Make sure to use 16 compute nodes in colocale performance testing

[e-kayrakli]

This PR passes --numLocales=32 to start_test to ensure that we get 16 compute nodes in colocale performance testing.

Note that colocale testing uses CHPL_RT_LOCALES_PER_NODE=2 to ask for colocales, which, with -nl16 results in using 8 compute nodes.

[jhh67]

How does this interact with tests that have their own .numlocales file?

[e-kayrakli]

How does this interact with tests that have their own .numlocales file?

Oh, that's a good question. I think .numlocales is for correctness and .ml.numlocales are for multilocale performance. There is also .cc.numlocales for comm counts.

Based on find test -name "*ml-numlocales" | xargs cat, I get a lot of 16s, some 8s and some 2s. I am not sure how to handle those. Some options:

• let them be: maybe there is something sensitive to numLocales in those tests
• change ml-numlocales to ml.numnodes and how it is interpreted by the nightly testing rig.
• make either of these decisions based on tests, but there are quite a few tests.

[e-kayrakli]

Another option: update all of these to have 16, 16x2 for both standard and colocale performance testing.