[Openmp-commits] [PATCH] D106438: [OpenMP] Add an option to disable function internalization

[Joseph Huber via Phabricator via Openmp-commits]

jhuber6 created this revision.
jhuber6 added reviewers: jdoerfert, tianshilei1992.
Herald added subscribers: ormris, guansong, hiraditya, yaxunl, mgorny.
jhuber6 requested review of this revision.
Herald added subscribers: llvm-commits, openmp-commits, sstefan1.
Herald added projects: OpenMP, LLVM.

Function internalization can sometimes occur in situations where we want to
keep the call sites intact. This patch adds an option to disable function
internalization and prevents the device runtime from being internalized while
creating the bitcode library.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D106438

Files:
  llvm/lib/Transforms/IPO/OpenMPOpt.cpp
  openmp/libomptarget/deviceRTLs/nvptx/CMakeLists.txt


Index: openmp/libomptarget/deviceRTLs/nvptx/CMakeLists.txt
===================================================================
--- openmp/libomptarget/deviceRTLs/nvptx/CMakeLists.txt
+++ openmp/libomptarget/deviceRTLs/nvptx/CMakeLists.txt
@@ -153,6 +153,7 @@
 
 # Set flags for LLVM Bitcode compilation.
 set(bc_flags -S -x c++ -O1 -std=c++14
+             -mllvm -openmp-opt-disable-internalization
              -target nvptx64
              -Xclang -emit-llvm-bc
              -Xclang -aux-triple -Xclang ${aux_triple}
Index: llvm/lib/Transforms/IPO/OpenMPOpt.cpp
===================================================================
--- llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -56,6 +56,11 @@
     cl::desc("Enable the OpenMP region merging optimization."), cl::Hidden,
     cl::init(false));
 
+static cl::opt<bool>
+    DisableInternalization("openmp-opt-disable-internalization", cl::ZeroOrMore,
+                           cl::desc("Disable function internalization."),
+                           cl::Hidden, cl::init(false));
+
 static cl::opt<bool> PrintICVValues("openmp-print-icv-values", cl::init(false),
                                     cl::Hidden);
 static cl::opt<bool> PrintOpenMPKernels("openmp-print-gpu-kernels",
@@ -3771,7 +3776,8 @@
   DenseSet<const Function *> InternalizedFuncs;
   if (isOpenMPDevice(M))
     for (Function &F : M)
-      if (!F.isDeclaration() && !Kernels.contains(&F) && IsCalled(F)) {
+      if (!F.isDeclaration() && !Kernels.contains(&F) && IsCalled(F) &&
+          !DisableInternalization) {
         if (Attributor::internalizeFunction(F, /* Force */ true)) {
           InternalizedFuncs.insert(&F);
         } else if (!F.hasLocalLinkage() && !F.hasFnAttribute(Attribute::Cold)) {


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D106438.360433.patch
Type: text/x-patch
Size: 1779 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/openmp-commits/attachments/20210721/f2c27f72/attachment-0001.bin>