From 81b60be587777e03baadb3a586f7f56de7c111d0 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 14:29:58 +0100
Subject: [PATCH 01/18] Add schaapspack and gracehopper runs
---
.gitlab-ci.yml | 20 ++++++++++++++++++++
CMakeLists.txt | 9 +++++++++
ci/das6/compile_and_run_native.sh | 20 ++++++++++++++++++++
3 files changed, 49 insertions(+)
create mode 100644 ci/das6/compile_and_run_native.sh
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index d11a266..d1d46da 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -96,6 +96,26 @@ performance-jetson: # This job runs in the test stage.
access: all
expire_in: 1 days
+performance-gracehopper: # This job runs in the test stage.
+ allow_failure: true
+ tags:
+ - das6-gpu
+ stage: benchmark # It only starts when the job in the build stage completes successfully.
+ script:
+ - sbatch --wait -p ghq -o output.txt -e error.txt ci/das6/compile_and_run_native_sh ghq_arm64
+ - cat output.txt >&1
+ - cat error.txt >&2
+
+ artifacts:
+ paths:
+ - ./results*.json
+ - ./output.txt
+ - ./error.txt
+ - ./*.tar
+ when: always
+ access: all
+ expire_in: 1 days
+
performance-generic:
stage: benchmark
image: "$CI_REGISTRY_IMAGE:latest"
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 73d5509..153e159 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -49,6 +49,15 @@ FetchContent_Declare(
FetchContent_Populate(aocommon)
+# Make schaapspack available
+FetchContent_Declare(
+ schaapcommon
+ GIT_REPOSITORY git@git.astron.nl:RD/schaapcommon.git
+ GIT_TAG master)
+
+FetchContent_Populate(schaapcommon)
+
+
set(COMPILER_FLAGS "-O3;-march=native;-ggdb;")
# List all kernel code
diff --git a/ci/das6/compile_and_run_native.sh b/ci/das6/compile_and_run_native.sh
new file mode 100644
index 0000000..4125cdc
--- /dev/null
+++ b/ci/das6/compile_and_run_native.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+
+set -e
+# compile the code and run it on das6
+# Specify the compiler version and architecture
+
+ARCHITECTURE=$1
+COMPILER_VERSION=$(gcc --version | head -n 1 | awk '{print $4}')
+
+
+echo RUNNING ON ${COMPILER_VERSION} AND ${ARCHITECTURE}
+BUILD_DIR=build-${COMPILER_VERSION}-${ARCHITECTURE}
+
+cmake -B ${BUILD_DIR} . -DCMAKE_BUILD_TYPE=Release
+
+make -C ${BUILD_DIR} -j
+
+tar -cvf asm-${ARCHITECTURE}-${COMPILER_VERSION}.tar ${BUILD_DIR}/*.s
+
+${BUILD_DIR}/microbenchmarks --benchmark_out=results-${COMPILER_VERSION}-${ARCHITECTURE}.json --benchmark_out_format=json
\ No newline at end of file
--
GitLab
From e038b2a35a4e2465c8be64b9d06b93b2012ab106 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 17:07:03 +0100
Subject: [PATCH 02/18] Fix build and add benchmark test
---
CMakeLists.txt | 18 +--
benchmarks/convolution.cpp | 62 ++++++++++
code/convolution.h | 26 +++++
code/convolution_reference.cpp | 8 ++
code/convolution_serial_fftw.cpp | 190 +++++++++++++++++++++++++++++++
5 files changed, 296 insertions(+), 8 deletions(-)
create mode 100644 benchmarks/convolution.cpp
create mode 100644 code/convolution.h
create mode 100644 code/convolution_reference.cpp
create mode 100644 code/convolution_serial_fftw.cpp
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 153e159..fdbe25d 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -45,9 +45,10 @@ include(Catch)
FetchContent_Declare(
aocommon
GIT_REPOSITORY https://gitlab.com/aroffringa/aocommon.git
- GIT_TAG master)
+ GIT_TAG master
+ EXCLUDE_FROM_ALL)
-FetchContent_Populate(aocommon)
+FetchContent_MakeAvailable(aocommon)
# Make schaapspack available
FetchContent_Declare(
@@ -55,8 +56,8 @@ FetchContent_Declare(
GIT_REPOSITORY git@git.astron.nl:RD/schaapcommon.git
GIT_TAG master)
-FetchContent_Populate(schaapcommon)
-
+FetchContent_MakeAvailable(schaapcommon)
+target_include_directories(schaapcommon PUBLIC ${aocommon_SOURCE_DIR}/include)
set(COMPILER_FLAGS "-O3;-march=native;-ggdb;")
@@ -73,8 +74,9 @@ find_package(OpenMP)
# Link against Google Benchmark
target_link_libraries(microbenchmarks PRIVATE benchmark::benchmark)
-target_include_directories(microbenchmarks PRIVATE ${aocommon_SOURCE_DIR}/include)
+target_include_directories(microbenchmarks PRIVATE ${aocommon_SOURCE_DIR}/include ${schaapcommon_SOURCE_DIR}/include)
target_include_directories(microbenchmarks PRIVATE code)
+target_link_libraries(microbenchmarks PRIVATE schaapcommon)
target_compile_options(microbenchmarks PUBLIC ${COMPILER_FLAGS})
if(OpenMP_CXX_FOUND)
target_link_libraries(microbenchmarks PRIVATE OpenMP::OpenMP_CXX)
