[FFmpeg-devel] [PATCH v3 2/2] dnn_backend_native_layer_conv2d.c: refine code.

xujunzz at sjtu.edu.cn xujunzz at sjtu.edu.cn
Wed Sep 16 13:07:19 EEST 2020


From: Xu Jun <xujunzz at sjtu.edu.cn>

Move thread area allocate out of thread function into
main thread.

Signed-off-by: Xu Jun <xujunzz at sjtu.edu.cn>
---
 .../dnn/dnn_backend_native_layer_conv2d.c     | 30 +++++++++----------
 1 file changed, 14 insertions(+), 16 deletions(-)

diff --git a/libavfilter/dnn/dnn_backend_native_layer_conv2d.c b/libavfilter/dnn/dnn_backend_native_layer_conv2d.c
index 5c313454f7..2aaa4162df 100644
--- a/libavfilter/dnn/dnn_backend_native_layer_conv2d.c
+++ b/libavfilter/dnn/dnn_backend_native_layer_conv2d.c
@@ -33,12 +33,11 @@ typedef struct thread_common_param{
     const void *parameters;
     NativeContext *ctx;
     float *output_data;
-    int thread_num;
 } thread_common_param;
 
 typedef struct thread_param{
     thread_common_param *thread_common_param;
-    int thread_index;
+    int thread_start, thread_end;
 } thread_param;
 
 int dnn_load_layer_conv2d(Layer *layer, AVIOContext *model_file_context, int file_size, int operands_num)
@@ -125,16 +124,12 @@ static void * dnn_execute_layer_conv2d_thread(void *threadarg)
     int filter_size = conv_params->kernel_size * filter_linesize;
     int pad_size = (conv_params->padding_method == VALID) ? (conv_params->kernel_size - 1) / 2 * conv_params->dilation : 0;
 
-    int thread_stride = (height - pad_size * 2) / thread_common_param->thread_num;
-    int thread_start = thread_stride * thread_param->thread_index + pad_size;
-    int thread_end = (thread_param->thread_index == thread_common_param->thread_num - 1) ? (height - pad_size) : (thread_start + thread_stride);
-
     float *output = thread_common_param->output_data;
-    output += (conv_params->output_num) * (width - 2 * pad_size) * (thread_start - pad_size);
+    output += (conv_params->output_num) * (width - 2 * pad_size) * (thread_param->thread_start - pad_size);
 
     av_assert0(channel == conv_params->input_num);
 
-    for (int y = thread_start; y < thread_end; ++y) {
+    for (int y = thread_param->thread_start; y < thread_param->thread_end; ++y) {
         for (int x = pad_size; x < width - pad_size; ++x) {
             for (int n_filter = 0; n_filter < conv_params->output_num; ++n_filter) {
                 if (conv_params->has_bias)
@@ -193,16 +188,19 @@ int dnn_execute_layer_conv2d(DnnOperand *operands, const int32_t *input_operand_
         ? (av_cpu_count() + 1) : (ctx->options.conv2d_threads);
 #if HAVE_PTHREAD_CANCEL
     pthread_t *thread_id = av_malloc(thread_num * sizeof(pthread_t));
+    int thread_stride;
 #endif
     thread_param **thread_param = av_malloc(thread_num * sizeof(*thread_param));
     thread_common_param thread_common_param;
     const ConvolutionalParams *conv_params = (const ConvolutionalParams *)(parameters);
+    int height = operands[input_operand_indexes[0]].dims[1];
+    int width = operands[input_operand_indexes[0]].dims[2];
     int pad_size = (conv_params->padding_method == VALID) ? (conv_params->kernel_size - 1) / 2 * conv_params->dilation : 0;
     DnnOperand *output_operand = &operands[output_operand_index];
 
     output_operand->dims[0] = operands[input_operand_indexes[0]].dims[0];
-    output_operand->dims[1] = operands[input_operand_indexes[0]].dims[1] - pad_size * 2;
-    output_operand->dims[2] = operands[input_operand_indexes[0]].dims[2] - pad_size * 2;
+    output_operand->dims[1] = height - pad_size * 2;
+    output_operand->dims[2] = width - pad_size * 2;
     output_operand->dims[3] = conv_params->output_num;
     output_operand->data_type = operands[input_operand_indexes[0]].data_type;
     output_operand->length = calculate_operand_data_length(output_operand);
@@ -223,13 +221,13 @@ int dnn_execute_layer_conv2d(DnnOperand *operands, const int32_t *input_operand_
     thread_common_param.ctx = ctx;
 
 #if HAVE_PTHREAD_CANCEL
-    thread_common_param.thread_num = thread_num;
-
+    thread_stride = (height - pad_size * 2) / thread_num;
     //create threads
     for (int i = 0; i < thread_num; i++){
         thread_param[i] = av_malloc(sizeof(**thread_param));
         thread_param[i]->thread_common_param = &thread_common_param;
-        thread_param[i]->thread_index = i;
+        thread_param[i]->thread_start = thread_stride * i + pad_size;
+        thread_param[i]->thread_end = (i == thread_num - 1) ? (height - pad_size) : (thread_param[i]->thread_start + thread_stride);
         pthread_create(&thread_id[i], NULL, dnn_execute_layer_conv2d_thread, (void *)thread_param[i]);
     }
 
@@ -245,10 +243,10 @@ int dnn_execute_layer_conv2d(DnnOperand *operands, const int32_t *input_operand_
         av_free(thread_param[i]);
     }
 #else
-    thread_common_param.thread_num = 1;
-    thread_param[0] = av_malloc(sizeof(thread_param));
+    thread_param[0] = av_malloc(sizeof(**thread_param));
     thread_param[0]->thread_common_param = &thread_common_param;
-    thread_param[0]->thread_index = 0;
+    thread_param[0]->thread_start = 0;
+    thread_param[0]->thread_end = height - pad_size;
     dnn_execute_layer_conv2d_thread((void *)thread_param[0]);
     av_free(thread_param[0]);
 #endif
-- 
2.28.0



More information about the ffmpeg-devel mailing list