[CodeGen] Do not allow opt-bisect-limit to skip ScalarizeMaskedMemIntrin.

[a.elovikov]

Summary

The pass is supposed to scalarize such intrinsics if the target does not support
them natively, so if the scalarization does not happen instruction selection
crashes due to inability to lower these intrinsics.

[andrew.w.kaylor]

lgtm

lib/CodeGen/ScalarizeMaskedMemIntrin.cpp

for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
     BranchInst::Create(CondBlock, NewIfBlock, Cmp, OldBr);
     OldBr->eraseFromParent();
     IfBlock = NewIfBlock;
   }
   CI->eraseFromParent();
 }
 
 bool ScalarizeMaskedMemIntrin::runOnFunction(Function &F) {
-  if (skipFunction(F))
-    return false;
-
   bool EverMadeChange = false;
 
   TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
 
   bool MadeChange = true;
   while (MadeChange) {
     MadeChange = false;
     for (Function::iterator I = F.begin(); I != F.end();) {

test/CodeGen/X86/replace_unsupported_masked_mem_intrin.ll

 ; RUN: llc -O0 -mtriple=x86_64-unknown-linux-gnu -mattr=+sse,+sse2 < %s -o /dev/null
 ; pr33001 - Check that llc doesn't crash when running with O0 option.
 
+; RUN: llc -O2 -opt-bisect-limit=0 -mtriple=x86_64-unknown-linux-gnu -mattr=+sse,+sse2 < %s -o /dev/null
+; Check that llc doesn't crash due to ScalarizeMaskedMemIntring not being run
+; because of opt-bisect-limit that in turn causes crash in instruction selection
+; for unsupported gather/scatter.
+
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 define <4 x i32> @test_masked_load(<4 x i32>* %base, <4 x i1> %mask) {
   %res = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* %base, i32 4, <4 x i1> %mask, <4 x i32> zeroinitializer)
   ret <4 x i32> %res
 }