][14] = ff_avg_h264_qpel4_mc23_mmi; + c->avg_h264_qpel_pixels_tab[2][15] = ff_avg_h264_qpel4_mc33_mmi; + } } -} -#endif // #if HAVE_MSA -#if HAVE_MMI -static av_cold void h264qpel_init_mmi(H264QpelContext *c, int bit_depth) -{ - if (8 == bit_depth) { - c->put_h264_qpel_pixels_tab[0][0] = ff_put_h264_qpel16_mc00_mmi; - c->put_h264_qpel_pixels_tab[0][1] = ff_put_h264_qpel16_mc10_mmi; - c->put_h264_qpel_pixels_tab[0][2] = ff_put_h264_qpel16_mc20_mmi; - c->put_h264_qpel_pixels_tab[0][3] = ff_put_h264_qpel16_mc30_mmi; - c->put_h264_qpel_pixels_tab[0][4] = ff_put_h264_qpel16_mc01_mmi; - c->put_h264_qpel_pixels_tab[0][5] = ff_put_h264_qpel16_mc11_mmi; - c->put_h264_qpel_pixels_tab[0][6] = ff_put_h264_qpel16_mc21_mmi; - c->put_h264_qpel_pixels_tab[0][7] = ff_put_h264_qpel16_mc31_mmi; - c->put_h264_qpel_pixels_tab[0][8] = ff_put_h264_qpel16_mc02_mmi; - c->put_h264_qpel_pixels_tab[0][9] = ff_put_h264_qpel16_mc12_mmi; - c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_mmi; - c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_mmi; - c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_mmi; - c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_mmi; - c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_mmi; - c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_mmi; + if (have_msa(cpu_flags)) { + if (bit_depth == 8) { + c->put_h264_qpel_pixels_tab[0][0] = ff_put_h264_qpel16_mc00_msa; + c->put_h264_qpel_pixels_tab[0][1] = ff_put_h264_qpel16_mc10_msa; + c->put_h264_qpel_pixels_tab[0][2] = ff_put_h264_qpel16_mc20_msa; + c->put_h264_qpel_pixels_tab[0][3] = ff_put_h264_qpel16_mc30_msa; + c->put_h264_qpel_pixels_tab[0][4] = ff_put_h264_qpel16_mc01_msa; + c->put_h264_qpel_pixels_tab[0][5] = ff_put_h264_qpel16_mc11_msa; + c->put_h264_qpel_pixels_tab[0][6] = ff_put_h264_qpel16_mc21_msa; + c->put_h264_qpel_pixels_tab[0][7] = ff_put_h264_qpel16_mc31_msa; + c->put_h264_qpel_pixels_tab[0][8] = ff_put_h264_qpel16_mc02_msa; + c->put_h264_qpel_pixels_tab[0][9] = ff_put_h264_qpel16_mc12_msa; + c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_msa; + c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_msa; + c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_msa; + c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_msa; + c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_msa; + c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_msa; - c->put_h264_qpel_pixels_tab[1][0] = ff_put_h264_qpel8_mc00_mmi; - c->put_h264_qpel_pixels_tab[1][1] = ff_put_h264_qpel8_mc10_mmi; - c->put_h264_qpel_pixels_tab[1][2] = ff_put_h264_qpel8_mc20_mmi; - c->put_h264_qpel_pixels_tab[1][3] = ff_put_h264_qpel8_mc30_mmi; - c->put_h264_qpel_pixels_tab[1][4] = ff_put_h264_qpel8_mc01_mmi; - c->put_h264_qpel_pixels_tab[1][5] = ff_put_h264_qpel8_mc11_mmi; - c->put_h264_qpel_pixels_tab[1][6] = ff_put_h264_qpel8_mc21_mmi; - c->put_h264_qpel_pixels_tab[1][7] = ff_put_h264_qpel8_mc31_mmi; - c->put_h264_qpel_pixels_tab[1][8] = ff_put_h264_qpel8_mc02_mmi; - c->put_h264_qpel_pixels_tab[1][9] = ff_put_h264_qpel8_mc12_mmi; - c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_mmi; - c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_mmi; - c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_mmi; - c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_mmi; - c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_mmi; - c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_mmi; + c->put_h264_qpel_pixels_tab[1][0] = ff_put_h264_qpel8_mc00_msa; + c->put_h264_qpel_pixels_tab[1][1] = ff_put_h264_qpel8_mc10_msa; + c->put_h264_qpel_pixels_tab[1][2] = ff_put_h264_qpel8_mc20_msa; + c->put_h264_qpel_pixels_tab[1][3] = ff_put_h264_qpel8_mc30_msa; + c->put_h264_qpel_pixels_tab[1][4] = ff_put_h264_qpel8_mc01_msa; + c->put_h264_qpel_pixels_tab[1][5] = ff_put_h264_qpel8_mc11_msa; + c->put_h264_qpel_pixels_tab[1][6] = ff_put_h264_qpel8_mc21_msa; + c->put_h264_qpel_pixels_tab[1][7] = ff_put_h264_qpel8_mc31_msa; + c->put_h264_qpel_pixels_tab[1][8] = ff_put_h264_qpel8_mc02_msa; + c->put_h264_qpel_pixels_tab[1][9] = ff_put_h264_qpel8_mc12_msa; + c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_msa; + c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_msa; + c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_msa; + c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_msa; + c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_msa; + c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_msa; - c->put_h264_qpel_pixels_tab[2][0] = ff_put_h264_qpel4_mc00_mmi; - c->put_h264_qpel_pixels_tab[2][1] = ff_put_h264_qpel4_mc10_mmi; - c->put_h264_qpel_pixels_tab[2][2] = ff_put_h264_qpel4_mc20_mmi; - c->put_h264_qpel_pixels_tab[2][3] = ff_put_h264_qpel4_mc30_mmi; - c->put_h264_qpel_pixels_tab[2][4] = ff_put_h264_qpel4_mc01_mmi; - c->put_h264_qpel_pixels_tab[2][5] = ff_put_h264_qpel4_mc11_mmi; - c->put_h264_qpel_pixels_tab[2][6] = ff_put_h264_qpel4_mc21_mmi; - c->put_h264_qpel_pixels_tab[2][7] = ff_put_h264_qpel4_mc31_mmi; - c->put_h264_qpel_pixels_tab[2][8] = ff_put_h264_qpel4_mc02_mmi; - c->put_h264_qpel_pixels_tab[2][9] = ff_put_h264_qpel4_mc12_mmi; - c->put_h264_qpel_pixels_tab[2][10] = ff_put_h264_qpel4_mc22_mmi; - c->put_h264_qpel_pixels_tab[2][11] = ff_put_h264_qpel4_mc32_mmi; - c->put_h264_qpel_pixels_tab[2][12] = ff_put_h264_qpel4_mc03_mmi; - c->put_h264_qpel_pixels_tab[2][13] = ff_put_h264_qpel4_mc13_mmi; - c->put_h264_qpel_pixels_tab[2][14] = ff_put_h264_qpel4_mc23_mmi; - c->put_h264_qpel_pixels_tab[2][15] = ff_put_h264_qpel4_mc33_mmi; + c->put_h264_qpel_pixels_tab[2][1] = ff_put_h264_qpel4_mc10_msa; + c->put_h264_qpel_pixels_tab[2][2] = ff_put_h264_qpel4_mc20_msa; + c->put_h264_qpel_pixels_tab[2][3] = ff_put_h264_qpel4_mc30_msa; + c->put_h264_qpel_pixels_tab[2][4] = ff_put_h264_qpel4_mc01_msa; + c->put_h264_qpel_pixels_tab[2][5] = ff_put_h264_qpel4_mc11_msa; + c->put_h264_qpel_pixels_tab[2][6] = ff_put_h264_qpel4_mc21_msa; + c->put_h264_qpel_pixels_tab[2][7] = ff_put_h264_qpel4_mc31_msa; + c->put_h264_qpel_pixels_tab[2][8] = ff_put_h264_qpel4_mc02_msa; + c->put_h264_qpel_pixels_tab[2][9] = ff_put_h264_qpel4_mc12_msa; + c->put_h264_qpel_pixels_tab[2][10] = ff_put_h264_qpel4_mc22_msa; + c->put_h264_qpel_pixels_tab[2][11] = ff_put_h264_qpel4_mc32_msa; + c->put_h264_qpel_pixels_tab[2][12] = ff_put_h264_qpel4_mc03_msa; + c->put_h264_qpel_pixels_tab[2][13] = ff_put_h264_qpel4_mc13_msa; + c->put_h264_qpel_pixels_tab[2][14] = ff_put_h264_qpel4_mc23_msa; + c->put_h264_qpel_pixels_tab[2][15] = ff_put_h264_qpel4_mc33_msa; - c->avg_h264_qpel_pixels_tab[0][0] = ff_avg_h264_qpel16_mc00_mmi; - c->avg_h264_qpel_pixels_tab[0][1] = ff_avg_h264_qpel16_mc10_mmi; - c->avg_h264_qpel_pixels_tab[0][2] = ff_avg_h264_qpel16_mc20_mmi; - c->avg_h264_qpel_pixels_tab[0][3] = ff_avg_h264_qpel16_mc30_mmi; - c->avg_h264_qpel_pixels_tab[0][4] = ff_avg_h264_qpel16_mc01_mmi; - c->avg_h264_qpel_pixels_tab[0][5] = ff_avg_h264_qpel16_mc11_mmi; - c->avg_h264_qpel_pixels_tab[0][6] = ff_avg_h264_qpel16_mc21_mmi; - c->avg_h264_qpel_pixels_tab[0][7] = ff_avg_h264_qpel16_mc31_mmi; - c->avg_h264_qpel_pixels_tab[0][8] = ff_avg_h264_qpel16_mc02_mmi; - c->avg_h264_qpel_pixels_tab[0][9] = ff_avg_h264_qpel16_mc12_mmi; - c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_mmi; - c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_mmi; - c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_mmi; - c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_mmi; - c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_mmi; - c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_mmi; + c->avg_h264_qpel_pixels_tab[0][0] = ff_avg_h264_qpel16_mc00_msa; + c->avg_h264_qpel_pixels_tab[0][1] = ff_avg_h264_qpel16_mc10_msa; + c->avg_h264_qpel_pixels_tab[0][2] = ff_avg_h264_qpel16_mc20_msa; + c->avg_h264_qpel_pixels_tab[0][3] = ff_avg_h264_qpel16_mc30_msa; + c->avg_h264_qpel_pixels_tab[0][4] = ff_avg_h264_qpel16_mc01_msa; + c->avg_h264_qpel_pixels_tab[0][5] = ff_avg_h264_qpel16_mc11_msa; + c->avg_h264_qpel_pixels_tab[0][6] = ff_avg_h264_qpel16_mc21_msa; + c->avg_h264_qpel_pixels_tab[0][7] = ff_avg_h264_qpel16_mc31_msa; + c->avg_h264_qpel_pixels_tab[0][8] = ff_avg_h264_qpel16_mc02_msa; + c->avg_h264_qpel_pixels_tab[0][9] = ff_avg_h264_qpel16_mc12_msa; + c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_msa; + c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_msa; + c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_msa; + c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_msa; + c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_msa; + c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_msa; - c->avg_h264_qpel_pixels_tab[1][0] = ff_avg_h264_qpel8_mc00_mmi; - c->avg_h264_qpel_pixels_tab[1][1] = ff_avg_h264_qpel8_mc10_mmi; - c->avg_h264_qpel_pixels_tab[1][2] = ff_avg_h264_qpel8_mc20_mmi; - c->avg_h264_qpel_pixels_tab[1][3] = ff_avg_h264_qpel8_mc30_mmi; - c->avg_h264_qpel_pixels_tab[1][4] = ff_avg_h264_qpel8_mc01_mmi; - c->avg_h264_qpel_pixels_tab[1][5] = ff_avg_h264_qpel8_mc11_mmi; - c->avg_h264_qpel_pixels_tab[1][6] = ff_avg_h264_qpel8_mc21_mmi; - c->avg_h264_qpel_pixels_tab[1][7] = ff_avg_h264_qpel8_mc31_mmi; - c->avg_h264_qpel_pixels_tab[1][8] = ff_avg_h264_qpel8_mc02_mmi; - c->avg_h264_qpel_pixels_tab[1][9] = ff_avg_h264_qpel8_mc12_mmi; - c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_mmi; - c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_mmi; - c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_mmi; - c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_mmi; - c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_mmi; - c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_mmi; + c->avg_h264_qpel_pixels_tab[1][0] = ff_avg_h264_qpel8_mc00_msa; + c->avg_h264_qpel_pixels_tab[1][1] = ff_avg_h264_qpel8_mc10_msa; + c->avg_h264_qpel_pixels_tab[1][2] = ff_avg_h264_qpel8_mc20_msa; + c->avg_h264_qpel_pixels_tab[1][3] = ff_avg_h264_qpel8_mc30_msa; + c->avg_h264_qpel_pixels_tab[1][4] = ff_avg_h264_qpel8_mc01_msa; + c->avg_h264_qpel_pixels_tab[1][5] = ff_avg_h264_qpel8_mc11_msa; + c->avg_h264_qpel_pixels_tab[1][6] = ff_avg_h264_qpel8_mc21_msa; + c->avg_h264_qpel_pixels_tab[1][7] = ff_avg_h264_qpel8_mc31_msa; + c->avg_h264_qpel_pixels_tab[1][8] = ff_avg_h264_qpel8_mc02_msa; + c->avg_h264_qpel_pixels_tab[1][9] = ff_avg_h264_qpel8_mc12_msa; + c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_msa; + c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_msa; + c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_msa; + c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_msa; + c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_msa; + c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_msa; - c->avg_h264_qpel_pixels_tab[2][0] = ff_avg_h264_qpel4_mc00_mmi; - c->avg_h264_qpel_pixels_tab[2][1] = ff_avg_h264_qpel4_mc10_mmi; - c->avg_h264_qpel_pixels_tab[2][2] = ff_avg_h264_qpel4_mc20_mmi; - c->avg_h264_qpel_pixels_tab[2][3] = ff_avg_h264_qpel4_mc30_mmi; - c->avg_h264_qpel_pixels_tab[2][4] = ff_avg_h264_qpel4_mc01_mmi; - c->avg_h264_qpel_pixels_tab[2][5] = ff_avg_h264_qpel4_mc11_mmi; - c->avg_h264_qpel_pixels_tab[2][6] = ff_avg_h264_qpel4_mc21_mmi; - c->avg_h264_qpel_pixels_tab[2][7] = ff_avg_h264_qpel4_mc31_mmi; - c->avg_h264_qpel_pixels_tab[2][8] = ff_avg_h264_qpel4_mc02_mmi; - c->avg_h264_qpel_pixels_tab[2][9] = ff_avg_h264_qpel4_mc12_mmi; - c->avg_h264_qpel_pixels_tab[2][10] = ff_avg_h264_qpel4_mc22_mmi; - c->avg_h264_qpel_pixels_tab[2][11] = ff_avg_h264_qpel4_mc32_mmi; - c->avg_h264_qpel_pixels_tab[2][12] = ff_avg_h264_qpel4_mc03_mmi; - c->avg_h264_qpel_pixels_tab[2][13] = ff_avg_h264_qpel4_mc13_mmi; - c->avg_h264_qpel_pixels_tab[2][14] = ff_avg_h264_qpel4_mc23_mmi; - c->avg_h264_qpel_pixels_tab[2][15] = ff_avg_h264_qpel4_mc33_mmi; + c->avg_h264_qpel_pixels_tab[2][0] = ff_avg_h264_qpel4_mc00_msa; + c->avg_h264_qpel_pixels_tab[2][1] = ff_avg_h264_qpel4_mc10_msa; + c->avg_h264_qpel_pixels_tab[2][2] = ff_avg_h264_qpel4_mc20_msa; + c->avg_h264_qpel_pixels_tab[2][3] = ff_avg_h264_qpel4_mc30_msa; + c->avg_h264_qpel_pixels_tab[2][4] = ff_avg_h264_qpel4_mc01_msa; + c->avg_h264_qpel_pixels_tab[2][5] = ff_avg_h264_qpel4_mc11_msa; + c->avg_h264_qpel_pixels_tab[2][6] = ff_avg_h264_qpel4_mc21_msa; + c->avg_h264_qpel_pixels_tab[2][7] = ff_avg_h264_qpel4_mc31_msa; + c->avg_h264_qpel_pixels_tab[2][8] = ff_avg_h264_qpel4_mc02_msa; + c->avg_h264_qpel_pixels_tab[2][9] = ff_avg_h264_qpel4_mc12_msa; + c->avg_h264_qpel_pixels_tab[2][10] = ff_avg_h264_qpel4_mc22_msa; + c->avg_h264_qpel_pixels_tab[2][11] = ff_avg_h264_qpel4_mc32_msa; + c->avg_h264_qpel_pixels_tab[2][12] = ff_avg_h264_qpel4_mc03_msa; + c->avg_h264_qpel_pixels_tab[2][13] = ff_avg_h264_qpel4_mc13_msa; + c->avg_h264_qpel_pixels_tab[2][14] = ff_avg_h264_qpel4_mc23_msa; + c->avg_h264_qpel_pixels_tab[2][15] = ff_avg_h264_qpel4_mc33_msa; + } } } -#endif /* HAVE_MMI */ - -av_cold void ff_h264qpel_init_mips(H264QpelContext *c, int bit_depth) -{ -#if HAVE_MMI - h264qpel_init_mmi(c, bit_depth); -#endif /* HAVE_MMI */ -#if HAVE_MSA - h264qpel_init_msa(c, bit_depth); -#endif // #if HAVE_MSA -} diff --git a/libavcodec/mips/hevcdsp_init_mips.c b/libavcodec/mips/hevcdsp_init_mips.c index 88337f462e..eb261e5adf 100644 --- a/libavcodec/mips/hevcdsp_init_mips.c +++ b/libavcodec/mips/hevcdsp_init_mips.c @@ -18,512 +18,500 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "libavcodec/mips/hevcdsp_mips.h" -#if HAVE_MMI -static av_cold void hevc_dsp_init_mmi(HEVCDSPContext *c, - const int bit_depth) +void ff_hevc_dsp_init_mips(HEVCDSPContext *c, const int bit_depth) { - if (8 == bit_depth) { - c->put_hevc_qpel[1][0][1] = ff_hevc_put_hevc_qpel_h4_8_mmi; - c->put_hevc_qpel[3][0][1] = ff_hevc_put_hevc_qpel_h8_8_mmi; - c->put_hevc_qpel[4][0][1] = ff_hevc_put_hevc_qpel_h12_8_mmi; - c->put_hevc_qpel[5][0][1] = ff_hevc_put_hevc_qpel_h16_8_mmi; - c->put_hevc_qpel[6][0][1] = ff_hevc_put_hevc_qpel_h24_8_mmi; - c->put_hevc_qpel[7][0][1] = ff_hevc_put_hevc_qpel_h32_8_mmi; - c->put_hevc_qpel[8][0][1] = ff_hevc_put_hevc_qpel_h48_8_mmi; - c->put_hevc_qpel[9][0][1] = ff_hevc_put_hevc_qpel_h64_8_mmi; - - c->put_hevc_qpel[1][1][1] = ff_hevc_put_hevc_qpel_hv4_8_mmi; - c->put_hevc_qpel[3][1][1] = ff_hevc_put_hevc_qpel_hv8_8_mmi; - c->put_hevc_qpel[4][1][1] = ff_hevc_put_hevc_qpel_hv12_8_mmi; - c->put_hevc_qpel[5][1][1] = ff_hevc_put_hevc_qpel_hv16_8_mmi; - c->put_hevc_qpel[6][1][1] = ff_hevc_put_hevc_qpel_hv24_8_mmi; - c->put_hevc_qpel[7][1][1] = ff_hevc_put_hevc_qpel_hv32_8_mmi; - c->put_hevc_qpel[8][1][1] = ff_hevc_put_hevc_qpel_hv48_8_mmi; - c->put_hevc_qpel[9][1][1] = ff_hevc_put_hevc_qpel_hv64_8_mmi; - - c->put_hevc_qpel_bi[1][0][1] = ff_hevc_put_hevc_qpel_bi_h4_8_mmi; - c->put_hevc_qpel_bi[3][0][1] = ff_hevc_put_hevc_qpel_bi_h8_8_mmi; - c->put_hevc_qpel_bi[4][0][1] = ff_hevc_put_hevc_qpel_bi_h12_8_mmi; - c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_qpel_bi_h16_8_mmi; - c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_qpel_bi_h24_8_mmi; - c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_qpel_bi_h32_8_mmi; - c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_qpel_bi_h48_8_mmi; - c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_qpel_bi_h64_8_mmi; - - c->put_hevc_qpel_bi[1][1][1] = ff_hevc_put_hevc_qpel_bi_hv4_8_mmi; - c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_qpel_bi_hv8_8_mmi; - c->put_hevc_qpel_bi[4][1][1] = ff_hevc_put_hevc_qpel_bi_hv12_8_mmi; - c->put_hevc_qpel_bi[5][1][1] = ff_hevc_put_hevc_qpel_bi_hv16_8_mmi; - c->put_hevc_qpel_bi[6][1][1] = ff_hevc_put_hevc_qpel_bi_hv24_8_mmi; - c->put_hevc_qpel_bi[7][1][1] = ff_hevc_put_hevc_qpel_bi_hv32_8_mmi; - c->put_hevc_qpel_bi[8][1][1] = ff_hevc_put_hevc_qpel_bi_hv48_8_mmi; - c->put_hevc_qpel_bi[9][1][1] = ff_hevc_put_hevc_qpel_bi_hv64_8_mmi; - - c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_pel_bi_pixels8_8_mmi; - c->put_hevc_qpel_bi[5][0][0] = ff_hevc_put_hevc_pel_bi_pixels16_8_mmi; - c->put_hevc_qpel_bi[6][0][0] = ff_hevc_put_hevc_pel_bi_pixels24_8_mmi; - c->put_hevc_qpel_bi[7][0][0] = ff_hevc_put_hevc_pel_bi_pixels32_8_mmi; - c->put_hevc_qpel_bi[8][0][0] = ff_hevc_put_hevc_pel_bi_pixels48_8_mmi; - c->put_hevc_qpel_bi[9][0][0] = ff_hevc_put_hevc_pel_bi_pixels64_8_mmi; - - c->put_hevc_epel_bi[3][0][0] = ff_hevc_put_hevc_pel_bi_pixels8_8_mmi; - c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_pel_bi_pixels16_8_mmi; - c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_pel_bi_pixels24_8_mmi; - c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_pel_bi_pixels32_8_mmi; - - c->put_hevc_epel_bi[1][1][1] = ff_hevc_put_hevc_epel_bi_hv4_8_mmi; - c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_epel_bi_hv8_8_mmi; - c->put_hevc_epel_bi[4][1][1] = ff_hevc_put_hevc_epel_bi_hv12_8_mmi; - c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_epel_bi_hv16_8_mmi; - c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_epel_bi_hv24_8_mmi; - c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_epel_bi_hv32_8_mmi; - - c->put_hevc_qpel_uni[1][1][1] = ff_hevc_put_hevc_qpel_uni_hv4_8_mmi; - c->put_hevc_qpel_uni[3][1][1] = ff_hevc_put_hevc_qpel_uni_hv8_8_mmi; - c->put_hevc_qpel_uni[4][1][1] = ff_hevc_put_hevc_qpel_uni_hv12_8_mmi; - c->put_hevc_qpel_uni[5][1][1] = ff_hevc_put_hevc_qpel_uni_hv16_8_mmi; - c->put_hevc_qpel_uni[6][1][1] = ff_hevc_put_hevc_qpel_uni_hv24_8_mmi; - c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_qpel_uni_hv32_8_mmi; - c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_qpel_uni_hv48_8_mmi; - c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_qpel_uni_hv64_8_mmi; + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + