matmul

cusparseLt - cusparseMatMul

Description

This sample demonstrates the usage of cuSPARSELt library and cusparseMatMul APIs for performing structured matrix - dense matrix multiplication by exploiting NVIDIA Sparse Tensor Cores, where the structured matrix is compressed with 50% sparsity ratio.

cusparseLt Documentation

D = alpha * A * B + beta * C

where A is an structured matrix, and B, C, D are dense matrices

Building

• Linux

  make CUSPARSELT_PATH=<cusparseLt_path> CUDA_TOOLKIT_PATH=<cuda_toolkit_path>

• or in alternative:

  mkdir build
  cd build
  cmake -DCUSPARSELT_PATH=<cusparseLt_path> -DCMAKE_CUDA_COMPILER=<nvcc_path> ..
  make

Support

• Supported SM Architectures: SM 8.0, SM 8.6, SM 8.7, SM 8.9, SM 9.0
• Supported OSes: Linux, Windows
• Supported CPU Architectures: x86_64, arm64
• Supported Compilers: gcc, clang, Microsoft msvc, Nvidia HPC SDK nvc
• Language: C++14

Prerequisites

• CUDA 12.0 toolkit (or above) and compatible driver (see CUDA Driver Release Notes).
• cusparseLt 0.6.2 or above
• CMake 3.18 or above