From patchwork Tue Dec 14 13:33:14 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: =?utf-8?b?6ZmI5piK?= X-Patchwork-Id: 32486 Delivered-To: ffmpegpatchwork2@gmail.com Received: by 2002:a6b:cd86:0:0:0:0:0 with SMTP id d128csp6965632iog; Tue, 14 Dec 2021 05:34:39 -0800 (PST) X-Google-Smtp-Source: ABdhPJx/7OmDfoefTWdQDVf4JnFfIMkGT6yvlyoNH3CVxRORdF2o2FH3cl9Epk9Wjbz8i6DMldvZ X-Received: by 2002:a05:6402:d73:: with SMTP id ec51mr7795815edb.175.1639488878923; Tue, 14 Dec 2021 05:34:38 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1639488878; cv=none; d=google.com; s=arc-20160816; b=yOs8SmAqcCZaR6x8tyW9K3eT/YgclI8Ss25zGb+AJsYe6EhPyz1kRvsRJ5UW/Y2QO6 m/MaZAIrx3u1vn/RsCAoKP+eZ0SQ8W1Svqm9bcQvZl4PpvVG1KyrMGj/Live/Vx3RWMZ OuUuJzhhNsNKNguf8wt0dGFLu+BvjBUgbOT0P19f6xIml3RMp4JvovvVgc1bmzuo5t6L y38F6ipd3qgJdHleAau0qNDkT4AwzbC/Gn1sYY5ProzHkSvk6hQ6yeMtdbfHqv9wkXa7 jat1fqqBaDjmda9fTbWB8tf/4RPz89iBh2rRe4ZCjKZ+H8rcy+aShXBkqbFPkuUlLTN/ RxRw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:mime-version:references:in-reply-to :message-id:date:to:from:delivered-to; bh=zm7BotcLoypr49oyv5CPH03K6va+Zj4BpRO9c8QxMjw=; b=RVIOaDKVJcU44g9T8+WWJu+j7EQkQgWdTMI5n80A/2YCDnitbxczJE6o6yVXLy6Ew8 sZU59+9ZUmB00id1MkXC1sUGLL336Nmq6y0TCzsJ5X/9Sm9zj+5WS771Nwu1URt33jGb emx2eioIIszKLDRKIemF+/JO/LBRRBzaJfJ0dJ8CBZ53OoS+4CkPE6g8p8stV+F34Fyl CojQ94TkbH2mbyV2I7VW0SU+bH80/w5zALLmFFoGCYZoMczpZmqGpR6TexuVjBP5a34H OcDH8Csctww3g2ANZ6xUFZ7r96cWtlSjHL8eQuohGexMOCEPGjqhS/j4ZgQ+LZa6H4gh FtfQ== ARC-Authentication-Results: i=1; mx.google.com; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id hd13si23457710ejc.148.2021.12.14.05.34.38; Tue, 14 Dec 2021 05:34:38 -0800 (PST) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 0E4C068AEEA; Tue, 14 Dec 2021 15:33:58 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from loongson.cn (mail.loongson.cn [114.242.206.163]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 24B8968AAD7 for ; Tue, 14 Dec 2021 15:33:46 +0200 (EET) Received: from localhost (unknown [36.33.26.144]) by mail.loongson.cn (Coremail) with SMTP id AQAAf9Dx7Nw4nbhhlqcAAA--.3411S3; Tue, 14 Dec 2021 21:33:44 +0800 (CST) From: Hao Chen To: ffmpeg-devel@ffmpeg.org Date: Tue, 14 Dec 2021 21:33:14 +0800 Message-Id: <20211214133316.8978-6-chenhao@loongson.cn> X-Mailer: git-send-email 2.20.1 In-Reply-To: <20211214133316.8978-1-chenhao@loongson.cn> References: <20211214133316.8978-1-chenhao@loongson.cn> MIME-Version: 1.0 X-CM-TRANSID: AQAAf9Dx7Nw4nbhhlqcAAA--.3411S3 X-Coremail-Antispam: 1UD129KBjvAXoWfWF13GFyxur4ktF1ktrWxCrg_yoW8trWDXo WUt392vr97Gw1Ivr95Ar9Yy3W8Cw43ur4UAw42qwsFya45Xa4qyrZ0kw4fJr17Krs7Wa43 Cry5XFy3ZrWFqr1Dn29KB7ZKAUJUUUU8529EdanIXcx71UUUUU7v73VFW2AGmfu7bjvjm3 AaLaJ3UjIYCTnIWjp_UUUY87k0a2IF6w4kM7kC6x804xWl14x267AKxVWUJVW8JwAFc2x0 x2IEx4CE42xK8VAvwI8IcIk0rVWrJVCq3wAFIxvE14AKwVWUJVWUGwA2ocxC64kIII0Yj4 1l84x0c7CEw4AK67xGY2AK021l84ACjcxK6xIIjxv20xvE14v26ryj6F1UM28EF7xvwVC0 I7IYx2IY6xkF7I0E14v26F4j6r4UJwA2z4x0Y4vEx4A2jsIE14v26rxl6s0DM28EF7xvwV