if (bit_depth == 8) { + c->put_hevc_qpel[1][0][1] = ff_hevc_put_hevc_qpel_h4_8_mmi; + c->put_hevc_qpel[3][0][1] = ff_hevc_put_hevc_qpel_h8_8_mmi; + c->put_hevc_qpel[4][0][1] = ff_hevc_put_hevc_qpel_h12_8_mmi; + c->put_hevc_qpel[5][0][1] = ff_hevc_put_hevc_qpel_h16_8_mmi; + c->put_hevc_qpel[6][0][1] = ff_hevc_put_hevc_qpel_h24_8_mmi; + c->put_hevc_qpel[7][0][1] = ff_hevc_put_hevc_qpel_h32_8_mmi; + c->put_hevc_qpel[8][0][1] = ff_hevc_put_hevc_qpel_h48_8_mmi; + c->put_hevc_qpel[9][0][1] = ff_hevc_put_hevc_qpel_h64_8_mmi; + + c->put_hevc_qpel[1][1][1] = ff_hevc_put_hevc_qpel_hv4_8_mmi; + c->put_hevc_qpel[3][1][1] = ff_hevc_put_hevc_qpel_hv8_8_mmi; + c->put_hevc_qpel[4][1][1] = ff_hevc_put_hevc_qpel_hv12_8_mmi; + c->put_hevc_qpel[5][1][1] = ff_hevc_put_hevc_qpel_hv16_8_mmi; + c->put_hevc_qpel[6][1][1] = ff_hevc_put_hevc_qpel_hv24_8_mmi; + c->put_hevc_qpel[7][1][1] = ff_hevc_put_hevc_qpel_hv32_8_mmi; + c->put_hevc_qpel[8][1][1] = ff_hevc_put_hevc_qpel_hv48_8_mmi; + c->put_hevc_qpel[9][1][1] = ff_hevc_put_hevc_qpel_hv64_8_mmi; + + c->put_hevc_qpel_bi[1][0][1] = ff_hevc_put_hevc_qpel_bi_h4_8_mmi; + c->put_hevc_qpel_bi[3][0][1] = ff_hevc_put_hevc_qpel_bi_h8_8_mmi; + c->put_hevc_qpel_bi[4][0][1] = ff_hevc_put_hevc_qpel_bi_h12_8_mmi; + c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_qpel_bi_h16_8_mmi; + c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_qpel_bi_h24_8_mmi; + c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_qpel_bi_h32_8_mmi; + c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_qpel_bi_h48_8_mmi; + c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_qpel_bi_h64_8_mmi; + + c->put_hevc_qpel_bi[1][1][1] = ff_hevc_put_hevc_qpel_bi_hv4_8_mmi; + c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_qpel_bi_hv8_8_mmi; + c->put_hevc_qpel_bi[4][1][1] = ff_hevc_put_hevc_qpel_bi_hv12_8_mmi; + c->put_hevc_qpel_bi[5][1][1] = ff_hevc_put_hevc_qpel_bi_hv16_8_mmi; + c->put_hevc_qpel_bi[6][1][1] = ff_hevc_put_hevc_qpel_bi_hv24_8_mmi; + c->put_hevc_qpel_bi[7][1][1] = ff_hevc_put_hevc_qpel_bi_hv32_8_mmi; + c->put_hevc_qpel_bi[8][1][1] = ff_hevc_put_hevc_qpel_bi_hv48_8_mmi; + c->put_hevc_qpel_bi[9][1][1] = ff_hevc_put_hevc_qpel_bi_hv64_8_mmi; + + c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_pel_bi_pixels8_8_mmi; + c->put_hevc_qpel_bi[5][0][0] = ff_hevc_put_hevc_pel_bi_pixels16_8_mmi; + c->put_hevc_qpel_bi[6][0][0] = ff_hevc_put_hevc_pel_bi_pixels24_8_mmi; + c->put_hevc_qpel_bi[7][0][0] = ff_hevc_put_hevc_pel_bi_pixels32_8_mmi; + c->put_hevc_qpel_bi[8][0][0] = ff_hevc_put_hevc_pel_bi_pixels48_8_mmi; + c->put_hevc_qpel_bi[9][0][0] = ff_hevc_put_hevc_pel_bi_pixels64_8_mmi; + + c->put_hevc_epel_bi[3][0][0] = ff_hevc_put_hevc_pel_bi_pixels8_8_mmi; + c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_pel_bi_pixels16_8_mmi; + c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_pel_bi_pixels24_8_mmi; + c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_pel_bi_pixels32_8_mmi; + + c->put_hevc_epel_bi[1][1][1] = ff_hevc_put_hevc_epel_bi_hv4_8_mmi; + c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_epel_bi_hv8_8_mmi; + c->put_hevc_epel_bi[4][1][1] = ff_hevc_put_hevc_epel_bi_hv12_8_mmi; + c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_epel_bi_hv16_8_mmi; + c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_epel_bi_hv24_8_mmi; + c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_epel_bi_hv32_8_mmi; + + c->put_hevc_qpel_uni[1][1][1] = ff_hevc_put_hevc_qpel_uni_hv4_8_mmi; + c->put_hevc_qpel_uni[3][1][1] = ff_hevc_put_hevc_qpel_uni_hv8_8_mmi; + c->put_hevc_qpel_uni[4][1][1] = ff_hevc_put_hevc_qpel_uni_hv12_8_mmi; + c->put_hevc_qpel_uni[5][1][1] = ff_hevc_put_hevc_qpel_uni_hv16_8_mmi; + c->put_hevc_qpel_uni[6][1][1] = ff_hevc_put_hevc_qpel_uni_hv24_8_mmi; + c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_qpel_uni_hv32_8_mmi; + c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_qpel_uni_hv48_8_mmi; + c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_qpel_uni_hv64_8_mmi; + } } -} -#endif // #if HAVE_MMI -#if HAVE_MSA -static av_cold void hevc_dsp_init_msa(HEVCDSPContext *c, - const int bit_depth) -{ - if (8 == bit_depth) { - c->put_hevc_qpel[1][0][0] = ff_hevc_put_hevc_pel_pixels4_8_msa; - c->put_hevc_qpel[2][0][0] = ff_hevc_put_hevc_pel_pixels6_8_msa; - c->put_hevc_qpel[3][0][0] = ff_hevc_put_hevc_pel_pixels8_8_msa; - c->put_hevc_qpel[4][0][0] = ff_hevc_put_hevc_pel_pixels12_8_msa; - c->put_hevc_qpel[5][0][0] = ff_hevc_put_hevc_pel_pixels16_8_msa; - c->put_hevc_qpel[6][0][0] = ff_hevc_put_hevc_pel_pixels24_8_msa; - c->put_hevc_qpel[7][0][0] = ff_hevc_put_hevc_pel_pixels32_8_msa; - c->put_hevc_qpel[8][0][0] = ff_hevc_put_hevc_pel_pixels48_8_msa; - c->put_hevc_qpel[9][0][0] = ff_hevc_put_hevc_pel_pixels64_8_msa; - - c->put_hevc_qpel[1][0][1] = ff_hevc_put_hevc_qpel_h4_8_msa; - c->put_hevc_qpel[3][0][1] = ff_hevc_put_hevc_qpel_h8_8_msa; - c->put_hevc_qpel[4][0][1] = ff_hevc_put_hevc_qpel_h12_8_msa; - c->put_hevc_qpel[5][0][1] = ff_hevc_put_hevc_qpel_h16_8_msa; - c->put_hevc_qpel[6][0][1] = ff_hevc_put_hevc_qpel_h24_8_msa; - c->put_hevc_qpel[7][0][1] = ff_hevc_put_hevc_qpel_h32_8_msa; - c->put_hevc_qpel[8][0][1] = ff_hevc_put_hevc_qpel_h48_8_msa; - c->put_hevc_qpel[9][0][1] = ff_hevc_put_hevc_qpel_h64_8_msa; - - c->put_hevc_qpel[1][1][0] = ff_hevc_put_hevc_qpel_v4_8_msa; - c->put_hevc_qpel[3][1][0] = ff_hevc_put_hevc_qpel_v8_8_msa; - c->put_hevc_qpel[4][1][0] = ff_hevc_put_hevc_qpel_v12_8_msa; - c->put_hevc_qpel[5][1][0] = ff_hevc_put_hevc_qpel_v16_8_msa; - c->put_hevc_qpel[6][1][0] = ff_hevc_put_hevc_qpel_v24_8_msa; - c->put_hevc_qpel[7][1][0] = ff_hevc_put_hevc_qpel_v32_8_msa; - c->put_hevc_qpel[8][1][0] = ff_hevc_put_hevc_qpel_v48_8_msa; - c->put_hevc_qpel[9][1][0] = ff_hevc_put_hevc_qpel_v64_8_msa; - - c->put_hevc_qpel[1][1][1] = ff_hevc_put_hevc_qpel_hv4_8_msa; - c->put_hevc_qpel[3][1][1] = ff_hevc_put_hevc_qpel_hv8_8_msa; - c->put_hevc_qpel[4][1][1] = ff_hevc_put_hevc_qpel_hv12_8_msa; - c->put_hevc_qpel[5][1][1] = ff_hevc_put_hevc_qpel_hv16_8_msa; - c->put_hevc_qpel[6][1][1] = ff_hevc_put_hevc_qpel_hv24_8_msa; - c->put_hevc_qpel[7][1][1] = ff_hevc_put_hevc_qpel_hv32_8_msa; - c->put_hevc_qpel[8][1][1] = ff_hevc_put_hevc_qpel_hv48_8_msa; - c->put_hevc_qpel[9][1][1] = ff_hevc_put_hevc_qpel_hv64_8_msa; - - c->put_hevc_epel[1][0][0] = ff_hevc_put_hevc_pel_pixels4_8_msa; - c->put_hevc_epel[2][0][0] = ff_hevc_put_hevc_pel_pixels6_8_msa; - c->put_hevc_epel[3][0][0] = ff_hevc_put_hevc_pel_pixels8_8_msa; - c->put_hevc_epel[4][0][0] = ff_hevc_put_hevc_pel_pixels12_8_msa; - c->put_hevc_epel[5][0][0] = ff_hevc_put_hevc_pel_pixels16_8_msa; - c->put_hevc_epel[6][0][0] = ff_hevc_put_hevc_pel_pixels24_8_msa; - c->put_hevc_epel[7][0][0] = ff_hevc_put_hevc_pel_pixels32_8_msa; - - c->put_hevc_epel[1][0][1] = ff_hevc_put_hevc_epel_h4_8_msa; - c->put_hevc_epel[2][0][1] = ff_hevc_put_hevc_epel_h6_8_msa; - c->put_hevc_epel[3][0][1] = ff_hevc_put_hevc_epel_h8_8_msa; - c->put_hevc_epel[4][0][1] = ff_hevc_put_hevc_epel_h12_8_msa; - c->put_hevc_epel[5][0][1] = ff_hevc_put_hevc_epel_h16_8_msa; - c->put_hevc_epel[6][0][1] = ff_hevc_put_hevc_epel_h24_8_msa; - c->put_hevc_epel[7][0][1] = ff_hevc_put_hevc_epel_h32_8_msa; - - c->put_hevc_epel[1][1][0] = ff_hevc_put_hevc_epel_v4_8_msa; - c->put_hevc_epel[2][1][0] = ff_hevc_put_hevc_epel_v6_8_msa; - c->put_hevc_epel[3][1][0] = ff_hevc_put_hevc_epel_v8_8_msa; - c->put_hevc_epel[4][1][0] = ff_hevc_put_hevc_epel_v12_8_msa; - c->put_hevc_epel[5][1][0] = ff_hevc_put_hevc_epel_v16_8_msa; - c->put_hevc_epel[6][1][0] = ff_hevc_put_hevc_epel_v24_8_msa; - c->put_hevc_epel[7][1][0] = ff_hevc_put_hevc_epel_v32_8_msa; - - c->put_hevc_epel[1][1][1] = ff_hevc_put_hevc_epel_hv4_8_msa; - c->put_hevc_epel[2][1][1] = ff_hevc_put_hevc_epel_hv6_8_msa; - c->put_hevc_epel[3][1][1] = ff_hevc_put_hevc_epel_hv8_8_msa; - c->put_hevc_epel[4][1][1] = ff_hevc_put_hevc_epel_hv12_8_msa; - c->put_hevc_epel[5][1][1] = ff_hevc_put_hevc_epel_hv16_8_msa; - c->put_hevc_epel[6][1][1] = ff_hevc_put_hevc_epel_hv24_8_msa; - c->put_hevc_epel[7][1][1] = ff_hevc_put_hevc_epel_hv32_8_msa; - - c->put_hevc_qpel_uni[3][0][0] = ff_hevc_put_hevc_uni_pel_pixels8_8_msa; - c->put_hevc_qpel_uni[4][0][0] = ff_hevc_put_hevc_uni_pel_pixels12_8_msa; - c->put_hevc_qpel_uni[5][0][0] = ff_hevc_put_hevc_uni_pel_pixels16_8_msa; - c->put_hevc_qpel_uni[6][0][0] = ff_hevc_put_hevc_uni_pel_pixels24_8_msa; - c->put_hevc_qpel_uni[7][0][0] = ff_hevc_put_hevc_uni_pel_pixels32_8_msa; - c->put_hevc_qpel_uni[8][0][0] = ff_hevc_put_hevc_uni_pel_pixels48_8_msa; - c->put_hevc_qpel_uni[9][0][0] = ff_hevc_put_hevc_uni_pel_pixels64_8_msa; - - c->put_hevc_qpel_uni[1][0][1] = ff_hevc_put_hevc_uni_qpel_h4_8_msa; - c->put_hevc_qpel_uni[3][0][1] = ff_hevc_put_hevc_uni_qpel_h8_8_msa; - c->put_hevc_qpel_uni[4][0][1] = ff_hevc_put_hevc_uni_qpel_h12_8_msa; - c->put_hevc_qpel_uni[5][0][1] = ff_hevc_put_hevc_uni_qpel_h16_8_msa; - c->put_hevc_qpel_uni[6][0][1] = ff_hevc_put_hevc_uni_qpel_h24_8_msa; - c->put_hevc_qpel_uni[7][0][1] = ff_hevc_put_hevc_uni_qpel_h32_8_msa; - c->put_hevc_qpel_uni[8][0][1] = ff_hevc_put_hevc_uni_qpel_h48_8_msa; - c->put_hevc_qpel_uni[9][0][1] = ff_hevc_put_hevc_uni_qpel_h64_8_msa; - - c->put_hevc_qpel_uni[1][1][0] = ff_hevc_put_hevc_uni_qpel_v4_8_msa; - c->put_hevc_qpel_uni[3][1][0] = ff_hevc_put_hevc_uni_qpel_v8_8_msa; - c->put_hevc_qpel_uni[4][1][0] = ff_hevc_put_hevc_uni_qpel_v12_8_msa; - c->put_hevc_qpel_uni[5][1][0] = ff_hevc_put_hevc_uni_qpel_v16_8_msa; - c->put_hevc_qpel_uni[6][1][0] = ff_hevc_put_hevc_uni_qpel_v24_8_msa; - c->put_hevc_qpel_uni[7][1][0] = ff_hevc_put_hevc_uni_qpel_v32_8_msa; - c->put_hevc_qpel_uni[8][1][0] = ff_hevc_put_hevc_uni_qpel_v48_8_msa; - c->put_hevc_qpel_uni[9][1][0] = ff_hevc_put_hevc_uni_qpel_v64_8_msa; - - c->put_hevc_qpel_uni[1][1][1] = ff_hevc_put_hevc_uni_qpel_hv4_8_msa; - c->put_hevc_qpel_uni[3][1][1] = ff_hevc_put_hevc_uni_qpel_hv8_8_msa; - c->put_hevc_qpel_uni[4][1][1] = ff_hevc_put_hevc_uni_qpel_hv12_8_msa; - c->put_hevc_qpel_uni[5][1][1] = ff_hevc_put_hevc_uni_qpel_hv16_8_msa; - c->put_hevc_qpel_uni[6][1][1] = ff_hevc_put_hevc_uni_qpel_hv24_8_msa; - c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_uni_qpel_hv32_8_msa; - c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_uni_qpel_hv48_8_msa; - c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_uni_qpel_hv64_8_msa; - - c->put_hevc_epel_uni[3][0][0] = ff_hevc_put_hevc_uni_pel_pixels8_8_msa; - c->put_hevc_epel_uni[4][0][0] = ff_hevc_put_hevc_uni_pel_pixels12_8_msa; - c->put_hevc_epel_uni[5][0][0] = ff_hevc_put_hevc_uni_pel_pixels16_8_msa; - c->put_hevc_epel_uni[6][0][0] = ff_hevc_put_hevc_uni_pel_pixels24_8_msa; - c->put_hevc_epel_uni[7][0][0] = ff_hevc_put_hevc_uni_pel_pixels32_8_msa; - - c->put_hevc_epel_uni[1][0][1] = ff_hevc_put_hevc_uni_epel_h4_8_msa; - c->put_hevc_epel_uni[2][0][1] = ff_hevc_put_hevc_uni_epel_h6_8_msa; - c->put_hevc_epel_uni[3][0][1] = ff_hevc_put_hevc_uni_epel_h8_8_msa; - c->put_hevc_epel_uni[4][0][1] = ff_hevc_put_hevc_uni_epel_h12_8_msa; - c->put_hevc_epel_uni[5][0][1] = ff_hevc_put_hevc_uni_epel_h16_8_msa; - c->put_hevc_epel_uni[6][0][1] = ff_hevc_put_hevc_uni_epel_h24_8_msa; - c->put_hevc_epel_uni[7][0][1] = ff_hevc_put_hevc_uni_epel_h32_8_msa; - - c->put_hevc_epel_uni[1][1][0] = ff_hevc_put_hevc_uni_epel_v4_8_msa; - c->put_hevc_epel_uni[2][1][0] = ff_hevc_put_hevc_uni_epel_v6_8_msa; - c->put_hevc_epel_uni[3][1][0] = ff_hevc_put_hevc_uni_epel_v8_8_msa; - c->put_hevc_epel_uni[4][1][0] = ff_hevc_put_hevc_uni_epel_v12_8_msa; - c->put_hevc_epel_uni[5][1][0] = ff_hevc_put_hevc_uni_epel_v16_8_msa; - c->put_hevc_epel_uni[6][1][0] = ff_hevc_put_hevc_uni_epel_v24_8_msa; - c->put_hevc_epel_uni[7][1][0] = ff_hevc_put_hevc_uni_epel_v32_8_msa; - - c->put_hevc_epel_uni[1][1][1] = ff_hevc_put_hevc_uni_epel_hv4_8_msa; - c->put_hevc_epel_uni[2][1][1] = ff_hevc_put_hevc_uni_epel_hv6_8_msa; - c->put_hevc_epel_uni[3][1][1] = ff_hevc_put_hevc_uni_epel_hv8_8_msa; - c->put_hevc_epel_uni[4][1][1] = ff_hevc_put_hevc_uni_epel_hv12_8_msa; - c->put_hevc_epel_uni[5][1][1] = ff_hevc_put_hevc_uni_epel_hv16_8_msa; - c->put_hevc_epel_uni[6][1][1] = ff_hevc_put_hevc_uni_epel_hv24_8_msa; - c->put_hevc_epel_uni[7][1][1] = ff_hevc_put_hevc_uni_epel_hv32_8_msa; - - c->put_hevc_qpel_uni_w[1][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels4_8_msa; - c->put_hevc_qpel_uni_w[3][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels8_8_msa; - c->put_hevc_qpel_uni_w[4][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels12_8_msa; - c->put_hevc_qpel_uni_w[5][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels16_8_msa; - c->put_hevc_qpel_uni_w[6][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels24_8_msa; - c->put_hevc_qpel_uni_w[7][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels32_8_msa; - c->put_hevc_qpel_uni_w[8][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels48_8_msa; - c->put_hevc_qpel_uni_w[9][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels64_8_msa; - - c->put_hevc_qpel_uni_w[1][0][1] = ff_hevc_put_hevc_uni_w_qpel_h4_8_msa; - c->put_hevc_qpel_uni_w[3][0][1] = ff_hevc_put_hevc_uni_w_qpel_h8_8_msa; - c->put_hevc_qpel_uni_w[4][0][1] = ff_hevc_put_hevc_uni_w_qpel_h12_8_msa; - c->put_hevc_qpel_uni_w[5][0][1] = ff_hevc_put_hevc_uni_w_qpel_h16_8_msa; - c->put_hevc_qpel_uni_w[6][0][1] = ff_hevc_put_hevc_uni_w_qpel_h24_8_msa; - c->put_hevc_qpel_uni_w[7][0][1] = ff_hevc_put_hevc_uni_w_qpel_h32_8_msa; - c->put_hevc_qpel_uni_w[8][0][1] = ff_hevc_put_hevc_uni_w_qpel_h48_8_msa; - c->put_hevc_qpel_uni_w[9][0][1] = ff_hevc_put_hevc_uni_w_qpel_h64_8_msa; - - c->put_hevc_qpel_uni_w[1][1][0] = ff_hevc_put_hevc_uni_w_qpel_v4_8_msa; - c->put_hevc_qpel_uni_w[3][1][0] = ff_hevc_put_hevc_uni_w_qpel_v8_8_msa; - c->put_hevc_qpel_uni_w[4][1][0] = ff_hevc_put_hevc_uni_w_qpel_v12_8_msa; - c->put_hevc_qpel_uni_w[5][1][0] = ff_hevc_put_hevc_uni_w_qpel_v16_8_msa; - c->put_hevc_qpel_uni_w[6][1][0] = ff_hevc_put_hevc_uni_w_qpel_v24_8_msa; - c->put_hevc_qpel_uni_w[7][1][0] = ff_hevc_put_hevc_uni_w_qpel_v32_8_msa; - c->put_hevc_qpel_uni_w[8][1][0] = ff_hevc_put_hevc_uni_w_qpel_v48_8_msa; - c->put_hevc_qpel_uni_w[9][1][0] = ff_hevc_put_hevc_uni_w_qpel_v64_8_msa; - - c->put_hevc_qpel_uni_w[1][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv4_8_msa; - c->put_hevc_qpel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv8_8_msa; - c->put_hevc_qpel_uni_w[4][1][1] = - ff_hevc_put_hevc_uni_w_qpel_hv12_8_msa; - c->put_hevc_qpel_uni_w[5][1][1] = - ff_hevc_put_hevc_uni_w_qpel_hv16_8_msa; - c->put_hevc_qpel_uni_w[6][1][1] = - ff_hevc_put_hevc_uni_w_qpel_hv24_8_msa; - c->put_hevc_qpel_uni_w[7][1][1] = - ff_hevc_put_hevc_uni_w_qpel_hv32_8_msa; - c->put_hevc_qpel_uni_w[8][1][1] = - ff_hevc_put_hevc_uni_w_qpel_hv48_8_msa; - c->put_hevc_qpel_uni_w[9][1][1] = - ff_hevc_put_hevc_uni_w_qpel_hv64_8_msa; - - c->put_hevc_epel_uni_w[1][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels4_8_msa; - c->put_hevc_epel_uni_w[2][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels6_8_msa; - c->put_hevc_epel_uni_w[3][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels8_8_msa; - c->put_hevc_epel_uni_w[4][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels12_8_msa; - c->put_hevc_epel_uni_w[5][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels16_8_msa; - c->put_hevc_epel_uni_w[6][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels24_8_msa; - c->put_hevc_epel_uni_w[7][0][0] = - ff_hevc_put_hevc_uni_w_pel_pixels32_8_msa; - - c->put_hevc_epel_uni_w[1][0][1] = ff_hevc_put_hevc_uni_w_epel_h4_8_msa; - c->put_hevc_epel_uni_w[2][0][1] = ff_hevc_put_hevc_uni_w_epel_h6_8_msa; - c->put_hevc_epel_uni_w[3][0][1] = ff_hevc_put_hevc_uni_w_epel_h8_8_msa; - c->put_hevc_epel_uni_w[4][0][1] = ff_hevc_put_hevc_uni_w_epel_h12_8_msa; - c->put_hevc_epel_uni_w[5][0][1] = ff_hevc_put_hevc_uni_w_epel_h16_8_msa; - c->put_hevc_epel_uni_w[6][0][1] = ff_hevc_put_hevc_uni_w_epel_h24_8_msa; - c->put_hevc_epel_uni_w[7][0][1] = ff_hevc_put_hevc_uni_w_epel_h32_8_msa; - - c->put_hevc_epel_uni_w[1][1][0] = ff_hevc_put_hevc_uni_w_epel_v4_8_msa; - c->put_hevc_epel_uni_w[2][1][0] = ff_hevc_put_hevc_uni_w_epel_v6_8_msa; - c->put_hevc_epel_uni_w[3][1][0] = ff_hevc_put_hevc_uni_w_epel_v8_8_msa; - c->put_hevc_epel_uni_w[4][1][0] = ff_hevc_put_hevc_uni_w_epel_v12_8_msa; - c->put_hevc_epel_uni_w[5][1][0] = ff_hevc_put_hevc_uni_w_epel_v16_8_msa; - c->put_hevc_epel_uni_w[6][1][0] = ff_hevc_put_hevc_uni_w_epel_v24_8_msa; - c->put_hevc_epel_uni_w[7][1][0] = ff_hevc_put_hevc_uni_w_epel_v32_8_msa; - - c->put_hevc_epel_uni_w[1][1][1] = ff_hevc_put_hevc_uni_w_epel_hv4_8_msa; - c->put_hevc_epel_uni_w[2][1][1] = ff_hevc_put_hevc_uni_w_epel_hv6_8_msa; - c->put_hevc_epel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_epel_hv8_8_msa; - c->put_hevc_epel_uni_w[4][1][1] = - ff_hevc_put_hevc_uni_w_epel_hv12_8_msa; - c->put_hevc_epel_uni_w[5][1][1] = - ff_hevc_put_hevc_uni_w_epel_hv16_8_msa; - c->put_hevc_epel_uni_w[6][1][1] = - ff_hevc_put_hevc_uni_w_epel_hv24_8_msa; - c->put_hevc_epel_uni_w[7][1][1] = - ff_hevc_put_hevc_uni_w_epel_hv32_8_msa; - - c->put_hevc_qpel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa; - c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa; - c->put_hevc_qpel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa; - c->put_hevc_qpel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_msa; - c->put_hevc_qpel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_msa; - c->put_hevc_qpel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_msa; - c->put_hevc_qpel_bi[8][0][0] = ff_hevc_put_hevc_bi_pel_pixels48_8_msa; - c->put_hevc_qpel_bi[9][0][0] = ff_hevc_put_hevc_bi_pel_pixels64_8_msa; - - c->put_hevc_qpel_bi[1][0][1] = ff_hevc_put_hevc_bi_qpel_h4_8_msa; - c->put_hevc_qpel_bi[3][0][1] = ff_hevc_put_hevc_bi_qpel_h8_8_msa; - c->put_hevc_qpel_bi[4][0][1] = ff_hevc_put_hevc_bi_qpel_h12_8_msa; - c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_bi_qpel_h16_8_msa; - c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_bi_qpel_h24_8_msa; - c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_bi_qpel_h32_8_msa; - c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_bi_qpel_h48_8_msa; - c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_bi_qpel_h64_8_msa; - - c->put_hevc_qpel_bi[1][1][0] = ff_hevc_put_hevc_bi_qpel_v4_8_msa; - c->put_hevc_qpel_bi[3][1][0] = ff_hevc_put_hevc_bi_qpel_v8_8_msa; - c->put_hevc_qpel_bi[4][1][0] = ff_hevc_put_hevc_bi_qpel_v12_8_msa; - c->put_hevc_qpel_bi[5][1][0] = ff_hevc_put_hevc_bi_qpel_v16_8_msa; - c->put_hevc_qpel_bi[6][1][0] = ff_hevc_put_hevc_bi_qpel_v24_8_msa; - c->put_hevc_qpel_bi[7][1][0] = ff_hevc_put_hevc_bi_qpel_v32_8_msa; - c->put_hevc_qpel_bi[8][1][0] = ff_hevc_put_hevc_bi_qpel_v48_8_msa; - c->put_hevc_qpel_bi[9][1][0] = ff_hevc_put_hevc_bi_qpel_v64_8_msa; - - c->put_hevc_qpel_bi[1][1][1] = ff_hevc_put_hevc_bi_qpel_hv4_8_msa; - c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_bi_qpel_hv8_8_msa; - c->put_hevc_qpel_bi[4][1][1] = ff_hevc_put_hevc_bi_qpel_hv12_8_msa; - c->put_hevc_qpel_bi[5][1][1] = ff_hevc_put_hevc_bi_qpel_hv16_8_msa; - c->put_hevc_qpel_bi[6][1][1] = ff_hevc_put_hevc_bi_qpel_hv24_8_msa; - c->put_hevc_qpel_bi[7][1][1] = ff_hevc_put_hevc_bi_qpel_hv32_8_msa; - c->put_hevc_qpel_bi[8][1][1] = ff_hevc_put_hevc_bi_qpel_hv48_8_msa; - c->put_hevc_qpel_bi[9][1][1] = ff_hevc_put_hevc_bi_qpel_hv64_8_msa; - - c->put_hevc_epel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa; - c->put_hevc_epel_bi[2][0][0] = ff_hevc_put_hevc_bi_pel_pixels6_8_msa; - c->put_hevc_epel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa; - c->put_hevc_epel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa; - c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_msa; - c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_msa; - c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_msa; - - c->put_hevc_epel_bi[1][0][1] = ff_hevc_put_hevc_bi_epel_h4_8_msa; - c->put_hevc_epel_bi[2][0][1] = ff_hevc_put_hevc_bi_epel_h6_8_msa; - c->put_hevc_epel_bi[3][0][1] = ff_hevc_put_hevc_bi_epel_h8_8_msa; - c->put_hevc_epel_bi[4][0][1] = ff_hevc_put_hevc_bi_epel_h12_8_msa; - c->put_hevc_epel_bi[5][0][1] = ff_hevc_put_hevc_bi_epel_h16_8_msa; - c->put_hevc_epel_bi[6][0][1] = ff_hevc_put_hevc_bi_epel_h24_8_msa; - c->put_hevc_epel_bi[7][0][1] = ff_hevc_put_hevc_bi_epel_h32_8_msa; - - c->put_hevc_epel_bi[1][1][0] = ff_hevc_put_hevc_bi_epel_v4_8_msa; - c->put_hevc_epel_bi[2][1][0] = ff_hevc_put_hevc_bi_epel_v6_8_msa; - c->put_hevc_epel_bi[3][1][0] = ff_hevc_put_hevc_bi_epel_v8_8_msa; - c->put_hevc_epel_bi[4][1][0] = ff_hevc_put_hevc_bi_epel_v12_8_msa; - c->put_hevc_epel_bi[5][1][0] = ff_hevc_put_hevc_bi_epel_v16_8_msa; - c->put_hevc_epel_bi[6][1][0] = ff_hevc_put_hevc_bi_epel_v24_8_msa; - c->put_hevc_epel_bi[7][1][0] = ff_hevc_put_hevc_bi_epel_v32_8_msa; - - c->put_hevc_epel_bi[1][1][1] = ff_hevc_put_hevc_bi_epel_hv4_8_msa; - c->put_hevc_epel_bi[2][1][1] = ff_hevc_put_hevc_bi_epel_hv6_8_msa; - c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_bi_epel_hv8_8_msa; - c->put_hevc_epel_bi[4][1][1] = ff_hevc_put_hevc_bi_epel_hv12_8_msa; - c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_bi_epel_hv16_8_msa; - c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_msa; - c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_bi_epel_hv32_8_msa; - - c->put_hevc_qpel_bi_w[1][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels4_8_msa; - c->put_hevc_qpel_bi_w[3][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels8_8_msa; - c->put_hevc_qpel_bi_w[4][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels12_8_msa; - c->put_hevc_qpel_bi_w[5][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels16_8_msa; - c->put_hevc_qpel_bi_w[6][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels24_8_msa; - c->put_hevc_qpel_bi_w[7][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels32_8_msa; - c->put_hevc_qpel_bi_w[8][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels48_8_msa; - c->put_hevc_qpel_bi_w[9][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels64_8_msa; - - c->put_hevc_qpel_bi_w[1][0][1] = ff_hevc_put_hevc_bi_w_qpel_h4_8_msa; - c->put_hevc_qpel_bi_w[3][0][1] = ff_hevc_put_hevc_bi_w_qpel_h8_8_msa; - c->put_hevc_qpel_bi_w[4][0][1] = ff_hevc_put_hevc_bi_w_qpel_h12_8_msa; - c->put_hevc_qpel_bi_w[5][0][1] = ff_hevc_put_hevc_bi_w_qpel_h16_8_msa; - c->put_hevc_qpel_bi_w[6][0][1] = ff_hevc_put_hevc_bi_w_qpel_h24_8_msa; - c->put_hevc_qpel_bi_w[7][0][1] = ff_hevc_put_hevc_bi_w_qpel_h32_8_msa; - c->put_hevc_qpel_bi_w[8][0][1] = ff_hevc_put_hevc_bi_w_qpel_h48_8_msa; - c->put_hevc_qpel_bi_w[9][0][1] = ff_hevc_put_hevc_bi_w_qpel_h64_8_msa; - - c->put_hevc_qpel_bi_w[1][1][0] = ff_hevc_put_hevc_bi_w_qpel_v4_8_msa; - c->put_hevc_qpel_bi_w[3][1][0] = ff_hevc_put_hevc_bi_w_qpel_v8_8_msa; - c->put_hevc_qpel_bi_w[4][1][0] = ff_hevc_put_hevc_bi_w_qpel_v12_8_msa; - c->put_hevc_qpel_bi_w[5][1][0] = ff_hevc_put_hevc_bi_w_qpel_v16_8_msa; - c->put_hevc_qpel_bi_w[6][1][0] = ff_hevc_put_hevc_bi_w_qpel_v24_8_msa; - c->put_hevc_qpel_bi_w[7][1][0] = ff_hevc_put_hevc_bi_w_qpel_v32_8_msa; - c->put_hevc_qpel_bi_w[8][1][0] = ff_hevc_put_hevc_bi_w_qpel_v48_8_msa; - c->put_hevc_qpel_bi_w[9][1][0] = ff_hevc_put_hevc_bi_w_qpel_v64_8_msa; - - c->put_hevc_qpel_bi_w[1][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv4_8_msa; - c->put_hevc_qpel_bi_w[3][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv8_8_msa; - c->put_hevc_qpel_bi_w[4][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv12_8_msa; - c->put_hevc_qpel_bi_w[5][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv16_8_msa; - c->put_hevc_qpel_bi_w[6][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv24_8_msa; - c->put_hevc_qpel_bi_w[7][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv32_8_msa; - c->put_hevc_qpel_bi_w[8][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv48_8_msa; - c->put_hevc_qpel_bi_w[9][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv64_8_msa; - - c->put_hevc_epel_bi_w[1][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels4_8_msa; - c->put_hevc_epel_bi_w[2][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels6_8_msa; - c->put_hevc_epel_bi_w[3][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels8_8_msa; - c->put_hevc_epel_bi_w[4][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels12_8_msa; - c->put_hevc_epel_bi_w[5][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels16_8_msa; - c->put_hevc_epel_bi_w[6][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels24_8_msa; - c->put_hevc_epel_bi_w[7][0][0] = - ff_hevc_put_hevc_bi_w_pel_pixels32_8_msa; - - c->put_hevc_epel_bi_w[1][0][1] = ff_hevc_put_hevc_bi_w_epel_h4_8_msa; - c->put_hevc_epel_bi_w[2][0][1] = ff_hevc_put_hevc_bi_w_epel_h6_8_msa; - c->put_hevc_epel_bi_w[3][0][1] = ff_hevc_put_hevc_bi_w_epel_h8_8_msa; - c->put_hevc_epel_bi_w[4][0][1] = ff_hevc_put_hevc_bi_w_epel_h12_8_msa; - c->put_hevc_epel_bi_w[5][0][1] = ff_hevc_put_hevc_bi_w_epel_h16_8_msa; - c->put_hevc_epel_bi_w[6][0][1] = ff_hevc_put_hevc_bi_w_epel_h24_8_msa; - c->put_hevc_epel_bi_w[7][0][1] = ff_hevc_put_hevc_bi_w_epel_h32_8_msa; - - c->put_hevc_epel_bi_w[1][1][0] = ff_hevc_put_hevc_bi_w_epel_v4_8_msa; - c->put_hevc_epel_bi_w[2][1][0] = ff_hevc_put_hevc_bi_w_epel_v6_8_msa; - c->put_hevc_epel_bi_w[3][1][0] = ff_hevc_put_hevc_bi_w_epel_v8_8_msa; - c->put_hevc_epel_bi_w[4][1][0] = ff_hevc_put_hevc_bi_w_epel_v12_8_msa; - c->put_hevc_epel_bi_w[5][1][0] = ff_hevc_put_hevc_bi_w_epel_v16_8_msa; - c->put_hevc_epel_bi_w[6][1][0] = ff_hevc_put_hevc_bi_w_epel_v24_8_msa; - c->put_hevc_epel_bi_w[7][1][0] = ff_hevc_put_hevc_bi_w_epel_v32_8_msa; - - c->put_hevc_epel_bi_w[1][1][1] = ff_hevc_put_hevc_bi_w_epel_hv4_8_msa; - c->put_hevc_epel_bi_w[2][1][1] = ff_hevc_put_hevc_bi_w_epel_hv6_8_msa; - c->put_hevc_epel_bi_w[3][1][1] = ff_hevc_put_hevc_bi_w_epel_hv8_8_msa; - c->put_hevc_epel_bi_w[4][1][1] = ff_hevc_put_hevc_bi_w_epel_hv12_8_msa; - c->put_hevc_epel_bi_w[5][1][1] = ff_hevc_put_hevc_bi_w_epel_hv16_8_msa; - c->put_hevc_epel_bi_w[6][1][1] = ff_hevc_put_hevc_bi_w_epel_hv24_8_msa; - c->put_hevc_epel_bi_w[7][1][1] = ff_hevc_put_hevc_bi_w_epel_hv32_8_msa; - - c->sao_band_filter[0] = - c->sao_band_filter[1] = - c->sao_band_filter[2] = - c->sao_band_filter[3] = - c->sao_band_filter[4] = ff_hevc_sao_band_filter_0_8_msa; - - c->sao_edge_filter[0] = - c->sao_edge_filter[1] = - c->sao_edge_filter[2] = - c->sao_edge_filter[3] = - c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_8_msa; - - c->hevc_h_loop_filter_luma = ff_hevc_loop_filter_luma_h_8_msa; - c->hevc_v_loop_filter_luma = ff_hevc_loop_filter_luma_v_8_msa; - - c->hevc_h_loop_filter_chroma = ff_hevc_loop_filter_chroma_h_8_msa; - c->hevc_v_loop_filter_chroma = ff_hevc_loop_filter_chroma_v_8_msa; - - c->hevc_h_loop_filter_luma_c = ff_hevc_loop_filter_luma_h_8_msa; - c->hevc_v_loop_filter_luma_c = ff_hevc_loop_filter_luma_v_8_msa; - - c->hevc_h_loop_filter_chroma_c = - ff_hevc_loop_filter_chroma_h_8_msa; - c->hevc_v_loop_filter_chroma_c = - ff_hevc_loop_filter_chroma_v_8_msa; - - c->idct[0] = ff_hevc_idct_4x4_msa; - c->idct[1] = ff_hevc_idct_8x8_msa; - c->idct[2] = ff_hevc_idct_16x16_msa; - c->idct[3] = ff_hevc_idct_32x32_msa; - c->idct_dc[0] = ff_hevc_idct_dc_4x4_msa; - c->idct_dc[1] = ff_hevc_idct_dc_8x8_msa; - c->idct_dc[2] = ff_hevc_idct_dc_16x16_msa; - c->idct_dc[3] = ff_hevc_idct_dc_32x32_msa; - c->add_residual[0] = ff_hevc_addblk_4x4_msa; - c->add_residual[1] = ff_hevc_addblk_8x8_msa; - c->add_residual[2] = ff_hevc_addblk_16x16_msa; - c->add_residual[3] = ff_hevc_addblk_32x32_msa; - c->transform_4x4_luma = ff_hevc_idct_luma_4x4_msa; + if (have_msa(cpu_flags)) { + if (bit_depth == 8) { + c->put_hevc_qpel[1][0][0] = ff_hevc_put_hevc_pel_pixels4_8_msa; + c->put_hevc_qpel[2][0][0] = ff_hevc_put_hevc_pel_pixels6_8_msa; + c->put_hevc_qpel[3][0][0] = ff_hevc_put_hevc_pel_pixels8_8_msa; + c->put_hevc_qpel[4][0][0] = ff_hevc_put_hevc_pel_pixels12_8_msa; + c->put_hevc_qpel[5][0][0] = ff_hevc_put_hevc_pel_pixels16_8_msa; + c->put_hevc_qpel[6][0][0] = ff_hevc_put_hevc_pel_pixels24_8_msa; + c->put_hevc_qpel[7][0][0] = ff_hevc_put_hevc_pel_pixels32_8_msa; + c->put_hevc_qpel[8][0][0] = ff_hevc_put_hevc_pel_pixels48_8_msa; + c->put_hevc_qpel[9][0][0] = ff_hevc_put_hevc_pel_pixels64_8_msa; + + c->put_hevc_qpel[1][0][1] = ff_hevc_put_hevc_qpel_h4_8_msa; + c->put_hevc_qpel[3][0][1] = ff_hevc_put_hevc_qpel_h8_8_msa; + c->put_hevc_qpel[4][0][1] = ff_hevc_put_hevc_qpel_h12_8_msa; + c->put_hevc_qpel[5][0][1] = ff_hevc_put_hevc_qpel_h16_8_msa; + c->put_hevc_qpel[6][0][1] = ff_hevc_put_hevc_qpel_h24_8_msa; + c->put_hevc_qpel[7][0][1] = ff_hevc_put_hevc_qpel_h32_8_msa; + c->put_hevc_qpel[8][0][1] = ff_hevc_put_hevc_qpel_h48_8_msa; + c->put_hevc_qpel[9][0][1] = ff_hevc_put_hevc_qpel_h64_8_msa; + + c->put_hevc_qpel[1][1][0] = ff_hevc_put_hevc_qpel_v4_8_msa; + c->put_hevc_qpel[3][1][0] = ff_hevc_put_hevc_qpel_v8_8_msa; + c->put_hevc_qpel[4][1][0] = ff_hevc_put_hevc_qpel_v12_8_msa; + c->put_hevc_qpel[5][1][0] = ff_hevc_put_hevc_qpel_v16_8_msa; + c->put_hevc_qpel[6][1][0] = ff_hevc_put_hevc_qpel_v24_8_msa; + c->put_hevc_qpel[7][1][0] = ff_hevc_put_hevc_qpel_v32_8_msa; + c->put_hevc_qpel[8][1][0] = ff_hevc_put_hevc_qpel_v48_8_msa; + c->put_hevc_qpel[9][1][0] = ff_hevc_put_hevc_qpel_v64_8_msa; + + c->put_hevc_qpel[1][1][1] = ff_hevc_put_hevc_qpel_hv4_8_msa; + c->put_hevc_qpel[3][1][1] = ff_hevc_put_hevc_qpel_hv8_8_msa; + c->put_hevc_qpel[4][1][1] = ff_hevc_put_hevc_qpel_hv12_8_msa; + c->put_hevc_qpel[5][1][1] = ff_hevc_put_hevc_qpel_hv16_8_msa; + c->put_hevc_qpel[6][1][1] = ff_hevc_put_hevc_qpel_hv24_8_msa; + c->put_hevc_qpel[7][1][1] = ff_hevc_put_hevc_qpel_hv32_8_msa; + c->put_hevc_qpel[8][1][1] = ff_hevc_put_hevc_qpel_hv48_8_msa; + c->put_hevc_qpel[9][1][1] = ff_hevc_put_hevc_qpel_hv64_8_msa; + + c->put_hevc_epel[1][0][0] = ff_hevc_put_hevc_pel_pixels4_8_msa; + c->put_hevc_epel[2][0][0] = ff_hevc_put_hevc_pel_pixels6_8_msa; + c->put_hevc_epel[3][0][0] = ff_hevc_put_hevc_pel_pixels8_8_msa; + c->put_hevc_epel[4][0][0] = ff_hevc_put_hevc_pel_pixels12_8_msa; + c->put_hevc_epel[5][0][0] = ff_hevc_put_hevc_pel_pixels16_8_msa; + c->put_hevc_epel[6][0][0] = ff_hevc_put_hevc_pel_pixels24_8_msa; + c->put_hevc_epel[7][0][0] = ff_hevc_put_hevc_pel_pixels32_8_msa; + + c->put_hevc_epel[1][0][1] = ff_hevc_put_hevc_epel_h4_8_msa; + c->put_hevc_epel[2][0][1] = ff_hevc_put_hevc_epel_h6_8_msa; + c->put_hevc_epel[3][0][1] = ff_hevc_put_hevc_epel_h8_8_msa; + c->put_hevc_epel[4][0][1] = ff_hevc_put_hevc_epel_h12_8_msa; + c->put_hevc_epel[5][0][1] = ff_hevc_put_hevc_epel_h16_8_msa; + c->put_hevc_epel[6][0][1] = ff_hevc_put_hevc_epel_h24_8_msa; + c->put_hevc_epel[7][0][1] = ff_hevc_put_hevc_epel_h32_8_msa; + + c->put_hevc_epel[1][1][0] = ff_hevc_put_hevc_epel_v4_8_msa; + c->put_hevc_epel[2][1][0] = ff_hevc_put_hevc_epel_v6_8_msa; + c->put_hevc_epel[3][1][0] = ff_hevc_put_hevc_epel_v8_8_msa; + c->put_hevc_epel[4][1][0] = ff_hevc_put_hevc_epel_v12_8_msa; + c->put_hevc_epel[5][1][0] = ff_hevc_put_hevc_epel_v16_8_msa; + c->put_hevc_epel[6][1][0] = ff_hevc_put_hevc_epel_v24_8_msa; + c->put_hevc_epel[7][1][0] = ff_hevc_put_hevc_epel_v32_8_msa; + + c->put_hevc_epel[1][1][1] = ff_hevc_put_hevc_epel_hv4_8_msa; + c->put_hevc_epel[2][1][1] = ff_hevc_put_hevc_epel_hv6_8_msa; + c->put_hevc_epel[3][1][1] = ff_hevc_put_hevc_epel_hv8_8_msa; + c->put_hevc_epel[4][1][1] = ff_hevc_put_hevc_epel_hv12_8_msa; + c->put_hevc_epel[5][1][1] = ff_hevc_put_hevc_epel_hv16_8_msa; + c->put_hevc_epel[6][1][1] = ff_hevc_put_hevc_epel_hv24_8_msa; + c->put_hevc_epel[7][1][1] = ff_hevc_put_hevc_epel_hv32_8_msa; + + c->put_hevc_qpel_uni[3][0][0] = ff_hevc_put_hevc_uni_pel_pixels8_8_msa; + c->put_hevc_qpel_uni[4][0][0] = ff_hevc_put_hevc_uni_pel_pixels12_8_msa; + c->put_hevc_qpel_uni[5][0][0] = ff_hevc_put_hevc_uni_pel_pixels16_8_msa; + c->put_hevc_qpel_uni[6][0][0] = ff_hevc_put_hevc_uni_pel_pixels24_8_msa; + c->put_hevc_qpel_uni[7][0][0] = ff_hevc_put_hevc_uni_pel_pixels32_8_msa; + c->put_hevc_qpel_uni[8][0][0] = ff_hevc_put_hevc_uni_pel_pixels48_8_msa; + c->put_hevc_qpel_uni[9][0][0] = ff_hevc_put_hevc_uni_pel_pixels64_8_msa; + + c->put_hevc_qpel_uni[1][0][1] = ff_hevc_put_hevc_uni_qpel_h4_8_msa; + c->put_hevc_qpel_uni[3][0][1] = ff_hevc_put_hevc_uni_qpel_h8_8_msa; + c->put_hevc_qpel_uni[4][0][1] = ff_hevc_put_hevc_uni_qpel_h12_8_msa; + c->put_hevc_qpel_uni[5][0][1] = ff_hevc_put_hevc_uni_qpel_h16_8_msa; + c->put_hevc_qpel_uni[6][0][1] = ff_hevc_put_hevc_uni_qpel_h24_8_msa; + c->put_hevc_qpel_uni[7][0][1] = ff_hevc_put_hevc_uni_qpel_h32_8_msa; + c->put_hevc_qpel_uni[8][0][1] = ff_hevc_put_hevc_uni_qpel_h48_8_msa; + c->put_hevc_qpel_uni[9][0][1] = ff_hevc_put_hevc_uni_qpel_h64_8_msa; + + c->put_hevc_qpel_uni[1][1][0] = ff_hevc_put_hevc_uni_qpel_v4_8_msa; + c->put_hevc_qpel_uni[3][1][0] = ff_hevc_put_hevc_uni_qpel_v8_8_msa; + c->put_hevc_qpel_uni[4][1][0] = ff_hevc_put_hevc_uni_qpel_v12_8_msa; + c->put_hevc_qpel_uni[5][1][0] = ff_hevc_put_hevc_uni_qpel_v16_8_msa; + c->put_hevc_qpel_uni[6][1][0] = ff_hevc_put_hevc_uni_qpel_v24_8_msa; + c->put_hevc_qpel_uni[7][1][0] = ff_hevc_put_hevc_uni_qpel_v32_8_msa; + c->put_hevc_qpel_uni[8][1][0] = ff_hevc_put_hevc_uni_qpel_v48_8_msa; + c->put_hevc_qpel_uni[9][1][0] = ff_hevc_put_hevc_uni_qpel_v64_8_msa; + + c->put_hevc_qpel_uni[1][1][1] = ff_hevc_put_hevc_uni_qpel_hv4_8_msa; + c->put_hevc_qpel_uni[3][1][1] = ff_hevc_put_hevc_uni_qpel_hv8_8_msa; + c->put_hevc_qpel_uni[4][1][1] = ff_hevc_put_hevc_uni_qpel_hv12_8_msa; + c->put_hevc_qpel_uni[5][1][1] = ff_hevc_put_hevc_uni_qpel_hv16_8_msa; + c->put_hevc_qpel_uni[6][1][1] = ff_hevc_put_hevc_uni_qpel_hv24_8_msa; + c->put_hevc_qpel_uni[7][1][1] = ff_hevc_put_hevc_uni_qpel_hv32_8_msa; + c->put_hevc_qpel_uni[8][1][1] = ff_hevc_put_hevc_uni_qpel_hv48_8_msa; + c->put_hevc_qpel_uni[9][1][1] = ff_hevc_put_hevc_uni_qpel_hv64_8_msa; + + c->put_hevc_epel_uni[3][0][0] = ff_hevc_put_hevc_uni_pel_pixels8_8_msa; + c->put_hevc_epel_uni[4][0][0] = ff_hevc_put_hevc_uni_pel_pixels12_8_msa; + c->put_hevc_epel_uni[5][0][0] = ff_hevc_put_hevc_uni_pel_pixels16_8_msa; + c->put_hevc_epel_uni[6][0][0] = ff_hevc_put_hevc_uni_pel_pixels24_8_msa; + c->put_hevc_epel_uni[7][0][0] = ff_hevc_put_hevc_uni_pel_pixels32_8_msa; + + c->put_hevc_epel_uni[1][0][1] = ff_hevc_put_hevc_uni_epel_h4_8_msa; + c->put_hevc_epel_uni[2][0][1] = ff_hevc_put_hevc_uni_epel_h6_8_msa; + c->put_hevc_epel_uni[3][0][1] = ff_hevc_put_hevc_uni_epel_h8_8_msa; + c->put_hevc_epel_uni[4][0][1] = ff_hevc_put_hevc_uni_epel_h12_8_msa; + c->put_hevc_epel_uni[5][0][1] = ff_hevc_put_hevc_uni_epel_h16_8_msa; + c->put_hevc_epel_uni[6][0][1] = ff_hevc_put_hevc_uni_epel_h24_8_msa; + c->put_hevc_epel_uni[7][0][1] = ff_hevc_put_hevc_uni_epel_h32_8_msa; + + c->put_hevc_epel_uni[1][1][0] = ff_hevc_put_hevc_uni_epel_v4_8_msa; + c->put_hevc_epel_uni[2][1][0] = ff_hevc_put_hevc_uni_epel_v6_8_msa; + c->put_hevc_epel_uni[3][1][0] = ff_hevc_put_hevc_uni_epel_v8_8_msa; + c->put_hevc_epel_uni[4][1][0] = ff_hevc_put_hevc_uni_epel_v12_8_msa; + c->put_hevc_epel_uni[5][1][0] = ff_hevc_put_hevc_uni_epel_v16_8_msa; + c->put_hevc_epel_uni[6][1][0] = ff_hevc_put_hevc_uni_epel_v24_8_msa; + c->put_hevc_epel_uni[7][1][0] = ff_hevc_put_hevc_uni_epel_v32_8_msa; + + c->put_hevc_epel_uni[1][1][1] = ff_hevc_put_hevc_uni_epel_hv4_8_msa; + c->put_hevc_epel_uni[2][1][1] = ff_hevc_put_hevc_uni_epel_hv6_8_msa; + c->put_hevc_epel_uni[3][1][1] = ff_hevc_put_hevc_uni_epel_hv8_8_msa; + c->put_hevc_epel_uni[4][1][1] = ff_hevc_put_hevc_uni_epel_hv12_8_msa; + c->put_hevc_epel_uni[5][1][1] = ff_hevc_put_hevc_uni_epel_hv16_8_msa; + c->put_hevc_epel_uni[6][1][1] = ff_hevc_put_hevc_uni_epel_hv24_8_msa; + c->put_hevc_epel_uni[7][1][1] = ff_hevc_put_hevc_uni_epel_hv32_8_msa; + + c->put_hevc_qpel_uni_w[1][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels4_8_msa; + c->put_hevc_qpel_uni_w[3][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels8_8_msa; + c->put_hevc_qpel_uni_w[4][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels12_8_msa; + c->put_hevc_qpel_uni_w[5][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels16_8_msa; + c->put_hevc_qpel_uni_w[6][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels24_8_msa; + c->put_hevc_qpel_uni_w[7][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels32_8_msa; + c->put_hevc_qpel_uni_w[8][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels48_8_msa; + c->put_hevc_qpel_uni_w[9][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels64_8_msa; + + c->put_hevc_qpel_uni_w[1][0][1] = ff_hevc_put_hevc_uni_w_qpel_h4_8_msa; + c->put_hevc_qpel_uni_w[3][0][1] = ff_hevc_put_hevc_uni_w_qpel_h8_8_msa; + c->put_hevc_qpel_uni_w[4][0][1] = ff_hevc_put_hevc_uni_w_qpel_h12_8_msa; + c->put_hevc_qpel_uni_w[5][0][1] = ff_hevc_put_hevc_uni_w_qpel_h16_8_msa; + c->put_hevc_qpel_uni_w[6][0][1] = ff_hevc_put_hevc_uni_w_qpel_h24_8_msa; + c->put_hevc_qpel_uni_w[7][0][1] = ff_hevc_put_hevc_uni_w_qpel_h32_8_msa; + c->put_hevc_qpel_uni_w[8][0][1] = ff_hevc_put_hevc_uni_w_qpel_h48_8_msa; + c->put_hevc_qpel_uni_w[9][0][1] = ff_hevc_put_hevc_uni_w_qpel_h64_8_msa; + + c->put_hevc_qpel_uni_w[1][1][0] = ff_hevc_put_hevc_uni_w_qpel_v4_8_msa; + c->put_hevc_qpel_uni_w[3][1][0] = ff_hevc_put_hevc_uni_w_qpel_v8_8_msa; + c->put_hevc_qpel_uni_w[4][1][0] = ff_hevc_put_hevc_uni_w_qpel_v12_8_msa; + c->put_hevc_qpel_uni_w[5][1][0] = ff_hevc_put_hevc_uni_w_qpel_v16_8_msa; + c->put_hevc_qpel_uni_w[6][1][0] = ff_hevc_put_hevc_uni_w_qpel_v24_8_msa; + c->put_hevc_qpel_uni_w[7][1][0] = ff_hevc_put_hevc_uni_w_qpel_v32_8_msa; + c->put_hevc_qpel_uni_w[8][1][0] = ff_hevc_put_hevc_uni_w_qpel_v48_8_msa; + c->put_hevc_qpel_uni_w[9][1][0] = ff_hevc_put_hevc_uni_w_qpel_v64_8_msa; + + c->put_hevc_qpel_uni_w[1][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv4_8_msa; + c->put_hevc_qpel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_qpel_hv8_8_msa; + c->put_hevc_qpel_uni_w[4][1][1] = + ff_hevc_put_hevc_uni_w_qpel_hv12_8_msa; + c->put_hevc_qpel_uni_w[5][1][1] = + ff_hevc_put_hevc_uni_w_qpel_hv16_8_msa; + c->put_hevc_qpel_uni_w[6][1][1] = + ff_hevc_put_hevc_uni_w_qpel_hv24_8_msa; + c->put_hevc_qpel_uni_w[7][1][1] = + ff_hevc_put_hevc_uni_w_qpel_hv32_8_msa; + c->put_hevc_qpel_uni_w[8][1][1] = + ff_hevc_put_hevc_uni_w_qpel_hv48_8_msa; + c->put_hevc_qpel_uni_w[9][1][1] = + ff_hevc_put_hevc_uni_w_qpel_hv64_8_msa; + + c->put_hevc_epel_uni_w[1][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels4_8_msa; + c->put_hevc_epel_uni_w[2][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels6_8_msa; + c->put_hevc_epel_uni_w[3][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels8_8_msa; + c->put_hevc_epel_uni_w[4][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels12_8_msa; + c->put_hevc_epel_uni_w[5][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels16_8_msa; + c->put_hevc_epel_uni_w[6][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels24_8_msa; + c->put_hevc_epel_uni_w[7][0][0] = + ff_hevc_put_hevc_uni_w_pel_pixels32_8_msa; + + c->put_hevc_epel_uni_w[1][0][1] = ff_hevc_put_hevc_uni_w_epel_h4_8_msa; + c->put_hevc_epel_uni_w[2][0][1] = ff_hevc_put_hevc_uni_w_epel_h6_8_msa; + c->put_hevc_epel_uni_w[3][0][1] = ff_hevc_put_hevc_uni_w_epel_h8_8_msa; + c->put_hevc_epel_uni_w[4][0][1] = ff_hevc_put_hevc_uni_w_epel_h12_8_msa; + c->put_hevc_epel_uni_w[5][0][1] = ff_hevc_put_hevc_uni_w_epel_h16_8_msa; + c->put_hevc_epel_uni_w[6][0][1] = ff_hevc_put_hevc_uni_w_epel_h24_8_msa; + c->put_hevc_epel_uni_w[7][0][1] = ff_hevc_put_hevc_uni_w_epel_h32_8_msa; + + c->put_hevc_epel_uni_w[1][1][0] = ff_hevc_put_hevc_uni_w_epel_v4_8_msa; + c->put_hevc_epel_uni_w[2][1][0] = ff_hevc_put_hevc_uni_w_epel_v6_8_msa; + c->put_hevc_epel_uni_w[3][1][0] = ff_hevc_put_hevc_uni_w_epel_v8_8_msa; + c->put_hevc_epel_uni_w[4][1][0] = ff_hevc_put_hevc_uni_w_epel_v12_8_msa; + c->put_hevc_epel_uni_w[5][1][0] = ff_hevc_put_hevc_uni_w_epel_v16_8_msa; + c->put_hevc_epel_uni_w[6][1][0] = ff_hevc_put_hevc_uni_w_epel_v24_8_msa; + c->put_hevc_epel_uni_w[7][1][0] = ff_hevc_put_hevc_uni_w_epel_v32_8_msa; + + c->put_hevc_epel_uni_w[1][1][1] = ff_hevc_put_hevc_uni_w_epel_hv4_8_msa; + c->put_hevc_epel_uni_w[2][1][1] = ff_hevc_put_hevc_uni_w_epel_hv6_8_msa; + c->put_hevc_epel_uni_w[3][1][1] = ff_hevc_put_hevc_uni_w_epel_hv8_8_msa; + c->put_hevc_epel_uni_w[4][1][1] = + ff_hevc_put_hevc_uni_w_epel_hv12_8_msa; + c->put_hevc_epel_uni_w[5][1][1] = + ff_hevc_put_hevc_uni_w_epel_hv16_8_msa; + c->put_hevc_epel_uni_w[6][1][1] = + ff_hevc_put_hevc_uni_w_epel_hv24_8_msa; + c->put_hevc_epel_uni_w[7][1][1] = + ff_hevc_put_hevc_uni_w_epel_hv32_8_msa; + + c->put_hevc_qpel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa; + c->put_hevc_qpel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa; + c->put_hevc_qpel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa; + c->put_hevc_qpel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_msa; + c->put_hevc_qpel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_msa; + c->put_hevc_qpel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_msa; + c->put_hevc_qpel_bi[8][0][0] = ff_hevc_put_hevc_bi_pel_pixels48_8_msa; + c->put_hevc_qpel_bi[9][0][0] = ff_hevc_put_hevc_bi_pel_pixels64_8_msa; + + c->put_hevc_qpel_bi[1][0][1] = ff_hevc_put_hevc_bi_qpel_h4_8_msa; + c->put_hevc_qpel_bi[3][0][1] = ff_hevc_put_hevc_bi_qpel_h8_8_msa; + c->put_hevc_qpel_bi[4][0][1] = ff_hevc_put_hevc_bi_qpel_h12_8_msa; + c->put_hevc_qpel_bi[5][0][1] = ff_hevc_put_hevc_bi_qpel_h16_8_msa; + c->put_hevc_qpel_bi[6][0][1] = ff_hevc_put_hevc_bi_qpel_h24_8_msa; + c->put_hevc_qpel_bi[7][0][1] = ff_hevc_put_hevc_bi_qpel_h32_8_msa; + c->put_hevc_qpel_bi[8][0][1] = ff_hevc_put_hevc_bi_qpel_h48_8_msa; + c->put_hevc_qpel_bi[9][0][1] = ff_hevc_put_hevc_bi_qpel_h64_8_msa; + + c->put_hevc_qpel_bi[1][1][0] = ff_hevc_put_hevc_bi_qpel_v4_8_msa; + c->put_hevc_qpel_bi[3][1][0] = ff_hevc_put_hevc_bi_qpel_v8_8_msa; + c->put_hevc_qpel_bi[4][1][0] = ff_hevc_put_hevc_bi_qpel_v12_8_msa; + c->put_hevc_qpel_bi[5][1][0] = ff_hevc_put_hevc_bi_qpel_v16_8_msa; + c->put_hevc_qpel_bi[6][1][0] = ff_hevc_put_hevc_bi_qpel_v24_8_msa; + c->put_hevc_qpel_bi[7][1][0] = ff_hevc_put_hevc_bi_qpel_v32_8_msa; + c->put_hevc_qpel_bi[8][1][0] = ff_hevc_put_hevc_bi_qpel_v48_8_msa; + c->put_hevc_qpel_bi[9][1][0] = ff_hevc_put_hevc_bi_qpel_v64_8_msa; + + c->put_hevc_qpel_bi[1][1][1] = ff_hevc_put_hevc_bi_qpel_hv4_8_msa; + c->put_hevc_qpel_bi[3][1][1] = ff_hevc_put_hevc_bi_qpel_hv8_8_msa; + c->put_hevc_qpel_bi[4][1][1] = ff_hevc_put_hevc_bi_qpel_hv12_8_msa; + c->put_hevc_qpel_bi[5][1][1] = ff_hevc_put_hevc_bi_qpel_hv16_8_msa; + c->put_hevc_qpel_bi[6][1][1] = ff_hevc_put_hevc_bi_qpel_hv24_8_msa; + c->put_hevc_qpel_bi[7][1][1] = ff_hevc_put_hevc_bi_qpel_hv32_8_msa; + c->put_hevc_qpel_bi[8][1][1] = ff_hevc_put_hevc_bi_qpel_hv48_8_msa; + c->put_hevc_qpel_bi[9][1][1] = ff_hevc_put_hevc_bi_qpel_hv64_8_msa; + + c->put_hevc_epel_bi[1][0][0] = ff_hevc_put_hevc_bi_pel_pixels4_8_msa; + c->put_hevc_epel_bi[2][0][0] = ff_hevc_put_hevc_bi_pel_pixels6_8_msa; + c->put_hevc_epel_bi[3][0][0] = ff_hevc_put_hevc_bi_pel_pixels8_8_msa; + c->put_hevc_epel_bi[4][0][0] = ff_hevc_put_hevc_bi_pel_pixels12_8_msa; + c->put_hevc_epel_bi[5][0][0] = ff_hevc_put_hevc_bi_pel_pixels16_8_msa; + c->put_hevc_epel_bi[6][0][0] = ff_hevc_put_hevc_bi_pel_pixels24_8_msa; + c->put_hevc_epel_bi[7][0][0] = ff_hevc_put_hevc_bi_pel_pixels32_8_msa; + + c->put_hevc_epel_bi[1][0][1] = ff_hevc_put_hevc_bi_epel_h4_8_msa; + c->put_hevc_epel_bi[2][0][1] = ff_hevc_put_hevc_bi_epel_h6_8_msa; + c->put_hevc_epel_bi[3][0][1] = ff_hevc_put_hevc_bi_epel_h8_8_msa; + c->put_hevc_epel_bi[4][0][1] = ff_hevc_put_hevc_bi_epel_h12_8_msa; + c->put_hevc_epel_bi[5][0][1] = ff_hevc_put_hevc_bi_epel_h16_8_msa; + c->put_hevc_epel_bi[6][0][1] = ff_hevc_put_hevc_bi_epel_h24_8_msa; + c->put_hevc_epel_bi[7][0][1] = ff_hevc_put_hevc_bi_epel_h32_8_msa; + + c->put_hevc_epel_bi[1][1][0] = ff_hevc_put_hevc_bi_epel_v4_8_msa; + c->put_hevc_epel_bi[2][1][0] = ff_hevc_put_hevc_bi_epel_v6_8_msa; + c->put_hevc_epel_bi[3][1][0] = ff_hevc_put_hevc_bi_epel_v8_8_msa; + c->put_hevc_epel_bi[4][1][0] = ff_hevc_put_hevc_bi_epel_v12_8_msa; + c->put_hevc_epel_bi[5][1][0] = ff_hevc_put_hevc_bi_epel_v16_8_msa; + c->put_hevc_epel_bi[6][1][0] = ff_hevc_put_hevc_bi_epel_v24_8_msa; + c->put_hevc_epel_bi[7][1][0] = ff_hevc_put_hevc_bi_epel_v32_8_msa; + + c->put_hevc_epel_bi[1][1][1] = ff_hevc_put_hevc_bi_epel_hv4_8_msa; + c->put_hevc_epel_bi[2][1][1] = ff_hevc_put_hevc_bi_epel_hv6_8_msa; + c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_bi_epel_hv8_8_msa; + c->put_hevc_epel_bi[4][1][1] = ff_hevc_put_hevc_bi_epel_hv12_8_msa; + c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_bi_epel_hv16_8_msa; + c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_msa; + c->put_hevc_epel_bi[7][1][1] = ff_hevc_put_hevc_bi_epel_hv32_8_msa; + + c->put_hevc_qpel_bi_w[1][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels4_8_msa; + c->put_hevc_qpel_bi_w[3][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels8_8_msa; + c->put_hevc_qpel_bi_w[4][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels12_8_msa; + c->put_hevc_qpel_bi_w[5][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels16_8_msa; + c->put_hevc_qpel_bi_w[6][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels24_8_msa; + c->put_hevc_qpel_bi_w[7][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels32_8_msa; + c->put_hevc_qpel_bi_w[8][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels48_8_msa; + c->put_hevc_qpel_bi_w[9][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels64_8_msa; + + c->put_hevc_qpel_bi_w[1][0][1] = ff_hevc_put_hevc_bi_w_qpel_h4_8_msa; + c->put_hevc_qpel_bi_w[3][0][1] = ff_hevc_put_hevc_bi_w_qpel_h8_8_msa; + c->put_hevc_qpel_bi_w[4][0][1] = ff_hevc_put_hevc_bi_w_qpel_h12_8_msa; + c->put_hevc_qpel_bi_w[5][0][1] = ff_hevc_put_hevc_bi_w_qpel_h16_8_msa; + c->put_hevc_qpel_bi_w[6][0][1] = ff_hevc_put_hevc_bi_w_qpel_h24_8_msa; + c->put_hevc_qpel_bi_w[7][0][1] = ff_hevc_put_hevc_bi_w_qpel_h32_8_msa; + c->put_hevc_qpel_bi_w[8][0][1] = ff_hevc_put_hevc_bi_w_qpel_h48_8_msa; + c->put_hevc_qpel_bi_w[9][0][1] = ff_hevc_put_hevc_bi_w_qpel_h64_8_msa; + + c->put_hevc_qpel_bi_w[1][1][0] = ff_hevc_put_hevc_bi_w_qpel_v4_8_msa; + c->put_hevc_qpel_bi_w[3][1][0] = ff_hevc_put_hevc_bi_w_qpel_v8_8_msa; + c->put_hevc_qpel_bi_w[4][1][0] = ff_hevc_put_hevc_bi_w_qpel_v12_8_msa; + c->put_hevc_qpel_bi_w[5][1][0] = ff_hevc_put_hevc_bi_w_qpel_v16_8_msa; + c->put_hevc_qpel_bi_w[6][1][0] = ff_hevc_put_hevc_bi_w_qpel_v24_8_msa; + c->put_hevc_qpel_bi_w[7][1][0] = ff_hevc_put_hevc_bi_w_qpel_v32_8_msa; + c->put_hevc_qpel_bi_w[8][1][0] = ff_hevc_put_hevc_bi_w_qpel_v48_8_msa; + c->put_hevc_qpel_bi_w[9][1][0] = ff_hevc_put_hevc_bi_w_qpel_v64_8_msa; + + c->put_hevc_qpel_bi_w[1][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv4_8_msa; + c->put_hevc_qpel_bi_w[3][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv8_8_msa; + c->put_hevc_qpel_bi_w[4][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv12_8_msa; + c->put_hevc_qpel_bi_w[5][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv16_8_msa; + c->put_hevc_qpel_bi_w[6][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv24_8_msa; + c->put_hevc_qpel_bi_w[7][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv32_8_msa; + c->put_hevc_qpel_bi_w[8][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv48_8_msa; + c->put_hevc_qpel_bi_w[9][1][1] = ff_hevc_put_hevc_bi_w_qpel_hv64_8_msa; + + c->put_hevc_epel_bi_w[1][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels4_8_msa; + c->put_hevc_epel_bi_w[2][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels6_8_msa; + c->put_hevc_epel_bi_w[3][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels8_8_msa; + c->put_hevc_epel_bi_w[4][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels12_8_msa; + c->put_hevc_epel_bi_w[5][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels16_8_msa; + c->put_hevc_epel_bi_w[6][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels24_8_msa; + c->put_hevc_epel_bi_w[7][0][0] = + ff_hevc_put_hevc_bi_w_pel_pixels32_8_msa; + + c->put_hevc_epel_bi_w[1][0][1] = ff_hevc_put_hevc_bi_w_epel_h4_8_msa; + c->put_hevc_epel_bi_w[2][0][1] = ff_hevc_put_hevc_bi_w_epel_h6_8_msa; + c->put_hevc_epel_bi_w[3][0][1] = ff_hevc_put_hevc_bi_w_epel_h8_8_msa; + c->put_hevc_epel_bi_w[4][0][1] = ff_hevc_put_hevc_bi_w_epel_h12_8_msa; + c->put_hevc_epel_bi_w[5][0][1] = ff_hevc_put_hevc_bi_w_epel_h16_8_msa; + c->put_hevc_epel_bi_w[6][0][1] = ff_hevc_put_hevc_bi_w_epel_h24_8_msa; + c->put_hevc_epel_bi_w[7][0][1] = ff_hevc_put_hevc_bi_w_epel_h32_8_msa; + + c->put_hevc_epel_bi_w[1][1][0] = ff_hevc_put_hevc_bi_w_epel_v4_8_msa; + c->put_hevc_epel_bi_w[2][1][0] = ff_hevc_put_hevc_bi_w_epel_v6_8_msa; + c->put_hevc_epel_bi_w[3][1][0] = ff_hevc_put_hevc_bi_w_epel_v8_8_msa; + c->put_hevc_epel_bi_w[4][1][0] = ff_hevc_put_hevc_bi_w_epel_v12_8_msa; + c->put_hevc_epel_bi_w[5][1][0] = ff_hevc_put_hevc_bi_w_epel_v16_8_msa; + c->put_hevc_epel_bi_w[6][1][0] = ff_hevc_put_hevc_bi_w_epel_v24_8_msa; + c->put_hevc_epel_bi_w[7][1][0] = ff_hevc_put_hevc_bi_w_epel_v32_8_msa; + + c->put_hevc_epel_bi_w[1][1][1] = ff_hevc_put_hevc_bi_w_epel_hv4_8_msa; + c->put_hevc_epel_bi_w[2][1][1] = ff_hevc_put_hevc_bi_w_epel_hv6_8_msa; + c->put_hevc_epel_bi_w[3][1][1] = ff_hevc_put_hevc_bi_w_epel_hv8_8_msa; + c->put_hevc_epel_bi_w[4][1][1] = ff_hevc_put_hevc_bi_w_epel_hv12_8_msa; + c->put_hevc_epel_bi_w[5][1][1] = ff_hevc_put_hevc_bi_w_epel_hv16_8_msa; + c->put_hevc_epel_bi_w[6][1][1] = ff_hevc_put_hevc_bi_w_epel_hv24_8_msa; + c->put_hevc_epel_bi_w[7][1][1] = ff_hevc_put_hevc_bi_w_epel_hv32_8_msa; + + c->sao_band_filter[0] = + c->sao_band_filter[1] = + c->sao_band_filter[2] = + c->sao_band_filter[3] = + c->sao_band_filter[4] = ff_hevc_sao_band_filter_0_8_msa; + + c->sao_edge_filter[0] = + c->sao_edge_filter[1] = + c->sao_edge_filter[2] = + c->sao_edge_filter[3] = + c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_8_msa; + + c->hevc_h_loop_filter_luma = ff_hevc_loop_filter_luma_h_8_msa; + c->hevc_v_loop_filter_luma = ff_hevc_loop_filter_luma_v_8_msa; + + c->hevc_h_loop_filter_chroma = ff_hevc_loop_filter_chroma_h_8_msa; + c->hevc_v_loop_filter_chroma = ff_hevc_loop_filter_chroma_v_8_msa; + + c->hevc_h_loop_filter_luma_c = ff_hevc_loop_filter_luma_h_8_msa; + c->hevc_v_loop_filter_luma_c = ff_hevc_loop_filter_luma_v_8_msa; + + c->hevc_h_loop_filter_chroma_c = + ff_hevc_loop_filter_chroma_h_8_msa; + c->hevc_v_loop_filter_chroma_c = + ff_hevc_loop_filter_chroma_v_8_msa; + + c->idct[0] = ff_hevc_idct_4x4_msa; + c->idct[1] = ff_hevc_idct_8x8_msa; + c->idct[2] = ff_hevc_idct_16x16_msa; + c->idct[3] = ff_hevc_idct_32x32_msa; + c->idct_dc[0] = ff_hevc_idct_dc_4x4_msa; + c->idct_dc[1] = ff_hevc_idct_dc_8x8_msa; + c->idct_dc[2] = ff_hevc_idct_dc_16x16_msa; + c->idct_dc[3] = ff_hevc_idct_dc_32x32_msa; + c->add_residual[0] = ff_hevc_addblk_4x4_msa; + c->add_residual[1] = ff_hevc_addblk_8x8_msa; + c->add_residual[2] = ff_hevc_addblk_16x16_msa; + c->add_residual[3] = ff_hevc_addblk_32x32_msa; + c->transform_4x4_luma = ff_hevc_idct_luma_4x4_msa; + } } } -#endif // #if HAVE_MSA - -void ff_hevc_dsp_init_mips(HEVCDSPContext *c, const int bit_depth) -{ -#if HAVE_MMI - hevc_dsp_init_mmi(c, bit_depth); -#endif // #if HAVE_MMI -#if HAVE_MSA - hevc_dsp_init_msa(c, bit_depth); -#endif // #if HAVE_MSA -} diff --git a/libavcodec/mips/hevcpred_init_mips.c b/libavcodec/mips/hevcpred_init_mips.c index e987698d66..f7ecb34dcc 100644 --- a/libavcodec/mips/hevcpred_init_mips.c +++ b/libavcodec/mips/hevcpred_init_mips.c @@ -18,32 +18,28 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "config.h" #include "libavutil/attributes.h" #include "libavcodec/mips/hevcpred_mips.h" -#if HAVE_MSA -static av_cold void hevc_pred_init_msa(HEVCPredContext *c, const int bit_depth) -{ - if (8 == bit_depth) { - c->intra_pred[2] = ff_intra_pred_8_16x16_msa; - c->intra_pred[3] = ff_intra_pred_8_32x32_msa; - c->pred_planar[0] = ff_hevc_intra_pred_planar_0_msa; - c->pred_planar[1] = ff_hevc_intra_pred_planar_1_msa; - c->pred_planar[2] = ff_hevc_intra_pred_planar_2_msa; - c->pred_planar[3] = ff_hevc_intra_pred_planar_3_msa; - c->pred_dc = ff_hevc_intra_pred_dc_msa; - c->pred_angular[0] = ff_pred_intra_pred_angular_0_msa; - c->pred_angular[1] = ff_pred_intra_pred_angular_1_msa; - c->pred_angular[2] = ff_pred_intra_pred_angular_2_msa; - c->pred_angular[3] = ff_pred_intra_pred_angular_3_msa; - } -} -#endif // #if HAVE_MSA - void ff_hevc_pred_init_mips(HEVCPredContext *c, const int bit_depth) { -#if HAVE_MSA - hevc_pred_init_msa(c, bit_depth); -#endif // #if HAVE_MSA + int cpu_flags = av_get_cpu_flags(); + + if (have_msa(cpu_flags)) { + if (bit_depth == 8) { + c->intra_pred[2] = ff_intra_pred_8_16x16_msa; + c->intra_pred[3] = ff_intra_pred_8_32x32_msa; + c->pred_planar[0] = ff_hevc_intra_pred_planar_0_msa; + c->pred_planar[1] = ff_hevc_intra_pred_planar_1_msa; + c->pred_planar[2] = ff_hevc_intra_pred_planar_2_msa; + c->pred_planar[3] = ff_hevc_intra_pred_planar_3_msa; + c->pred_dc = ff_hevc_intra_pred_dc_msa; + c->pred_angular[0] = ff_pred_intra_pred_angular_0_msa; + c->pred_angular[1] = ff_pred_intra_pred_angular_1_msa; + c->pred_angular[2] = ff_pred_intra_pred_angular_2_msa; + c->pred_angular[3] = ff_pred_intra_pred_angular_3_msa; + } + } } diff --git a/libavcodec/mips/hpeldsp_init_mips.c b/libavcodec/mips/hpeldsp_init_mips.c index d6f7a9793d..77cbe99fa4 100644 --- a/libavcodec/mips/hpeldsp_init_mips.c +++ b/libavcodec/mips/hpeldsp_init_mips.c @@ -19,104 +19,94 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "../hpeldsp.h" #include "libavcodec/mips/hpeldsp_mips.h" -#if HAVE_MSA -static void ff_hpeldsp_init_msa(HpelDSPContext *c, int flags) -{ - c->put_pixels_tab[0][0] = ff_put_pixels16_msa; - c->put_pixels_tab[0][1] = ff_put_pixels16_x2_msa; - c->put_pixels_tab[0][2] = ff_put_pixels16_y2_msa; - c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_msa; - - c->put_pixels_tab[1][0] = ff_put_pixels8_msa; - c->put_pixels_tab[1][1] = ff_put_pixels8_x2_msa; - c->put_pixels_tab[1][2] = ff_put_pixels8_y2_msa; - c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_msa; - - c->put_pixels_tab[2][1] = ff_put_pixels4_x2_msa; - c->put_pixels_tab[2][2] = ff_put_pixels4_y2_msa; - c->put_pixels_tab[2][3] = ff_put_pixels4_xy2_msa; - - c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_msa; - c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_msa; - c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_msa; - c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_msa; - - c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_msa; - c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_msa; - c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_msa; - c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_msa; - - c->avg_pixels_tab[0][0] = ff_avg_pixels16_msa; - c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_msa; - c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_msa; - c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_msa; - - c->avg_pixels_tab[1][0] = ff_avg_pixels8_msa; - c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_msa; - c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_msa; - c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_msa; - - c->avg_pixels_tab[2][0] = ff_avg_pixels4_msa; - c->avg_pixels_tab[2][1] = ff_avg_pixels4_x2_msa; - c->avg_pixels_tab[2][2] = ff_avg_pixels4_y2_msa; - c->avg_pixels_tab[2][3] = ff_avg_pixels4_xy2_msa; -} -#endif // #if HAVE_MSA - -#if HAVE_MMI -static void ff_hpeldsp_init_mmi(HpelDSPContext *c, int flags) -{ - c->put_pixels_tab[0][0] = ff_put_pixels16_8_mmi; - c->put_pixels_tab[0][1] = ff_put_pixels16_x2_8_mmi; - c->put_pixels_tab[0][2] = ff_put_pixels16_y2_8_mmi; - c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_8_mmi; - - c->put_pixels_tab[1][0] = ff_put_pixels8_8_mmi; - c->put_pixels_tab[1][1] = ff_put_pixels8_x2_8_mmi; - c->put_pixels_tab[1][2] = ff_put_pixels8_y2_8_mmi; - c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_8_mmi; - - c->put_pixels_tab[2][0] = ff_put_pixels4_8_mmi; - c->put_pixels_tab[2][1] = ff_put_pixels4_x2_8_mmi; - c->put_pixels_tab[2][2] = ff_put_pixels4_y2_8_mmi; - c->put_pixels_tab[2][3] = ff_put_pixels4_xy2_8_mmi; - - c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_8_mmi; - c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_8_mmi; - c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_8_mmi; - c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_8_mmi; - - c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_8_mmi; - c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_8_mmi; - c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_8_mmi; - c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_8_mmi; - - c->avg_pixels_tab[0][0] = ff_avg_pixels16_8_mmi; - c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_8_mmi; - c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_8_mmi; - c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_8_mmi; - - c->avg_pixels_tab[1][0] = ff_avg_pixels8_8_mmi; - c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_8_mmi; - c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_8_mmi; - c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_8_mmi; - - c->avg_pixels_tab[2][0] = ff_avg_pixels4_8_mmi; - c->avg_pixels_tab[2][1] = ff_avg_pixels4_x2_8_mmi; - c->avg_pixels_tab[2][2] = ff_avg_pixels4_y2_8_mmi; - c->avg_pixels_tab[2][3] = ff_avg_pixels4_xy2_8_mmi; -} -#endif // #if HAVE_MMI - void ff_hpeldsp_init_mips(HpelDSPContext *c, int flags) { -#if HAVE_MMI - ff_hpeldsp_init_mmi(c, flags); -#endif // #if HAVE_MMI -#if HAVE_MSA - ff_hpeldsp_init_msa(c, flags); -#endif // #if HAVE_MSA + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + c->put_pixels_tab[0][0] = ff_put_pixels16_8_mmi; + c->put_pixels_tab[0][1] = ff_put_pixels16_x2_8_mmi; + c->put_pixels_tab[0][2] = ff_put_pixels16_y2_8_mmi; + c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_8_mmi; + + c->put_pixels_tab[1][0] = ff_put_pixels8_8_mmi; + c->put_pixels_tab[1][1] = ff_put_pixels8_x2_8_mmi; + c->put_pixels_tab[1][2] = ff_put_pixels8_y2_8_mmi; + c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_8_mmi; + + c->put_pixels_tab[2][0] = ff_put_pixels4_8_mmi; + c->put_pixels_tab[2][1] = ff_put_pixels4_x2_8_mmi; + c->put_pixels_tab[2][2] = ff_put_pixels4_y2_8_mmi; + c->put_pixels_tab[2][3] = ff_put_pixels4_xy2_8_mmi; + + c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_8_mmi; + c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_8_mmi; + c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_8_mmi; + c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_8_mmi; + + c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_8_mmi; + c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_8_mmi; + c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_8_mmi; + c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_8_mmi; + + c->avg_pixels_tab[0][0] = ff_avg_pixels16_8_mmi; + c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_8_mmi; + c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_8_mmi; + c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_8_mmi; + + c->avg_pixels_tab[1][0] = ff_avg_pixels8_8_mmi; + c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_8_mmi; + c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_8_mmi; + c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_8_mmi; + + c->avg_pixels_tab[2][0] = ff_avg_pixels4_8_mmi; + c->avg_pixels_tab[2][1] = ff_avg_pixels4_x2_8_mmi; + c->avg_pixels_tab[2][2] = ff_avg_pixels4_y2_8_mmi; + c->avg_pixels_tab[2][3] = ff_avg_pixels4_xy2_8_mmi; + } + + if (have_msa(cpu_flags)) { + c->put_pixels_tab[0][0] = ff_put_pixels16_msa; + c->put_pixels_tab[0][1] = ff_put_pixels16_x2_msa; + c->put_pixels_tab[0][2] = ff_put_pixels16_y2_msa; + c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_msa; + + c->put_pixels_tab[1][0] = ff_put_pixels8_msa; + c->put_pixels_tab[1][1] = ff_put_pixels8_x2_msa; + c->put_pixels_tab[1][2] = ff_put_pixels8_y2_msa; + c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_msa; + + c->put_pixels_tab[2][1] = ff_put_pixels4_x2_msa; + c->put_pixels_tab[2][2] = ff_put_pixels4_y2_msa; + c->put_pixels_tab[2][3] = ff_put_pixels4_xy2_msa; + + c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_msa; + c->put_no_rnd_pixels_tab[0][1] = ff_put_no_rnd_pixels16_x2_msa; + c->put_no_rnd_pixels_tab[0][2] = ff_put_no_rnd_pixels16_y2_msa; + c->put_no_rnd_pixels_tab[0][3] = ff_put_no_rnd_pixels16_xy2_msa; + + c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_msa; + c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_msa; + c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_msa; + c->put_no_rnd_pixels_tab[1][3] = ff_put_no_rnd_pixels8_xy2_msa; + + c->avg_pixels_tab[0][0] = ff_avg_pixels16_msa; + c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_msa; + c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_msa; + c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_msa; + + c->avg_pixels_tab[1][0] = ff_avg_pixels8_msa; + c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_msa; + c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_msa; + c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_msa; + + c->avg_pixels_tab[2][0] = ff_avg_pixels4_msa; + c->avg_pixels_tab[2][1] = ff_avg_pixels4_x2_msa; + c->avg_pixels_tab[2][2] = ff_avg_pixels4_y2_msa; + c->avg_pixels_tab[2][3] = ff_avg_pixels4_xy2_msa; + } } diff --git a/libavcodec/mips/idctdsp_init_mips.c b/libavcodec/mips/idctdsp_init_mips.c index 85b76ca478..23efd9ed58 100644 --- a/libavcodec/mips/idctdsp_init_mips.c +++ b/libavcodec/mips/idctdsp_init_mips.c @@ -19,56 +19,44 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "idctdsp_mips.h" #include "xvididct_mips.h" -#if HAVE_MSA -static av_cold void idctdsp_init_msa(IDCTDSPContext *c, AVCodecContext *avctx, - unsigned high_bit_depth) +av_cold void ff_idctdsp_init_mips(IDCTDSPContext *c, AVCodecContext *avctx, + unsigned high_bit_depth) { - if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) && - (avctx->bits_per_raw_sample != 10) && - (avctx->bits_per_raw_sample != 12) && - (avctx->idct_algo == FF_IDCT_AUTO)) { - c->idct_put = ff_simple_idct_put_msa; - c->idct_add = ff_simple_idct_add_msa; - c->idct = ff_simple_idct_msa; - c->perm_type = FF_IDCT_PERM_NONE; - } + int cpu_flags = av_get_cpu_flags(); - c->put_pixels_clamped = ff_put_pixels_clamped_msa; - c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_msa; - c->add_pixels_clamped = ff_add_pixels_clamped_msa; -} -#endif // #if HAVE_MSA + if (have_mmi(cpu_flags)) { + if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) && + (avctx->bits_per_raw_sample != 10) && + (avctx->bits_per_raw_sample != 12) && + ((avctx->idct_algo == FF_IDCT_AUTO) || (avctx->idct_algo == FF_IDCT_SIMPLE))) { + c->idct_put = ff_simple_idct_put_8_mmi; + c->idct_add = ff_simple_idct_add_8_mmi; + c->idct = ff_simple_idct_8_mmi; + c->perm_type = FF_IDCT_PERM_NONE; + } -#if HAVE_MMI -static av_cold void idctdsp_init_mmi(IDCTDSPContext *c, AVCodecContext *avctx, - unsigned high_bit_depth) -{ - if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) && - (avctx->bits_per_raw_sample != 10) && - (avctx->bits_per_raw_sample != 12) && - ((avctx->idct_algo == FF_IDCT_AUTO) || (avctx->idct_algo == FF_IDCT_SIMPLE))) { - c->idct_put = ff_simple_idct_put_8_mmi; - c->idct_add = ff_simple_idct_add_8_mmi; - c->idct = ff_simple_idct_8_mmi; - c->perm_type = FF_IDCT_PERM_NONE; + c->put_pixels_clamped = ff_put_pixels_clamped_mmi; + c->add_pixels_clamped = ff_add_pixels_clamped_mmi; + c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmi; } - c->put_pixels_clamped = ff_put_pixels_clamped_mmi; - c->add_pixels_clamped = ff_add_pixels_clamped_mmi; - c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmi; -} -#endif /* HAVE_MMI */ + if (have_msa(cpu_flags)) { + if ((avctx->lowres != 1) && (avctx->lowres != 2) && (avctx->lowres != 3) && + (avctx->bits_per_raw_sample != 10) && + (avctx->bits_per_raw_sample != 12) && + (avctx->idct_algo == FF_IDCT_AUTO)) { + c->idct_put = ff_simple_idct_put_msa; + c->idct_add = ff_simple_idct_add_msa; + c->idct = ff_simple_idct_msa; + c->perm_type = FF_IDCT_PERM_NONE; + } -av_cold void ff_idctdsp_init_mips(IDCTDSPContext *c, AVCodecContext *avctx, - unsigned high_bit_depth) -{ -#if HAVE_MMI - idctdsp_init_mmi(c, avctx, high_bit_depth); -#endif /* HAVE_MMI */ -#if