Skip to content

Latest commit

 

History

History
338 lines (287 loc) · 18.8 KB

README.md

File metadata and controls

338 lines (287 loc) · 18.8 KB
Raw

AMD ROCm Performance Primitives (RPP) Test Suite

This repository contains four test suites for the AMD ROCm Performance Primitives (RPP) library: for image/voxel/audio/miscellaneous augmentations processing. These test suites can be used to validate the functionality and performance of the AMD ROCm Performance Primitives (RPP) image/voxel/audio/miscellaneous functionalities.

Test suite prerequisites

  • OpenCV 3.4.0/4.5.5

    • Install OpenCV prerequisites:

      sudo apt-get update
sudo -S apt-get -y --allow-unauthenticated install build-essential libgtk2.0-dev libavcodec-dev libavformat-dev libswscale-dev python-dev python-numpy
sudo -S apt-get -y --allow-unauthenticated install libtbb2 libtbb-dev libjpeg-dev libpng-dev libtiff-dev libdc1394-22-dev unzip wget

    • Download OpenCV 3.4.0 /4.5.5:

      wget https://github.com/opencv/opencv/archive/3.4.0.zip
unzip 3.4.0.zip
cd opencv-3.4.0/

      OR

      wget https://github.com/opencv/opencv/archive/4.5.5.zip
unzip 4.5.5.zip
cd opencv-4.5.5/

    • Install OpenCV:

      mkdir build
cd build
cmake -D WITH_GTK=ON -D WITH_JPEG=ON -D BUILD_JPEG=ON -D WITH_OPENCL=OFF -D WITH_OPENCLAMDFFT=OFF -D WITH_OPENCLAMDBLAS=OFF -D WITH_VA_INTEL=OFF -D WITH_OPENCL_SVM=OFF -D CMAKE_INSTALL_PREFIX=/usr/local ..
sudo -S make -j128 <Or other number of threads to use>
sudo -S make install
sudo -S ldconfig

  • Install TurboJpeg:

    sudo apt-get install nasm
sudo apt-get install wget
git clone -b 2.0.6.1 https://github.com/rrawther/libjpeg-turbo.git
cd libjpeg-turbo
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=/usr \
      -DCMAKE_BUILD_TYPE=RELEASE  \
      -DENABLE_STATIC=FALSE       \
      -DCMAKE_INSTALL_DOCDIR=/usr/share/doc/libjpeg-turbo-2.0.3 \
      -DCMAKE_INSTALL_DEFAULT_LIBDIR=lib  \
      ..
make -j$nproc
sudo make install

  • Libsndfile installation

    sudo apt-get update
sudo apt-get install libsndfile1-dev

  • Imagemagick

    sudo apt-get install imagemagick

  • Nifti-Imaging nifti_clib

    git clone [email protected]:NIFTI-Imaging/nifti_clib.git
cd nifti_clib
mkdir build
cd build
cmake ..
sudo make -j$nproc install

  • Openpyxl

    pip install openpyxl

Rpp Image Test Suite

The image test suite can be executed under 2 backend scenarios - (HOST/HIP):

  • HOST backend - (On a CPU with HOST backend)
  • HIP backend - (On a GPU with HIP backend)

Command Line Arguments (RPP Image Test Suite)

The image test suite accepts the following command line arguments:

  • input_path1: The path to the input folder 1. Default is $cwd/../TEST_IMAGES/three_images_mixed_src1
  • input_path2: The path to the input folder 2. Default is $cwd/../TEST_IMAGES/three_images_mixed_src2
  • case_start: The starting case number for the test range (0-91). Default is 0
  • case_end: The ending case number for the test range (0-91). Default is 91
  • test_type: The type of test to run (0 = Unit tests, 1 = Performance tests). Default is 0
  • case_list: A list of specific case numbers to run. Must be used in conjunction with --test_type
  • profiling: Run the tests with a profiler (YES/NO). Default is NO. This option is only available with HIP backend
  • qa_mode: Output images from tests will be compared with golden outputs - (0 / 1). Default is 0
  • decoder_type: Type of Decoder to decode the input data - (0 = TurboJPEG / 1 = OpenCV). Default is 0
  • num_runs: Specifies the number of runs for running the performance tests
  • preserve_output: preserves the output images or performance logs generated from the previous test suite run - (0 = remove output images or performance logs / 1 = preserve output images or performance logs). Default is 1
  • batch_size: Specifies the batch size to use for running tests. Default is 1

Running the Tests for HOST Backend (RPP Image Test Suite)

