diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -4654,8 +4654,13 @@
   }
 
   if (!Call.use_empty() && !Call.isMustTailCall())
-    if (Value *ReturnedArg = Call.getReturnedArgOperand())
-      return replaceInstUsesWith(Call, ReturnedArg);
+    if (Value *ReturnedArg = Call.getReturnedArgOperand()) {
+      Type *CallTy = Call.getType();
+      Type *RetArgTy = ReturnedArg->getType();
+      if (RetArgTy->canLosslesslyBitCastTo(CallTy))
+        return replaceInstUsesWith(
+            Call, Builder.CreateBitOrPointerCast(ReturnedArg, CallTy));
+    }
 
   if (isAllocLikeFn(&Call, &TLI))
     return visitAllocSite(Call);
diff --git a/llvm/test/Transforms/InstCombine/call-returned.ll b/llvm/test/Transforms/InstCombine/call-returned.ll
--- a/llvm/test/Transforms/InstCombine/call-returned.ll
+++ b/llvm/test/Transforms/InstCombine/call-returned.ll
@@ -3,6 +3,8 @@
 
 declare i32 @passthru_i32(i32 returned)
 declare i8* @passthru_p8(i8* returned)
+declare i8* @passthru_p8_from_p32(i32* returned)
+declare <8 x i8> @passthru_8i8v_from_2i32v(<2 x i32> returned)
 
 define i32 @returned_const_int_arg() {
 ; CHECK-LABEL: @returned_const_int_arg(
@@ -22,6 +24,46 @@
   ret i8* %x
 }
 
+define i8* @returned_const_ptr_arg_casted() {
+; CHECK-LABEL: @returned_const_ptr_arg_casted(
+; CHECK-NEXT:    [[X:%.*]] = call i8* @passthru_p8_from_p32(i32* null)
+; CHECK-NEXT:    ret i8* null
+;
+  %x = call i8* @passthru_p8_from_p32(i32* null)
+  ret i8* %x
+}
+
+define i8* @returned_ptr_arg_casted(i32* %a) {
+; CHECK-LABEL: @returned_ptr_arg_casted(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[A:%.*]] to i8*
+; CHECK-NEXT:    [[X:%.*]] = call i8* @passthru_p8_from_p32(i32* [[A]])
+; CHECK-NEXT:    ret i8* [[TMP1]]
+;
+  %x = call i8* @passthru_p8_from_p32(i32* %a)
+  ret i8* %x
+}
+
+@GV = constant <2 x i32> zeroinitializer
+define <8 x i8> @returned_const_vec_arg_casted() {
+; CHECK-LABEL: @returned_const_vec_arg_casted(
+; CHECK-NEXT:    [[X:%.*]] = call <8 x i8> @passthru_8i8v_from_2i32v(<2 x i32> zeroinitializer)
+; CHECK-NEXT:    ret <8 x i8> zeroinitializer
+;
+  %v = load <2 x i32>, <2 x i32>* @GV
+  %x = call <8 x i8> @passthru_8i8v_from_2i32v(<2 x i32> %v)
+  ret <8 x i8> %x
+}
+
+define <8 x i8> @returned_vec_arg_casted(<2 x i32> %a) {
+; CHECK-LABEL: @returned_vec_arg_casted(
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <2 x i32> [[A:%.*]] to <8 x i8>
+; CHECK-NEXT:    [[X:%.*]] = call <8 x i8> @passthru_8i8v_from_2i32v(<2 x i32> [[A]])
+; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
+;
+  %x = call <8 x i8> @passthru_8i8v_from_2i32v(<2 x i32> %a)
+  ret <8 x i8> %x
+}
+
 define i32 @returned_var_arg(i32 %arg) {
 ; CHECK-LABEL: @returned_var_arg(
 ; CHECK-NEXT:    [[X:%.*]] = call i32 @passthru_i32(i32 [[ARG:%.*]])
@@ -48,3 +90,4 @@
   %x = musttail call i32 @passthru_i32(i32 %arg)
   ret i32 %x
 }
+