Implement SplitK and StreamK algorithm for Intel PVC #132
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…tlass-fork into intel_pvc_streamk
* Need to fix splitK for batch > 1
muhammad-tanvir-1211
force-pushed
the
intel_pvc_streamk
branch
from
b6d7abf
to
58d082e
Compare
muhammad-tanvir-1211
force-pushed
the
intel_pvc_streamk
branch
from
58d082e
to
f559e31
Compare
muhammad-tanvir-1211
force-pushed
the
intel_pvc_streamk
branch
2 times, most recently
from
cef3fec
to
e4fefc2
Compare
muhammad-tanvir-1211
force-pushed
the
intel_pvc_streamk
branch
from
e4fefc2
to
2ab1cd8
Compare
* Instantiate new accumulator registers per iteration
|
In light of #138 , could you rename the newly added |
| BlockStripedReduceT::store(reduction_workspace_array, *accumulator_array, barrier_group_thread_idx); | ||
| } | ||
| else { | ||
| // Wait until the preceding split added its accumulators |
There was a problem hiding this comment.
why does this need to wait on the preceding one even for non-deterministic?
| BarrierManager::wait_eq(barrier_idx, lock_workspace, barrier_group_thread_idx, lock_idx, work_tile_info.K_idx); | ||
|
|
||
| // Perform reduction in workspace | ||
| BlockStripedReduceT::reduce(reduction_workspace_array, *accumulator_array, barrier_group_thread_idx); |
There was a problem hiding this comment.
for the deterministic case, why does that need to be reduce(/atomic_add) and load_add isn't sufficient? My understanding is that for deterministic case only a single K_idx is adding at the same time and if so atomic shouldn't be needed.
AD2605
reviewed
Sep 12, 2024
include/cutlass/gemm/kernel/intel_pvc_tile_scheduler_streamk.hpp
Outdated
Show resolved
Hide resolved
…tlass-fork into intel_pvc_streamk
muhammad-tanvir-1211
force-pushed
the
intel_pvc_streamk
branch
from
82d7092
to
db705da
Compare
Done. |
* Removed l2 workspace alignment
muhammad-tanvir-1211
force-pushed
the
intel_pvc_streamk
branch
from
db705da
to
93d87fc
Compare
aacostadiaz
reviewed
Oct 21, 2024
aacostadiaz
reviewed
Oct 21, 2024
aacostadiaz
reviewed
Oct 21, 2024
aacostadiaz
reviewed
Oct 21, 2024
aacostadiaz
reviewed
Oct 21, 2024
aacostadiaz
reviewed
Oct 21, 2024
aacostadiaz
reviewed
Oct 21, 2024
muhammad-tanvir-1211
force-pushed
the
intel_pvc_streamk
branch
from
29fe8a5
to
8e20733
Compare
…tlass-fork into intel_pvc_streamk
…tlass-fork into intel_pvc_streamk
…tlass-fork into intel_pvc_streamk
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds the required code for implementing the SplitK and StreamK work distribution algorithms for Intel PVC.