[PATCH] D124718: [VPlan] Do not create VPWidenCall recipes for scalar vector factors.

[Florian Hahn via Phabricator via llvm-commits]

fhahn created this revision.
fhahn added reviewers: Ayal, gilr, rengolin.
Herald added subscribers: tschuett, psnobl, rogfer01, bollu, hiraditya.
Herald added a project: All.
fhahn requested review of this revision.
Herald added a subscriber: vkmr.
Herald added a project: LLVM.

'Widen' recipe are only used when actual vector values are generated.
Fix tryToWidenCall to do not create VPWidenCallRecipes for scalar vector
factors.

This was exposed by D123720 <https://reviews.llvm.org/D123720>, because the widened recipes are considered
vector users.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D124718

Files:
  llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
  llvm/test/Transforms/LoopVectorize/AArch64/strict-fadd.ll
  llvm/test/Transforms/LoopVectorize/interleave-with-call.ll


Index: llvm/test/Transforms/LoopVectorize/interleave-with-call.ll
===================================================================
--- llvm/test/Transforms/LoopVectorize/interleave-with-call.ll
+++ llvm/test/Transforms/LoopVectorize/interleave-with-call.ll
@@ -12,7 +12,7 @@
 ; CHECK-NEXT:   vector.body:
 ; CHECK-NEXT:     EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
 ; CHECK-NEXT:     vp<[[IV_STEPS:%.]]>    = SCALAR-STEPS vp<[[CAN_IV]]>, ir<%start>, ir<1>
-; CHECK-NEXT:     WIDEN-CALL ir<%min> = call @llvm.smin.i32(vp<[[IV_STEPS]]>, ir<65535>)
+; CHECK-NEXT:     CLONE ir<%min> = call vp<[[IV_STEPS]]>, ir<65535>, ir<@llvm.smin.i32>
 ; CHECK-NEXT:     CLONE ir<%arrayidx> = getelementptr ir<%dst>, vp<[[IV_STEPS]]>
 ; CHECK-NEXT:     CLONE store ir<%min>, ir<%arrayidx>
 ; CHECK-NEXT:     EMIT vp<[[INC:%.+]]> = VF * UF +(nuw)  vp<[[CAN_IV]]>
@@ -27,8 +27,8 @@
 ; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i32 %start, [[INDEX]]
 ; CHECK-NEXT:    [[INDUCTION:%.*]] = add i32 [[OFFSET_IDX]], 0
 ; CHECK-NEXT:    [[INDUCTION1:%.*]] = add i32 [[OFFSET_IDX]], 1
-; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @llvm.smin.i32(i32 [[INDUCTION]], i32 65535)
-; CHECK-NEXT:    [[TMP2:%.*]] = call i32 @llvm.smin.i32(i32 [[INDUCTION1]], i32 65535)
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call i32 @llvm.smin.i32(i32 [[INDUCTION]], i32 65535)
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call i32 @llvm.smin.i32(i32 [[INDUCTION1]], i32 65535)
 ; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[DST:%.*]], i32 [[INDUCTION]]
 ; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[INDUCTION1]]
 ; CHECK-NEXT:    store i32 [[TMP1]], ptr [[TMP3]], align 8
Index: llvm/test/Transforms/LoopVectorize/AArch64/strict-fadd.ll
===================================================================
--- llvm/test/Transforms/LoopVectorize/AArch64/strict-fadd.ll
+++ llvm/test/Transforms/LoopVectorize/AArch64/strict-fadd.ll
@@ -1139,10 +1139,10 @@
 ; CHECK-UNORDERED: [[LOAD5:%.*]] = load float, float*
 ; CHECK-UNORDERED: [[LOAD6:%.*]] = load float, float*
 ; CHECK-UNORDERED: [[LOAD7:%.*]] = load float, float*
-; CHECK-UNORDERED: [[FMULADD]] = call float @llvm.fmuladd.f32(float [[LOAD]], float [[LOAD4]], float [[VEC_PHI]])
-; CHECK-UNORDERED: [[FMULADD1]] = call float @llvm.fmuladd.f32(float [[LOAD1]], float [[LOAD5]], float [[VEC_PHI1]])
-; CHECK-UNORDERED: [[FMULADD2]] = call float @llvm.fmuladd.f32(float [[LOAD2]], float [[LOAD6]], float [[VEC_PHI2]])
-; CHECK-UNORDERED: [[FMULADD3]] = call float @llvm.fmuladd.f32(float [[LOAD3]], float [[LOAD7]], float [[VEC_PHI3]])
+; CHECK-UNORDERED: [[FMULADD]] = tail call float @llvm.fmuladd.f32(float [[LOAD]], float [[LOAD4]], float [[VEC_PHI]])
+; CHECK-UNORDERED: [[FMULADD1]] = tail call float @llvm.fmuladd.f32(float [[LOAD1]], float [[LOAD5]], float [[VEC_PHI1]])
+; CHECK-UNORDERED: [[FMULADD2]] = tail call float @llvm.fmuladd.f32(float [[LOAD2]], float [[LOAD6]], float [[VEC_PHI2]])
+; CHECK-UNORDERED: [[FMULADD3]] = tail call float @llvm.fmuladd.f32(float [[LOAD3]], float [[LOAD7]], float [[VEC_PHI3]])
 ; CHECK-UNORDERED-NOT: llvm.vector.reduce.fadd
 ; CHECK-UNORDERED: middle.block:
 ; CHECK-UNORDERED: [[BIN_RDX:%.*]] = fadd float [[FMULADD1]], [[FMULADD]]
Index: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -8338,7 +8338,7 @@
     InstructionCost CallCost = CM.getVectorCallCost(CI, VF, NeedToScalarize);
     InstructionCost IntrinsicCost = ID ? CM.getVectorIntrinsicCost(CI, VF) : 0;
     bool UseVectorIntrinsic = ID && IntrinsicCost <= CallCost;
-    return UseVectorIntrinsic || !NeedToScalarize;
+    return VF.isVector() && (UseVectorIntrinsic || !NeedToScalarize);
   };
 
   if (!LoopVectorizationPlanner::getDecisionAndClampRange(willWiden, Range))


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D124718.426245.patch
Type: text/x-patch
Size: 3917 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220430/5f226d29/attachment.bin>