diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -8363,7 +8363,8 @@
           }
         }
       } else {
-        Builder.SetInsertPoint(cast<Instruction>(User));
+        Builder.SetInsertPoint(VecI->getParent(),
+                               std::next(VecI->getIterator()));
         Value *NewInst = ExtractAndExtendIfNeeded(Vec);
         CSEBlocks.insert(cast<Instruction>(User)->getParent());
         User->replaceUsesOfWith(Scalar, NewInst);
diff --git a/llvm/test/Transforms/SLPVectorizer/pr55796.ll b/llvm/test/Transforms/SLPVectorizer/pr55796.ll
new file mode 100644
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/pr55796.ll
@@ -0,0 +1,105 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=slp-vectorizer -S < %s | FileCheck %s
+; target triple = "x86_64-unknown-linux-gnu"
+
+; REQUIRES: asserts
+
+define i32 @"foo"(i8 addrspace(3)* %tmp_buf) #0 {
+; CHECK-LABEL: @foo(
+; CHECK-NEXT:  always_continue:
+; CHECK-NEXT:    [[LOCAL_TMP_6:%.*]] = load i32, i32 addrspace(3)* undef, align 4
+; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> poison, i32 [[LOCAL_TMP_6]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> [[TMP0]], i32 [[LOCAL_TMP_6]], i32 1
+; CHECK-NEXT:    [[TMP2:%.*]] = sub nsw <2 x i32> <i32 110, i32 0>, [[TMP1]]
+; CHECK-NEXT:    [[SHUFFLE10:%.*]] = shufflevector <2 x i32> [[TMP2]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 1, i32 1>
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i32> [[SHUFFLE10]], i32 1
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x i32> [[SHUFFLE10]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x i32> poison, i32 [[TMP3]], i32 0
+; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x i32> [[TMP5]], i32 [[TMP4]], i32 1
+; CHECK-NEXT:    [[SHUFFLE14:%.*]] = shufflevector <2 x i32> [[TMP6]], <2 x i32> poison, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1>
+; CHECK-NEXT:    br i1 undef, label [[GUARDED209_7:%.*]], label [[DEOPT210_SPLIT_LOOP_EXIT1140:%.*]]
+; CHECK:       deopt210.split.loop.exit1140:
+; CHECK-NEXT:    [[TMP7:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> <i32 0, i32 90, i32 72, i32 56, i32 42, i32 30, i32 20, i32 12>)
+; CHECK-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[SHUFFLE14]])
+; CHECK-NEXT:    [[OP_RDX15:%.*]] = add i32 [[TMP7]], [[TMP8]]
+; CHECK-NEXT:    br label [[DEOPT210:%.*]]
+; CHECK:       deopt210.split.loop.exit1164:
+; CHECK-NEXT:    br label [[DEOPT210]]
+; CHECK:       deopt210:
+; CHECK-NEXT:    [[LOCAL_7_3370_LCSSA1006:%.*]] = phi i32 [ [[OP_RDX15]], [[DEOPT210_SPLIT_LOOP_EXIT1140]] ], [ undef, [[DEOPT210_SPLIT_LOOP_EXIT1164:%.*]] ]
+; CHECK-NEXT:    ret i32 undef
+; CHECK:       deopt215.split.loop.exit:
+; CHECK-NEXT:    [[TMP9:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> <i32 90, i32 72, i32 56, i32 42, i32 30, i32 20, i32 12, i32 6>)
+; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <4 x i32> poison, i32 [[TMP3]], i32 0
+; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[TMP10]], <4 x i32> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = add <4 x i32> [[SHUFFLE]], [[SHUFFLE10]]
+; CHECK-NEXT:    [[TMP12:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP11]])
+; CHECK-NEXT:    [[OP_RDX11:%.*]] = add i32 [[TMP9]], undef
+; CHECK-NEXT:    [[OP_RDX12:%.*]] = add i32 [[OP_RDX11]], [[TMP3]]
+; CHECK-NEXT:    [[OP_RDX13:%.*]] = add i32 [[TMP12]], [[OP_RDX12]]
