F16 variants - Update loads and stores to AVX2 - Group 6 #649
+501
−230
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
F16 load store group 6
Contributor
There was a problem hiding this comment.
Pull request overview
This PR optimizes FP16 image processing operations by replacing scalar loads/stores with AVX2 vectorized intrinsics. The changes enable processing 24 elements at once with AVX2 (versus 12 with SSE), achieving 28-58% performance gains for PKD3 to PLN3 variants.
Key changes:
- Replaced SSE implementations with AVX2 variants for six augmentations (Copy, Color Jitter, Crop, Gridmask, Ricap, Crop and Patch)
- Direct FP16-to-FP32 AVX2 conversions eliminate intermediate scalar conversion loops
- Unified vectorization constants via conditional compilation for AVX2/SSE paths
Reviewed changes
Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| src/modules/tensor/cpu/kernel/saturation.cpp | Added AVX2 fast path with direct F16-to-F32 SIMD operations for saturation adjustments across all layout combinations |
| src/modules/tensor/cpu/kernel/ricap.cpp | Implemented AVX2 vectorization for RICAP augmentation with unified vector increment handling |
| src/modules/tensor/cpu/kernel/hue.cpp | Added AVX2 support for hue adjustments with direct F16 conversions; fixed incorrect cast in scalar fallback path |
| src/modules/tensor/cpu/kernel/gridmask.cpp | Introduced AVX2 mask computation functions and integrated them across all gridmask layout paths |
| src/modules/tensor/cpu/kernel/crop_and_patch.cpp | Added AVX2 vectorization for crop-and-patch operations with unified alignment calculations |
| src/modules/tensor/cpu/kernel/crop.cpp | Implemented AVX2 fast path for crop operations across layout toggle scenarios |
| src/modules/tensor/cpu/kernel/copy.cpp | Added AVX2 support for copy operations with direct F16 SIMD loads/stores |
| src/modules/tensor/cpu/kernel/color_jitter.cpp | Implemented AVX2 color jitter computation with proper F16 boundary checking |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| else if ((srcDescPtr->c == 3) && (srcDescPtr->layout == RpptLayout::NHWC) && (dstDescPtr->layout == RpptLayout::NHWC)) | ||
| { | ||
| Rpp32u alignedLength = bufferLength & ~3; | ||
| // Rpp32u alignedLength = bufferLength & ~3; |
There was a problem hiding this comment.
Commented-out code should be removed rather than left in the codebase. The alignedLength calculation is now handled by the conditional compilation block above.
Suggested change
| // Rpp32u alignedLength = bufferLength & ~3; |
Comment on lines
1255
to
1256
| // Rpp32u alignedLength = bufferLength & ~3; | ||
|
|
There was a problem hiding this comment.
Commented-out code should be removed rather than left in the codebase. The alignedLength calculation is now handled by the conditional compilation block above.
Suggested change
| // Rpp32u alignedLength = bufferLength & ~3; |
Hue and Saturation SSE Updates - F16 Copilot Fix
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #649 +/- ##
===========================================
+ Coverage 88.16% 88.38% +0.21%
===========================================
Files 195 195
Lines 82723 82420 -303
===========================================
- Hits 72932 72839 -93
+ Misses 9791 9581 -210
🚀 New features to boost your workflow:
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 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.
This PR includes replacement of scalar loads/stores and conversion to FP32, with AVX2 intrinsics - no additions or removals to external user API.