mmr/tkDNN
Archived
1
0
Fork 0
Code Issues 46 Pull Requests 12 Releases 4 Activity

memory release

Browse Source
This commit is contained in:
Francesco Gatti
2020-06-02 12:43:06 +02:00
parent 8c79b75227
commit 4f907dde40
7 changed files with 20 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/LayerWgs.cpp
+3 -3
View File
@@ -80,7 +80,7 @@ LayerWgs::LayerWgs(Network *net, int inputs, int outputs,
variance16_h = new __half[b_size];
scales16_h = new __half[b_size];
//cudaMalloc(&power16_d, b_size*sizeof(__half));
cudaMalloc(&power16_d, b_size*sizeof(__half));
cudaMalloc(&mean16_d, b_size*sizeof(__half));
cudaMalloc(&variance16_d, b_size*sizeof(__half));
cudaMalloc(&scales16_d, b_size*sizeof(__half));
@@ -91,8 +91,8 @@ LayerWgs::LayerWgs(Network *net, int inputs, int outputs,
//init power array of ones
cudaMemcpy(tmp_d, power_h, b_size*sizeof(float), cudaMemcpyHostToDevice);
//float2half(tmp_d, power16_d, b_size);
//cudaMemcpy(power16_h, power16_d, b_size*sizeof(__half), cudaMemcpyDeviceToHost);
float2half(tmp_d, power16_d, b_size);
cudaMemcpy(power16_h, power16_d, b_size*sizeof(__half), cudaMemcpyDeviceToHost);
//mean array
cudaMemcpy(tmp_d, mean_h, b_size*sizeof(float), cudaMemcpyHostToDevice);