Merge pull request #8 from pravirkr/fdmt_gpu

GPU code for FDMT

Author: pravirkr

.clang-format

@@ -6,6 +6,7 @@ Language: Cpp
 AccessModifierOffset: -4
 AlignConsecutiveAssignments:
   Enabled: true
+BinPackParameters: false
 DerivePointerAlignment: false
 PackConstructorInitializers: CurrentLine
 PointerAlignment: Left

.clang-tidy

@@ -24,35 +24,34 @@ Checks: >
   -readability-isolate-declaration,
   -readability-magic-numbers,
 
-WarningsAsErrors: ''
+WarningsAsErrors: ""
 HeaderFilterRegex: ".*"
-FormatStyle: 'file'
+FormatStyle: "file"
 CheckOptions:
-  - { key: readability-identifier-naming.ClassCase,               value: CamelCase  }
-  - { key: readability-identifier-naming.ClassMemberCase,         value: lower_case }
-  - { key: readability-identifier-naming.EnumCase,                value: CamelCase  }
-  - { key: readability-identifier-naming.FunctionCase,            value: lower_case }
-  - { key: readability-identifier-naming.MacroDefinitionCase,     value: UPPER_CASE }
-  - { key: readability-identifier-naming.MemberCase,              value: lower_case }
-  - { key: readability-identifier-naming.PrivateMemberPrefix,     value: m_         }
-  - { key: readability-identifier-naming.ProtectedMemberPrefix,   value: m_         }
-  - { key: readability-identifier-naming.NamespaceCase,           value: lower_case }
-  - { key: readability-identifier-naming.ParameterCase,           value: lower_case }
-  - { key: readability-identifier-naming.TypeAliasCase,           value: CamelCase  }
-  - { key: readability-identifier-naming.TypedefCase,             value: CamelCase  }
-  - { key: readability-identifier-naming.VariableCase,            value: lower_case }
-  - { key: readability-identifier-naming.StaticVariableCase,      value: lower_case }
-  - { key: readability-identifier-naming.ConstexprVariableCase,   value: CamelCase  }
-  - { key: readability-identifier-naming.ConstexprVariablePrefix, value: k          }
-  - { key: readability-identifier-naming.EnumConstantCase,        value: CamelCase  }
-  - { key: readability-identifier-naming.EnumConstantPrefix,      value: k          }
-  - { key: readability-identifier-naming.GlobalConstantCase,      value: CamelCase  }
-  - { key: readability-identifier-naming.GlobalConstantPrefix,    value: k          }
-  - { key: readability-identifier-naming.MemberConstantCase,      value: CamelCase  }
-  - { key: readability-identifier-naming.MemberConstantPrefix,    value: k          }
-  - { key: readability-identifier-naming.StaticConstantCase,      value: CamelCase  }
-  - { key: readability-identifier-naming.StaticConstantPrefix,    value: k          }
-  - { key: readability-identifier-naming.StructCase,              value: CamelCase  }
-  - { key: readability-identifier-naming.TemplateParameterCase,   value: CamelCase  }
-...
-
+  readability-identifier-naming.ClassCase: "CamelCase"
+  readability-identifier-naming.ClassMemberCase: "lower_case"
+  readability-identifier-naming.EnumCase: "CamelCase"
+  readability-identifier-naming.FunctionCase: "lower_case"
+  readability-identifier-naming.MacroDefinitionCase: "UPPER_CASE"
+  readability-identifier-naming.MemberCase: "lower_case"
+  readability-identifier-naming.PrivateMemberPrefix: "m_"
+  readability-identifier-naming.ProtectedMemberPrefix: "m_"
+  readability-identifier-naming.NamespaceCase: "lower_case"
+  readability-identifier-naming.ParameterCase: "lower_case"
+  readability-identifier-naming.TypeAliasCase: "CamelCase"
+  readability-identifier-naming.TypedefCase: "CamelCase"
+  readability-identifier-naming.VariableCase: "lower_case"
+  readability-identifier-naming.StaticVariableCase: "lower_case"
+  readability-identifier-naming.ConstexprVariableCase: "CamelCase"
+  readability-identifier-naming.ConstexprVariablePrefix: "k"
+  readability-identifier-naming.EnumConstantCase: "CamelCase"
+  readability-identifier-naming.EnumConstantPrefix: "k"
+  readability-identifier-naming.GlobalConstantCase: "CamelCase"
+  readability-identifier-naming.GlobalConstantPrefix: "k"
+  readability-identifier-naming.MemberConstantCase: "CamelCase"
+  readability-identifier-naming.MemberConstantPrefix: "k"
+  readability-identifier-naming.StaticConstantCase: "CamelCase"
+  readability-identifier-naming.StaticConstantPrefix: "k"
+  readability-identifier-naming.StructCase: "CamelCase"
+  readability-identifier-naming.TemplateParameterCase: "CamelCase"
+  readability-function-cognitive-complexity.Threshold: "30"

