Fixed multi-GPU training for Tensor Cores

This commit is contained in:
AlexeyAB
2018-03-09 19:44:46 +03:00
parent a6c51e3b75
commit 880cf187d8
5 changed files with 48 additions and 35 deletions

View File

@ -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);