Issues with native execution on Intel

[maleadt]

Intel is one of the few platforms that also supports IL execution, however something seems wrong:

using OpenCL

cl.platform!("intel")

function memset(a, val)
    @inbounds a[1] = val
    return
end

function main()
    a = CLArray{Int}(undef, 1)
    @opencl memset(a, 42)
    Array(a)
end

julia> main()
1-element Vector{Int64}:
 42

julia> main()
1-element Vector{Int64}:
 0

julia> main()
1-element Vector{Int64}:
 0

Not sure yet what the issue is here.

[maleadt]

Looks like this works when not using uninitialized memory:

function main()
    # broken
    a = CLArray{Int}(undef, 1)
    # working
    a = CLArray{Int}([1])
    
    @opencl memset(a, 42)
    println("Result: ", Array(a)[])
end

julia> uninitialized()
libopencl.clSVMAlloc(OpenCL.cl.Context(id=0x0000000032fd8f20), 1, 8, 0) = Ptr{Nothing} @0x0000791e9b200000
libopencl.clSetKernelArg(OpenCL.cl.Kernel(id=0x00000000349537e0), 0x00000000, 32, Ref{OpenCL.CLDeviceArray{Int64, 1, 1}}) = 0
 4: 1-element device array at Core.LLVMPtr{Int64, 1}(0x0000791e9b200000)
libopencl.clSetKernelArg(OpenCL.cl.Kernel(id=0x00000000349537e0), 0x00000001, 8, Ref{Int64}) = 0
 4: 42
libopencl.clGetDeviceInfo(OpenCL.cl.Device(id=0x0000000032cf3020), 0x1003, 4, Ref{UInt32}, 0x0000000000000000) = 0
 4: 3
libopencl.clEnqueueNDRangeKernel(OpenCL.cl.CmdQueue(id=0x0000000031b9d5d0), OpenCL.cl.Kernel(id=0x00000000349537e0), 0x00000001, 0x0000000000000000, 1-element Vector{UInt64}, 0x0000000000000000, 0x00000000, 0x0000000000000000, Ref{Ptr{OpenCL.cl._cl_event}}) = 0
 9: Ptr{OpenCL.cl._cl_event} @0x0000000037d36c40
libopencl.clEnqueueSVMMemcpy(OpenCL.cl.CmdQueue(id=0x0000000031b9d5d0), true, 0x0000791f7799da00, 0x0000791e9b200000, 8, 0, 0x0000000000000000, Ref{Ptr{OpenCL.cl._cl_event}}) = 0
 8: Ptr{OpenCL.cl._cl_event} @0x0000000037c7bf60
Result: 0

julia> initialized()
libopencl.clSVMAlloc(OpenCL.cl.Context(id=0x0000000032fd8f20), 1, 8, 0) = Ptr{Nothing} @0x0000791e9ae00000
libopencl.clEnqueueSVMMemcpy(OpenCL.cl.CmdQueue(id=0x0000000031b9d5d0), true, 0x0000791e9ae00000, 0x0000791f77f0ee40, 8, 0, 0x0000000000000000, Ref{Ptr{OpenCL.cl._cl_event}}) = 0
 8: Ptr{OpenCL.cl._cl_event} @0x0000000033a15c00
libopencl.clSetKernelArg(OpenCL.cl.Kernel(id=0x00000000349537e0), 0x00000000, 32, Ref{OpenCL.CLDeviceArray{Int64, 1, 1}}) = 0
 4: 1-element device array at Core.LLVMPtr{Int64, 1}(0x0000791e9ae00000)
libopencl.clSetKernelArg(OpenCL.cl.Kernel(id=0x00000000349537e0), 0x00000001, 8, Ref{Int64}) = 0
 4: 42
libopencl.clGetDeviceInfo(OpenCL.cl.Device(id=0x0000000032cf3020), 0x1003, 4, Ref{UInt32}, 0x0000000000000000) = 0
 4: 3
libopencl.clEnqueueNDRangeKernel(OpenCL.cl.CmdQueue(id=0x0000000031b9d5d0), OpenCL.cl.Kernel(id=0x00000000349537e0), 0x00000001, 0x0000000000000000, 1-element Vector{UInt64}, 0x0000000000000000, 0x00000000, 0x0000000000000000, Ref{Ptr{OpenCL.cl._cl_event}}) = 0
 9: Ptr{OpenCL.cl._cl_event} @0x0000000037acbbd0
libopencl.clEnqueueSVMMemcpy(OpenCL.cl.CmdQueue(id=0x0000000031b9d5d0), true, 0x0000791f77f14f00, 0x0000791e9ae00000, 8, 0, 0x0000000000000000, Ref{Ptr{OpenCL.cl._cl_event}}) = 0
 8: Ptr{OpenCL.cl._cl_event} @0x0000000037c8df30
Result: 42