ILGPU icon indicating copy to clipboard operation
ILGPU copied to clipboard

Sample of "AlgorithmsRadixSort" failed on OpenCL device

Open GeoBIM2020 opened this issue 1 year ago • 4 comments

The OpenCL device is 11th Gen Intel(R) Core i9-11980HK.

Following is error message:

Performing operations on 11th Gen Intel(R) Core(TM) i9-11980HK @ 2.60GHz [Type: OpenCL, WarpSize: 0, MaxNumThreadsPerGroup: 8192, MemorySize: 34046320640] Unhandled exception. ILGPU.Runtime.OpenCL.CLException: Exception of type 'ILGPU.Runtime.OpenCL.CLException' was thrown. at ILGPULauncher(Kernel, AcceleratorStream, KernelConfig, ArrayView1D2, ArrayView1, SpecializedValue1, Int32, Int32, Int32) at ILGPULauncher(SpecializationCache3, AcceleratorStream, KernelConfig, ArrayView1D2, ArrayView1, SpecializedValue1, Int32, Int32, Int32) at ILGPU.Algorithms.RadixSortExtensions.<>c__DisplayClass23_23.<CreateRadixSort>b__1(AcceleratorStream stream, ArrayView1D2 input, ArrayView1 tempView) in C:\Users\17058\source\ILGPU\Nov\Src\ILGPU.Algorithms\RadixSortExtensions.cs:line 1261 at AlgorithmsRadixSort.Program.Main() in C:\Users\17058\source\ILGPU\Nov\Samples\AlgorithmsRadixSort\Program.cs:line 77

GeoBIM2020 avatar Oct 30 '23 18:10 GeoBIM2020

Hi @GeoBIM2020, welcome to the ILGPU community. Thank you very much for raising this issue. Unfortunately, I'm not able to reproduce your issue at the moment. It may be an issue with an actual driver (version) being used. @MoFtZ maybe you have additional insights into this issue at the moment?

m4rs-mt avatar Nov 02 '23 11:11 m4rs-mt

@m4rs-mt I currently have access to a 10th Gen Intel Laptop CPU, and a 7th Gen Intel Desktop CPU, and have not been able to reproduce the problem either.

However, it is interesting that @GeoBIM2020 's CPU has MaxNumThreadsPerGroup: 8192. It is higher than I would expect, and I'm wondering if there is a driver issue, or if ILGPU is doing something wrong with OpenCL?

MoFtZ avatar Nov 14 '23 06:11 MoFtZ

@MoFtZ I fully agree regarding the MaxNumThreadsPerGroup. In my opinion, this value should ideally fall within the range of [4, 128] (usually).

m4rs-mt avatar Nov 14 '23 12:11 m4rs-mt

@m4rs-mt @MoFtZ There is no updated driver available for my CPU. Please let me know if you need me to do something.

GeoBIM2020 avatar Nov 14 '23 18:11 GeoBIM2020