From patchwork Tue Apr 4 16:48:15 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Ronald S. Bultje" X-Patchwork-Id: 3288 Delivered-To: ffmpegpatchwork@gmail.com Received: by 10.103.44.195 with SMTP id s186csp275762vss; Tue, 4 Apr 2017 09:50:24 -0700 (PDT) X-Received: by 10.223.177.138 with SMTP id q10mr22887514wra.164.1491324624587; Tue, 04 Apr 2017 09:50:24 -0700 (PDT) Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id t14si25285073wrb.43.2017.04.04.09.50.24; Tue, 04 Apr 2017 09:50:24 -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=@gmail.com; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=gmail.com Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 25E2A68015B; Tue, 4 Apr 2017 19:50:13 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mail-qk0-f195.google.com (mail-qk0-f195.google.com [209.85.220.195]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id B0DBD680D1D for ; Tue, 4 Apr 2017 19:50:05 +0300 (EEST) Received: by mail-qk0-f195.google.com with SMTP id p22so24518354qka.0 for ; Tue, 04 Apr 2017 09:50:07 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=sOiEOI/wlQlh8IcLjJdyrsqDB4SeOfzp3hrMGEavYNk=; b=DWf0cK4wi21gSLPHTI+03fHgz1ezYoSGFhIlJVuP0ZcguRQtq/OdtJGUNyx1ZRwc8M CSkXpJZ1yaPKa3bOXzCRP2lZ6HaQynpzH5cNPOVzN+eK/coWsmJooCx/Lr02D+nUkNPU oQwdjUYTgJE31wUrjgdIv0aqBF4C+Rtj1Ql9Mx8jxXSAwuYPuRTUPFTM2p3L21Usx1d4 pHxRKzLYm2Z2XPTH4J208SAnDf4+XGbkoi+SbIf78TtkHvU9fO3sWC0xljwImZ+hWV5c MZuDLvL/9K294BwIx5CtMhnMkr65s6Q4N5RU1G6Uo3KxxxSqPCjm/LN+DxgYFBPz1MVJ Qr3w== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=sOiEOI/wlQlh8IcLjJdyrsqDB4SeOfzp3hrMGEavYNk=; b=I6RGG7o5V3WusYpV+exAlo/s95CX0HqNC8PCL0vdl1iQjVhhn6n8euF+oZXo5s1DDJ T4mBCBeZDeyvO4v6iFfXYe7UOsAJvlImaXJfw5ohWaYQ0h+lLHIj9ita6mW+SV7TOdna EikakUoBzhGmAqeoNB7uJpONNOxFNc5M8ZMlolPO8EH0bVh+o61ObPMtF9SnEqclwtTX ToObjWB1Ol0+1dsnTIvkHnM/Zozy3QIt+8gnuye0GY1yPcAk8Y4QtuaELnDO2MgxbgiK WdT/OoYW8dNQzDpmAwBjaJlxFS4TnaoHbgKIGuEewptmDb9WHZ5Z85WZC17lEs5MBOm3 lHaA== X-Gm-Message-State: AFeK/H2fC0oPbwdWpktEHt/I8UEb5NoCpg3RPyk/WCs5zA1bT53zKGed1nfRmfs1BSHBjg== X-Received: by 10.55.106.199 with SMTP id f190mr21595148qkc.151.1491324606623; Tue, 04 Apr 2017 09:50:06 -0700 (PDT) Received: from localhost.localdomain ([65.206.95.146]) by smtp.gmail.com with ESMTPSA id x19sm12205798qtc.23.2017.04.04.09.50.05 (version=TLS1 cipher=AES128-SHA bits=128/128); Tue, 04 Apr 2017 09:50:06 -0700 (PDT) From: "Ronald S. Bultje" To: ffmpeg-devel@ffmpeg.org Date: Tue, 4 Apr 2017 12:48:15 -0400 Message-Id: <1491324498-26655-3-git-send-email-rsbultje@gmail.com> X-Mailer: git-send-email 2.8.1 In-Reply-To: <1491324498-26655-1-git-send-email-rsbultje@gmail.com> References: <1491324498-26655-1-git-send-email-rsbultje@gmail.com> Subject: [FFmpeg-devel] [PATCH 3/6] cavs: convert idct from inline asm to yasm. 