[PATCH] D100763: [LoopVectorize] Don't create unnecessary vscale intrinsic calls

[David Sherwood via Phabricator via llvm-commits]

david-arm created this revision.
david-arm added reviewers: sdesmalen, c-rhodes, frasercrmck, kmclaughlin.
Herald added a subscriber: hiraditya.
david-arm requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

In quite a few cases in LoopVectorize.cpp we call createStepForVF
with a step value of 0, which leads to unnecessary generation of
llvm.vscale intrinsic calls. I've optimised IRBuilder::CreateVScale
and createStepForVF to return 0 when attempting to multiply
vscale by 0.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D100763

Files:
  llvm/lib/IR/IRBuilder.cpp
  llvm/lib/Transforms/Vectorize/LoopVectorize.cpp


Index: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1108,10 +1108,12 @@
 /// Return a value for Step multiplied by VF.
 static Value *createStepForVF(IRBuilder<> &B, Constant *Step, ElementCount VF) {
   assert(isa<ConstantInt>(Step) && "Expected an integer step");
-  Constant *StepVal = ConstantInt::get(
-      Step->getType(),
-      cast<ConstantInt>(Step)->getSExtValue() * VF.getKnownMinValue());
-  return VF.isScalable() ? B.CreateVScale(StepVal) : StepVal;
+  uint64_t StepVal = cast<ConstantInt>(Step)->getSExtValue();
+  if (StepVal == 0)
+    return Step;
+  Constant *NewStep =
+      ConstantInt::get(Step->getType(), StepVal * VF.getKnownMinValue());
+  return VF.isScalable() ? B.CreateVScale(NewStep) : NewStep;
 }
 
 namespace llvm {
@@ -4768,10 +4770,9 @@
              "Currently unsupported for scalable vectors");
       unsigned Lanes = IsUniform ? 1 : State.VF.getFixedValue();
 
-      Value *RuntimeVF = getRuntimeVF(Builder, PtrInd->getType(), VF);
       for (unsigned Part = 0; Part < UF; ++Part) {
-        Value *PartStart = Builder.CreateMul(
-            RuntimeVF, ConstantInt::get(PtrInd->getType(), Part));
+        Value *PartStart = createStepForVF(
+            Builder, ConstantInt::get(PtrInd->getType(), Part), VF);
         for (unsigned Lane = 0; Lane < Lanes; ++Lane) {
           Value *Idx = Builder.CreateAdd(
               PartStart, ConstantInt::get(PtrInd->getType(), Lane));
Index: llvm/lib/IR/IRBuilder.cpp
===================================================================
--- llvm/lib/IR/IRBuilder.cpp
+++ llvm/lib/IR/IRBuilder.cpp
@@ -81,8 +81,10 @@
 }
 
 Value *IRBuilderBase::CreateVScale(Constant *Scaling, const Twine &Name) {
-  Module *M = GetInsertBlock()->getParent()->getParent();
   assert(isa<ConstantInt>(Scaling) && "Expected constant integer");
+  if (cast<ConstantInt>(Scaling)->isZero())
+    return Scaling;
+  Module *M = GetInsertBlock()->getParent()->getParent();
   Function *TheFn =
       Intrinsic::getDeclaration(M, Intrinsic::vscale, {Scaling->getType()});
   CallInst *CI = createCallHelper(TheFn, {}, this, Name);


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D100763.338515.patch
Type: text/x-patch
Size: 2280 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210419/2292137a/attachment.bin>