@@ -88,15 +90,15 @@ list(APPEND CMAKE_MODULE_PATH ${catch2_SOURCE_DIR}/extras)
catch_discover_tests(unittests WORKING_DIRECTORY)
target_link_libraries(unittests PRIVATE Catch2::Catch2WithMain)
target_include_directories(unittests PRIVATE code)
-target_include_directories(unittests PRIVATE ${aocommon_SOURCE_DIR}/include)
-
+target_include_directories(unittests PRIVATE ${aocommon_SOURCE_DIR}/include ${schaapcommon_SOURCE_DIR}/include)
+target_link_libraries(unittests PRIVATE schaapcommon)
target_compile_options(unittests PUBLIC ${COMPILER_FLAGS})
foreach(KERNEL_SOURCE ${KERNEL_SOURCES})
get_filename_component(KERNEL_NAME ${KERNEL_SOURCE} NAME_WE)
add_precompile_target(TARGET_NAME ${KERNEL_NAME}
- INCLUDE_DIRS code ${aocommon_SOURCE_DIR}/include
+ INCLUDE_DIRS code ${aocommon_SOURCE_DIR}/include ${schaapcommon_SOURCE_DIR}/include
SOURCES ${KERNEL_SOURCE}
COMPILER_FLAGS ${COMPILER_FLAGS})
endforeach()
diff --git a/benchmarks/convolution.cpp b/benchmarks/convolution.cpp
new file mode 100644
index 0000000..ae82e9f
--- /dev/null
+++ b/benchmarks/convolution.cpp
@@ -0,0 +1,62 @@
+#include <benchmark/benchmark.h>
+#include <convolution.h>
+#include <memory>
+#include <vector>
+
+namespace {
+
+class InitializeInput : public benchmark::Fixture {
+ public:
+ void SetUp(::benchmark::State& state) {
+ size_t width = state.range(0);
+ size_t height = state.range(1);
+ image = std::make_unique<std::vector<float>>(width * height);
+ kernel = std::make_unique<std::vector<float>>(width * height);
+
+ Initialize(image->data(), width, height);
+ Initialize(kernel->data(), width, height);
+ }
+ void TearDown(::benchmark::State& state) {
+ image.reset();
+ kernel.reset();
+ }
+
+ std::unique_ptr<std::vector<float>> image;
+ std::unique_ptr<std::vector<float>> kernel;
+};
+} // namespace
+
+// Reference standard
+BENCHMARK_DEFINE_F(InitializeInput, ConvolveReference)
+(benchmark::State& state) {
+ for (auto _ : state) {
+ ConvolveReference(image->data(), kernel->data(), state.range(0),
+ state.range(1));
+ }
+}
+BENCHMARK_REGISTER_F(InitializeInput, ConvolveReference)
+ ->Args({64, 32})
+ ->Args({128, 64})
+ ->Args({256, 126})
+ ->Args({512, 754})
+ ->Args({1024, 124})
+ ->Args({2048, 1000})
+ ->Args({4096, 5000});
+
+// FFTW serial standard
+BENCHMARK_DEFINE_F(InitializeInput, ConvolveSerial)
+(benchmark::State& state) {
+ for (auto _ : state) {
+ ConvolveSerial(image->data(), kernel->data(), state.range(0),
+ state.range(1));
+ }
+}
+
+BENCHMARK_REGISTER_F(InitializeInput, ConvolveSerial)
+ ->Args({64, 32})
+ ->Args({128, 64})
+ ->Args({256, 126})
+ ->Args({512, 754})
+ ->Args({1024, 124})
+ ->Args({2048, 1000})
+ ->Args({4096, 5000});
diff --git a/code/convolution.h b/code/convolution.h
new file mode 100644
index 0000000..66e8c8c
--- /dev/null
+++ b/code/convolution.h
@@ -0,0 +1,26 @@
+#ifndef CONVOLUTION_H_
+#define CONVOLUTION_H_
+
+#include <complex>
+#include <iomanip>
+#include <iostream>
+#include <random>
+
+inline void Initialize(float* a, const size_t width, const size_t height) {
+ // Initialize matrices with random complex values
+ std::seed_seq seed({42});
+ std::mt19937 gen(seed);
+ std::uniform_real_distribution<float> dis(-1.0, 1.0);
+ const size_t linear_size = width * height;
+ for (int i = 0; i < linear_size; i++) {
+ a[i] = dis(gen);
+ }
+}
+
+// Function to perform matrix multiplication for 2x2 complex matrices
+void ConvolveReference(float* image, const float* kernel, size_t width,
+ size_t height);
+
+void ConvolveSerial(float* image, const float* kernel, size_t width,
+ size_t height);
+#endif
diff --git a/code/convolution_reference.cpp b/code/convolution_reference.cpp
new file mode 100644
index 0000000..e83c8d0
--- /dev/null
+++ b/code/convolution_reference.cpp
@@ -0,0 +1,8 @@
+#include "convolution.h"
+
+#include <schaapcommon/math/convolution.h>
+
+void ConvolveReference(float* image, const float* kernel, size_t width,
+ size_t height) {
+ schaapcommon::math::Convolve(image, kernel, width, height);
+}
\ No newline at end of file
diff --git a/code/convolution_serial_fftw.cpp b/code/convolution_serial_fftw.cpp
new file mode 100644
index 0000000..e097730
--- /dev/null
+++ b/code/convolution_serial_fftw.cpp
@@ -0,0 +1,190 @@
+#include "convolution.h"
+#include <fftw3.h>
+#include <algorithm>
+// Partially unroll rows/columns with a factor of kUnroll
+constexpr size_t kUnroll = 4;
+
+// With kUnroll > 1, the temporary buffers need to be aligned
+// for FFTW to work correctly.
