Test suite - Multi-dimensional tensor tests support #644

r-abishek · 2025-12-03T02:49:43Z

Adds Multi-dimensional tensor tests support:

Introduces test support for multiple dims at the same time with num_dims_list parameter instead of num_dims.
QA test support extended for U8 and F32 bit depths for all kernels in misc test suite.
Minor fixes for QA match inside individual kernels.

Test suite - Multi-dimensional tensor tests support

Copilot

Pull request overview

This PR adds comprehensive support for multi-dimensional tensor testing across the test suite. Key changes include enabling simultaneous testing of multiple dimensions (2D, 3D, 4D) via a num_dims_list parameter, extending QA test coverage to U8 and F32 bit depths for all misc test kernels, and fixing various issues in QA validation and kernel implementations.

Key Changes:

Replaced single num_dims parameter with num_dims_list to support testing multiple dimensions in one run
Extended bit depth testing support for U8 and F32 across all misc kernels
Fixed QA comparison logic and various kernel-specific issues (normalize, concat, log operations)

Reviewed changes

Copilot reviewed 12 out of 57 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
utilities/test_suite/rpp_test_suite_misc.h	Updated path generation, data reading functions to support multiple bit depths; fixed QA comparison logic; added 4D support for normalize/transpose operations
utilities/test_suite/rpp_test_suite_common.h	Added new bit depth test modes (I8_TO_F32, I16_TO_F32) for extended test coverage
utilities/test_suite/common.py	Added bit depth dictionary entries for new I8_TO_F32 and I16_TO_F32 modes
utilities/test_suite/HOST/runMiscTests.py	Changed from single `num_dims` to `num_dims_list` parameter; updated bit depth configurations for unit and performance tests
utilities/test_suite/HOST/Tensor_misc_host.cpp	Refactored memory allocation and data handling to support new bit depths; improved function naming with bit depth suffixes
utilities/test_suite/HIP/runMiscTests.py	Mirrored HOST changes for HIP backend with `num_dims_list` support and bit depth updates
utilities/test_suite/HIP/Tensor_misc_hip.cpp	Updated device memory handling and QA comparison for multiple bit depths; fixed memory allocation sizing
src/modules/tensor/rppt_tensor_geometric_augmentations.cpp	Removed unnecessary layout mismatch check in concat operations
src/modules/tensor/hip/kernel/normalize.cpp	Fixed shared memory sizing, improved stddev computation, added memory zero-initialization
src/modules/tensor/cpu/kernel/normalize.cpp	Removed extraneous braces; added pixel clamping for U8 output; fixed code formatting
src/modules/tensor/cpu/kernel/concat.cpp	Fixed stride usage bugs in 2D/3D concat operations affecting different axis masks
src/include/common/cpu/rpp_cpu_simd_math.hpp	Replaced fast approximation with standard math library for inverse square root calculation

Comments suppressed due to low confidence (1)

utilities/test_suite/rpp_test_suite_misc.h:1

The ternary operator defaults to 1.0f when stdDev is zero, but this may not be the intended behavior for all normalization cases. Consider using 0.0f or adding a comment explaining why 1.0f is the correct default.

/*

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-12-03T02:51:20Z

utilities/test_suite/HOST/Tensor_misc_host.cpp

    // fill roi based on mode and number of dimensions
    Rpp32u *roiTensor = static_cast<Rpp32u *>(calloc(nDim * 2 * batchSize, sizeof(Rpp32u)));
-    Rpp32u *roiTensorSecond ;
+    Rpp32u *roiTensorSecond = nullptr;


roiTensorSecond is declared and initialized to nullptr, but later at line 97 it's used in fill_roi_values without being allocated memory for the CONCAT case. This will cause a segmentation fault.

Before fill_roi_values() is invoked, a calloc call allocates and zero-initializes the buffer. This guarantees that the function always operates on valid, initialized memory, preventing uninitialized-access or out-of-bounds-related bugs. Resolved

Copilot · 2025-12-03T02:51:20Z

utilities/test_suite/HIP/Tensor_misc_hip.cpp

    if(testCase == CONCAT)
+    {
+        iBufferSizeSecondInBytes = iBufferSizeSecond * get_size_of_data_type(srcDescriptorPtrNDSecond->dataType);
+        inputSecond = calloc(iBufferSizeSecond, get_size_of_data_type(srcDescriptorPtrNDSecond->dataType));


The calloc arguments are reversed. The first argument should be the size of each element, and the second should be the number of elements. This should be: calloc(get_size_of_data_type(srcDescriptorPtrNDSecond->dataType), iBufferSizeSecond) or use calloc(iBufferSizeSecondInBytes, 1) for consistency with other allocations.

