search

ROCm / Tensile

Stretching GPU performance for GEMMs and tensor contractions.
MIT License
218 stars 147 forks source link

Revert "Optimization for ShadowLimit (#1829)" #1855

Closed nakajee closed 10 months ago

nakajee commented 10 months ago

This reverts commit 9e8f1ab9b708c52271dd53d5e20d45c59d3b5993.

nakajee commented 10 months ago

This logic does not work in edge case. We cannot uniformly remove ShadowLimit code. We need some way to ensue no edge case by using assertion.

nakajee commented 10 months ago

I confirmed that 90a precheckin test fail was not caused by this change. I could reproduce the issue without my change.