--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/example/Cuda/main.cc	Fri Jan 24 07:16:26 2014 +0900
@@ -0,0 +1,76 @@
+#include <stdio.h>
+
+#include <cuda.h>
+
+#define LENGTH 1000
+
+void check_data(float* A,float* B,float* C) {
+    for (int i=0; i<LENGTH; i++) {
+        if (A[i]*B[i]!=C[i]) {
+            puts("failure.");
+            return;
+        }
+    }
+    puts("success.");
+    return;
+}
+
+
+
+int main() {
+    CUdevice device;
+    CUcontext context;
+    CUmodule module;
+    CUfunction function;
+    //    CUresult result;
+
+    cuInit(0);
+    cuDeviceGet(&device, 0);
+    cuCtxCreate(&context, 0, device);
+    cuModuleLoad(&module, "multiply.cu");
+    cuModuleGetFunction(&function, module, "multiply");
+
+    float* A = new float[LENGTH];
+    float* B = new float[LENGTH];
+    float* C = new float[LENGTH];
+
+    for (int i=0; i<LENGTH; i++) {
+        A[i] = (float)(i+1000);
+        B[i] = (float)(i+1)/10.f;
+    }
+
+    CUdeviceptr devA,devB,devC;
+
+    cuMemAlloc(&devA, LENGTH*sizeof(float));
+    cuMemAlloc(&devB, LENGTH*sizeof(float));
+    cuMemAlloc(&devC, LENGTH*sizeof(float));
+
+    cuMemcpyHtoD(devA, A, LENGTH*sizeof(float));
+    cuMemcpyHtoD(devB, B, LENGTH*sizeof(float));
+    cuMemcpyHtoD(devC, C, LENGTH*sizeof(float));
+
+    cuParamSetv(function, 0, A, LENGTH*sizeof(float));
+    cuParamSetv(function, 0, B, LENGTH*sizeof(float));
+    cuParamSetv(function, 0, C, LENGTH*sizeof(float));
+
+    cuLaunchKernel(function,
+                   LENGTH, 1, 1,
+                   1, 1, 1,
+                   0, NULL, NULL, NULL);
+
+    cuMemcpyDtoH(C, devC, LENGTH*sizeof(float));
+
+    check_data(A, B, C);
+
+    delete[] A;
+    delete[] B;
+    delete[] C;
+    cuModuleUnload(module);
+    cuMemFree(devA);
+    cuMemFree(devB);
+    cuMemFree(devC);
+    
+
+    return 0;
+}
+