LowerTypeTests: Generate simpler IR for br(llvm.type.test, then, else).

[pcc]

Summary

This makes it so that the code quality for CFI checks when compiling
with -O2 and linking with --lto-O0 is similar to that of the rest of
the code.

Reduces the size of a chrome binary built with -O2/--lto-O0 by
about 750KB.

llvm/lib/Transforms/IPO/LowerTypeTests.cpp

Value *LowerTypeTestsModule::lowerTypeTestCall(Metadata *TypeId, CallInst *CI,
   Value *BitOffset = B.CreateOr(OffsetSHR, OffsetSHL);
 
   Value *OffsetInRange = B.CreateICmpULE(BitOffset, TIL.SizeM1);
 
   // If the bit set is all ones, testing against it is unnecessary.
   if (TIL.TheKind == TypeTestResolution::AllOnes)
     return OffsetInRange;
 
-  TerminatorInst *Term = SplitBlockAndInsertIfThen(OffsetInRange, CI, false);
-  IRBuilder<> ThenB(Term);
+  // See if the intrinsic is used in the following common pattern:
+  //   br(llvm.type.test(...), thenbb, elsebb)
+  // where nothing happens between the type test and the br.
+  // If so, create slightly simpler IR.
+  if (CI->hasOneUse())
+    if (auto *Br = dyn_cast<BranchInst>(*CI->user_begin()))
+      if (CI->getNextNode() == Br) {
+        BasicBlock *Then = InitialBB->splitBasicBlock(CI->getIterator());
+        BasicBlock *Else = Br->getSuccessor(1);
+        BranchInst *NewBr = BranchInst::Create(Then, Else, OffsetInRange);
+        NewBr->setMetadata(LLVMContext::MD_prof,
+                           Br->getMetadata(LLVMContext::MD_prof));
+        ReplaceInstWithInst(InitialBB->getTerminator(), NewBr);
+
+        IRBuilder<> ThenB(CI);
+        return createBitSetTest(ThenB, TIL, BitOffset);
+      }
+
+  IRBuilder<> ThenB(SplitBlockAndInsertIfThen(OffsetInRange, CI, false));
 
   // Now that we know that the offset is in range and aligned, load the
   // appropriate bit from the bitset.
   Value *Bit = createBitSetTest(ThenB, TIL, BitOffset);
 
   // The value we want is 0 if we came directly from the initial block
   // (having failed the range or alignment checks), or the loaded bit if
   // we came from the block in which we loaded it.

llvm/test/Transforms/LowerTypeTests/simplify.ll

+; RUN: opt -S -lowertypetests -lowertypetests-summary-action=import -lowertypetests-read-summary=%S/Inputs/import.yaml < %s | FileCheck %s
+
+target datalayout = "e-p:64:64"
+
+declare i1 @llvm.type.test(i8* %ptr, metadata %bitset) nounwind readnone
+
+; CHECK: define i1 @bytearray7(i8* [[p:%.*]])
+define i1 @bytearray7(i8* %p) {
+  ; CHECK-NEXT: [[pi:%.*]] = ptrtoint i8* [[p]] to i64
+  ; CHECK-NEXT: [[sub:%.*]] = sub i64 [[pi]], ptrtoint (i8* @__typeid_bytearray7_global_addr to i64)
+  ; CHECK-NEXT: [[lshr:%.*]] = lshr i64 [[sub]], zext (i8 ptrtoint (i8* @__typeid_bytearray7_align to i8) to i64)
+  ; CHECK-NEXT: [[shl:%.*]] = shl i64 [[sub]], zext (i8 sub (i8 64, i8 ptrtoint (i8* @__typeid_bytearray7_align to i8)) to i64)
+  ; CHECK-NEXT: [[or:%.*]] = or i64 [[lshr]], [[shl]]
+  ; CHECK-NEXT: [[ule:%.*]] = icmp ule i64 [[or]], ptrtoint (i8* @__typeid_bytearray7_size_m1 to i64)
+  ; CHECK-NEXT: br i1 [[ule]], label %[[t1:.*]], label %[[f:.*]]
+
+  ; CHECK: [[t1]]:
+  ; CHECK-NEXT: [[gep:%.*]] = getelementptr i8, i8* @__typeid_bytearray7_byte_array, i64 [[or]]
+  ; CHECK-NEXT: [[load:%.*]] = load i8, i8* [[gep]]
+  ; CHECK-NEXT: [[and:%.*]] = and i8 [[load]], ptrtoint (i8* @__typeid_bytearray7_bit_mask to i8)
+  ; CHECK-NEXT: [[ne:%.*]] = icmp ne i8 [[and]], 0
+  ; CHECK-NEXT: br i1 [[ne]], label %[[t:.*]], label %[[f:.*]]
+
+  ; CHECK: [[t]]:
+  ; CHECK-NEXT: ret i1 true
+
+  ; CHECK: [[f]]:
+  ; CHECK-NEXT: ret i1 false
+  %x = call i1 @llvm.type.test(i8* %p, metadata !"bytearray7")
+  br i1 %x, label %t, label %f
+
+t:
+  ret i1 true
+
+f:
+  ret i1 false
+}