CUDA_ERROR_LAUNCH_FAILED when running GMMNLSE_driver_gpu_1550_linear.m

[SeveNOlogy7]

Hi,

I came across a CUDA_ERROR_LAUNCH_FAILED error when trying to run the GMMNLSE_driver_gpu_1550_linear.m file. The full error output is as follows.

Warning: An unexpected error occurred during CUDA execution. The CUDA error was:
CUDA_ERROR_LAUNCH_FAILED 
> In GMMNLSE_propagate (line 37)
  In GMMNLSE_driver_gpu_1550_linear (line 74) 
Error using gpuArray/ifft
An error occurred during PTX compilation of <image>.
The information log was:

The error log was:

The CUDA error code was: CUDA_ERROR_LAUNCH_FAILED.

Error in GMMNLSE_MPA_step (line 96)
    Vpl = dt*fft(hrw.*ifft(Vpl));

Error in GMMNLSE_propagate (line 308)
        [num_it, last_result] = GMMNLSE_MPA_step(last_result, initial_condition.dt, sim, nonlin_const, mode_info, omegas, D_pos,
        D_neg, hrw);

Error in GMMNLSE_driver_gpu_1550_linear (line 74)
prop_output = GMMNLSE_propagate(fiber, initial_condition, sim); % This actually does the propagation

I believe the CUDA and VC toolchains I installed were working just fine. And other .m files like GMMNLSE_driver_gpu_1550_GRINMMS_XPM.m etc worked normally on my PC.

The error may be related to Line 22, fiber.SR = 0*SR; , in the GMMNLSE_driver_gpu_1550_linear.m file. Once I changed that line to like fiber.SR = 1e-6*SR; , the error disappeared and the result is almost the same to those shown in the Advanced Example pdf.

This could be a temporary solution for me and I don't know if other people had the same issue. Also, it would be nice if anyone knows a properer solution for this error. Thanks.

I will post my computing environment here for your reference.

Intel Xeon CPU E5-2620 v4@2.1GHz

CUDADevice with properties:

                  Name: 'Quadro K620'
                 Index: 1
     ComputeCapability: '5.0'
        SupportsDouble: 1
         DriverVersion: 9.1000
        ToolkitVersion: 8
    MaxThreadsPerBlock: 1024
      MaxShmemPerBlock: 49152
    MaxThreadBlockSize: [1024 1024 64]
           MaxGridSize: [2.1475e+09 65535 65535]
             SIMDWidth: 32
           TotalMemory: 2.1475e+09
   MultiprocessorCount: 3
          ClockRateKHz: 1124000
           ComputeMode: 'Default'
  GPUOverlapsTransfers: 1
KernelExecutionTimeout: 0
      CanMapHostMemory: 1
       DeviceSupported: 1
        DeviceSelected: 1

Windows10 Ver.1709

Cuda compilation tools, release 9.1, V9.1.85

Visual C++ 2015 toolset

[AaHaHaa]

This is a bug of the "GMMNLSE_driver_gpu_1550_linear.m" code.

The zeros of SR are excluded to save the computational time, so SR can't be all zero, which will make the solver find no SR and generates an error. For this case, the SR needs to to "small" to have a "linear" effect, as if propagating in a really large core fiber.