RPP JPEG Compression Distortion HOST by r-abishek · Pull Request #531 · ROCm/rpp

r-abishek · 2025-03-26T06:29:11Z

Adds JPEG Compression Distortion augmentation for HOST on U8/F32/F16/I8.
Adds relevant unit/perf/QA tests.

Add inverse DCT function Add ycbCr to RGB conversion Add test suite support

RPP Jpeg Compression distortion Tensor Support

Fix CI failures - JPEG HOST

Jpeg compression distortion - resolved review comments

Copilot

Pull Request Overview

This PR introduces a new JPEG Compression Distortion augmentation for the HOST backend, including support for U8, F32, F16, and I8 data types, along with corresponding unit, performance, and QA tests.

Added new augmentation mappings and enum entries across header and source files.
Implemented the jpeg_compression_distortion functions for various data types in tensor geometric augmentation modules.
Updated test suites and API documentation to integrate and validate the new augmentation.

Reviewed Changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
utilities/test_suite/rpp_test_suite_image.h	Added mapping entry and enum for JPEG Compression Distortion.
utilities/test_suite/common.py	Updated augmentation map with the new JPEG compression case.
utilities/test_suite/HOST/runImageTests.py	Adjusted the test range to include the new augmentation.
utilities/test_suite/HOST/Tensor_image_host.cpp	Introduced a new test case branch for JPEG Compression Distortion.
src/modules/tensor/rppt_tensor_geometric_augmentations.cpp	Added the main JPEG compression distortion function for HOST.
src/include/tensor/host_tensor_executors.hpp	Declared new function prototypes for the augmentation methods.
src/include/common/cpu/rpp_cpu_simd_math.hpp	Added a helper function for rounding away from zero.
src/include/common/cpu/rpp_cpu_simd_load_store.hpp	Included new load/store functions handling larger packed data blocks.
api/rppt_tensor_geometric_augmentations.h	Provided API documentation for the new augmentation function.
CHANGELOG.md	Updated the changelog to include JPEG Compression Distortion.

api/rppt_tensor_geometric_augmentations.h

rrawther

I have added few comments. Let's discuss the rest over a call

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

RPP JPEG Compression distortion HOST - review comments resolution

codecov · 2025-04-23T01:03:42Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #531      +/-   ##
===========================================
+ Coverage    54.82%   55.56%   +0.74%     
===========================================
  Files          293      294       +1     
  Lines       118516   120606    +2090     
===========================================
+ Hits         64972    67007    +2035     
- Misses       53544    53599      +55

Files with missing lines	Coverage Δ
src/include/common/cpu/rpp_cpu_simd_load_store.hpp	`92.60% <100.00%> (+0.55%)`	⬆️
.../tensor/cpu/kernel/jpeg_compression_distortion.cpp	`95.88% <ø> (ø)`
...les/tensor/rppt_tensor_geometric_augmentations.cpp	`91.76% <100.00%> (+0.18%)`	⬆️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

utilities/test_suite/HOST/Tensor_image_host.cpp

utilities/test_suite/HOST/runImageTests.py

rrawther · 2025-04-24T19:32:05Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+    temp[6] = x[2] - x[5];
+    temp[7] = x[4] - x[3];
+
+    temp[8] = temp[0] + temp[3];


please combine temp[0] and temp[3] computation here and use only T temp[4] like
temp[8] = x[0] + x[7] + x[3] + x[4];

rrawther · 2025-04-24T22:15:20Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+            if (row < colLimit && col < rowLimit)
+            {
+                Rpp32s idx = srcRowIdx + col * wStride;
+                if constexpr (std::is_same<T, Rpp32f>::value || std::is_same<T, Rpp16f>::value)


I think if we have a way to combine this with the formula it is better code
mul_factor = 1.0 * mul_table(rgb_data_type) , then use it in line 354

rrawther

Added few comments

rrawther · 2025-04-24T22:19:43Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+                        r *= ONE_OVER_255;
+                        g *= ONE_OVER_255;
+                        b *= ONE_OVER_255;
+                        dstPtrR[dstIdx] = static_cast<T>((r < 0.0f) ? 0.0f : (r > 1.0f) ? 1.0f : r);


check if std::max std::min combo is better. Should use an inline clamp function

Changed to an inline function, and computing the clamp function using conditional operators is faster than using the combination of std::min and std::max.

