microsoft/mscclpp v0.6.0

MSCCL++ v0.6.0

Repository: microsoft/mscclpp

Tag: v0.6.0

Published: 2024-12-23T19:01:13Z

Prerelease: no

Release notes:

Highlight

• Improved NCCL API integration in MSCCL++ for better performance and usability
• Enhanced execution plan-based executor in MSCCL++
• Fixed several bugs to improve stability and reliability

What's Changed

• Add support for different vector sizes in multimem instructions by @roshandathathri in https://github.com/microsoft/mscclpp/pull/332
• NCCL API Executor Integration by @caiomcbr in https://github.com/microsoft/mscclpp/pull/331
• Fix missing import in executor test by @yzygitzh in https://github.com/microsoft/mscclpp/pull/334
• bfloat16 support by @chhwang in https://github.com/microsoft/mscclpp/pull/336
• Dynamically load libibverbs by @caiomcbr in https://github.com/microsoft/mscclpp/pull/337
• Auto-tune vector sizes for NVLS allreduce6 by @roshandathathri in https://github.com/microsoft/mscclpp/pull/338
• Make ibverbs optional at compile time by @chhwang in https://github.com/microsoft/mscclpp/pull/340
• ProxyChannel Support in Executor by @caiomcbr in https://github.com/microsoft/mscclpp/pull/342
• Support executors to send packets over ProxyChannel by @caiomcbr in https://github.com/microsoft/mscclpp/pull/344
• Fix for ROCm 6.0 by @chhwang in https://github.com/microsoft/mscclpp/pull/347
• Fix bug for construct sempaphore by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/341
• Add proxy channel related operations by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/351
• Add CI for rocm by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/346
• Tune threads per block for mscclpp executor by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/345
• Fix NPKit exit event offset by @yzygitzh in https://github.com/microsoft/mscclpp/pull/356
• Use IB transport flags only when an IB device exists by @chhwang in https://github.com/microsoft/mscclpp/pull/355
• Update ROCm CI by @chhwang in https://github.com/microsoft/mscclpp/pull/357
• Fixing RegisterMemory Allocation for ProxyChannels by @caiomcbr in https://github.com/microsoft/mscclpp/pull/353
• Fix NCCL API bugs by @chhwang in https://github.com/microsoft/mscclpp/pull/363
• Perf optimization & support clipping by @chhwang in https://github.com/microsoft/mscclpp/pull/364
• Fix copyright messages by @chhwang in https://github.com/microsoft/mscclpp/pull/367
• [Doc] mscclpp docs by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/348
• Executor AllGather In-Place Support by @caiomcbr in https://github.com/microsoft/mscclpp/pull/365
• Fix algo repo name by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/369
• Update docker image for cuda12.4 by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/370
• Fix in-place all-gather input buffer in executor_test by @yzygitzh in https://github.com/microsoft/mscclpp/pull/372
• [docs] fix quickstart link by @jeffra in https://github.com/microsoft/mscclpp/pull/374
• Add kernel-based verification for executor_test by @yzygitzh in https://github.com/microsoft/mscclpp/pull/378
• Lazily create the context stream by @chhwang in https://github.com/microsoft/mscclpp/pull/381
• Fixing Bug Const Offset in Execution Plan by @caiomcbr in https://github.com/microsoft/mscclpp/pull/380
• Fix light load bug by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/379
• Small Adjust in Test Data AllGather at Executor Test by @caiomcbr in https://github.com/microsoft/mscclpp/pull/384
• Fix missing packet parameter for executor by @yzygitzh in https://github.com/microsoft/mscclpp/pull/385
• NVLS support for msccl++ executor by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/375
• Fix typo by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/389
• Improve CMake options by @chhwang in https://github.com/microsoft/mscclpp/pull/376
• Fixing Message Boundary AllReduce Fallback Code by @caiomcbr in https://github.com/microsoft/mscclpp/pull/391
• Fix mscclpp_benchmark by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/392
• Add cross threadblock barrier by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/383
• AllGather Executor Support in NCCL Interface by @caiomcbr in https://github.com/microsoft/mscclpp/pull/393
• Providing reduce-scatter test support by @caiomcbr in https://github.com/microsoft/mscclpp/pull/390
• Select algo according to json config by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/396
• Add connection events for NPKit by @yzygitzh in https://github.com/microsoft/mscclpp/pull/386
• Revised ProxyChannel interfaces by @chhwang in https://github.com/microsoft/mscclpp/pull/400
• Setup pipeline for mscclpp over nccl by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/401
• Exception Max Number Operation per Tb by @caiomcbr in https://github.com/microsoft/mscclpp/pull/405
• Reduce memory usage for scratch buffer by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/403
• [Cherry-pick] Move pipeline to official org (#406) by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/416
• [Cherry-pick] trigger ci for release branches (#426) by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/427
• [Cherry-pick] Disable CuMemMap check for ROCm (#411) by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/424
• [Cherry-pick] NVLS support for NCCL API (#410) by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/425
• [Cherry-pick] Fix nccl-test failure issue (#421) by @Binyang2014 in https://github.com/microsoft/mscclpp/pull/429

New Contributors

• @jeffra made their first contribution in https://github.com/microsoft/mscclpp/pull/374

Full Changelog: https://github.com/microsoft/mscclpp/compare/v0.5.2...v0.6.0