+constexpr size_t kAlignment = 64;
+
+size_t RoundUp(size_t a, size_t b) { return ((a + b) / b) * b; }
+
+void FftR2CComposite(fftwf_plan plan_r2c, fftwf_plan plan_c2c,
+ size_t image_height, size_t image_width, const float* in,
+ fftwf_complex* out) {
+ const size_t complex_width = image_width / 2 + 1;
+ const size_t complex_size = image_height * complex_width;
+
+ fftwf_complex* temp1 = fftwf_alloc_complex(complex_size);
+
+ fftwf_complex* temp2 = fftwf_alloc_complex(complex_width);
+ float* temp2_ptr = reinterpret_cast<float*>(temp2);
+ for (size_t y = 0; y < image_height; y++) {
+ float* temp1_ptr = reinterpret_cast<float*>(&temp1[y * complex_width]);
+ std::copy_n(&in[y * image_width], image_width, temp2_ptr);
+ fftwf_execute_dft_r2c(plan_r2c, temp2_ptr, temp2);
+ std::copy_n(temp2_ptr, 2 * complex_width, temp1_ptr);
+ }
+ fftwf_free(temp2);
+
+ // Partially kUnroll over columns
+ size_t padded_height = RoundUp(image_height, kAlignment);
+ temp2 = fftwf_alloc_complex(kUnroll * padded_height);
+
+ for (size_t x = 0; x < complex_width; x += kUnroll) {
+ // Copy input
+ for (size_t y = 0; y < image_height; y++) {
+ for (size_t i = 0; i < kUnroll; i++) {
+ if ((x + i) < complex_width) {
+ float* temp1_ptr =
+ reinterpret_cast<float*>(&temp1[y * complex_width + x + i]);
+ float* temp2_ptr =
+ reinterpret_cast<float*>(&temp2[i * padded_height + y]);
+ std::copy_n(temp1_ptr, 2, temp2_ptr);
+ }
+ }
+ }
+
+ // Perform 1D FFT over columns
+ for (size_t i = 0; i < kUnroll; i++) {
+ fftwf_complex* temp2_ptr = &temp2[i * padded_height];
+ fftwf_execute_dft(plan_c2c, temp2_ptr, temp2_ptr);
+ }
+
+ // Transpose output
+ for (size_t y = 0; y < image_height; y++) {
+ for (size_t i = 0; i < kUnroll; i++) {
+ if ((x + i) < complex_width) {
+ float* temp2_ptr =
+ reinterpret_cast<float*>(&temp2[i * padded_height + y]);
+ float* out_ptr =
+ reinterpret_cast<float*>(&out[y * complex_width + x + i]);
+ std::copy_n(temp2_ptr, 2, out_ptr);
+ }
+ }
+ }
+ }
+
+ fftwf_free(temp2);
+ fftwf_free(temp1);
+}
+
+void FftC2RComposite(fftwf_plan plan_c2c, fftwf_plan plan_c2r,
+ size_t image_height, size_t image_width,
+ const fftwf_complex* in, float* out) {
+ const size_t complex_width = image_width / 2 + 1;
+
+ size_t padded_height = RoundUp(image_height, kAlignment);
+ size_t padded_size = padded_height * complex_width;
+ fftwf_complex* temp1 = fftwf_alloc_complex(padded_size);
+
+ for (size_t x = 0; x < complex_width; x += kUnroll) {
+ // Transpose input
+ for (size_t y = 0; y < image_height; y++) {
+ for (size_t i = 0; i < kUnroll; i++) {
+ if ((x + i) < complex_width) {
+ const float* in_ptr =
+ reinterpret_cast<const float*>(&in[y * complex_width + x + i]);
+ float* temp1_ptr =
+ reinterpret_cast<float*>(&temp1[(x + i) * padded_height + y]);
+ std::copy_n(in_ptr, 2, temp1_ptr);
+ }
+ }
+ }
+
+ // Perform 1D C2C FFT over columns
+ for (size_t i = 0; i < kUnroll; i++) {
+ if ((x + i) < complex_width) {
+ fftwf_complex* temp1_ptr = &temp1[(x + i) * padded_height];
+ fftwf_execute_dft(plan_c2c, temp1_ptr, temp1_ptr);
+ }
+ }
+ }
+
+ size_t paddedWidth = RoundUp(complex_width, kAlignment);
+ fftwf_complex* temp2 = fftwf_alloc_complex(kUnroll * paddedWidth);
+
+ for (size_t y = 0; y < image_height; y += kUnroll) {
+ // Transpose input
+ for (size_t x = 0; x < complex_width; x++) {
+ for (size_t i = 0; i < kUnroll; i++) {
+ if ((y + i) < image_height) {
+ float* temp1_ptr =
+ reinterpret_cast<float*>(&temp1[x * padded_height + y + i]);
+ float* temp2_ptr =
+ reinterpret_cast<float*>(&temp2[i * paddedWidth + x]);
+ std::copy_n(temp1_ptr, 2, temp2_ptr);
+ }
+ }
+ }
+
+ // Perform 1D C2R FFT over rows
