#include <cuda_runtime_api.h>
#include <pthread.h>
#include <stdio.h>
#include <sys/time.h>

int size = 1024*1024*300;

double gettime() {
	timeval t;
	gettimeofday(&t, NULL);
	return t.tv_sec + (1.e-6)*t.tv_usec;
}

pthread_barrier_t barrier;

void *test(void *varg) {
	int arg = *( (int*)varg );
	cudaSetDevice(arg);
	char *h_buf;
	cudaMallocHost((void **)&h_buf, size);
	char *d_buf;
	cudaMalloc((void **)&d_buf, size);

	pthread_barrier_wait(&barrier);
	cudaMemcpy(d_buf, h_buf, size, cudaMemcpyHostToDevice);
	cudaMemcpy(h_buf, d_buf, size, cudaMemcpyDeviceToHost);
	cudaThreadSynchronize();
	pthread_barrier_wait(&barrier);
	pthread_exit(NULL);
}

int main(int argc, char **argv) {
	pthread_t threads[2];
	int args[2];
	int cards = 0;

	int deviceCount;
	cudaGetDeviceCount(&deviceCount);
	for (int dev = 0; dev < deviceCount; ++dev) {
		cudaDeviceProp deviceProp;
		cudaGetDeviceProperties(&deviceProp, dev);
		if (strcmp(deviceProp.name, "GeForce 9800 GX2") == 0) {
			args[cards++] = dev;
		}
		if (cards == 2)
			break;
	}

	if (cards != 2) {
		printf("Only detected %i GX2s, exiting.\n", cards);
		exit(1);
	}

	printf("Testing %i and %i\n", args[0], args[1]);


	pthread_barrier_init(&barrier, NULL, cards+1);
	for (int i = 0; i < cards; i++) {
		int rc = pthread_create(&threads[i], NULL, test, 
				(void *)(args+i));
		if (rc) {
			perror("pthread_create\n");
			exit(-1);
		}
	}
	pthread_barrier_wait(&barrier);
	double time = -1*gettime();
	pthread_barrier_wait(&barrier);
	time += gettime();
	printf("%lf MB/s\n", double(size*2*cards)/time/1024.0/1024.0);
	pthread_exit(NULL);
}