[Openmp-commits] [clang] [llvm] [mlir] [openmp] [OpenMP][offload] Cross-team reductions with variable number of teams (PR #195102)

[Robert Imschweiler via Openmp-commits]

================
@@ -452,11 +452,14 @@ struct GenericKernelTy {
   }
 
   /// Return a device pointer to a new kernel launch environment.
-  Expected<KernelLaunchEnvironmentTy *>
-  getKernelLaunchEnvironment(GenericDeviceTy &GenericDevice,
-                             const KernelArgsTy &KernelArgs,
-                             const DynBlockMemConfTy &DynBlockMemConf,
-                             AsyncInfoWrapperTy &AsyncInfoWrapper) const;
+  ///
+  /// \p NumBlocks0 is the number of teams for this launch and is used to size
+  /// the teams-reduction buffer when the kernel environment's
+  /// ReductionBufferLength is 0 (i.e. unspecified at compile time).
----------------
ro-i wrote:

yep, thanks

https://github.com/llvm/llvm-project/pull/195102