COSMA-v2.6.6

Fix linking against cray-libsci.

COSMA-v2.6.5

• fix a bug in tiled-mm API
• fix cmake related to nccl/rccl

COSMA-v2.6.4

Update submodules, minor fixes in cmake.

COSMA-v2.6.3

Improvements in cmake config. Update to new tiled-mm API.

COSMA-v2.6.2

This release fixes a bug in find_package(cosma) (cmake).

COSMA-v2.6.1

This release fixes the issues of COSMA-v2.6.0 coming from resizing the memory pool, as reported here.

2.6.0-fixed

Fixed a bug with memory pool resizing.

COSMA-v2.6.0

This release enables COSMA to take advantage of fast GPU-to-GPU interconnects like NVLink, to efficiently utilize modern Multi-GPU Systems. This is achieved in 2 ways:

• Using NCCL/RCCL Libraries: by specifying -DCOSMA_WITH_NCCL=ON cmake option.
• Using GPU-aware MPI: by specifying -DCOSMA_WITH_GPU_AWARE_MPI=ON cmake option, as proposed here.
  See README and INSTALL for more info on how to build.

In addition, the following performance improvemets have been made:

• Improved Caching:
  • all nccl buffers, MPI comms, nccl comms are cached and reused when appropriate.
  • all device memory is cached and reused.
• Reduced Data Trasfers: the GPU backend of COSMA called Tiled-MM is extended to offer the possibility to the user to leave the resulting matrix C on the GPU. In that case, there is no need to trasfer matrix C from device to host, which not only reduces the communication, but also speeds up the whole cpu->gpu pipeline as no additional synchronizations are needed. Furthermore, reduce_scatter operation does not have to wait for C to be transfered back to host but is immediately invoked with GPU pointers, thus utilizing fast inter-gpu links. This way, there is no unnecessary data transfers between cpu<->gpu.
• All collectives updated: both all-gather and reduce-scatter collectives are improved.
• Reduced Data Reshuffling: avoids double reshuffling of data, i.e. the data from NCCL/RCCL GPU buffers is immediately copied in the right layout, without additional reshuffling.
• Works for variable blocks: NCCL/RCCL' reduce_scatter operation assumes that all the blocks are of the same size and is hence not completely equivalent to MPI_Reduce_scatterv which we previously used. We padded all the blocks to be able to overcome this issue.
• Portability: Supports both NVIDIA and AMD GPUs.
• Tiled-MM: Updated submodule
• COSTA: Updated submodule

COSMA-v2.5.1

Fixes the building issue with cmake versions prior to 3.12.2.

COSMA-v2.5.0

This version brings the following improvements:

• [feature] Adds COSMA_DIM_THRESHOLD environment variable to cosma_prefixed_pxgemm.
• [improvements] Fixes the building issues and dependency handling in CMake.
• [bugfix] Fixes OpenMP race conditions.
• [bugfix] Resolves the problem with setting devices when running COSMA on multigpu systems.