mkadavil 7114376519 New kernels for int4 B matrix reordering following BF16 kernel schema. ...

-To enable Weight-only-Quantization (WOQ) workflow, new LPGEMM APIs
are required where data types are A:bf16, B:int4 and C:f32/bf16. It
is expected that the BF16 kernels will be reused within this API and
subsequently the B matrix needs to be reordered following the BF16
kernel schema, but with the reordered matrix type still being int4. To
address this, new BF16 reorder kernels enabling the same are added.

AMD-Internal: [SWLCSG-2943]
Change-Id: Ib770ecbf90a3d906deafece94b1a96e0b9412738

2024-07-25 01:10:13 -04:00

..

1

Optimization for DCOPY and SCOPY API

2024-07-24 08:23:07 -04:00

1f

DGEMV Optimizations for NO_TRANSPOSE cases

2024-07-24 15:59:36 +05:30

1m

DGEMM optimizations for Turin Classic

2024-07-09 07:53:27 -04:00

3

Added 24x8 triangular kernels for DGEMMT SUP

2024-07-22 12:02:30 -04:00

aocl_smart

CMake: Adding new portable CMake system.

2023-11-09 15:49:45 +05:30

lpgemm

New kernels for int4 B matrix reordering following BF16 kernel schema.

2024-07-25 01:10:13 -04:00

bli_kernels_zen4.h

DGEMV Optimizations for NO_TRANSPOSE cases

2024-07-24 15:59:36 +05:30