AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
76,348 Commits 3 Branches 0 Tags
f227312867f9417cce3800a959de6bb587445a49
Commit Graph

15 Commits

Author SHA1 Message Date
Wenkai Du 75a69211a0 Add all_reduce_bias_perf to support All Reduce with Bias (#130) 
Use dynamic symbol loading of ncclAllReduceWithBias

Co-authored-by: mberenjk <146776561+mberenjk@users.noreply.github.com>

[ROCm/rccl-tests commit: db6ea5a594]
 2025-10-13 16:09:10 -05:00
Mustafa Abduljabbar cb4b286d2b Enable viewing algo/proto/channels used in rccl-tests output (#151) 
* Enable algo/proto/channel viewing 

* Use dynamic symbol loading to avoid build/runtime issues with non-compatible RCCL versions

* Reduce code duplication

[ROCm/rccl-tests commit: 0c94d4d2b3]
 2025-09-26 18:09:01 -04:00
nileshnegi 8d887aad0d Merge remote-tracking branch 'nccl-tests/master' into develop 
[ROCm/rccl-tests commit: 5625599dda]
 2025-04-21 19:46:10 -05:00
John Bachan 69b9a05e71 Fixes to all tests that divide buffers by nranks so that they trim buffer sizes to be multiples of 16 bytes. 
This ensures non-pow2 ranks have buffer addresses aligned suitably for performance.


[ROCm/rccl-tests commit: 29f4114f02]
 2024-12-18 11:20:28 -08:00
Bertan Dogancay 882a96f5cb Add hipify steps prior to build (#62) 
* Add hipify steps prior to build

[ROCm/rccl-tests commit: 88cf7dbf45]
 2024-03-05 09:47:18 -07:00
Wenkai Du b49f6da1ec Merge remote-tracking branch 'nccl-tests/master' into HEAD 
[ROCm/rccl-tests commit: 621dde544d]
 2024-03-01 18:34:44 +00:00
Edgar Gabriel 08f9435e5a Merge remote-tracking branch 'nccl-tests/master' into topic/v2.13.4-sync 
[ROCm/rccl-tests commit: 3ae371cce7]
 2022-10-14 16:02:54 -05:00
Sylvain Jeaugey fdaa88710b Update NCCL tests 
[ROCm/rccl-tests commit: d313d20a26]
 2022-09-23 01:13:29 -07:00
John Bachan b5d746b58e Resync with NCCL 2.13 
* Added "verifiable", a suite of kernels for generating and verifying reduction
  input and output arrays in a bit-precise way.
* Data corruption errors now reported in number of wrong elements instead of max
  deviation.
* Use ncclGetLastError.
* Don't run hypercube on non-powers of 2 ranks.
* Fix to hypercube data verification.
* Use "thread local" as the defaut CUDA capture mode.
* Replaced pthread_yield -> sched_yield()
* Bugfix to the cpu-side barrier/allreduce implementations.


[ROCm/rccl-tests commit: 51af5572bf]
 2022-08-22 17:51:06 -07:00
Edgar dad6d819d0 implementation of multi-rank support in rccl-tests. 
[ROCm/rccl-tests commit: 0500f2f132]
 2022-06-10 14:54:10 -04:00
Wenkai Du 669a778efb Merge remote-tracking branch 'nccl/master' into develop 
[ROCm/rccl-tests commit: 213abee002]
 2021-09-20 14:01:22 -07:00
David Addison 56ff821802 Resync with NCCL 2.11 
New operator: mulsum
New test: gather


[ROCm/rccl-tests commit: f773748b46]
 2021-09-17 09:02:45 -07:00
Wenkai Du 99d8ddc145 Change scatter and gather bandwidth calculation to match alltoall (#7) 
[ROCm/rccl-tests commit: 346cb16442]
 2020-07-06 17:12:50 -07:00
Wenkai Du 0f10b6ceeb Add option to use alltoall, gather and scatter API 
These APIs launche RCCL kernel implementation by default. If environmental
variable RCCL_ALLTOALL_KERNEL_DISABLE=1, then the APIs use wrapper around
ncclSend and ncclRecv.


[ROCm/rccl-tests commit: 2813c96826]
 2020-05-11 15:21:19 -07:00
Wenkai Du af08abefb0 Add gather and scatter test 
[ROCm/rccl-tests commit: 6932a583e7]
 2020-04-10 14:45:10 -07:00