[PATCH] D62937: [AMDGPU] Mark s_endpgm convergent

Stanislav Mekhanoshin via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Jun 5 19:10:08 PDT 2019


rampitec added a comment.

In D62937#1531865 <https://reviews.llvm.org/D62937#1531865>, @arsenm wrote:

> The concept of convergence we have breaks down at the machine level. At the machine level, we don't exactly need to think about the operations treated in the IR as convergent as related to control flow anymore. s_endpgm doesn't really have the convergent properties anyway


You cannot duplicate it by moving into the control flow. If you move it into the 'if..then' block that will prevent subsequent 'else' from from execution. How do you propose to shield it from duplication?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62937/new/

https://reviews.llvm.org/D62937





More information about the llvm-commits mailing list