From patchwork Sat Sep 5 02:25:16 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Shiyou Yin X-Patchwork-Id: 22110 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id E862544AD7E for ; Sat, 5 Sep 2020 05:25:41 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id BC1E368A9AA; Sat, 5 Sep 2020 05:25:41 +0300 (EEST) 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 1E5E268029C for ; Sat, 5 Sep 2020 05:25:33 +0300 (EEST) Received: from localhost (unknown [36.33.26.144]) by mail.loongson.cn (Coremail) with SMTP id AQAAf9Dxj8cW91JfRggRAA--.452S3; Sat, 05 Sep 2020 10:25:27 +0800 (CST) From: Shiyou Yin To: ffmpeg-devel@ffmpeg.org Date: Sat, 5 Sep 2020 10:25:16 +0800 Message-Id: <1599272718-13893-2-git-send-email-yinshiyou-hf@loongson.cn> X-Mailer: git-send-email 2.1.0 In-Reply-To: <1599272718-13893-1-git-send-email-yinshiyou-hf@loongson.cn> References: <1599272718-13893-1-git-send-email-yinshiyou-hf@loongson.cn> X-CM-TRANSID: AQAAf9Dxj8cW91JfRggRAA--.452S3 X-Coremail-Antispam: 1UD129KBjvJXoW7Aw13Aw15Aw13Cr4UWw47twb_yoW8Ww17pF 1UAF17Gr4xGrZakr4xGr4Y9Fy5Aw1FqryYyF9rKan7A3y5Xr4xKr9xJrykCrWUJFZFkF95 Xryqgry5G3WkCFUanT9S1TB71UUUUU7qnTZGkaVYY2UrUUUUjbIjqfuFe4nvWSU5nxnvy2 9KBjDU0xBIdaVrnRJUUUk0b7Iv0xC_Cr1lb4IE77IF4wAFF20E14v26r1j6r4UM7CY07I2 0VC2zVCF04k26cxKx2IYs7xG6rWj6s0DM7CIcVAFz4kK6r1j6r18M28lY4IEw2IIxxk0rw A2F7IY1VAKz4vEj48ve4kI8wA2z4x0Y4vE2Ix0cI8IcVAFwI0_JFI_Gr1l84ACjcxK6xII jxv20xvEc7CjxVAFwI0_Gr0_Cr1l84ACjcxK6I8E87Iv67AKxVW8JVWxJwA2z4x0Y4vEx4 A2jsIEc7CjxVAFwI0_Gr0_Gr1UM2AIxVAIcxkEcVAq07x20xvEncxIr21l5I8CrVACY4xI 64kE6c02F40Ex7xfMcIj6xIIjxv20xvE14v26r126r1DMcIj6I8E87Iv67AKxVWUJVW8Jw Am72CE4IkC6x0Yz7v_Jr0_Gr1lF7xvr2IYc2Ij64vIr41lc2xSY4AK67AK6r4UMxAIw28I cxkI7VAKI48JMxC20s026xCaFVCjc4AY6r1j6r4UMI8I3I0E5I8CrVAFwI0_Jr0_Jr4lx2 IqxVCjr7xvwVAFwI0_JrI_JrWlx4CE17CEb7AF67AKxVWUJVWUXwCIc40Y0x0EwIxGrwCI 42IY6xIIjxv20xvE14v26r1I6r4UMIIF0xvE2Ix0cI8IcVCY1x0267AKxVWUJVW8JwCI42 IY6xAIw20EY4v20xvaj40_Jr0_JF4lIxAIcVC2z280aVAFwI0_Jr0_Gr1lIxAIcVC2z280 aVCY1x0267AKxVWUJVW8JbIYCTnIWIevJa73UjIFyTuYvjxU2Z2-UUUUU X-CM-SenderInfo: p1lq2x5l1r3gtki6z05rqj20fqof0/ Subject: [FFmpeg-devel] [PATCH 1/3] avcodec/mips: Restore the initialization sequence of MSA and MMI in ff_h264chroma_init_mips. 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 MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Speed of decoding H264: 5.14x ==> 5.23x (tested on 3A4000). --- libavcodec/mips/h264chroma_init_mips.c | 19 +++++++++---------- 1 file changed, 9 insertions(+), 10 deletions(-) diff --git a/libavcodec/mips/h264chroma_init_mips.c b/libavcodec/mips/h264chroma_init_mips.c index 6bb19d3..755cc04 100644 --- a/libavcodec/mips/h264chroma_init_mips.c +++ b/libavcodec/mips/h264chroma_init_mips.c @@ -28,7 +28,15 @@ 