Merge branch 'development/speed-tests-tools' into 'development/ranges'

Non-regression speed test tools

See merge request !38

.gitlab-ci.yml

@@ -12,7 +12,7 @@ build-doc-images:
     stage: build
     script:
         - mkdir build && cd build
-        - conan install .. -pr buildfarm
+        - conan install .. -pr buildfarm --build=missing
         - cmake -DPYLENE_BUILD_BENCHMARKS=NO .. -DCMAKE_BUILD_TYPE=$PYLENE_CONFIGURATION
         - cmake --build .
                 --config Release
@@ -29,16 +29,18 @@ build-doc-images:
     stage: test
     script:
         - mkdir build && cd build
-        - conan install .. -pr buildfarm
+        - conan install .. -pr buildfarm  --build=missing
         - cmake .. -DCMAKE_BUILD_TYPE=$PYLENE_CONFIGURATION
         - cmake --build .
                 --config $PYLENE_CONFIGURATION
                 --target check
+    dependencies: []
     artifacts:
         reports:
             junit: build/tests/UT*.xml
 
 
+
 distcheck-linux-gcc-release:
     <<: *distcheck-linux-base
     variables:
@@ -76,15 +78,15 @@ distcheck-linux-clang-debug:
     stage: bench
     script:
         - mkdir build && cd build
-        - conan install .. -pr buildfarm
+        - conan install .. -pr buildfarm  --build=missing
         - cmake .. -DCMAKE_BUILD_TYPE=$PYLENE_CONFIGURATION
+        - cmake --build . --target fetch-external-data
         - cmake --build .
                 --config $PYLENE_CONFIGURATION
-                --target benchmarks
-        - cmake --build .
-                --config $PYLENE_CONFIGURATION
-                --target benchmarks-run
-                -- -j1  # Force serial execution
+                --target run-all-benchmarks
+        - ctest -L SpeedTests -V
+    when: manual
+    dependencies: []
     artifacts:
          name: "benchmark-results"
          paths:
@@ -115,6 +117,7 @@ distbench-linux-clang-release:
         - cd doc
         - doxygen
         - sphinx-build -b html source ../public
+    dependencies: [build-doc-images]
     artifacts:
         name: "documentation"
         paths:

bench/BMErosion.cpp

-#ifdef _MSC_VER
-#define __restrict__ __restrict
-#endif
-
-#include <mln/core/image/image2d.hpp>
-#include <mln/core/win2d.hpp>
-#include <mln/core/wrt_offset.hpp>
-#include <mln/morpho/structural/erode.hpp>
-
-using namespace mln;
-
-void erode_c8_mlnstyle(const mln::image2d<mln::uint8>& f, mln::image2d<mln::uint8>& out)
-{
-  mln_pixter(px, f);
-  mln_pixter(pxout, out);
-  mln_iter(nx, winc8(px));
-
-  mln_forall (px, pxout)
-  {
-    uint8 min = 255;
-    //#pragma clang loop vectorize(enable)
-    mln_simple_forall(nx) min = std::min(min, nx->val());
-    pxout->val()              = min;
-  }
-}
-
-void erode_c8_kernel(const mln::image2d<mln::uint8>& f, mln::image2d<mln::uint8>& out)
-{
-  (void)f;
-  (void)out;
-  mln::morpho::structural::erode(f, winc8, out, std::less<mln::uint8>());
-}
-
-void erode_c8_cstyle(const mln::image2d<mln::uint8>& f, mln::image2d<mln::uint8>& out)
-{
-  constexpr int sz      = 9;
-  auto          dpoints = winc8_t::dpoints;
-
-  const uint8* __restrict__ inptr = &f({0, 0});
-  uint8* __restrict__ outptr      = &out({0, 0});
-
-  ptrdiff_t offsets[sz];
-  wrt_offset(f, dpoints, offsets);
-
-  int stride = static_cast<int>(f.strides()[0]);
-  int nr     = f.nrows();
-  int nc     = f.ncols();
-
-  for (int y = 0; y < nr; ++y)
-  {
-    const uint8* inlineptr  = inptr + stride * y;
-    uint8*       outlineptr = outptr + stride * y;
-    for (int x = 0; x < nc; ++x)
-    {
-      uint8 min = 255;
-      for (int k = 0; k < sz; ++k)
-        min = std::min(min, inlineptr[x + offsets[k]]);
-      outlineptr[x] = min;
-    }
-  }
-}
-
-void erode_c8_cstyle_restrict(const mln::image2d<mln::uint8>& f, mln::image2d<mln::uint8>& out)
-{
-  constexpr int sz      = 9;
-  auto          dpoints = winc8_t::dpoints;
-
-  const uint8* __restrict__ inptr = &f({0, 0});
-  uint8* __restrict__ outptr      = &out({0, 0});
-
-  ptrdiff_t offsets[sz];
-  wrt_offset(f, dpoints, offsets);
-
-  int stride = static_cast<int>(f.strides()[0]);
-  int nr     = f.nrows();
-  int nc     = f.ncols();
-
-  auto applyLine = [nc, offsets](const uint8* __restrict__ inlineptr, uint8* __restrict__ outlineptr) {
-    for (int x = 0; x < nc; ++x)
-    {
-      uint8 min = 255;
-      for (int k = 0; k < sz; ++k)
-        min = std::min(min, inlineptr[x + offsets[k]]);
-      outlineptr[x] = min;
-    }
-  };
-
-  for (int y = 0; y < nr; ++y)
-  {
-    const uint8* inlineptr  = inptr + stride * y;
-    uint8*       outlineptr = outptr + stride * y;
-    applyLine(inlineptr, outlineptr);
-  }
-}

