From: Erich Keane <erich.keane@intel.com>
Date: Fri, 7 Dec 2018 15:31:23 +0000 (+0000)
Subject: Make CPUDispatch resolver emit dependent functions.
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=28ce64f1231252c541527446e77f54b52702dd84;p=clang

Make CPUDispatch resolver emit dependent functions.

Inline cpu_specific versions referenced before the cpu_dispatch function
weren't properly emitted, since they hadn't been referred to.  This
patch ensures that during resolver generation that all appropriate
versions are emitted.

Change-Id: I94c3766aaf9c75ca07a0ad8258efdbb834654ff8

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@348600 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/CodeGen/CodeGenModule.cpp b/lib/CodeGen/CodeGenModule.cpp
index 81304493a3..b80a43b7eb 100644
--- a/lib/CodeGen/CodeGenModule.cpp
+++ b/lib/CodeGen/CodeGenModule.cpp
@@ -2585,11 +2585,22 @@ void CodeGenModule::emitCPUDispatchDefinition(GlobalDecl GD) {
 
     llvm::Constant *Func = GetGlobalValue(MangledName);
 
-    if (!Func)
+    if (!Func) {
+      GlobalDecl ExistingDecl = Manglings.lookup(MangledName);
+      if (ExistingDecl.getDecl() &&
+          ExistingDecl.getDecl()->getAsFunction()->isDefined()) {
+        EmitGlobalFunctionDefinition(ExistingDecl, nullptr);
+        Func = GetGlobalValue(MangledName);
+      } else {
+        if (!ExistingDecl.getDecl())
+          ExistingDecl = GD.getWithMultiVersionIndex(Index);
+
       Func = GetOrCreateLLVMFunction(
-          MangledName, DeclTy, GD.getWithMultiVersionIndex(Index),
+          MangledName, DeclTy, ExistingDecl,
           /*ForVTable=*/false, /*DontDefer=*/true,
           /*IsThunk=*/false, llvm::AttributeList(), ForDefinition);
+      }
+    }
 
     llvm::SmallVector<StringRef, 32> Features;
     Target.getCPUSpecificCPUDispatchFeatures(II->getName(), Features);
diff --git a/test/CodeGen/attr-target-mv.c b/test/CodeGen/attr-target-mv.c
index 372b14053f..363dea6a2f 100644
--- a/test/CodeGen/attr-target-mv.c
+++ b/test/CodeGen/attr-target-mv.c
@@ -223,10 +223,12 @@ void bar5() {
 // WINDOWS: define linkonce_odr dso_local void @foo_decls()
 // WINDOWS: define linkonce_odr dso_local void @foo_decls.sse4.2()
 
+// LINUX: define linkonce void @foo_multi(i32 %{{[^,]+}}, double %{{[^\)]+}})
 // LINUX: define linkonce void @foo_multi.avx_sse4.2(i32 %{{[^,]+}}, double %{{[^\)]+}})
 // LINUX: define linkonce void @foo_multi.fma4_sse4.2(i32 %{{[^,]+}}, double %{{[^\)]+}})
 // LINUX: define linkonce void @foo_multi.arch_ivybridge_fma4_sse4.2(i32 %{{[^,]+}}, double %{{[^\)]+}})
 
+// WINDOWS: define linkonce_odr dso_local void @foo_multi(i32 %{{[^,]+}}, double %{{[^\)]+}})
 // WINDOWS: define linkonce_odr dso_local void @foo_multi.avx_sse4.2(i32 %{{[^,]+}}, double %{{[^\)]+}})
 // WINDOWS: define linkonce_odr dso_local void @foo_multi.fma4_sse4.2(i32 %{{[^,]+}}, double %{{[^\)]+}})
 // WINDOWS: define linkonce_odr dso_local void @foo_multi.arch_ivybridge_fma4_sse4.2(i32 %{{[^,]+}}, double %{{[^\)]+}})
diff --git a/test/CodeGenCXX/attr-cpuspecific.cpp b/test/CodeGenCXX/attr-cpuspecific.cpp
index 57b45e1da2..bfee49ca34 100644
--- a/test/CodeGenCXX/attr-cpuspecific.cpp
+++ b/test/CodeGenCXX/attr-cpuspecific.cpp
@@ -16,8 +16,11 @@ void foo() {
 // LINUX: define void (%struct.S*)* @_ZN1S4FuncEv.resolver
 // LINUX: ret void (%struct.S*)* @_ZN1S4FuncEv.S
 // LINUX: ret void (%struct.S*)* @_ZN1S4FuncEv.O
+// LINUX: declare void @_ZN1S4FuncEv.S
+// LINUX: define linkonce_odr void @_ZN1S4FuncEv.O
 
 // WINDOWS: define dso_local void @"?Func@S@@QEAAXXZ"(%struct.S*)
 // WINDOWS: musttail call void @"?Func@S@@QEAAXXZ.S"(%struct.S* %0)
 // WINDOWS: musttail call void @"?Func@S@@QEAAXXZ.O"(%struct.S* %0)
-
+// WINDOWS: declare dso_local void @"?Func@S@@QEAAXXZ.S"
+// WINDOWS: define linkonce_odr dso_local void @"?Func@S@@QEAAXXZ.O"