search

NVlabs / timeloop

Timeloop performs modeling, mapping and code-generation for tensor algebra workloads on various accelerator architectures.
https://timeloop.csail.mit.edu/
BSD 3-Clause "New" or "Revised" License
340 stars 104 forks source link

Cutlass permutation is KMN? #262

Open ziyuhuang123 opened 6 months ago

ziyuhuang123 commented 6 months ago

I am puzzled about the permutation concept... I know cutlass way to compute GEMM, https://developer.nvidia.com/blog/cutlass-linear-algebra-cuda/ In timeloop's permutation idea, this is in shared memory KMN and DRAM KMN, right?

angshuman-parashar commented 6 months ago

I'm not sure I understand the question. Timeloop models a multi-level hierarchy. At each level the permutation describes the pattern in which tiles are sent to the next level in the hierarchy.