C2z280aVCY1x0267AKxVW0oVCq3wAS0I0E0xvYzxvE52x082IY62kv0487Mc02F40EFcxC 0VAKzVAqx4xG6I80ewAv7VC0I7IYx2IY67AKxVWUtVWrXwAv7VC2z280aVAFwI0_Gr1j6F 4UJwAm72CE4IkC6x0Yz7v_Jr0_Gr1lF7xvr2IYc2Ij64vIr41lc2xSY4AK67AK6ry8MxAI w28IcxkI7VAKI48JMxC20s026xCaFVCjc4AY6r1j6r4UMI8I3I0E5I8CrVAFwI0_Jr0_Jr 4lx2IqxVCjr7xvwVAFwI0_JrI_JrWlx4CE17CEb7AF67AKxVWUXVWUAwCIc40Y0x0EwIxG rwCI42IY6xIIjxv20xvE14v26r4j6ryUMIIF0xvE2Ix0cI8IcVCY1x0267AKxVW8JVWxJw CI42IY6xAIw20EY4v20xvaj40_Jr0_JF4lIxAIcVC2z280aVAFwI0_Gr0_Cr1lIxAIcVC2 z280aVCY1x0267AKxVW8JVW8JrUvcSsGvfC2KfnxnUUI43ZEXa7IU5tl1PUUUUU== X-CM-SenderInfo: hfkh0xtdr6z05rqj20fqof0/ Subject: [FFmpeg-devel] [PATCH v2 5/7] avcodec: [loongarch] Optimize h264idct with LASX. X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches Cc: Lu Wang Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" X-TUID: qB8rmJEZtOYZ From: Lu Wang ./ffmpeg -i ../1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -y /dev/null -an before:282 after :293 Change-Id: Ia8889935a6359630dd5dbb61263287f1cb24a0a4 --- libavcodec/loongarch/Makefile | 3 +- libavcodec/loongarch/h264dsp_init_loongarch.c | 15 + libavcodec/loongarch/h264dsp_lasx.h | 23 + libavcodec/loongarch/h264idct_lasx.c | 498 ++++++++++++++++++ 4 files changed, 538 insertions(+), 1 deletion(-) create mode 100644 libavcodec/loongarch/h264idct_lasx.c diff --git a/libavcodec/loongarch/Makefile b/libavcodec/loongarch/Makefile index df43151dbd..242a2be290 100644 --- a/libavcodec/loongarch/Makefile +++ b/libavcodec/loongarch/Makefile @@ -3,4 +3,5 @@ OBJS-$(CONFIG_H264QPEL) += loongarch/h264qpel_init_loongarch.o OBJS-$(CONFIG_H264DSP) += loongarch/h264dsp_init_loongarch.o LASX-OBJS-$(CONFIG_H264CHROMA) += loongarch/h264chroma_lasx.o LASX-OBJS-$(CONFIG_H264QPEL) += loongarch/h264qpel_lasx.o -LASX-OBJS-$(CONFIG_H264DSP) += loongarch/h264dsp_lasx.o +LASX-OBJS-$(CONFIG_H264DSP) += loongarch/h264dsp_lasx.o \ + loongarch/h264idct_lasx.o diff --git a/libavcodec/loongarch/h264dsp_init_loongarch.c b/libavcodec/loongarch/h264dsp_init_loongarch.c index ddc0877a74..0985c2fe8a 100644 --- a/libavcodec/loongarch/h264dsp_init_loongarch.c +++ b/libavcodec/loongarch/h264dsp_init_loongarch.c @@ -53,6 +53,21 @@ av_cold void ff_h264dsp_init_loongarch(H264DSPContext *c, const int bit_depth, c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16_8_lasx; c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_8_lasx; c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels4_8_lasx; + + c->h264_idct_add = ff_h264_idct_add_lasx; + c->h264_idct8_add = ff_h264_idct8_addblk_lasx; + c->h264_idct_dc_add = ff_h264_idct4x4_addblk_dc_lasx; + c->h264_idct8_dc_add = ff_h264_idct8_dc_addblk_lasx; + c->h264_idct_add16 = ff_h264_idct_add16_lasx; + c->h264_idct8_add4 = ff_h264_idct8_add4_lasx; + + if (chroma_format_idc <= 1) + c->h264_idct_add8 = ff_h264_idct_add8_lasx; + else + c->h264_idct_add8 = ff_h264_idct_add8_422_lasx; + + c->h264_idct_add16intra = ff_h264_idct_add16_intra_lasx; + c->h264_luma_dc_dequant_idct = ff_h264_deq_idct_luma_dc_lasx; } } } diff --git a/libavcodec/loongarch/h264dsp_lasx.h b/libavcodec/loongarch/h264dsp_lasx.h index 538c14c936..bfd567fffa 100644 --- a/libavcodec/loongarch/h264dsp_lasx.h +++ b/libavcodec/loongarch/h264dsp_lasx.h @@ -65,4 +65,27 @@ void ff_weight_h264_pixels4_8_lasx(uint8_t *src, ptrdiff_t stride, void ff_h264_add_pixels4_8_lasx(uint8_t *_dst, int16_t *_src, int stride); void ff_h264_add_pixels8_8_lasx(uint8_t *_dst, int16_t *_src, int stride); +void ff_h264_idct_add_lasx(uint8_t *dst, int16_t *src, int32_t dst_stride); +void ff_h264_idct8_addblk_lasx(uint8_t *dst, int16_t *src, int32_t dst_stride); +void ff_h264_idct4x4_addblk_dc_lasx(uint8_t *dst, int16_t *src, + int32_t dst_stride); +void ff_h264_idct8_dc_addblk_lasx(uint8_t *dst, int16_t *src, + int32_t dst_stride); +void ff_h264_idct_add16_lasx(uint8_t *dst, const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]); +void ff_h264_idct8_add4_lasx(uint8_t *dst, const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]); +void ff_h264_idct_add8_lasx(uint8_t **dst, const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]); +void ff_h264_idct_add8_422_lasx(uint8_t **dst, const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]); +void ff_h264_idct_add16_intra_lasx(uint8_t *dst, const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]); +void ff_h264_deq_idct_luma_dc_lasx(int16_t *dst, int16_t *src, + int32_t de_qval); #endif // #ifndef AVCODEC_LOONGARCH_H264DSP_LASX_H diff --git a/libavcodec/loongarch/h264idct_lasx.c b/libavcodec/loongarch/h264idct_lasx.c new file mode 100644 index 0000000000..46bd3b74d5 --- /dev/null +++ b/libavcodec/loongarch/h264idct_lasx.c @@ -0,0 +1,498 @@ +/* + * Loongson LASX optimized h264dsp + * + * Copyright (c) 2021 Loongson Technology Corporation Limited + * Contributed by Shiyou Yin + * Xiwei Gu + * + * 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/loongarch/loongson_intrinsics.h" +#include "h264dsp_lasx.h" +#include "libavcodec/bit_depth_template.c" + +#define AVC_ITRANS_H(in0, in1, in2, in3, out0, out1, out2, out3) \ +{ \ + __m256i tmp0_m, tmp1_m, tmp2_m, tmp3_m; \ + \ + tmp0_m = __lasx_xvadd_h(in0, in2); \ + tmp1_m = __lasx_xvsub_h(in0, in2); \ + tmp2_m = __lasx_xvsrai_h(in1, 1); \ + tmp2_m = __lasx_xvsub_h(tmp2_m, in3); \ + tmp3_m = __lasx_xvsrai_h(in3, 1); \ + tmp3_m = __lasx_xvadd_h(in1, tmp3_m); \ + \ + LASX_BUTTERFLY_4_H(tmp0_m, tmp1_m, tmp2_m, tmp3_m, \ + out0, out1, out2, out3); \ +} + +void ff_h264_idct_add_lasx(uint8_t *dst, int16_t *src, int32_t dst_stride) +{ + __m256i src0_m, src1_m, src2_m, src3_m; + __m256i dst0_m, dst1_m; + __m256i hres0, hres1, hres2, hres3, vres0, vres1, vres2, vres3; + __m256i inp0_m, inp1_m, res0_m, src1, src3; + __m256i src0 = __lasx_xvld(src, 0); + __m256i src2 = __lasx_xvld(src, 16); + __m256i zero = __lasx_xvldi(0); + int32_t dst_stride_2x = dst_stride << 1; + int32_t dst_stride_3x = dst_stride_2x + dst_stride; + + __lasx_xvst(zero, src, 