Flatten apply functions

Author: OliverRietmann

.github/workflows/standalone-benchmark.yml

@@ -18,7 +18,7 @@ jobs:
       STANDALONE_DIR: /root/standalone
       BENCHMARK_CSV: standalone_cpu.csv
       PROFILER_CSV: profiler_cpu.csv
-      TIMING_CA: ./ca -e 50kHz -c --seed 0 --sync --runsInit 0 --PROCresetTimers 1 --PROCdebugMarkdown 1 --debug 1 # Add --runs 42 for benchmark runs
+      TIMING_CA: ./ca -e 50kHz -c --seed 0 --sync --runsInit 0 --PROCresetTimers 1 --PROCdebugMarkdown 1 # Add --runs 42 for benchmark runs
       CC: /opt/gcc/bin/gcc #/opt/clang-p2996/bin/clang
       CXX: /opt/gcc/bin/g++ #/opt/clang-p2996/bin/clang++
 
@@ -73,7 +73,7 @@ jobs:
           module load ninja/fortran-v1.11.1.g9-15 Vc/1.4.5-10 boost/v1.83.0-alice2-57 fmt/11.1.2-14 CMake/v3.31.6-10 ms_gsl/4.2.1-3 Clang/v20.1.7-9 TBB/v2022.3.0-3 ROOT/v6-36-04-alice9-15 ONNXRuntime/v1.22.0-71 GLFW/3.3.2-25
           export PATH=/opt/gcc/bin:$PATH && export LD_LIBRARY_PATH=/opt/gcc/lib64:/opt/gcc/lib:$LD_LIBRARY_PATH
           cd ${STANDALONE_DIR}
-          ${TIMING_CA} --runs 42 --PROCdebugCSV /root/${BENCHMARK_CSV}
+          ${TIMING_CA} --runs 12 --PROCdebugCSV /root/${BENCHMARK_CSV} --debug 1
           python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_standalone.py --discard 2 --input /root/${BENCHMARK_CSV} --output /root/summary_${BENCHMARK_CSV}
           rm -rf ${STANDALONE_DIR}/*.out ${STANDALONE_DIR}/events/o2-simple ${STANDALONE_DIR}/build
 

GPU/Common/MemLayout.h

@@ -100,33 +100,33 @@ constexpr auto apply_unary(Self &self, FunctionObject&& f) {
 
 // apply on skeleton struct S<F>
 template <class FunctionObject, template <template <class> class> class S, template <class> class F>
-constexpr auto apply(S<F> &self, FunctionObject&& f) {
+__attribute__((flatten)) constexpr auto apply(S<F> &self, FunctionObject&& f) {
     return apply_unary(self, std::forward<FunctionObject&&>(f));
 }
 
 template <class FunctionObject, template <template <class> class> class S, template <class> class F>
-constexpr auto apply(const S<F> &self, FunctionObject&& f) {
+__attribute__((flatten)) constexpr auto apply(const S<F> &self, FunctionObject&& f) {
     return apply_unary(self, std::forward<FunctionObject&&>(f));
 }
 
 // apply on wrappers, forwarding to the base type
 template <class FunctionObject, class Self> 
     requires requires { typename Self::Base; }
-constexpr auto apply(Self &self, FunctionObject&& f) {
+__attribute__((flatten)) constexpr auto apply(Self &self, FunctionObject&& f) {
     return apply_unary<typename Self::Base>(self, std::forward<FunctionObject&&>(f));
 }
 
 template <class FunctionObject, class Self> 
     requires requires { typename Self::Base; }
-constexpr auto apply(const Self &self, FunctionObject&& f) {
+__attribute__((flatten)) constexpr auto apply(const Self &self, FunctionObject&& f) {
     return apply_unary<const typename Self::Base>(self, std::forward<FunctionObject&&>(f));
 }
 
 
 // template <class FunctionObject, class Self, class Other>
 // constexpr auto apply(Self &self, Other &other, FunctionObject&& f) {
 template <class Self, class Other, class FunctionObject>
-constexpr auto apply_binary(Self &self, Other &other, FunctionObject&& f) {
+__attribute__((flatten)) constexpr auto apply_binary(Self &self, Other &other, FunctionObject&& f) {
     auto construct_output = [&]<size_t... Is>(std::index_sequence<Is...>) -> Self {
         return {f(
             self.[:nsdms(^^Self)[Is]:], other.[:nsdms(^^Other)[Is]:])...};
@@ -136,25 +136,24 @@ constexpr auto apply_binary(Self &self, Other &other, FunctionObject&& f) {
 }
 
 template <class FunctionObject, template <template <class> class> class S, template <class> class F_self, template <class> class F_other>
-constexpr auto apply(S<F_self> &self, S<F_other> &other, FunctionObject&& f) {
+__attribute__((flatten)) constexpr auto apply(S<F_self> &self, S<F_other> &other, FunctionObject&& f) {
     return apply_binary(self, other, std::forward<FunctionObject&&>(f));
 }
 
 template <class FunctionObject, template <template <class> class> class S, template <class> class F_self, template <class> class F_other>
-constexpr auto apply(S<F_self> &self, const S<F_other> &other, FunctionObject&& f) {
+__attribute__((flatten)) constexpr auto apply(S<F_self> &self, const S<F_other> &other, FunctionObject&& f) {
     return apply_binary(self, other, std::forward<FunctionObject&&>(f));
 }
 
 template <class Self, class Other, class FunctionObject>
     requires requires { typename Self::Base; typename Other::Base; }
-constexpr auto apply(Self &self, Other &other, FunctionObject&& f) {
+__attribute__((flatten)) constexpr auto apply(Self &self, Other &other, FunctionObject&& f) {
     return apply_binary<typename Self::Base, typename Other::Base>(self, other, std::forward<FunctionObject&&>(f));
 }
 
 template <class Self, class Other, class FunctionObject>
     requires requires { typename Self::Base; typename Other::Base; }
-constexpr auto apply(Self &self, const Other &other, FunctionObject&& f) {
-    static_assert(count_members<typename Self::Base>() == 4);
+__attribute__((flatten)) constexpr auto apply(Self &self, const Other &other, FunctionObject&& f) {
     return apply_binary<typename Self::Base, const typename Other::Base>(self, other, std::forward<FunctionObject&&>(f));
 }