NDS

This is the source code used for NDS: N-Dimensional Storage

Install

make

Structure

• cuda_samples: Copied from cuda samples that helps to learn TCU wmma

• data: Contains matrix generator in sequential/NDS format

• lib:

  • gemm: offers APIs binding cublasGemm with NDS format
  • spdkrpc: communicates to SPDK app to fetch matrix from remote server
  • tensorstore: converts data format between sequential and NDS formats

• test:

  • cublas_mm_verification: Verify the resulting matrix is correct by using cublasGEMM
  • matrix_transpose: Verify and check the performance of fetching data from SPDK transposing matrix.
  • mm_verification: Verify the resulting matrix is correct by using blockmm on CPU
  • read_test: Verify tensorstore functionalities
  • tensorstore_mm_perftest: Performance test of APIs from gemm library
  • tensorstore_mm_verification: Verifications of APIs from gemm library
  • transpose_perftest: Performance test for cublasGEMM after applying normal/transposed matrix operands