nki(matrix_transpose): NKI (Trainium) implementation - #291

Open

bowencui123 wants to merge 8 commits into

mainfrom

bowen/nki/matrix_transpose

Commits on Aug 29, 2026

nki(matrix_transpose): NKI (Trainium) implementation

committed
8b226e2

Commits on Aug 30, 2026

Merge branch 'main' into bowen/nki/matrix_transpose
Cecilia123li
authored
e902a55

Commits on Sep 9, 2026

nki(matrix_transpose): add NKI benchmark results (trn2.3xlarge, LNC2)

Ubuntu
and
claude
committed
cab665e

Commits on Sep 15, 2026

Merge remote-tracking branch 'origin/main' into bowen/nki/matrix_transpose
bowencui123
committed
990bd07

Commits on Sep 16, 2026

nki(matrix_transpose): LNC2 dma_transpose kernel (int8 via bf16 nc_transpose blocks), no host pad; exact int8 XLA torch baseline (int16 transpose + clamp); rerun default+autotune benchmarks

bowencui123
and
claude
committed
c3e4526

Commits on Sep 21, 2026

Commits on Sep 22, 2026

nki(matrix_transpose): drop docstrings and inline comments

bowencui123
and
claude
committed
ec1bd2e