F16 variants - Update loads and stores to AVX2 - Group 5#637
Merged
kiritigowda merged 12 commits intoROCm:developfrom Nov 21, 2025
Merged
F16 variants - Update loads and stores to AVX2 - Group 5#637kiritigowda merged 12 commits intoROCm:developfrom
kiritigowda merged 12 commits intoROCm:developfrom
Conversation
FP16 Load/Store Updates
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR updates F16 (half-precision floating point) kernels to use AVX2 intrinsics for loading and storing data, replacing scalar conversions to FP32. The changes target the blend, color_cast, flip, and crop_mirror_normalize kernels, delivering performance improvements of 28.7% to 48.5% for FP16 operations.
- Replaces scalar F16 to F32 conversions with AVX2 SIMD intrinsics
- Introduces new AVX2 load/store functions for F16 data with mirroring support
- Updates conditional branching for better code clarity
Reviewed Changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| src/modules/tensor/cpu/kernel/flip.cpp | Updates flip kernel to use AVX2 F16 load/store functions, removes temporary F32 buffers, adjusts flip factor calculation for RGB channels |
| src/modules/tensor/cpu/kernel/crop_mirror_normalize.cpp | Replaces scalar F16 conversions with AVX2 intrinsics, improves conditional structure with else if |
| src/modules/tensor/cpu/kernel/color_cast.cpp | Adds AVX2 code paths with preprocessing directives, updates to use F16 load/store functions |
| src/modules/tensor/cpu/kernel/blend.cpp | Converts to AVX2 F16 operations, adds compile-time AVX2 feature detection |
| src/include/common/cpu/rpp_cpu_simd_load_store.hpp | Adds new F16 mirror load functions for AVX2 (pkd3 and pln3 variants) |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| _MM_TRANSPOSE4_PS(p128[4], p128[5], p128[6], p128[7]); /* Transpose the 4x4 matrix and forms [[R05 R06 R07 R08][B05 B06 B07 B08][G05 G06 G07 G08][R06 R07 R08 R09]] */ | ||
| p[0] = _mm256_setr_m128(p128[0], p128[4]); /* packs as R01-R08 */ | ||
| p[1] = _mm256_setr_m128(p128[1], p128[5]); /* packs as G01-G08 */ | ||
| p[2] = _mm256_setr_m128(p128[2], p128[6]); /* packs as B01-R08 */ |
There was a problem hiding this comment.
The comment incorrectly states "B01-R08" when it should be "B01-B08" to match the pattern of the other channels and correctly describe what is being packed.
Suggested change
| p[2] = _mm256_setr_m128(p128[2], p128[6]); /* packs as B01-R08 */ | |
| p[2] = _mm256_setr_m128(p128[2], p128[6]); /* packs as B01-B08 */ |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #637 +/- ##
===========================================
+ Coverage 88.24% 88.28% +0.04%
===========================================
Files 195 195
Lines 82712 82619 -93
===========================================
- Hits 72985 72934 -51
+ Misses 9727 9685 -42
🚀 New features to boost your workflow:
|
Fix comment in common function
Member
Author
|
@rrawther @LakshmiKumar23 CI failure is only due to (-0.09%) reduction in coverage. |
rrawther
approved these changes
Nov 20, 2025
ManasaDattaT
pushed a commit
to RooseweltMcW/rpp
that referenced
this pull request
Dec 19, 2025
* Updates for crop mirror normalize * Updated flip F16 rawC and load store modifications * Updated blend with AVX support for F16 bitdepth * Updated color cast with AVX support for F16 bitdepth * Remove empty lines * Update comments * Fix comment in common function --------- Co-authored-by: Srihari-mcw <srihari@multicorewareinc.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com>
HazarathKumarM
pushed a commit
to HazarathKumarM/rpp
that referenced
this pull request
Jan 6, 2026
* Updates for crop mirror normalize * Updated flip F16 rawC and load store modifications * Updated blend with AVX support for F16 bitdepth * Updated color cast with AVX support for F16 bitdepth * Remove empty lines * Update comments * Fix comment in common function --------- Co-authored-by: Srihari-mcw <srihari@multicorewareinc.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com>
JeniferC99
pushed a commit
that referenced
this pull request
Jan 22, 2026
