From 45a0534517e58aa7a3e6956593cf21409e201c67 Mon Sep 17 00:00:00 2001 From: David Addison Date: Tue, 7 Jul 2026 10:43:21 -0700 Subject: [PATCH 1/2] Fix blocking event timing synchronization Signed-off-by: David Addison --- src/common.cu | 3 +++ 1 file changed, 3 insertions(+) diff --git a/src/common.cu b/src/common.cu index 52b2842..39f13c5 100644 --- a/src/common.cu +++ b/src/common.cu @@ -772,6 +772,9 @@ testResult_t BenchTime(struct threadArgs* args, ncclDataType_t type, ncclRedOp_t if (cudaGraphLaunches >= 1) deltaSec = deltaSec/cudaGraphLaunches; if (record) { + // completeColl skips the stream sync in blocking mode with a single aggregated iteration. + if (blocking_coll && agg_iters <= 1) + TESTCHECK(testStreamSynchronize(args->nGpus, args->streams, args->comms)); TESTCHECK(getElapsedTimes(args, iters, agg_iters)); TESTCHECK(destroyEvents(args, iters)); } From 5bcd45d2ee8365e37895e118c9d11b0ebc15aa69 Mon Sep 17 00:00:00 2001 From: David Addison Date: Tue, 7 Jul 2026 10:48:17 -0700 Subject: [PATCH 2/2] NCCL_TESTS_VERSION 2.19.5 Signed-off-by: David Addison --- src/common.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/common.h b/src/common.h index 8566b48..4edeb68 100644 --- a/src/common.h +++ b/src/common.h @@ -6,7 +6,7 @@ #ifndef __COMMON_H__ #define __COMMON_H__ -#define NCCL_TESTS_VERSION "2.19.4" +#define NCCL_TESTS_VERSION "2.19.5" #include "nccl.h" #if NCCL_VERSION_CODE >= NCCL_VERSION(2,28,0)