+ for (size_t i = 0; i < kUnroll; i++) {
+ if ((y + i) < image_height) {
+ fftwf_complex* temp2_ptr = &temp2[i * paddedWidth];
+ fftwf_execute_dft_c2r(plan_c2r, temp2_ptr,
+ reinterpret_cast<float*>(temp2_ptr));
+ }
+ }
+
+ // Copy output
+ for (size_t i = 0; i < kUnroll; i++) {
+ if ((y + i) < image_height) {
+ float* temp2_ptr = reinterpret_cast<float*>(&temp2[i * paddedWidth]);
+ std::copy_n(temp2_ptr, image_width, &out[(y + i) * image_width]);
+ }
+ }
+ }
+
+ fftwf_free(temp2);
+
+ fftwf_free(temp1);
+}
+
+void ConvolveSerial(float* image, const float* kernel, size_t image_width,
+ size_t image_height) {
+ const size_t image_size = image_width * image_height;
+ const size_t complex_width = image_width / 2 + 1;
+ const size_t complex_size = complex_width * image_height;
+ float* temp_data = fftwf_alloc_real(image_size);
+ fftwf_complex* fft_image_data = fftwf_alloc_complex(complex_size);
+ fftwf_complex* fft_kernel_data = fftwf_alloc_complex(complex_size);
+
+ fftwf_plan plan_r2c =
+ fftwf_plan_dft_r2c_1d(image_width, nullptr, nullptr, FFTW_ESTIMATE);
+ fftwf_plan plan_c2c_forward = fftwf_plan_dft_1d(
+ image_height, nullptr, nullptr, FFTW_FORWARD, FFTW_ESTIMATE);
+ fftwf_plan plan_c2c_backward = fftwf_plan_dft_1d(
+ image_height, nullptr, nullptr, FFTW_BACKWARD, FFTW_ESTIMATE);
+ fftwf_plan plan_c2r =
+ fftwf_plan_dft_c2r_1d(image_width, nullptr, nullptr, FFTW_ESTIMATE);
+
+ FftR2CComposite(plan_r2c, plan_c2c_forward, image_height, image_width, image,
+ fft_image_data);
+
+ std::copy_n(kernel, image_size, temp_data);
+ FftR2CComposite(plan_r2c, plan_c2c_forward, image_height, image_width,
+ temp_data, fft_kernel_data);
+
+ const float fact = 1.0 / image_size;
+ for (size_t y = 0; y != image_height; ++y) {
+ for (size_t x = 0; x != complex_width; ++x) {
+ const size_t i = y * complex_width + x;
+ reinterpret_cast<std::complex<float>*>(fft_image_data)[i] *=
+ fact * reinterpret_cast<std::complex<float>*>(fft_kernel_data)[i];
+ }
+ }
+
+ FftC2RComposite(plan_c2c_backward, plan_c2r, image_height, image_width,
+ fft_image_data, image);
+
+ fftwf_free(fft_image_data);
+ fftwf_free(fft_kernel_data);
+ fftwf_free(temp_data);
+
+ fftwf_destroy_plan(plan_r2c);
+ fftwf_destroy_plan(plan_c2c_forward);
+ fftwf_destroy_plan(plan_c2c_backward);
+ fftwf_destroy_plan(plan_c2r);
+}
\ No newline at end of file
--
GitLab
From 21f86690899ad17e18156251dba4c6d395364c54 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 17:32:50 +0100
Subject: [PATCH 03/18] Add tests
---
code/convolution.h | 1 +
test/helpers.cpp | 49 ++++++++++++++++++++++++++++++++++++++-
test/helpers.h | 7 ++++++
test/test_convolution.cpp | 37 +++++++++++++++++++++++++++++
4 files changed, 93 insertions(+), 1 deletion(-)
create mode 100644 test/test_convolution.cpp
diff --git a/code/convolution.h b/code/convolution.h
index 66e8c8c..5e0ceb4 100644
--- a/code/convolution.h
+++ b/code/convolution.h
@@ -23,4 +23,5 @@ void ConvolveReference(float* image, const float* kernel, size_t width,
void ConvolveSerial(float* image, const float* kernel, size_t width,
size_t height);
+
#endif
diff --git a/test/helpers.cpp b/test/helpers.cpp
index 60cabef..9b4b04b 100644
--- a/test/helpers.cpp
+++ b/test/helpers.cpp
@@ -16,6 +16,14 @@ void compareSingle(const std::vector<T>& lv, const std::vector<T>& rv,
REQUIRE_THAT(lv, Catch::Matchers::WithinAbs(rv, precision));
}
+template <typename T>
+void compareMulti(const std::vector<T>& lv, const std::vector<T>& rv,
+ float precision) {
+ for (size_t idx = 0; idx < lv.size(); idx++) {
+ REQUIRE_THAT(lv[idx], Catch::Matchers::WithinAbs(rv[idx], precision));
+ }
+}
+
template <>
