In caffe-opencl branch, it seems that the dropout kernel in dropout_layer.cu has not been executed. There should be some codes like 'kernel->Execute(group, local);'
Steps to reproduce
We could add a dropout layer with dropout ratio 1.0. And the network will be trained without any problems.
For example, in examples/mnist/lenet_train_test.prototxt, we could add a new dropout layer with dropout ratio 1.0, and then trains lenet.
Tried solutions
Add 'kernel->Execute(group, local);' statements in the Forward_gpu and Backward_gpu methods in dropoutlayer.cu. Right after the statement 'this->device->get_threads(&work_size, &group, &local, kernel.get(), true);'.
Issue summary
In caffe-opencl branch, it seems that the dropout kernel in dropout_layer.cu has not been executed. There should be some codes like 'kernel->Execute(group, local);'
Steps to reproduce
We could add a dropout layer with dropout ratio 1.0. And the network will be trained without any problems. For example, in examples/mnist/lenet_train_test.prototxt, we could add a new dropout layer with dropout ratio 1.0, and then trains lenet.
Tried solutions
Add 'kernel->Execute(group, local);' statements in the Forward_gpu and Backward_gpu methods in dropoutlayer.cu. Right after the statement 'this->device->get_threads(&work_size, &group, &local, kernel.get(), true);'.
System configuration