asynchronous - Why am I not getting I/O-compute overlap with this code?

Question

The following program:

#include <iostream>
#include <array>

using clock_value_t = long long;

__device__ void gpu_sleep(clock_value_t sleep_cycles)
{
    clock_value_t start = clock64();
    clock_value_t cycles_elapsed;
    do { cycles_elapsed = clock64() - start; }
    while (cycles_elapsed < sleep_cycles);
}

__global__ void dummy(clock_value_t duration_in_cycles)
{
    gpu_sleep(duration_in_cycles);
}

int main()
{
    const clock_value_t duration_in_clocks = 1e7;
    const size_t buffer_size = 5e7;
    constexpr const auto num_streams = 2;

    std::array<char*, num_streams> host_ptrs;
    std::array<char*, num_streams> device_ptrs;
    std::array<cudaStream_t, num_streams> streams;
    for (auto i=0; i<num_streams; i++) {
        cudaMallocHost(&host_ptrs[i], buffer_size);
        cudaMalloc(&device_ptrs[i], buffer_size);
        cudaStreamCreateWithFlags(&streams[i], cudaStreamNonBlocking);
    }
    cudaDeviceSynchronize();
    for (auto i=0; i<num_streams; i++) {
        cudaMemcpyAsync(device_ptrs[i], host_ptrs[i], buffer_size, 
            cudaMemcpyDefault, streams[i]);
        dummy<<<128, 128, 0, streams[i]>>>(duration_in_clocks);
        cudaMemcpyAsync(host_ptrs[i], device_ptrs[i], buffer_size, 
            cudaMemcpyDefault, streams[i]);
    }
    for (auto i=0; i<num_streams; i++) { cudaStreamSynchronize(streams[i]); }
    for (auto i=0; i<num_streams; i++) {
        cudaFreeHost(host_ptrs[i]);
        cudaFree(device_ptrs[i]);
    }
}

should result in overlapping I/O and Compute between the work on the first and second streams: When the first stream's Host-to-Device ends, the first stream's kernel can start, but so can the second stream's Host-to-Device transfer. Instead, I get the following timeline, with no overlap:

I think I've covered my bases to ensure overlap. The streams are non-blocking (and indeed the enqueueing of work concludes well before the first HtoD does); the host memory is pinned... so what's missing for me to see overlap?

Using CUDA 8.0.61 on GNU/Linux Mint 18.2 with an NVIDIA GTX 650 Ti Boost. But the driver is v384.59.

Answer 1

Ok, it must be something with my GPU model, because with Fedora 25, and a GTX Titan X, I get: