From patchwork Wed Nov 22 13:59:10 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Shengbin Meng X-Patchwork-Id: 6274 Delivered-To: ffmpegpatchwork@gmail.com Received: by 10.2.161.94 with SMTP id m30csp202300jah; Wed, 22 Nov 2017 06:00:12 -0800 (PST) X-Google-Smtp-Source: AGs4zMZDphgpvr04ef+IV1F7450Sfe/VeWfFjCiomjJmUp+5HlKjdgBasGeUv4QGRD9tP4l1HEpS X-Received: by 10.223.181.131 with SMTP id c3mr17803693wre.198.1511359212534; Wed, 22 Nov 2017 06:00:12 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1511359212; cv=none; d=google.com; s=arc-20160816; b=U++dNy2Jiv2ObZPER7O4craLGseLoROs13yDsPeNNY0pi7JNVo3UZdeap2Yg/h45CR qTxEvufef4JvFCi2Juvpsvs1AxNU1gKR2AzGrQKFZvTIdeprUIe9Cl7BYTljo2P1GTaS lS4Jrvt1HN7VRohEkJqfXDla/EZNxWBP+cesYlF2WeB/U8/MVqwOyOtzv1fxhS6pDfKF xmu+DoAcU6vdZ0X4EE7fbAnL8FvsTgIpDuzeA102UaeAD6Hg5j44cTRvk2O6HL4Ixenw eEF6ZqovFS4TzOOMfgPDjCd8oKcfiDN+2of0Q0x21MIV4r5WYd8t2XgmZvLfBY9NT7tE s6VQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:mime-version:cc:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:references:in-reply-to:message-id:date :to:from:dkim-signature:delivered-to:arc-authentication-results; bh=gJRpX/yjrjeEOGBbl1N/KpQO971ba5OvYiqu/JE/h8o=; b=rCir8babxRy8dm3TynUyZWyQYWPzahWNjcvQeAKIvyDR72rBjh1imt8pRBrpTDTXe4 eHZeaeQtnzYup11iXMaNfsOhwgVbf48dfUTFDV28s3iVmnAGBFghyDH6Zxna/ebBtSHd qQGgGKlaWSHp9pgbWO2wk6Ju8aOmIetnkmd8gdq1HbDYyAkcXq5RJd7bttJxY4QWW3up AkMKZ2WhU1IiytzPvddo8hlv3JMtmuTyMTXqx6N27ZGJfbxlFUD/Y98DboY0dA3qtAKL 9Sp6sRba1GEKLbUNxftlvMgQa9EM/fnEhGgqEaLldeTbYvW/fNiYmq9Xru0s/+LORqM0 rcig== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@gmail.com header.s=20161025 header.b=q0jna9/j; 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 Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id 82si3443771wmg.22.2017.11.22.06.00.12; Wed, 22 Nov 2017 06:00:12 -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; dkim=neutral (body hash did not verify) header.i=@gmail.com header.s=20161025 header.b=q0jna9/j; 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 7B02268A192; Wed, 22 Nov 2017 15:59:34 +0200 (EET) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mail-pg0-f67.google.com (mail-pg0-f67.google.com [74.125.83.67]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 9BFF868A17C for ; Wed, 22 Nov 2017 15:59:25 +0200 (EET) Received: by mail-pg0-f67.google.com with SMTP id q7so14750pgr.8 for ; Wed, 22 Nov 2017 05:59:25 -0800 (PST) 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=3DtSlN7a66Ybiyd6weYz9hT0luXgV96fwNyTJhcLzEI=; b=q0jna9/j9EXPpdy7nEWLsMe1B8czdivg+kbWWFNoDdzNen3cubvr2Dp7MnQIdpau20 7Qm7WQpPgDP4O6QNnPcc7incfjpNoXQC7nbZOKBu91l3c0XuHLXyfqbT1KrhXtS3OCY/ i31GIO7GJCE3Yc1+uoL6yBMZGhL1bndtYAw1zeW/LkIP3UaX8G2s2+iSkcuIAkIreZU6 