Open
Description
Backend: oneAPIBackend()
[ Info: ... downloaded 17.0 B (100% complete, 213496.504 days)
[ Info: ... downloaded 17.0 B (100% complete, 213496.504 days)
[ Info: ... downloaded 17.0 B (100% complete, 213496.504 days)
[ Info: ... downloaded 137.1 MiB (30% complete, 213496.504 days)
[ Info: ... downloaded 182.8 MiB (40% complete, 213496.504 days)
[ Info: ... downloaded 228.5 MiB (50% complete, 213496.504 days)
[ Info: ... downloaded 274.2 MiB (60% complete, 213496.504 days)
[ Info: ... downloaded 319.9 MiB (70% complete, 213496.504 days)
[ Info: ... downloaded 365.5 MiB (80% complete, 213496.504 days)
[ Info: ... downloaded 411.2 MiB (90% complete, 213496.504 days)
[ Info: ... downloaded 456.1 MiB (100% complete, 213496.504 days)
┌ Error: Module compilation failed:
│
│ error: Total size of kernel arguments exceeds limit! Total arguments size: 2056, limit: 2048
Thanks to the help from ALCF (@kris-rowe), it could be resolved with:
export IGC_OverrideOCLMaxParamSize=4096
Here was the issue with CUDA.jl https://github.com/JuliaGPU/CUDA.jl/pull/2180/files#diff-ecbfaf5b99ab10dcafff5717c7cc5f856768e4313446fa3fb58b839a25b17cfc .
I guess we should find a way to poll for the hardware/compiler capabilities.