0); + DUP2_ARG2(__lasx_xvilvh_d, src0, src0, src2, src2, src1, src3); + AVC_ITRANS_H(src0, src1, src2, src3, hres0, hres1, hres2, hres3); + LASX_TRANSPOSE4x4_H(hres0, hres1, hres2, hres3, hres0, hres1, hres2, hres3); + AVC_ITRANS_H(hres0, hres1, hres2, hres3, vres0, vres1, vres2, vres3); + DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x, + dst, dst_stride_3x, src0_m, src1_m, src2_m, src3_m); + DUP4_ARG2(__lasx_xvld, dst, 0, dst + dst_stride, 0, dst + dst_stride_2x, + 0, dst + dst_stride_3x, 0, src0_m, src1_m, src2_m, src3_m); + DUP2_ARG2(__lasx_xvilvl_d, vres1, vres0, vres3, vres2, inp0_m, inp1_m); + inp0_m = __lasx_xvpermi_q(inp1_m, inp0_m, 0x20); + inp0_m = __lasx_xvsrari_h(inp0_m, 6); + DUP2_ARG2(__lasx_xvilvl_w, src1_m, src0_m, src3_m, src2_m, dst0_m, dst1_m); + dst0_m = __lasx_xvilvl_d(dst1_m, dst0_m); + res0_m = __lasx_vext2xv_hu_bu(dst0_m); + res0_m = __lasx_xvadd_h(res0_m, inp0_m); + res0_m = __lasx_xvclip255_h(res0_m); + dst0_m = __lasx_xvpickev_b(res0_m, res0_m); + __lasx_xvstelm_w(dst0_m, dst, 0, 0); + __lasx_xvstelm_w(dst0_m, dst + dst_stride, 0, 1); + __lasx_xvstelm_w(dst0_m, dst + dst_stride_2x, 0, 4); + __lasx_xvstelm_w(dst0_m, dst + dst_stride_3x, 0, 5); +} + +void ff_h264_idct8_addblk_lasx(uint8_t *dst, int16_t *src, + int32_t dst_stride) +{ + __m256i src0, src1, src2, src3, src4, src5, src6, src7; + __m256i vec0, vec1, vec2, vec3; + __m256i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + __m256i res0, res1, res2, res3, res4, res5, res6, res7; + __m256i dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; + __m256i zero = __lasx_xvldi(0); + int32_t dst_stride_2x = dst_stride << 1; + int32_t dst_stride_4x = dst_stride << 2; + int32_t dst_stride_3x = dst_stride_2x + dst_stride; + + src[0] += 32; + DUP4_ARG2(__lasx_xvld, src, 0, src, 16, src, 32, src, 48, + src0, src1, src2, src3); + DUP4_ARG2(__lasx_xvld, src, 64, src, 80, src, 96, src, 112, + src4, src5, src6, src7); + __lasx_xvst(zero, src, 0); + __lasx_xvst(zero, src, 32); + __lasx_xvst(zero, src, 64); + __lasx_xvst(zero, src, 96); + + vec0 = __lasx_xvadd_h(src0, src4); + vec1 = __lasx_xvsub_h(src0, src4); + vec2 = __lasx_xvsrai_h(src2, 1); + vec2 = __lasx_xvsub_h(vec2, src6); + vec3 = __lasx_xvsrai_h(src6, 1); + vec3 = __lasx_xvadd_h(src2, vec3); + + LASX_BUTTERFLY_4_H(vec0, vec1, vec2, vec3, tmp0, tmp1, tmp2, tmp3); + + vec0 = __lasx_xvsrai_h(src7, 1); + vec0 = __lasx_xvsub_h(src5, vec0); + vec0 = __lasx_xvsub_h(vec0, src3); + vec0 = __lasx_xvsub_h(vec0, src7); + + vec1 = __lasx_xvsrai_h(src3, 1); + vec1 = __lasx_xvsub_h(src1, vec1); + vec1 = __lasx_xvadd_h(vec1, src7); + vec1 = __lasx_xvsub_h(vec1, src3); + + vec2 = __lasx_xvsrai_h(src5, 1); + vec2 = __lasx_xvsub_h(vec2, src1); + vec2 = __lasx_xvadd_h(vec2, src7); + vec2 = __lasx_xvadd_h(vec2, src5); + + vec3 = __lasx_xvsrai_h(src1, 1); + vec3 = __lasx_xvadd_h(src3, vec3); + vec3 = __lasx_xvadd_h(vec3, src5); + vec3 = __lasx_xvadd_h(vec3, src1); + + tmp4 = __lasx_xvsrai_h(vec3, 2); + tmp4 = __lasx_xvadd_h(tmp4, vec0); + tmp5 = __lasx_xvsrai_h(vec2, 