+; CHECK-NEXT:    [[DEOPTCALL216:%.*]] = call i32 (...) @llvm.experimental.deoptimize.i32(i32 12) [ "deopt"(i32 0, i32 1, i32 0, i32 142, i32 4, i32 19, i32 1, i32 0, i8 addrspace(1)* undef, i32 3, i32 undef, i32 0, i8 addrspace(1)* undef, i32 3, i32 undef, i32 4, i64 undef, i32 7, i8* null, i32 4, i64 undef, i32 7, i8* null, i32 4, double undef, i32 7, i8* null, i32 3, i32 [[LOCAL_TMP_6]], i32 3, i32 [[OP_RDX13]], i32 3, i32 undef, i32 3, i32 undef, i32 3, i32 undef, i32 7, i8* null, i32 3, i32 undef, i32 3, i32 undef, i32 3, float undef, i32 3, i32 undef, i32 0, i8 addrspace(1)* undef, i32 7, i8* null, i32 7, i8* null, i32 0, i8 addrspace(1)* undef) ]
+; CHECK-NEXT:    ret i32 [[DEOPTCALL216]]
+; CHECK:       guarded209.7:
+; CHECK-NEXT:    br label [[DEOPT215_SPLIT_LOOP_EXIT:%.*]]
+;
+always_continue:
+  %local.tmp.6 = load i32, i32 addrspace(3)* undef, align 4
+  %0 = sub nsw i32 110, %local.tmp.6
+  %local_6_.lcssa.neg = sub nsw i32 0, %local.tmp.6
+  br i1 undef, label %guarded209.7, label %deopt210.split.loop.exit1140
+
+deopt210.split.loop.exit1140:                     ; preds = %always_continue
+  %1 = add i32 0, %0
+  %2 = add i32 %1, %local_6_.lcssa.neg
+  %3 = add i32 %2, 90
+  %4 = add i32 %3, %local_6_.lcssa.neg
+  %5 = add i32 %4, 72
+  %6 = add i32 %5, %local_6_.lcssa.neg
+  %7 = add i32 %6, 56
+  %8 = add i32 %7, %local_6_.lcssa.neg
+  %9 = add i32 %8, 42
+  %10 = add i32 %9, %local_6_.lcssa.neg
+  %11 = add i32 %10, 30
+  %12 = add i32 %11, %local_6_.lcssa.neg
+  %13 = add i32 %12, 20
+  %14 = add i32 %13, %local_6_.lcssa.neg
+  %15 = add i32 %14, 12
+  br label %deopt210
+
+deopt210.split.loop.exit1164:                     ; No predecessors!
+  br label %deopt210
+
+deopt210:                                         ; preds = %deopt210.split.loop.exit1164, %deopt210.split.loop.exit1140
+  %local_7_3370.lcssa1006 = phi i32 [ %15, %deopt210.split.loop.exit1140 ], [ undef, %deopt210.split.loop.exit1164 ]
+  ret i32 undef
+
+deopt215.split.loop.exit:                         ; preds = %guarded209.7
+  %16 = add i32 undef, %0
+  %17 = add i32 %16, %local_6_.lcssa.neg
+  %18 = add i32 %17, 90
+  %19 = add i32 %18, %local_6_.lcssa.neg
+  %20 = add i32 %19, 72
+  %21 = add i32 %20, %local_6_.lcssa.neg
+  %22 = add i32 %21, 56
+  %23 = add i32 %22, %local_6_.lcssa.neg
+  %24 = add i32 %23, 42
+  %25 = add i32 %24, %local_6_.lcssa.neg
+  %26 = add i32 %25, 30
+  %27 = add i32 %26, %local_6_.lcssa.neg
+  %28 = add i32 %27, 20
+  %29 = add i32 %28, %local_6_.lcssa.neg
+  %30 = add i32 %29, 12
+  %31 = add i32 %30, %local_6_.lcssa.neg
+  %32 = add i32 %31, 6
+  %deoptcall216 = call i32 (...) @llvm.experimental.deoptimize.i32(i32 12) [ "deopt"(i32 0, i32 1, i32 0, i32 142, i32 4, i32 19, i32 1, i32 0, i8 addrspace(1)* undef, i32 3, i32 undef, i32 0, i8 addrspace(1)* undef, i32 3, i32 undef, i32 4, i64 undef, i32 7, i8* null, i32 4, i64 undef, i32 7, i8* null, i32 4, double undef, i32 7, i8* null, i32 3, i32 %local.tmp.6, i32 3, i32 %32, i32 3, i32 undef, i32 3, i32 undef, i32 3, i32 undef, i32 7, i8* null, i32 3, i32 undef, i32 3, i32 undef, i32 3, float undef, i32 3, i32 undef, i32 0, i8 addrspace(1)* undef, i32 7, i8* null, i32 7, i8* null, i32 0, i8 addrspace(1)* undef) ]
+  ret i32 %deoptcall216
+
+guarded209.7:                                     ; preds = %always_continue
+  br label %deopt215.split.loop.exit
+}
+
+
+declare i32 @llvm.experimental.deoptimize.i32(...)
+
+attributes #0 = { "target-features"="-avx512pf,+avx512f,+avx512bw" }
+attributes #1 = { willreturn "target-features"="-avx512pf,+avx512f,+avx512bw" }