[SimplifyLibcalls] Realloc(null, N) -> Malloc(N)

[xbolva00]

Summary

realloc(null, N) -> malloc(N)

[xbolva00]

I would like to hear your feedback on these three changes

1. malloc(N) + realloc(ptr, N + X) -> fold to -> malloc(N + X), remove realloc
2. malloc(N) + realloc(ptr, N - X) -> fold to -> malloc(N), remove realloc
3. realloc(NULL, N) -> malloc(N)

@efriedma @lebedev.ri @spatel and others...

Edit: I think only 3. should stay in this patch..

lib/Transforms/Utils/SimplifyLibCalls.cpp

823

This is somehow problematic, if I remove this check, this code will crash on 32bit.

define i8* @foo(i32 %n) #0 {

%call = call noalias i8* @malloc(i32 %n) #2
%add = add nsw i32 %n, 2
%call1 = call i8* @realloc(i8* %call, i32 %add) #2
ret i8* %call1

}

Any ideas why, @spatel @efriedma ?

[efriedma]

lib/Transforms/Utils/BuildLibCalls.cpp

1059 ↗

(On Diff #142131)

Why are the implementations of emitMalloc and emitCalloc different? (I'm specifically concerned about missing call to inferLibFuncAttributes.)

test/Transforms/InstCombine/realloc.ll

14

This doesn't look like it's what you meant to test.

[xbolva00]

test/Transforms/InstCombine/realloc.ll

14

oh yes, I will rework test soon.

[xbolva00]

test/Transforms/InstCombine/realloc.ll

14

Done.

[efriedma]

lib/Transforms/Utils/BuildLibCalls.cpp

1062 ↗

(On Diff #142247)

This isn't right; should be TLI->has(LibFunc_calloc).

[xbolva00]

lib/Transforms/Utils/BuildLibCalls.cpp

1062 ↗

(On Diff #142247)

Fixed.

Anyway, emitCalloc function I just copied from SimplifyLibCalls since it has no sense to have it there. So some time ago when it was merged and there was without proper review?

[xbolva00]

Any review or merge? @spatel ?

[efriedma]

LGTM

[xbolva00]

In D45413#1068982, @efriedma wrote:

LGTM

I have no write access, so somebody must merge it. @spatel ?

[spatel]

The patch doesn't compile as-is. Please post a corrected patch.

/llvm/lib/Transforms/Utils/BuildLibCalls.cpp:1061:11: error: member reference type 'const llvm::TargetLibraryInfo' is not a pointer; did you mean to use
      '.'?
  if (!TLI->has(LibFunc_calloc))
       ~~~^~
          .
/llvm/lib/Transforms/Utils/BuildLibCalls.cpp:1069:53: error: indirection requires pointer operand ('const llvm::TargetLibraryInfo' invalid)
  inferLibFuncAttributes(*M->getFunction("calloc"), *TLI);

lib/Transforms/Utils/SimplifyLibCalls.cpp

 
 Value *LibCallSimplifier::optimizeMemMove(CallInst *CI, IRBuilder<> &B) {
   // memmove(x, y, n) -> llvm.memmove(align 1 x, align 1 y, n)
   B.CreateMemMove(CI->getArgOperand(0), 1, CI->getArgOperand(1), 1,
                   CI->getArgOperand(2));
   return CI->getArgOperand(0);
 }
 
