cuda thrust::remove_if throws "thrust::system::system_error" for device_vector?

Question

I am currently using CUDA 7.5 under VS 2013. Today I needed to remove some of the elements from a device_vector, thus decided to use remove_if. But however I modify the code, the program just compiles well but throws "thrust::system::system_error" at run time.

Firstly I tried my own code:

int main()
{
    thrust::host_vector<int> AA(10, 1);
    thrust::sequence(AA.begin(), AA.end());
    thrust::host_vector<bool> SS(10,false);
    thrust::fill(SS.begin(), SS.begin() + 5, true);
    thrust::device_vector<int> devAA=AA;
    thrust::device_vector<bool> devSS = SS;
    thrust::device_vector<int>::iterator new_end = thrust::remove_if(thrust::device,
    devAA.begin(), devAA.end(), devSS.begin(), thrust::identity<int>());
}

But it throws thrust::system::system_error at run time. However, if I use two host_vector, i.e. AA and SS to perform remove_if, everything goes fine.

Then, I tried the code I found on stackoverflow here, the code in Robert Crovella's answer seemed work fine, but on my machine, it still throws thrust::system::system_error.

Did new version of thrust modify anything? Or I should try some other way? I am using cmake to organise the code, is there any thing special?

Could you explain your problem a little better? Is the code you have posted working for you or not? When you say " However, if I use two host_vector, i.e. AA and SS to perform remove_if, everything goes fine" what does that mean relative to the code you have posted? — talonmies
I don't have any trouble with what you have posted. Here is my test case. You might have a problem with your machine setup or CUDA install. — Robert Crovella
If you add precise thrust error checking like this, then it's likely you'll get more information about why the system error is occurring. For example, if I compile that code for compute_30,sm_30 but run on my cc2.0 device, I get the following: "thrust system error: cudaFuncGetAttributes in function_attributes: invalid device function" and the "invalid device function" message is a good indicator that I have compiled for an incorrect GPU type. — Robert Crovella
I opened your project file. Are you building a Win32 project? You should be building a x64 release project. Anyway, I don't have any trouble building and running your project on my machine if I build an x64 release project. Yes, if you build a win32 project it will not work. — Robert Crovella
Official support for 32-bit is disappearing in newer CUDA versions. In CUDA 7.5 it is almost gone. — Robert Crovella

Robert Crovella Robert Crovella · Accepted Answer · 2015-12-02T02:52:51

The problem appears to be that OP was building a 32-bit project. The issue was resolved when switching to a 64-bit project.

My recommendation for CUDA 7.5 and beyond is to only use 64-bit projects. If you review the current state of 32-bit support on windows and linux you'll find it's quite limited.

Purely as a matter of conjecture, this issue may be related to thrust issue #715

cuda thrust::remove_if throws "thrust::system::system_error" for device_vector?

1 Answers