[llvm-branch-commits] [llvm] [openmp] [OpenMP][offload] Trade threads for blocks in Generic-SPMD kernels (PR #234191)

[Robert Imschweiler via llvm-branch-commits]

================
@@ -101,6 +101,10 @@ struct DeviceGeometryPropsTy {
   uint32_t DefaultNumBlocks = 0;
   /// Maximum amount of shared memory, in bytes, a block can use in a kernel.
   uint64_t MaxBlockSharedMemSize = 0;
+  /// Number of compute units, or zero if the device does not report it.
+  uint32_t NumComputeUnits = 0;
+  /// Backend of the platform of the device.
+  ol_platform_backend_t Backend = OL_PLATFORM_BACKEND_UNKNOWN;
----------------
ro-i wrote:

factored it out to https://github.com/llvm/llvm-project/pull/234973.

I wonder if the multiverse implodes if I try to use GitHubs stacked PR feature for this situation where two PRs share the same base PR...

https://github.com/llvm/llvm-project/pull/234191