From patchwork Tue Apr 2 14:29:09 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Guo, Yejun" X-Patchwork-Id: 12570 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 5716644702A for ; Tue, 2 Apr 2019 09:39:36 +0300 (EEST) Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id 469FC68AC4C; Tue, 2 Apr 2019 09:39:36 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mga06.intel.com (mga06.intel.com [134.134.136.31]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id 938DB68AC3E for ; Tue, 2 Apr 2019 09:39:34 +0300 (EEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga003.jf.intel.com ([10.7.209.27]) by orsmga104.jf.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 01 Apr 2019 23:39:32 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.60,298,1549958400"; d="scan'208";a="139235275" Received: from yguo18-skl-u1604.sh.intel.com ([10.239.13.25]) by orsmga003.jf.intel.com with ESMTP; 01 Apr 2019 23:39:32 -0700 From: "Guo, Yejun" To: ffmpeg-devel@ffmpeg.org Date: Tue, 2 Apr 2019 22:29:09 +0800 Message-Id: <1554215349-13962-1-git-send-email-yejun.guo@intel.com> X-Mailer: git-send-email 2.7.4 Subject: [FFmpeg-devel] [PATCH 5/8] libavfilter/dnn: avoid memcpy for tensorflow dnn output 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: yejun.guo@intel.com MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" use TF_Tensor's cpu address to avoid extra memcpy. Signed-off-by: Guo, Yejun --- libavfilter/dnn_backend_tf.c | 37 ++++++++++++------------------------- libavfilter/vf_sr.c | 3 --- 2 files changed, 12 insertions(+), 28 deletions(-) diff --git a/libavfilter/dnn_backend_tf.c b/libavfilter/dnn_backend_tf.c index 7966688..be8401e 100644 --- a/libavfilter/dnn_backend_tf.c +++ b/libavfilter/dnn_backend_tf.c @@ -35,6 +35,7 @@ typedef struct TFModel{ TF_Status *status; TF_Output input, output; TF_Tensor *input_tensor; + TF_Tensor *output_tensor; } TFModel; static void free_buffer(void *data, size_t length) @@ -460,14 +461,11 @@ DNNModel *ff_dnn_load_model_tf(const char *model_filename) return NULL; } - tf_model = av_malloc(sizeof(TFModel)); + tf_model = av_mallocz(sizeof(TFModel)); if (!tf_model){ av_freep(&model); return NULL; } - tf_model->session = NULL; - tf_model->input_tensor = NULL; - tf_model->output_data = NULL; if (load_tf_model(tf_model, model_filename) != DNN_SUCCESS){ if (load_native_model(tf_model, model_filename) != DNN_SUCCESS){ @@ -489,36 +487,22 @@ DNNModel *ff_dnn_load_model_tf(const char *model_filename) DNNReturnType ff_dnn_execute_model_tf(const DNNModel *model, DNNData *output) { TFModel *tf_model = (TFModel *)model->model; - TF_Tensor *output_tensor; - uint32_t count; - uint32_t old_count = output->height * output->width * output->channels * sizeof(float); + if (tf_model->output_tensor) + TF_DeleteTensor(tf_model->output_tensor); TF_SessionRun(tf_model->session, NULL, &tf_model->input, &tf_model->input_tensor, 1, - &tf_model->output, &output_tensor, 1, + &tf_model->output, &tf_model->output_tensor, 1, NULL, 0, NULL, tf_model->status); if (TF_GetCode(tf_model->status) != TF_OK){ return DNN_ERROR; } - output->height = TF_Dim(output_tensor, 1); - output->width = TF_Dim(output_tensor, 2); - output->channels = TF_Dim(output_tensor, 3); - count = output->height * output->width * output->channels * sizeof(float); - if (output->data) { - if (count > old_count) { - av_freep(&output->data); - } - } - if (!output->data) { - output->data = av_malloc(count); - if (!output->data){ - return DNN_ERROR; - } - } - memcpy(output->data, TF_TensorData(output_tensor), count); - TF_DeleteTensor(output_tensor); + output->height = TF_Dim(tf_model->output_tensor, 1); + output->width = TF_Dim(tf_model->output_tensor, 2); + output->channels = TF_Dim(tf_model->output_tensor, 3); + output->data = TF_TensorData(tf_model->output_tensor); return DNN_SUCCESS; } @@ -542,6 +526,9 @@ void ff_dnn_free_model_tf(DNNModel **model) if (tf_model->input_tensor){ TF_DeleteTensor(tf_model->input_tensor); } + if (tf_model->output_tensor){ + TF_DeleteTensor(tf_model->output_tensor); + } av_freep(&tf_model); av_freep(model); } diff --git a/libavfilter/vf_sr.c b/libavfilter/vf_sr.c index 7c92730..53bd8ea 100644 --- a/libavfilter/vf_sr.c +++ b/libavfilter/vf_sr.c @@ -277,9 +277,6 @@ static av_cold void uninit(AVFilterContext *context) int i; SRContext *sr_context = context->priv; - if (sr_context->backend_type == DNN_TF) - av_freep(&sr_context->output.data); - if (sr_context->dnn_module){ (sr_context->dnn_module->free_model)(&sr_context->model); av_freep(&sr_context->dnn_module);