Fixes from Antonio. 1. I had -D_GLIBCXX_DEBUG=1 -D_GLIBCXX_DEBUG_PEDANTIC=1 in the meson.build. 2. The moa_many_for test had a copy-pasta extra extra for-loop that made it slow. Fix those two and the results match expectation with auto&:for doing better than indexing.

meson.build

@@ -2,7 +2,7 @@ project('sol2_test', 'cpp',
   version: '0.1.0',
   default_options: [
     'cpp_std=c++23',
-    'cpp_args=-D_GLIBCXX_DEBUG=1 -D_GLIBCXX_DEBUG_PEDANTIC=1',
+    # 'cpp_args=-D_GLIBCXX_DEBUG=1 -D_GLIBCXX_DEBUG_PEDANTIC=1',
   ])
 
 # use this for common options only for our executables

tests/perf_tests.cpp

@@ -360,56 +360,57 @@ namespace perf_tests {
     Stats moa_many_for;
     for(int i = 0; i < sample_number; i++) {
       auto start = moa_many_for.time_start();
+
+      for(auto& el : moa->s1) {
+        el.a01 += el.a02 + i;
+        el.a02 += el.a03 + i;
+        el.a03 += el.a04 + i;
+      }
+
+      for(int i = 0; i < sample_size; i++) {
+        moa->s1[i].a04 += moa->s2[i].a05 + i;
+      }
+
+      for(auto& el : moa->s2) {
+        el.a05 += el.a06 + i;
+        el.a06 += el.a07 + i;
+        el.a07 += el.a08 + i;
+      }
+
+      for(int i = 0; i < sample_size; i++) {
+        moa->s2[i].a08 += moa->s3[i].a09 + i;
+      }
+
+
+      for(auto& el : moa->s3) {
+        el.a09 += el.a10 + i;
+        el.a10 += el.a11 + i;
+        el.a11 += el.a12 + i;
+      }
+
       for(int i = 0; i < sample_size; i++) {
-        for(auto& el : moa->s1) {
+        moa->s3[i].a12 += moa->s4[i].a13 + i;
-          el.a01 += el.a02 + i;
-          el.a02 += el.a03 + i;
-          el.a03 += el.a04 + i;
-        }
-
-        for(int i = 0; i < sample_size; i++) {
-          moa->s1[i].a04 += moa->s2[i].a05 + i;
-        }
-
-        for(auto& el : moa->s2) {
-          el.a05 += el.a06 + i;
-          el.a06 += el.a07 + i;
-          el.a07 += el.a08 + i;
-        }
-
-        for(int i = 0; i < sample_size; i++) {
-          moa->s2[i].a08 += moa->s3[i].a09 + i;
-        }
-
-
-        for(auto& el : moa->s3) {
-          el.a09 += el.a10 + i;
-          el.a10 += el.a11 + i;
-          el.a11 += el.a12 + i;
-        }
-
-        for(int i = 0; i < sample_size; i++) {
-          moa->s3[i].a12 += moa->s4[i].a13 + i;
-        }
-
-        for(auto& el : moa->s4) {
-          el.a13 += el.a14 + i;
-          el.a14 += el.a15 + i;
-          el.a15 += el.a16 + i;
-        }
-
-        for(int i = 0; i < sample_size; i++) {
-          moa->s4[i].a16 += moa->s5[i].a17 + i;
-        }
-
-        for(auto& el : moa->s5) {
-          el.a17 += el.a18 + i;
-          el.a18 += el.a19 + i;
-          el.a19 += el.a20 + i;
-        }
       }
+
+      for(auto& el : moa->s4) {
+        el.a13 += el.a14 + i;
+        el.a14 += el.a15 + i;
+        el.a15 += el.a16 + i;
+      }
+
+      for(int i = 0; i < sample_size; i++) {
+        moa->s4[i].a16 += moa->s5[i].a17 + i;
+      }
+
+      for(auto& el : moa->s5) {
+        el.a17 += el.a18 + i;
+        el.a18 += el.a19 + i;
+        el.a19 += el.a20 + i;
+      }
+
       moa_many_for.sample_time(start);
     }
+
     moa_many_for.dump("moa many for");
     fmt::println("-------------\n");
     samples.moa_many_for.push_back(moa_many_for);