search

DavidDiazGuerra / gpuRIR

Python library for Room Impulse Response (RIR) simulation with GPU acceleration
GNU Affero General Public License v3.0
488 stars 94 forks source link

A problem occured when do simulation #29

Closed lsg1213 closed 2 years ago

lsg1213 commented 2 years ago

https://github.com/DavidDiazGuerra/gpuRIR/blob/master/examples/time_vs_T60.py#L39 When execute above line, an error message is occured GPUassert: an illegal memory access was encountered /tmp/pip-req-build-y9g99fpi/src/gpuRIR_cuda.cu 822

My environment is CUDA=11.4, UBUNTU 18.04, python=3.7.12, gpu is titan rtx

please help me

lsg1213 commented 2 years ago

However, a weird thing is that https://github.com/DavidDiazGuerra/gpuRIR/blob/master/examples/example.py is executing well.

DavidDiazGuerra commented 2 years ago

Are you executing the time_vs_T60.py script as it is without any modifications? Do you know in which iteration (the values of i and j) the program crashed?

lsg1213 commented 2 years ago

Yes I didn't modificate it except CUDA_VISIBLE_DEVICES=gpunumber and i=0, j=0 crashed

DavidDiazGuerra commented 2 years ago

Sorry for the late reply. It seems like the last commit introduced some kind of bug. I'll try to fix it in the following days, as a workaround till then, you can install the previous version of the library with:

pip uninstall gpuRIR
pip install https://github.com/DavidDiazGuerra/gpuRIR/zipball/57463ffe3dd16e2bb10b9624b044e8216a956018
DavidDiazGuerra commented 2 years ago

93b52e5367b4578e977388a83e144d565e65c1fd had a bug concerning the shared memory reservation for the orientation of the speakers, but it should be fixed now with the last commit. However, this increases the shared memory consumption and therefore reduces the maximum number of RIRs that can be computed in parallel. I'll try to modify the kernel so the orientation of the source/receivers is not copied to shared memory in the case of omnidirectional directivity patterns.

I'm closing the issue now since it should be already fixed, but let me know if you still get that error.