[PATCH] D81114: [OpenMPOpt] initial tests for ICV tracking. Only nthreads is used.

[Stefan Stipanovic via Phabricator via llvm-commits]

sstefan1 marked an inline comment as done.
sstefan1 added a comment.

In D81114#2072328 <https://reviews.llvm.org/D81114#2072328>, @JonChesterfield wrote:

> It might be worth removing some more of the noise from the IR, e.g. the 'noalias nocapture' stuff on the arguments is unlikely to be relevant.


Agreed. I already cleaned some stuff up. Will do more.



================
Comment at: llvm/test/Transforms/OpenMP/icv_tracking.ll:61
+  tail call void @omp_set_num_threads(i32 10) #1
+  %5 = tail call i32 @omp_get_max_threads() #1
+  tail call void @use(i32 %5) #1
----------------
jdoerfert wrote:
> I guess `%5` should be replaced with `min(%3, 10)`, right? We might want to use SCEV to represent the values and combine them.
Not sure I understand why `min(%3, 10)`? Without any transformations `%5` is 10. Maybe I'm missing something?

Will look into using SCEV.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D81114/new/

https://reviews.llvm.org/D81114