Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix joint_matrix implementation to match latest api #491

Merged
merged 22 commits into from
Feb 28, 2024
Merged

Fix joint_matrix implementation to match latest api #491

merged 22 commits into from
Feb 28, 2024

Conversation

muhammad-tanvir-1211
Copy link
Collaborator

@muhammad-tanvir-1211 muhammad-tanvir-1211 commented Jan 5, 2024 •
edited
Loading

This PR updates the joint_matrix based GEMM implementation to match the api from intel/llvm (commit # 2a828f49283145433dc9bbbff74cefcb2d2b10dc).

  • Removed the deprecated get_wi_data() calls and replaced them with joint_matrix_apply() and joint_matrix_copy() calls.
  • Changed the leading dimensions for shared memory accesses to avoid excessive bank conflicts.
  • Refactored the code to write output data to Global Memory in a coalesced way (this wasn't possible earlier due to the joint_matrix_store() operation performed on the Global pointer C)
  • Fixed race condition for batch size > 1.

@muhammad-tanvir-1211 muhammad-tanvir-1211 mentioned this pull request Jan 5, 2024
Closed
@muhammad-tanvir-1211 muhammad-tanvir-1211 force-pushed the joint_matrix_fix branch 2 times, most recently from 00d00d7 to cb06e62 Compare January 12, 2024 14:57
OuadiElfarouki
OuadiElfarouki approved these changes Jan 24, 2024
View reviewed changes
Copy link
Collaborator

@OuadiElfarouki OuadiElfarouki left a comment

Choose a reason for hiding this comment

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

LGTM!

@muhammad-tanvir-1211 muhammad-tanvir-1211 force-pushed the joint_matrix_fix branch 3 times, most recently from ba3ca46 to 7985cbd Compare February 1, 2024 14:22
@muhammad-tanvir-1211
Fix joint_matrix_mad api call
cdc3486 
* Remove get_wi_data() and replace with joint_matrix_copy and joint_matrix_apply
@muhammad-tanvir-1211
Fixed output for matrices with no corner cases
548321c
@muhammad-tanvir-1211
Added support for corner cases
65d720b 
* Need to fix transpose case
@muhammad-tanvir-1211
Fixed the transpose implementation
1f1b41c
@muhammad-tanvir-1211
Fixed corner case output
95b11bd
@muhammad-tanvir-1211
Code working for gemm and gemm_batched
e85a2fa
@muhammad-tanvir-1211
Unrolled the Global Memory write loops
6d659af
@muhammad-tanvir-1211
Reorder computation loops
277d787
@muhammad-tanvir-1211
Updated comments and removed redundant code
eca4634
@muhammad-tanvir-1211
Fixed build with release compiler
37d26e1
@muhammad-tanvir-1211
Fix loop bounds
9d2f540
@pgorlani @muhammad-tanvir-1211
Add functions for checking lower precision
5e8d8c6
@muhammad-tanvir-1211
Fixed synchronization for double buffering
60fa20a
@muhammad-tanvir-1211
Restriced VectorSize to 1 for joint_matrix
ba4cd87 
* Updated the load/store file with proper vectorized load/store for future use.
pgorlani
pgorlani reviewed Feb 19, 2024
View reviewed changes
test/blas_test.hpp Outdated Show resolved Hide resolved
pgorlani
pgorlani reviewed Feb 19, 2024
View reviewed changes
test/blas_test.hpp Outdated Show resolved Hide resolved
@muhammad-tanvir-1211
Address test commit feedback
fe55302 
* Fix half tests by changing the initialization values
* Reduced the no. of tests executed
@muhammad-tanvir-1211
Reduced the initializer range
50c150b 
* Increase error margin
@muhammad-tanvir-1211
Merge branch 'master' into joint_matrix_fix
6b87355
@muhammad-tanvir-1211 muhammad-tanvir-1211 merged commit 861b310 into codeplaysoftware:master Feb 28, 2024
3 checks passed
@muhammad-tanvir-1211 muhammad-tanvir-1211 mentioned this pull request Mar 28, 2024
Merged
@muhammad-tanvir-1211 muhammad-tanvir-1211 deleted the joint_matrix_fix branch April 11, 2024 17:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@pgorlani pgorlani pgorlani left review comments

@s-Nick s-Nick s-Nick approved these changes

@OuadiElfarouki OuadiElfarouki OuadiElfarouki approved these changes

@aacostadiaz aacostadiaz Awaiting requested review from aacostadiaz

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@muhammad-tanvir-1211 @s-Nick @pgorlani @OuadiElfarouki