[PATCH] D29214: [AMDGPU] Internalize non-kernel symbols

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

I don't see why the backend needs to do this. Why isn't internalize run when the library is linked?



================
Comment at: lib/Target/AMDGPU/AMDGPUTargetMachine.cpp:225-226
+          if (const Function *F = dyn_cast<Function>(&GV)) {
+            if (F->isDeclaration())
+                return true;
+            switch (F->getCallingConv()) {
----------------
Shouldn't this be return false?


Repository:
  rL LLVM

https://reviews.llvm.org/D29214