Suggested change

inputSecond = calloc(iBufferSizeSecond, get_size_of_data_type(srcDescriptorPtrNDSecond->dataType));

inputSecond = calloc(iBufferSizeSecondInBytes, 1);

Copilot · 2025-12-03T02:51:21Z

utilities/test_suite/HIP/Tensor_misc_hip.cpp

-                if (BitDepthTestMode == U8_TO_U8 || BitDepthTestMode == F16_TO_F16 || BitDepthTestMode == F32_TO_F32 || BitDepthTestMode == I8_TO_I8)
-                    errorCodeCapture = rppt_transpose_gpu(d_input, srcDescriptorPtrND, d_output, dstDescriptorPtrND, permTensor, roiTensor, handle);
+                if(BitDepthTestMode == U8_TO_U8 || BitDepthTestMode == F16_TO_F16 || BitDepthTestMode == F32_TO_F32 || BitDepthTestMode == I8_TO_I8)
+                    rppt_transpose_gpu(d_input, srcDescriptorPtrND, d_output, dstDescriptorPtrND, permTensor, roiTensor, handle);


The return value from rppt_transpose_gpu is not captured in errorCodeCapture, unlike other function calls in this switch statement. This inconsistency could lead to errors being silently ignored.

Suggested change

rppt_transpose_gpu(d_input, srcDescriptorPtrND, d_output, dstDescriptorPtrND, permTensor, roiTensor, handle);

errorCodeCapture = rppt_transpose_gpu(d_input, srcDescriptorPtrND, d_output, dstDescriptorPtrND, permTensor, roiTensor, handle);

Copilot · 2025-12-03T02:51:21Z

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

 // Computes concatenation for 2D tensors (Supports Rpp32f and Rpp8u)
 template <typename T, typename SIMD_LOAD, typename SIMD_STORE>
-void concat_2D_tensor(T *srcPtr1, T *srcPtr2, SIMD_LOAD simd_load, SIMD_STORE simd_store, RpptGenericDescPtr srcDescPtr, RpptGenericDescPtr srcDescPtr1, T *dstPtr, RpptGenericDescPtr dstDescPtr, Rpp32u *dims, Rpp32u *strides,Rpp32u *dims1, Rpp32u *strides1, Rpp32u axisMask)
+void concat_2D_tensor(T *srcPtr1, T *srcPtr2, SIMD_LOAD simd_load, SIMD_STORE simd_store, RpptGenericDescPtr srcDescPtr, RpptGenericDescPtr srcDescPtr1, T *dstPtr, RpptGenericDescPtr dstDescPtr, Rpp32u *dims, Rpp32u *strides1, Rpp32u *dims1, Rpp32u *strides2, Rpp32u axisMask)


The parameter names are confusing: 'dims' and 'strides1' appear to be for the first source, while 'dims1' and 'strides2' appear to be for the second source. Consider renaming to 'dims1/strides1' and 'dims2/strides2' for clarity.

Copilot · 2025-12-03T02:51:21Z

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

+        Rpp8u *srcPtrRowTemp1 = srcPtrRow1;
        Rpp8u *dstPtrRow = dstPtr;


This line declares srcPtrRowTemp1 but it's immediately reassigned in the for loop at line 294. The declaration at line 290 is redundant and should be removed.