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

RPP JPEG compression HOST : Resolve review comments

kiritigowda · 2025-04-30T17:35:07Z

@r-abishek - needs conflicts resolved

r-abishek · 2025-04-30T20:00:38Z

@kiritigowda Resolved. Docs CI failing to pull commit.

rrawther

Added few more comments

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

rrawther · 2025-04-30T22:48:34Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+            Rpp32f avgB = (b[id1] + b[id2] + b[id3] + b[id4]) / 4.0f;
+
+            // Convert to Cb/Cr and center around 0
+            Rpp32s chromaIdx = (row / 2) * 8 + (col / 2);


please use ( >> 1) instead of /2

rrawther · 2025-04-30T22:50:29Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+        for (Rpp32s col = 0; col < 16; col += 2)
+        {
+            Rpp32s id1 = (row * 16) + col;
+            Rpp32s id2 = (row * 16) + col + 1;


please change it to id2 = id1 + 1, also u can use in place in line#392 and get rid of id2. Same for id4

rrawther · 2025-04-30T22:50:52Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+            Rpp32s id3 = (row + 1) * 16 + col;
+            Rpp32s id4 = (row + 1) * 16 + col + 1;
+
+            Rpp32f avgR = (r[id1] + r[id2] + r[id3] + r[id4]) / 4.0f;


change /4.0 to * 0.25

rrawther · 2025-04-30T22:53:52Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+
+                    // Convert YCbCr to RGB
+                    Rpp32f yVal = y[yIdx] + 128.0f;
+                    yVal = clamp(yVal, 0.0f, 255.0f);


why is this required. Please remove

removed the clamp

rrawther · 2025-04-30T22:54:35Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+                    Rpp32f r = yVal + 1.402f * currCr;
+                    Rpp32f g = yVal - 0.344136f * currCb - 0.714136f * currCr;
+                    Rpp32f b = yVal + 1.772f * currCb;
+


I suggest clamping values here so it is int clamp between 0 and 255

moved the all the clamps

rrawther · 2025-04-30T22:54:52Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+                        r *= ONE_OVER_255;
+                        g *= ONE_OVER_255;
+                        b *= ONE_OVER_255;
+                        dstPtrR[dstIdx] = static_cast<T>(clamp(r, 0.0f, 1.0f));


remove clamp from here

rrawther · 2025-04-30T22:55:11Z

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

+                        r -= 128;
+                        g -= 128;
+                        b -= 128;
+                        dstPtrR[dstIdx] = static_cast<T>(clamp(r, -128.0f, 127.0f));


remove clamp from here

src/modules/tensor/cpu/kernel/jpeg_compression_distortion.cpp

src/include/common/cpu/rpp_cpu_simd_math.hpp

rrawther

@HazarathKumarM : Are all the comments resolved for this PR?

HazarathKumarM · 2025-05-14T14:59:27Z

@HazarathKumarM : Are all the comments resolved for this PR?