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; - /* MMI apears to be faster than MSA here */ + 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; + } + } + if (have_msa(cpu_flags)) { if (!high_bit_depth) { c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_msa; @@ -40,13 +48,4 @@ av_cold void ff_h264chroma_init_mips(H264ChromaContext *c, int bit_depth) c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_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; - } - } } From patchwork Sat Sep 5 02:25:17 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Shiyou Yin X-Patchwork-Id: 22111 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id 2D68B44AD7E for ; Sat, 5 Sep 2020 05:25:43 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 1554F68AEBC; Sat, 5 Sep 2020 05:25:43 +0300 (EEST) 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 8CC116808E9 for ; Sat, 5 Sep 2020 05:25:33 +0300 (EEST) Received: from localhost (unknown [36.33.26.144]) by mail.loongson.cn (Coremail) with SMTP id AQAAf9DxyMUX91JfRwgRAA--.493S3; Sat, 05 Sep 2020 10:25:28 +0800 (CST) From: Shiyou Yin To: ffmpeg-devel@ffmpeg.org Date: Sat, 5 Sep 2020 10:25:17 +0800 Message-Id: <1599272718-13893-3-git-send-email-yinshiyou-hf@loongson.cn> X-Mailer: git-send-email 2.1.0 In-Reply-To: <1599272718-13893-1-git-send-email-yinshiyou-hf@loongson.cn> References: <1599272718-13893-1-git-send-email-yinshiyou-hf@loongson.cn> X-CM-TRANSID: AQAAf9DxyMUX91JfRwgRAA--.493S3 X-Coremail-Antispam: 1UD129KBjvJXoWxtr15Gw4xZw13XFWrKr17ZFb_yoWxKF1fpr y8Grs0yrWavFW7CFZxJF4xGrnxZF48tw1UWFyUtF18Ars0yr1rurZ7GryxWw1rGFykuFWa vF1UZFy3CF17Zw7anT9S1TB71UUUUU7qnTZGkaVYY2UrUUUUjbIjqfuFe4nvWSU5nxnvy2 9KBjDU0xBIdaVrnRJUUUk0b7Iv0xC_tr1lb4IE77IF4wAFF20E14v26r1j6r4UM7CY07I2 0VC2zVCF04k26cxKx2IYs7xG6rWj6s0DM7CIcVAFz4kK6r1j6r18M28lY4IEw2IIxxk0rw A2F7IY1VAKz4vEj48ve4kI8wA2z4x0Y4vE2Ix0cI8IcVAFwI0_JFI_Gr1l84ACjcxK6xII jxv20xvEc7CjxVAFwI0_Gr0_Cr1l84ACjcxK6I8E87Iv67AKxVW8JVWxJwA2z4x0Y4vEx4 A2jsIEc7CjxVAFwI0_Gr0_Gr1UM2AIxVAIcxkEcVAq07x20xvEncxIr21l5I8CrVACY4xI 64kE6c02F40Ex7xfMcIj6xIIjxv20xvE14v26r126r1DMcIj6I8E87Iv67AKxVWUJVW8Jw Am72CE4IkC6x0Yz7v_Jr0_Gr1lF7xvr2IYc2Ij64vIr41lc2xSY4AK67AK6r4UMxAIw28I cxkI7VAKI48JMxC20s026xCaFVCjc4AY6r1j6r4UMI8I3I0E5I8CrVAFwI0_Jr0_Jr4lx2 IqxVCjr7xvwVAFwI0_JrI_JrWlx4CE17CEb7AF67AKxVWUJVWUXwCIc40Y0x0EwIxGrwCI 42IY6xIIjxv20xvE14v26r1I6r4UMIIF0xvE2Ix0cI8IcVCY1x0267AKxVWUJVW8JwCI42 IY6xAIw20EY4v20xvaj40_Jr0_JF4lIxAIcVC2z280aVAFwI0_Jr0_Gr1lIxAIcVC2z280 aVCY1x0267AKxVWUJVW8JbIYCTnIWIevJa73UjIFyTuYvjxUc2-eUUUUU X-CM-SenderInfo: p1lq2x5l1r3gtki6z05rqj20fqof0/ Subject: [FFmpeg-devel] [PATCH 2/3] avcodec/mips: Refine get_cabac_inline_mips. 