This patch reverts the fix added in #85218 in favour of a better approach

[TIFitis]

This patch reverts the fix added in #85218 in favour of a better approach.

[llvmbot]

@llvm/pr-subscribers-flang-openmp

Author: Akash Banerjee (TIFitis)

Changes

This patch reverts the fix added in #85218 in favour of a better approach.

---

Full diff: https://github.com/llvm/llvm-project/pull/88118.diff

1 Files Affected:

• (modified) llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp (+1-4)

diff --git a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
index 7fd8474c2ec890..6b5446182a772f 100644
--- a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -4886,11 +4886,8 @@ OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createTargetData(
     return InsertPointTy();
 
   // Disable TargetData CodeGen on Device pass.
-  if (Config.IsTargetDevice.value_or(false)) {
-    if (BodyGenCB)
-      Builder.restoreIP(BodyGenCB(Builder.saveIP(), BodyGenTy::NoPriv));
+  if (Config.IsTargetDevice.value_or(false))
     return Builder.saveIP();
-  }
 
   Builder.restoreIP(CodeGenIP);
   bool IsStandAlone = !BodyGenCB;

[TIFitis]

@jdoerfert @skatrak I believe @jsjodin is working on a patch which is a more general approach for codegen of target constructs nested within host operations. Once that patch is upstream, I'll merge this to get rid of the placeholder fix we have.

Does that sound reasonable?

[skatrak]

Sounds good to me. Did #85239 not completely cover this case?