[clang] [llvm] [mlir] [OpenMP] Reserve the main thread's warp for generic mode kernels (PR #218790)

[Robert Imschweiler via cfe-commits]

================
@@ -279,10 +279,7 @@ Error NativeRecordReplayTy::recordDescImpl(
 
   // Export minimum and maximum for allowed number of threads. If zero, it means
   // there was no restriction provided by the program.
-  uint32_t UserThreads = std::max(LaunchArgs.UserThreadLimit[0], uint32_t(0));
-  uint32_t MaxThreads = UserThreads
-                            ? std::min(UserThreads, Kernel.getMaxThreads())
-                            : Kernel.getMaxThreads();
+  uint32_t MaxThreads = Kernel.getMaxThreads();
----------------
ro-i wrote:

ping @kevinsala for awareness (see https://github.com/llvm/llvm-project/pull/200639)

https://github.com/llvm/llvm-project/pull/218790