void compareSingle(const std::vector<std::complex<float>>& lv,
const std::vector<std::complex<float>>& rv,
@@ -49,7 +57,7 @@ void compareArrays(const std::string& test, unsigned line, std::array<T, N> lhs,
std::stringstream ss;
ss << "Expected : \n";
- for (size_t idx = 0; idx < N; idx++) {
+ for (size_t idx = 0; idx < lv.size(); idx++) {
ss << valueToString(lhs[idx]) << "\t";
}
@@ -59,14 +67,53 @@ void compareArrays(const std::string& test, unsigned line, std::array<T, N> lhs,
}
ss << "\n";
INFO("Reason: \n" << ss.str());
+
compareSingle(lv, rv, precision);
}
+template <typename T>
+void compareVectors(const std::string& test, unsigned line, std::vector<T> lhs,
+ std::vector<T> rhs, float precision) {
+ INFO("Test case [" << test << "] failed at line "
+ << line); // Reported only if REQUIRE fails
+
+ std::stringstream ss;
+ if (lhs.size() != rhs.size()) {
+ ss << " Size mismatch\n";
+ ss << "Expected size : " << lhs.size() << "\n";
+ ss << "Obtained size : " << rhs.size() << "\n";
+ INFO("Reason: \n" << ss.str());
+ }
+
+ CHECK(lhs.size() == rhs.size());
+ const size_t N = lhs.size();
+ ss << "Expected : \n";
+ for (size_t idx = 0; idx < N; idx++) {
+ ss << valueToString(lhs[idx]) << "\t";
+ }
+
+ ss << "\nObtained : \n";
+ for (size_t idx = 0; idx < N; idx++) {
+ ss << valueToString(rhs[idx]) << "\t";
+ }
+ ss << "\n";
+ INFO("Reason: \n" << ss.str());
+ compareMulti(lhs, rhs, precision);
+}
+
template void compareArrays(const std::string& test, unsigned line,
std::array<std::complex<float>, 4ul> lhs,
std::array<std::complex<float>, 4ul> rhs,
float precision);
+template void compareVectors(const std::string& test, unsigned line,
+ std::vector<float> lhs, std::vector<float> rhs,
+ float precision);
+
+template void compareVectors(const std::string& test, unsigned line,
+ std::vector<double> lhs, std::vector<double> rhs,
+ float precision);
+
void AssertEqual(const aocommon::Matrix4x4& a, const aocommon::Matrix4x4& b,
float precision) {
for (size_t i = 0; i < 16; i++) {
diff --git a/test/helpers.h b/test/helpers.h
index 83f306f..f3ce809 100644
--- a/test/helpers.h
+++ b/test/helpers.h
@@ -14,6 +14,9 @@
#define COMPARE_ARRAYS(lhs, rhs, precision) \
compareArrays(Catch::getResultCapture().getCurrentTestName(), __LINE__, lhs, \
rhs, precision)
+#define COMPARE_VECTORS(lhs, rhs, precision) \
+ compareVectors(Catch::getResultCapture().getCurrentTestName(), __LINE__, \
+ lhs, rhs, precision)
template <typename T>
void compareSingle(const std::vector<T>& lv, const std::vector<T>& rv,
@@ -32,6 +35,10 @@ template <typename T, size_t N>
void compareArrays(const std::string& test, unsigned line, std::array<T, N> lhs,
std::array<T, N> rhs, float precision);
+template <typename T>
+void compareVectors(const std::string& test, unsigned line, std::vector<T> lhs,
+ std::vector<T> rhs, float precision);
+
void AssertEqual(const aocommon::Matrix4x4& a, const aocommon::Matrix4x4& b,
float precision);
diff --git a/test/test_convolution.cpp b/test/test_convolution.cpp
new file mode 100644
index 0000000..4227e34
--- /dev/null
+++ b/test/test_convolution.cpp
@@ -0,0 +1,37 @@
+#include <convolution.h>
+
+#include <catch2/catch_test_macros.hpp>
+#include <catch2/matchers/catch_matchers_floating_point.hpp>
+
+#include "helpers.h"
+
+TEST_CASE("test convolution", "[float]") {
+ // This setup will be done 4 times in total, once for each section
+ size_t width = 16;
+ size_t height = 32;
+
+ std::vector<float> image(width * height);
+ std::vector<float> expected_image(width * height);
+
+ std::vector<float> kernel(width * height);
+ Initialize(image.data(), width, height);
+ std::copy(image.begin(), image.end(), expected_image.begin());