bench/BMErosion.hpp

-#include <mln/core/image/image2d.hpp>
-
-void erode_c8_mlnstyle(const mln::image2d<mln::uint8>& f, mln::image2d<mln::uint8>& out);
-
-void erode_c8_kernel(const mln::image2d<mln::uint8>& f, mln::image2d<mln::uint8>& out);
-
-void erode_c8_cstyle(const mln::image2d<mln::uint8>& f, mln::image2d<mln::uint8>& out);
-
-void erode_c8_cstyle_restrict(const mln::image2d<mln::uint8>& f, mln::image2d<mln::uint8>& out);

bench/BMErosion_main.cpp

-#include <mln/core/extension/fill.hpp>
-#include <mln/core/image/image2d.hpp>
-#include <mln/io/imread.hpp>
-#include <mln/io/imsave.hpp>
-
-#include <fixtures/ImagePath/image_path.hpp>
-
-#include <benchmark/benchmark.h>
-
-#include "BMErosion.hpp"
-
-using namespace mln;
-
-struct BMErosion : public benchmark::Fixture
-{
-  std::string filename = fixtures::ImagePath::concat_with_filename("lena.pgm");
-
-  BMErosion()
-  {
-    io::imread(filename, m_input);
-    int nr = m_input.nrows();
-    int nc = m_input.ncols();
-    resize(m_output, m_input);
-    m_bytes = nr * nc * sizeof(uint8);
-    mln::extension::fill(m_input, 255);
-  }
-
-  image2d<uint8> m_input;
-  image2d<uint8> m_output;
-  std::size_t    m_bytes;
-};
-
-BENCHMARK_F(BMErosion, Erosion_CStyle_Uint8)(benchmark::State& st)
-{
-  while (st.KeepRunning())
-  {
-    erode_c8_cstyle(m_input, m_output);
-  }
-  st.SetBytesProcessed(st.iterations() * m_bytes);
-}
-
-BENCHMARK_F(BMErosion, Erosion_CStyle_Restrict_Uint8)(benchmark::State& st)
-{
-  while (st.KeepRunning())
-  {
-    erode_c8_cstyle_restrict(m_input, m_output);
-  }
-  st.SetBytesProcessed(st.iterations() * m_bytes);
-}
-
-BENCHMARK_F(BMErosion, Erosion_Iterators_Uint8)(benchmark::State& st)
-{
-  while (st.KeepRunning())
-  {
-    erode_c8_mlnstyle(m_input, m_output);
-  }
-  st.SetBytesProcessed(st.iterations() * m_bytes);
-}
-
-BENCHMARK_F(BMErosion, Erosion_Kernel_Uint8)(benchmark::State& st)
-{
-  while (st.KeepRunning())
-  {
-    erode_c8_kernel(m_input, m_output);
-  }
-  st.SetBytesProcessed(st.iterations() * m_bytes);
-}
-
-BENCHMARK_MAIN();

bench/BMReference_Neighborhood_test.cpp

@@ -34,10 +34,9 @@ void Anisotropic_Diffusion_C(const mln::uint8* __restrict ibuffer, mln::uint8* _
 
 class TestNeighborhood : public ::testing::Test
 {
-  static std::string filename = fixtures::ImagePath::concat_with_filename("lena.ppm");
-
   virtual void SetUp() override
   {
+    std::string filename = fixtures::ImagePath::concat_with_filename("lena.ppm");
     mln::image2d<mln::rgb8> tmp(0, 0);
     mln::io::imread(filename, tmp);
 