-// TODO: Does this belong in BuildLibCalls or should all of those similar
-// functions be moved here?
-static Value *emitCalloc(Value *Num, Value *Size, const AttributeList &Attrs,
-                         IRBuilder<> &B, const TargetLibraryInfo &TLI) {
-  LibFunc Func;
-  if (!TLI.getLibFunc("calloc", Func) || !TLI.has(Func))
-    return nullptr;
-
-  Module *M = B.GetInsertBlock()->getModule();
-  const DataLayout &DL = M->getDataLayout();
-  IntegerType *PtrType = DL.getIntPtrType((B.GetInsertBlock()->getContext()));
-  Value *Calloc = M->getOrInsertFunction("calloc", Attrs, B.getInt8PtrTy(),
-                                         PtrType, PtrType);
-  CallInst *CI = B.CreateCall(Calloc, { Num, Size }, "calloc");
-
-  if (const auto *F = dyn_cast<Function>(Calloc->stripPointerCasts()))
-    CI->setCallingConv(F->getCallingConv());
-
-  return CI;
-}
-
 /// Fold memset[_chk](malloc(n), 0, n) --> calloc(1, n).
 static Value *foldMallocMemset(CallInst *Memset, IRBuilder<> &B,
                                const TargetLibraryInfo &TLI) {
   // This has to be a memset of zeros (bzero).
   auto *FillValue = dyn_cast<ConstantInt>(Memset->getArgOperand(1));
   if (!FillValue || FillValue->getZExtValue() != 0)
     return nullptr;
 
   // TODO: We should handle the case where the malloc has more than one use.
   // This is necessary to optimize common patterns such as when the result of
   // the malloc is checked against null or when a memset intrinsic is used in
   // place of a memset library call.
   auto *Malloc = dyn_cast<CallInst>(Memset->getArgOperand(0));
   if (!Malloc || !Malloc->hasOneUse())
     return nullptr;
 
   // Is the inner call really malloc()?
   Function *InnerCallee = Malloc->getCalledFunction();
if (auto *Calloc = foldMallocMemset(CI, B, *TLI))
     return Calloc;
 
   // memset(p, v, n) -> llvm.memset(align 1 p, v, n)
   Value *Val = B.CreateIntCast(CI->getArgOperand(1), B.getInt8Ty(), false);
   B.CreateMemSet(CI->getArgOperand(0), Val, CI->getArgOperand(2), 1);
   return CI->getArgOperand(0);
 }
 
+Value *LibCallSimplifier::optimizeRealloc(CallInst *CI, IRBuilder<> &B) {
+  if (isa<ConstantPointerNull>(CI->getArgOperand(0)))
+    return emitMalloc(CI->getArgOperand(1), B, DL, TLI);
+
+  return nullptr;
+}
+
 //===----------------------------------------------------------------------===//
 // Math Library Optimizations
 //===----------------------------------------------------------------------===//
 
 /// Return a variant of Val with float type.
 /// Currently this works in two cases: If Val is an FPExtension of a float
 /// value to something bigger, simply return the operand.
 /// If Val is a ConstantFP but can be converted to a float ConstantFP without
if (TLI->getLibFunc(*Callee, Func) && TLI->has(Func)) {
     case LibFunc_memcmp:
       return optimizeMemCmp(CI, Builder);
     case LibFunc_memcpy:
       return optimizeMemCpy(CI, Builder);
     case LibFunc_memmove:
       return optimizeMemMove(CI, Builder);
     case LibFunc_memset:
       return optimizeMemSet(CI, Builder);
+    case LibFunc_realloc:
+      return optimizeRealloc(CI, Builder);
     case LibFunc_wcslen:
       return optimizeWcslen(CI, Builder);
     default:
       break;
     }
   }
   return nullptr;
 }

test/Transforms/InstCombine/realloc.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+declare i8* @realloc(i8*, i64) #1
+declare noalias i8* @malloc(i64) #1
+
+
+define i8* @realloc_null_ptr() #0 {
+; CHECK-LABEL: @realloc_null_ptr(
+; CHECK-NEXT:    [[MALLOC:%.*]] = call i8* @malloc(i64 100)
+; CHECK-NEXT:    ret i8* [[MALLOC]]
+;
+  %malloc = call i8* @malloc(i64 100)
+  ret i8* %malloc
+}
+
+define i8* @realloc_unknown_ptr(i8* %ptr) #0 {
+; CHECK-LABEL: @realloc_unknown_ptr(
+; CHECK-NEXT:    [[CALL:%.*]] = call i8* @realloc(i8* [[PTR:%.*]], i64 100)
+; CHECK-NEXT:    ret i8* [[CALL]]
+;
+  %call = call i8* @realloc(i8* %ptr, i64 100) #2
+  ret i8* %call
+}