+ Initialize(kernel.data(), width, height);
+
+ ConvolveReference(expected_image.data(), kernel.data(), width, height);
+
+ SECTION("test correctness of reference implementation") {
+ ConvolveReference(image.data(), kernel.data(), width, height);
+ COMPARE_VECTORS(expected_image, image, 1.e-5);
+ }
+
+ SECTION("test correctness of reference implementation twice") {
+ ConvolveReference(image.data(), kernel.data(), width, height);
+ COMPARE_VECTORS(expected_image, image, 1.e-5);
+ }
+
+ SECTION("test correctness of serial implementation") {
+ ConvolveSerial(image.data(), kernel.data(), width, height);
+ COMPARE_VECTORS(expected_image, image, 1.e-5);
+ }
+}
\ No newline at end of file
--
GitLab
From 09fc57641cd73355524fb48419acc53c1056735e Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 17:45:56 +0100
Subject: [PATCH 04/18] Use HTTPS instead of SSH to clone schaapcommon
---
CMakeLists.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index fdbe25d..26c5e45 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -53,7 +53,7 @@ FetchContent_MakeAvailable(aocommon)
# Make schaapspack available
FetchContent_Declare(
schaapcommon
- GIT_REPOSITORY git@git.astron.nl:RD/schaapcommon.git
+ GIT_REPOSITORY https://git.astron.nl/RD/schaapcommon.git
GIT_TAG master)
FetchContent_MakeAvailable(schaapcommon)
--
GitLab
From 247218cff598ce95e5ed5264c1e158692952559c Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 17:48:37 +0100
Subject: [PATCH 05/18] Add missing dependency
---
docker/Dockerfile | 1 +
1 file changed, 1 insertion(+)
diff --git a/docker/Dockerfile b/docker/Dockerfile
index 67a2ef8..a60a0d8 100644
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -9,6 +9,7 @@ RUN apt-get update -qq &&\
git \
libblas-dev liblapack-dev \
libboost-date-time-dev \
+ libbost-filesystem-dev \
libboost-test-dev \
libboost-dev \
libcfitsio-dev \
--
GitLab
From ac9acc622bee1834db0a622e7605a4ed71306c64 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 17:49:34 +0100
Subject: [PATCH 06/18] Add fftw3
---
docker/Dockerfile | 1 +
1 file changed, 1 insertion(+)
diff --git a/docker/Dockerfile b/docker/Dockerfile
index a60a0d8..4f51d02 100644
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -18,6 +18,7 @@ RUN apt-get update -qq &&\
libhdf5-dev \
libopenmpi-dev \
libpython3-dev \
+ libfftw3-dev \
pkg-config \
python3-dev python3-numpy \
python3-sphinx \
--
GitLab
From 0f361b708b7dc360f73009666c978b6fbe08eca1 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 17:51:33 +0100
Subject: [PATCH 07/18] Fix name of library
---
docker/Dockerfile | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/docker/Dockerfile b/docker/Dockerfile
index 4f51d02..6b72250 100644
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -9,7 +9,7 @@ RUN apt-get update -qq &&\
git \
libblas-dev liblapack-dev \
libboost-date-time-dev \
- libbost-filesystem-dev \
+ libboost-filesystem-dev \
libboost-test-dev \
libboost-dev \
libcfitsio-dev \
--
GitLab
From 1131c9a79e460da3b7a062a8950159f2106ec0e9 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 19:56:14 +0100
Subject: [PATCH 08/18] Force pull policy
---
.gitlab-ci.yml | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index d1d46da..28fda75 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -25,7 +25,9 @@ docker-base:
- docker/Dockerfile
.build_docker:
- image: "$CI_REGISTRY_IMAGE:latest"
+ image:
+ name: "$CI_REGISTRY_IMAGE:latest"
+ pull_policy: always
before_script:
- cmake -B build . -DCMAKE_BUILD_TYPE=Release
- make -C build -j
@@ -33,7 +35,9 @@ docker-base:
build-job: # This job runs in the build stage, which runs first.
