From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org [79.124.17.100]) by master.gitmailbox.com (Postfix) with ESMTP id C18E240932 for ; Thu, 25 Jul 2024 20:25:32 +0000 (UTC) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 546D268D67A; Thu, 25 Jul 2024 23:25:29 +0300 (EEST) Received: from ursule.remlab.net (vps-a2bccee9.vps.ovh.net [51.75.19.47]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 7C51068D42D for ; Thu, 25 Jul 2024 23:25:23 +0300 (EEST) Received: from basile.remlab.net (localhost [IPv6:::1]) by ursule.remlab.net (Postfix) with ESMTP id 1A3C5C013D for ; Thu, 25 Jul 2024 23:25:23 +0300 (EEST) From: =?UTF-8?q?R=C3=A9mi=20Denis-Courmont?= To: ffmpeg-devel@ffmpeg.org Date: Thu, 25 Jul 2024 23:25:16 +0300 Message-ID: <20240725202522.276182-2-remi@remlab.net> X-Mailer: git-send-email 2.45.2 In-Reply-To: <20240725202522.276182-1-remi@remlab.net> References: <20240725202522.276182-1-remi@remlab.net> MIME-Version: 1.0 Subject: [FFmpeg-devel] [PATCH 2/6] lavc/audiodsp: properly unroll vector_clipf 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 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Archived-At: List-Archive: List-Post: Given that source and destination can alias, the compiler was forced to perform each read-modify-write sequentially. We cannot use the `restrict` qualifier to avoid this here because the AC-3 encoder uses the function in-place. Instead this commit provides an explicit guarantee to the compiler that batches of 8 elements will not overlap, so that it can interleave calculations. In practice contemporary optimising compilers are able to unroll and keep the temporary array in FPU registers (without spilling). On SiFive-U74, this speeds the same signs branch by 4x, and the opposite signs branch 1.5x. --- libavcodec/audiodsp.c | 40 +++++++++++++++++----------------------- 1 file changed, 17 insertions(+), 23 deletions(-) diff --git a/libavcodec/audiodsp.c b/libavcodec/audiodsp.c index c5427d3535..9e83f06aaa 100644 --- a/libavcodec/audiodsp.c +++ b/libavcodec/audiodsp.c @@ -38,41 +38,35 @@ static inline float clipf_c_one(float a, uint32_t mini, static void vector_clipf_c_opposite_sign(float *dst, const float *src, float min, float max, int len) { - int i; uint32_t mini = av_float2int(min); uint32_t maxi = av_float2int(max); uint32_t maxisign = maxi ^ (1U << 31); - for (i = 0; i < len; i += 8) { - dst[i + 0] = clipf_c_one(src[i + 0], mini, maxi, maxisign); - dst[i + 1] = clipf_c_one(src[i + 1], mini, maxi, maxisign); - dst[i + 2] = clipf_c_one(src[i + 2], mini, maxi, maxisign); - dst[i + 3] = clipf_c_one(src[i + 3], mini, maxi, maxisign); - dst[i + 4] = clipf_c_one(src[i + 4], mini, maxi, maxisign); - dst[i + 5] = clipf_c_one(src[i + 5], mini, maxi, maxisign); - dst[i + 6] = clipf_c_one(src[i + 6], mini, maxi, maxisign); - dst[i + 7] = clipf_c_one(src[i + 7], mini, maxi, maxisign); + for (int i = 0; i < len; i += 8) { + float tmp[8]; + + for (int j = 0; j < 8; j++) + tmp[j]= clipf_c_one(src[i + j], mini, maxi, maxisign); + for (int j = 0; j < 8; j++) + dst[i + j] = tmp[j]; } } static void vector_clipf_c(float *dst, const float *src, int len, float min, float max) { - int i; - if (min < 0 && max > 0) { vector_clipf_c_opposite_sign(dst, src, min, max, len); - } else { - for (i = 0; i < len; i += 8) { - dst[i] = av_clipf(src[i], min, max); - dst[i + 1] = av_clipf(src[i + 1], min, max); - dst[i + 2] = av_clipf(src[i + 2], min, max); - dst[i + 3] = av_clipf(src[i + 3], min, max); - dst[i + 4] = av_clipf(src[i + 4], min, max); - dst[i + 5] = av_clipf(src[i + 5], min, max); - dst[i + 6] = av_clipf(src[i + 6], min, max); - dst[i + 7] = av_clipf(src[i + 7], min, max); - } + return; + } + + for (int i = 0; i < len; i += 8) { + float tmp[8]; + + for (int j = 0; j < 8; j++) + tmp[j]= av_clipf(src[i + j], min, max); + for (int j = 0; j < 8; j++) + dst[i + j] = tmp[j]; } } -- 2.45.2 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".