mirror of
https://github.com/pjreddie/darknet.git
synced 2023-08-10 21:13:14 +03:00
Fixed multi-GPU training for Tensor Cores
This commit is contained in:
@ -305,8 +305,8 @@ convolutional_layer make_convolutional_layer(int batch, int h, int w, int c, int
|
||||
|
||||
l.weights_gpu = cuda_make_array(l.weights, c*n*size*size);
|
||||
#ifdef CUDNN_HALF
|
||||
l.weights_gpu16 = cuda_make_array(l.weights, c*n*size*size / 2);
|
||||
l.weight_updates_gpu16 = cuda_make_array(l.weight_updates, c*n*size*size / 2);
|
||||
l.weights_gpu16 = cuda_make_array(NULL, c*n*size*size / 2); //cuda_make_array(l.weights, c*n*size*size / 2);
|
||||
l.weight_updates_gpu16 = cuda_make_array(NULL, c*n*size*size / 2); //cuda_make_array(l.weight_updates, c*n*size*size / 2);
|
||||
#endif
|
||||
l.weight_updates_gpu = cuda_make_array(l.weight_updates, c*n*size*size);
|
||||
|
||||
|
Reference in New Issue
Block a user