bench/BenchmarkMacros.cmake

+# .rst:
+#
+# This file defines the following functions:
+#
+# * add_benchmark(Executable src_1 src_2 ... src_n)
+#
+# It creates a target with the name BM${Executable} which is the binary file
+# And a target BM${Executable}_run that runs the benchmark and is added to
+# the list of executables to be run.
+# A test is also added with the same name and belongs to the group "SpeedTests"
+# They can be run with ctest -L SpeedTests.
+# 
+
+add_custom_target(run-all-benchmarks)
+
+
+function(add_benchmark Executable)
+  set(Benchmark_SOURCES ${ARGN})
+  set(Benchmark_NAME ${Executable})
+
+  # Create benchmark exe
+  add_executable(${Benchmark_NAME} ${Benchmark_SOURCES})
+  target_link_libraries(${Benchmark_NAME} PRIVATE Fixtures::ImagePath Pylene::Pylene benchmark::benchmark ${FreeImage_LIBRARIES})
+  target_include_directories(${Benchmark_NAME} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
+
+
+  # Retrieve reference
+  set(Benchmark_OUTPUT_FILE ${Benchmark_NAME}-${CMAKE_CXX_COMPILER_ID}-${CMAKE_CXX_COMPILER_VERSION}.json)
+
+
+  # Create a target to run the benchmark
+  add_custom_target(run-${Benchmark_NAME}
+    COMMAND ${Benchmark_NAME} --benchmark_out_format=json --benchmark_out=${Benchmark_OUTPUT_FILE} --benchmark_repetitions=9 #--benchmark_display_aggregates_only=true
+    DEPENDS ${Benchmark_NAME}
+    WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
+    USES_TERMINAL)
+  add_dependencies(run-all-benchmarks run-${Benchmark_NAME})
+
+  # Create a speed test to run the benchmark and compare to the reference
+  if(EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/references/${Benchmark_OUTPUT_FILE}.md5")
+    message(STATUS "Test ${Benchmark_NAME} found.")
+    externaldata_add_test(
+      fetch-external-data
+      NAME ${Benchmark_NAME}
+      COMMAND python3 ${CMAKE_CURRENT_SOURCE_DIR}/tools/compare.py --display_aggregates_only benchmarks DATA{references/${Benchmark_OUTPUT_FILE}} ${Benchmark_OUTPUT_FILE})
+    set_tests_properties(${Benchmark_NAME} PROPERTIES
+      LABELS SpeedTests
+      RUN_SERIAL Yes)
+  endif()
+endfunction(add_benchmark)

bench/CMakeLists.txt

 find_package(benchmark REQUIRED)
 find_package(GTest REQUIRED)
 
+include(CTest)
+include(ExternalData)
+include(BenchmarkMacros.cmake)
+
 # Download file dependency
-# Big images are stored at https://www.lrde.epita.fr/~mroynard/big_images/
-file(DOWNLOAD "https://www.lrde.epita.fr/~mroynard/big_images/Space1_20MB.jpg" "${CMAKE_BINARY_DIR}/bench/Space1_20MB.jpg")
+set(ExternalData_URL_TEMPLATES
+  "file:///lrde/dload/olena/pylene/data/%(algo)/%(hash)"
+  "https://www.lrde.epita.fr/dload/olena/pylene/data/%(algo)/%(hash)")
+
+ExternalData_Expand_Arguments(
+  fetch-external-data
+  images
+  DATA{Space1_20MB.jpg}
+  )
 
+# Extra compiler options
 
 if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
   set(STANDALONE_COMPILE_FLAGS "-Rpass=loop-vectorize -Rpass-missed=loop-vectorize -Rpass-analysis=loop-vectorize -gline-tables-only -gcolumn-info ")
@@ -29,37 +41,18 @@ set(src_standalone
 set_source_files_properties(${src_standalone} PROPERTIES COMPILE_FLAGS ${STANDALONE_COMPILE_FLAGS})
 
 
-add_custom_target(benchmarks ALL COMMENT "Build all the benchmarks.")
-
-add_custom_target(benchmarks-run
-        DEPENDS benchmarks
-        COMMENT "Build and run all the benchmarks.")
-
-macro(add_benchmark Executable)
-    set(Sources ${ARGN})
-    add_executable(BM${Executable} ${Sources})
 
-    target_link_libraries(BM${Executable} PRIVATE Fixtures::ImagePath Pylene::Pylene benchmark::benchmark ${FreeImage_LIBRARIES})
-    target_include_directories(BM${Executable} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
+enable_testing()
 
-    add_dependencies(benchmarks BM${Executable})
+add_benchmark(BMNeighborhood            BMNeighborhood.cpp BMNeighborhood_main.cpp)
+add_benchmark(BMRotation                BMRotation.cpp)
+add_benchmark(BMDilation                BMDilation.cpp)
+add_benchmark(BMMorphers                BMMorphers.cpp BMMorphers_main.cpp)
+add_benchmark(BMReference_Linear        BMReference_Linear.cpp BMReference_Linear_Reversed.cpp BMReference_Linear_main.cpp)
+add_benchmark(BMReference_Neighborhood  BMReference_Neighborhood.cpp BMReference_Neighborhood_main.cpp)
 
