Skip to content

Add gemm_shuffle host api#71

Merged
asroy merged 5 commits into
developfrom
gemm-shuffle-host-api
Jan 21, 2022
Merged

Add gemm_shuffle host api#71
asroy merged 5 commits into
developfrom
gemm-shuffle-host-api

Conversation

@rocking5566
Copy link
Copy Markdown
Collaborator

@rocking5566 rocking5566 commented Jan 21, 2022

[What]

  1. Add DeviceGemmShuffleXdl
  2. Revise example of gemm_xdl
    [Why] Prepare to add shuuffle version of D = alpha * (A * B) + beta * C
    [How] Imitate DeviceGemmXdl and device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp

1. Add DeviceGemmShuffleXdl
 2. Revise example of gemm_xdl
[Why] Prepare to add shuuffle version of D = alpha * (A * B) + beta * C
[How] Imitate DeviceGemmXdl and device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
@rocking5566 rocking5566 requested a review from asroy January 21, 2022 05:55
@rocking5566 rocking5566 changed the title [What] Add gemm_shuffle host api Jan 21, 2022
asroy
asroy previously approved these changes Jan 21, 2022
@asroy
Copy link
Copy Markdown
Contributor

asroy commented Jan 21, 2022

@rocking5566 Need to do clang-format next time

@asroy asroy merged commit 4d40b19 into develop Jan 21, 2022
@rocking5566 rocking5566 deleted the gemm-shuffle-host-api branch January 21, 2022 07:40
@rocking5566 rocking5566 restored the gemm-shuffle-host-api branch January 21, 2022 07:42
MXdlPerWave,
NXdlPerWave,
ABlockTransferThreadClusterLengths_K0_M_K1,
Sequence<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder,
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@rocking5566 There shouldn't be hardcoded parameters here.

}
else
{
const auto kernel = kernel_gemm_xdlops_v2r3<
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@rocking5566 Is v2r3 a typo here, since you were invoking v3r1 kernel when condition is true? @asroy can you confirm if this is intended?

@illsilin illsilin deleted the gemm-shuffle-host-api branch December 7, 2023 18:39
carlushuang added a commit that referenced this pull request Jan 31, 2024
* compute correct

* improve perf, but seems pipeline has duplicated ISA

* refactor generate_kernel

* remove duplicated GetBlockQKGemm

* finialize a more generic codegen

* refactor into autogen API

* fix some comment

* Use occupancy=1 for hdim=256

* support hdim=256

* modify some comment

* we no longer need to change target inside a file

* update bench script

* add readme

* modify

---------

Co-authored-by: Po Yen, Chen <PoYen.Chen@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants