Quantify FFT numerical error

.gitattributes

@@ -0,0 +1,13 @@
+csv/xpa2.csv filter=lfs diff=lfs merge=lfs -text
+csv/acars.csv filter=lfs diff=lfs merge=lfs -text
+csv/ads-b.csv filter=lfs diff=lfs merge=lfs -text
+csv/flex.csv filter=lfs diff=lfs merge=lfs -text
+csv/jorn.csv filter=lfs diff=lfs merge=lfs -text
+csv/lofar-20m-x.csv filter=lfs diff=lfs merge=lfs -text
+csv/test-reverse.csv filter=lfs diff=lfs merge=lfs -text
+csv/xpa1.csv filter=lfs diff=lfs merge=lfs -text
+wav/lofar-20m-x.wav filter=lfs diff=lfs merge=lfs -text
+wav/xpa1-8000.wav filter=lfs diff=lfs merge=lfs -text
+wav/acars1-iq-48000.wav filter=lfs diff=lfs merge=lfs -text
+wav/dvb-t-iq-5333333.wav filter=lfs diff=lfs merge=lfs -text
+wav/flex-iq-39062.wav filter=lfs diff=lfs merge=lfs -text

.gitlab-ci.yml

@@ -1,23 +1,77 @@
+image: $CI_REGISTRY_IMAGE/ci-build-runner:$CI_COMMIT_REF_SLUG
+variables:
+  GIT_SUBMODULE_STRATEGY: recursive
+
 stages:
+  - prepare
   - build-lib
   - build
   - test
+  - release
+
+include:
+  - template: Security/SAST.gitlab-ci.yml
+  - template: Security/Dependency-Scanning.gitlab-ci.yml
+  - template: Security/Secret-Detection.gitlab-ci.yml
+
+default:
+  before_script:
+    - mkdir -p build
+    - cd build || exit
+  cache:
+    paths:
+      - "build/*"
+
+trigger-prepare:
+  stage: prepare
+  trigger:
+    strategy: depend
+    include: .prepare.gitlab-ci.yml
+
+prepare-release:
+  stage: prepare
+  image: alpine:latest
+  rules:
+    - if: '$CI_COMMIT_TAG && $CI_COMMIT_REF_PROTECTED == "true"'
+  before_script:
+    - apk add curl jq
+  script:
+    - 'curl -H "PRIVATE-TOKEN: $CI_API_TOKEN" "$CI_API_V4_URL/projects/$CI_PROJECT_ID/repository/changelog?version=$CI_COMMIT_TAG" | jq -r .notes > release_notes.md'
+    - cat release_notes.md
+  artifacts:
+    paths:
+      - release_notes.md
+
 build-library:
   stage: build-lib
+  needs:
+    - trigger-prepare
   script:
-    - cd $CI_PROJECT_DIR
-    - mkdir -p build
-    - cd build || exit
     - cmake ..
     - make oCLFFTsup -j $(nproc)
   artifacts:
     paths:
       - build/lib/liboCLFFTsup.so
-  cache:
+
+build-lookup:
+  stage: build-lib
+  needs:
+    - trigger-prepare
+  script:
+    - cmake ..
+    - make play-lookup play-lookup-fp32
+    - ./playground/play-lookup > ../oclfft/ard-ocl/src/lookup.cl
+    - ./playground/play-lookup > ../oclfft/bit-ocl/src/lookup.cl
+    - ./playground/play-lookup > ../oclfft/flk-ocl/src/lookup.cl
+    - ./playground/play-lookup-fp32 > ../oclfft/flk-ocl-fp32/src/lookup.cl
+  artifacts:
     paths:
-      - "build/*"
+      - oclfft/**/lookup.cl
+
 build-documents:
   stage: build
+  needs:
+    - trigger-prepare
   script:
     - cd $CI_PROJECT_DIR/proposal || exit
     - texi2pdf proposal.tex
@@ -26,40 +80,70 @@ build-documents:
   artifacts:
     paths:
       - "*/*.pdf"
+
 build-documentation:
   stage: build
+  needs:
+    - trigger-prepare
   script:
-    - cd $CI_PROJECT_DIR
-    - mkdir -p build
-    - cd build || exit
     - cmake -DENABLE_DOXYGEN=on ..
     - make doxygen
   artifacts:
     paths:
       - "docs/*"
+
 build-binaries:
   stage: build
+  needs:
+    - trigger-prepare
+    - build-lookup
+    - build-library
   script:
-    - cd $CI_PROJECT_DIR || exit
-    - mkdir -p build
-    - cd build || exit
     - cmake -DENABLE_DOXYGEN=off ..
-    - make -j $(nproc)
+    - make ard-ocl
+    - make ard-omp
+    - make ard-seq
+    - make bit-ocl
+    - make dft-seq
+    - make dft-seq-fp32
+    - make flk-ocl
+    - make flk-ocl-fp32
+    - make numerical-analysis
+    - make play-objcopy
+    - make play-example
+    - make play-reversegen
   artifacts:
     paths:
       - "*/**/play-objcopy"
       - "*/**/play-example"
-      - "*/**/testexample"
-  cache:
-    paths:
-      - "build/*"
+
 test:
   stage: test
+  needs:
+    - trigger-prepare
+    - build-lookup
+    - build-library
+    - build-binaries
+  variables:
+    CTEST_OUTPUT_ON_FAILURE: 1
   script:
-    - cd $CI_PROJECT_DIR || exit
-    - mkdir -p build
-    - cd build || exit
-    - make test
-  cache:
-    paths:
-      - "build/*"
+    - make run_tests
+
+release_job:
+  stage: release
+  image: registry.gitlab.com/gitlab-org/release-cli:latest
+  environment: gitlab-releases
+  needs:
+    - job: prepare-release
+      artifacts: true
+  rules:
+    - if: '$CI_COMMIT_TAG && $CI_COMMIT_REF_PROTECTED == "true"'
+  before_script:
+    - echo "Creating release ${CI_COMMIT_TAG}"
+  script:
+    - cat release_notes.md
+  release:
+    name: '${CI_COMMIT_TAG}'
+    description: release_notes.md
+    tag_name: '${CI_COMMIT_TAG}'
+    ref: '$CI_COMMIT_SHA'

.prepare.gitlab-ci.yml

@@ -0,0 +1,21 @@
+stages:
+  - build
+
+build_ci_runner_image:
+  stage: build
+  image: docker:latest
+  tags:
+    - dind
+  before_script:
+    - apk add yq
+  script:
+    - docker login -u $CI_REGISTRY_USER -p $CI_REGISTRY_PASSWORD $CI_REGISTRY
+    - docker context create my-builder
+    - docker buildx create my-builder --driver docker-container --use
+    - docker buildx build --push -t $CI_REGISTRY_IMAGE/ci-build-runner:$CI_COMMIT_REF_SLUG --cache-to type=registry,ref=$CI_REGISTRY_IMAGE/ci-build-runner-cache:$CI_COMMIT_REF_SLUG,mode=max --cache-from type=registry,ref=$CI_REGISTRY_IMAGE/ci-build-runner-cache:$CI_COMMIT_REF_SLUG  docker/
+
+
+
+workflow:
+  rules:
+    - when: always

CMakeLists.txt

@@ -14,7 +14,7 @@
 # You should have received a copy of the GNU General Public License
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
-CMAKE_MINIMUM_REQUIRED(VERSION 3.0)
+CMAKE_MINIMUM_REQUIRED(VERSION 3.10)
 PROJECT(oCLFFT)
 
 # 3.0 is minimum version as it is the first version to include FindOpenCL
@@ -43,16 +43,16 @@ use_cxx_warning_pedantic()
 # Configurable options and defaults
 OPTION(ENABLE_TESTS "Enable unit tests" ON)
 OPTION(ENABLE_CODECOV "Measure code coverage" OFF)
-OPTION(ENABLE_DOXYGEN "Generate documentation" ON)
+OPTION(ENABLE_DOXYGEN "Generate documentation" OFF)
 OPTION(ENABLE_CI "Enable continuous integration" OFF)
 
-include_directories(${CMAKE_SOURCE_DIR}/lib/include)
+#include_directories(${CMAKE_SOURCE_DIR}/lib/include)
 
 # Required packages
 set(CMAKE_THREAD_PREFER_PTHREAD TRUE)
 set(THREADS_PREFER_PTHREAD_FLAG TRUE)
 find_package(Threads REQUIRED)
-find_package(OpenCL 2.0 REQUIRED)
+find_package(OpenCL 2.2 REQUIRED)
 
 # Include Commonly required headers
 # include_directories(${ZeroMQ_INCLUDE_DIR})
@@ -89,8 +89,8 @@ if(CMAKE_BUILD_TYPE STREQUAL "Debug")
 	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g3 -O0")
 elseif(CMAKE_BUILD_TYPE STREQUAL "Release")
 	message("${PRJ_PRX}: Enabling extensive optimizations")
-	set(CMAKE_CXX_FLAGS 
-		"${CMAKE_CXX_FLAGS} -Ofast -ffast-math -ftree-vectorize")
+#	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native -Ofast -ffast-math -ftree-vectorize")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native -O3 -ftree-vectorize")
 endif()
 
 # Commonly shared library

README.md

@@ -5,6 +5,10 @@ Project for University of Amsterdam (UvA) performance engineering course,
 boilerplate CMake files are from
 [a previous project](https://github.com/hexoxide/O2-Balancer2)
 
+![Performance results on Ryzen 5900X & Radeon 5700XT](measurements/results-march-2024-cpu-vs-gpu.png)
+
+![Numerical error comparison](numerical-analysis/fft-numerical-analysis-2025-10-30.png)
+
 ### Directory structure
 
 * cmake - small cmake snippets to enable various features
@@ -13,6 +17,7 @@ boilerplate CMake files are from
 * lib - support library with functions and definitions
 * measurements - performance evaluation scripts as well as measured values
 * midterm - LaTeX files for the presentation at the first half of the course
+* numerical-analysis - Results and measurements to determine numerical error
 * playground - small toy examples or other experiments
 * [python](python/README.md) - python scripts to aid in visualization or measurements
 * proposal - LaTeX source for project proposal
@@ -23,31 +28,44 @@ boilerplate CMake files are from
 
 ### Implementations
 
-| Executable | Hardware Target   | Base Algorithm | Properties                    |
-|------------|-------------------|----------------|-------------------------------|
-| ard-seq    | CPU Single Thread | Cooley-Tukey   | Bit-reversal in-place radix-2 |
-| dft-seq    | CPU Single Thread | DFT            |                               |
-| ard-omp    | CPU OpenMP        | Cooley-Tukey   | Bit-reversal in-place radix-2 |
-| ard-ocl    | GPU               | Cooley-Tukey   | Slow bit-reversal, fft lookup |
-| bit-ocl    | GPU               | Cooley-Tukey   | 2D bit-reversal, fft lookup   |
-| flk-ocl    | GPU               | Cooley-Tukey   | 2D bit-reversal, fast lookup  |
+| Executable         | Hardware Target   | Base Algorithm | Precision    | Properties                                                 |
+|--------------------|-------------------|----------------|--------------|------------------------------------------------------------|
+| ard-seq            | CPU Single Thread | Cooley-Tukey   | 64bit double | Bit-reversal in-place radix-2                              |
+| dft-seq            | CPU Single Thread | DFT            | 64bit double |                                                            |
+| dft-seq-fp32       | CPU Single Thread | DFT            | 32bit float  |                                                            |
+| ard-omp            | CPU OpenMP        | Cooley-Tukey   | 64bit double | Bit-reversal in-place radix-2                              |
+| ard-ocl            | GPU               | Cooley-Tukey   | 64bit double | Slow bit-reversal, fft lookup                              |
+| bit-ocl            | GPU               | Cooley-Tukey   | 64bit double | 2D bit-reversal, fft lookup                                |
+| flk-ocl            | GPU               | Cooley-Tukey   | 64bit double | 2D bit-reversal, fast lookup                               |
+| flk-ocl-fp32       | GPU               | Cooley-Tukey   | 32bit float  | 2D bit-reversal, fast lookup, optimized local groups       |
+| numerical-analysis | CPU/GPU           | N.A            | arbitrary    | Arbitrary precision MPFR FFR, compute L1,L2 and Linf norms |
 
