diff --git a/test/vector_endianness.ll b/test/vector_endianness.ll
new file mode 100644
index 00000000..a8980ec8
--- /dev/null
+++ b/test/vector_endianness.ll
@@ -0,0 +1,181 @@
+; This file is part of SymCC.
+;
+; SymCC is free software: you can redistribute it and/or modify it under the
+; terms of the GNU General Public License as published by the Free Software
+; Foundation, either version 3 of the License, or (at your option) any later
+; version.
+;
+; SymCC is distributed in the hope that it will be useful, but WITHOUT ANY
+; WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
+; A PARTICULAR PURPOSE. See the GNU General Public License for more details.
+;
+; You should have received a copy of the GNU General Public License along with
+; SymCC. If not, see <https://www.gnu.org/licenses/>.
+
+; Verify that symbolic vector values use the same lane and byte order no matter
+; whether they come from memory or are built by vector instructions. Loads and
+; stores treat a vector like an integer of the same width (on little-endian
+; targets, lane 0 is in the least significant bits), so extracting, inserting
+; and computing on lanes must agree with that layout. Otherwise the solver
+; reasons about the wrong lane, or about byte-swapped lane values, and generates
+; inputs that don't take the branch.
+;
+; The 16-byte input is a <4 x i32>. Each scenario uses a distinct lane, so each
+; expected solution is unique:
+;   1. vector load, extractelement of lane 1 == 0x11223344
+;        -> bytes 4..7 = 44 33 22 11
+;   2. vector load, add 1 to every lane (carry crosses a byte), vector store,
+;      scalar load of lane 2 == 0x100
+;        -> bytes 8..11 = ff 00 00 00
+;   3. scalar load of lane 3, insertelement into lane 3, vector store, scalar
+;      load of lane 3 == 0x55667788
+;        -> bytes 12..15 = 88 77 66 55
+;
+; The simple backend prints its solutions, so we check them directly. The QSYM
+; backend reports them through the test-case handler, which prints "sN solved"
+; only if the generated input really satisfies the condition of scenario N.
+;
+; Since the bitcode is written by hand, we first run llc on it because it
+; performs a validity check, whereas Clang doesn't.
+
+; RUN: llc %s -o /dev/null
+; RUN: %symcc %s -o %t
+; RUN: env SYMCC_MEMORY_INPUT=1 %t 2>&1 | %filecheck %s
+
+target triple = "x86_64-pc-linux-gnu"
+
+@s1_hit = private constant [7 x i8] c"s1 hit\0A"
+@s2_hit = private constant [7 x i8] c"s2 hit\0A"
+@s3_hit = private constant [7 x i8] c"s3 hit\0A"
+@s1_end = private constant [7 x i8] c"s1 end\0A"
+@s2_end = private constant [7 x i8] c"s2 end\0A"
+@s3_end = private constant [7 x i8] c"s3 end\0A"
+@s1_solved = private constant [10 x i8] c"s1 solved\0A"
+@s2_solved = private constant [10 x i8] c"s2 solved\0A"
+@s3_solved = private constant [10 x i8] c"s3 solved\0A"
+
+define i32 @main(i32 %argc, ptr %argv) {
+entry:
+  call void @symcc_set_test_case_handler(ptr @handle_test_case)
+  %in = alloca <4 x i32>, align 16
+  %out = alloca <4 x i32>, align 16
+  store <4 x i32> zeroinitializer, ptr %in, align 16
+  call void @symcc_make_symbolic(ptr %in, i64 16)
+
+  ; Scenario 1: extract a lane from a vector loaded from memory.
+  %v = load <4 x i32>, ptr %in, align 16
+  %lane1 = extractelement <4 x i32> %v, i32 1
+  %c1 = icmp eq i32 %lane1, 287454020
+  ; SIMPLE: Trying to solve
+  ; SIMPLE: Found diverging input
+  ; SIMPLE-DAG: stdin4 -> #x44
+  ; SIMPLE-DAG: stdin5 -> #x33
+  ; SIMPLE-DAG: stdin6 -> #x22
+  ; SIMPLE-DAG: stdin7 -> #x11
+  ; QSYM: s1 solved
+  br i1 %c1, label %s1_hit, label %s1_end
+
+s1_hit:
+  call i64 @write(i32 2, ptr @s1_hit, i64 7)
+  br label %s1_end
+
+s1_end:
+  call i64 @write(i32 2, ptr @s1_end, i64 7)
+  ; ANY: s1 end
+
+  ; Scenario 2: lane-wise arithmetic with a carry across bytes, stored back to
+  ; memory and read as a scalar.
