Feature: Optimize dngvd_op with CPU/GPU Branching Based on nstart #5919
Conversation
This file contains hidden or 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
Collaborator
|
please fix the compiling error |
dyzheng
reviewed
Feb 25, 2025
dyzheng
reviewed
Feb 25, 2025
dyzheng
reviewed
Feb 25, 2025
dyzheng
reviewed
Feb 25, 2025
dyzheng
reviewed
Feb 25, 2025
dyzheng
reviewed
Feb 25, 2025
| // copied from ../cuda/dngvd_op.cu, "dngvd_op" | ||
| assert(nstart == ldh); | ||
|
|
||
| if (nstart > N_DCU){ |
Collaborator
There was a problem hiding this comment.
please add notes for this kernel has intersection point of the performance curves between CPU and DCU.
dyzheng
reviewed
Feb 25, 2025
| // copied from ../cuda/dngvd_op.cu, "dngvd_op" | ||
| assert(nstart == ldh); | ||
|
|
||
| if (nstart > N_DCU){ |
Collaborator
There was a problem hiding this comment.
please add notes for this kernel has intersection point of the performance curves between CPU and DCU.
Collaborator
There was a problem hiding this comment.
to be honest, N_DCU is tested only for "complex" kernel, but maybe not good intersection point for "double" and "complex" kernel.
mohanchen
added
Features Needed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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Outdated suggestions cannot be applied.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.
Currently,
dngvd_opinmodule_hsolver/kernels/rocm/dngvd_op.hip.cuuses the ROCm implementation for all input sizes (nstart). Performance analysis shows that the CPU implementation (dngvd_op.cpp) is faster for smaller nstart values.This PR proposes adding a conditional branch within the
dngvd_op<double, base_device::DEVICE_GPU>::operator()function to select the optimal implementation based onnstart:If nstart > 234, use the existing ROCm implementation.If nstart <= 234, call the CPU implementation (dngvd_op<double, base_device::DEVICE_CPU>).This change requires:
Adding an
if (nstart > 234) { ... } else { ... }block within the GPU operator().Inside the else block, calling the CPU implementation with appropriate type casts.
This optimization is expected to improve performance, especially for bigger matrix sizes.