Consistent HOST and HIP/pinned buffers for respective API #628
+844
−350
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
… rcm, color temperature
Mem copy elimination
r-abishek
changed the title
1 task
…memcpy_removal
Mem copy elimination version change and Review comments resolved
Contributor
|
@r-abishek please resolve merge conflicts |
Update patch version from 2.2.0 to 2.2.1
Copilot finished reviewing on behalf of
kiritigowda
November 17, 2025 18:07
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR removes internal host-to-HIP buffer copy functionality from RPP to ensure consistent memory management. GPU APIs now require users to provide HIP/pinned memory buffers directly, eliminating the copy_param_float(), copy_param_uint(), and similar helper functions that previously performed host-to-device copies within RPP.
Key changes include:
- Memory allocation updated from stack arrays to hipHostMalloc in test suite
- API function signatures updated to pass tensor pointers directly to HIP kernels
- Memory management responsibility shifted entirely to the user
Reviewed Changes
Copilot reviewed 28 out of 28 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| utilities/test_suite/HIP/Tensor_image_hip.cpp | Updated to use hipHostMalloc for parameter buffers instead of stack arrays; added cleanup code |
| src/modules/tensor/rppt_tensor_geometric_augmentations.cpp | Removed copy_param calls; parameters now passed directly to kernels |
| src/modules/tensor/rppt_tensor_filter_augmentations.cpp | Removed copy_param calls for gaussian_filter |
| src/modules/tensor/rppt_tensor_effects_augmentations.cpp | Removed copy_param calls; added hipHostMalloc for spatter mask arrays |
| src/modules/tensor/rppt_tensor_color_augmentations.cpp | Removed copy_param calls for all color augmentations |
| src/modules/tensor/hip/kernel/*.cpp | Updated function signatures to accept tensor pointers directly |
| src/include/tensor/hip_tensor_executors.hpp | Updated function declarations with new parameters |
| CMakeLists.txt | Version bump from 2.2.0 to 2.2.1; trailing whitespace cleanup |
| CHANGELOG.md | Added entry for memory copy elimination |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Address copilot comments for HIP HOST consistent allocation
Contributor
|
@r-abishek please check and resolve conflicts |
Contributor
|
@Srihari-mcw @HazarathKumarM please add the doc changes as we discussed offline |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 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.
RPP was originally also responsible for host to hip buffer conversions. This was removed during the course of tensor implementations to ensure all RPP HOST API only have HOST buffers, and GPU API only have HIP buffers (or pinned memory for smaller argument buffers).
The following functionality were still using the old style host->hip memcopy within RPP, and this is now being removed. After this, RPP tensor API will no longer be responsible for any HOST -> HIP buffer copy. The user is responsible to provide HOST buffers for HOST API, and HIP/Pinned memory for GPU API.
copy_param_float(), copy_param_uint() etc perform these copies and are now eliminated.
Just like all other tensor functionalities, pinned memory allocation from test suite is used for samller argument buffers.
These are the changed functionalities:
exposure
blend
brightness
color cast
color twist
constrast
crop mirror normalize
gamma_correction
gaussian_filter
noise
non_linear_blend
resize_mirror_normalize
water
@rrawther Please note equivalent changes in MIVisionX would need to be merged together with this PR.
A patch version change has been done for this tentatively from 2.2.0 to 2.2.1