Fixed CPU-RAM + GPU-processing bug for Dropout-layer

src/network_kernels.cu

@@ -112,11 +112,13 @@ void backward_network_gpu(network net, network_state state)
         if (l.onlyforward) continue;
         l.backward_gpu(l, state);
 
+        if (i != 0) {
             layer prev = net.layers[i - 1];
-        if (net.optimized_memory && state.delta && !prev.keep_delta_gpu) {
+            if (net.optimized_memory && state.delta && !prev.keep_delta_gpu && prev.delta_gpu != l.delta_gpu) {
                 simple_copy_ongpu(prev.outputs*prev.batch, state.delta, prev.delta_gpu);
                 fill_ongpu(prev.outputs*prev.batch, 0, net.state_delta_gpu, 1);
             }
+        }
 
         /*
         if(i != 0)

src/parser.c

@@ -1148,6 +1148,7 @@ network parse_network_cfg_custom(char *filename, int batch, int time_steps)
 #ifdef GPU
             l.output_gpu = net.layers[count-1].output_gpu;
             l.delta_gpu = net.layers[count-1].delta_gpu;
+            l.keep_delta_gpu = 1;
 #endif
         }
         else if (lt == EMPTY) {
@@ -1168,7 +1169,7 @@ network parse_network_cfg_custom(char *filename, int batch, int time_steps)
 
 #ifdef GPU
         // futher GPU-memory optimization: net.optimized_memory == 2
-        if (net.optimized_memory >= 2 && params.train)
+        if (net.optimized_memory >= 2 && params.train && l.type != DROPOUT)
         {
             l.optimized_memory = net.optimized_memory;
             if (l.output_gpu) {
@@ -1187,7 +1188,7 @@ network parse_network_cfg_custom(char *filename, int batch, int time_steps)
             }
 
             // maximum optimization
-            if (net.optimized_memory >= 3) {
+            if (net.optimized_memory >= 3 && l.type != DROPOUT) {
                 if (l.delta_gpu) {
                     cuda_free(l.delta_gpu);
                     //l.delta_gpu = cuda_make_array_pinned_preallocated(NULL, l.batch*l.outputs); // l.steps
@@ -1257,7 +1258,7 @@ network parse_network_cfg_custom(char *filename, int batch, int time_steps)
             }
 
             // maximum optimization
-            if (net.optimized_memory >= 3) {
+            if (net.optimized_memory >= 3 && l.type != DROPOUT) {
                 if (l.delta_gpu && l.keep_delta_gpu) {
                     //cuda_free(l.delta_gpu);   // already called above
                     l.delta_gpu = cuda_make_array_pinned_preallocated(NULL, l.batch*l.outputs); // l.steps