Teuchos: TimeMonitor::report to YAML prints on each MPI process

[ppraegla]

Bug Report

@bartlettroscoe

Description

Using TimeMonitor:.report with output to YAML, writes the output on each MPI process. Using the Report format: Table writes correctly only on rank 0. The YAML output should also be only written once, i.e., on rank 0.

There seems to be an if (myRank == 0) missing when writing to the output stream. It may also be that the YAML and table report are different #1917. I have not looked into that

Steps to Reproduce

1. SHA1: d8f7113
2. Write TimeMonitor report

Teuchos::ParameterList params_report;
params_report.set("Report format", "YAML");
Teuchos::TimeMonitor::report(std::cout, Teuchos::rcpFromRef(params_report));

3. Output

Output report format YAML

# Teuchos::TimeMonitor report
---
Output mode: spacious
Number of processes: 2
Time unit: s
Statistics collected: 
  - MinOverProcs
  - MeanOverProcs
  - MaxOverProcs
  - MeanOverCallCounts
Timer names: 
  - fun
Total times: 
  fun: 
    MinOverProcs: 1.8344e-05
    MeanOverProcs: 1.91755e-05
    MaxOverProcs: 2.0007e-05
    MeanOverCallCounts: 9.58775e-06
Call counts:
  fun: 
    MinOverProcs: 2
    MeanOverProcs: 2
    MaxOverProcs: 2
    MeanOverCallCounts: 2
# Teuchos::TimeMonitor report
---
Output mode: spacious
Number of processes: 2
Time unit: s
Statistics collected: 
  - MinOverProcs
  - MeanOverProcs
  - MaxOverProcs
  - MeanOverCallCounts
Timer names: 
  - fun
Total times: 
  fun: 
    MinOverProcs: 1.8344e-05
    MeanOverProcs: 1.91755e-05
    MaxOverProcs: 2.0007e-05
    MeanOverCallCounts: 9.58775e-06
Call counts:
  fun: 
    MinOverProcs: 2
    MeanOverProcs: 2
    MaxOverProcs: 2
    MeanOverCallCounts: 2

Output report format Table

=======================================================================================

                         TimeMonitor results over 2 processors

Timer Name    MinOverProcs     MeanOverProcs    MaxOverProcs     MeanOverCallCounts    
---------------------------------------------------------------------------------------
fun           2.808e-05 (2)    2.901e-05 (2)    2.994e-05 (2)    1.45e-05 (2)          
=======================================================================================

[bartlettroscoe]

@ppraegla, you might just consider using the Teuchos::FancyOStream class as that can ensure that output only gets printed from proc 0. See:

https://trilinos.github.io/docs/teuchos/classTeuchos_1_1basic__FancyOStream.html#details

https://trilinos.github.io/docs/teuchos/classTeuchos_1_1basic__FancyOStream__buf.html#a25ec06139d190ace75b189677dd4efaa

The magic is shown in:

https://trilinos.github.io/docs/teuchos/Teuchos__FancyOStream_8hpp_source.html#l00995

You can wrap any std::ostream object with Teuchos::FancyOStream and Teuchos::FancyOStream implements the std::ostream interface.

When doing parallel programming in C++, it is really nice to not worry about what proc you are print on and just use Teuchos::FancyOStream to handle what to keep and what to throw away.

[ppraegla]

Thanks for the quick reply and the hint to the Teuchos::FancyOStream.

This

Teuchos::RCP<Teuchos::FancyOStream> pfout = getFancyOStream(Teuchos::rcpFromRef(std::cout));
pfout->setOutputToRootOnly(0);
Teuchos::TimeMonitor::report(*pfout, Teuchos::rcpFromRef(params_report));

worked.

I was just confused because the TimeMonitor::report docs explicitly said to use the `std::ostream only on rank 0

out [out] Output stream to which to write. This will only be used on the process with rank 0 in the communicator.

[bartlettroscoe]

I was just confused because the TimeMonitor::report docs explicitly said to use the `std::ostream only on rank 0

The documentation needs to be improved.