Oc4ZQ+JyfCoNAV1vh0hLjQF9YOq9jWZpHZfC2Thv1Irs/M7yvnhEQkmA4AS5dVs4ppEZ IEyP9DQVCoBCCkfheZkuB39f4iQNgiVLAAslyqJ0O6L6oQYX7yNuAcdzwG2fFTJUkUPh XL3w== 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=3DtSlN7a66Ybiyd6weYz9hT0luXgV96fwNyTJhcLzEI=; b=ZV5Ajy9o8QCmjevGNINFbUxR+4LUxLI94xw6DWbVR0lPwDzfoXB0hnqIuMZCy7dwkB BJQtueolV6Eppy6ZEoooTzzHEauymt/0fEEiC0gGhK4M9yZYmVSN1jleRBYti/U3FLul BvL4fhjYTxfoqrYd2NZlbG/nvQ4mFkD/7+QhdnRBvQbiwtPX7A93VzC/P8JB/JtOIB+0 DIH64IXespYryOi63bdgpuD+OlVDOOY4nXkXFFZqJafihR4t38MXQhTX1eCO8rJNAWZB M7Ume3AmhZ6BSYtUh4rbQQx9cwnsUdneUucnMoxXAxqV53b6ZJRzn9Soze67hKm/4MQd cJ1w== X-Gm-Message-State: AJaThX47oBaRnPLGfXBWWiz6VqQae/tEVMD/C6FeYbsBQmxLVXb2VX+3 yQbaiEKFuWOn0TJ4FaikaIHTbw== X-Received: by 10.101.92.66 with SMTP id v2mr20553438pgr.151.1511359164149; Wed, 22 Nov 2017 05:59:24 -0800 (PST) Received: from Shengbins-Mac-mini.corp.bytedance.com ([103.48.142.82]) by smtp.googlemail.com with ESMTPSA id 13sm13111692pfs.112.2017.11.22.05.59.22 (version=TLS1_2 cipher=ECDHE-RSA-AES128-SHA bits=128/128); Wed, 22 Nov 2017 05:59:23 -0800 (PST) From: Shengbin Meng To: ffmpeg-devel@ffmpeg.org Date: Wed, 22 Nov 2017 21:59:10 +0800 Message-Id: <20171122135910.20719-5-shengbinmeng@gmail.com> X-Mailer: git-send-email 2.13.6 (Apple Git-96) In-Reply-To: <20171122135910.20719-1-shengbinmeng@gmail.com> References: <20171122135910.20719-1-shengbinmeng@gmail.com> Subject: [FFmpeg-devel] [PATCH v2 5/5] avcodec/hevcdsp: Add NEON optimization for sao 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: Meng Wang MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" From: Meng Wang Signed-off-by: Meng Wang --- libavcodec/arm/Makefile | 3 +- libavcodec/arm/hevcdsp_init_neon.c | 62 +++++++++++++ libavcodec/arm/hevcdsp_sao_neon.S | 181 +++++++++++++++++++++++++++++++++++++ 3 files changed, 245 insertions(+), 1 deletion(-) create mode 100644 libavcodec/arm/hevcdsp_sao_neon.S diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile index 1acda0b1f8..fc4c0147c5 100644 --- a/libavcodec/arm/Makefile +++ b/libavcodec/arm/Makefile @@ -137,7 +137,8 @@ NEON-OBJS-$(CONFIG_HEVC_DECODER) += arm/hevcdsp_init_neon.o \ arm/hevcdsp_deblock_neon.o \ arm/hevcdsp_idct_neon.o \ arm/hevcdsp_qpel_neon.o \ - arm/hevcdsp_epel_neon.o + arm/hevcdsp_epel_neon.o \ + arm/hevcdsp_sao_neon.o NEON-OBJS-$(CONFIG_RV30_DECODER) += arm/rv34dsp_neon.o NEON-OBJS-$(CONFIG_RV40_DECODER) += arm/rv34dsp_neon.o \ arm/rv40dsp_neon.o diff --git a/libavcodec/arm/hevcdsp_init_neon.c b/libavcodec/arm/hevcdsp_init_neon.c index 4e57422ad4..f7efff28e1 100644 --- a/libavcodec/arm/hevcdsp_init_neon.c +++ b/libavcodec/arm/hevcdsp_init_neon.c @@ -23,6 +23,13 @@ #include "libavcodec/hevcdsp.h" #include "hevcdsp_arm.h" +void ff_hevc_sao_band_filter_neon_wrapper(uint8_t *_dst, uint8_t *_src, + ptrdiff_t stride_dst, ptrdiff_t stride_src, + int16_t *sao_offset_val, int sao_left_class, + int width, int height); +void ff_hevc_sao_edge_filter_neon_wrapper(uint8_t *_dst, uint8_t *_src, ptrdiff_t stride_dst, int16_t *sao_offset_val, + int eo, int width, int height); + void ff_hevc_v_loop_filter_luma_neon(uint8_t *_pix, ptrdiff_t _stride, int _beta, int *_tc, uint8_t *_no_p, uint8_t *_no_q); void ff_hevc_h_loop_filter_luma_neon(uint8_t *_pix, ptrdiff_t _stride, int _beta, int *_tc, uint8_t *_no_p, uint8_t *_no_q); void ff_hevc_v_loop_filter_chroma_neon(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q); @@ -414,6 +421,51 @@ EPEL_FUNC_WT(ff_hevc_put_epel_wt_h6v7_neon_8); EPEL_FUNC_WT(ff_hevc_put_epel_wt_h7v7_neon_8); #undef EPEL_FUNC_WT +void ff_hevc_sao_band_filter_neon_8(uint8_t *dst, uint8_t *src, ptrdiff_t stride_dst, ptrdiff_t stride_src, int width, int height, int16_t *offset_table); + +void ff_hevc_sao_band_filter_neon_wrapper(uint8_t *_dst, uint8_t *_src, + ptrdiff_t stride_dst, ptrdiff_t stride_src, + int16_t *sao_offset_val, int sao_left_class, + int width, int height) { + uint8_t *dst = (uint8_t *)_dst; + uint8_t *src = (uint8_t *)_src; + int16_t offset_table[32] = {0}; + int k; + + stride_dst /= sizeof(uint8_t); + stride_src /= sizeof(uint8_t); + + for (k = 0; k < 4; k++) { + offset_table[(k + sao_left_class) & 31] = sao_offset_val[k + 1]; + } + + ff_hevc_sao_band_filter_neon_8(dst, src, stride_dst, stride_src, width, height, offset_table); +} + +void ff_hevc_sao_edge_filter_neon_8(uint8_t *dst, uint8_t *src, ptrdiff_t stride_dst, ptrdiff_t stride_src, int width, int height, + int a_stride, int b_stride, int16_t *sao_offset_val, uint8_t *edge_idx); + +void ff_hevc_sao_edge_filter_neon_wrapper(uint8_t *_dst, uint8_t *_src, ptrdiff_t stride_dst, int16_t *sao_offset_val, + int eo, int width, int height) { + static uint8_t edge_idx[] = { 1, 2, 0, 3, 4 }; + static const int8_t pos[4][2][2] = { + { { -1, 0 }, { 1, 0 } }, // horizontal + { { 0, -1 }, { 0, 1 } }, // vertical + { { -1, -1 }, { 1, 1 } }, // 45 degree + { { 1, -1 }, { -1, 1 } }, // 135 degree + }; + uint8_t *dst = (uint8_t *)_dst; + uint8_t *src = (uint8_t *)_src; + int a_stride, b_stride; + ptrdiff_t stride_src = (2*64 + 32) / sizeof(uint8_t); + stride_dst /= sizeof(uint8_t); + + a_stride = pos[eo][0][0] + pos[eo][0][1] * stride_src; + b_stride = pos[eo][1][0] + pos[eo][1][1] * stride_src; + + ff_hevc_sao_edge_filter_neon_8(dst, src, stride_dst, stride_src, width, height, a_stride, b_stride, sao_offset_val, edge_idx); +} + void ff_hevc_put_qpel_neon_wrapper(int16_t *dst, uint8_t *src, ptrdiff_t srcstride, int height, intptr_t mx, intptr_t my, int width) { @@ -505,6 +557,16 @@ av_cold void ff_hevc_dsp_init_neon(HEVCDSPContext *c, const int bit_depth) c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_neon; c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_neon; c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_neon; + c->sao_band_filter[0] = ff_hevc_sao_band_filter_neon_wrapper; + c->sao_band_filter[1] = ff_hevc_sao_band_filter_neon_wrapper; + c->sao_band_filter[2] = ff_hevc_sao_band_filter_neon_wrapper; + c->sao_band_filter[3] = ff_hevc_sao_band_filter_neon_wrapper; + c->sao_band_filter[4] = ff_hevc_sao_band_filter_neon_wrapper; + c->sao_edge_filter[0] = ff_hevc_sao_edge_filter_neon_wrapper; + c->sao_edge_filter[1] = ff_hevc_sao_edge_filter_neon_wrapper; + c->sao_edge_filter[2] = ff_hevc_sao_edge_filter_neon_wrapper; + c->sao_edge_filter[3] = ff_hevc_sao_edge_filter_neon_wrapper; + c->sao_edge_filter[4] = ff_hevc_sao_edge_filter_neon_wrapper; c->add_residual[0] = ff_hevc_add_residual_4x4_8_neon; c->add_residual[1] = ff_hevc_add_residual_8x8_8_neon; c->add_residual[2] = ff_hevc_add_residual_16x16_8_neon; diff --git a/libavcodec/arm/hevcdsp_sao_neon.S b/libavcodec/arm/hevcdsp_sao_neon.S new file mode 100644 index 0000000000..347167951b --- /dev/null +++ b/libavcodec/arm/hevcdsp_sao_neon.S @@ -0,0 +1,181 @@ +/* + * Copyright (c) 2017 Meng Wang + * + * 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/arm/asm.S" +#include "neon.S" + +function ff_hevc_sao_band_filter_neon_8, export=1 + push {r4-r10} + ldr r5, [sp, #28] // width + ldr r4, [sp, #32] // height + ldr r8, [sp, #36] // offset_table + vpush {d8-d15} + mov r12, r4 // r12 = height + mov r6, r0 // r6 = r0 = dst + mov r7, r1 // r7 = r1 = src + vldm r8, {q0-q3} + vmov.u16 q15, #1 + vmov.u8 q14, #32 +0: pld [r1] + vld1.8 {d16}, [r1], r3 + cmp r5, #4 + beq 4f +8: subs r4, #1 + vshr.u8 d17, d16, #3 // index = [src>>3] + vshll.u8 q9, d17, #1 // lowIndex = 2*index + vadd.u16 q11, q9, q15 // highIndex = (2*index+1) << 8 + vshl.u16 q10, q11, #8 // q10: highIndex; q9: lowIndex; + vadd.u16 q10, q9 // combine high and low index; + // Look-up Table Round 1; index range: 0-15 + vtbx.8 d24, {q0-q1}, d20 + vtbx.8 d25, {q0-q1}, d21 + // Look-up Table Round 2; index range: 16-31 + vsub.u8 q10, q14 // Look-up with 8bit + vtbx.8 d24, {q2-q3}, d20 + vtbx.8 d25, {q2-q3}, d21 + vaddw.u8 q13, q12, d16 + vqmovun.s16 d8, q13 + vst1.8 d8, [r0], r2 + vld1.8 {d16}, [r1], r3 + bne 8b + subs r5, #8 + beq 99f + mov r4, r12 + add r6, #8 + mov r0, r6 + add r7, #8 + mov r1, r7 + b 0b +4: subs r4, #1 + vshr.u8 d17, d16, #3 // src>>3 + vshll.u8 q9, d17, #1 // lowIndex = 2*index + vadd.u16 q11, q9, q15 // highIndex = (2*index+1) << 8 + vshl.u16 q10, q11, #8 // q10: highIndex; q9: lowIndex; + vadd.u16 q10, q9 // combine high and low index; + // Look-up Table Round 