CODEX: Add production-boundary timing gates

Co-authored-by: Codex <codex@elphel.com>

jna/test_lma_products_jna.cu

@@ -7,12 +7,14 @@
 //   cd tile_processor_gpu && /usr/local/cuda-12.8/bin/nvcc -std=c++17 -O2 \
 //     jna/test_lma_products_jna.cu -o tests_bin/test_lma_products_jna \
 //     -L jna -ltileproc -Xlinker -rpath -Xlinker $PWD/jna
+// Run: tests_bin/test_lma_products_jna [srcdir] [libcudadevrt.a] [profile_calls]
 //
 // Created Jul 14, 2026 by Codex for Elphel.
 
 #include <cstdio>
 #include <cstdlib>
 #include <cstring>
+#include <chrono>
 #include <vector>
 
 extern "C" {
@@ -37,6 +39,8 @@ int main(int argc, char** argv)
   const char* srcdir = (argc > 1) ? argv[1] : "src";
   const char* devrt = (argc > 2) ? argv[2] :
       "/usr/local/cuda-12.8/targets/x86_64-linux/lib/libcudadevrt.a";
+  const int profile_calls = (argc > 3) ? std::atoi(argv[3]) : 0;
+  if(profile_calls < 0){ std::fprintf(stderr,"profile_calls must be >= 0\n"); return EXIT_FAILURE; }
   void* module = tp_create_module(srcdir, devrt);
   if(!module){ std::fprintf(stderr,"tp_create_module: %s\n",tp_last_error()); return EXIT_FAILURE; }
   void* proc = tp_proc_create(module);
@@ -79,6 +83,25 @@ int main(int argc, char** argv)
   }
   std::printf("NVRTC lma_normal_products: %zu doubles, bit-exact -> %s\n",
       got.size(),bad?"FAIL":"PASS");
+  if(!bad && profile_calls > 0){
+    for(int i=0;i<20;i++) if(tp_proc_exec_lma_products(
+        proc,weights.data(),jt.data(),ymfx.data(),np,nv,got.data())){
+      std::fprintf(stderr,"profile warmup: %s\n",tp_last_error()); bad=1; break;
+    }
+    const auto profile = [&](){
+      const auto t0=std::chrono::steady_clock::now();
+      for(int i=0;i<profile_calls;i++) if(tp_proc_exec_lma_products(
+          proc,weights.data(),jt.data(),ymfx.data(),np,nv,got.data())) return -1.0;
+      const auto t1=std::chrono::steady_clock::now();
+      return std::chrono::duration<double,std::milli>(t1-t0).count()/profile_calls;
+    };
+    if(!bad){
+      const double a=profile(), b=profile();
+      if((a < 0)||(b < 0)){ std::fprintf(stderr,"profile: %s\n",tp_last_error()); bad=1; }
+      else std::printf("LMA products profile (%d calls, 3x303 -> 12 doubles): rounds=[%.6f, %.6f] ms, avg=%.6f ms/call\n",
+          profile_calls,a,b,0.5*(a+b));
+    }
+  }
   tp_proc_destroy(proc); tp_destroy_module(module);
   return bad?EXIT_FAILURE:EXIT_SUCCESS;
 }