iarray.matmul_params#

iarray.matmul_params(ashape, bshape, dtype=None, l2_size=None, chunk_size=134217728)#

Given a matrix multiplication of two arrays, it computes the chunks and the blocks of the operands to use an optimized version of the matmul algorithm.

Parameters
  • ashape (tuple or list) – The shape of the operand a.

  • bshape (tuple or list) – The shape of the operand b.

  • dtype – The dtype of each item.

  • l2_size (int) – The size of the l2 cache. It is used to compute the size of the blocks.

  • chunk_size (int) – The maximum chunksize allowed. It is used to compute the size of the chunks.

Returns

params – A tuple specifying the chunks and the blocks of the matmul operands a and b (achunks, ablocks, bchunks, bblocks).

Return type

tuple