[WIP] Update backprop OpenMP GPU version

After hours of debugging, it's still not working.

1. Fix missing headers, function declarations, etc.
2. Use 1D array instead of 2D array due to OpenMP data mapping issue.
3. Fix data allocation and usage. Original code maps data[0:size] but use data[0:size+1] on the device.

The memory allocation is problematic. Depending on the input size, e.g. 64, 128, 1024, 65536 ..., the program will crash at different places.
The default input size 65536 may cause data overflow.

openmp-gpu/backprop/Makefile

@@ -1,12 +1,9 @@
 # C compiler
-#CC = pathcc
-#CC_FLAGS = -acc -O2
-CC = pgcc
-CC_FLAGS = -acc -O3 -Minfo -I /usr/include/i386-linux-gnu/ -L /usr/lib/i386-linux-gnu/
+CC = clang
+CC_FLAGS = -g -O3 -fopenmp -fopenmp-targets=nvptx64
 
-
-backprop: backprop.o facetrain.o imagenet.o backprop_kernel.o 
-	$(CC) $(CC_FLAGS) backprop.o facetrain.o imagenet.o backprop_kernel.o -o backprop -lm
+backprop.out: backprop.o facetrain.o imagenet.o backprop_kernel.o
+	$(CC) $(CC_FLAGS) backprop.o facetrain.o imagenet.o backprop_kernel.o -o backprop.out -lm
 
 %.o: %.[ch]
 	$(CC) $(CC_FLAGS) $< -c
@@ -23,6 +20,9 @@ backprop_kernel.o: backprop_kernel.c backprop.h
 imagenet.o: imagenet.c backprop.h
 	$(CC) $(CC_FLAGS) imagenet.c -c
 
+.PHONY: run clean
+run: backprop.out
+	./backprop.out 65536
 
 clean:
-	rm -f *.o *~ backprop backprop_cuda.linkinfo
+	rm -f *.o *.out