.github/workflows/ci.yml

@@ -13,9 +13,9 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest]
+        os: [ubuntu-22.04]
         build_type: [Debug]
-        compiler: [{cpp: g++-13, code-cov: true, gcov: gcov-13}, {cpp: clang++-15}]
+        compiler: [{cpp: g++-12, code-cov: true, gcov: gcov-12}, {cpp: clang++-15}]
     steps:
     - uses: actions/checkout@v4
     - name: Set reusable strings

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.15)
+cmake_minimum_required(VERSION 3.18)
 cmake_policy(SET CMP0148 OLD)
 
 project(
@@ -13,25 +13,43 @@ option(BUILD_PYTHON "Build Python bindings" ON)
 option(BUILD_TESTING "Build tests" OFF)
 option(BUILD_BENCHMARKS "Build benchmarks" OFF)
 option(CODE_COVERAGE "Enable coverage reporting" OFF)
-option(ENABLE_FAST_MATH "Enable fast math flags" ON)
 
 set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
+set(CMAKE_INTERPROCEDURAL_OPTIMIZATION ON)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${PROJECT_SOURCE_DIR}/cmake)
 
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wextra")
-set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -g")
-set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -O3 -DNDEBUG -march=native")
-
-if(ENABLE_FAST_MATH)
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ffast-math")
-endif()
+set(CMAKE_CXX_FLAGS "-Wall -Wextra")
+set(CMAKE_CXX_FLAGS_DEBUG "-g")
+set(CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG -march=native -ffast-math")
 
 if(NOT CMAKE_BUILD_TYPE)
   set(CMAKE_BUILD_TYPE Release)
 endif()
 
+# Find CUDA and set up the project
+include(CheckLanguage)
+check_language(CUDA)
+
+if(CMAKE_CUDA_COMPILER)
+  enable_language(CUDA)
+  if(CMAKE_CUDA_COMPILER_VERSION VERSION_LESS_EQUAL 10.0.0)
+    message(FATAL_ERROR "Found CUDA ${CMAKE_CUDA_COMPILER_VERSION}. Need CUDA 10+")
+  else()
+    message(STATUS "Found CUDA ${CMAKE_CUDA_COMPILER_VERSION}.")
+  endif()
+
+  set(CMAKE_CUDA_STANDARD 17)
+  set(CMAKE_CUDA_STANDARD_REQUIRED ON)
+  set(CMAKE_CUDA_FLAGS "-c -Xcompiler=-Wall,-Wextra --expt-extended-lambda")
+  set(CMAKE_CUDA_FLAGS_DEBUG "-g")
+  set(CMAKE_CUDA_FLAGS_RELEASE "-O3 -DNDEBUG -use_fast_math -Xcompiler=-march=native")
+  set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_INCLUDES 0)
+elseif(NOT CMAKE_CUDA_COMPILER)
+  message(STATUS "Not building gpu code. No CUDA compiler found...")
+endif(CMAKE_CUDA_COMPILER)
+
 # Initialize some default paths
 include(GNUInstallDirs)
 include(cmake/CPM.cmake)
@@ -45,6 +63,7 @@ endif()
 
 # Build the C++ library
 set(LIBRARY_NAME dmt)
+set(PYLIBRARY_NAME dmtlib)
 add_subdirectory(lib)
 # Build the Python bindings
 if(BUILD_PYTHON)

README.md

@@ -23,10 +23,10 @@ pip install -U git+https://github.com/pravirkr/dmt
 ## Usage
 
 ```python
-from dmt.libdmt import FDMT
+from dmtlib import FDMTCPU
 
 frb = np.ones((nchans, nsamps), dtype=np.float32)
-thefdmt = FDMT(f_min, f_max, nchans, nsamps, tsamp, dt_max=dt_max, dt_min=0, dt_step=1)
+thefdmt = FDMTCPU(f_min, f_max, nchans, nsamps, tsamp, dt_max=dt_max, dt_min=0, dt_step=1)
 dmt_transform = thefdmt.execute(frb.astype(np.float32))
 ```
 
