[llvm] [AMDGPU] Improve execz branch removal for async load-to-LDS (PR #192341)

[Juan Manuel Martinez Caamaño via llvm-commits]

jmmartinez wrote:

Left a comment on the ticket.

This problem is likely the case for other instructions. At the time, we played a bit with using the `MachineTraceAnalysis` to obtain a better estimate.

Have you tried using it to see if it could fix your problem?

https://github.com/llvm/llvm-project/pull/192341