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 MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" 1. Refined function get_cabac_inline_mips. 2. Optimize function get_cabac_bypass and get_cabac_bypass_sign. Speed of decoding h264: 5.23x ==> 5.45x(tested on 3A4000). --- libavcodec/mips/cabac.h | 131 +++++++++++++++++++++++++++++++++++++----------- 1 file changed, 102 insertions(+), 29 deletions(-) diff --git a/libavcodec/mips/cabac.h b/libavcodec/mips/cabac.h index 3d09e93..579d4ee 100644 --- a/libavcodec/mips/cabac.h +++ b/libavcodec/mips/cabac.h @@ -2,7 +2,8 @@ * Loongson SIMD optimized h264chroma * * Copyright (c) 2018 Loongson Technology Corporation Limited - * Copyright (c) 2018 Shiyou Yin + * Contributed by Shiyou Yin + * Gu Xiwei(guxiwei-hf@loongson.cn) * * This file is part of FFmpeg. * @@ -25,18 +26,18 @@ #define AVCODEC_MIPS_CABAC_H #include "libavcodec/cabac.h" -#include "libavutil/mips/asmdefs.h" +#include "libavutil/mips/mmiutils.h" #include "config.h" #define get_cabac_inline get_cabac_inline_mips static av_always_inline int get_cabac_inline_mips(CABACContext *c, - uint8_t * const state){ + uint8_t * const state){ mips_reg tmp0, tmp1, tmp2, bit; __asm__ volatile ( "lbu %[bit], 0(%[state]) \n\t" "and %[tmp0], %[c_range], 0xC0 \n\t" - PTR_ADDU "%[tmp0], %[tmp0], %[tmp0] \n\t" + PTR_SLL "%[tmp0], %[tmp0], 0x01 \n\t" PTR_ADDU "%[tmp0], %[tmp0], %[tables] \n\t" PTR_ADDU "%[tmp0], %[tmp0], %[bit] \n\t" /* tmp1: RangeLPS */ @@ -44,18 +45,11 @@ static av_always_inline int get_cabac_inline_mips(CABACContext *c, PTR_SUBU "%[c_range], %[c_range], %[tmp1] \n\t" PTR_SLL "%[tmp0], %[c_range], 0x11 \n\t" - PTR_SUBU "%[tmp0], %[tmp0], %[c_low] \n\t" - - /* tmp2: lps_mask */ - PTR_SRA "%[tmp2], %[tmp0], 0x1F \n\t" - /* If tmp0 < 0, lps_mask == 0xffffffff*/ - /* If tmp0 >= 0, lps_mask == 0x00000000*/ + "slt %[tmp2], %[tmp0], %[c_low] \n\t" "beqz %[tmp2], 1f \n\t" - PTR_SLL "%[tmp0], %[c_range], 0x11 \n\t" + "move %[c_range], %[tmp1] \n\t" + "not %[bit], %[bit] \n\t" PTR_SUBU "%[c_low], %[c_low], %[tmp0] \n\t" - PTR_SUBU "%[tmp0], %[tmp1], %[c_range] \n\t" - PTR_ADDU "%[c_range], %[c_range], %[tmp0] \n\t" - "xor %[bit], %[bit], %[tmp2] \n\t" "1: \n\t" /* tmp1: *state */ @@ -70,23 +64,18 @@ static av_always_inline int get_cabac_inline_mips(CABACContext *c, PTR_SLL "%[c_range], %[c_range], %[tmp2] \n\t" PTR_SLL "%[c_low], %[c_low], %[tmp2] \n\t" - "and %[tmp0], %[c_low], %[cabac_mask] \n\t" - "bnez %[tmp0], 1f \n\t" - PTR_ADDIU "%[tmp0], %[c_low], -0x01 \n\t" + "and %[tmp1], %[c_low], %[cabac_mask] \n\t" + "bnez %[tmp1], 1f \n\t" + PTR_ADDIU "%[tmp0], %[c_low], -0X01 \n\t" "xor %[tmp0], %[c_low], %[tmp0] \n\t" PTR_SRA "%[tmp0], %[tmp0], 0x0f \n\t" PTR_ADDU "%[tmp0], %[tmp0], %[tables] \n\t" + /* tmp2: ff_h264_norm_shift[x >> (CABAC_BITS - 1)] */ "lbu %[tmp2], %[norm_off](%[tmp0]) \n\t" -#if CABAC_BITS == 16 - "lbu %[tmp0], 0(%[c_bytestream]) \n\t" - "lbu %[tmp1], 1(%[c_bytestream]) \n\t" - PTR_SLL "%[tmp0], %[tmp0], 0x09 \n\t" - PTR_SLL "%[tmp1], %[tmp1], 0x01 \n\t" - PTR_ADDU "%[tmp0], %[tmp0], %[tmp1] \n\t" -#else - "lbu %[tmp0], 0(%[c_bytestream]) \n\t" + + "lhu %[tmp0], 0(%[c_bytestream]) \n\t" + "dsbh %[tmp0], %[tmp0] \n\t" PTR_SLL "%[tmp0], %[tmp0], 0x01 \n\t" -#endif PTR_SUBU "%[tmp0], %[tmp0], %[cabac_mask] \n\t" "li %[tmp1], 0x07 \n\t" @@ -94,10 +83,13 @@ static av_always_inline int get_cabac_inline_mips(CABACContext *c, PTR_SLL "%[tmp0], %[tmp0], %[tmp1] \n\t" PTR_ADDU "%[c_low], %[c_low], %[tmp0] \n\t" -#if !UNCHECKED_BITSTREAM_READER - "bge %[c_bytestream], %[c_bytestream_end], 1f \n\t" +#if UNCHECKED_BITSTREAM_READER + PTR_ADDIU "%[c_bytestream], %[c_bytestream], 0x02 \n\t" +#else + "slt %[tmp0], %[c_bytestream], %[c_bytestream_end] \n\t" + PTR_ADDIU "%[tmp2], %[c_bytestream], 0x02 \n\t" + "movn %[c_bytestream], %[tmp2], %[tmp0] \n\t" #endif - PTR_ADDIU "%[c_bytestream], %[c_bytestream], 0X02 \n\t" "1: \n\t" : [bit]"=&r"(bit), [tmp0]"=&r"(tmp0), [tmp1]"=&r"(tmp1), [tmp2]"=&r"(tmp2), [c_range]"+&r"(c->range), [c_low]"+&r"(c->low), @@ -116,4 +108,85 @@ static av_always_inline int get_cabac_inline_mips(CABACContext *c, return bit; } +#define get_cabac_bypass get_cabac_bypass_mips +static av_always_inline int get_cabac_bypass_mips(CABACContext *c) +{ + mips_reg tmp0, tmp1; + int res = 0; + __asm__ volatile( + PTR_SLL "%[c_low], %[c_low], 0x01 \n\t" + "and %[tmp0], %[c_low], %[cabac_mask] \n\t" + "bnez %[tmp0], 1f \n\t" + "lhu %[tmp1], 0(%[c_bytestream]) \n\t" + "dsbh %[tmp1], %[tmp1] \n\t" + PTR_SLL "%[tmp1], %[tmp1], 0x01 \n\t" + PTR_SUBU "%[tmp1], %[tmp1], %[cabac_mask] \n\t" + PTR_ADDU "%[c_low], %[c_low], %[tmp1] \n\t" +#if UNCHECKED_BITSTREAM_READER + PTR_ADDIU "%[c_bytestream], %[c_bytestream], 0x02 \n\t" +#else + "slt %[tmp0], %[c_bytestream], %[c_bytestream_end] \n\t" + PTR_ADDIU "%[tmp1], %[c_bytestream], 0x02 \n\t" + "movn %[c_bytestream], %[tmp1], %[tmp0] \n\t" +#endif + "1: \n\t" + PTR_SLL "%[tmp1], %[c_range], 0x11 \n\t" + "slt %[tmp0], %[c_low], %[tmp1] \n\t" + PTR_SUBU "%[tmp1], %[c_low], %[tmp1] \n\t" + "movz %[res], %[one], %[tmp0] \n\t" + "movz %[c_low], %[tmp1], %[tmp0] \n\t" + : [tmp0]"=&r"(tmp0), [tmp1]"=&r"(tmp1), [res]"+&r"(res), + [c_range]"+&r"(c->range), [c_low]"+&r"(c->low), + [c_bytestream]"+&r"(c->bytestream) + : [cabac_mask]"r"(CABAC_MASK), +#if !UNCHECKED_BITSTREAM_READER + [c_bytestream_end]"r"(c->bytestream_end), +#endif + [one]"r"(0x01) + : "memory" + ); + return res; +} + +#define get_cabac_bypass_sign get_cabac_bypass_sign_mips +static av_always_inline int