[OPENMP]Fix PR45911: Data sharing and lambda capture.

[ABataev]

Summary

No need to generate inlined OpenMP region for variables captured in
lambdas or block decls, only for implicitly captured variables in the
OpenMP region.

[ABataev]

Rebase

[jdoerfert]

LGTM

clang/lib/CodeGen/CGOpenMPRuntime.cpp

if (Init && (!ForDup || (isa<CXXConstructExpr>(Init) &&
               CGF.MakeAddrLValue(CGF.GetAddrOfLocalVar(OriginalVD), Type);
         } else if (ForDup) {
           SharedRefLValue = CGF.EmitLValueForField(SrcBase, SharedField);
           SharedRefLValue = CGF.MakeAddrLValue(
               Address(SharedRefLValue.getPointer(CGF),
                       C.getDeclAlign(OriginalVD)),
               SharedRefLValue.getType(), LValueBaseInfo(AlignmentSource::Decl),
               SharedRefLValue.getTBAAInfo());
+        } else if (CGF.LambdaCaptureFields.count(
+                       Pair.second.Original->getCanonicalDecl()) > 0 ||
+                   dyn_cast_or_null<BlockDecl>(CGF.CurCodeDecl)) {
+          SharedRefLValue = CGF.EmitLValue(Pair.second.OriginalRef);
         } else {
+          // Processing for implicitly captured variables.
           InlinedOpenMPRegionRAII Region(
               CGF, [](CodeGenFunction &, PrePostActionTy &) {}, OMPD_unknown,
               /*HasCancel=*/false);
-          SharedRefLValue =  CGF.EmitLValue(Pair.second.OriginalRef);
+          SharedRefLValue = CGF.EmitLValue(Pair.second.OriginalRef);
         }
         if (Type->isArrayType()) {
           // Initialize firstprivate array.
           if (!isa<CXXConstructExpr>(Init) || CGF.isTrivialInitializer(Init)) {
             // Perform simple memcpy.
             CGF.EmitAggregateAssign(PrivateLValue, SharedRefLValue, Type);
           } else {
             // Initialize firstprivate array using element-by-element

clang/test/OpenMP/task_firstprivate_codegen.cpp

#pragma omp task firstprivate(t_var, vec, s_arr, s_arr, var, var)
     vec[0] = t_var;
     s_arr[0] = var;
   }
   return T();
 }
 
 int main() {
   static int sivar;
+  float local = 0;
 #ifdef LAMBDA
   // LAMBDA: [[G:@.+]] = global double
   // LAMBDA: [[SIVAR:@.+]] = internal global i{{[0-9]+}} 0,
   // LAMBDA-LABEL: @main
   // LAMBDA: call{{( x86_thiscallcc)?}} void [[OUTER_LAMBDA:@.+]](
   [&]() {
   // LAMBDA: define{{.*}} internal{{.*}} void [[OUTER_LAMBDA]](
   // LAMBDA: [[RES:%.+]] = call i8* @__kmpc_omp_task_alloc(%{{[^ ]+}} @{{[^,]+}}, i32 %{{[^,]+}}, i32 1, i64 56, i64 1, i32 (i32, i8*)* bitcast (i32 (i32, %{{[^*]+}}*)* [[TASK_ENTRY:@[^ ]+]] to i32 (i32, i8*)*))
 // LAMBDA: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
 // LAMBDA: [[G_VAL:%.+]] = load volatile double, double* [[G]],
 // LAMBDA: store volatile double [[G_VAL]], double* [[G_PRIVATE_ADDR]]
 
 // LAMBDA: [[SIVAR_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
 // LAMBDA: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR]],
 // LAMBDA: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]]
 
+// LAMBDA: [[LOCAL_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
+// LAMBDA: store float %{{.+}}, float* [[LOCAL_PRIVATE_ADDR]]
+
 // LAMBDA: call i32 @__kmpc_omp_task(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
 // LAMBDA: ret
-#pragma omp task firstprivate(g, sivar)
+#pragma omp task firstprivate(g, sivar, local)
   {
     // LAMBDA: define {{.+}} void [[INNER_LAMBDA:@.+]](%{{.+}}* [[ARG_PTR:%.+]])
     // LAMBDA: store %{{.+}}* [[ARG_PTR]], %{{.+}}** [[ARG_PTR_REF:%.+]],
     // LAMBDA: [[ARG_PTR:%.+]] = load %{{.+}}*, %{{.+}}** [[ARG_PTR_REF]]
     // LAMBDA: [[G_PTR_REF:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[ARG_PTR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
     // LAMBDA: [[G_REF:%.+]] = load double*, double** [[G_PTR_REF]]
     // LAMBDA: store volatile double 2.0{{.+}}, double* [[G_REF]]
 
#elif defined(BLOCKS)
   // BLOCKS: [[PRIVATES:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* %{{.+}}, i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[G_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 0
   // BLOCKS: [[G_VAL:%.+]] = load volatile double, double* [[G]],
   // BLOCKS: store volatile double [[G_VAL]], double* [[G_PRIVATE_ADDR]]
 
   // BLOCKS: [[SIVAR_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 1
   // BLOCKS: [[SIVAR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[SIVAR]],
   // BLOCKS: store i{{[0-9]+}} [[SIVAR_VAL]], i{{[0-9]+}}* [[SIVAR_PRIVATE_ADDR]],
+
+  // BLOCKS: [[LOCAL_PRIVATE_ADDR:%.+]] = getelementptr inbounds %{{.+}}, %{{.+}}* [[PRIVATES]], i{{.+}} 0, i{{.+}} 2
+  // BLOCKS: store float %{{.+}}, float* [[LOCAL_PRIVATE_ADDR]]
+
   // BLOCKS: call i32 @__kmpc_omp_task(%{{.+}}* @{{.+}}, i32 %{{.+}}, i8* [[RES]])
   // BLOCKS: ret
-#pragma omp task firstprivate(g, sivar)
+#pragma omp task firstprivate(g, sivar, local)
   {
     // BLOCKS: define {{.+}} void {{@.+}}(i8*
     // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
     // BLOCKS: store volatile double 2.0{{.+}}, double*
     // BLOCKS-NOT: [[G]]{{[[^:word:]]}}
     // BLOCKS-NOT: [[ISVAR]]{{[[^:word:]]}}
     // BLOCKS: store i{{[0-9]+}} 22, i{{[0-9]+}}*
     // BLOCKS-NOT: [[SIVAR]]{{[[^:word:]]}}