-    add_custom_target(BM${Executable}_run
-        COMMAND BM${Executable} --benchmark_out_format=json --benchmark_out=BM${Executable}.json
-        DEPENDS BM${Executable}
-        WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
-        USES_TERMINAL)
-
-    add_dependencies(benchmarks-run BM${Executable}_run)
-endmacro(add_benchmark)
+add_executable(UTReference_Neighborhood BMReference_Neighborhood.cpp BMReference_Neighborhood_test.cpp)
+target_link_libraries(UTReference_Neighborhood PRIVATE Fixtures::ImagePath Pylene::Pylene GTest::Main GTest::GTest ${FreeImage_LIBRARIES})
 
-add_benchmark(Erosion                 BMErosion.cpp BMErosion_main.cpp)
-add_benchmark(Neighborhood            BMNeighborhood.cpp BMNeighborhood_main.cpp)
-add_benchmark(Rotation                BMRotation.cpp)
-add_benchmark(Dilation                BMDilation.cpp)
-add_benchmark(Morphers                BMMorphers.cpp BMMorphers_main.cpp)
-add_benchmark(Reference_Linear        BMReference_Linear.cpp BMReference_Linear_Reversed.cpp BMReference_Linear_main.cpp)
-add_benchmark(Reference_Neighborhood  BMReference_Neighborhood.cpp BMReference_Neighborhood_main.cpp)
 
-add_executable(UTReference_Neighborhood BMReference_Neighborhood.cpp BMReference_Neighborhood_test.cpp)
-target_link_libraries(UTReference_Neighborhood Pylene::Pylene GTest::Main GTest::GTest ${FreeImage_LIBRARIES})
+ExternalData_Add_Target(fetch-external-data)

bench/Readme.md

+# Compiling, running, comparing benchmarks
+
+## Fetching external data
+
+Before compiling and running the benchmarks, external data are required:
+* external image samples (that can be quite huge for testing)
+* reference benchmarks
+So before everything else, fetch external data with the rules:
+
+```sh
+cmake --build . --target fetch-external-data
+```
+
+## Compiling and executing benchmarks
+
+Benchmarks are compiled automatically when the target
+`run-all-benchmarks` is triggered and then run.
+
+
+```sh
+cmake --build . --target run-all-benchmarks --config Release
+```
+
+It will output for each benchmark, a speed report with the name:
+`{BenchmarkName}-{Compiler}-{Version}.json`
+
+
+## Non-regressions tests
+
+Once the benchmarks have been run, you can compare with the references by running the tests. Note that a test is ran
+only if there is a reference file.
+
+
+```sh
+ctest -L SpeedTests -V
+```
+
+# Adding & Updating references
+
+* Download benchmark artifacts from the CI
+* Copy artifacts to `/lrde/dload/olena/pylene/data/benchmarks`
+* Create links to `/lrde/dload/olena/pylene/data/MD5` to new benchmarks (use the script `install.sh`)
+* Update MD5 checksum in `CVS_ROOT/bench/references` (just move all `*.md5` that have been generated by the `install.sh` script)
+
+

bench/Space1_20MB.jpg.md5

+d72f20e1cd3c273dcdbd5660dc8e0ffc

bench/references/BMDilation-Clang-7.0.0.json.md5

+8b99eb45aafca6477b7ae888d8390b79

bench/references/BMDilation-Clang-7.0.1.json.md5

+f7673886b29b0b1c6c4ee524e30a6485

bench/references/BMDilation-GNU-8.2.1.json.md5

+d0f76f7c435caa6feb87b64f246b0563

bench/references/BMMorphers-Clang-7.0.0.json.md5

+babdf051a3853ad96a0f4e03b9f85a51

bench/references/BMMorphers-Clang-7.0.1.json.md5

+2b5ece70d9e6bf82255c18acc74403a6

bench/references/BMMorphers-GNU-8.2.1.json.md5

+0e52ca75826c746e4a2c02f677eadbef

bench/references/BMNeighborhood-Clang-7.0.0.json.md5

+4f90c3d75a72c3c47d8e5bcc90cfc6f2

bench/references/BMNeighborhood-Clang-7.0.1.json.md5

+a8f834e2db1f04ee9e576229d0987ba0

bench/references/BMNeighborhood-GNU-8.2.1.json.md5

+954d1bb38b850720b5bc1c5e0a74bf61

bench/references/BMReference_Linear-Clang-7.0.0.json.md5

+0b4d02981ceec4ae4342a3343b0f68fd

bench/references/BMReference_Linear-Clang-7.0.1.json.md5

+2374df6f0ac6e675d94fc5cd8ea6e636