218 warpsize patch - #2916

Open

i-chaochen wants to merge 5 commits into

r2.18-rocm-enhancedfrom

218_warpsize_patch

Commits on Apr 1, 2025

Remove CUDA 12.1 workaround from reduction logic

beckerhe
authored and
i-chaochen
committed
5674bfd

Commits on Apr 6, 2025

Use DeviceDescription instead of hard-coding warp size as 32 + remove legacy emitter
i-chaochen
committed
5e154de
PR #17814: [ROCM] buffer_comparator init bugfix

pemeliya
authored and
i-chaochen
committed
ee2233c
[XLA:GPU] Fix check failure when number of blocks is larger than the hardware limit per dim.

olegshyshkov
authored and
i-chaochen
committed
5b94d14

Commits on Apr 14, 2025

fixed build error
i-chaochen
committed
aa686cf