The test suite can be run with the following command:

python runTests.py --input_path1 <input_path1> --input_path2 <input_path2> --case_start <case_start> --case_end <case_end> --test_type <test_type>

Running the Tests for HIP Backend (RPP Image Test Suite)

The test suite can be run with the following command:

python runTests.py --input_path1 <input_path1> --input_path2 <input_path2> --case_start <case_start> --case_end <case_end> --test_type <test_type> --profiling <profiling>

Modes of operation (Rpp Image Test Suite)

  • QA mode (Unit tests) - Tolerance based PASS/FAIL tests for RPP HIP/HOST functionalities checking pixelwise match between C/SSE/AVX/HIP versions after comparison to preset golden outputs. Please note that QA mode is only supported with a batch size of 3. Note: QA mode is not supported for case 84 due to run-to-run variation of outputs.
python runTests.py --case_start 0 --case_end 91 --test_type 0 --qa_mode 1 --batch_size 3
  • QA mode (Performance tests) - Tolerance based PASS/FAIL tests for RPP HIP/HOST functionalities checking achieved improvement in performance percentage over BatchPD versions after comparison to a threshold percentage of improvement
python runTests.py --case_list 21 36 63 --test_type 1 --qa_mode 1 --batch_size 8 --num_runs 100
  • Unit test mode - Unit tests allowing users to pass a path to a folder containing images, to execute the desired functionality and variant once, report RPP execution wall time, save and view output images Note: For testcase 82(RICAP) Please use images of same resolution and Batchsize > 1 RICAP dataset path: rpp/utilities/test_suite/TEST_IMAGES/three_images_150x150_src1
python runTests.py --case_start 0 --case_end 91 --test_type 0 --qa_mode 0
  • Performance test mode - Performance tests that execute the desired functionality and variant 100 times by default, and report max/min/avg RPP execution wall time, or optionally, AMD rocprof kernel profiler max/min/avg time for HIP backend variants. Note: For testcase 82(RICAP) Please use images of same resolution and Batchsize > 1 RICAP dataset path: rpp/utilities/test_suite/TEST_IMAGES/three_images_150x150_src1
python runTests.py --case_start 0 --case_end 91 --test_type 1

To run the unit tests / performance tests for specific case numbers. please case use case_list parameter. Example as below

  • To run unittests for case numbers 0, 2, 4
python runTests.py --case_list 0 2 4 --test_type 0
  • To run performance tests for case numbers 0, 2, 4
python runTests.py --case_list 0 2 4 --test_type 1

To run performance tests with AMD rocprof kernel profiler for HIP backend variants. This will generate profiler times for HIP backend variants

python runTests.py --test_type 1 --profiling YES

Summary of features (RPP Image Test Suite)

The image test suite includes:

  • Unit tests that execute the desired functionality and variant once, report RPP execution wall time and save output images
  • Performance tests that execute the desired functionality and variant 100 times by default, and report max/min/avg RPP execution wall time, or optionally, AMD rocprof kernel profiler max/min/avg time for HIP backend variants.
  • Unit and Performance tests are included for three layouts - PLN1 (1 channel planar NCHW), PLN3 (3 channel planar NCHW) and PKD3 (3 channel packed/interrleaved NHWC).
  • Unit and Performance tests are included for various input/output bitdepths including U8/F32/F16/I8.
  • Support for pixelwise output referencing against golden outputs, and functionality validation checking, by tolerance-based pass/fail criterions for each variant. (Current support only for U8 variants)
  • Support for TurboJPEG and OpenCV decoder for decoding input images

RPP Voxel Test Suite

The 3D Voxel test suite can be executed under 2 backend scenarios - (HOST/HIP):

  • HOST backend - (On a CPU with HOST backend)
  • HIP backend - (On a GPU with HIP backend)

Command Line Arguments (RPP Voxel Test Suite)

The voxel test suite accepts the following command line arguments:

  • header_path: Path to the nii header
  • data_path: Path to the nii data file
  • case_start: The starting case number for the test range (0-38). Default is 0
  • case_end: The ending case number for the test range (0-38). Default is 4
  • test_type: The type of test to run (0 = Unit tests, 1 = Performance tests). Default is 0
  • case_list: A list of specific case numbers to run. Must be used in conjunction with --test_type
  • profiling: Run the tests with a profiler (YES/NO). Default is NO. This option is only available with HIP backend
  • qa_mode: Output images from tests will be compared with golden outputs - (0 / 1). Default is 0
  • num_runs: Specifies the number of runs for running the performance tests
  • preserve_output: preserves the output images or performance logs generated from the previous test suite run - (0 = remove output images or performance logs / 1 = preserve output images or performance logs). Default is 1
  • batch_size: Specifies the batch size to use for running tests. Default is 1