yes Rajy

HazarathKumarM and others added 20 commits January 9, 2025 03:57

Intial implementation-JPEG compression distortion

2274e01

Adds Zig Zag Ordering

4b9fb05

Remove Zig Zag ordering

29fedad

Add inverse DCT function Add ycbCr to RGB conversion Add test suite support

Merge branch 'r-abishek:master' into hk/jpeg_distortion

24a7617

Add test suite support and PLN3-PLN3 implementation

d7b911e

Add PKD3 implementation

fa9106c

Add F32 bitdepth support

f713be5

clean up the code and added new helper functions

d7bfc7d

update documentation of Jpeg compression distortion

2d56ee5

Merge remote-tracking branch 'develop' into hk/jpeg_distortion

0d5c0a7

address review comments

94a7c2d

Address review comments

19039b1

Merge remote-tracking branch 'develop' into hk/jpeg_distortion

fb35df7

Fix build errors

e95cb30

cleanup

c17d01f

cleanup and remove trailing spaces

9ae00be

Add golden outputs and remove the trailing spaces

0d356ea

resolve review comments

738da5d

Resolve review comments

c89bc46

Merge pull request #408 from HazarathKumarM/hk/jpeg_distortion

4778243

RPP Jpeg Compression distortion Tensor Support

r-abishek added the ci:precheckin label Mar 26, 2025

HazarathKumarM and others added 2 commits March 26, 2025 07:08

add avx_px6 and avx_px7 variables

b009463

Merge pull request #418 from HazarathKumarM/hk/jpeg_distortion

ea66e54

Fix CI failures - JPEG HOST

kiritigowda requested a review from AryanSalmanpour March 26, 2025 17:05

kiritigowda self-assigned this Mar 26, 2025

kiritigowda and others added 4 commits March 26, 2025 14:28

Merge branch 'develop' into ar/opt_jpeg_distortion_host

1ecea49

resolve review comments

6403a7d

fix paranthesis checks

e5a6d4f

Merge pull request #419 from HazarathKumarM/hk/jpeg_distortion

73f51a3

Jpeg compression distortion - resolved review comments

kiritigowda requested a review from Copilot April 1, 2025 04:18

Merge branch 'develop' into ar/opt_jpeg_distortion_host

66e4c29

kiritigowda requested review from Copilot and rrawther April 19, 2025 17:45

Copilot AI reviewed Apr 19, 2025

View reviewed changes

api/rppt_tensor_geometric_augmentations.h Outdated Show resolved Hide resolved

rrawther requested changes Apr 21, 2025

View reviewed changes

HazarathKumarM and others added 2 commits April 21, 2025 19:11

resolved review comments

54b01ab

Merge pull request #423 from HazarathKumarM/hk/jpeg_distortion_host

1303c61

RPP JPEG Compression distortion HOST - review comments resolution

rrawther reviewed Apr 24, 2025

View reviewed changes

utilities/test_suite/HOST/Tensor_image_host.cpp Show resolved Hide resolved

rrawther reviewed Apr 24, 2025

View reviewed changes

utilities/test_suite/HOST/runImageTests.py Outdated Show resolved Hide resolved

rrawther reviewed Apr 24, 2025

View reviewed changes

rrawther requested changes Apr 24, 2025

View reviewed changes

HazarathKumarM and others added 2 commits April 25, 2025 18:24

resolve review comments

20e2b32

Merge pull request #426 from HazarathKumarM/hk/jpeg_distortion_host

10306c0

RPP JPEG compression HOST : Resolve review comments

Merge branch 'develop' into ar/opt_jpeg_distortion_host

3f67c6d

rrawther requested changes Apr 30, 2025

View reviewed changes

HazarathKumarM and others added 3 commits May 5, 2025 17:24

Resolve review comments

0113cf2

Merge branch 'develop' into ar/opt_jpeg_distortion_host

dfc0571

Merge branch 'develop' into ar/opt_jpeg_distortion_host

92bedc6

rrawther reviewed May 5, 2025

View reviewed changes

src/include/common/cpu/rpp_cpu_simd_math.hpp Outdated Show resolved Hide resolved

HazarathKumarM and others added 2 commits May 6, 2025 04:45

Remove redundant functions

c49c20e

Merge pull request #430 from HazarathKumarM/hk/jpeg_distortion_host

16b22d0

rrawther reviewed May 14, 2025

View reviewed changes

Merge branch 'develop' into ar/opt_jpeg_distortion_host

dcd55b4

rrawther approved these changes May 29, 2025

View reviewed changes

kiritigowda merged commit 4e65665 into ROCm:develop May 30, 2025
22 of 24 checks passed

Conversation

r-abishek commented Mar 26, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Uh oh!

rrawther left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

codecov bot commented Apr 23, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rrawther left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

kiritigowda commented Apr 30, 2025

Uh oh!

r-abishek commented Apr 30, 2025

Uh oh!

rrawther left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

rrawther left a comment

Choose a reason for hiding this comment

Uh oh!

HazarathKumarM commented May 14, 2025

Uh oh!

Uh oh!

Reviewers

codecov bot commented Apr 23, 2025 •

edited

Loading