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

Johannes Doerfert llvmlistbot at llvm.org
Thu May 7 17:46:55 PDT 2026


================
@@ -268,8 +289,9 @@ Error GenericKernelTy::launch(GenericDeviceTy &GenericDevice, void **ArgPtrs,
     AsyncInfoWrapper.freeAllocationAfterSynchronization(
         DynBlockMemConf.FallbackPtr);
 
----------------
jdoerfert wrote:

This is an (existing) mess. We should not have new arrays for NumThreads/Blocks that might carry different values than the ones in KernelArgs. Maybe we can address that first and thereby certainly avoid the extra argument.

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


More information about the Mlir-commits mailing list