[llvm] [VPlan] Use llvm.masked.{u, s}{div, rem} for predicated division (PR #191377)

[Florian Hahn via llvm-commits]

================
@@ -6700,28 +6705,21 @@ bool VPRecipeBuilder::shouldWiden(Instruction *I, VFRange &Range) const {
                                                              Range);
 }
 
-VPWidenRecipe *VPRecipeBuilder::tryToWiden(VPInstruction *VPI) {
+VPRecipeWithIRFlags *VPRecipeBuilder::tryToWiden(VPInstruction *VPI) {
   auto *I = VPI->getUnderlyingInstr();
   switch (VPI->getOpcode()) {
   default:
     return nullptr;
   case Instruction::SDiv:
   case Instruction::UDiv:
   case Instruction::SRem:
-  case Instruction::URem: {
-    // If not provably safe, use a select to form a safe divisor before widening the
-    // div/rem operation itself.  Otherwise fall through to general handling below.
-    if (CM.isPredicatedInst(I)) {
-      SmallVector<VPValue *> Ops(VPI->operandsWithoutMask());
-      VPValue *Mask = VPI->getMask();
-      VPValue *One = Plan.getConstantInt(I->getType(), 1u);
-      auto *SafeRHS =
-          Builder.createSelect(Mask, Ops[1], One, VPI->getDebugLoc());
-      Ops[1] = SafeRHS;
-      return new VPWidenRecipe(*I, Ops, *VPI, *VPI, VPI->getDebugLoc());
----------------
fhahn wrote:

One thing we lose with the intrinsics is things like the `exact` flag. I don't think there anything we can do about that, but it would probably be good to have a test for that?

https://github.com/llvm/llvm-project/pull/191377