HAVE_MSA - idctdsp_init_msa(c, avctx, high_bit_depth); -#endif // #if HAVE_MSA + c->put_pixels_clamped = ff_put_pixels_clamped_msa; + c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_msa; + c->add_pixels_clamped = ff_add_pixels_clamped_msa; + } } diff --git a/libavcodec/mips/me_cmp_init_mips.c b/libavcodec/mips/me_cmp_init_mips.c index 219a0dc00c..e3e33b8e5e 100644 --- a/libavcodec/mips/me_cmp_init_mips.c +++ b/libavcodec/mips/me_cmp_init_mips.c @@ -18,39 +18,35 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "me_cmp_mips.h" -#if HAVE_MSA -static av_cold void me_cmp_msa(MECmpContext *c, AVCodecContext *avctx) +av_cold void ff_me_cmp_init_mips(MECmpContext *c, AVCodecContext *avctx) { + int cpu_flags = av_get_cpu_flags(); + + if (have_msa(cpu_flags)) { #if BIT_DEPTH == 8 - c->pix_abs[0][0] = ff_pix_abs16_msa; - c->pix_abs[0][1] = ff_pix_abs16_x2_msa; - c->pix_abs[0][2] = ff_pix_abs16_y2_msa; - c->pix_abs[0][3] = ff_pix_abs16_xy2_msa; - c->pix_abs[1][0] = ff_pix_abs8_msa; - c->pix_abs[1][1] = ff_pix_abs8_x2_msa; - c->pix_abs[1][2] = ff_pix_abs8_y2_msa; - c->pix_abs[1][3] = ff_pix_abs8_xy2_msa; + c->pix_abs[0][0] = ff_pix_abs16_msa; + c->pix_abs[0][1] = ff_pix_abs16_x2_msa; + c->pix_abs[0][2] = ff_pix_abs16_y2_msa; + c->pix_abs[0][3] = ff_pix_abs16_xy2_msa; + c->pix_abs[1][0] = ff_pix_abs8_msa; + c->pix_abs[1][1] = ff_pix_abs8_x2_msa; + c->pix_abs[1][2] = ff_pix_abs8_y2_msa; + c->pix_abs[1][3] = ff_pix_abs8_xy2_msa; - c->hadamard8_diff[0] = ff_hadamard8_diff16_msa; - c->hadamard8_diff[1] = ff_hadamard8_diff8x8_msa; + c->hadamard8_diff[0] = ff_hadamard8_diff16_msa; + c->hadamard8_diff[1] = ff_hadamard8_diff8x8_msa; - c->hadamard8_diff[4] = ff_hadamard8_intra16_msa; - c->hadamard8_diff[5] = ff_hadamard8_intra8x8_msa; + c->hadamard8_diff[4] = ff_hadamard8_intra16_msa; + c->hadamard8_diff[5] = ff_hadamard8_intra8x8_msa; - c->sad[0] = ff_pix_abs16_msa; - c->sad[1] = ff_pix_abs8_msa; - c->sse[0] = ff_sse16_msa; - c->sse[1] = ff_sse8_msa; - c->sse[2] = ff_sse4_msa; + c->sad[0] = ff_pix_abs16_msa; + c->sad[1] = ff_pix_abs8_msa; + c->sse[0] = ff_sse16_msa; + c->sse[1] = ff_sse8_msa; + c->sse[2] = ff_sse4_msa; #endif -} -#endif // #if HAVE_MSA - -av_cold void ff_me_cmp_init_mips(MECmpContext *c, AVCodecContext *avctx) -{ -#if HAVE_MSA - me_cmp_msa(c, avctx); -#endif // #if HAVE_MSA + } } diff --git a/libavcodec/mips/mpegvideo_init_mips.c b/libavcodec/mips/mpegvideo_init_mips.c index be77308140..bfda90bbcc 100644 --- a/libavcodec/mips/mpegvideo_init_mips.c +++ b/libavcodec/mips/mpegvideo_init_mips.c @@ -18,41 +18,31 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "h263dsp_mips.h" #include "mpegvideo_mips.h" -#if HAVE_MSA -static av_cold void dct_unquantize_init_msa(MpegEncContext *s) +av_cold void ff_mpv_common_init_mips(MpegEncContext *s) { - s->dct_unquantize_h263_intra = ff_dct_unquantize_h263_intra_msa; - s->dct_unquantize_h263_inter = ff_dct_unquantize_h263_inter_msa; - if (!s->q_scale_type) - s->dct_unquantize_mpeg2_inter = ff_dct_unquantize_mpeg2_inter_msa; -} -#endif // #if HAVE_MSA + int cpu_flags = av_get_cpu_flags(); -#if HAVE_MMI -static av_cold void dct_unquantize_init_mmi(MpegEncContext *s) -{ - s->dct_unquantize_h263_intra = ff_dct_unquantize_h263_intra_mmi; - s->dct_unquantize_h263_inter = ff_dct_unquantize_h263_inter_mmi; - s->dct_unquantize_mpeg1_intra = ff_dct_unquantize_mpeg1_intra_mmi; - s->dct_unquantize_mpeg1_inter = ff_dct_unquantize_mpeg1_inter_mmi; + if (have_mmi(cpu_flags)) { + s->dct_unquantize_h263_intra = ff_dct_unquantize_h263_intra_mmi; + s->dct_unquantize_h263_inter = ff_dct_unquantize_h263_inter_mmi; + s->dct_unquantize_mpeg1_intra = ff_dct_unquantize_mpeg1_intra_mmi; + s->dct_unquantize_mpeg1_inter = ff_dct_unquantize_mpeg1_inter_mmi; - if (!(s->avctx->flags & AV_CODEC_FLAG_BITEXACT)) - if (!s->q_scale_type) - s->dct_unquantize_mpeg2_intra = ff_dct_unquantize_mpeg2_intra_mmi; + if (!(s->avctx->flags & AV_CODEC_FLAG_BITEXACT)) + if (!s->q_scale_type) + s->dct_unquantize_mpeg2_intra = ff_dct_unquantize_mpeg2_intra_mmi; - s->denoise_dct= ff_denoise_dct_mmi; -} -#endif /* HAVE_MMI */ + s->denoise_dct= ff_denoise_dct_mmi; + } -av_cold void ff_mpv_common_init_mips(MpegEncContext *s) -{ -#if HAVE_MMI - dct_unquantize_init_mmi(s); -#endif /* HAVE_MMI */ -#if HAVE_MSA - dct_unquantize_init_msa(s); -#endif // #if HAVE_MSA + if (have_msa(cpu_flags)) { + s->dct_unquantize_h263_intra = ff_dct_unquantize_h263_intra_msa; + s->dct_unquantize_h263_inter = ff_dct_unquantize_h263_inter_msa; + if (!s->q_scale_type) + s->dct_unquantize_mpeg2_inter = ff_dct_unquantize_mpeg2_inter_msa; + } } diff --git a/libavcodec/mips/mpegvideoencdsp_init_mips.c b/libavcodec/mips/mpegvideoencdsp_init_mips.c index 9bfe94e4cd..71831a61ac 100644 --- a/libavcodec/mips/mpegvideoencdsp_init_mips.c +++ b/libavcodec/mips/mpegvideoencdsp_init_mips.c @@ -18,23 +18,18 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "libavcodec/bit_depth_template.c" #include "h263dsp_mips.h" -#if HAVE_MSA -static av_cold void mpegvideoencdsp_init_msa(MpegvideoEncDSPContext *c, - AVCodecContext *avctx) -{ -#if BIT_DEPTH == 8 - c->pix_sum = ff_pix_sum_msa; -#endif -} -#endif // #if HAVE_MSA - av_cold void ff_mpegvideoencdsp_init_mips(MpegvideoEncDSPContext *c, AVCodecContext *avctx) { -#if HAVE_MSA - mpegvideoencdsp_init_msa(c, avctx); -#endif // #if HAVE_MSA + int cpu_flags = av_get_cpu_flags(); + + if (have_msa(cpu_flags)) { +#if BIT_DEPTH == 8 + c->pix_sum = ff_pix_sum_msa; +#endif + } } diff --git a/libavcodec/mips/pixblockdsp_init_mips.c b/libavcodec/mips/pixblockdsp_init_mips.c index fd0238d79b..2e2d70953b 100644 --- a/libavcodec/mips/pixblockdsp_init_mips.c +++ b/libavcodec/mips/pixblockdsp_init_mips.c @@ -19,51 +19,38 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "pixblockdsp_mips.h" -#if HAVE_MSA -static av_cold void pixblockdsp_init_msa(PixblockDSPContext *c, - AVCodecContext *avctx, - unsigned high_bit_depth) +void ff_pixblockdsp_init_mips(PixblockDSPContext *c, AVCodecContext *avctx, + unsigned high_bit_depth) { - c->diff_pixels = ff_diff_pixels_msa; + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + c->diff_pixels = ff_diff_pixels_mmi; - switch (avctx->bits_per_raw_sample) { - case 9: - case 10: - case 12: - case 14: - c->get_pixels = ff_get_pixels_16_msa; - break; - default: - if (avctx->bits_per_raw_sample <= 8 || avctx->codec_type != - AVMEDIA_TYPE_VIDEO) { - c->get_pixels = ff_get_pixels_8_msa; + if (!high_bit_depth || avctx->codec_type != AVMEDIA_TYPE_VIDEO) { + c->get_pixels = ff_get_pixels_8_mmi; } - break; } -} -#endif // #if HAVE_MSA -#if HAVE_MMI -static av_cold void pixblockdsp_init_mmi(PixblockDSPContext *c, - AVCodecContext *avctx, unsigned high_bit_depth) -{ - c->diff_pixels = ff_diff_pixels_mmi; + if (have_msa(cpu_flags)) { + c->diff_pixels = ff_diff_pixels_msa; - if (!high_bit_depth || avctx->codec_type != AVMEDIA_TYPE_VIDEO) { - c->get_pixels = ff_get_pixels_8_mmi; + switch (avctx->bits_per_raw_sample) { + case 9: + case 10: + case 12: + case 14: + c->get_pixels = ff_get_pixels_16_msa; + break; + default: + if (avctx->bits_per_raw_sample <= 8 || avctx->codec_type != + AVMEDIA_TYPE_VIDEO) { + c->get_pixels = ff_get_pixels_8_msa; + } + break; + } } } -#endif /* HAVE_MMI */ - -void ff_pixblockdsp_init_mips(PixblockDSPContext *c, AVCodecContext *avctx, - unsigned high_bit_depth) -{ -#if HAVE_MMI - pixblockdsp_init_mmi(c, avctx, high_bit_depth); -#endif /* HAVE_MMI */ -#if HAVE_MSA - pixblockdsp_init_msa(c, avctx, high_bit_depth); -#endif // #if HAVE_MSA -} diff --git a/libavcodec/mips/qpeldsp_init_mips.c b/libavcodec/mips/qpeldsp_init_mips.c index 140e8f89c9..cccf9d4429 100644 --- a/libavcodec/mips/qpeldsp_init_mips.c +++ b/libavcodec/mips/qpeldsp_init_mips.c @@ -18,150 +18,146 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "qpeldsp_mips.h" -#if HAVE_MSA -static av_cold void qpeldsp_init_msa(QpelDSPContext *c) +void ff_qpeldsp_init_mips(QpelDSPContext *c) { - c->put_qpel_pixels_tab[0][0] = ff_copy_16x16_msa; - c->put_qpel_pixels_tab[0][1] = ff_horiz_mc_qpel_aver_src0_16width_msa; - c->put_qpel_pixels_tab[0][2] = ff_horiz_mc_qpel_16width_msa; - c->put_qpel_pixels_tab[0][3] = ff_horiz_mc_qpel_aver_src1_16width_msa; - c->put_qpel_pixels_tab[0][4] = ff_vert_mc_qpel_aver_src0_16x16_msa; - c->put_qpel_pixels_tab[0][5] = ff_hv_mc_qpel_aver_hv_src00_16x16_msa; - c->put_qpel_pixels_tab[0][6] = ff_hv_mc_qpel_aver_v_src0_16x16_msa; - c->put_qpel_pixels_tab[0][7] = ff_hv_mc_qpel_aver_hv_src10_16x16_msa; - c->put_qpel_pixels_tab[0][8] = ff_vert_mc_qpel_16x16_msa; - c->put_qpel_pixels_tab[0][9] = ff_hv_mc_qpel_aver_h_src0_16x16_msa; - c->put_qpel_pixels_tab[0][10] = ff_hv_mc_qpel_16x16_msa; - c->put_qpel_pixels_tab[0][11] = ff_hv_mc_qpel_aver_h_src1_16x16_msa; - c->put_qpel_pixels_tab[0][12] = ff_vert_mc_qpel_aver_src1_16x16_msa; - c->put_qpel_pixels_tab[0][13] = ff_hv_mc_qpel_aver_hv_src01_16x16_msa; - c->put_qpel_pixels_tab[0][14] = ff_hv_mc_qpel_aver_v_src1_16x16_msa; - c->put_qpel_pixels_tab[0][15] = ff_hv_mc_qpel_aver_hv_src11_16x16_msa; + int cpu_flags = av_get_cpu_flags(); - c->put_qpel_pixels_tab[1][0] = ff_copy_8x8_msa; - c->put_qpel_pixels_tab[1][1] = ff_horiz_mc_qpel_aver_src0_8width_msa; - c->put_qpel_pixels_tab[1][2] = ff_horiz_mc_qpel_8width_msa; - c->put_qpel_pixels_tab[1][3] = ff_horiz_mc_qpel_aver_src1_8width_msa; - c->put_qpel_pixels_tab[1][4] = ff_vert_mc_qpel_aver_src0_8x8_msa; - c->put_qpel_pixels_tab[1][5] = ff_hv_mc_qpel_aver_hv_src00_8x8_msa; - c->put_qpel_pixels_tab[1][6] = ff_hv_mc_qpel_aver_v_src0_8x8_msa; - c->put_qpel_pixels_tab[1][7] = ff_hv_mc_qpel_aver_hv_src10_8x8_msa; - c->put_qpel_pixels_tab[1][8] = ff_vert_mc_qpel_8x8_msa; - c->put_qpel_pixels_tab[1][9] = ff_hv_mc_qpel_aver_h_src0_8x8_msa; - c->put_qpel_pixels_tab[1][10] = ff_hv_mc_qpel_8x8_msa; - c->put_qpel_pixels_tab[1][11] = ff_hv_mc_qpel_aver_h_src1_8x8_msa; - c->put_qpel_pixels_tab[1][12] = ff_vert_mc_qpel_aver_src1_8x8_msa; - c->put_qpel_pixels_tab[1][13] = ff_hv_mc_qpel_aver_hv_src01_8x8_msa; - c->put_qpel_pixels_tab[1][14] = ff_hv_mc_qpel_aver_v_src1_8x8_msa; - c->put_qpel_pixels_tab[1][15] = ff_hv_mc_qpel_aver_hv_src11_8x8_msa; + if (have_msa(cpu_flags)) { + c->put_qpel_pixels_tab[0][0] = ff_copy_16x16_msa; + c->put_qpel_pixels_tab[0][1] = ff_horiz_mc_qpel_aver_src0_16width_msa; + c->put_qpel_pixels_tab[0][2] = ff_horiz_mc_qpel_16width_msa; + c->put_qpel_pixels_tab[0][3] = ff_horiz_mc_qpel_aver_src1_16width_msa; + c->put_qpel_pixels_tab[0][4] = ff_vert_mc_qpel_aver_src0_16x16_msa; + c->put_qpel_pixels_tab[0][5] = ff_hv_mc_qpel_aver_hv_src00_16x16_msa; + c->put_qpel_pixels_tab[0][6] = ff_hv_mc_qpel_aver_v_src0_16x16_msa; + c->put_qpel_pixels_tab[0][7] = ff_hv_mc_qpel_aver_hv_src10_16x16_msa; + c->put_qpel_pixels_tab[0][8] = ff_vert_mc_qpel_16x16_msa; + c->put_qpel_pixels_tab[0][9] = ff_hv_mc_qpel_aver_h_src0_16x16_msa; + c->put_qpel_pixels_tab[0][10] = ff_hv_mc_qpel_16x16_msa; + c->put_qpel_pixels_tab[0][11] = ff_hv_mc_qpel_aver_h_src1_16x16_msa; + c->put_qpel_pixels_tab[0][12] = ff_vert_mc_qpel_aver_src1_16x16_msa; + c->put_qpel_pixels_tab[0][13] = ff_hv_mc_qpel_aver_hv_src01_16x16_msa; + c->put_qpel_pixels_tab[0][14] = ff_hv_mc_qpel_aver_v_src1_16x16_msa; + c->put_qpel_pixels_tab[0][15] = ff_hv_mc_qpel_aver_hv_src11_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][0] = ff_copy_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][1] = - ff_horiz_mc_qpel_no_rnd_aver_src0_16width_msa; - c->put_no_rnd_qpel_pixels_tab[0][2] = ff_horiz_mc_qpel_no_rnd_16width_msa; - c->put_no_rnd_qpel_pixels_tab[0][3] = - ff_horiz_mc_qpel_no_rnd_aver_src1_16width_msa; - c->put_no_rnd_qpel_pixels_tab[0][4] = - ff_vert_mc_qpel_no_rnd_aver_src0_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][5] = - ff_hv_mc_qpel_no_rnd_aver_hv_src00_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][6] = - ff_hv_mc_qpel_no_rnd_aver_v_src0_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][7] = - ff_hv_mc_qpel_no_rnd_aver_hv_src10_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][8] = ff_vert_mc_qpel_no_rnd_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][9] = - ff_hv_mc_qpel_no_rnd_aver_h_src0_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][10] = ff_hv_mc_qpel_no_rnd_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][11] = - ff_hv_mc_qpel_no_rnd_aver_h_src1_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][12] = - ff_vert_mc_qpel_no_rnd_aver_src1_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][13] = - ff_hv_mc_qpel_no_rnd_aver_hv_src01_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][14] = - ff_hv_mc_qpel_no_rnd_aver_v_src1_16x16_msa; - c->put_no_rnd_qpel_pixels_tab[0][15] = - ff_hv_mc_qpel_no_rnd_aver_hv_src11_16x16_msa; + c->put_qpel_pixels_tab[1][0] = ff_copy_8x8_msa; + c->put_qpel_pixels_tab[1][1] = ff_horiz_mc_qpel_aver_src0_8width_msa; + c->put_qpel_pixels_tab[1][2] = ff_horiz_mc_qpel_8width_msa; + c->put_qpel_pixels_tab[1][3] = ff_horiz_mc_qpel_aver_src1_8width_msa; + c->put_qpel_pixels_tab[1][4] = ff_vert_mc_qpel_aver_src0_8x8_msa; + c->put_qpel_pixels_tab[1][5] = ff_hv_mc_qpel_aver_hv_src00_8x8_msa; + c->put_qpel_pixels_tab[1][6] = ff_hv_mc_qpel_aver_v_src0_8x8_msa; + c->put_qpel_pixels_tab[1][7] = ff_hv_mc_qpel_aver_hv_src10_8x8_msa; + c->put_qpel_pixels_tab[1][8] = ff_vert_mc_qpel_8x8_msa; + c->put_qpel_pixels_tab[1][9] = ff_hv_mc_qpel_aver_h_src0_8x8_msa; + c->put_qpel_pixels_tab[1][10] = ff_hv_mc_qpel_8x8_msa; + c->put_qpel_pixels_tab[1][11] = ff_hv_mc_qpel_aver_h_src1_8x8_msa; + c->put_qpel_pixels_tab[1][12] = ff_vert_mc_qpel_aver_src1_8x8_msa; + c->put_qpel_pixels_tab[1][13] = ff_hv_mc_qpel_aver_hv_src01_8x8_msa; + c->put_qpel_pixels_tab[1][14] = ff_hv_mc_qpel_aver_v_src1_8x8_msa; + c->put_qpel_pixels_tab[1][15] = ff_hv_mc_qpel_aver_hv_src11_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][0] = ff_copy_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][1] = - ff_horiz_mc_qpel_no_rnd_aver_src0_8width_msa; - c->put_no_rnd_qpel_pixels_tab[1][2] = ff_horiz_mc_qpel_no_rnd_8width_msa; - c->put_no_rnd_qpel_pixels_tab[1][3] = - ff_horiz_mc_qpel_no_rnd_aver_src1_8width_msa; - c->put_no_rnd_qpel_pixels_tab[1][4] = - ff_vert_mc_qpel_no_rnd_aver_src0_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][5] = - ff_hv_mc_qpel_no_rnd_aver_hv_src00_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][6] = - ff_hv_mc_qpel_no_rnd_aver_v_src0_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][7] = - ff_hv_mc_qpel_no_rnd_aver_hv_src10_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][8] = ff_vert_mc_qpel_no_rnd_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][9] = - ff_hv_mc_qpel_no_rnd_aver_h_src0_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][10] = ff_hv_mc_qpel_no_rnd_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][11] = - ff_hv_mc_qpel_no_rnd_aver_h_src1_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][12] = - ff_vert_mc_qpel_no_rnd_aver_src1_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][13] = - ff_hv_mc_qpel_no_rnd_aver_hv_src01_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][14] = - ff_hv_mc_qpel_no_rnd_aver_v_src1_8x8_msa; - c->put_no_rnd_qpel_pixels_tab[1][15] = - ff_hv_mc_qpel_no_rnd_aver_hv_src11_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[0][0] = ff_copy_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][1] = + ff_horiz_mc_qpel_no_rnd_aver_src0_16width_msa; + c->put_no_rnd_qpel_pixels_tab[0][2] = ff_horiz_mc_qpel_no_rnd_16width_msa; + c->put_no_rnd_qpel_pixels_tab[0][3] = + ff_horiz_mc_qpel_no_rnd_aver_src1_16width_msa; + c->put_no_rnd_qpel_pixels_tab[0][4] = + ff_vert_mc_qpel_no_rnd_aver_src0_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][5] = + ff_hv_mc_qpel_no_rnd_aver_hv_src00_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][6] = + ff_hv_mc_qpel_no_rnd_aver_v_src0_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][7] = + ff_hv_mc_qpel_no_rnd_aver_hv_src10_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][8] = ff_vert_mc_qpel_no_rnd_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][9] = + ff_hv_mc_qpel_no_rnd_aver_h_src0_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][10] = ff_hv_mc_qpel_no_rnd_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][11] = + ff_hv_mc_qpel_no_rnd_aver_h_src1_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][12] = + ff_vert_mc_qpel_no_rnd_aver_src1_