search

ROCm / composable_kernel

Composable Kernel: Performance Portable Programming Model for Machine Learning Tensor Operators
https://rocm.docs.amd.com/projects/composable_kernel/en/latest/
Other
321 stars 129 forks source link

[Issue]: tile_distribution_encoding.hpp parameters explaination #1693

Open demonsan opened 4 days ago

demonsan commented 4 days ago

Problem Description

It is hard to figure out the template parameters here. RsLengths, HsLengthss, Ps2RHssMajor, Ps2RHssMinor, Ys2RHsMajor, Ys2RHsMinor. Need a Readme for this to explain the meaning and how they are transformed.

Operating System

Ubuntu22.04 LTS

CPU

AMD Ryzen9 5900X

GPU

AMD Radeon RX 7900 XTX

Other

No response

ROCm Version

ROCm 6.0.0

ROCm Component

No response

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

ppanchad-amd commented 3 days ago

Hi @demonsan. Internal ticket has been created to assist with your issue. Thanks!

demonsan commented 3 days ago

From tile_distribution, i found some comments. It seems R means replication, H means hidden, P means partition. I'm tring to figure out Y dimension meanning and how they are composed together. It will be really helpful if an example illustrating the mechnasim of tile_distribution can be provided. thx :)

schung-amd commented 1 day ago

Hi @demonsan, unfortunately we don't have any guidance prepared for external users about many aspects of CK at the moment, including this. If you have specific questions I can pass them to the internal team and see if there are straightforward answers.