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

[Larry Meadows via cfe-commits]

================
@@ -283,6 +283,9 @@ Error NativeRecordReplayTy::recordDescImpl(
   uint32_t MaxThreads = UserThreads
                             ? std::min(UserThreads, Kernel.getMaxThreads())
                             : Kernel.getMaxThreads();
+  // Record a bound the launch fits in: generic mode runs a warp more than the
+  // program asked for, and replay has to be allowed the same block.
+  MaxThreads = std::max(MaxThreads, Instance.NumThreads);
----------------
lfmeadow wrote:

This PR introduced this problem, and it is now resolved the same way as the launch.

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