Suggested change

Rpp8u *srcPtrRowTemp1 = srcPtrRow1;

Rpp8u *dstPtrRow = dstPtr;

Rpp8u *dstPtrRow = dstPtr;

Copilot · 2025-12-03T02:51:22Z

src/include/common/cpu/rpp_cpu_simd_math.hpp

+inline Rpp32f rpp_math_inverse_square_root(Rpp32f x)
 {
-    __m128 X = _mm_set_ss(x);
-    __m128 tmp = _mm_rsqrt_ss(X);
-    Rpp32f y = _mm_cvtss_f32(tmp);
-    return y * (1.5f - x * 0.5f * y * y);
+    //raw C version of scalar
+    return 1.0f / std::sqrt(x);
 }


[nitpick] The comment 'raw C version of scalar' is unclear. This is actually C++ using std::sqrt. Consider clarifying: 'Standard library implementation using C++ std::sqrt' or removing the comment if it doesn't add value.

codecov · 2025-12-03T05:18:20Z

Codecov Report

❌ Patch coverage is 71.71717% with 28 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/modules/tensor/cpu/kernel/concat.cpp	65.75%	25 Missing ⚠️
src/include/common/cpu/rpp_cpu_simd_math.hpp	57.14%	3 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #644      +/-   ##
===========================================
+ Coverage    88.16%   89.85%   +1.69%     
===========================================
  Files          195      195              
  Lines        82723    82687      -36     
===========================================
+ Hits         72932    74295    +1363     
+ Misses        9791     8392    -1399

Files with missing lines	Coverage Δ
src/modules/tensor/cpu/kernel/normalize.cpp	`84.33% <100.00%> (+36.65%)`	⬆️
src/modules/tensor/hip/kernel/normalize.cpp	`99.20% <100.00%> (+64.36%)`	⬆️
...les/tensor/rppt_tensor_geometric_augmentations.cpp	`96.69% <ø> (+0.12%)`	⬆️
src/include/common/cpu/rpp_cpu_simd_math.hpp	`46.17% <57.14%> (-0.17%)`	⬇️
src/modules/tensor/cpu/kernel/concat.cpp	`85.25% <65.75%> (+42.19%)`	⬆️

... and 12 files with indirect coverage changes

🚀 New features to boost your workflow:

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

Fix copilot comments for misc test suite updates

Srihari-mcw and others added 7 commits October 29, 2025 21:17

Changes to kernels for QA fixes

664d8b1

Fixes to test suite

a1e00d8

Updated bin files with QA support

9ad4982

Add additional comments

4826ccd

Rename function to rpp_math_inverse_square_root

360996c

Merge branch 'current_develop' into misc_test_suite_changes

983dac9

Merge pull request #514 from Srihari-mcw/misc_test_suite_changes

426b2f4

Test suite - Multi-dimensional tensor tests support

r-abishek requested a review from Copilot December 3, 2025 02:49

r-abishek added the ci:precheckin label Dec 3, 2025

Copilot AI reviewed Dec 3, 2025

View reviewed changes

Srihari-mcw and others added 2 commits December 3, 2025 17:40

Fix copilot comments for misc test suite

efb413b

Merge pull request #542 from Srihari-mcw/misc_test_suite_copilot

af29985

Fix copilot comments for misc test suite updates

LakshmiKumar23 requested review from LakshmiKumar23 and rrawther December 8, 2025 21:26

LakshmiKumar23 assigned kiritigowda Dec 8, 2025

Merge branch 'develop' into ar/test_suite_upgrade_11_misc_suite

cd2556a

	inputSecond = calloc(iBufferSizeSecond, get_size_of_data_type(srcDescriptorPtrNDSecond->dataType));
	inputSecond = calloc(iBufferSizeSecondInBytes, 1);

	rppt_transpose_gpu(d_input, srcDescriptorPtrND, d_output, dstDescriptorPtrND, permTensor, roiTensor, handle);
	errorCodeCapture = rppt_transpose_gpu(d_input, srcDescriptorPtrND, d_output, dstDescriptorPtrND, permTensor, roiTensor, handle);

		Rpp8u *srcPtrRowTemp1 = srcPtrRow1;
		Rpp8u *dstPtrRow = dstPtr;

Test suite - Multi-dimensional tensor tests support #644

Are you sure you want to change the base?

Test suite - Multi-dimensional tensor tests support #644

Uh oh!

Conversation

r-abishek commented Dec 3, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

Srihari-mcw Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

Srihari-mcw Dec 4, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

Srihari-mcw Dec 4, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

Srihari-mcw Dec 4, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

Srihari-mcw Dec 4, 2025

Choose a reason for hiding this comment

Uh oh!

Copilot AI Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

Srihari-mcw Dec 4, 2025

Choose a reason for hiding this comment

Uh oh!

codecov bot commented Dec 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov bot commented Dec 3, 2025 •

edited

Loading