2); + tmp5 = __lasx_xvadd_h(tmp5, vec1); + tmp6 = __lasx_xvsrai_h(vec1, 2); + tmp6 = __lasx_xvsub_h(tmp6, vec2); + tmp7 = __lasx_xvsrai_h(vec0, 2); + tmp7 = __lasx_xvsub_h(vec3, tmp7); + + LASX_BUTTERFLY_8_H(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, + res0, res1, res2, res3, res4, res5, res6, res7); + LASX_TRANSPOSE8x8_H(res0, res1, res2, res3, res4, res5, res6, res7, + res0, res1, res2, res3, res4, res5, res6, res7); + + DUP4_ARG1(__lasx_vext2xv_w_h, res0, res1, res2, res3, + tmp0, tmp1, tmp2, tmp3); + DUP4_ARG1(__lasx_vext2xv_w_h, res4, res5, res6, res7, + tmp4, tmp5, tmp6, tmp7); + vec0 = __lasx_xvadd_w(tmp0, tmp4); + vec1 = __lasx_xvsub_w(tmp0, tmp4); + + vec2 = __lasx_xvsrai_w(tmp2, 1); + vec2 = __lasx_xvsub_w(vec2, tmp6); + vec3 = __lasx_xvsrai_w(tmp6, 1); + vec3 = __lasx_xvadd_w(vec3, tmp2); + + tmp0 = __lasx_xvadd_w(vec0, vec3); + tmp2 = __lasx_xvadd_w(vec1, vec2); + tmp4 = __lasx_xvsub_w(vec1, vec2); + tmp6 = __lasx_xvsub_w(vec0, vec3); + + vec0 = __lasx_xvsrai_w(tmp7, 1); + vec0 = __lasx_xvsub_w(tmp5, vec0); + vec0 = __lasx_xvsub_w(vec0, tmp3); + vec0 = __lasx_xvsub_w(vec0, tmp7); + + vec1 = __lasx_xvsrai_w(tmp3, 1); + vec1 = __lasx_xvsub_w(tmp1, vec1); + vec1 = __lasx_xvadd_w(vec1, tmp7); + vec1 = __lasx_xvsub_w(vec1, tmp3); + + vec2 = __lasx_xvsrai_w(tmp5, 1); + vec2 = __lasx_xvsub_w(vec2, tmp1); + vec2 = __lasx_xvadd_w(vec2, tmp7); + vec2 = __lasx_xvadd_w(vec2, tmp5); + + vec3 = __lasx_xvsrai_w(tmp1, 1); + vec3 = __lasx_xvadd_w(tmp3, vec3); + vec3 = __lasx_xvadd_w(vec3, tmp5); + vec3 = __lasx_xvadd_w(vec3, tmp1); + + tmp1 = __lasx_xvsrai_w(vec3, 2); + tmp1 = __lasx_xvadd_w(tmp1, vec0); + tmp3 = __lasx_xvsrai_w(vec2, 2); + tmp3 = __lasx_xvadd_w(tmp3, vec1); + tmp5 = __lasx_xvsrai_w(vec1, 2); + tmp5 = __lasx_xvsub_w(tmp5, vec2); + tmp7 = __lasx_xvsrai_w(vec0, 2); + tmp7 = __lasx_xvsub_w(vec3, tmp7); + + LASX_BUTTERFLY_4_W(tmp0, tmp2, tmp5, tmp7, res0, res1, res6, res7); + LASX_BUTTERFLY_4_W(tmp4, tmp6, tmp1, tmp3, res2, res3, res4, res5); + + DUP4_ARG2(__lasx_xvsrai_w, res0, 6, res1, 6, res2, 6, res3, 6, + res0, res1, res2, res3); + DUP4_ARG2(__lasx_xvsrai_w, res4, 6, res5, 6, res6, 6, res7, 6, + res4, res5, res6, res7); + DUP4_ARG2(__lasx_xvpickev_h, res1, res0, res3, res2, res5, res4, res7, + res6, res0, res1, res2, res3); + DUP4_ARG2(__lasx_xvpermi_d, res0, 0xd8, res1, 0xd8, res2, 0xd8, res3, 0xd8, + res0, res1, res2, res3); + + DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x, + dst, dst_stride_3x, dst0, dst1, dst2, dst3); + dst += dst_stride_4x; + DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x, + dst, dst_stride_3x, dst4, dst5, dst6, dst7); + dst -= dst_stride_4x; + DUP4_ARG2(__lasx_xvilvl_b, zero, dst0, zero, dst1, zero, dst2, zero, dst3, + dst0, dst1, dst2, dst3); + DUP4_ARG2(__lasx_xvilvl_b, zero, dst4, zero, dst5, zero, dst6, zero, dst7, + dst4, dst5, dst6, dst7); + DUP4_ARG3(__lasx_xvpermi_q, dst1, dst0, 0x20, dst3, dst2, 0x20, dst5, + dst4, 0x20, dst7, dst6, 0x20, dst0, dst1, dst2, dst3); + res0 = __lasx_xvadd_h(res0, dst0); + res1 = __lasx_xvadd_h(res1, dst1); + res2 = __lasx_xvadd_h(res2, dst2); + res3 = __lasx_xvadd_h(res3, dst3); + DUP4_ARG1(__lasx_xvclip255_h, res0, res1, res2, res3, res0, res1, + res2, res3); + DUP2_ARG2(__lasx_xvpickev_b, res1, res0, res3, res2, res0, res1); + __lasx_xvstelm_d(res0, dst, 0, 0); + __lasx_xvstelm_d(res0, dst + dst_stride, 0, 2); + __lasx_xvstelm_d(res0, dst + dst_stride_2x, 0, 1); + __lasx_xvstelm_d(res0, dst + dst_stride_3x, 0, 3); + dst += dst_stride_4x; + __lasx_xvstelm_d(res1, dst, 0, 0); + __lasx_xvstelm_d(res1, dst + dst_stride, 0, 2); + __lasx_xvstelm_d(res1, dst + dst_stride_2x, 0, 1); + __lasx_xvstelm_d(res1, dst + dst_stride_3x, 0, 3); +} + +void ff_h264_idct4x4_addblk_dc_lasx(uint8_t *dst, int16_t *src, + int32_t dst_stride) +{ + const int16_t dc = (src[0] + 32) >> 6; + int32_t dst_stride_2x = dst_stride << 1; + int32_t dst_stride_3x = dst_stride_2x + dst_stride; + __m256i pred, out; + __m256i src0, src1, src2, src3; + __m256i input_dc = __lasx_xvreplgr2vr_h(dc); + + src[0] = 0; + DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x, + dst, dst_stride_3x, src0, src1, src2, src3); + DUP2_ARG2(__lasx_xvilvl_w, src1, src0, src3, src2, src0, src1); + + pred = __lasx_xvpermi_q(src0, src1, 0x02); + pred = __lasx_xvaddw_h_h_bu(input_dc, pred); + pred = __lasx_xvclip255_h(pred); + out = __lasx_xvpickev_b(pred, pred); + __lasx_xvstelm_w(out, dst, 0, 0); + __lasx_xvstelm_w(out, dst + dst_stride, 0, 1); + __lasx_xvstelm_w(out, dst + dst_stride_2x, 0, 4); + __lasx_xvstelm_w(out, dst + dst_stride_3x, 0, 5); +} + +void ff_h264_idct8_dc_addblk_lasx(uint8_t *dst, int16_t *src, + int32_t dst_stride) +{ + int32_t dc_val; + int32_t dst_stride_2x = dst_stride << 1; + int32_t dst_stride_4x = dst_stride << 2; + int32_t dst_stride_3x = dst_stride_2x + dst_stride; + __m256i dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; + __m256i dc; + + dc_val = (src[0] + 32) >> 6; + dc = __lasx_xvreplgr2vr_h(dc_val); + + src[0] = 0; + + DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x, + dst, dst_stride_3x, dst0, dst1, dst2, dst3); + dst += dst_stride_4x; + DUP4_ARG2(__lasx_xvldx, dst, 0, dst, dst_stride, dst, dst_stride_2x, + dst, dst_stride_3x, dst4, dst5, dst6, dst7); + dst -= dst_stride_4x; + DUP4_ARG1(__lasx_vext2xv_hu_bu, dst0, dst1, dst2, dst3, + dst0, dst1, dst2, dst3); + DUP4_ARG1(__lasx_vext2xv_hu_bu, dst4, dst5, dst6, dst7, + dst4, dst5, dst6, dst7); + DUP4_ARG3(__lasx_xvpermi_q, dst1, dst0, 0x20, dst3, dst2, 0x20, dst5, + dst4, 0x20, dst7, dst6, 0x20, dst0, dst1, dst2, dst3); + dst0 = __lasx_xvadd_h(dst0, dc); + dst1 = __lasx_xvadd_h(dst1, dc); + dst2 = __lasx_xvadd_h(dst2, dc); + dst3 = __lasx_xvadd_h(dst3, dc); + DUP4_ARG1(__lasx_xvclip255_h, dst0, dst1, dst2, dst3, + dst0, dst1, dst2, dst3); + DUP2_ARG2(__lasx_xvpickev_b, dst1, dst0, dst3, dst2, dst0, dst1); + __lasx_xvstelm_d(dst0, dst, 0, 0); + __lasx_xvstelm_d(dst0, dst + dst_stride, 0, 2); + __lasx_xvstelm_d(dst0, dst + dst_stride_2x, 0, 1); + __lasx_xvstelm_d(dst0, dst + dst_stride_3x, 0, 3); + dst += dst_stride_4x; + __lasx_xvstelm_d(dst1, dst, 0, 0); + __lasx_xvstelm_d(dst1, dst + dst_stride, 0, 2); + __lasx_xvstelm_d(dst1, dst + dst_stride_2x, 0, 1); + __lasx_xvstelm_d(dst1, dst + dst_stride_3x, 0, 3); +} + +void ff_h264_idct_add16_lasx(uint8_t *dst, + const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]) +{ + int32_t i; + + for (i = 0; i < 16; i++) { + int32_t nnz = nzc[scan8[i]]; + + if (nnz) { + if (nnz == 1 && ((dctcoef *) block)[i * 16]) + ff_h264_idct4x4_addblk_dc_lasx(dst + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + else + ff_h264_idct_add_lasx(dst + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + } + } +} + +void ff_h264_idct8_add4_lasx(uint8_t *dst, const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]) +{ + int32_t cnt; + + for (cnt = 0; cnt < 16; cnt += 4) { + int32_t nnz = nzc[scan8[cnt]]; + + if (nnz) { + if (nnz == 1 && ((dctcoef *) block)[cnt * 16]) + ff_h264_idct8_dc_addblk_lasx(dst + blk_offset[cnt], + block + cnt * 16 * sizeof(pixel), + dst_stride); + else + ff_h264_idct8_addblk_lasx(dst + blk_offset[cnt], + block + cnt * 16 * sizeof(pixel), + dst_stride); + } + } +} + + +void ff_h264_idct_add8_lasx(uint8_t **dst, + const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]) +{ + int32_t i; + + for (i = 16; i < 20; i++) { + if (nzc[scan8[i]]) + ff_h264_idct_add_lasx(dst[0] + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + else if (((dctcoef *) block)[i * 16]) + ff_h264_idct4x4_addblk_dc_lasx(dst[0] + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + } + for (i = 32; i < 36; i++) { + if (nzc[scan8[i]]) + ff_h264_idct_add_lasx(dst[1] + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + else if (((dctcoef *) block)[i * 16]) + ff_h264_idct4x4_addblk_dc_lasx(dst[1] + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + } +} + +void ff_h264_idct_add8_422_lasx(uint8_t **dst, + const int32_t *blk_offset, + int16_t *block, int32_t dst_stride, + const uint8_t nzc[15 * 8]) +{ + int32_t i; + + for (i = 16; i < 20; i++) { + if (nzc[scan8[i]]) + ff_h264_idct_add_lasx(dst[0] + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + else if (((dctcoef *) block)[i * 16]) + ff_h264_idct4x4_addblk_dc_lasx(dst[0] + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + } + for (i = 32; i < 36; i++) { + if (nzc[scan8[i]]) + ff_h264_idct_add_lasx(dst[1] + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + else if (((dctcoef *) block)[i * 16]) + ff_h264_idct4x4_addblk_dc_lasx(dst[1] + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + } + for (i = 20; i < 24; i++) { + if (nzc[scan8[i + 4]]) + ff_h264_idct_add_lasx(dst[0] + blk_offset[i + 4], + block + i * 16 * sizeof(pixel), + dst_stride); + else if (((dctcoef *) block)[i * 16]) + ff_h264_idct4x4_addblk_dc_lasx(dst[0] + blk_offset[i + 4], + block + i * 16 * sizeof(pixel), + dst_stride); + } + for (i = 36; i < 40; i++) { + if (nzc[scan8[i + 