TargetLibraryInfo: Stop guessing the wchar_t size

[MatzeB]

Summary

Usually the frontend communicates the size of wchar_t via metadata and
we can optimize wcslen (and possibly other calls in the future). In
cases without the wchar_size metadata we would previously try to guess
the correct size based on the target triple; however this is fragile to
keep up to date and may miss users manually changing the size via flags.
Better be safe and stop guessing and optimizing if the frontend didn't
communicate the size.

[MatzeB]

After thinking about it some more, I decided that we better leave the logic to decide on the wchar_t size to a C frontend.

Related/motivated by the discussion in https://reviews.llvm.org/D37891

[MatzeB]

Add a unittest to ensure we are indeed not guessing the wchar_t size.

[compnerd]

This is pretty awesome. I assume that you will do a corresponding change in clang to remove the single use of that function?

[MatzeB]

Thanks, committed in r314185

include/llvm/Analysis/TargetLibraryInfo.h

public:
   }
 
   /// Set to true iff i32 parameters to library functions should have signext
   /// attribute if they correspond to C-level int or unsigned int.
   void setShouldSignExtI32Param(bool Val) {
     ShouldSignExtI32Param = Val;
   }
 
-  /// Returns the size of the wchar_t type in bytes.
+  /// Returns the size of the wchar_t type in bytes or 0 if the size is unknown.
+  /// This queries the 'wchar_size' metadata.
   unsigned getWCharSize(const Module &M) const;
-
-  /// Returns size of the default wchar_t type on target \p T. This is mostly
-  /// intended to verify that the size in the frontend matches LLVM. All other
-  /// queries should use getWCharSize() instead.
-  static unsigned getTargetWCharSize(const Triple &T);
 };
 
 /// Provides information about what library functions are available for
 /// the current target.
 ///
 /// This both allows optimizations to handle them specially and frontends to
 /// disable such optimizations through -fno-builtin etc.
 class TargetLibraryInfo {

lib/Analysis/TargetLibraryInfo.cpp

TargetLibraryInfoImpl &TargetLibraryAnalysis::lookupInfoImpl(const Triple &T) {
   std::unique_ptr<TargetLibraryInfoImpl> &Impl =
       Impls[T.normalize()];
   if (!Impl)
     Impl.reset(new TargetLibraryInfoImpl(T));
 
   return *Impl;
 }
 
-unsigned TargetLibraryInfoImpl::getTargetWCharSize(const Triple &T) {
-  // See also clang/lib/Basic/Targets.cpp.
-  if (T.isPS4() || T.isOSWindows() || T.isArch16Bit())
-    return 2;
-  if (T.getArch() == Triple::xcore)
-    return 1;
-  return 4;
-}
-
 unsigned TargetLibraryInfoImpl::getWCharSize(const Module &M) const {
   if (auto *ShortWChar = cast_or_null<ConstantAsMetadata>(
       M.getModuleFlag("wchar_size")))
     return cast<ConstantInt>(ShortWChar->getValue())->getZExtValue();
-  return getTargetWCharSize(Triple(M.getTargetTriple()));
+  return 0;
 }
 
 TargetLibraryInfoWrapperPass::TargetLibraryInfoWrapperPass()
     : ImmutablePass(ID), TLIImpl(), TLI(TLIImpl) {
   initializeTargetLibraryInfoWrapperPassPass(*PassRegistry::getPassRegistry());
 }
 
 TargetLibraryInfoWrapperPass::TargetLibraryInfoWrapperPass(const Triple &T)

lib/Transforms/Utils/SimplifyLibCalls.cpp

 
 Value *LibCallSimplifier::optimizeStrLen(CallInst *CI, IRBuilder<> &B) {
   return optimizeStringLength(CI, B, 8);
 }
 
 Value *LibCallSimplifier::optimizeWcslen(CallInst *CI, IRBuilder<> &B) {
   Module &M = *CI->getParent()->getParent()->getParent();
   unsigned WCharSize = TLI->getWCharSize(M) * 8;
+  // We cannot perform this optimization without wchar_size metadata.
+  if (WCharSize == 0)
+    return nullptr;
 
   return optimizeStringLength(CI, B, WCharSize);
 }
 
 Value *LibCallSimplifier::optimizeStrPBrk(CallInst *CI, IRBuilder<> &B) {
   StringRef S1, S2;
   bool HasS1 = getConstantStringInfo(CI->getArgOperand(0), S1);
   bool HasS2 = getConstantStringInfo(CI->getArgOperand(1), S2);

test/Transforms/InstCombine/wcslen-1.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; Test that the wcslen library call simplifier works correctly.
 ;
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
 
 declare i64 @wcslen(i32*)
 
+!0 = !{i32 1, !"wchar_size", i32 4}
+!llvm.module.flags = !{!0}
+
 @hello = constant [6 x i32] [i32 104, i32 101, i32 108, i32 108, i32 111, i32 0]
 @longer = constant [7 x i32] [i32 108, i32 111, i32 110, i32 103, i32 101, i32 114, i32 0]
 @null = constant [1 x i32] zeroinitializer
 @null_hello = constant [7 x i32] [i32 0, i32 104, i32 101, i32 108, i32 108, i32 111, i32 0]
 @nullstring = constant i32 0
 @a = common global [32 x i32] zeroinitializer, align 1
 @null_hello_mid = constant [13 x i32] [i32 104, i32 101, i32 108, i32 108, i32 111, i32 32, i32 119, i32 111, i32 114, i32 0, i32 108, i32 100, i32 0]
 

test/Transforms/InstCombine/wcslen-2.ll

 ; Test that the wcslen library call simplifier works correctly.
 ;
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 
 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
 
+!0 = !{i32 1, !"wchar_size", i32 4}
+!llvm.module.flags = !{!0}
+
 @hello = constant [6 x i32] [i32 104, i32 101, i32 108, i32 108, i32 111, i32 0]
 
 declare i64 @wcslen(i32*, i32)
 
 define i64 @test_no_simplify1() {
 ; CHECK-LABEL: @test_no_simplify1(
   %hello_p = getelementptr [6 x i32], [6 x i32]* @hello, i64 0, i64 0
   %hello_l = call i64 @wcslen(i32* %hello_p, i32 187)
 ; CHECK-NEXT: %hello_l = call i64 @wcslen
   ret i64 %hello_l
 ; CHECK-NEXT: ret i64 %hello_l
 }