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][13] = + ff_hv_mc_qpel_no_rnd_aver_hv_src01_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][14] = + ff_hv_mc_qpel_no_rnd_aver_v_src1_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[0][15] = + ff_hv_mc_qpel_no_rnd_aver_hv_src11_16x16_msa; - c->avg_qpel_pixels_tab[0][0] = ff_avg_width16_msa; - c->avg_qpel_pixels_tab[0][1] = - ff_horiz_mc_qpel_avg_dst_aver_src0_16width_msa; - c->avg_qpel_pixels_tab[0][2] = ff_horiz_mc_qpel_avg_dst_16width_msa; - c->avg_qpel_pixels_tab[0][3] = - ff_horiz_mc_qpel_avg_dst_aver_src1_16width_msa; - c->avg_qpel_pixels_tab[0][4] = ff_vert_mc_qpel_avg_dst_aver_src0_16x16_msa; - c->avg_qpel_pixels_tab[0][5] = - ff_hv_mc_qpel_avg_dst_aver_hv_src00_16x16_msa; - c->avg_qpel_pixels_tab[0][6] = ff_hv_mc_qpel_avg_dst_aver_v_src0_16x16_msa; - c->avg_qpel_pixels_tab[0][7] = - ff_hv_mc_qpel_avg_dst_aver_hv_src10_16x16_msa; - c->avg_qpel_pixels_tab[0][8] = ff_vert_mc_qpel_avg_dst_16x16_msa; - c->avg_qpel_pixels_tab[0][9] = ff_hv_mc_qpel_avg_dst_aver_h_src0_16x16_msa; - c->avg_qpel_pixels_tab[0][10] = ff_hv_mc_qpel_avg_dst_16x16_msa; - c->avg_qpel_pixels_tab[0][11] = ff_hv_mc_qpel_avg_dst_aver_h_src1_16x16_msa; - c->avg_qpel_pixels_tab[0][12] = ff_vert_mc_qpel_avg_dst_aver_src1_16x16_msa; - c->avg_qpel_pixels_tab[0][13] = - ff_hv_mc_qpel_avg_dst_aver_hv_src01_16x16_msa; - c->avg_qpel_pixels_tab[0][14] = ff_hv_mc_qpel_avg_dst_aver_v_src1_16x16_msa; - c->avg_qpel_pixels_tab[0][15] = - ff_hv_mc_qpel_avg_dst_aver_hv_src11_16x16_msa; + c->put_no_rnd_qpel_pixels_tab[1][0] = ff_copy_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][1] = + ff_horiz_mc_qpel_no_rnd_aver_src0_8width_msa; + c->put_no_rnd_qpel_pixels_tab[1][2] = ff_horiz_mc_qpel_no_rnd_8width_msa; + c->put_no_rnd_qpel_pixels_tab[1][3] = + ff_horiz_mc_qpel_no_rnd_aver_src1_8width_msa; + c->put_no_rnd_qpel_pixels_tab[1][4] = + ff_vert_mc_qpel_no_rnd_aver_src0_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][5] = + ff_hv_mc_qpel_no_rnd_aver_hv_src00_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][6] = + ff_hv_mc_qpel_no_rnd_aver_v_src0_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][7] = + ff_hv_mc_qpel_no_rnd_aver_hv_src10_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][8] = ff_vert_mc_qpel_no_rnd_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][9] = + ff_hv_mc_qpel_no_rnd_aver_h_src0_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][10] = ff_hv_mc_qpel_no_rnd_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][11] = + ff_hv_mc_qpel_no_rnd_aver_h_src1_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][12] = + ff_vert_mc_qpel_no_rnd_aver_src1_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][13] = + ff_hv_mc_qpel_no_rnd_aver_hv_src01_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][14] = + ff_hv_mc_qpel_no_rnd_aver_v_src1_8x8_msa; + c->put_no_rnd_qpel_pixels_tab[1][15] = + ff_hv_mc_qpel_no_rnd_aver_hv_src11_8x8_msa; - c->avg_qpel_pixels_tab[1][0] = ff_avg_width8_msa; - c->avg_qpel_pixels_tab[1][1] = - ff_horiz_mc_qpel_avg_dst_aver_src0_8width_msa; - c->avg_qpel_pixels_tab[1][2] = ff_horiz_mc_qpel_avg_dst_8width_msa; - c->avg_qpel_pixels_tab[1][3] = - ff_horiz_mc_qpel_avg_dst_aver_src1_8width_msa; - c->avg_qpel_pixels_tab[1][4] = ff_vert_mc_qpel_avg_dst_aver_src0_8x8_msa; - c->avg_qpel_pixels_tab[1][5] = ff_hv_mc_qpel_avg_dst_aver_hv_src00_8x8_msa; - c->avg_qpel_pixels_tab[1][6] = ff_hv_mc_qpel_avg_dst_aver_v_src0_8x8_msa; - c->avg_qpel_pixels_tab[1][7] = ff_hv_mc_qpel_avg_dst_aver_hv_src10_8x8_msa; - c->avg_qpel_pixels_tab[1][8] = ff_vert_mc_qpel_avg_dst_8x8_msa; - c->avg_qpel_pixels_tab[1][9] = ff_hv_mc_qpel_avg_dst_aver_h_src0_8x8_msa; - c->avg_qpel_pixels_tab[1][10] = ff_hv_mc_qpel_avg_dst_8x8_msa; - c->avg_qpel_pixels_tab[1][11] = ff_hv_mc_qpel_avg_dst_aver_h_src1_8x8_msa; - c->avg_qpel_pixels_tab[1][12] = ff_vert_mc_qpel_avg_dst_aver_src1_8x8_msa; - c->avg_qpel_pixels_tab[1][13] = ff_hv_mc_qpel_avg_dst_aver_hv_src01_8x8_msa; - c->avg_qpel_pixels_tab[1][14] = ff_hv_mc_qpel_avg_dst_aver_v_src1_8x8_msa; - c->avg_qpel_pixels_tab[1][15] = ff_hv_mc_qpel_avg_dst_aver_hv_src11_8x8_msa; -} -#endif // #if HAVE_MSA + c->avg_qpel_pixels_tab[0][0] = ff_avg_width16_msa; + c->avg_qpel_pixels_tab[0][1] = + ff_horiz_mc_qpel_avg_dst_aver_src0_16width_msa; + c->avg_qpel_pixels_tab[0][2] = ff_horiz_mc_qpel_avg_dst_16width_msa; + c->avg_qpel_pixels_tab[0][3] = + ff_horiz_mc_qpel_avg_dst_aver_src1_16width_msa; + c->avg_qpel_pixels_tab[0][4] = ff_vert_mc_qpel_avg_dst_aver_src0_16x16_msa; + c->avg_qpel_pixels_tab[0][5] = + ff_hv_mc_qpel_avg_dst_aver_hv_src00_16x16_msa; + c->avg_qpel_pixels_tab[0][6] = ff_hv_mc_qpel_avg_dst_aver_v_src0_16x16_msa; + c->avg_qpel_pixels_tab[0][7] = + ff_hv_mc_qpel_avg_dst_aver_hv_src10_16x16_msa; + c->avg_qpel_pixels_tab[0][8] = ff_vert_mc_qpel_avg_dst_16x16_msa; + c->avg_qpel_pixels_tab[0][9] = ff_hv_mc_qpel_avg_dst_aver_h_src0_16x16_msa; + c->avg_qpel_pixels_tab[0][10] = ff_hv_mc_qpel_avg_dst_16x16_msa; + c->avg_qpel_pixels_tab[0][11] = ff_hv_mc_qpel_avg_dst_aver_h_src1_16x16_msa; + c->avg_qpel_pixels_tab[0][12] = ff_vert_mc_qpel_avg_dst_aver_src1_16x16_msa; + c->avg_qpel_pixels_tab[0][13] = + ff_hv_mc_qpel_avg_dst_aver_hv_src01_16x16_msa; + c->avg_qpel_pixels_tab[0][14] = ff_hv_mc_qpel_avg_dst_aver_v_src1_16x16_msa; + c->avg_qpel_pixels_tab[0][15] = + ff_hv_mc_qpel_avg_dst_aver_hv_src11_16x16_msa; -void ff_qpeldsp_init_mips(QpelDSPContext *c) -{ -#if HAVE_MSA - qpeldsp_init_msa(c); -#endif // #if HAVE_MSA + c->avg_qpel_pixels_tab[1][0] = ff_avg_width8_msa; + c->avg_qpel_pixels_tab[1][1] = + ff_horiz_mc_qpel_avg_dst_aver_src0_8width_msa; + c->avg_qpel_pixels_tab[1][2] = ff_horiz_mc_qpel_avg_dst_8width_msa; + c->avg_qpel_pixels_tab[1][3] = + ff_horiz_mc_qpel_avg_dst_aver_src1_8width_msa; + c->avg_qpel_pixels_tab[1][4] = ff_vert_mc_qpel_avg_dst_aver_src0_8x8_msa; + c->avg_qpel_pixels_tab[1][5] = ff_hv_mc_qpel_avg_dst_aver_hv_src00_8x8_msa; + c->avg_qpel_pixels_tab[1][6] = ff_hv_mc_qpel_avg_dst_aver_v_src0_8x8_msa; + c->avg_qpel_pixels_tab[1][7] = ff_hv_mc_qpel_avg_dst_aver_hv_src10_8x8_msa; + c->avg_qpel_pixels_tab[1][8] = ff_vert_mc_qpel_avg_dst_8x8_msa; + c->avg_qpel_pixels_tab[1][9] = ff_hv_mc_qpel_avg_dst_aver_h_src0_8x8_msa; + c->avg_qpel_pixels_tab[1][10] = ff_hv_mc_qpel_avg_dst_8x8_msa; + c->avg_qpel_pixels_tab[1][11] = ff_hv_mc_qpel_avg_dst_aver_h_src1_8x8_msa; + c->avg_qpel_pixels_tab[1][12] = ff_vert_mc_qpel_avg_dst_aver_src1_8x8_msa; + c->avg_qpel_pixels_tab[1][13] = ff_hv_mc_qpel_avg_dst_aver_hv_src01_8x8_msa; + c->avg_qpel_pixels_tab[1][14] = ff_hv_mc_qpel_avg_dst_aver_v_src1_8x8_msa; + c->avg_qpel_pixels_tab[1][15] = ff_hv_mc_qpel_avg_dst_aver_hv_src11_8x8_msa; + } } diff --git a/libavcodec/mips/vc1dsp_init_mips.c b/libavcodec/mips/vc1dsp_init_mips.c index c0007ff650..94126f3a9d 100644 --- a/libavcodec/mips/vc1dsp_init_mips.c +++ b/libavcodec/mips/vc1dsp_init_mips.c @@ -18,6 +18,7 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "libavutil/attributes.h" #include "libavcodec/vc1dsp.h" #include "vc1dsp_mips.h" @@ -27,104 +28,93 @@ dsp->OP##vc1_mspel_pixels_tab[1][X+4*Y] = ff_##OP##vc1_mspel_mc##X##Y##INSN; \ dsp->OP##vc1_mspel_pixels_tab[0][X+4*Y] = ff_##OP##vc1_mspel_mc##X##Y##_16##INSN -#if HAVE_MMI -static av_cold void vc1dsp_init_mmi(VC1DSPContext *dsp) -{ -#if _MIPS_SIM != _ABIO32 - dsp->vc1_inv_trans_8x8 = ff_vc1_inv_trans_8x8_mmi; - dsp->vc1_inv_trans_4x8 = ff_vc1_inv_trans_4x8_mmi; - dsp->vc1_inv_trans_8x4 = ff_vc1_inv_trans_8x4_mmi; -#endif - dsp->vc1_inv_trans_4x4 = ff_vc1_inv_trans_4x4_mmi; - dsp->vc1_inv_trans_8x8_dc = ff_vc1_inv_trans_8x8_dc_mmi; - dsp->vc1_inv_trans_4x8_dc = ff_vc1_inv_trans_4x8_dc_mmi; - dsp->vc1_inv_trans_8x4_dc = ff_vc1_inv_trans_8x4_dc_mmi; - dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_mmi; - - dsp->vc1_h_overlap = ff_vc1_h_overlap_mmi; - dsp->vc1_v_overlap = ff_vc1_v_overlap_mmi; - dsp->vc1_h_s_overlap = ff_vc1_h_s_overlap_mmi; - dsp->vc1_v_s_overlap = ff_vc1_v_s_overlap_mmi; - - dsp->vc1_v_loop_filter4 = ff_vc1_v_loop_filter4_mmi; - dsp->vc1_h_loop_filter4 = ff_vc1_h_loop_filter4_mmi; - dsp->vc1_v_loop_filter8 = ff_vc1_v_loop_filter8_mmi; - dsp->vc1_h_loop_filter8 = ff_vc1_h_loop_filter8_mmi; - dsp->vc1_v_loop_filter16 = ff_vc1_v_loop_filter16_mmi; - dsp->vc1_h_loop_filter16 = ff_vc1_h_loop_filter16_mmi; - - FN_ASSIGN(put_, 0, 0, _mmi); - FN_ASSIGN(put_, 0, 1, _mmi); - FN_ASSIGN(put_, 0, 2, _mmi); - FN_ASSIGN(put_, 0, 3, _mmi); - - FN_ASSIGN(put_, 1, 0, _mmi); - //FN_ASSIGN(put_, 1, 1, _mmi);//FIXME - //FN_ASSIGN(put_, 1, 2, _mmi);//FIXME - //FN_ASSIGN(put_, 1, 3, _mmi);//FIXME - - FN_ASSIGN(put_, 2, 0, _mmi); - //FN_ASSIGN(put_, 2, 1, _mmi);//FIXME - //FN_ASSIGN(put_, 2, 2, _mmi);//FIXME - //FN_ASSIGN(put_, 2, 3, _mmi);//FIXME - - FN_ASSIGN(put_, 3, 0, _mmi); - //FN_ASSIGN(put_, 3, 1, _mmi);//FIXME - //FN_ASSIGN(put_, 3, 2, _mmi);//FIXME - //FN_ASSIGN(put_, 3, 3, _mmi);//FIXME - - FN_ASSIGN(avg_, 0, 0, _mmi); - FN_ASSIGN(avg_, 0, 1, _mmi); - FN_ASSIGN(avg_, 0, 2, _mmi); - FN_ASSIGN(avg_, 0, 3, _mmi); - - FN_ASSIGN(avg_, 1, 0, _mmi); - //FN_ASSIGN(avg_, 1, 1, _mmi);//FIXME - //FN_ASSIGN(avg_, 1, 2, _mmi);//FIXME - //FN_ASSIGN(avg_, 1, 3, _mmi);//FIXME - - FN_ASSIGN(avg_, 2, 0, _mmi); - //FN_ASSIGN(avg_, 2, 1, _mmi);//FIXME - //FN_ASSIGN(avg_, 2, 2, _mmi);//FIXME - //FN_ASSIGN(avg_, 2, 3, _mmi);//FIXME - - FN_ASSIGN(avg_, 3, 0, _mmi); - //FN_ASSIGN(avg_, 3, 1, _mmi);//FIXME - //FN_ASSIGN(avg_, 3, 2, _mmi);//FIXME - //FN_ASSIGN(avg_, 3, 3, _mmi);//FIXME - - dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = ff_put_no_rnd_vc1_chroma_mc8_mmi; - dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_no_rnd_vc1_chroma_mc8_mmi; - dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = ff_put_no_rnd_vc1_chroma_mc4_mmi; - dsp->avg_no_rnd_vc1_chroma_pixels_tab[1] = ff_avg_no_rnd_vc1_chroma_mc4_mmi; -} -#endif /* HAVE_MMI */ - -#if HAVE_MSA -static av_cold void vc1dsp_init_msa(VC1DSPContext *dsp) -{ - dsp->vc1_inv_trans_8x8 = ff_vc1_inv_trans_8x8_msa; - dsp->vc1_inv_trans_4x8 = ff_vc1_inv_trans_4x8_msa; - dsp->vc1_inv_trans_8x4 = ff_vc1_inv_trans_8x4_msa; - - FN_ASSIGN(put_, 1, 1, _msa); - FN_ASSIGN(put_, 1, 2, _msa); - FN_ASSIGN(put_, 1, 3, _msa); - FN_ASSIGN(put_, 2, 1, _msa); - FN_ASSIGN(put_, 2, 2, _msa); - FN_ASSIGN(put_, 2, 3, _msa); - FN_ASSIGN(put_, 3, 1, _msa); - FN_ASSIGN(put_, 3, 2, _msa); - FN_ASSIGN(put_, 3, 3, _msa); -} -#endif /* HAVE_MSA */ - av_cold void ff_vc1dsp_init_mips(VC1DSPContext *dsp) { -#if HAVE_MMI - vc1dsp_init_mmi(dsp); -#endif /* HAVE_MMI */ -#if HAVE_MSA - vc1dsp_init_msa(dsp); -#endif /* HAVE_MSA */ + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + #if _MIPS_SIM != _ABIO32 + dsp->vc1_inv_trans_8x8 = ff_vc1_inv_trans_8x8_mmi; + dsp->vc1_inv_trans_4x8 = ff_vc1_inv_trans_4x8_mmi; + dsp->vc1_inv_trans_8x4 = ff_vc1_inv_trans_8x4_mmi; +#endif + dsp->vc1_inv_trans_4x4 = ff_vc1_inv_trans_4x4_mmi; + dsp->vc1_inv_trans_8x8_dc = ff_vc1_inv_trans_8x8_dc_mmi; + dsp->vc1_inv_trans_4x8_dc = ff_vc1_inv_trans_4x8_dc_mmi; + dsp->vc1_inv_trans_8x4_dc = ff_vc1_inv_trans_8x4_dc_mmi; + dsp->vc1_inv_trans_4x4_dc = ff_vc1_inv_trans_4x4_dc_mmi; + + dsp->vc1_h_overlap = ff_vc1_h_overlap_mmi; + dsp->vc1_v_overlap = ff_vc1_v_overlap_mmi; + dsp->vc1_h_s_overlap = ff_vc1_h_s_overlap_mmi; + dsp->vc1_v_s_overlap = ff_vc1_v_s_overlap_mmi; + + dsp->vc1_v_loop_filter4 = ff_vc1_v_loop_filter4_mmi; + dsp->vc1_h_loop_filter4 = ff_vc1_h_loop_filter4_mmi; + dsp->vc1_v_loop_filter8 = ff_vc1_v_loop_filter8_mmi; + dsp->vc1_h_loop_filter8 = ff_vc1_h_loop_filter8_mmi; + dsp->vc1_v_loop_filter16 = ff_vc1_v_loop_filter16_mmi; + dsp->vc1_h_loop_filter16 = ff_vc1_h_loop_filter16_mmi; + + FN_ASSIGN(put_, 0, 0, _mmi); + FN_ASSIGN(put_, 0, 1, _mmi); + FN_ASSIGN(put_, 0, 2, _mmi); + FN_ASSIGN(put_, 0, 3, _mmi); + + FN_ASSIGN(put_, 1, 0, _mmi); + //FN_ASSIGN(put_, 1, 1, _mmi);//FIXME + //FN_ASSIGN(put_, 1, 2, _mmi);//FIXME + //FN_ASSIGN(put_, 1, 3, _mmi);//FIXME + + FN_ASSIGN(put_, 2, 0, _mmi); + //FN_ASSIGN(put_, 2, 1, _mmi);//FIXME + //FN_ASSIGN(put_, 2, 2, _mmi);//FIXME + //FN_ASSIGN(put_, 2, 3, _mmi);//FIXME + + FN_ASSIGN(put_, 3, 0, _mmi); + //FN_ASSIGN(put_, 3, 1, _mmi);//FIXME + //FN_ASSIGN(put_, 3, 2, _mmi);//FIXME + //FN_ASSIGN(put_, 3, 3, _mmi);//FIXME + + FN_ASSIGN(avg_, 0, 0, _mmi); + FN_ASSIGN(avg_, 0, 1, _mmi); + FN_ASSIGN(avg_, 0, 2, _mmi); + FN_ASSIGN(avg_, 0, 3, _mmi); + + FN_ASSIGN(avg_, 1, 0, _mmi); + //FN_ASSIGN(avg_, 1, 1, _mmi);//FIXME + //FN_ASSIGN(avg_, 1, 2, _mmi);//FIXME + //FN_ASSIGN(avg_, 1, 3, _mmi);//FIXME + + FN_ASSIGN(avg_, 2, 0, _mmi); + //FN_ASSIGN(avg_, 2, 1, _mmi);//FIXME + //FN_ASSIGN(avg_, 2, 2, _mmi);//FIXME + //FN_ASSIGN(avg_, 2, 3, _mmi);//FIXME + + FN_ASSIGN(avg_, 3, 0, _mmi); + //FN_ASSIGN(avg_, 3, 1, _mmi);//FIXME + //FN_ASSIGN(avg_, 3, 2, _mmi);//FIXME + //FN_ASSIGN(avg_, 3, 3, _mmi);//FIXME + + dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = ff_put_no_rnd_vc1_chroma_mc8_mmi; + dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = ff_avg_no_rnd_vc1_chroma_mc8_mmi; + dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = ff_put_no_rnd_vc1_chroma_mc4_mmi; + dsp->avg_no_rnd_vc1_chroma_pixels_tab[1] = ff_avg_no_rnd_vc1_chroma_mc4_mmi; + } + + if (have_msa(cpu_flags)) { + dsp->vc1_inv_trans_8x8 = ff_vc1_inv_trans_8x8_msa; + dsp->vc1_inv_trans_4x8 = ff_vc1_inv_trans_4x8_msa; + dsp->vc1_inv_trans_8x4 = ff_vc1_inv_trans_8x4_msa; + + FN_ASSIGN(put_, 1, 1, _msa); + FN_ASSIGN(put_, 1, 2, _msa); + FN_ASSIGN(put_, 1, 3, _msa); + FN_ASSIGN(put_, 2, 1, _msa); + FN_ASSIGN(put_, 2, 2, _msa); + FN_ASSIGN(put_, 2, 3, _msa); + FN_ASSIGN(put_, 3, 1, _msa); + FN_ASSIGN(put_, 3, 2, _msa); + FN_ASSIGN(put_, 3, 3, _msa); + } } diff --git a/libavcodec/mips/videodsp_init.c b/libavcodec/mips/videodsp_init.c index 817040420b..07c23bcf7e 100644 --- a/libavcodec/mips/videodsp_init.c +++ b/libavcodec/mips/videodsp_init.c @@ -18,12 +18,12 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "config.h" #include "libavutil/attributes.h" #include "libavutil/mips/asmdefs.h" #include "libavcodec/videodsp.h" -#if HAVE_MSA static void prefetch_mips(uint8_t *mem, ptrdiff_t stride, int h) { register const uint8_t *p = mem; @@ -41,11 +41,11 @@ static void prefetch_mips(uint8_t *mem, ptrdiff_t stride, int h) : [stride] "r" (stride) ); } -#endif // #if HAVE_MSA av_cold void ff_videodsp_init_mips(VideoDSPContext *ctx, int bpc) { -#if HAVE_MSA - ctx->prefetch = prefetch_mips; -#endif // #if HAVE_MSA + int cpu_flags = av_get_cpu_flags(); + + if (have_msa(cpu_flags)) + ctx->prefetch = prefetch_mips; } diff --git a/libavcodec/mips/vp3dsp_init_mips.c b/libavcodec/mips/vp3dsp_init_mips.c index e183db35b6..4252ff790e 100644 --- a/libavcodec/mips/vp3dsp_init_mips.c +++ b/libavcodec/mips/vp3dsp_init_mips.c @@ -19,42 +19,32 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "config.h" #include "libavutil/attributes.h" #include "libavcodec/avcodec.h" #include "libavcodec/vp3dsp.h" #include "vp3dsp_mips.h" -#if HAVE_MSA -static av_cold void vp3dsp_init_msa(VP3DSPContext *c, int flags) +av_cold void ff_vp3dsp_init_mips(VP3DSPContext *c, int flags) { - c->put_no_rnd_pixels_l2 = ff_put_no_rnd_pixels_l2_msa; + int cpu_flags = av_get_cpu_flags(); - c->idct_add = ff_vp3_idct_add_msa; - c->idct_put = ff_vp3_idct_put_msa; - c->idct_dc_add = ff_vp3_idct_dc_add_msa; - c->v_loop_filter = ff_vp3_v_loop_filter_msa; - c->h_loop_filter = ff_vp3_h_loop_filter_msa; -} -#endif /* HAVE_MSA */ + if (have_mmi(cpu_flags)) { + c->put_no_rnd_pixels_l2 = ff_put_no_rnd_pixels_l2_mmi; -#if HAVE_MMI -static av_cold void vp3dsp_init_mmi(VP3DSPContext *c, int flags) -{ - c->put_no_rnd_pixels_l2 = ff_put_no_rnd_pixels_l2_mmi; + c->idct_add = ff_vp3_idct_add_mmi; + c->idct_put = ff_vp3_idct_put_mmi; + c->idct_dc_add = ff_vp3_idct_dc_add_mmi; + } - c->idct_add = ff_vp3_idct_add_mmi; - c->idct_put = ff_vp3_idct_put_mmi; - c->idct_dc_add = ff_vp3_idct_dc_add_mmi; -} -#endif /* HAVE_MMI */ + if (have_msa(cpu_flags)) { + c->put_no_rnd_pixels_l2 = ff_put_no_rnd_pixels_l2_msa; -av_cold void ff_vp3dsp_init_mips(VP3DSPContext *c, int flags) -{ -#if HAVE_MMI - vp3dsp_init_mmi(c, flags); -#endif /* HAVE_MMI */ -#if HAVE_MSA - vp3dsp_init_msa(c, flags); -#endif /* HAVE_MSA */ + c->idct_add = ff_vp3_idct_add_msa; + c->idct_put = ff_vp3_idct_put_msa; + c->idct_dc_add = ff_vp3_idct_dc_add_msa; + c->v_loop_filter = ff_vp3_v_loop_filter_msa; + c->h_loop_filter = ff_vp3_h_loop_filter_msa; + } } diff --git a/libavcodec/mips/vp8dsp_init_mips.c b/libavcodec/mips/vp8dsp_init_mips.c index 7fd8fb0d32..92d8c792cb 100644 --- a/libavcodec/mips/vp8dsp_init_mips.c +++ b/libavcodec/mips/vp8dsp_init_mips.c @@ -24,6 +24,7 @@ * VP8 compatible video decoder */ +#include "libavutil/mips/cpu.h" #include "config.h" #include "libavutil/attributes.h" #include "libavcodec/vp8dsp.h" @@ -71,132 +72,123 @@ dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = \ ff_put_vp8_pixels##SIZE##_msa; -#if HAVE_MSA -static av_cold void vp8dsp_init_msa(VP8DSPContext *dsp) -{ - dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_msa; - dsp->vp8_idct_add = ff_vp8_idct_add_msa; - dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_msa; - dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_msa; - dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_msa; - - VP8_MC_MIPS_FUNC(0, 16); - VP8_MC_MIPS_FUNC(1, 8); - VP8_MC_MIPS_FUNC(2, 4); - - VP8_BILINEAR_MC_MIPS_FUNC(0, 16); - VP8_BILINEAR_MC_MIPS_FUNC(1, 8); - VP8_BILINEAR_MC_MIPS_FUNC(2, 4); - - VP8_MC_MIPS_COPY(0, 16); - VP8_MC_MIPS_COPY(1, 8); - - dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_msa; - dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_msa; - dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_msa; - dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_msa; - - dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_msa; - dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_msa; - dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_msa; - dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_msa; - - dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_msa; - dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_msa; -} -#endif // #if HAVE_MSA -#if HAVE_MMI -static av_cold void vp8dsp_init_mmi(VP8DSPContext *dsp) -{ - dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmi; - dsp->vp8_luma_dc_wht_dc = ff_vp8_luma_dc_wht_dc_mmi; - dsp->vp8_idct_add = ff_vp8_idct_add_mmi; - dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_mmi; - dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_mmi; - dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_mmi; - - dsp->put_vp8_epel_pixels_tab[0][0][1] = ff_put_vp8_epel16_h4_mmi; - dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_mmi; - dsp->put_vp8_epel_pixels_tab[0][1][0] = ff_put_vp8_epel16_v4_mmi; - dsp->put_vp8_epel_pixels_tab[0][1][1] = ff_put_vp8_epel16_h4v4_mmi; - dsp->put_vp8_epel_pixels_tab[0][1][2] = ff_put_vp8_epel16_h6v4_mmi; - dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_mmi; - dsp->put_vp8_epel_pixels_tab[0][2][1] = ff_put_vp8_epel16_h4v6_mmi; - dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_mmi; - - dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_mmi; - dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_mmi; - dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_mmi; - dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_mmi; - dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_mmi; - dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_mmi; - dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_mmi; - dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_mmi; - - dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_mmi; - dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_mmi; - dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_mmi; - dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_mmi; - dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_mmi; - dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_mmi; - dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_mmi; - dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_mmi; - - dsp->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilinear16_h_mmi; - dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilinear16_h_mmi; - dsp->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilinear16_v_mmi; - dsp->put_vp8_bilinear_pixels_tab[0][1][1] = ff_put_vp8_bilinear16_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[0][1][2] = ff_put_vp8_bilinear16_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilinear16_v_mmi; - dsp->put_vp8_bilinear_pixels_tab[0][2][1] = ff_put_vp8_bilinear16_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilinear16_hv_mmi; - - dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilinear8_h_mmi; - dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilinear8_h_mmi; - dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilinear8_v_mmi; - dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilinear8_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilinear8_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilinear8_v_mmi; - dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilinear8_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilinear8_hv_mmi; - - dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilinear4_h_mmi; - dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilinear4_h_mmi; - dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilinear4_v_mmi; - dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilinear4_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilinear4_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilinear4_v_mmi; - dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilinear4_hv_mmi; - dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilinear4_hv_mmi; - - dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmi; - dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmi; - - dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmi; - dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmi; - - dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_mmi; - dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_mmi; - dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mmi; - dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mmi; - - dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_mmi; - dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_mmi; - dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmi; - dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmi; - - dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmi; - dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmi; -} -#endif /* HAVE_MMI */ av_cold void ff_vp8dsp_init_mips(VP8DSPContext *dsp) { -#if HAVE_MMI - vp8dsp_init_mmi(dsp); -#endif /* HAVE_MMI */ -#if HAVE_MSA - vp8dsp_init_msa(dsp); -#endif // #if HAVE_MSA + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_mmi; + dsp->vp8_luma_dc_wht_dc = ff_vp8_luma_dc_wht_dc_mmi; + dsp->vp8_idct_add = ff_vp8_idct_add_mmi; + dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_mmi; + dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_mmi; + dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_mmi; + + dsp->put_vp8_epel_pixels_tab[0][0][1] = ff_put_vp8_epel16_h4_mmi; + dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_mmi; + dsp->put_vp8_epel_pixels_tab[0][1][0] = ff_put_vp8_epel16_v4_mmi; + dsp->put_vp8_epel_pixels_tab[0][1][1] = ff_put_vp8_epel16_h4v4_mmi; + dsp->put_vp8_epel_pixels_tab[0][1][2] = ff_put_vp8_epel16_h6v4_mmi; + dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_mmi; + dsp->put_vp8_epel_pixels_tab[0][2][1] = ff_put_vp8_epel16_h4v6_mmi; + dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_mmi; + + dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_mmi; + dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_mmi; + dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_mmi; + dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_mmi; + dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_mmi; + dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_mmi; + dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_mmi; + dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_mmi; + + dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_mmi; + dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_mmi; + dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_mmi; + dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_mmi; + dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_mmi; + dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_mmi; + dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_mmi; + dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_mmi; + + dsp->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilinear16_h_mmi; + dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilinear16_h_mmi; + dsp->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilinear16_v_mmi; + dsp->put_vp8_bilinear_pixels_tab[0][1][1] = ff_put_vp8_bilinear16_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[0][1][2] = ff_put_vp8_bilinear16_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilinear16_v_mmi; + dsp->put_vp8_bilinear_pixels_tab[0][2][1] = ff_put_vp8_bilinear16_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilinear16_hv_mmi; + + dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilinear8_h_mmi; + dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilinear8_h_mmi; + dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilinear8_v_mmi; + dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilinear8_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilinear8_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilinear8_v_mmi; + dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilinear8_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilinear8_hv_mmi; + + dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilinear4_h_mmi; + dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilinear4_h_mmi; + dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilinear4_v_mmi; + dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilinear4_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilinear4_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilinear4_v_mmi; + dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilinear4_hv_mmi; + dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilinear4_hv_mmi; + + dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmi; + dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmi; + + dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmi; + dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmi; + + dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_mmi; + dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_mmi; + dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_mmi; + dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_mmi; + + dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_mmi; + dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_mmi; + dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmi; + dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmi; + + dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmi; + dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmi; + } + + if (have_msa(cpu_flags)) { + dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_msa; + dsp->vp8_idct_add = ff_vp8_idct_add_msa; + dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_msa; + dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_msa; + dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_msa; + + VP8_MC_MIPS_FUNC(0, 16); + VP8_MC_MIPS_FUNC(1, 8); + VP8_MC_MIPS_FUNC(2, 4); + + VP8_BILINEAR_MC_MIPS_FUNC(0, 16); + VP8_BILINEAR_MC_MIPS_FUNC(1, 8); + VP8_BILINEAR_MC_MIPS_FUNC(2, 4); + + VP8_MC_MIPS_COPY(0, 16); + VP8_MC_MIPS_COPY(1, 8); + + dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_msa; + dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_msa; + dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_msa; + dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_msa; + + dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_msa; + dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_msa; + dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_msa; + dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_msa; + + dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_msa; + dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_msa; + } } diff --git a/libavcodec/mips/vp9dsp_init_mips.c b/libavcodec/mips/vp9dsp_init_mips.c index 5990fa6952..5a8c599e7e 100644 --- a/libavcodec/mips/vp9dsp_init_mips.c +++ b/libavcodec/mips/vp9dsp_init_mips.c @@ -18,6 +18,7 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "config.h" #include "libavutil/common.h" #include "libavcodec/vp9dsp.h" @@ -209,10 +210,17 @@ static av_cold void vp9dsp_init_mmi(VP9DSPContext *dsp, int bpp) av_cold void ff_vp9dsp_init_mips(VP9DSPContext *dsp, int bpp) { +#if HAVE_MSA || HAVE_MMI + int cpu_flags = av_get_cpu_flags(); +#endif + #if HAVE_MMI - vp9dsp_init_mmi(dsp, bpp); -#endif // #if HAVE_MMI + if (have_mmi(cpu_flags)) + vp9dsp_init_mmi(dsp, bpp); +#endif + #if HAVE_MSA - vp9dsp_init_msa(dsp, bpp); -#endif // #if HAVE_MSA + if (have_msa(cpu_flags)) + vp9dsp_init_msa(dsp, bpp); +#endif } diff --git a/libavcodec/mips/wmv2dsp_init_mips.c b/libavcodec/mips/wmv2dsp_init_mips.c index 51dd2078d9..af1400731a 100644 --- a/libavcodec/mips/wmv2dsp_init_mips.c +++ b/libavcodec/mips/wmv2dsp_init_mips.c @@ -18,21 +18,17 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "config.h" #include "libavutil/attributes.h" #include "wmv2dsp_mips.h" -#if HAVE_MMI -static av_cold void wmv2dsp_init_mmi(WMV2DSPContext *c) -{ - c->idct_add = ff_wmv2_idct_add_mmi; - c->idct_put = ff_wmv2_idct_put_mmi; -} -#endif /* HAVE_MMI */ - av_cold void ff_wmv2dsp_init_mips(WMV2DSPContext *c) { -#if HAVE_MMI - wmv2dsp_init_mmi(c); -#endif /* HAVE_MMI */ + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + c->idct_add = ff_wmv2_idct_add_mmi; + c->idct_put = ff_wmv2_idct_put_mmi; + } } diff --git a/libavcodec/mips/wmv2dsp_mips.h b/libavcodec/mips/wmv2dsp_mips.h index 22894c505d..f7313460fb 100644 --- a/libavcodec/mips/wmv2dsp_mips.h +++ b/libavcodec/mips/wmv2dsp_mips.h @@ -23,7 +23,7 @@ #include "libavcodec/wmv2dsp.h" -void ff_wmv2_idct_add_mmi(uint8_t *dest, int line_size, int16_t *block); -void ff_wmv2_idct_put_mmi(uint8_t *dest, int line_size, int16_t *block); +void ff_wmv2_idct_add_mmi(uint8_t *dest, long int line_size, int16_t *block); +void ff_wmv2_idct_put_mmi(uint8_t *dest, long int line_size, int16_t *block); #endif /* AVCODEC_MIPS_WMV2DSP_MIPS_H */ diff --git a/libavcodec/mips/wmv2dsp_mmi.c b/libavcodec/mips/wmv2dsp_mmi.c index 1f6ccb299b..8796ebe195 100644 --- a/libavcodec/mips/wmv2dsp_mmi.c +++ b/libavcodec/mips/wmv2dsp_mmi.c @@ -95,7 +95,7 @@ static void wmv2_idct_col_mmi(short * b) b[56] = (a0 + a2 - a1 - a5 + 8192) >> 14; } -void ff_wmv2_idct_add_mmi(uint8_t *dest, int line_size, int16_t *block) +void ff_wmv2_idct_add_mmi(uint8_t *dest, long int line_size, int16_t *block) { int i; double ftmp[11]; @@ -212,7 +212,7 @@ void ff_wmv2_idct_add_mmi(uint8_t *dest, int line_size, int16_t *block) ); } -void ff_wmv2_idct_put_mmi(uint8_t *dest, int line_size, int16_t *block) +void ff_wmv2_idct_put_mmi(uint8_t *dest, long int line_size, int16_t *block) { int i; double ftmp[8]; diff --git a/libavcodec/mips/xvididct_init_mips.c b/libavcodec/mips/xvididct_init_mips.c index c1d82cc30c..ed545cfe17 100644 --- a/libavcodec/mips/xvididct_init_mips.c +++ b/libavcodec/mips/xvididct_init_mips.c @@ -18,28 +18,23 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/mips/cpu.h" #include "xvididct_mips.h" -#if HAVE_MMI -static av_cold void xvid_idct_init_mmi(IDCTDSPContext *c, AVCodecContext *avctx, +av_cold void ff_xvid_idct_init_mips(IDCTDSPContext *c, AVCodecContext *avctx, unsigned high_bit_depth) { - if (!high_bit_depth) { - if (avctx->idct_algo == FF_IDCT_AUTO || - avctx->idct_algo == FF_IDCT_XVID) { - c->idct_put = ff_xvid_idct_put_mmi; - c->idct_add = ff_xvid_idct_add_mmi; - c->idct = ff_xvid_idct_mmi; - c->perm_type = FF_IDCT_PERM_NONE; + int cpu_flags = av_get_cpu_flags(); + + if (have_mmi(cpu_flags)) { + if (!high_bit_depth) { + if (avctx->idct_algo == FF_IDCT_AUTO || + avctx->idct_algo == FF_IDCT_XVID) { + c->idct_put = ff_xvid_idct_put_mmi; + c->idct_add = ff_xvid_idct_add_mmi; + c->idct = ff_xvid_idct_mmi; + c->perm_type = FF_IDCT_PERM_NONE; + } } } } -#endif /* HAVE_MMI */ - -av_cold void ff_xvid_idct_init_mips(IDCTDSPContext *c, AVCodecContext *avctx, - unsigned high_bit_depth) -{ -#if HAVE_MMI - xvid_idct_init_mmi(c, avctx, high_bit_depth); -#endif /* HAVE_MMI */ -}