mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-11 17:00:18 +00:00
8c13df07bf
* Improve formatting of docs; Add a note about the DL_KERNELS flag * Change the recommended version of ROCm to 5.6
53 lines
1.1 KiB
ReStructuredText
53 lines
1.1 KiB
ReStructuredText
|
|
*******************
|
|
API Reference Guide
|
|
*******************
|
|
|
|
=================
|
|
Introduction
|
|
=================
|
|
|
|
This document contains details of the APIs for the Composable Kernel (CK) library and introduces
|
|
some of the key design principles that are used to write new classes that extend CK functionality.
|
|
|
|
=================
|
|
Using CK API
|
|
=================
|
|
|
|
This section describes how to use the CK library API.
|
|
|
|
=================
|
|
CK Datatypes
|
|
=================
|
|
|
|
-----------------
|
|
DeviceMem
|
|
-----------------
|
|
|
|
.. doxygenstruct:: DeviceMem
|
|
|
|
---------------------------
|
|
Kernels For Flashattention
|
|
---------------------------
|
|
|
|
The Flashattention algorithm is defined in :cite:t:`dao2022flashattention`. This sections lists
|
|
the classes that are used in the CK GPU implementation of Flashattention.
|
|
|
|
**Gridwise classes**
|
|
|
|
.. doxygenstruct:: ck::GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
|
|
|
|
**Blockwise classes**
|
|
|
|
.. doxygenstruct:: ck::ThreadGroupTensorSliceTransfer_v4r1
|
|
|
|
.. doxygenstruct:: ck::BlockwiseGemmXdlops_v2
|
|
|
|
.. doxygenstruct:: ck::BlockwiseSoftmax
|
|
|
|
**Threadwise classes**
|
|
|
|
.. doxygenstruct:: ck::ThreadwiseTensorSliceTransfer_StaticToStatic
|
|
|
|
.. bibliography::
|