diff --git a/library/src/rocfft_aot_helper.cpp b/library/src/rocfft_aot_helper.cpp
index f0a889f4..452eb37f 100644
--- a/library/src/rocfft_aot_helper.cpp
+++ b/library/src/rocfft_aot_helper.cpp
@@ -771,26 +771,22 @@ int main(int argc, char** argv)
     for(size_t i = 0; i < NUM_THREADS; ++i)
     {
         threads.emplace_back([&queue, &gpu_archs]() {
+            int compile_count = 0;
             while(true)
             {
                 auto item = queue.pop();
                 if(item.kernel_name.empty())
                     break;
 
+                if(++compile_count % 16 == 0)
+                    std::cerr << "rocfft_aot_helper processing " << item.kernel_name << std::endl << std::flush;
                 for(const auto& gpu_arch : gpu_archs)
                 {
-                    if(item.sol_arch_name.empty())
+                    if(item.sol_arch_name.empty() || gpu_arch.find(item.sol_arch_name) != std::string::npos)
                     {
                         RTCCache::cached_compile(
                             item.kernel_name, gpu_arch, item.generate_src, generator_sum());
                     }
-                    else if(gpu_arch.find(item.sol_arch_name) != std::string::npos)
-                    {
-                        // std::cout << "arch: " << gpu_arch
-                        //           << ", solution-kernel: " << item.kernel_name << std::endl;
-                        RTCCache::cached_compile(
-                            item.kernel_name, gpu_arch, item.generate_src, generator_sum());
-                    }
                 }
             }
         });