[SLP]Do not require external uses for roots and single use for other instructions in computeMinimumValueSizes. (#72679)

After changes, that does not require support from InstCombine, we can
drop some extra requirements for values-to-be-demoted. No need to check
for external uses for roots/other instructions, just check that the
  no non-vectorized insertelement instruction, which may require
  widening.

Review: #72679

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -12017,8 +12017,12 @@ Value *BoUpSLP::vectorizeTree(
               IRBuilder<>::InsertPointGuard Guard(Builder);
               if (auto *IVec = dyn_cast<Instruction>(Vec))
                 Builder.SetInsertPoint(IVec->getNextNonDebugInstruction());
-              Vec = Builder.CreateIntCast(Vec, VU->getType(),
-                                          BWIt->second.second);
+              Vec = Builder.CreateIntCast(
+                  Vec,
+                  FixedVectorType::get(
+                      cast<VectorType>(VU->getType())->getElementType(),
+                      cast<FixedVectorType>(Vec->getType())->getNumElements()),
+                  BWIt->second.second);
               VectorCasts.try_emplace(Scalar, Vec);
             } else {
               Vec = VecIt->second;
@@ -13096,10 +13100,14 @@ bool BoUpSLP::collectValuesToDemote(
   if (isa<Constant>(V))
     return true;
 
-  // If the value is not a vectorized instruction in the expression with only
-  // one use, it cannot be demoted.
+  // If the value is not a vectorized instruction in the expression and not used
+  // by the insertelement instruction and not used in multiple vector nodes, it
+  // cannot be demoted.
   auto *I = dyn_cast<Instruction>(V);
-  if (!I || !I->hasOneUse() || !getTreeEntry(I) || !Visited.insert(I).second)
+  if (!I || !getTreeEntry(I) || MultiNodeScalars.contains(I) ||
+      !Visited.insert(I).second || all_of(I->users(), [&](User *U) {
+        return isa<InsertElementInst>(U) && !getTreeEntry(U);
+      }))
     return false;
 
   unsigned Start = 0;
@@ -13170,11 +13178,6 @@ bool BoUpSLP::collectValuesToDemote(
 }
 
 void BoUpSLP::computeMinimumValueSizes() {
-  // If there are no external uses, the expression tree must be rooted by a
-  // store. We can't demote in-memory values, so there is nothing to do here.
-  if (ExternalUses.empty())
-    return;
-
   // We only attempt to truncate integer expressions.
   auto &TreeRoot = VectorizableTree[0]->Scalars;
   auto *TreeRootIT = dyn_cast<IntegerType>(TreeRoot[0]->getType());

llvm/test/Transforms/SLPVectorizer/X86/int-bitcast-minbitwidth.ll

@@ -5,19 +5,14 @@ define void @t(i64 %v) {
 ; CHECK-LABEL: define void @t(
 ; CHECK-SAME: i64 [[V:%.*]]) {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CONV12_1_I:%.*]] = trunc i64 [[V]] to i32
-; CHECK-NEXT:    [[MUL_I_1_I:%.*]] = mul i32 [[CONV12_1_I]], 2
-; CHECK-NEXT:    [[CONV12_I:%.*]] = trunc i64 [[V]] to i32
-; CHECK-NEXT:    [[MUL_I_I:%.*]] = mul i32 [[CONV12_I]], 3
-; CHECK-NEXT:    [[CONV14104_I:%.*]] = or i32 [[MUL_I_1_I]], [[MUL_I_I]]
-; CHECK-NEXT:    [[CONV12_1_I_1:%.*]] = trunc i64 [[V]] to i32
-; CHECK-NEXT:    [[MUL_I_1_I_1:%.*]] = mul i32 [[CONV12_1_I_1]], 6
-; CHECK-NEXT:    [[CONV12_I_1:%.*]] = trunc i64 [[V]] to i32
-; CHECK-NEXT:    [[MUL_I_I_1:%.*]] = mul i32 [[CONV12_I_1]], 5
-; CHECK-NEXT:    [[CONV14104_I_1:%.*]] = or i32 [[MUL_I_1_I_1]], [[MUL_I_I_1]]
-; CHECK-NEXT:    [[TMP0:%.*]] = or i32 [[CONV14104_I]], [[CONV14104_I_1]]
-; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[TMP0]], 65535
-; CHECK-NEXT:    store i32 [[TMP1]], ptr null, align 4
+; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <4 x i64> poison, i64 [[V]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i64> [[TMP0]], <4 x i64> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc <4 x i64> [[TMP1]] to <4 x i16>
+; CHECK-NEXT:    [[TMP3:%.*]] = mul <4 x i16> [[TMP2]], <i16 5, i16 6, i16 3, i16 2>
+; CHECK-NEXT:    [[TMP4:%.*]] = call i16 @llvm.vector.reduce.or.v4i16(<4 x i16> [[TMP3]])
+; CHECK-NEXT:    [[TMP5:%.*]] = sext i16 [[TMP4]] to i32
+; CHECK-NEXT:    [[TMP6:%.*]] = and i32 [[TMP5]], 65535
+; CHECK-NEXT:    store i32 [[TMP6]], ptr null, align 4
 ; CHECK-NEXT:    ret void
 ;
 entry:

llvm/test/Transforms/SLPVectorizer/X86/minbitwidth-multiuse-with-insertelement.ll

@@ -6,15 +6,18 @@ define void @test(i8 %0) {
 ; CHECK-SAME: i8 [[TMP0:%.*]]) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i8> <i8 0, i8 poison>, i8 [[TMP0]], i32 1
-; CHECK-NEXT:    [[TMP2:%.*]] = sext <2 x i8> [[TMP1]] to <2 x i32>
-; CHECK-NEXT:    [[TMP3:%.*]] = mul <2 x i32> [[TMP2]], zeroinitializer
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP3]], i32 0
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1
-; CHECK-NEXT:    [[ADD:%.*]] = or i32 [[TMP4]], [[TMP5]]
+; CHECK-NEXT:    [[TMP2:%.*]] = sext <2 x i8> [[TMP1]] to <2 x i16>
+; CHECK-NEXT:    [[TMP3:%.*]] = sext <2 x i16> [[TMP2]] to <2 x i32>
+; CHECK-NEXT:    [[TMP4:%.*]] = mul <2 x i16> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i16> [[TMP4]], i32 0
+; CHECK-NEXT:    [[TMP6:%.*]] = zext i16 [[TMP5]] to i32
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i16> [[TMP4]], i32 1
+; CHECK-NEXT:    [[TMP8:%.*]] = zext i16 [[TMP7]] to i32
+; CHECK-NEXT:    [[ADD:%.*]] = or i32 [[TMP6]], [[TMP8]]
 ; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 [[ADD]], 1
 ; CHECK-NEXT:    [[CONV9:%.*]] = trunc i32 [[SHR]] to i8
 ; CHECK-NEXT:    store i8 [[CONV9]], ptr null, align 1
-; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x i32> [[TMP2]], <2 x i32> poison, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <2 x i32> [[TMP3]], <2 x i32> poison, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
 ; CHECK-NEXT:    ret void
 ;
 entry:

llvm/test/Transforms/SLPVectorizer/X86/minbitwidth-transformed-operand.ll

@@ -5,24 +5,16 @@ define void @test(i64 %d.promoted.i) {
 ; CHECK-LABEL: define void @test(
 ; CHECK-SAME: i64 [[D_PROMOTED_I:%.*]]) {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x i64> <i64 poison, i64 0>, i64 [[D_PROMOTED_I]], i32 0
-; CHECK-NEXT:    [[TMP1:%.*]] = trunc <2 x i64> [[TMP0]] to <2 x i1>
-; CHECK-NEXT:    [[TMP2:%.*]] = and <2 x i1> zeroinitializer, [[TMP1]]
-; CHECK-NEXT:    [[TMP3:%.*]] = mul <2 x i1> [[TMP2]], zeroinitializer
-; CHECK-NEXT:    [[TMP4:%.*]] = or <2 x i1> [[TMP3]], zeroinitializer
-; CHECK-NEXT:    [[TMP5:%.*]] = or <2 x i1> [[TMP4]], zeroinitializer
-; CHECK-NEXT:    [[TMP6:%.*]] = or <2 x i1> [[TMP5]], zeroinitializer
-; CHECK-NEXT:    [[TMP7:%.*]] = or <2 x i1> [[TMP6]], zeroinitializer
-; CHECK-NEXT:    [[TMP8:%.*]] = or <2 x i1> [[TMP7]], zeroinitializer
-; CHECK-NEXT:    [[TMP9:%.*]] = or <2 x i1> [[TMP8]], zeroinitializer
-; CHECK-NEXT:    [[TMP10:%.*]] = or <2 x i1> [[TMP9]], zeroinitializer
-; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i1> [[TMP10]], i32 0
-; CHECK-NEXT:    [[TMP12:%.*]] = sext i1 [[TMP11]] to i32
-; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x i1> [[TMP10]], i32 1
-; CHECK-NEXT:    [[TMP14:%.*]] = sext i1 [[TMP13]] to i32
-; CHECK-NEXT:    [[TMP15:%.*]] = or i32 [[TMP12]], [[TMP14]]
-; CHECK-NEXT:    [[TMP16:%.*]] = and i32 [[TMP15]], 0
-; CHECK-NEXT:    store i32 [[TMP16]], ptr null, align 4
+; CHECK-NEXT:    [[AND_1_I:%.*]] = and i64 0, [[D_PROMOTED_I]]
+; CHECK-NEXT:    [[AND_1_I_1:%.*]] = and i64 0, 0
+; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <16 x i64> <i64 0, i64 poison, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0, i64 poison, i64 0, i64 0, i64 0, i64 0, i64 0, i64 0>, i64 [[AND_1_I_1]], i32 1
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <16 x i64> [[TMP0]], i64 [[AND_1_I]], i32 9
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc <16 x i64> [[TMP1]] to <16 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = mul <16 x i1> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = call i1 @llvm.vector.reduce.or.v16i1(<16 x i1> [[TMP3]])
+; CHECK-NEXT:    [[TMP5:%.*]] = zext i1 [[TMP4]] to i32
+; CHECK-NEXT:    [[TMP6:%.*]] = and i32 [[TMP5]], 0
+; CHECK-NEXT:    store i32 [[TMP6]], ptr null, align 4
 ; CHECK-NEXT:    ret void
 ;
 entry:

llvm/test/Transforms/SLPVectorizer/X86/root-trunc-extract-reuse.ll

@@ -8,20 +8,18 @@ define i1 @test() {
 ; CHECK:       then:
 ; CHECK-NEXT:    br label [[ELSE]]
 ; CHECK:       else:
-; CHECK-NEXT:    [[TMP0:%.*]] = phi <2 x i1> [ zeroinitializer, [[THEN]] ], [ zeroinitializer, [[ENTRY:%.*]] ]
-; CHECK-NEXT:    [[TMP1:%.*]] = zext <2 x i1> [[TMP0]] to <2 x i32>
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i1> [[TMP0]], i32 0
-; CHECK-NEXT:    [[TMP3:%.*]] = zext i1 [[TMP2]] to i32
-; CHECK-NEXT:    [[BF_CAST162:%.*]] = and i32 [[TMP3]], 0
-; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> zeroinitializer, <2 x i32> [[TMP1]], <2 x i32> <i32 3, i32 1>
-; CHECK-NEXT:    [[T13:%.*]] = and <2 x i32> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP0:%.*]] = phi <2 x i32> [ zeroinitializer, [[THEN]] ], [ zeroinitializer, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x i32> [[TMP0]], i32 0
+; CHECK-NEXT:    [[BF_CAST162:%.*]] = and i32 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> zeroinitializer, <2 x i32> [[TMP0]], <2 x i32> <i32 3, i32 1>
+; CHECK-NEXT:    [[T13:%.*]] = and <2 x i32> [[TMP2]], zeroinitializer
 ; CHECK-NEXT:    br label [[ELSE1:%.*]]
 ; CHECK:       else1:
-; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x i32> [[T13]], <2 x i32> poison, <2 x i32> <i32 poison, i32 0>
-; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x i32> [[TMP5]], i32 [[BF_CAST162]], i32 0
-; CHECK-NEXT:    [[TMP7:%.*]] = icmp ugt <2 x i32> [[TMP6]], zeroinitializer
-; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1
-; CHECK-NEXT:    ret i1 [[TMP8]]
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <2 x i32> [[T13]], <2 x i32> poison, <2 x i32> <i32 poison, i32 0>
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x i32> [[TMP3]], i32 [[BF_CAST162]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ugt <2 x i32> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i1> [[TMP5]], i32 1
+; CHECK-NEXT:    ret i1 [[TMP6]]
 ;
 entry:
   br i1 false, label %then, label %else