[SLP]Fix PR101213: Reuse extractelement, only if its vector operand comes before new vector value.

When trying to reuse extractelement instruction, need to check that it
is inserted into proper position. Its original vector operand should
come before new vector value, otherwise new extractelement instruction
must be generated.

Fixes llvm#101213

(cherry picked from commit f70f122)

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -13889,11 +13889,16 @@ Value *BoUpSLP::vectorizeTree(
         }
         if (!Ex) {
           // "Reuse" the existing extract to improve final codegen.
-          if (auto *ES = dyn_cast<ExtractElementInst>(Scalar)) {
+          if (auto *ES = dyn_cast<ExtractElementInst>(Scalar);
+              ES && isa<Instruction>(Vec)) {
             Value *V = ES->getVectorOperand();
             if (const TreeEntry *ETE = getTreeEntry(V))
               V = ETE->VectorizedValue;
-            Ex = Builder.CreateExtractElement(V, ES->getIndexOperand());
+            if (auto *IV = dyn_cast<Instruction>(V);
+                !IV || IV == Vec || (IV->getParent() == cast<Instruction>(Vec)->getParent() && IV->comesBefore(cast<Instruction>(Vec))))
+              Ex = Builder.CreateExtractElement(V, ES->getIndexOperand());
+            else
+              Ex = Builder.CreateExtractElement(Vec, Lane);
           } else if (ReplaceGEP) {
             // Leave the GEPs as is, they are free in most cases and better to
             // keep them as GEPs.

llvm/test/Transforms/SLPVectorizer/X86/extract-vectorized-operand.ll

@@ -0,0 +1,49 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -slp-threshold=-99999 < %s -mtriple=x86_64-unknown-linux-gnu | FileCheck %s
+
+define void @test() {
+; CHECK-LABEL: define void @test() {
+; CHECK-NEXT:  [[BB:.*]]:
+; CHECK-NEXT:    br label %[[BB43:.*]]
+; CHECK:       [[BB20:.*]]:
+; CHECK-NEXT:    br label %[[BB105:.*]]
+; CHECK:       [[BB43]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = phi <2 x ptr addrspace(1)> [ [[TMP1:%.*]], %[[BB51:.*]] ], [ zeroinitializer, %[[BB]] ]
+; CHECK-NEXT:    br i1 false, label %[[BB105]], label %[[BB51]]
+; CHECK:       [[BB51]]:
+; CHECK-NEXT:    [[TMP1]] = phi <2 x ptr addrspace(1)> [ poison, %[[BB54:.*]] ], [ zeroinitializer, %[[BB43]] ]
+; CHECK-NEXT:    br label %[[BB43]]
+; CHECK:       [[BB54]]:
+; CHECK-NEXT:    br label %[[BB51]]
+; CHECK:       [[BB105]]:
+; CHECK-NEXT:    [[PHI106:%.*]] = phi ptr addrspace(1) [ null, %[[BB20]] ], [ null, %[[BB43]] ]
+; CHECK-NEXT:    ret void
+;
+bb:
+  %0 = shufflevector <2 x ptr addrspace(1)> zeroinitializer, <2 x ptr addrspace(1)> zeroinitializer, <2 x i32> <i32 1, i32 0>
+  %1 = extractelement <2 x ptr addrspace(1)> %0, i32 0
+  %2 = extractelement <2 x ptr addrspace(1)> %0, i32 1
+  br label %bb43
+
+bb20:
+  br label %bb105
+
+bb43:
+  %phi441 = phi ptr addrspace(1) [ %4, %bb51 ], [ %2, %bb ]
+  %phi452 = phi ptr addrspace(1) [ %5, %bb51 ], [ %1, %bb ]
+  br i1 false, label %bb105, label %bb51
+
+bb51:
+  %3 = phi <2 x ptr addrspace(1)> [ poison, %bb54 ], [ zeroinitializer, %bb43 ]
+  %4 = extractelement <2 x ptr addrspace(1)> %3, i32 0
+  %5 = extractelement <2 x ptr addrspace(1)> %3, i32 1
+  br label %bb43
+
+bb54:
+  br label %bb51
+
+bb105:
+  %phi106 = phi ptr addrspace(1) [ %1, %bb20 ], [ null, %bb43 ]
+  ret void
+}
+