* F16 variants - Update loads and stores to AVX2 - Group 4 (#627) * Make changes for exposure, log and spatter * Updates for crop mirror normalize * Fix memory issues with log 1D * Remove changes for crop mirror normalize and restore rpp_cpu_simd_load_store.hpp * Update the alignedLength for log --------- Co-authored-by: Srihari-mcw <srihari@multicorewareinc.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com> Co-authored-by: Lakshmi Kumar <lakshmi.kumar@amd.com> * Package - Enable Lintian Support rpp (#633) * fix lintian errors * fix lintian overrides static error * lintian errors fixed * move lintian overrides into if deb check * use existing changelog. fix formatting * not installing lintian overrides. keeping original changelog name * remove overrides --------- Co-authored-by: Lakshmi Kumar <lakshmi.kumar@amd.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com> * Docs - Bump rocm-docs-core[api_reference] from 1.27.0 to 1.29.0 in /docs/sphinx (#638) Bumps [rocm-docs-core[api_reference]](https://github.com/ROCm/rocm-docs-core) from 1.27.0 to 1.29.0. - [Release notes](https://github.com/ROCm/rocm-docs-core/releases) - [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md) - [Commits](ROCm/rocm-docs-core@v1.27.0...v1.29.0) --- updated-dependencies: - dependency-name: rocm-docs-core[api_reference] dependency-version: 1.29.0 dependency-type: direct:production update-type: version-update:semver-minor ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com> * Test suite - Add QA pass/fail tests for F32 bit depth (#631) * Added golden outputs and resolved HOST backend * Updated bin files for median filter and resize crop mirror * Fix for median filter F32 QA * Updated bin files * Updated rcm review comments * Updated comments for rmn * Modified bitdepths and resolved review comments * Fix typo * resolve review comments --------- Co-authored-by: sampath117 <snehaa@multicorewareinc.com> Co-authored-by: HazarathKumarM <hazarathkumar@multicorewareinc.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com> Co-authored-by: Lakshmi Kumar <lakshmi.kumar@amd.com> * Test Suite - Error Code Capture for all tests (#635) * Updates to capture error code * Intialize RPP_SUCCESS as default value * Update the code to display error status as part of the C++ code execution * Update rpp_test_suite_common.h * Update utilities/test_suite/HIP/Tensor_audio_hip.cpp Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update utilities/test_suite/HIP/Tensor_image_hip.cpp Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update utilities/test_suite/HIP/Tensor_misc_hip.cpp Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update utilities/test_suite/HIP/Tensor_voxel_hip.cpp Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update utilities/test_suite/HOST/Tensor_audio_host.cpp Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update utilities/test_suite/HOST/Tensor_image_host.cpp Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update utilities/test_suite/HOST/Tensor_misc_host.cpp Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Update utilities/test_suite/HOST/Tensor_voxel_host.cpp Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Fixes for CI issues * Restore naming convention in voxel test suite * Fix compilation issues * Update the code to use func for funcName * Modify error message * Modify the print statements --------- Co-authored-by: Srihari-mcw <srihari@multicorewareinc.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com> * F16 variants - Update loads and stores to AVX2 - Group 5 (#637) * Updates for crop mirror normalize * Updated flip F16 rawC and load store modifications * Updated blend with AVX support for F16 bitdepth * Updated color cast with AVX support for F16 bitdepth * Remove empty lines * Update comments * Fix comment in common function --------- Co-authored-by: Srihari-mcw <srihari@multicorewareinc.