level-zero icon indicating copy to clipboard operation
level-zero copied to clipboard

Potential problem with zeKernelSuggestGroupSize

Open jczaja opened this issue 2 years ago • 1 comments

My use case is that I have a buffer to be processed of a size 2^40 and I want to use zeKernelSuggestGroupSize to get optimal setting of work group to be used. Problem is that spec claims that globalSizeX is of u32 type and this is not enough to hold 2^40 value. My spirv is processng linearly whole buffer (elementwise operation). So shouldn't zeKernelSuggestGroupSize's args globalSizeX, globalSizeY,globalSizeZ be of type of u64 or size_t ?

jczaja avatar Oct 25 '23 12:10 jczaja

+1 OpenCL function clGetKernelSuggestedLocalWorkSizeKHR's input global_work_size and output suggested_local_work_size have type size_t.

wenju-he avatar Dec 27 '23 23:12 wenju-he