stage: build
- image: "$CI_REGISTRY_IMAGE:latest"
+ image:
+ name: "$CI_REGISTRY_IMAGE:latest"
+ pull_policy: always
script:
- cmake -B build . -DCMAKE_BUILD_TYPE=Release
- make -C build -j
--
GitLab
From 5ae0648f742bf7dda19cafb2228d3af07f5fc92b Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 19:59:13 +0100
Subject: [PATCH 09/18] Summarize convolution results
---
.gitlab-ci.yml | 1 +
1 file changed, 1 insertion(+)
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 28fda75..d4e9793 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -155,4 +155,5 @@ collect-performance:
- python3 ci/summarize-results.py --filter MatrixMultiplication results*.json result-summary-matrix-multiplication
- python3 ci/summarize-results.py --filter HermitianSquare results*.json result-summary-hermitian-square
- python3 ci/summarize-results.py --filter KroneckerSquare results*.json result-summary-kronecker-square
+ - python3 ci/summarize-results.py --filter Convolution results*.json result-summary-convolution
--
GitLab
From 5ea6c8fb99c1d840c8890e335b346afddd4b1c36 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 20:01:31 +0100
Subject: [PATCH 10/18] Nodes need boost
---
ci/das6/compile_and_run.sh | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ci/das6/compile_and_run.sh b/ci/das6/compile_and_run.sh
index 5d89c01..3641f71 100755
--- a/ci/das6/compile_and_run.sh
+++ b/ci/das6/compile_and_run.sh
@@ -11,7 +11,7 @@ COMPILER_VERSION=$2
echo RUNNING ON ${COMPILER_VERSION} AND ${ARCHITECTURE}
BUILD_DIR=build-${COMPILER_VERSION}-${ARCHITECTURE}
module load spack/${COMPILER_VERSION}
-module load cmake
+module load cmake boost
cmake -B ${BUILD_DIR} . -DCMAKE_BUILD_TYPE=Release
--
GitLab
From a4e5c2ca5de48f44df701e08439fda963f829fae Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 20:07:28 +0100
Subject: [PATCH 11/18] Add more dependencies
---
ci/das6/compile_and_run.sh | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ci/das6/compile_and_run.sh b/ci/das6/compile_and_run.sh
index 3641f71..4b8175d 100755
--- a/ci/das6/compile_and_run.sh
+++ b/ci/das6/compile_and_run.sh
@@ -11,7 +11,7 @@ COMPILER_VERSION=$2
echo RUNNING ON ${COMPILER_VERSION} AND ${ARCHITECTURE}
BUILD_DIR=build-${COMPILER_VERSION}-${ARCHITECTURE}
module load spack/${COMPILER_VERSION}
-module load cmake boost
+module load cmake boost casacore fftw3
cmake -B ${BUILD_DIR} . -DCMAKE_BUILD_TYPE=Release
--
GitLab
From bc4fd251874b2bcb640171defb78a685bbc184b5 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Thu, 6 Feb 2025 20:13:13 +0100
Subject: [PATCH 12/18] Fix wrong module name
---
ci/das6/compile_and_run.sh | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ci/das6/compile_and_run.sh b/ci/das6/compile_and_run.sh
index 4b8175d..34d5ef9 100755
--- a/ci/das6/compile_and_run.sh
+++ b/ci/das6/compile_and_run.sh
@@ -11,7 +11,7 @@ COMPILER_VERSION=$2
echo RUNNING ON ${COMPILER_VERSION} AND ${ARCHITECTURE}
BUILD_DIR=build-${COMPILER_VERSION}-${ARCHITECTURE}
module load spack/${COMPILER_VERSION}
-module load cmake boost casacore fftw3
+module load cmake boost casacore fftw
cmake -B ${BUILD_DIR} . -DCMAKE_BUILD_TYPE=Release
--
GitLab
From 0e46a84507c32c4bc8d94848f760e6809c428815 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Fri, 7 Feb 2025 09:00:50 +0100
Subject: [PATCH 13/18] Add hdf5
---
ci/das6/compile_and_run.sh | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ci/das6/compile_and_run.sh b/ci/das6/compile_and_run.sh
index 34d5ef9..9838a5d 100755
--- a/ci/das6/compile_and_run.sh
+++ b/ci/das6/compile_and_run.sh
@@ -11,7 +11,7 @@ COMPILER_VERSION=$2
echo RUNNING ON ${COMPILER_VERSION} AND ${ARCHITECTURE}
BUILD_DIR=build-${COMPILER_VERSION}-${ARCHITECTURE}
module load spack/${COMPILER_VERSION}
-module load cmake boost casacore fftw
+module load cmake boost casacore fftw hdf5
cmake -B ${BUILD_DIR} . -DCMAKE_BUILD_TYPE=Release
--
GitLab
From 6de58f42860c3f481b2ecbc60381ae6017ee04f0 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Fri, 7 Feb 2025 09:16:58 +0100
Subject: [PATCH 14/18] Add explicitly fftw3
---
CMakeLists.txt | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 26c5e45..f694c8f 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -71,12 +71,14 @@ file(GLOB TEST_SOURCES "test/*.cpp")
add_executable(unittests ${TEST_SOURCES} ${KERNEL_SOURCES})
find_package(OpenMP)
+find_package(PkgConfig REQUIRED)
+pkg_search_module(FFTW REQUIRED fftw3 IMPORTED_TARGET)