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com> * Docs - Bump rocm-docs-core[api_reference] from 1.29.0 to 1.30.0 in /docs/sphinx (#640) Bumps [rocm-docs-core[api_reference]](https://github.com/ROCm/rocm-docs-core) from 1.29.0 to 1.30.0. - [Release notes](https://github.com/ROCm/rocm-docs-core/releases) - [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md) - [Commits](ROCm/rocm-docs-core@v1.29.0...v1.30.0) --- updated-dependencies: - dependency-name: rocm-docs-core[api_reference] dependency-version: 1.30.0 dependency-type: direct:production update-type: version-update:semver-minor ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * HOST and HIP - pinned buffers for respective API (#628) * Removed memcpy and used hipHostMalloc for allocation : blend * Removed memcpy and used hipHostMalloc for allocation : brightness * Removed memcpy and used hipHostMalloc for allocation : color cast * Removed memcpy and used hipHostMalloc for allocation : color twist * Removed memcpy and used hipHostMalloc for allocation : contrast * Removed memcpy and used hipHostMalloc for allocation : crop mirror normalize * Removed memcpy and used hipHostMalloc for allocation : Exposure * Removed memcpy and used hipHostMalloc for allocation : Gamma correction * Removed memcpy and used hipHostMalloc for allocation : gaussian filter * Removed memcpy and used hipHostMalloc for allocation : Noise * Removed memcpy and used hipHostMalloc for allocation : Non linear blend * Removed memcpy and used hipHostMalloc for allocation : Resize mirror normalize * Removed memcpy and used hipHostMalloc for allocation : Water * Added hipHostFree for all kernels in test suite * Added hipHostFree for all kernels in test suite * Removed memcpy and used hipHostMalloc for allocation : Flip, spatter, rcm, color temperature * Resolved copilot review comments * Updated version * Removed unused parameter * Updated version in cmakeList * removed the host to device mem copies for warp affine and rotate * Updated version * Removed comment * Updated Chnagelog file * Update patch version from 2.2.0 to 2.2.1 * Update CHANGELOG * Address copilot comments for HIP HOST consistent allocation * Documentation changes for updated memcpy changes * Update ricap outer API to use pinned memory and remove mem copy * Fix memory allocation and deallocation for permutationTensor * Update api/rppt_tensor_effects_augmentations.h Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Fix spelling of noiseProbability and saltProbability * Fix deallocation --------- Co-authored-by: HazarathKumarM <hazarathkumar@multicorewareinc.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com> Co-authored-by: Srihari-mcw <srihari@multicorewareinc.com> Co-authored-by: hmaddise <HazarathKumar.Maddisetty@amd.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> * Docs - Bump rocm-docs-core[api_reference] from 1.30.0 to 1.30.1 in /docs/sphinx (#643) Bumps [rocm-docs-core[api_reference]](https://github.com/ROCm/rocm-docs-core) from 1.30.0 to 1.30.1. - [Release notes](https://github.com/ROCm/rocm-docs-core/releases) - [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md) - [Commits](ROCm/rocm-docs-core@v1.30.0...v1.30.1) --- updated-dependencies: - dependency-name: rocm-docs-core[api_reference] dependency-version: 1.30.1 dependency-type: direct:production update-type: version-update:semver-patch ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * CMakelists - Add optional GPU targets (#641) * add optional gpu targets * add addiitonal gpu targets * Rename function - hip_exec_roi_converison_ltrb_to_xywh to hip_exec_roi_conversion_ltrb_to_xywh (#645) Co-authored-by: Srihari-mcw <srihari@multicorewareinc.com> * Docs - Update CHANGELOG.md (#646) Updates --------- Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: Abishek <52214183+r-abishek@users.noreply.github.com> Co-authored-by: Srihari-mcw <srihari@multicorewareinc.com> Co-authored-by: Lakshmi Kumar <lakshmi.kumar@amd.com> Co-authored-by: jonatluu <jonatluu@amd.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: sampath117 <snehaa@multicorewareinc.com> Co-authored-by: HazarathKumarM <hazarathkumar@multicorewareinc.com> Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com> Co-authored-by: hmaddise <HazarathKumar.Maddisetty@amd.com>
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.
F16 Load/Store updates for blend, color_cast, flip, crop_mirror_normalize.