search

ORNL / cpp-proposals-pub

Collaborating on papers for the ISO C++ committee - public repo
26 stars 26 forks source link

P2901 LEWG presentation: 2023/10/24 #419

Closed mhoemmen closed 10 months ago

mhoemmen commented 10 months ago

P2901 LEWG presentation: 2023/10/24

Batched: Solve multiple problems at a time

Why this library?

  1. Exposes intent; exposes optimizations (parallelization, vectorization)
  2. Easiest way to optimize (or at least amortize view overhead) for small problems
  3. Covers tensor case: Batched matrix multiply expresses every tensor contraction

C[i, j, b] = \sum_k A[i, k, b] * B[k, j, b]