microsoft/mscclpp

Issues

• [Feature] Ethernet connection

  #121 opened 8 months ago by saeedmaleki
  0

• [Bug] Stronger correctness check in mscclpp-test

  #198 opened a year ago by chhwang
  0

• [Feature] Support python-based mscclpp-test

  #187 opened a year ago by Binyang2014
  0

• [Feature] fp16 allreduce.

  #192 opened a year ago by saeedmaleki
  1

• MSCCL++ v0.4.0 Release Plan (Released)

  #160 opened a year ago by chhwang
  1

• [Feature] a warning for when CQ is about to be full and ask user to flush it.

  #194 opened a year ago by saeedmaleki
  0

• [Bug] GIL release for python bindings that require waiting.

  #175 opened a year ago by saeedmaleki
  0

• [Bug] check on nv_peer module installation during cmake

  #191 opened a year ago by saeedmaleki
  0

• [Feature] waits should have a way to specify inf waiting

  #193 opened a year ago by saeedmaleki
  2

• MSCCL++ v0.3.0 Release Plan

  #89 opened a year ago by chhwang
  1

• [Feature] `getPacket` arg list does not match the `get` function from `sm_channels`

  #158 opened a year ago by saeedmaleki
  0

• [feature] `ProxyChannel` should not be taking device handles in for constructors.

  #155 opened a year ago by saeedmaleki
  0

• [Bug] Bootstrap occasionally returns "Address in use" error during `initialize()`

  #163 opened a year ago by chhwang
  2

• [Feature] semaphores do not take a timeout for wait.

  #161 opened a year ago by saeedmaleki
  1

• [feature] `poll` instead of wait.

  #176 opened a year ago by saeedmaleki
  0

• [Bug] Compilation fails with `CMAKE_BUILD_TYPE=Debug`

  #174 opened a year ago by chhwang
  0

• [Performance] Improve single-node AllReduce latency

  #164 opened a year ago by chhwang
  2

• [Design] SM channel doesn't cover important use cases

  #133 opened a year ago by olsaarik
  2

• [Unit Test] Listing unit tests needed

  #115 opened a year ago by chhwang
  0

• [Feature] Support custom `RegisteredMemory::Impl`

  #135 opened a year ago by chhwang
  1

• [bug] atomic operations do not have a signal option

  #177 opened a year ago by saeedmaleki
  0

• [Design] Hide `RegisteredMemory::Impl` from the user interface

  #168 opened a year ago by chhwang
  1

• [Feature] Connection without Bootstrap

  #137 opened a year ago by olsaarik
  2

• [Unit Test] Context client-server Connection forming test needed

  #173 opened a year ago by olsaarik
  0

• [Bug] Need to call cudaIpcCloseMemHandle to release remote registered memory

  #165 opened a year ago by Binyang2014
  1

• [Unit Test] Add Python unit tests

  #134 opened a year ago by chhwang
  0

• [Feature] Support `POLL_PRINT_ON_STUCK`

  #99 opened a year ago by chhwang
  0

• [feature] all of `#define` stuff need to be an arg

  #149 opened a year ago by saeedmaleki
  3

• [feature] Enable custom fifo size and tail flush

  #138 opened a year ago by saeedmaleki
  2

• [feature] `sm_channel` constructors need to be improved.

  #148 opened a year ago by saeedmaleki
  0

• [bug] device functions really need to be in a different file

  #139 opened a year ago by saeedmaleki
  1

• [bug] `LLPacket` needs a jailbreak for reading a packet.

  #157 opened a year ago by saeedmaleki
  1

• [bug] `fifo` has 128bit atomic reading problems.

  #154 opened a year ago by saeedmaleki
  1

• [feature] `ProxyService` does not need communicator in constructor.

  #152 opened a year ago by saeedmaleki
  1

• [feature] add `.def_prop_ro` for all device components so that they can be passed to GPU kernels.

  #144 opened a year ago by saeedmaleki
  1

• [bug] change all `__CUDACC__` to `__CUDA_ARCH__`

  #143 opened a year ago by saeedmaleki
  1

• [feature] python binding for DeviceSyncer

  #156 opened a year ago by saeedmaleki
  0

• [bug] the `fifo` from proxy cannot be bound by `nb` to python

  #153 opened a year ago by saeedmaleki
  0

• [feature] `sm_channel.put` can only copy from global memory

  #151 opened a year ago by saeedmaleki
  0

• [feature] find the right folder for include directory

  #140 opened a year ago by saeedmaleki
  0

• [Bug] SmChannel and ownership of semaphores

  #126 opened a year ago by saeedmaleki
  0

• [Bug] When `nvcc` is not in `$PATH`, `cmake` fails.

  #128 opened a year ago by saeedmaleki
  1

• [Feature] Add Python bindings

  #120 opened a year ago by olsaarik
  0

• [Bug] Python binding cannot be ctrl+c'ed while waiting for bootstrap

  #127 opened a year ago by saeedmaleki
  1

• [Bug] nanobind is leaking with connection

  #130 opened a year ago by saeedmaleki
  2

• pickle for uniq_id

  #129 opened a year ago by saeedmaleki
  1

• [Bug] Performance downgrade for allgather kernel 2

  #105 opened 2 years ago by Binyang2014
  0

• [Bug] Hanging bootstrap communication

  #92 opened 2 years ago by chhwang
  4

• [Performance] AllReduce performance debugging

  #90 opened 2 years ago by chhwang
  5

• A test for the core assumption of data + signal of MSCCL++

  #91 opened 2 years ago by saeedmaleki
  2