cudaGraph_t graph;
cudaGraphExec_t graphExec;
cudaStream_t stream;
cudaStreamCreate(&stream);

cudaStreamBeginCapture(stream, cudaStreamCaptureModeGlobal);

kernelA<<<blocks, threads, 0, stream>>>(d_data);
kernelB<<<blocks, threads, 0, stream>>>(d_data);
kernelC<<<blocks, threads, 0, stream>>>(d_data);

cudaStreamEndCapture(stream, &graph);
cudaGraphInstantiate(&graphExec, graph, nullptr, nullptr, 0);

for (int iter = 0; iter < numIterations; ++iter) {
    cudaGraphLaunch(graphExec, stream);   // replays kernelA, B, C with one call
}
cudaStreamSynchronize(stream);

cudaGraphExecDestroy(graphExec);
cudaGraphDestroy(graph);