get_cabac_bypass_sign_mips(CABACContext *c, int val) +{ + mips_reg tmp0, tmp1; + int res = val; + __asm__ volatile( + PTR_SLL "%[c_low], %[c_low], 0x01 \n\t" + "and %[tmp0], %[c_low], %[cabac_mask] \n\t" + "bnez %[tmp0], 1f \n\t" + "lhu %[tmp1], 0(%[c_bytestream]) \n\t" + "dsbh %[tmp1], %[tmp1] \n\t" + PTR_SLL "%[tmp1], %[tmp1], 0x01 \n\t" + PTR_SUBU "%[tmp1], %[tmp1], %[cabac_mask] \n\t" + PTR_ADDU "%[c_low], %[c_low], %[tmp1] \n\t" +#if UNCHECKED_BITSTREAM_READER + PTR_ADDIU "%[c_bytestream], %[c_bytestream], 0x02 \n\t" +#else + "slt %[tmp0], %[c_bytestream], %[c_bytestream_end] \n\t" + PTR_ADDIU "%[tmp1], %[c_bytestream], 0x02 \n\t" + "movn %[c_bytestream], %[tmp1], %[tmp0] \n\t" +#endif + "1: \n\t" + PTR_SLL "%[tmp1], %[c_range], 0x11 \n\t" + "slt %[tmp0], %[c_low], %[tmp1] \n\t" + PTR_SUBU "%[tmp1], %[c_low], %[tmp1] \n\t" + "movz %[c_low], %[tmp1], %[tmp0] \n\t" + PTR_SUBU "%[tmp1], %[zero], %[res] \n\t" + "movn %[res], %[tmp1], %[tmp0] \n\t" + : [tmp0]"=&r"(tmp0), [tmp1]"=&r"(tmp1), [res]"+&r"(res), + [c_range]"+&r"(c->range), [c_low]"+&r"(c->low), + [c_bytestream]"+&r"(c->bytestream) + : [cabac_mask]"r"(CABAC_MASK), +#if !UNCHECKED_BITSTREAM_READER + [c_bytestream_end]"r"(c->bytestream_end), +#endif + [zero]"r"(0x0) + : "memory" + ); + + return res; +} #endif /* AVCODEC_MIPS_CABAC_H */ From patchwork Sat Sep 5 02:25:18 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Shiyou Yin X-Patchwork-Id: 22112 Return-Path: X-Original-To: patchwork@ffaux-bg.ffmpeg.org Delivered-To: patchwork@ffaux-bg.ffmpeg.org Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by ffaux.localdomain (Postfix) with ESMTP id 469A544AD7E for ; Sat, 5 Sep 2020 05:25:45 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 3067468B2E0; Sat, 5 Sep 2020 05:25:45 +0300 (EEST) 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 1782B687F43 for ; Sat, 5 Sep 2020 05:25:34 +0300 (EEST) Received: from localhost (unknown [36.33.26.144]) by mail.loongson.cn (Coremail) with SMTP id AQAAf9Dx38cb91JfSQgRAA--.434S3; Sat, 05 Sep 2020 10:25:32 +0800 (CST) From: Shiyou Yin To: ffmpeg-devel@ffmpeg.org Date: Sat, 5 Sep 2020 10:25:18 +0800 Message-Id: <1599272718-13893-4-git-send-email-yinshiyou-hf@loongson.cn> X-Mailer: git-send-email 2.1.0 In-Reply-To: <1599272718-13893-1-git-send-email-yinshiyou-hf@loongson.cn> References: <1599272718-13893-1-git-send-email-yinshiyou-hf@loongson.cn> X-CM-TRANSID: AQAAf9Dx38cb91JfSQgRAA--.434S3 X-Coremail-Antispam: 1UD129KBjvJXoW3Jr1UKr1xAF1DAFWUJFy7ZFb_yoW3KFy3p3 WUu3W3Gr18J3yakas3Jr4kC3WayF95tFy8AFyUXw1jq39rX348ZFZ3KFZ3X3WkWr9FqF13 Ww1Fg347A3W7C3DanT9S1TB71UUUUU7qnTZGkaVYY2UrUUUUjbIjqfuFe4nvWSU5nxnvy2 9KBjDU0xBIdaVrnRJUUUka14x267AKxVWUJVW8JwAFc2x0x2IEx4CE42xK8VAvwI8IcIk0 