Running the Tests for HOST Backend (RPP Voxel Test Suite)

The test suite can be run with the following command:

python runVoxelTests.py --header_path <header_path> --data_path <data_path> --case_start <case_start> --case_end <case_end> --test_type <test_type>

Running the Tests for HIP Backend (RPP Voxel Test Suite)

The test suite can be run with the following command:

python runVoxelTests.py --header_path <header_path> --data_path <data_path> --case_start <case_start> --case_end <case_end> --test_type <test_type> --profiling <profiling>

Modes of operation (RPP Voxel Test Suite)

  • QA mode - Tolerance based PASS/FAIL tests for RPP HIP/HOST functionalities checking pixelwise match between C/SSE/AVX/HIP versions after comparison to preset golden outputs.
python runVoxelTests.py --case_start 0 --case_end 4 --test_type 0 --qa_mode 1 --batch_size 3
  • Unit test mode - Unit tests allowing users to pass a path to a folder containing nii fikes, to execute the desired functionality and variant once, report RPP execution wall time, save and view output images, gifs and nii files
python runVoxelTests.py --case_start 0 --case_end 4 --test_type 0 --qa_mode 0
  • Performance test mode - Performance tests that execute the desired functionality and variant 100 times by default, and report max/min/avg RPP execution wall time, or optionally, AMD rocprof kernel profiler max/min/avg time for HIP backend variants.
python runVoxelTests.py --case_start 0 --case_end 4 --test_type 1

To run the unit tests / performance tests for specific case numbers. please case use case_list parameter. Example as below

  • To run unittests for case numbers 0, 2, 4
python runVoxelTests.py --case_list 0 2 4 --test_type 0
  • To run performance tests for case numbers 0, 2, 4
python runVoxelTests.py --case_list 0 2 4 --test_type 1

To run performance tests with AMD rocprof kernel profiler for HIP backend variants. This will generate profiler times for HIP backend variants

python runVoxelTests.py --test_type 1 --profiling YES

Summary of features (RPP Voxel Test Suite)

The image test suite includes:

  • Unit tests that execute the desired functionality and variant once, report RPP execution wall time and save output images, gifs and nii files
  • Performance tests that execute the desired functionality and variant 100 times by default, and report max/min/avg RPP execution wall time, or optionally, AMD rocprof kernel profiler max/min/avg time for HIP backend variants.
  • Unit and Performance tests are included for three layouts - PLN1 (1 channel planar NCHW), PLN3 (3 channel planar NCHW) and PKD3 (3 channel packed/interrleaved NHWC).
  • Unit and Performance tests are included for one input/output bitdepth F32.
  • Support for pixelwise output referencing against golden outputs, and functionality validation checking, by tolerance-based pass/fail criterions for each variant.

References (RPP Voxel Test Suite)

RPP test suite uses a sample ".nii" image usage from the BraTS2020 Dataset at https://www.kaggle.com/code/rastislav/3d-mri-brain-tumor-segmentation-u-net/input as per the following Data Usage Agreement present at the dataset link above.

Data Usage Agreement / Citations: You are free to use and/or refer to the BraTS datasets in your own research, provided that you always cite the following three manuscripts:

  • B. H. Menze, A. Jakab, S. Bauer, J. Kalpathy-Cramer, K. Farahani, J. Kirby, et al. "The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS)", IEEE Transactions on Medical Imaging 34(10), 1993-2024 (2015) DOI: 10.1109/TMI.2014.2377694
  • S. Bakas, H. Akbari, A. Sotiras, M. Bilello, M. Rozycki, J.S. Kirby, et al., "Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features", Nature Scientific Data, 4:170117 (2017) DOI: 10.1038/sdata.2017.117
  • S. Bakas, M. Reyes, A. Jakab, S. Bauer, M. Rempfler, A. Crimi, et al., "Identifying the Best Machine Learning Algorithms for Brain Tumor Segmentation, Progression Assessment, and Overall Survival Prediction in the BRATS Challenge", arXiv preprint arXiv:1811.02629 (2018)

RPP Audio Test Suite

The audio test suite can be executed to validate the functionality and performance of the AMD ROCm Performance Primitives (RPP) audio library.

