[Clang][OpenMP] Enable device-mapped constexpr class members to not be optimized out

[doru1004]

Summary

This patch fixes an issue whereby a constexpr class member which is mapped to the device is being optimized out thus leading to a runtime error:

Libomptarget error: Unable to generate entries table for device id 0.
Libomptarget error: Failed to init globals on device 0

This is due to the optimized-out variable not being present when host entry table values are matched with their device counterparts.

A currently failing example is included in the runtime test included in this patch.

[jhuber6]

We should have a clang test as well

[doru1004]

In D146552#4210757, @jhuber6 wrote:

We should have a clang test as well

Agreed, working on one currently.

[jdoerfert]

clang/lib/CodeGen/CGOpenMPRuntime.cpp

10394

I really think we need a different solution to this than the "ref" variables... alas, not with this patch

[jdoerfert]

This is as good as it gets until we rip out $ref.

[doru1004]

Updated lit test to show variable added to compiler used vars.

[jhuber6]

LG, thanks.

If you include Fixes: <bug url> in the commit message it'll automatically close it by the way.

[doru1004]

Commit: 0eabf59528f3c3f64923900cae740d9f26c45ae8

clang/lib/CodeGen/CGOpenMPRuntime.cpp

if (VD->hasDefinition(CGM.getContext()) != VarDecl::DeclarationOnly) {
       VarSize =
           CGM.getContext().getTypeSizeInChars(VD->getType()).getQuantity();
       assert(VarSize != 0 && "Expected non-zero size of the variable");
     } else {
       VarSize = 0;
     }
     Linkage = CGM.getLLVMLinkageVarDefinition(VD, /*IsConstant=*/false);
     // Temp solution to prevent optimizations of the internal variables.
-    if (CGM.getLangOpts().OpenMPIsDevice && !VD->isExternallyVisible()) {
+    if (CGM.getLangOpts().OpenMPIsDevice &&
+        (!VD->isExternallyVisible() ||
+         Linkage == llvm::GlobalValue::LinkOnceODRLinkage)) {
       // Do not create a "ref-variable" if the original is not also available
       // on the host.
       if (!OffloadEntriesInfoManager.hasDeviceGlobalVarEntryInfo(VarName))
         return;
       std::string RefName = getName({VarName, "ref"});
       if (!CGM.GetGlobalValue(RefName)) {
         llvm::Constant *AddrRef =
             OMPBuilder.getOrCreateInternalVariable(Addr->getType(), RefName);
         auto *GVAddrRef = cast<llvm::GlobalVariable>(AddrRef);
         GVAddrRef->setConstant(/*Val=*/true);

clang/test/OpenMP/declare_target_constexpr_codegen.cpp

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --check-globals --prefix-filecheck-ir-name _ --global-value-regex "llvm.compiler.used" "_[0-9a-zA-Z]+A[0-9a-zA-Z]+pi[0-9a-zA-Z]+" "_[0-9a-zA-Z]+anotherPi" --version 2
+// REQUIRES: amdgpu-registered-target
+
+
+// Test target codegen - host bc file has to be created first.
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple nvptx64-unknown-unknown -fopenmp-targets=nvptx64-nvidia-cuda -emit-llvm %s -fopenmp-target-debug -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=CHECK
+
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+#pragma omp declare target
+class A {
+public:
+  static constexpr double pi = 3.141592653589793116;
+//.
+// CHECK: @_ZN1A2piE = linkonce_odr constant double 0x400921FB54442D18, comdat, align 8
+// CHECK: @_ZL9anotherPi = internal constant double 3.140000e+00, align 8
+// CHECK: @llvm.compiler.used = appending global [2 x ptr] [ptr @"__ZN1A2piE$ref", ptr @"__ZL9anotherPi$ref"], section "llvm.metadata"
+//.
+  A() { ; }
+  ~A() { ; }
+};
+#pragma omp end declare target
+
+void F(const double &);
+void Test() { F(A::pi); }
+
+#pragma omp declare target
+constexpr static double anotherPi = 3.14;
+#pragma omp end declare target
+
+#endif
+
+
+//
+//// NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+// CHECK: {{.*}}

openmp/libomptarget/test/offloading/target_constexpr_mapping.cpp

+// RUN: %libomptarget-compileoptxx-run-and-check-generic
+
+#include <omp.h>
+#include <stdio.h>
+
+#pragma omp declare target
+class A {
+public:
+  constexpr static double pi = 3.141592653589793116;
+  A() { ; }
+  ~A() { ; }
+};
+#pragma omp end declare target
+
+#pragma omp declare target
+constexpr static double anotherPi = 3.14;
+#pragma omp end declare target
+
+int main() {
+  double a[2];
+#pragma omp target map(tofrom : a[:2])
+  {
+    a[0] = A::pi;
+    a[1] = anotherPi;
+  }
+
+  // CHECK: pi = 3.141592653589793116
+  printf("pi = %.18f\n", a[0]);
+
+  // CHECK: anotherPi = 3.14
+  printf("anotherPi = %.2f\n", a[1]);
+
+  return 0;
+}