diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -4094,7 +4094,7 @@
     if (!IsScalableVec && Ty->isSized()) {
       Value *P;
       uint64_t C;
-      uint64_t TyAllocSize = Q.DL.getTypeAllocSize(Ty);
+      uint64_t TyAllocSize = Q.DL.getTypeAllocSize(Ty).getFixedSize();
       // getelementptr P, N -> P if P points to a type of zero size.
       if (TyAllocSize == 0 && Ops[0]->getType() == GEPTy)
         return Ops[0];
@@ -4139,12 +4139,16 @@
     }
   }
 
-  if (!IsScalableVec && Q.DL.getTypeAllocSize(LastType) == 1 &&
+  // The following folds are only valid when GEP index types are scalar types.
+  // Bailout early if GEP index type is vector type.
+  if (!IsScalableVec && !isa<VectorType>(Ops.back()->getType()) &&
+      Q.DL.getTypeAllocSize(LastType).getFixedSize() == 1 &&
       all_of(Ops.slice(1).drop_back(1),
              [](Value *Idx) { return match(Idx, m_Zero()); })) {
     unsigned IdxWidth =
         Q.DL.getIndexSizeInBits(Ops[0]->getType()->getPointerAddressSpace());
-    if (Q.DL.getTypeSizeInBits(Ops.back()->getType()) == IdxWidth) {
+    if (Q.DL.getTypeSizeInBits(Ops.back()->getType()).getFixedSize() ==
+        IdxWidth) {
       APInt BasePtrOffset(IdxWidth, 0);
       Value *StrippedBasePtr =
           Ops[0]->stripAndAccumulateInBoundsConstantOffsets(Q.DL,
diff --git a/llvm/test/Transforms/InstSimplify/cast.ll b/llvm/test/Transforms/InstSimplify/cast.ll
--- a/llvm/test/Transforms/InstSimplify/cast.ll
+++ b/llvm/test/Transforms/InstSimplify/cast.ll
@@ -27,6 +27,7 @@
 ; CHECK: ret i8* %V
 }
 
+; This test checks for fold "gep (gep V, C), (sub 0, V) -> C"
 define i32 @test4() {
 ; CHECK-LABEL: @test4(
   %alloca = alloca i32, align 4                                     ; alloca + 0
@@ -40,6 +41,7 @@
 ; CHECK-NEXT: ret i32 4
 }
 
+; This test checks for fold "gep (gep V, C), (xor V, -1) -> C-1"
 define i32 @test5() {
 ; CHECK-LABEL: @test5(
   %alloca = alloca i32, align 4                                     ; alloca + 0
@@ -48,7 +50,27 @@
   %pti = ptrtoint i32* %alloca to i32                               ; alloca
   %sub = xor i32 %pti, -1                                           ; ~alloca
   %add = getelementptr [4 x i8], [4 x i8]* %bc, i32 0, i32 %sub     ; alloca + 4 - alloca - 1 == 3
-  %add_to_int = ptrtoint i8* %add to i32                            ; 4
-  ret i32 %add_to_int                                               ; 4
+  %add_to_int = ptrtoint i8* %add to i32                            ; 3
+  ret i32 %add_to_int                                               ; 3
 ; CHECK-NEXT: ret i32 3
 }
+
+; Negative tests
+
+; The above folds are only valid when gep index type is scalar type. These negative
+; tests check the folding codepath won't be executed, and assertion won't be triggered
+; that we are not trying to get fixed size on a scalable type.
+
+define <1 x i8*> @gep_index_vector(i8* %a, <1 x i32> %offset) {
+; CHECK-LABEL: @gep_index_vector
+  %v = getelementptr i8, i8* %a, <1 x i32> %offset
+  ret <1 x i8*> %v
+; CHECK: ret <1 x i8*> %v
+}
+
+define <vscale x 1 x i8*> @gep_index_vector_scalable(i8* %a, <vscale x 1 x i32> %offset) {
+; CHECK-LABEL: @gep_index_vector_scalable
+  %v = getelementptr i8, i8* %a, <vscale x 1 x i32> %offset
+  ret <vscale x 1 x i8*> %v
+; CHECK: ret <vscale x 1 x i8*> %v
+}