+  %sum = add <4 x i32> %v, <i32 1, i32 1, i32 1, i32 1>
+  store <4 x i32> %sum, ptr %out, align 16
+  %p2 = getelementptr i32, ptr %out, i64 2
+  %lane2 = load i32, ptr %p2, align 8
+  %c2 = icmp eq i32 %lane2, 256
+  ; SIMPLE: Trying to solve
+  ; SIMPLE: Found diverging input
+  ; SIMPLE-DAG: stdin8 -> #xff
+  ; SIMPLE-DAG: stdin9 -> #x00
+  ; SIMPLE-DAG: stdin10 -> #x00
+  ; SIMPLE-DAG: stdin11 -> #x00
+  ; QSYM: s2 solved
+  br i1 %c2, label %s2_hit, label %s2_end
+
+s2_hit:
+  call i64 @write(i32 2, ptr @s2_hit, i64 7)
+  br label %s2_end
+
+s2_end:
+  call i64 @write(i32 2, ptr @s2_end, i64 7)
+  ; ANY: s2 end
+
+  ; Scenario 3: insert a symbolic scalar into a concrete vector, store the
+  ; vector and read the lane back as a scalar.
+  %p3in = getelementptr i32, ptr %in, i64 3
+  %x3 = load i32, ptr %p3in, align 4
+  %ins = insertelement <4 x i32> zeroinitializer, i32 %x3, i32 3
+  store <4 x i32> %ins, ptr %out, align 16
+  %p3 = getelementptr i32, ptr %out, i64 3
+  %lane3 = load i32, ptr %p3, align 4
+  %c3 = icmp eq i32 %lane3, 1432778632
+  ; SIMPLE: Trying to solve
+  ; SIMPLE: Found diverging input
+  ; SIMPLE-DAG: stdin12 -> #x88
+  ; SIMPLE-DAG: stdin13 -> #x77
+  ; SIMPLE-DAG: stdin14 -> #x66
+  ; SIMPLE-DAG: stdin15 -> #x55
+  ; QSYM: s3 solved
+  br i1 %c3, label %s3_hit, label %s3_end
+
+s3_hit:
+  call i64 @write(i32 2, ptr @s3_hit, i64 7)
+  br label %s3_end
+
+s3_end:
+  call i64 @write(i32 2, ptr @s3_end, i64 7)
+  ; ANY: s3 end
+  ret i32 0
+}
+
+; Print "sN solved" for each scenario whose condition the generated input
+; satisfies (QSYM backend only; the simple backend doesn't call handlers).
+define void @handle_test_case(ptr %data, i64 %len) {
+entry:
+  %long_enough = icmp uge i64 %len, 16
+  br i1 %long_enough, label %check1, label %done
+
+check1:
+  %q1 = getelementptr i8, ptr %data, i64 4
+  %d1 = load i32, ptr %q1, align 1
+  %k1 = icmp eq i32 %d1, 287454020
+  br i1 %k1, label %ok1, label %check2
+
+ok1:
+  call i64 @write(i32 2, ptr @s1_solved, i64 10)
+  br label %check2
+
+check2:
+  %q2 = getelementptr i8, ptr %data, i64 8
+  %d2 = load i32, ptr %q2, align 1
+  %k2 = icmp eq i32 %d2, 255
+  br i1 %k2, label %ok2, label %check3
+
+ok2:
+  call i64 @write(i32 2, ptr @s2_solved, i64 10)
+  br label %check3
+
+check3:
+  %q3 = getelementptr i8, ptr %data, i64 12
+  %d3 = load i32, ptr %q3, align 1
+  %k3 = icmp eq i32 %d3, 1432778632
+  br i1 %k3, label %ok3, label %done
+
+ok3:
+  call i64 @write(i32 2, ptr @s3_solved, i64 10)
+  br label %done
+
+done:
+  ret void
+}
+
+declare void @symcc_make_symbolic(ptr, i64)
+declare void @symcc_set_test_case_handler(ptr)
+declare i64 @write(i32, ptr, i64)