From 05a36771df3e0d74869902fedf0cdff22a936d8b Mon Sep 17 00:00:00 2001 From: Douglas Jacobsen Date: Wed, 2 Oct 2024 10:52:31 -0600 Subject: [PATCH] Add NCCL_TESTS_SPLIT_MASK to nccl-tests application --- .../applications/nccl-tests/application.py | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/var/ramble/repos/builtin/applications/nccl-tests/application.py b/var/ramble/repos/builtin/applications/nccl-tests/application.py index fe863629d..8a59c3cd8 100644 --- a/var/ramble/repos/builtin/applications/nccl-tests/application.py +++ b/var/ramble/repos/builtin/applications/nccl-tests/application.py @@ -124,6 +124,21 @@ class NcclTests(ExecutableApplication): workloads=all_workloads, ) + workload_variable( + "nccl_tests_split_mask", + default="", + description='How NCCL communicators should be split, if at all. "0x7" for rail-aligned, "0x0" for world-level.', + workloads=all_workloads, + expandable=False, + ) + + environment_variable( + "NCCL_TESTS_SPLIT_MASK", + "{nccl_tests_split_mask}", + 'How NCCL communicators should be split, if at all. "0x7" for rail-aligned, "0x0" for world-level.', + workloads=all_workloads, + ) + # (output_name, units, group_name, regex) regex_parts = [ ("Size", "B", "size", "[0-9]+"),