# Link against Google Benchmark
target_link_libraries(microbenchmarks PRIVATE benchmark::benchmark)
target_include_directories(microbenchmarks PRIVATE ${aocommon_SOURCE_DIR}/include ${schaapcommon_SOURCE_DIR}/include)
target_include_directories(microbenchmarks PRIVATE code)
-target_link_libraries(microbenchmarks PRIVATE schaapcommon)
+target_link_libraries(microbenchmarks PRIVATE schaapcommon PkgConfig::FFTW)
target_compile_options(microbenchmarks PUBLIC ${COMPILER_FLAGS})
if(OpenMP_CXX_FOUND)
target_link_libraries(microbenchmarks PRIVATE OpenMP::OpenMP_CXX)
@@ -91,7 +93,7 @@ catch_discover_tests(unittests WORKING_DIRECTORY)
target_link_libraries(unittests PRIVATE Catch2::Catch2WithMain)
target_include_directories(unittests PRIVATE code)
target_include_directories(unittests PRIVATE ${aocommon_SOURCE_DIR}/include ${schaapcommon_SOURCE_DIR}/include)
-target_link_libraries(unittests PRIVATE schaapcommon)
+target_link_libraries(unittests PRIVATE schaapcommon PkgConfig::FFTW)
target_compile_options(unittests PUBLIC ${COMPILER_FLAGS})
foreach(KERNEL_SOURCE ${KERNEL_SOURCES})
--
GitLab
From b590d541623bbe866c5457fdeafae072d6c0b940 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Fri, 7 Feb 2025 09:48:31 +0100
Subject: [PATCH 15/18] Add include to precompile
---
CMakeLists.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index f694c8f..cac174a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -100,7 +100,7 @@ foreach(KERNEL_SOURCE ${KERNEL_SOURCES})
get_filename_component(KERNEL_NAME ${KERNEL_SOURCE} NAME_WE)
add_precompile_target(TARGET_NAME ${KERNEL_NAME}
- INCLUDE_DIRS code ${aocommon_SOURCE_DIR}/include ${schaapcommon_SOURCE_DIR}/include
+ INCLUDE_DIRS code ${aocommon_SOURCE_DIR}/include ${schaapcommon_SOURCE_DIR}/include ${FFTW_INCLUDE_DIRS}
SOURCES ${KERNEL_SOURCE}
COMPILER_FLAGS ${COMPILER_FLAGS})
endforeach()
--
GitLab
From 070d07dbabcc3d0ea759f5071e382bd943e9eca6 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Fri, 7 Feb 2025 10:31:23 +0100
Subject: [PATCH 16/18] Add cfitsio
---
ci/das6/compile_and_run.sh | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ci/das6/compile_and_run.sh b/ci/das6/compile_and_run.sh
index 9838a5d..295945e 100755
--- a/ci/das6/compile_and_run.sh
+++ b/ci/das6/compile_and_run.sh
@@ -11,7 +11,7 @@ COMPILER_VERSION=$2
echo RUNNING ON ${COMPILER_VERSION} AND ${ARCHITECTURE}
BUILD_DIR=build-${COMPILER_VERSION}-${ARCHITECTURE}
module load spack/${COMPILER_VERSION}
-module load cmake boost casacore fftw hdf5
+module load cmake boost casacore fftw hdf5 cfitsio
cmake -B ${BUILD_DIR} . -DCMAKE_BUILD_TYPE=Release
--
GitLab
From 7a250ed25e1e8afc144cd32970b750987da21186 Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Fri, 7 Feb 2025 10:42:00 +0100
Subject: [PATCH 17/18] Fix naming
---
.gitlab-ci.yml | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index d4e9793..5f614c6 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -155,5 +155,5 @@ collect-performance:
- python3 ci/summarize-results.py --filter MatrixMultiplication results*.json result-summary-matrix-multiplication
- python3 ci/summarize-results.py --filter HermitianSquare results*.json result-summary-hermitian-square
- python3 ci/summarize-results.py --filter KroneckerSquare results*.json result-summary-kronecker-square
- - python3 ci/summarize-results.py --filter Convolution results*.json result-summary-convolution
+ - python3 ci/summarize-results.py --filter Convolve results*.json result-summary-convolution
--
GitLab
From 75da170fbd79219faf2a70f4f9a22a47fd75a9dd Mon Sep 17 00:00:00 2001
From: mancini <mancini@astron.nl>
Date: Fri, 7 Feb 2025 11:08:22 +0100
Subject: [PATCH 18/18] Reduce the number of tests
---
benchmarks/convolution.cpp | 10 ----------
1 file changed, 10 deletions(-)
diff --git a/benchmarks/convolution.cpp b/benchmarks/convolution.cpp
index ae82e9f..d1dbb4a 100644
--- a/benchmarks/convolution.cpp
+++ b/benchmarks/convolution.cpp
@@ -35,11 +35,6 @@ BENCHMARK_DEFINE_F(InitializeInput, ConvolveReference)
}
}
BENCHMARK_REGISTER_F(InitializeInput, ConvolveReference)
- ->Args({64, 32})
- ->Args({128, 64})
- ->Args({256, 126})
- ->Args({512, 754})
- ->Args({1024, 124})
->Args({2048, 1000})
->Args({4096, 5000});
@@ -53,10 +48,5 @@ BENCHMARK_DEFINE_F(InitializeInput, ConvolveSerial)
}
BENCHMARK_REGISTER_F(InitializeInput, ConvolveSerial)
- ->Args({64, 32})
- ->Args({128, 64})
- ->Args({256, 126})
- ->Args({512, 754})
- ->Args({1024, 124})
->Args({2048, 1000})
->Args({4096, 5000});
--
GitLab