Use --debug only with standalone

Author: OliverRietmann

.github/workflows/standalone-benchmark.yml

@@ -53,7 +53,7 @@ jobs:
       STANDALONE_DIR: /root/standalone
       BENCHMARK_CSV: standalone_${{ matrix.name }}.csv
       PROFILER_CSV: profiler_${{ matrix.name }}.csv
-      TIMING_CA: ./ca -e 50kHz ${{ matrix.cpu_gpu }} --seed 0 --sync --runsInit 0 --PROCresetTimers 1 --PROCdebugMarkdown 1 --debug 1 # Add --runs 42 for benchmark runs
+      TIMING_CA: ./ca -e 50kHz ${{ matrix.cpu_gpu }} --seed 0 --sync --runsInit 0 --PROCresetTimers 1 --PROCdebugMarkdown 1 # Add --runs 42 for benchmark runs
       LD_LIBRARY_PATH: /usr/local/cuda-13.0/compat
 
     name: ${{ matrix.name }}
@@ -99,13 +99,13 @@ jobs:
         env:
           DETERMINISTIC_MODE: OFF
 
-      - name: Benchmark Track Reconstruction
+      - name: Standalone Benchmark Track
         run: |
           source /etc/profile.d/modules.sh
           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
           cd ${STANDALONE_DIR}
-          ${TIMING_CA} --runs ${{ matrix.standalone_runs }} --PROCdebugCSV /root/${BENCHMARK_CSV}
-          python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_standalone.py --discard 0 --input /root/${BENCHMARK_CSV} --output /root/summary_${BENCHMARK_CSV}
+          ${TIMING_CA} --runs ${{ matrix.standalone_runs }} --PROCdebugCSV /root/${BENCHMARK_CSV} --debug 1
+          python3 ${GITHUB_WORKSPACE}/.github/scripts/profiler_standalone.py --discard 2 --input /root/${BENCHMARK_CSV} --output /root/summary_${BENCHMARK_CSV}
       
       - name: Profiler - Nsight Compute
         if: ${{ matrix.name == 'nvidia-h100' }}

GPU/Common/MemLayout.h

@@ -292,20 +292,20 @@ struct interface<S, F, Flag::soa> { using type = wrapper<S, F>; };
 
 #define MEMLAYOUT_APPLY_UNARY(...)\
     template <class Function>\
-    constexpr auto apply(Function&& f) { return f(__VA_ARGS__); }\
+    __attribute__((flatten)) constexpr auto apply(Function&& f) { return f(__VA_ARGS__); }\
     template <class Function>\
-    constexpr auto apply(Function&& f) const { return f(__VA_ARGS__); }\
+    __attribute__((flatten)) constexpr auto apply(Function&& f) const { return f(__VA_ARGS__); }\
 
 #define MEMLAYOUT_EXPAND(m) f(m, other.m)
 
 #define MEMLAYOUT_APPLY_BINARY(STRUCT_NAME, ...)\
     template <template <class> class F_other, class Function>\
-    constexpr STRUCT_NAME apply(STRUCT_NAME<F_other>& other, Function&& f) { return {__VA_ARGS__}; }\
+    __attribute__((flatten)) constexpr STRUCT_NAME apply(STRUCT_NAME<F_other>& other, Function&& f) { return {__VA_ARGS__}; }\
     template <template <class> class F_other, class Function>\
-    constexpr STRUCT_NAME apply(STRUCT_NAME<F_other>& other, Function&& f) const { return {__VA_ARGS__}; }\
+    __attribute__((flatten)) constexpr STRUCT_NAME apply(STRUCT_NAME<F_other>& other, Function&& f) const { return {__VA_ARGS__}; }\
     template <template <class> class F_other, class Function>\
-    constexpr STRUCT_NAME apply(const STRUCT_NAME<F_other>& other, Function&& f) { return {__VA_ARGS__}; }\
+    __attribute__((flatten)) constexpr STRUCT_NAME apply(const STRUCT_NAME<F_other>& other, Function&& f) { return {__VA_ARGS__}; }\
     template <template <class> class F_other, class Function>\
-    constexpr STRUCT_NAME apply(const STRUCT_NAME<F_other>& other, Function&& f) const { return {__VA_ARGS__}; }\
+    __attribute__((flatten)) constexpr STRUCT_NAME apply(const STRUCT_NAME<F_other>& other, Function&& f) const { return {__VA_ARGS__}; }\
 
 #endif // MEMLAYOUT_H