Getting Started

Introduction

cuBLASMp aims to provide GPU-accelerated PBLAS-like basic linear algebra functionality.

cuBLASMp leverages the 2D block-cyclic data layout for load balancing and to maximize compatibility with PBLAS routines.

The library assumes the data is available on the device memory. It is the responsibility of the developer to allocate memory and to copy data between GPU memory and CPU memory using standard CUDA runtime APIs, such as cudaMallocAsync(), cudaFreeAsync(), cudaMemcpyAsync(), etc.

Hardware and Software Requirements

GPU Architectures	Compute Capability 7.0 (Volta) and above
CUDA	12.8 recommended, 12.x compatible
CPU Architectures	x86_64, arm64-sbsa
Operating System	Linux

• Recommended NVIDIA InfiniBand solutions for accelerated inter-node communication

Required Packages

• CUDA Toolkit 12.x (https://developer.nvidia.com/cuda-downloads)

• HPC-X v2.17.1 and above (https://developer.nvidia.com/networking/hpc-x) - contains OpenUCC and OpenUCX that satisfy cuBLASMp requirements.

• NVSHMEM v3.1 and above (https://developer.nvidia.com/nvshmem)

• CAL (https://developer.download.nvidia.com/compute/cublasmp/redist/libcal) - a companion library used for communication.

Recommended Packages

• NCCL v2.18.5 and above (https://developer.nvidia.com/nccl) - required to achieve better performance in functions not relying on NVSHMEM.

• GDRCopy v2.0+ (NVIDIA/gdrcopy) and nv_peer_mem (Mellanox/nv_peer_memory) - Allows underlying communication packages to use GPUDirect RDMA. If you install OpenUCX yourself, it should be configured with GDRCopy support.

• Mellanox OFED (https://www.mellanox.com/products/infiniband-drivers/linux/mlnx_ofed) - drivers for NVIDIA Infiniband Adapters (https://www.nvidia.com/en-us/networking/products/infiniband). If you install OpenUCX yourself, it should be configured with IB communications support.

Data Layout of Local Matrices

cuBLASMp assumes that local matrices are stored in column-major format.

Workflow

cuBLASMp’s workflow can be broken down as follows:

1. Bootstrap CAL communicator: cal_comm_create().

2. Initialize the library handle: cublasMpCreate().

3. Initialize grid descriptors: cublasMpGridCreate().

4. Initialize matrix descriptors: cublasMpMatrixDescriptorCreate().

5. Query the host and device buffer sizes for a given routine.

6. Allocate host and device workspace buffers for a given routine.

7. Execute the routine to perform the desired computation.

8. Synchronize local stream to make sure the result is available, if required: cal_stream_sync().

9. Deallocate host and device workspace.

10. Destroy matrix descriptors: cublasMpMatrixDescriptorDestroy().

11. Destroy grid descriptors: cublasMpGridDestroy().

12. Destroy cuBLASMp library handle: cublasMpDestroy().

13. Destroy CAL library handle: cal_comm_destroy().

Code Samples

Code samples can be found in the CUDALibrarySamples repository.