[llvm] [Offload] Allow CUDA Kernels to use arbitrarily large shared memory (PR #145963)

[Shilei Tian via llvm-commits]

================
@@ -1302,6 +1305,16 @@ Error CUDAKernelTy::launchImpl(GenericDeviceTy &GenericDevice,
   if (GenericDevice.getRPCServer())
     GenericDevice.Plugin.getRPCServer().Thread->notify();
 
+  // In case we require more memory than the current limit.
+  if (MaxDynCGroupMem >= MaxDynCGroupMemLimit) {
----------------
shiltian wrote:

Hmm, is this code dedicated for OpenMP? I know it is, for now, but we might want to consider to check what the actual "user" here.

https://github.com/llvm/llvm-project/pull/145963