Infinite partial_mapreduce_device recursion

[michel2323]

When running

julia --project -e 'using oneAPI; sum(oneAPI.zeros(30000)) ; sum(oneAPI.zeros(2182))'

the reduction for the 30,000 case runs fine. However, for the 2182 it crashes with an InvalidIRError.

I then added some debug output into

oneAPI.jl/src/mapreduce.jl

Lines 164 to 178 in 7973a89

	if reduce_groups == 1
	# we can cover the dimensions to reduce using a single group
	@oneapi items groups partial_mapreduce_device(
	f, op, init, Val(items), Rreduce, Rother, R′, A)
	else
	# we need multiple steps to cover all values to reduce
	partial = similar(R, (size(R)..., reduce_groups))
	if init === nothing
	# without an explicit initializer we need to copy from the output container
	partial .= R
	end
	@oneapi items groups partial_mapreduce_device(
	f, op, init, Val(items), Rreduce, Rother, partial, A)
	GPUArrays.mapreducedim!(identity, op, R′, partial; init=init)

.

It seems the recursion for the 2182 is infinite. I would suspect it to be shallower than for the 30000 case. I attached a log with some debug output. Maybe there is something obvious to you @maleadt .

out.log

[maleadt]

Looks like Level Zero's zeKernelSuggestGroupSize doesn't like prime-sized global sizes:

julia> k = @oneapi launch=false identity(nothing)

julia> oneL0.suggest_groupsize(k.fun, 521)
oneAPI.oneL0.ZeDim3(1, 1, 1)

julia> oneL0.suggest_groupsize(k.fun, 7877)
oneAPI.oneL0.ZeDim3(1, 1, 1)

julia> oneL0.suggest_groupsize(k.fun, 7919)
oneAPI.oneL0.ZeDim3(1, 1, 1)

These are really bad launch configurations... Maybe I'm misinterpreting the API?

EDIT: it seems to suggest really bad configurations for non-prime inputs too:

julia> oneL0.suggest_groupsize(k.fun, 8000)
oneAPI.oneL0.ZeDim3(64, 1, 1)

julia> oneL0.suggest_groupsize(k.fun, 512)
oneAPI.oneL0.ZeDim3(512, 1, 1)