Wenkai Du 07453ebfaf Improve RCCL kernel coll trace (#2061) ...

[ROCm/rccl commit: 1d22c87167]

2026-01-08 16:07:18 -08:00

..

network/unpack

Merge remote-tracking branch 'nccl/master' into develop

2025-04-23 20:47:43 -07:00

symmetric

Add copyright to src/device/symmetric/all_reduce.cuh (#2080)

2025-11-27 14:29:21 -07:00

all_gather.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

all_reduce.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

alltoall_pivot.h

Added useAcc as a template parameter to address the performance regression (#1856)

2025-08-14 15:58:54 -05:00

broadcast.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

common_kernel.h

[Device] Add dynamic fetch/reduce pipelining for reduction collectives - Simple protocol (#1861)

2025-08-26 15:03:54 -04:00

common.cu

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

common.h

Improve RCCL kernel coll trace (#2061)

2026-01-08 16:07:18 -08:00

generate.py

Disable Bfloatf16 pipelining for reduction collectives for gfx950 (#2047)

2025-11-13 14:55:09 -06:00

Makefile

NCCL 2.27.5-1

2025-06-18 10:34:47 -07:00

msccl_kernel_impl.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

onerank.cu

Added useAcc as a template parameter to address the performance regression (#1856)

2025-08-14 15:58:54 -05:00

op128.h

Fix compilation when enabling indirect function calls (#1994)

2025-11-11 09:36:48 -05:00

primitives.h

[Device] Adjust threadblock size for gfx950 to increase LL64/Simple performance for AR, RS and AG (#1978)

2025-10-29 23:24:32 -05:00

prims_ll128.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

prims_ll.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

prims_simple.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

rccl_metadata.h

MSCCL: Unland PR1788 + Fix for MSCCL Data Corruption (#1960)

2025-10-15 10:32:25 -07:00

rccl_ptr.h

Remove nontemporality from stores, put in casts to global address space (#1982)

2025-10-28 10:34:48 -07:00

reduce_kernel.h

Using hip_bf16.h instead of hip_bfloat16.h for the __bf16 intrinsic (#2037)

2025-11-13 15:56:18 -05:00

reduce_scatter.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

reduce.h

[Device] WarpSpeed enablement and single node CU and perf opt for MI350 (#2073)

2025-12-11 19:04:35 -05:00

sendrecv.h

Use batched P2P to enhance alltoall small message performance (#1902)

2025-09-22 16:25:10 -04:00