[PreISelIntrinsicLowering] Use TLI for correct function

[nikic]

Summary

We should query the subtarget of the calling function, not of the intrinsic.

This probably makes no functional difference (as libcalls are unlikely to vary across subtargets), but fixes minor compile-time regressions from unnecessary subtarget instantiations.

Followup to D157567.

[arsenm]

I would assume looking up the subtarget per-use is worse for compile time?

[arsenm]

RuntimeLibcalls probably shouldn't belong to the subtarget in the first place

[nikic]

In D157848#4585069, @arsenm wrote:

I would assume looking up the subtarget per-use is worse for compile time?

Looking up the subtarget is cheap if it's already instantiated. Usually all functions have the same features, so there's just the single subtarget. But of course the intrinsics will have different (no) features and will instantiate a separate subtarget. And apparently subtarget instantiation is super expensive.

Doesn't really matter unless you're compiling tiny files, it's just how I happened to notice this.

llvm/lib/CodeGen/PreISelIntrinsicLowering.cpp

uint64_t Threshold = MemIntrinsicExpandSizeThresholdOpt.getNumOccurrences()
                            : TTI.getMaxMemIntrinsicInlineSizeThreshold();
   uint64_t SizeVal = CI->getZExtValue();
 
   // Treat a threshold of 0 as a special case to force expansion of all
   // intrinsics, including size 0.
   return SizeVal > Threshold || Threshold == 0;
 }
 
-static bool canEmitLibcall(const TargetLowering &TLI, RTLIB::Libcall LC) {
+static bool canEmitLibcall(const TargetMachine &TM, Function *F,
+                           RTLIB::Libcall LC) {
   // TODO: Should this consider the address space of the memcpy?
-  return TLI.getLibcallName(LC) != nullptr;
+  const TargetLowering *TLI = TM.getSubtargetImpl(*F)->getTargetLowering();
+  return TLI->getLibcallName(LC) != nullptr;
 }
 
 // TODO: Handle atomic memcpy and memcpy.inline
 // TODO: Pass ScalarEvolution
 bool PreISelIntrinsicLowering::expandMemIntrinsicUses(Function &F) const {
-  const TargetLowering *TLI = TM.getSubtargetImpl(F)->getTargetLowering();
   Intrinsic::ID ID = F.getIntrinsicID();
   bool Changed = false;
 
   for (User *U : llvm::make_early_inc_range(F.users())) {
     Instruction *Inst = cast<Instruction>(U);
 
     switch (ID) {
     case Intrinsic::memcpy: {
       auto *Memcpy = cast<MemCpyInst>(Inst);
       Function *ParentFunc = Memcpy->getFunction();
       const TargetTransformInfo &TTI = LookupTTI(*ParentFunc);
       if (shouldExpandMemIntrinsicWithSize(Memcpy->getLength(), TTI)) {
-        if (UseMemIntrinsicLibFunc && canEmitLibcall(*TLI, RTLIB::MEMCPY))
+        if (UseMemIntrinsicLibFunc &&
+            canEmitLibcall(TM, ParentFunc, RTLIB::MEMCPY))
           break;
 
         // TODO: For optsize, emit the loop into a separate function
         expandMemCpyAsLoop(Memcpy, TTI);
         Changed = true;
         Memcpy->eraseFromParent();
       }
 
       break;
     }
     case Intrinsic::memmove: {
       auto *Memmove = cast<MemMoveInst>(Inst);
       Function *ParentFunc = Memmove->getFunction();
       const TargetTransformInfo &TTI = LookupTTI(*ParentFunc);
       if (shouldExpandMemIntrinsicWithSize(Memmove->getLength(), TTI)) {
-        if (UseMemIntrinsicLibFunc && canEmitLibcall(*TLI, RTLIB::MEMMOVE))
+        if (UseMemIntrinsicLibFunc &&
+            canEmitLibcall(TM, ParentFunc, RTLIB::MEMMOVE))
           break;
 
         if (expandMemMoveAsLoop(Memmove, TTI)) {
           Changed = true;
           Memmove->eraseFromParent();
         }
       }
 
       break;
     }
     case Intrinsic::memset: {
       auto *Memset = cast<MemSetInst>(Inst);
       Function *ParentFunc = Memset->getFunction();
       const TargetTransformInfo &TTI = LookupTTI(*ParentFunc);
       if (shouldExpandMemIntrinsicWithSize(Memset->getLength(), TTI)) {
-        if (UseMemIntrinsicLibFunc && canEmitLibcall(*TLI, RTLIB::MEMSET))
+        if (UseMemIntrinsicLibFunc &&
+            canEmitLibcall(TM, ParentFunc, RTLIB::MEMSET))
           break;
 
         expandMemSetAsLoop(Memset);
         Changed = true;
         Memset->eraseFromParent();
       }
 
       break;