 #### Dependencies
 
-* cmake 3.0 or higher
-* cmake 3.9 or higher (ard-omp)
-* doxygen (documentation)
+* cmake 3.10 or higher
 * boost 1.32 or higher
 * boost 1.53 or higher (unit tests)
+* OpenCL 2.0 or higher
+* OpenCL C++ headers
+* MPFR
+* MPFR C++, MPREAL (rms-error)
+* doxygen (documentation)
 * python 3.x (python)
 * virtualenv (python)
 
 #### Setup
 
+Initialize lfs:
+
+```bash
+git lfs install
+git lfs checkout
+```
+
 Python environment:
 
 ```bash
-virtualenv -p python3 python/e
+virtualenv -p python3 python/
 cd python
 source bin/activate
 pip install -r requirements.txt
@@ -58,11 +76,14 @@ Generating lookup tables:
 ```bash
 mkdir build
 cd build
-cmake ..
+cmake -DCMAKE_BUILD_TYPE=Release ..
 make play-lookup
 ./playground/play-lookup > ../oclfft/ard-ocl/src/lookup.cl
 ./playground/play-lookup > ../oclfft/bit-ocl/src/lookup.cl
 ./playground/play-lookup > ../oclfft/flk-ocl/src/lookup.cl
+make play-lookup-fp32
+./playground/play-lookup-fp32 > ../oclfft/flk-ocl-fp32/src/lookup.cl
+make
 ```
 
 #### Licensing
@@ -81,13 +102,24 @@ A few examples of files licensed under different authors:
 
 * [Emebedding source files into binaries](https://www.linuxjournal.com/content/embedding-file-executable-aka-hello-world-version-5967)
 * [OpenCL, SyCL and SPIR-V progress 2016](https://www.youtube.com/watch?v=TYp1d6yzHUQ)
+* [RDNA whitepaper](https://www.amd.com/system/files/documents/rdna-whitepaper.pdf)
+* Arbitrary Precision Arithmetic (FFTs)
+  * [Python Arbitrary Precision FFT](https://github.com/dbstein/apfft/tree/master)
+  * [C++ Arbitrary Precision Arithmetic](https://github.com/anxieuse/Arbitrary-precision-arithmetic/tree/main)
+  * [BenchFFT](https://fftw.org/benchfft/)
+  * [RRMSE](https://stats.stackexchange.com/questions/413209/is-there-something-like-a-root-mean-square-relative-error-rmsre-or-what-is-t#413217)
 
 #### Snippets
 
 ```bash
-../cmake-build-debug/oclfft/bit-ocl/bit-ocl -f ../csv/ads-b.csv -s 262144  -o real | python evaluate-error.py
+cd python
+../cmake-build-debug/oclfft/flk-ocl-fp32/flk-ocl-fp32 -w 256 -f ../csv/lofar-20m-x.csv -s 4194304  -o real | python evaluate-error.py
 ```
 
 ```bash
- /opt/rocm/bin/rocprof --obj-tracking on --hsa-trace binary
-```
+/opt/rocm/bin/rocprofv3 --obj-tracking on --hsa-trace binary
+rocprofv2 --sys-trace --kernel-trace --plugin perfetto -o results.json binary
+rocprofv3 -r -s --output-format pftrace -i ../counters.txt -- oclfft/flk-ocl-fp32/flk-ocl-fp32 -w 256 -f ../csv/lofar-20m-x.csv -s 4194304 -o TIME
+```
+
+Open pftrace in https://ui.perfetto.dev