[Optimization]: Use dynamic HIP kernel block sizes#122
Open
zacharyvincze wants to merge 32 commits intoROCm:developfrom
Open
[Optimization]: Use dynamic HIP kernel block sizes#122zacharyvincze wants to merge 32 commits intoROCm:developfrom
zacharyvincze wants to merge 32 commits intoROCm:developfrom
Conversation
zacharyvincze
requested review from
Copilot,
jeffqjiangNew,
msiddaiah and
paveltc
There was a problem hiding this comment.
Pull request overview
This PR introduces dynamic HIP kernel block size determination to optimize performance across different GPU architectures. It adds helper functions GetMaximumPotentialBlockSize2D and GetGridSize2D to replace hardcoded block sizes throughout the codebase.
Changes:
- Added
GetMaximumPotentialBlockSize2DandGetGridSize2Dhelper functions ininclude/core/detail/hip_utils.hpp - Refactored 14 operator files to use dynamic block sizes instead of hardcoded values
- Cleaned up unused includes and reorganized some headers
Reviewed changes
Copilot reviewed 20 out of 20 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| include/core/detail/hip_utils.hpp | Adds helper functions for dynamic 2D block size calculation using HIP occupancy API |
| include/kernels/host/bnd_box_host.hpp | Adds missing math_vector.hpp include |
| include/kernels/device/bnd_box_device.hpp | Adds missing math_vector.hpp include |
| src/op_warp_perspective.cpp | Replaces hardcoded (64,16) block size with dynamic calculation |
| src/op_thresholding.cpp | Replaces hardcoded (64,16) block size with dynamic calculation per threshold type |
| src/op_rotate.cpp | Replaces hardcoded (32,16) block size with dynamic calculation |
| src/op_resize.cpp | Replaces hardcoded (64,16) block size with dynamic calculation; reformats function map |
| src/op_remap.cpp | Replaces hardcoded (64,16) block size with dynamic calculation |
| src/op_normalize.cpp | Replaces hardcoded (32,8) block size with dynamic calculation; reformats line breaks |
| src/op_non_max_suppression.cpp | Removes unused include (no block size changes as this uses 1D blocks) |
| src/op_histogram.cpp | Removes unused includes (no block size changes) |
| src/op_gamma_contrast.cpp | Replaces hardcoded (64,16) block size with dynamic calculation |
| src/op_flip.cpp | Replaces hardcoded (64,16) block size with dynamic calculation |
| src/op_cvt_color.cpp | Replaces hardcoded (32,16) block size with dynamic calculation per color conversion type |
| src/op_custom_crop.cpp | Replaces hardcoded (64,16) block size with dynamic calculation |
| src/op_copy_make_border.cpp | Replaces hardcoded (64,16) block size with dynamic calculation |
| src/op_convert_to.cpp | Replaces hardcoded (64,16) block size with dynamic calculation; reformats function signatures |
| src/op_composite.cpp | Replaces hardcoded (64,16) block size with dynamic calculation |
| src/op_bnd_box.cpp | Replaces hardcoded (32,32) block size with dynamic calculation |
| src/op_bilateral_filter.cpp | Replaces hardcoded (8,8) block size with dynamic calculation |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## develop #122 +/- ##
===========================================
- Coverage 74.40% 74.39% -0.01%
===========================================
Files 79 79
Lines 3355 3381 +26
Branches 738 740 +2
===========================================
+ Hits 2496 2515 +19
- Misses 378 382 +4
- Partials 481 484 +3
🚀 New features to boost your workflow:
|
jeffqjiangNew
approved these changes
Feb 19, 2026
| @@ -1,2 +1,2 @@ | |||
| /** | |||
| Copyright (c) 2025 Advanced Micro Devices, Inc. All rights reserved. | |||
Contributor
There was a problem hiding this comment.
Update (c) year to 2026 here and other places.
paveltc
approved these changes
Feb 20, 2026
…aryvincze/rocCV into zv/optimization/dynamic-block-sizes
Review: [Optimization] Dynamic HIP kernel block sizesPerformance optimization for GPU kernel launches: What's changed:
Assessment: APPROVED - Performance improvement with no API changes. Already approved, this is a solid optimization. Removing hard-coded block sizes improves portability across different GPU architectures (MI series, RDNA, etc.). The approach of querying wavefront size at runtime is the right way to handle this. Nice cleanup of magic numbers throughout the codebase. |
…aryvincze/rocCV into zv/optimization/dynamic-block-sizes
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
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.
PR Description
Notes