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: "Ronald S. Bultje" MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" --- libavcodec/x86/Makefile | 2 +- libavcodec/x86/cavsdsp.c | 165 ++------------------------------------------ libavcodec/x86/cavsidct.asm | 165 ++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 171 insertions(+), 161 deletions(-) create mode 100644 libavcodec/x86/cavsidct.asm diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index 0295a9f..56df530 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -48,7 +48,7 @@ OBJS-$(CONFIG_ADPCM_G722_DECODER) += x86/g722dsp_init.o OBJS-$(CONFIG_ADPCM_G722_ENCODER) += x86/g722dsp_init.o OBJS-$(CONFIG_ALAC_DECODER) += x86/alacdsp_init.o OBJS-$(CONFIG_APNG_DECODER) += x86/pngdsp_init.o -OBJS-$(CONFIG_CAVS_DECODER) += x86/cavsdsp.o +OBJS-$(CONFIG_CAVS_DECODER) += x86/cavsdsp.o x86/cavsidct.o OBJS-$(CONFIG_DCA_DECODER) += x86/dcadsp_init.o x86/synth_filter_init.o OBJS-$(CONFIG_DNXHD_ENCODER) += x86/dnxhdenc_init.o OBJS-$(CONFIG_HEVC_DECODER) += x86/hevcdsp_init.o diff --git a/libavcodec/x86/cavsdsp.c b/libavcodec/x86/cavsdsp.c index ecb9b23..add4536 100644 --- a/libavcodec/x86/cavsdsp.c +++ b/libavcodec/x86/cavsdsp.c @@ -34,172 +34,19 @@ #include "idctdsp.h" #include "config.h" -#if HAVE_MMX_INLINE -/* in/out: mma=mma+mmb, mmb=mmb-mma */ -#define SUMSUB_BA( a, b ) \ - "paddw "#b", "#a" \n\t"\ - "paddw "#b", "#b" \n\t"\ - "psubw "#a", "#b" \n\t" - -/***************************************************************************** - * - * inverse transform - * - ****************************************************************************/ - -static inline void cavs_idct8_1d(int16_t *block, uint64_t bias) -{ - __asm__ volatile( - "movq 112(%0), %%mm4 \n\t" /* mm4 = src7 */ - "movq 16(%0), %%mm5 \n\t" /* mm5 = src1 */ - "movq 80(%0), %%mm2 \n\t" /* mm2 = src5 */ - "movq 48(%0), %%mm7 \n\t" /* mm7 = src3 */ - "movq %%mm4, %%mm0 \n\t" - "movq %%mm5, %%mm3 \n\t" - "movq %%mm2, %%mm6 \n\t" - "movq %%mm7, %%mm1 \n\t" - - "paddw %%mm4, %%mm4 \n\t" /* mm4 = 2*src7 */ - "paddw %%mm3, %%mm3 \n\t" /* mm3 = 2*src1 */ - "paddw %%mm6, %%mm6 \n\t" /* mm6 = 2*src5 */ - "paddw %%mm1, %%mm1 \n\t" /* mm1 = 2*src3 */ - "paddw %%mm4, %%mm0 \n\t" /* mm0 = 3*src7 */ - "paddw %%mm3, %%mm5 \n\t" /* mm5 = 3*src1 */ - "paddw %%mm6, %%mm2 \n\t" /* mm2 = 3*src5 */ - "paddw %%mm1, %%mm7 \n\t" /* mm7 = 3*src3 */ - "psubw %%mm4, %%mm5 \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */ - "paddw %%mm6, %%mm7 \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */ - "psubw %%mm2, %%mm1 \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */ - "paddw %%mm0, %%mm3 \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */ - - "movq %%mm5, %%mm4 \n\t" - "movq %%mm7, %%mm6 \n\t" - "movq %%mm3, %%mm0 \n\t" - "movq %%mm1, %%mm2 \n\t" - SUMSUB_BA( %%mm7, %%mm5 ) /* mm7 = a0 + a1 mm5 = a0 - a1 */ - "paddw %%mm3, %%mm7 \n\t" /* mm7 = a0 + a1 + a3 */ - "paddw %%mm1, %%mm5 \n\t" /* mm5 = a0 - a1 + a2 */ - "paddw %%mm7, %%mm7 \n\t" - "paddw %%mm5, %%mm5 \n\t" - "paddw %%mm6, %%mm7 \n\t" /* mm7 = b4 */ - "paddw %%mm4, %%mm5 \n\t" /* mm5 = b5 */ - - SUMSUB_BA( %%mm1, %%mm3 ) /* mm1 = a3 + a2 mm3 = a3 - a2 */ - "psubw %%mm1, %%mm4 \n\t" /* mm4 = a0 - a2 - a3 */ - "movq %%mm4, %%mm1 \n\t" /* mm1 = a0 - a2 - a3 */ - "psubw %%mm6, %%mm3 \n\t" /* mm3 = a3 - a2 - a1 */ - "paddw %%mm1, %%mm1 \n\t" - "paddw %%mm3, %%mm3 \n\t" - "psubw %%mm2, %%mm1 \n\t" /* mm1 = b7 */ - "paddw %%mm0, %%mm3 \n\t" /* mm3 = b6 */ - - "movq 32(%0), %%mm2 \n\t" /* mm2 = src2 */ - "movq 96(%0), %%mm6 \n\t" /* mm6 = src6 */ - "movq %%mm2, %%mm4 \n\t" - "movq %%mm6, %%mm0 \n\t" - "psllw $2, %%mm4 \n\t" /* mm4 = 4*src2 */ - "psllw $2, %%mm6 \n\t" /* mm6 = 4*src6 */ - "paddw %%mm4, %%mm2 \n\t" /* mm2 = 5*src2 */ - "paddw %%mm6, %%mm0 \n\t" /* mm0 = 5*src6 */ - "paddw %%mm2, %%mm2 \n\t" - "paddw %%mm0, %%mm0 \n\t" - "psubw %%mm0, %%mm4 \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */ - "paddw %%mm2, %%mm6 \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */ - - "movq (%0), %%mm2 \n\t" /* mm2 = src0 */ - "movq 64(%0), %%mm0 \n\t" /* mm0 = src4 */ - SUMSUB_BA( %%mm0, %%mm2 ) /* mm0 = src0+src4 mm2 = src0-src4 */ - "psllw $3, %%mm0 \n\t" - "psllw $3, %%mm2 \n\t" - "paddw %1, %%mm0 \n\t" /* add rounding bias */ - "paddw %1, %%mm2 \n\t" /* add rounding bias */ - - SUMSUB_BA( %%mm6, %%mm0 ) /* mm6 = a4 + a6 mm0 = a4 - a6 */ - SUMSUB_BA( %%mm4, %%mm2 ) /* mm4 = a5 + a7 mm2 = a5 - a7 */ - SUMSUB_BA( %%mm7, %%mm6 ) /* mm7 = dst0 mm6 = dst7 */ - SUMSUB_BA( %%mm5, %%mm4 ) /* mm5 = dst1 mm4 = dst6 */ - SUMSUB_BA( %%mm3, %%mm2 ) /* mm3 = dst2 mm2 = dst5 */ - SUMSUB_BA( %%mm1, %%mm0 ) /* mm1 = dst3 mm0 = dst4 */ - :: "r"(block), "m"(bias) - ); -} - -#define SBUTTERFLY(a,b,t,n,m)\ - "mov" #m " " #a ", " #t " \n\t" /* abcd */\ - "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\ - "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */\ +#if HAVE_MMX_EXTERNAL -#define TRANSPOSE4(a,b,c,d,t)\ - SBUTTERFLY(a,b,t,wd,q) /* a=aebf t=cgdh */\ - SBUTTERFLY(c,d,b,wd,q) /* c=imjn b=kolp */\ - SBUTTERFLY(a,c,d,dq,q) /* a=aeim d=bfjn */\ - SBUTTERFLY(t,b,c,dq,q) /* t=cgko c=dhlp */ +void ff_cavs_idct8_mmx(int16_t *out, const int16_t *in); static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, ptrdiff_t stride) { - int i; LOCAL_ALIGNED(16, int16_t, b2, [64]); - - for(i=0; i<2; i++){ - cavs_idct8_1d(block + 4 * i, ff_pw_4.a); - - __asm__ volatile( - "psraw $3, %%mm7 \n\t" - "psraw $3, %%mm6 \n\t" - "psraw $3, %%mm5 \n\t" - "psraw $3, %%mm4 \n\t" - "psraw $3, %%mm3 \n\t" - "psraw $3, %%mm2 \n\t" - "psraw $3, %%mm1 \n\t" - "psraw $3, %%mm0 \n\t" - "movq %%mm7, (%0) \n\t" - TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 ) - "movq %%mm0, 8(%0) \n\t" - "movq %%mm6, 24(%0) \n\t" - "movq %%mm7, 40(%0) \n\t" - "movq %%mm4, 56(%0) \n\t" - "movq (%0), %%mm7 \n\t" - TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 ) - "movq %%mm7, (%0) \n\t" - "movq %%mm1, 16(%0) \n\t" - "movq %%mm0, 32(%0) \n\t" - "movq %%mm3, 48(%0) \n\t" - : - : "r"(b2 + 32 * i) - : "memory" - ); - } - - for(i=0; i<2; i++){ - cavs_idct8_1d(b2+4*i, ff_pw_64.a); - - __asm__ volatile( - "psraw $7, %%mm7 \n\t" - "psraw $7, %%mm6 \n\t" - "psraw $7, %%mm5 \n\t" - "psraw $7, %%mm4 \n\t" - "psraw $7, %%mm3 \n\t" - "psraw $7, %%mm2 \n\t" - "psraw $7, %%mm1 \n\t" - "psraw $7, %%mm0 \n\t" - "movq %%mm7, (%0) \n\t" - "movq %%mm5, 16(%0) \n\t" - "movq %%mm3, 32(%0) \n\t" - "movq %%mm1, 48(%0) \n\t" - "movq %%mm0, 64(%0) \n\t" - "movq %%mm2, 80(%0) \n\t" - "movq %%mm4, 96(%0) \n\t" - "movq %%mm6, 112(%0) \n\t" - :: "r"(b2+4*i) - : "memory" - ); - } - + ff_cavs_idct8_mmx(b2, block); ff_add_pixels_clamped(b2, dst, stride); } -#endif /* HAVE_MMX_INLINE */ +#endif /* HAVE_MMX_EXTERNAL */ #if (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE) @@ -529,12 +376,10 @@ static av_cold void cavsdsp_init_mmx(CAVSDSPContext *c, c->put_cavs_qpel_pixels_tab[1][0] = put_cavs_qpel8_mc00_mmx; c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_mmx; c->avg_cavs_qpel_pixels_tab[1][0] = avg_cavs_qpel8_mc00_mmx; -#endif -#if HAVE_MMX_INLINE c->cavs_idct8_add = cavs_idct8_add_mmx; c->idct_perm = FF_IDCT_PERM_TRANSPOSE; -#endif /* HAVE_MMX_INLINE */ +#endif /* HAVE_MMX_EXTERNAL */ } #define DSPFUNC(PFX, IDX, NUM, EXT) \ diff --git a/libavcodec/x86/cavsidct.asm b/libavcodec/x86/cavsidct.asm new file mode 100644 index 0000000..5421196 --- /dev/null +++ b/libavcodec/x86/cavsidct.asm @@ -0,0 +1,165 @@ +; Chinese AVS video (AVS1-P2, JiZhun profile) decoder +; Copyright (c) 2006 Stefan Gehrer +; +; MMX-optimized DSP functions, based on H.264 optimizations by +; Michael Niedermayer and Loren Merritt +; Conversion from gcc syntax to x264asm syntax with modifications +; by Ronald S. Bultje +; +; This file is part of FFmpeg. +; +; FFmpeg is free software; you can redistribute it and/or +; modify it under the terms of the GNU Lesser General Public +; License as published by the Free Software Foundation; either +; version 2.1 of the License, or (at your option) any later version. +; +; FFmpeg is distributed in the hope that it will be useful, +; but WITHOUT ANY WARRANTY; without even the implied warranty of +; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +; Lesser General Public License for more details. +; +; You should have received a copy of the GNU Lesser General Public License +; along with FFmpeg; if not, write to the Free Software Foundation, +; Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + +%include "libavutil/x86/x86util.asm" + +cextern pw_4 +cextern pw_64 + +SECTION .text + +%macro CAVS_IDCT8_1D 2 ; source, round + mova m4, [%1+7*16] ; m4 = src7 + mova m5, [%1+1*16] ; m5 = src1 + mova