pygsti.baseobjs.profiler
Defines the Profiler class and supporting functionality
Module Contents
Classes
Profiler objects are used for tracking both time and memory usage. |
|
A dummy profiler that doesn't do anything. |
- class pygsti.baseobjs.profiler.Profiler(comm=None, default_print_memcheck=False)
Bases:
objectProfiler objects are used for tracking both time and memory usage.
Parameters
- commmpi4py.MPI.Comm optional
The active MPI communicator.
- default_print_memcheckbool, optional
Whether to print memory checks.
Construct a new Profiler instance.
Parameters
- commmpi4py.MPI.Comm, optional
MPI communicator so only profile and print messages on root proc.
- comm
- timers
- counters
- mem_checkpoints
- print_memchecks
- add_time(name, start_time, prefix=0)
Adds an elapsed time to a named “timer”-type accumulator.
Parameters
- namestring
The name of the timer to add elapsed time into (if the name doesn’t exist, one is created and initialized to the elapsed time).
- start_timefloat
The starting time used to compute the elapsed, i.e. the value time.time()-start_time, which is added to the named timer.
- prefixint, optional
Prefix to the timer name the current stack depth and this number of function names, starting with the current function and moving the call stack. When zero, no prefix is added. For example, with prefix == 1, “Total” might map to “ 3: myFunc: Total”.
Returns
None
- add_count(name, inc=1, prefix=0)
Adds a given value to a named “counter”-type accumulator.
Parameters
- namestring
The name of the counter to add val into (if the name doesn’t exist, one is created and initialized to val).
- incint, optional
The increment (the value to add to the counter).
- prefixint, optional
Prefix to the timer name the current stack depth and this number of function names, starting with the current function and moving the call stack. When zero, no prefix is added. For example, with prefix == 1, “Total” might map to “ 3: myFunc: Total”.
Returns
None
- memory_check(name, printme=None, prefix=0)
Record the memory usage at this point and tag with a name.
Parameters
- namestring
The name of the memory checkpoint. (Later, memory information can be organized by checkpoint name.)
- printmebool, optional
Whether or not to print the memory usage during this function call (if None, the default, then the value of default_print_memcheck specified during Profiler construction is used).
- prefixint, optional
Prefix to the timer name the current stack depth and this number of function names, starting with the current function and moving the call stack. When zero, no prefix is added. For example, with prefix == 1, “Total” might map to “ 3: myFunc: Total”.
Returns
None
- print_memory(name, show_minmax=False)
Prints the current memory usage (but doesn’t store it).
Useful for debugging, this function prints the current memory usage - optionally giving the mininum, maximum, and average across all the processors.
Parameters
- namestring
A label to print before the memory usage number(s).
- show_minmaxbool, optional
If True and there are multiple processors, print the min, average, and max memory usage from among the processors. Note that this will invoke MPI collective communication and so this print_memory call must be executed by all the processors. If False and there are multiple processors, only the rank 0 processor prints output.
Returns
None
- print_message(msg, all_ranks=False)
Prints a message to stdout, possibly from all ranks.
A utility function used in debugging, this function offers a convenient way to print a message on only the root processor or on all processors.
Parameters
- msgstring
The message to print.
- all_ranksbool, optional
If True, all processors will print msg, preceded by their rank label (e.g. “Rank4: “). If False, only the rank 0 processor will print the message.
Returns
None
- class pygsti.baseobjs.profiler.DummyProfiler
Bases:
objectA dummy profiler that doesn’t do anything.
A class which implements the same interface as Profiler but which doesn’t actually do any profiling (consists of stub functions).
Construct a new DummyProfiler instance.
- add_time(name, start_time, prefix=0)
Stub function that does nothing
Parameters
- namestring
The name of the timer to add elapsed time into (if the name doesn’t exist, one is created and initialized to the elapsed time).
- start_timefloat
The starting time used to compute the elapsed, i.e. the value time.time()-start_time, which is added to the named timer.
- prefixint, optional
Prefix to the timer name the current stack depth and this number of function names, starting with the current function and moving the call stack. When zero, no prefix is added. For example, with prefix == 1, “Total” might map to “ 3: myFunc: Total”.
Returns
None
- add_count(name, inc=1, prefix=0)
Stub function that does nothing
Parameters
- namestring
The name of the counter to add val into (if the name doesn’t exist, one is created and initialized to val).
- incint, optional
The increment (the value to add to the counter).
- prefixint, optional
Prefix to the timer name the current stack depth and this number of function names, starting with the current function and moving the call stack. When zero, no prefix is added. For example, with prefix == 1, “Total” might map to “ 3: myFunc: Total”.
Returns
None
- memory_check(name, printme=None, prefix=0)
Stub function that does nothing
Parameters
- namestring
The name of the memory checkpoint. (Later, memory information can be organized by checkpoint name.)
- printmebool, optional
Whether or not to print the memory usage during this function call (if None, the default, then the value of default_print_memcheck specified during Profiler construction is used).
- prefixint, optional
Prefix to the timer name the current stack depth and this number of function names, starting with the current function and moving the call stack. When zero, no prefix is added. For example, with prefix == 1, “Total” might map to “ 3: myFunc: Total”.
Returns
None