Aviral Goel b948026e16 feat: add split_k support for block scale gemm bquant mode. (#3653) ...

* WIP: add splitk to bquant

* feat: add support for bf8i4 and fp8i4 by calculating correct stride for packed data types

* chore: remove temporary test script

* fix: incorrect tile window length for splitted bq tensor window

* chore: improve comments

* test: add unit tests to cover bquant splitk functionality

* fix: conflict resolution by renaming variables

[ROCm/composable_kernel commit: 3e77721755]

2026-02-02 14:41:53 -08:00

..

block

[CK_Tile] Support for a4w4 (fp4) in block scale gemm AB quant (#3603)

2026-01-30 04:40:50 -07:00

kernel

feat: add split_k support for block scale gemm bquant mode. (#3653)

2026-02-02 14:41:53 -08:00

pipeline

[CK_Tile] Support for a4w4 (fp4) in block scale gemm AB quant (#3603)

2026-01-30 04:40:50 -07:00