m2, [%1+5*16] ; m2 = src5 + mova m7, [%1+3*16] ; m7 = src3 + mova m0, m4 + mova m3, m5 + mova m6, m2 + mova m1, m7 + + paddw m4, m4 ; m4 = 2*src7 + paddw m3, m3 ; m3 = 2*src1 + paddw m6, m6 ; m6 = 2*src5 + paddw m1, m1 ; m1 = 2*src3 + paddw m0, m4 ; m0 = 3*src7 + paddw m5, m3 ; m5 = 3*src1 + paddw m2, m6 ; m2 = 3*src5 + paddw m7, m1 ; m7 = 3*src3 + psubw m5, m4 ; m5 = 3*src1 - 2*src7 = a0 + paddw m7, m6 ; m7 = 3*src3 - 2*src5 = a1 + psubw m1, m2 ; m1 = 2*src3 - 3*src5 = a2 + paddw m3, m0 ; m3 = 2*src1 - 3*src7 = a3 + + mova m4, m5 + mova m6, m7 + mova m0, m3 + mova m2, m1 + SUMSUB_BA w, 7, 5 ; m7 = a0 + a1, m5 = a0 - a1 + paddw m7, m3 ; m7 = a0 + a1 + a3 + paddw m5, m1 ; m5 = a0 - a1 + a2 + paddw m7, m7 + paddw m5, m5 + paddw m7, m6 ; m7 = b4 + paddw m5, m4 ; m5 = b5 + + SUMSUB_BA w, 1, 3 ; m1 = a3 + a2, m3 = a3 - a2 + psubw m4, m1 ; m4 = a0 - a2 - a3 + mova m1, m4 ; m1 = a0 - a2 - a3 + psubw m3, m6 ; m3 = a3 - a2 - a1 + paddw m1, m1 + paddw m3, m3 + psubw m1, m2 ; m1 = b7 + paddw m3, m0 ; m3 = b6 + + mova m2, [%1+2*16] ; m2 = src2 + mova m6, [%1+6*16] ; m6 = src6 + mova m4, m2 + mova m0, m6 + psllw m4, 2 ; m4 = 4*src2 + psllw m6, 2 ; m6 = 4*src6 + paddw m2, m4 ; m2 = 5*src2 + paddw m0, m6 ; m0 = 5*src6 + paddw m2, m2 + paddw m0, m0 + psubw m4, m0 ; m4 = 4*src2 - 10*src6 = a7 + paddw m6, m2 ; m6 = 4*src6 + 10*src2 = a6 + + mova m2, [%1+0*16] ; m2 = src0 + mova m0, [%1+4*16] ; m0 = src4 + SUMSUB_BA w, 0, 2 ; m0 = src0 + src4, m2 = src0 - src4 + psllw m0, 3 + psllw m2, 3 + paddw m0, %2 ; add rounding bias + paddw m2, %2 ; add rounding bias + + SUMSUB_BA w, 6, 0 ; m6 = a4 + a6, m0 = a4 - a6 + SUMSUB_BA w, 4, 2 ; m4 = a5 + a7, m2 = a5 - a7 + SUMSUB_BA w, 7, 6 ; m7 = dst0, m6 = dst7 + SUMSUB_BA w, 5, 4 ; m5 = dst1, m4 = dst6 + SUMSUB_BA w, 3, 2 ; m3 = dst2, m2 = dst5 + SUMSUB_BA w, 1, 0 ; m1 = dst3, m0 = dst4 +%endmacro + +INIT_MMX mmx +cglobal cavs_idct8, 2, 4, 8, 8 * 16, out, in, cnt, tmp + mov cntd, 2 + mov tmpq, rsp + +.loop_1: + CAVS_IDCT8_1D inq, [pw_4] + psraw m7, 3 + psraw m6, 3 + psraw m5, 3 + psraw m4, 3 + psraw m3, 3 + psraw m2, 3 + psraw m1, 3 + psraw m0, 3 + mova [tmpq], m7 + TRANSPOSE4x4W 0, 2, 4, 6, 7 + mova [tmpq+1*8], m0 + mova [tmpq+3*8], m2 + mova [tmpq+5*8], m4 + mova [tmpq+7*8], m6 + mova m7, [tmpq] + TRANSPOSE4x4W 7, 5, 3, 1, 0 + mova [tmpq+0*8], m7 + mova [tmpq+2*8], m5 + mova [tmpq+4*8], m3 + mova [tmpq+6*8], m1 + + add inq, mmsize + add tmpq, 64 + dec cntd + jg .loop_1 + + mov cntd, 2 + mov tmpq, rsp +.loop_2: + CAVS_IDCT8_1D tmpq, [pw_64] + psraw m7, 7 + psraw m6, 7 + psraw m5, 7 + psraw m4, 7 + psraw m3, 7 + psraw m2, 7 + psraw m1, 7 + psraw m0, 7 + + mova [outq+0*16], m7 + mova [outq+1*16], m5 + mova [outq+2*16], m3 + mova [outq+3*16], m1 + mova [outq+4*16], m0 + mova [outq+5*16], m2 + mova [outq+6*16], m4 + mova [outq+7*16], m6 + + add outq, mmsize + add tmpq, mmsize + dec cntd + jg .loop_2 + + RET