taucmdr
taucmdr copied to clipboard
tau init --track-memory-footprint fails
[TAU] XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
[TAU]
[TAU] CRITICAL
[TAU]
[TAU] track_memory_footprint = True in measurement is incompatible with baseline = True in measurement
[TAU]
[TAU] Hint: Try `tau --help`
[TAU]
[TAU] TAU cannot proceed with the given inputs.
[TAU] Please check the configuration for errors or contact <[email protected]> for assistance.
[TAU]
[TAU] XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
tau init
also fails when specifying the --heap-usage
or --memory-alloc
options.
@nchaimov @khsa1 Is tau init
supposed to be compatible with all measurement options? This was never clear to me.
I assumed that was the case but maybe not. I think that passing options to tau init
should at least not cause initialization of the project to fail.
Yes I agree. We should figure out a sane approach. On Thu, Sep 6, 2018 at 8:55 AM khsa1 [email protected] wrote:
I assumed that was the case but maybe not. I think that passing options to tau init should at least not cause initialization of the project to fail.
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/ParaToolsInc/taucmdr/issues/308#issuecomment-419082569, or mute the thread https://github.com/notifications/unsubscribe-auth/AAREPIMpH_Had2AP7SavGxBvxmXLjrHhks5uYRtFgaJpZM4WbIR6 .
I'm not completely sure what was intended, but given that tau init --help
lists --track-memory-footprint
as an option, it ought to be supported (or, if it's not intended to be supported, not listed in the help).
I think there just needs to be some handling to not set things in the baseline measurement that can't be used with baseline. Other measurement properties seem to have this. For example, tau init --openmp=ompt
sets that property in all the default measurements except baseline.
| Name | Profile | Trace | Sample | Source Inst. | Compiler Inst. | OpenMP | CUDA | I/O | MPI | SHMEM |
+===============+=========+=======+========+==============+================+========+======+=====+=====+=======+
| baseline | tau | none | No | never | never | ignore | No | No | No | No |
+---------------+---------+-------+--------+--------------+----------------+--------+------+-----+-----+-------+
| sample | tau | none | Yes | never | never | ompt | No | No | No | No |
+---------------+---------+-------+--------+--------------+----------------+--------+------+-----+-----+-------+
| profile | tau | none | No | automatic | fallback | ompt | No | No | No | No |
+---------------+---------+-------+--------+--------------+----------------+--------+------+-----+-----+-------+
| source-inst | tau | none | No | automatic | never | ompt | No | No | No | No |
+---------------+---------+-------+--------+--------------+----------------+--------+------+-----+-----+-------+
| compiler-inst | tau | none | No | never | always | ompt | No | No | No | No |
+---------------+---------+-------+--------+--------------+----------------+--------+------+-----+-----+-------+
| trace | none | otf2 | No | automatic | fallback | ompt | No | No | No | No |
+---------------+---------+-------+--------+--------------+----------------+--------+------+-----+-----+-------+```
The best way I found to do this was to manually disable it when the baseline
measurement is created in the _populate_project
function of cli/commands/initialize.py
. I also tried to disable any other conflicting options in 57af4450fbf92e1b3a671bcbea9f79294aedb828.
Thanks Sam! I think we'll still get warning though. I'll take a pass at silencing those