[clang] [llvm] [SPIR-V] Report branch divergence in TTI (PR #214624)

[Arseniy Obolenskiy via cfe-commits]

aobolensk wrote:

> But ultimately.... do we have perf numbers with and without the patch? Better not on ML related workloads (where occupancy might be set in stone), but on some HPC applications?

In HeCbench: all binaries are the same, no effect of this change :/

Do you have any suggestions on other suite?

https://github.com/llvm/llvm-project/pull/214624