4]]) + ff_h264_idct_add_lasx(dst[1] + blk_offset[i + 4], + block + i * 16 * sizeof(pixel), + dst_stride); + else if (((dctcoef *) block)[i * 16]) + ff_h264_idct4x4_addblk_dc_lasx(dst[1] + blk_offset[i + 4], + block + i * 16 * sizeof(pixel), + dst_stride); + } +} + +void ff_h264_idct_add16_intra_lasx(uint8_t *dst, + const int32_t *blk_offset, + int16_t *block, + int32_t dst_stride, + const uint8_t nzc[15 * 8]) +{ + int32_t i; + + for (i = 0; i < 16; i++) { + if (nzc[scan8[i]]) + ff_h264_idct_add_lasx(dst + blk_offset[i], + block + i * 16 * sizeof(pixel), dst_stride); + else if (((dctcoef *) block)[i * 16]) + ff_h264_idct4x4_addblk_dc_lasx(dst + blk_offset[i], + block + i * 16 * sizeof(pixel), + dst_stride); + } +} + +void ff_h264_deq_idct_luma_dc_lasx(int16_t *dst, int16_t *src, + int32_t de_qval) +{ +#define DC_DEST_STRIDE 16 + + __m256i src0, src1, src2, src3; + __m256i vec0, vec1, vec2, vec3; + __m256i tmp0, tmp1, tmp2, tmp3; + __m256i hres0, hres1, hres2, hres3; + __m256i vres0, vres1, vres2, vres3; + __m256i de_q_vec = __lasx_xvreplgr2vr_w(de_qval); + + DUP4_ARG2(__lasx_xvld, src, 0, src, 8, src, 16, src, 24, + src0, src1, src2, src3); + LASX_TRANSPOSE4x4_H(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3); + LASX_BUTTERFLY_4_H(tmp0, tmp2, tmp3, tmp1, vec0, vec3, vec2, vec1); + LASX_BUTTERFLY_4_H(vec0, vec1, vec2, vec3, hres0, hres3, hres2, hres1); + LASX_TRANSPOSE4x4_H(hres0, hres1, hres2, hres3, + hres0, hres1, hres2, hres3); + LASX_BUTTERFLY_4_H(hres0, hres1, hres3, hres2, vec0, vec3, vec2, vec1); + LASX_BUTTERFLY_4_H(vec0, vec1, vec2, vec3, vres0, vres1, vres2, vres3); + DUP4_ARG1(__lasx_vext2xv_w_h, vres0, vres1, vres2, vres3, + vres0, vres1, vres2, vres3); + DUP2_ARG3(__lasx_xvpermi_q, vres1, vres0, 0x20, vres3, vres2, 0x20, + vres0, vres1); + + vres0 = __lasx_xvmul_w(vres0, de_q_vec); + vres1 = __lasx_xvmul_w(vres1, de_q_vec); + + vres0 = __lasx_xvsrari_w(vres0, 8); + vres1 = __lasx_xvsrari_w(vres1, 8); + vec0 = __lasx_xvpickev_h(vres1, vres0); + vec0 = __lasx_xvpermi_d(vec0, 0xd8); + __lasx_xvstelm_h(vec0, dst + 0 * DC_DEST_STRIDE, 0, 0); + __lasx_xvstelm_h(vec0, dst + 2 * DC_DEST_STRIDE, 0, 1); + __lasx_xvstelm_h(vec0, dst + 8 * DC_DEST_STRIDE, 0, 2); + __lasx_xvstelm_h(vec0, dst + 10 * DC_DEST_STRIDE, 0, 3); + __lasx_xvstelm_h(vec0, dst + 1 * DC_DEST_STRIDE, 0, 4); + __lasx_xvstelm_h(vec0, dst + 3 * DC_DEST_STRIDE, 0, 5); + __lasx_xvstelm_h(vec0, dst + 9 * DC_DEST_STRIDE, 0, 6); + __lasx_xvstelm_h(vec0, dst + 11 * DC_DEST_STRIDE, 0, 7); + __lasx_xvstelm_h(vec0, dst + 4 * DC_DEST_STRIDE, 0, 8); + __lasx_xvstelm_h(vec0, dst + 6 * DC_DEST_STRIDE, 0, 9); + __lasx_xvstelm_h(vec0, dst + 12 * DC_DEST_STRIDE, 0, 10); + __lasx_xvstelm_h(vec0, dst + 14 * DC_DEST_STRIDE, 0, 11); + __lasx_xvstelm_h(vec0, dst + 5 * DC_DEST_STRIDE, 0, 12); + __lasx_xvstelm_h(vec0, dst + 7 * DC_DEST_STRIDE, 0, 13); + __lasx_xvstelm_h(vec0, dst + 13 * DC_DEST_STRIDE, 0, 14); + __lasx_xvstelm_h(vec0, dst + 15 * DC_DEST_STRIDE, 0, 15); + +#undef DC_DEST_STRIDE +}