@@ -36,6 +36,6 @@ dmt_transform = thefdmt.execute(frb.astype(np.float32))
 f_min = 704.0, f_max = 1216.0, nchans = 4096, tsamp = 0.00008192, dt_max = 2048, nsamps = n;
 nthreads = 1, 8;
 ```
-![](bench/results/bench_m1.png)
+![](bench/results/bench.png)
 
 

bench/CMakeLists.txt

@@ -1,13 +1,19 @@
 CPMAddPackage(
   NAME benchmark
-  VERSION 1.8.3
+  VERSION 1.8.4
   GITHUB_REPOSITORY google/benchmark
   OPTIONS "BENCHMARK_ENABLE_WERROR OFF" "BENCHMARK_FORCE_WERROR OFF" "BENCHMARK_ENABLE_TESTING OFF"
 )
 
 set(TARGET_BENCHMARKS ${PROJECT_NAME}_bench)
-file(GLOB TEST_SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/*.cpp)
+file(GLOB BENCH_SOURCES ${CMAKE_CURRENT_SOURCE_DIR}/*.cpp)
 
-add_executable(${TARGET_BENCHMARKS} ${TEST_SOURCES})
+# Check if CUDA was found
+if(CMAKE_CUDA_COMPILER)
+  file(GLOB BENCH_SOURCES_CUDA ${CMAKE_CURRENT_SOURCE_DIR}/*.cu)
+  list(APPEND BENCH_SOURCES ${BENCH_SOURCES_CUDA})
+endif()
+
+add_executable(${TARGET_BENCHMARKS} ${BENCH_SOURCES})
 target_include_directories(${TARGET_BENCHMARKS} PUBLIC ${CMAKE_SOURCE_DIR}/include)
-target_link_libraries(${TARGET_BENCHMARKS} PRIVATE ${LIBRARY_NAME} benchmark::benchmark)
+target_link_libraries(${TARGET_BENCHMARKS} PRIVATE ${LIBRARY_NAME} benchmark::benchmark_main)

bench/bench_plots.py

@@ -22,7 +22,7 @@
     "/BM_|_|/",
     expand=True,
 ).apply(
-    lambda x: ["_".join([x[1], x[2]]), "_".join([x[3], x[4]]), x[5]],
+    lambda x: ["_".join([x[1], x[2]]), x[3], x[4]],
     axis=1,
     result_type="expand",
 )
@@ -48,8 +48,8 @@
     ax.set_title(benchmark)
     ax.set_xscale("log")
     ax.set_yscale("log")
-    ax.set_xlabel("n")
-    ax.set_ylabel("Time (s)")
+    ax.set_xlabel("nsamples")
+    ax.set_ylabel("Time (ms)")
     ax.legend()
 
 fig.tight_layout()

bench/fdmt_b.cpp renamed to bench/fdmt_cpu_b.cpp

@@ -5,7 +5,7 @@
 #include <random>
 #include <vector>
 
-class FDMTFixture : public benchmark::Fixture {
+class FDMTCPUFixture : public benchmark::Fixture {
 public:
     void SetUp(const ::benchmark::State& state) override {
         f_min  = 704.0F;
@@ -34,13 +34,22 @@ class FDMTFixture : public benchmark::Fixture {
     size_t nsamps{};
 };
 
-BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_plan_seq_cpu)(benchmark::State& state) {
+BENCHMARK_DEFINE_F(FDMTCPUFixture, BM_fdmt_planBuffer_seq)
+(benchmark::State& state) {
+    for (auto _ : state) {
+        FDMTCPU fdmt(f_min, f_max, nchans, nsamps, tsamp, dt_max);
+    }
+}
+
+BENCHMARK_DEFINE_F(FDMTCPUFixture, BM_fdmt_planBuffer_par)
+(benchmark::State& state) {
+    FDMTCPU::set_num_threads(8);
     for (auto _ : state) {
         FDMTCPU fdmt(f_min, f_max, nchans, nsamps, tsamp, dt_max);
     }
 }
 
-BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_initialise_seq_cpu)
+BENCHMARK_DEFINE_F(FDMTCPUFixture, BM_fdmt_initialise_seq)
 (benchmark::State& state) {
     FDMTCPU::set_num_threads(1);
     FDMTCPU fdmt(f_min, f_max, nchans, nsamps, tsamp, dt_max);
@@ -52,11 +61,12 @@ BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_initialise_seq_cpu)
     const auto state_size = plan.state_shape[0][3] * plan.state_shape[0][4];
     std::vector<float> state_init(state_size, 0.0F);
     for (auto _ : state) {
-        fdmt.initialise(waterfall.data(), state_init.data());
+        fdmt.initialise(waterfall.data(), waterfall.size(), state_init.data(),
+                        state_init.size());
     }
 }
 
-BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_initialise_par_cpu)