1; index range: 0-15 + vtbx.8 d24, {q0-q1}, d20 + vtbx.8 d25, {q0-q1}, d21 + // Look-up Table Round 2; index range: 16-32 + vsub.u8 q10, q14 // Look-up with 8bit + vtbx.8 d24, {q2-q3}, d20 + vtbx.8 d25, {q2-q3}, d21 + vaddw.u8 q13, q12, d16 + vqmovun.s16 d14, q13 + vst1.32 d14[0], [r0], r2 + vld1.32 {d16[0]}, [r1], r3 + bne 4b + b 99f +99: + vpop {d8-d15} + pop {r4-r10} + bx lr +endfunc + +function ff_hevc_sao_edge_filter_neon_8, export=1 + push {r4-r11} + ldr r5, [sp, #32] // width + ldr r4, [sp, #36] // height + ldr r8, [sp, #40] // a_stride + ldr r9, [sp, #44] // b_stride + ldr r10, [sp, #48] // sao_offset_val + ldr r11, [sp, #52] // edge_idx + vpush {d8-d15} + mov r12, r4 // r12 = height + mov r6, r0 // r6 = r0 = dst + mov r7, r1 // r7 = r1 = src + vld1.8 {d0}, [r11] // edge_idx tabel load in d0 5x8bit + vld1.16 {q1}, [r10] // sao_offset_val table load in q1, 5x16bit + vmov.u8 d1, #2 + vmov.u16 q2, #1 +0: mov r10, r1 + add r10, r8 // src[x + a_stride] + mov r11, r1 + add r11, r9 // src[x + b_stride] + pld [r1] + vld1.8 {d16}, [r1], r3 // src[x] 8x8bit + vld1.8 {d17}, [r10], r3 // src[x + a_stride] + vld1.8 {d18}, [r11], r3 // src[x + b_stride] + cmp r5, #4 + beq 4f +8: subs r4, #1 + vcgt.u8 d8, d16, d17 + vshr.u8 d9, d8, #7 + vclt.u8 d8, d16, d17 + vadd.u8 d8, d9 // diff0 + vcgt.u8 d10, d16, d18 + vshr.u8 d11, d10, #7 + vclt.u8 d10, d16, d18 + vadd.u8 d10, d11 // diff1 + vadd.s8 d8, d10 + vadd.s8 d8, d1 + vtbx.8 d9, {d0}, d8 // offset_val + vshll.u8 q6, d9, #1 // lowIndex + vadd.u16 q7, q6, q2 + vshl.u16 q10, q7, #8 // highIndex + vadd.u16 q10, q6 // combine lowIndex and highIndex, offset_val + vtbx.8 d22, {q1}, d20 + vtbx.8 d23, {q1}, d21 + vaddw.u8 q12, q11, d16 + vqmovun.s16 d26, q12 + vst1.8 d26, [r0], r2 + vld1.8 {d16}, [r1], r3 // src[x] 8x8bit + vld1.8 {d17}, [r10], r3 // src[x + a_stride] + vld1.8 {d18}, [r11], r3 // src[x + b_stride] + bne 8b + subs r5, #8 + beq 99f + mov r4, r12 + add r6, #8 + mov r0, r6 + add r7, #8 + mov r1, r7 + b 0b +4: subs r4, #1 + vcgt.u8 d8, d16, d17 + vshr.u8 d9, d8, #7 + vclt.u8 d8, d16, d17 + vadd.u8 d8, d9 // diff0 + vcgt.u8 d10, d16, d18 + vshr.u8 d11, d10, #7 + vclt.u8 d10, d16, d18 + vadd.u8 d10, d11 // diff1 + vadd.s8 d8, d10 + vadd.s8 d8, d1 + vtbx.8 d9, {d0}, d8 // offset_val + vshll.u8 q6, d9, #1 // lowIndex + vadd.u16 q7, q6, q2 + vshl.u16 q10, q7, #8 // highIndex + vadd.u16 q10, q6 // combine lowIndex and highIndex, offset_val + vtbx.8 d22, {q1}, d20 + vtbx.8 d23, {q1}, d21 + vaddw.u8 q12, q11, d16 + vqmovun.s16 d26, q12 + vst1.32 d26[0], [r0], r2 + vld1.32 {d16[0]}, [r1], r3 + vld1.32 {d17[0]}, [r10], r3 // src[x + a_stride] + vld1.32 {d18[0]}, [r11], r3 // src[x + b_stride] + bne 4b + b 99f +99: + vpop {d8-d15} + pop {r4-r11} + bx lr +endfunc