  • HOST backend - (On a CPU with HOST backend)
  • F32 Bit Depth

Command Line Arguments (RPP Audio Test Suite)

The audio test suite accepts the following command line arguments:

  • input_path: The path to the input folder. Default is $cwd/../TEST_AUDIO_FILES/eight_samples_single_channel_src1
  • case_start: The starting case number for the test range (0-0). Default is 0
  • case_end: The ending case number for the test range (0-0). Default is 0
  • test_type: The type of test to run (0 = QA tests, 1 = Performance tests). Default is 0
  • qa_mode: Output audio data from tests will be compared with golden outputs - (0 / 1). Default is 0
  • case_list: A list of specific case numbers to run. Must be used in conjunction with --test_type
  • num_runs: Specifies the number of runs for running the performance tests
  • preserve_output: preserves the output or performance logs generated from the previous test suite run - (0 = remove output or performance logs / 1 = preserve output or performance logs). Default is 1
  • batch_size: Specifies the batch size to use for running tests. Default is 1

Running the Tests for HOST Backend (RPP Audio Test Suite)

The test suite can be run with the following command:

python runAudioTests.py --input_path <input_path> --case_start <case_start> --case_end <case_end> --test_type <test_type>

Modes of operation (RPP Audio Test Suite)

  • QA mode - Tolerance based PASS/FAIL tests for RPP AUDIO HOST functionalities checking match between output and preset golden outputs. Please note that QA mode is only supported with a batch size of 3.
python runAudioTests.py --case_start 0 --case_end 0 --qa_mode 1 --batch_size 3
  • Performance test mode - Performance tests that execute the desired functionality and variant 100 times by default, and report max/min/avg RPP execution wall time.
python runAudioTests.py --case_start 0 --case_end 0 --test_type 1

To run the QA tests / performance tests for specific case numbers. please case use case_list parameter. Example as below

  • To run QA tests for case numbers 0, 1, 2
python runAudioTests.py --case_list 0 1 2 --qa_mode 1 --batch_size 3
  • To run performance tests for case numbers 0, 1, 2
python runAudioTests.py --case_list 0 1 2 --test_type 1

Summary of features (RPP Audio Test Suite)

The audio test suite includes:

  • Performance tests that execute the desired functionality and variant 100 times by default, and report max/min/avg RPP execution wall time.
  • QA and Performance tests are included for one input/output bitdepth F32.
  • Support for output referencing against golden outputs, and functionality validation checking, by tolerance-based pass/fail criterions for each variant.

RPP Miscellaneous Test Suite

The miscellaneous test suite can be executed to validate the functionality and performance of the AMD ROCm Performance Primitives (RPP) generic libraries that can process N-Dimensional inputs

  • HOST backend - (On a CPU with HOST backend)
  • HIP backend - (On a GPU with HIP backend)
  • F32 Bit Depth

Command Line Arguments (RPP Miscellaneous Test Suite)

The miscellaneous test suite accepts the following command line arguments:

  • case_start: The starting case number for the test range (1-1). Default is 1
  • case_end: The ending case number for the test range (1-1). Default is 1
  • test_type: The type of test to run (0 = QA tests, 1 = Performance tests). Default is 0
  • qa_mode: Output audio data from tests will be compared with golden outputs - (0 / 1). Default is 0
  • case_list: A list of specific case numbers to run. Must be used in conjunction with --test_type
  • num_dims: The number of dimensions of input. Default is 2
  • num_runs: Specifies the number of runs for running the performance tests
  • preserve_output: preserves the output or performance logs generated from the previous test suite run - (0 = remove output or performance logs / 1 = preserve output or performance logs). Default is 1
  • batch_size: Specifies the batch size to use for running tests. Default is 1
  • profiling: Run the tests with a profiler (YES/NO). Default is NO. This option is only available with HIP backend

Running the Tests for HOST Backend (RPP Miscellaneous Test Suite)

The test suite can be run with the following command:

python runMiscTests.py --case_start <case_start> --case_end <case_end> --test_type <test_type>

Modes of operation (RPP Miscellaneous Test Suite)

  • QA mode - Tolerance based PASS/FAIL tests for RPP MISC HOST functionalities checking match between output and preset golden outputs. Please note that QA mode is only supported with a batch size of 3 and num dims 2 or 3
python runMiscTests.py --case_start 1 --case_end 1 --qa_mode 1 --batch_size 3
  • Performance test mode - Performance tests that execute the desired functionality and variant 100 times by default, and report max/min/avg RPP execution wall time.
python runMiscTests.py --case_start 1 --case_end 1 --test_type 1

Summary of features (RPP Miscellaneous Test Suite)

The miscellaneous test suite includes:

  • Performance tests that execute the desired functionality and variant 100 times by default, and report max/min/avg RPP execution wall time.
  • QA and Performance tests are included for one input/output bitdepth F32.
  • Support for output referencing against golden outputs, and functionality validation checking, by tolerance-based pass/fail criterions for each variant.