+BENCHMARK_DEFINE_F(FDMTCPUFixture, BM_fdmt_initialise_par)
 (benchmark::State& state) {
     FDMTCPU::set_num_threads(8);
     FDMTCPU fdmt(f_min, f_max, nchans, nsamps, tsamp, dt_max);
@@ -68,11 +78,12 @@ BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_initialise_par_cpu)
     const auto state_size = plan.state_shape[0][3] * plan.state_shape[0][4];
     std::vector<float> state_init(state_size, 0.0F);
     for (auto _ : state) {
-        fdmt.initialise(waterfall.data(), state_init.data());
+        fdmt.initialise(waterfall.data(), waterfall.size(), state_init.data(),
+                        state_init.size());
     }
 }
 
-BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_execute_seq_cpu)
+BENCHMARK_DEFINE_F(FDMTCPUFixture, BM_fdmt_execute_seq)
 (benchmark::State& state) {
     FDMTCPU::set_num_threads(1);
     FDMTCPU fdmt(f_min, f_max, nchans, nsamps, tsamp, dt_max);
@@ -87,7 +98,7 @@ BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_execute_seq_cpu)
     }
 }
 
-BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_execute_par_cpu)
+BENCHMARK_DEFINE_F(FDMTCPUFixture, BM_fdmt_execute_par)
 (benchmark::State& state) {
     FDMTCPU::set_num_threads(8);
     FDMTCPU fdmt(f_min, f_max, nchans, nsamps, tsamp, dt_max);
@@ -102,7 +113,7 @@ BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_execute_par_cpu)
     }
 }
 
-BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_overall_seq_cpu)
+BENCHMARK_DEFINE_F(FDMTCPUFixture, BM_fdmt_overall_seq)
 (benchmark::State& state) {
     std::random_device rd;
     std::mt19937 gen(rd());
@@ -119,7 +130,7 @@ BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_overall_seq_cpu)
     }
 }
 
-BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_overall_par_cpu)
+BENCHMARK_DEFINE_F(FDMTCPUFixture, BM_fdmt_overall_par)
 (benchmark::State& state) {
     std::random_device rd;
     std::mt19937 gen(rd());
@@ -139,26 +150,29 @@ BENCHMARK_DEFINE_F(FDMTFixture, BM_fdmt_overall_par_cpu)
 constexpr size_t kMinNsamps = 1 << 11;
 constexpr size_t kMaxNsamps = 1 << 16;
 
-BENCHMARK_REGISTER_F(FDMTFixture, BM_fdmt_plan_seq_cpu)
+BENCHMARK_REGISTER_F(FDMTCPUFixture, BM_fdmt_planBuffer_seq)
+    ->RangeMultiplier(2)
+    ->Range(kMinNsamps, kMaxNsamps);
+BENCHMARK_REGISTER_F(FDMTCPUFixture, BM_fdmt_planBuffer_par)
     ->RangeMultiplier(2)
     ->Range(kMinNsamps, kMaxNsamps);
-BENCHMARK_REGISTER_F(FDMTFixture, BM_fdmt_initialise_seq_cpu)
+BENCHMARK_REGISTER_F(FDMTCPUFixture, BM_fdmt_initialise_seq)
     ->RangeMultiplier(2)
     ->Range(kMinNsamps, kMaxNsamps);
-BENCHMARK_REGISTER_F(FDMTFixture, BM_fdmt_initialise_par_cpu)
+BENCHMARK_REGISTER_F(FDMTCPUFixture, BM_fdmt_initialise_par)
     ->RangeMultiplier(2)
     ->Range(kMinNsamps, kMaxNsamps);
-BENCHMARK_REGISTER_F(FDMTFixture, BM_fdmt_execute_seq_cpu)
+BENCHMARK_REGISTER_F(FDMTCPUFixture, BM_fdmt_execute_seq)
     ->RangeMultiplier(2)
     ->Range(kMinNsamps, kMaxNsamps);
-BENCHMARK_REGISTER_F(FDMTFixture, BM_fdmt_execute_par_cpu)
+BENCHMARK_REGISTER_F(FDMTCPUFixture, BM_fdmt_execute_par)
     ->RangeMultiplier(2)
     ->Range(kMinNsamps, kMaxNsamps);
-BENCHMARK_REGISTER_F(FDMTFixture, BM_fdmt_overall_seq_cpu)
+BENCHMARK_REGISTER_F(FDMTCPUFixture, BM_fdmt_overall_seq)
     ->RangeMultiplier(2)
     ->Range(kMinNsamps, kMaxNsamps);
-BENCHMARK_REGISTER_F(FDMTFixture, BM_fdmt_overall_par_cpu)
+BENCHMARK_REGISTER_F(FDMTCPUFixture, BM_fdmt_overall_par)
     ->RangeMultiplier(2)
     ->Range(kMinNsamps, kMaxNsamps);
 
-BENCHMARK_MAIN();
+// BENCHMARK_MAIN();