rVWrJVCq3wAFIxvE14AKwVWUJVWUGwA2ocxC64kIII0Yj41l84x0c7CEw4AK67xGY2AK02 1l84ACjcxK6xIIjxv20xvE14v26r4j6ryUM28EF7xvwVC0I7IYx2IY6xkF7I0E14v26r4j 6F4UM28EF7xvwVC2z280aVAFwI0_Gr0_Cr1l84ACjcxK6I8E87Iv6xkF7I0E14v26r4j6r 4UJwAS0I0E0xvYzxvE52x082IY62kv0487Mc02F40EFcxC0VAKzVAqx4xG6I80ewAv7VC0 I7IYx2IY67AKxVWUAVWUtwAv7VC2z280aVAFwI0_Jr0_Gr1lOx8S6xCaFVCjc4AY6r1j6r 4UM4x0Y48IcxkI7VAKI48JM4x0x7Aq67IIx4CEVc8vx2IErcIFxwCY02Avz4vE14v_Gr1l 42xK82IYc2Ij64vIr41l4I8I3I0E4IkC6x0Yz7v_Jr0_Gr1lx2IqxVAqx4xG67AKxVWUJV WUGwC20s026x8GjcxK67AKxVWUGVWUWwC2zVAF1VAY17CE14v26r1Y6r17MIIYrxkI7VAK I48JMIIF0xvE2Ix0cI8IcVAFwI0_JFI_Gr1lIxAIcVC0I7IYx2IY6xkF7I0E14v26r1j6r 4UMIIF0xvE42xK8VAvwI8IcIk0rVWUJVWUCwCI42IY6I8E87Iv67AKxVWUJVW8JwCI42IY 6I8E87Iv6xkF7I0E14v26r1j6r4UYxBIdaVFxhVjvjDU0xZFpf9x0JUkb18UUUUU= X-CM-SenderInfo: p1lq2x5l1r3gtki6z05rqj20fqof0/ Subject: [FFmpeg-devel] [PATCH 3/3] avcodec/mips: Optimize function ff_h264_loop_filter_strength_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: gxw MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" From: gxw Speed of decoding H264: 5.45x ==> 5.53x Signed-off-by: Shiyou Yin --- libavcodec/mips/Makefile | 3 +- libavcodec/mips/h264_deblock_msa.c | 153 ++++++++++++++++++++++++++++++++++++ libavcodec/mips/h264dsp_init_mips.c | 2 + libavcodec/mips/h264dsp_mips.h | 4 + 4 files changed, 161 insertions(+), 1 deletion(-) create mode 100644 libavcodec/mips/h264_deblock_msa.c diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile index 2be4d9b..81a73a4 100644 --- a/libavcodec/mips/Makefile +++ b/libavcodec/mips/Makefile @@ -57,7 +57,8 @@ MSA-OBJS-$(CONFIG_VP8_DECODER) += mips/vp8_mc_msa.o \ mips/vp8_lpf_msa.o MSA-OBJS-$(CONFIG_VP3DSP) += mips/vp3dsp_idct_msa.o MSA-OBJS-$(CONFIG_H264DSP) += mips/h264dsp_msa.o \ - mips/h264idct_msa.o + mips/h264idct_msa.o \ + mips/h264_deblock_msa.o MSA-OBJS-$(CONFIG_H264QPEL) += mips/h264qpel_msa.o MSA-OBJS-$(CONFIG_H264CHROMA) += mips/h264chroma_msa.o MSA-OBJS-$(CONFIG_H264PRED) += mips/h264pred_msa.o diff --git a/libavcodec/mips/h264_deblock_msa.c b/libavcodec/mips/h264_deblock_msa.c new file mode 100644 index 0000000..4fed55c --- /dev/null +++ b/libavcodec/mips/h264_deblock_msa.c @@ -0,0 +1,153 @@ +/* + * MIPS SIMD optimized H.264 deblocking code + * + * Copyright (c) 2020 Loongson Technology Corporation Limited + * Gu Xiwei + * + * 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 "libavcodec/bit_depth_template.c" +#include "h264dsp_mips.h" +#include "libavutil/mips/generic_macros_msa.h" +#include "libavcodec/mips/h264dsp_mips.h" + +#define h264_loop_filter_strength_iteration_msa(edges, step, mask_mv, dir, \ + d_idx, mask_dir) \ +do { \ + int b_idx = 0; \ + int step_x4 = step << 2; \ + int d_idx_12 = d_idx + 12; \ + int d_idx_52 = d_idx + 52; \ + int d_idx_x4 = d_idx << 2; \ + int d_idx_x4_48 = d_idx_x4 + 48; \ + int dir_x32 = dir * 32; \ + uint8_t *ref_t = (uint8_t*)ref; \ + uint8_t *mv_t = (uint8_t*)mv; \ + uint8_t *nnz_t = (uint8_t*)nnz; \ + uint8_t *bS_t = (uint8_t*)bS; \ + mask_mv <<= 3; \ + for (; b_idx < edges; b_idx += step) { \ + out &= mask_dir; \ + if (!