[clang] [llvm] Fix Windows EH IP2State tables (remove +1 bias) (PR #144745)

[via cfe-commits]

sivadeilra wrote:

That was easier than expected -- thanks again, @efriedma-quic .  I've updated the PR.  I've also avoided adding NOPs after tail calls, which were happening in my first version of this.  There's never a need for NOP padding after TAILJMP* because there won't be a return address pointing into this frame.

https://github.com/llvm/llvm-project/pull/144745