implement NVTXT

[vchuravy]

So the idea is that we have a common output format for the CPU backend as well as the GPU backend.
This implements NVTXT so that you can load these annotations with nsys

export KERNELABSTRACTIONS_TIMELINE=true

julia> import KernelAbstractions.Extras.Timeline.NVTXT
julia> NVTXT.push_range("Costly computation")
julia> NVTXT.pop_range()

These leaves a file in the cwd:

cat ka-31089.nvtxt
SetFileDisplayName, KernelAbstractions
@RangeStartEnd, Start, End, ThreadId, Message
ProcessId = 31089
CategoryId = 1
Color = Blue
TimeBase = Manual
@RangePush, Time, ThreadId, Message
ProcessId = 31089
CategoryId = 1
Color = Blue
TimeBase = Manual
@RangePop, Time, ThreadId
ProcessId = 31089
TimeBase = Manual
@Marker, Time, ThreadId, Message
ProcessId = 31089
CategoryId = 1
Color = Blue
TimeBase = Manual
RangePush, 515251367655304, 1, "Costly computation"
RangePop, 515256007484357, 1

Convert to qdrep

LD_LIBRARY_PATH=/opt/nsight-systems-2020.1.1/host-linux-x64/ /opt/nsight-systems-2020.1.1/host-linux-x64/ImportNvtxt --cmd create --nvtxt ka-31089.nvtxt -o report.qdrep

LD_LIBRARY_PATH=/opt/nsight-systems-2020.1.1/host-linux-x64/ /opt/nsight-systems-2020.1.1/host-linux-x64/ImportNvtxt --cmd info -i report.qdrep 
Analysis start (ns)     515251367000000
Analysis end (ns)       515256008000000

[vchuravy]

@lcw is that roughly what you were thinking? I am questioning whether CPU code should have ranges auto-injected or if I should leave that up to the user.