(mask_mv & b_idx)) { \ + if (bidir) { \ + ref_2 = LD_SB(ref_t + d_idx_12); \ + ref_3 = LD_SB(ref_t + d_idx_52); \ + ref_0 = LD_SB(ref_t + 12); \ + ref_1 = LD_SB(ref_t + 52); \ + ref_2 = (v16i8)__msa_ilvr_w((v4i32)ref_3, (v4i32)ref_2); \ + ref_0 = (v16i8)__msa_ilvr_w((v4i32)ref_0, (v4i32)ref_0); \ + ref_1 = (v16i8)__msa_ilvr_w((v4i32)ref_1, (v4i32)ref_1); \ + ref_3 = (v16i8)__msa_shf_h((v8i16)ref_2, 0x4e); \ + ref_0 -= ref_2; \ + ref_1 -= ref_3; \ + ref_0 = (v16i8)__msa_or_v((v16u8)ref_0, (v16u8)ref_1); \ +\ + tmp_2 = LD_SH(mv_t + d_idx_x4_48); \ + tmp_3 = LD_SH(mv_t + 48); \ + tmp_4 = LD_SH(mv_t + 208); \ + tmp_5 = tmp_2 - tmp_3; \ + tmp_6 = tmp_2 - tmp_4; \ + SAT_SH2_SH(tmp_5, tmp_6, 7); \ + tmp_0 = __msa_pckev_b((v16i8)tmp_6, (v16i8)tmp_5); \ + tmp_0 += cnst_1; \ + tmp_0 = (v16i8)__msa_subs_u_b((v16u8)tmp_0, (v16u8)cnst_0);\ + tmp_0 = (v16i8)__msa_sat_s_h((v8i16)tmp_0, 7); \ + tmp_0 = __msa_pckev_b(tmp_0, tmp_0); \ + out = (v16i8)__msa_or_v((v16u8)ref_0, (v16u8)tmp_0); \ +\ + tmp_2 = LD_SH(mv_t + 208 + d_idx_x4); \ + tmp_5 = tmp_2 - tmp_3; \ + tmp_6 = tmp_2 - tmp_4; \ + SAT_SH2_SH(tmp_5, tmp_6, 7); \ + tmp_1 = __msa_pckev_b((v16i8)tmp_6, (v16i8)tmp_5); \ + tmp_1 += cnst_1; \ + tmp_1 = (v16i8)__msa_subs_u_b((v16u8)tmp_1, (v16u8)cnst_0); \ + tmp_1 = (v16i8)__msa_sat_s_h((v8i16)tmp_1, 7); \ + tmp_1 = __msa_pckev_b(tmp_1, tmp_1); \ +\ + tmp_1 = (v16i8)__msa_shf_h((v8i16)tmp_1, 0x4e); \ + out = (v16i8)__msa_or_v((v16u8)out, (v16u8)tmp_1); \ + tmp_0 = (v16i8)__msa_shf_h((v8i16)out, 0x4e); \ + out = (v16i8)__msa_min_u_b((v16u8)out, (v16u8)tmp_0); \ + } else { \ + ref_0 = LD_SB(ref_t + d_idx_12); \ + ref_3 = LD_SB(ref_t + 12); \ + tmp_2 = LD_SH(mv_t + d_idx_x4_48); \ + tmp_3 = LD_SH(mv_t + 48); \ + tmp_4 = tmp_3 - tmp_2; \ + tmp_1 = (v16i8)__msa_sat_s_h(tmp_4, 7); \ + tmp_1 = __msa_pckev_b(tmp_1, tmp_1); \ + tmp_1 += cnst_1; \ + out = (v16i8)__msa_subs_u_b((v16u8)tmp_1, (v16u8)cnst_0); \ + out = (v16i8)__msa_sat_s_h((v8i16)out, 7); \ + out = __msa_pckev_b(out, out); \ + ref_0 = ref_3 - ref_0; \ + out = (v16i8)__msa_or_v((v16u8)out, (v16u8)ref_0); \ + } \ + } \ + tmp_0 = LD_SB(nnz_t + 12); \ + tmp_1 = LD_SB(nnz_t + d_idx_12); \ + tmp_0 = (v16i8)__msa_or_v((v16u8)tmp_0, (v16u8)tmp_1); \ + tmp_0 = (v16i8)__msa_min_u_b((v16u8)tmp_0, (v16u8)cnst_2); \ + out = (v16i8)__msa_min_u_b((v16u8)out, (v16u8)cnst_2); \ + tmp_0 = (v16i8)((v8i16)tmp_0 << 1); \ + tmp_0 = (v16i8)__msa_max_u_b((v16u8)out, (v16u8)tmp_0); \ + tmp_0 = __msa_ilvr_b(zero, tmp_0); \ + ST_D1(tmp_0, 0, bS_t + dir_x32); \ + ref_t += step; \ + mv_t += step_x4; \ + nnz_t += step; \ + bS_t += step; \ + } \ +} while(0) + +void ff_h264_loop_filter_strength_msa(int16_t bS[2][4][4], uint8_t nnz[40], + int8_t ref[2][40], int16_t mv[2][40][2], + int bidir, int edges, int step, + int mask_mv0, int mask_mv1, int field) +{ + v16i8 out; + v16i8 ref_0, ref_1, ref_2, ref_3; + v16i8 tmp_0, tmp_1; + v8i16 tmp_2, tmp_3, tmp_4, tmp_5, tmp_6; + v16i8 cnst_0, cnst_1, cnst_2; + v16i8 zero = { 0 }; + v16i8 one = __msa_fill_b(0xff); + if (field) { + cnst_0 = (v16i8)__msa_fill_h(0x206); + cnst_1 = (v16i8)__msa_fill_h(0x103); + cnst_2 = (v16i8)__msa_fill_h(0x101); + } else { + cnst_0 = __msa_fill_b(0x6); + cnst_1 = __msa_fill_b(0x3); + cnst_2 = __msa_fill_b(0x1); + } + step <<= 3; + edges <<= 3; + + h264_loop_filter_strength_iteration_msa(edges, step, mask_mv1, 1, -8, zero); + h264_loop_filter_strength_iteration_msa(32, 8, mask_mv0, 0, -1, one); + + LD_SB2((int8_t*)bS, 16, tmp_0, tmp_1); + tmp_2 = (v8i16)__msa_ilvl_d((v2i64)tmp_0, (v2i64)tmp_0); + tmp_3 = (v8i16)__msa_ilvl_d((v2i64)tmp_1, (v2i64)tmp_1); + TRANSPOSE4x4_SH_SH(tmp_0, tmp_2, tmp_1, tmp_3, tmp_2, tmp_3, tmp_4, tmp_5); + tmp_0 = (v16i8)__msa_ilvr_d((v2i64)tmp_3, (v2i64)tmp_2); + tmp_1 = (v16i8)__msa_ilvr_d((v2i64)tmp_5, (v2i64)tmp_4); + ST_SB2(tmp_0, tmp_1, (int8_t*)bS, 16); +} diff --git a/libavcodec/mips/h264dsp_init_mips.c b/libavcodec/mips/h264dsp_init_mips.c index 9cd05e0..f822c38 100644 --- a/libavcodec/mips/h264dsp_init_mips.c +++ b/libavcodec/mips/h264dsp_init_mips.c @@ -78,6 +78,8 @@ av_cold void ff_h264dsp_init_mips(H264DSPContext *c, const int bit_depth, } if (have_msa(cpu_flags)) { + if (chroma_format_idc <= 1) + c->h264_loop_filter_strength = ff_h264_loop_filter_strength_msa; 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; diff --git a/libavcodec/mips/h264dsp_mips.h b/libavcodec/mips/h264dsp_mips.h index 35e16c4..5847ef3 100644 --- a/libavcodec/mips/h264dsp_mips.h +++ b/libavcodec/mips/h264dsp_mips.h @@ -319,6 +319,10 @@ void ff_vp8_pred8x8_129_dc_8_msa(uint8_t *src, ptrdiff_t stride); void ff_vp8_pred16x16_127_dc_8_msa(uint8_t *src, ptrdiff_t stride); void ff_vp8_pred16x16_129_dc_8_msa(uint8_t *src, ptrdiff_t stride); +void ff_h264_loop_filter_strength_msa(int16_t bS[2][4][4], uint8_t nnz[40], + int8_t ref[2][40], int16_t mv[2][40][2], int bidir, int edges, + int step, int mask_mv0, int mask_mv1, int field); + void ff_h264_add_pixels4_8_mmi(uint8_t *_dst, int16_t *_src, int stride); void ff_h264_idct_add_8_mmi(uint8_t *dst, int16_t *block, int stride); void ff_h264_idct8_add_8_mmi(uint8_t *dst, int16_t *block, int stride);