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

Fix conv2d with additional bias for tensorRT. Fix reshape deserialize. Mb2512 works with tensorRT

Browse Source 
Signed-off-by: Micaela Verucchi <micaelaverucchi@gmail.com>
This commit is contained in:
Micaela Verucchi
2020-03-17 13:18:05 +01:00
parent 14d850751d
commit 1a50065bde
4 changed files with 23 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/LayerWgs.cpp
+8
View File
@@ -59,6 +59,14 @@ LayerWgs::LayerWgs(Network *net, int inputs, int outputs,
float2half(data_d, data16_d, w_size);
cudaMemcpy(data16_h, data16_d, w_size*sizeof(__half), cudaMemcpyDeviceToHost);
if(additional_bias){
int b2_size = outputs;
bias216_h = new __half[b2_size];
cudaMalloc(&bias216_d, w_size*sizeof(__half));
float2half(bias2_d, bias216_d, b2_size);
cudaMemcpy(bias216_h, bias216_d, b2_size*sizeof(__half), cudaMemcpyDeviceToHost);
}
int b_size = outputs;
bias16_h = new __half[b_size];
cudaMalloc(&bias16_d, w_size*sizeof(__half));