Precisely track blend selects to see if they have been simplified, add test case

Author: lukel97

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -7020,6 +7020,7 @@ static bool planContainsAdditionalSimplifications(VPlan &Plan,
   };
 
   DenseSet<Instruction *> SeenInstrs;
+  SmallDenseMap<PHINode *, unsigned> BlendPhis;
   auto Iter = vp_depth_first_deep(Plan.getVectorLoopRegion()->getEntry());
   for (VPBasicBlock *VPBB : VPBlockUtils::blocksOnly<VPBasicBlock>(Iter)) {
     for (VPRecipeBase &R : *VPBB) {
@@ -7048,11 +7049,13 @@ static bool planContainsAdditionalSimplifications(VPlan &Plan,
         return true;
 
       // VPBlendRecipes are converted to selects and may have been simplified.
+      // Keep track of how many selects each phi has been converted to.
       using namespace VPlanPatternMatch;
       if (match(&R, m_VPInstruction<Instruction::Select>(
-                        m_VPValue(), m_VPValue(), m_VPValue())) &&
-          isa_and_nonnull<PHINode>(R.getVPSingleValue()->getUnderlyingValue()))
-        return true;
+                        m_VPValue(), m_VPValue(), m_VPValue())))
+        if (auto *Phi = dyn_cast_if_present<PHINode>(
+                R.getVPSingleValue()->getUnderlyingValue()))
+          BlendPhis[Phi]++;
 
       /// If a VPlan transform folded a recipe to one producing a single-scalar,
       /// but the original instruction wasn't uniform-after-vectorization in the
@@ -7077,6 +7080,12 @@ static bool planContainsAdditionalSimplifications(VPlan &Plan,
     }
   }
 
+  // If a phi has been simplified then it will have less selects than the number
+  // of incoming values.
+  for (auto [Phi, NumSelects] : BlendPhis)
+    if (NumSelects != Phi->getNumIncomingValues() - 1)
+      return true;
+
   // Return true if the loop contains any instructions that are not also part of
   // the VPlan or are skipped for VPlan-based cost computations. This indicates
   // that the VPlan contains extra simplifications.

llvm/test/Transforms/LoopVectorize/RISCV/blend-simplified.ll

@@ -0,0 +1,81 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --version 5
+; RUN: opt < %s -S -p loop-vectorize | FileCheck %s
+
+; VPlanTransforms::simplifyRecipes will simplify some of selects stemming from
+; the blend recipe, which will cause a difference between the legacy and VPlan
+; based cost models. Make sure we account for this simplifcation. This is
+; extracted from sqlite3 in llvm-test-suite.
+
+target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128"
+target triple = "riscv64-unknown-linux-gnu"
+
+define i64 @html_encode() #0 {
+; CHECK-LABEL: define i64 @html_encode(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = load i8, ptr null, align 1
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i8> poison, i8 [[TMP0]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i8> [[BROADCAST_SPLATINSERT]], <2 x i8> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq <2 x i8> [[BROADCAST_SPLAT]], splat (i8 38)
+; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP1]], <2 x i64> zeroinitializer, <2 x i64> splat (i64 1)
+; CHECK-NEXT:    [[TMP2:%.*]] = or <2 x i64> zeroinitializer, [[PREDPHI]]
+; CHECK-NEXT:    br label %[[MIDDLE_BLOCK:.*]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i64> [[TMP2]], i32 1
+; CHECK-NEXT:    br i1 true, label %[[FOR_END_LOOPEXIT:.*]], label %[[SCALAR_PH]]
+; CHECK:       [[SCALAR_PH]]:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 2, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT:    br label %[[FOR_BODY:.*]]
+; CHECK:       [[SW_BB6:.*]]:
+; CHECK-NEXT:    br label %[[FOR_INC:.*]]
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[J_031:%.*]] = phi i64 [ [[INC12:%.*]], %[[FOR_INC]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
+; CHECK-NEXT:    [[TMP4:%.*]] = load i8, ptr null, align 1
+; CHECK-NEXT:    switch i8 [[TMP4]], label %[[FOR_INC]] [
+; CHECK-NEXT:      i8 62, label %[[FOR_INC]]
+; CHECK-NEXT:      i8 1, label %[[FOR_INC]]
+; CHECK-NEXT:      i8 38, label %[[SW_BB6]]
+; CHECK-NEXT:      i8 0, label %[[FOR_INC]]
+; CHECK-NEXT:    ]
+; CHECK:       [[FOR_INC]]:
+; CHECK-NEXT:    [[DOTSINK:%.*]] = phi i64 [ 0, %[[SW_BB6]] ], [ 1, %[[FOR_BODY]] ], [ 1, %[[FOR_BODY]] ], [ 1, %[[FOR_BODY]] ], [ 1, %[[FOR_BODY]] ]
+; CHECK-NEXT:    [[INC:%.*]] = or i64 0, [[DOTSINK]]
+; CHECK-NEXT:    [[INC12]] = add i64 [[J_031]], 1
+; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[J_031]], 1
+; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label %[[FOR_END_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       [[FOR_END_LOOPEXIT]]:
+; CHECK-NEXT:    [[INC_LCSSA:%.*]] = phi i64 [ [[INC]], %[[FOR_INC]] ], [ [[TMP3]], %[[MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    ret i64 [[INC_LCSSA]]
+;
+entry:
+  br label %for.body
+
+sw.bb6:                                           ; preds = %for.body
+  br label %for.inc
+
+for.body:                                         ; preds = %for.inc, %entry
+  %j.031 = phi i64 [ %inc12, %for.inc ], [ 0, %entry ]
+  %0 = load i8, ptr null, align 1
+  switch i8 %0, label %for.inc [
+  i8 62, label %for.inc
+  i8 1, label %for.inc
+  i8 38, label %sw.bb6
+  i8 0, label %for.inc
+  ]
+
+for.inc:                                          ; preds = %for.body, %for.body, %for.body, %for.body, %sw.bb6
+  %.sink = phi i64 [ 0, %sw.bb6 ], [ 1, %for.body ], [ 1, %for.body ], [ 1, %for.body ], [ 1, %for.body ]
+  %inc = or i64 0, %.sink
+  %inc12 = add i64 %j.031, 1
+  %exitcond.not = icmp eq i64 %j.031, 1
+  br i1 %exitcond.not, label %for.end.loopexit, label %for.body
+
+for.end.loopexit:                                 ; preds = %for.inc
+  ret i64 %inc
+}
+
+attributes #0 = { "target-features"="+v" }