-
Notifications
You must be signed in to change notification settings - Fork 17
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Out of bounds write in kernel20 #4
Comments
Hi Alex, I cannot reproduce the issue with the default setting. Have you changed any options or anything? The only memory writing access in matupd::kernel20 is writing to buf_array_p at Line 112 of matupd.cu. According to the cuda-memcheck info, for block (135, 0, 0), the writing address is simply 135, while buf_array_p should have size initialized with 8272 (Line 143 of lbfgsbcuda.cpp). Then there should not be a memory access error as long as buf_array_p is correctly allocated. Therefore, the only reason I can think of is that the GPU memory is somehow not correctly allocated. You may verify if buf_array_p has been initialized into the correct size at Line 143 of lbfgsbcuda.cpp. Could it be some driver issue? |
Thanks for looking into it. git clone [email protected]:nepluno/lbfgsb-gpu.git
cd lbfgsb-gpu
mkdir build
cd build
cmake ..
make
./bin/dsscfg That size
It looks like the actual exception is:
So I agree that this looks like a memory allocation failure. Any further suggestion for how to track down the root cause? |
Hello,
I am looking to use this library in a project so I just compiled it and ran the test
dsscfg
program and it failed with what looks like an out-of-bounds write to a global memory array somewhere in thematupd::kernel20
kernel function.This is on a GeForce RTX 2060 Super (CC 7.5) with CUDA 11.2.
Here's the cuda-memcheck output:
The text was updated successfully, but these errors were encountered: