Debug self host: check whether the original kernel produces errors

kchristin22 · Oct 8, 2024 · 54c8954 · 54c8954
1 parent 40ead4c
commit 54c8954
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/test/CUDA/GradientKernels.cu b/test/CUDA/GradientKernels.cu
@@ -619,6 +619,10 @@ int main(void) {
   cudaMemcpy(dy, fives, 10 * sizeof(double), cudaMemcpyHostToDevice);
   cudaMemcpy(d_val, zeros, sizeof(double), cudaMemcpyHostToDevice);
 
+  device_pullback<<<1, 10>>>(y, x, 5);
+  cudaDeviceSynchronize();
+  printf("%s\n", cudaGetErrorString(cudaGetLastError())); // CHECK-EXEC: no error
+
   auto test_device = clad::gradient(device_pullback, "out, val");
   test_device.execute_kernel(dim3(1), dim3(10, 1, 1), x, y, 5, dy, d_val);
   cudaDeviceSynchronize();