We were only constructing the 64-bit element, and not
replicating the 64-bit element across the rest of the vector.

Cc: qemu-stable@nongnu.org
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---

This bug is visible running RISU for aarch64 on x86_32.


r~

---
 tcg/i386/tcg-target.inc.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/tcg/i386/tcg-target.inc.c b/tcg/i386/tcg-target.inc.c
index 223dba9c8c..7f61eeedd0 100644
--- a/tcg/i386/tcg-target.inc.c
+++ b/tcg/i386/tcg-target.inc.c
@@ -2855,9 +2855,13 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
         goto gen_simd;
 #if TCG_TARGET_REG_BITS == 32
     case INDEX_op_dup2_vec:
-        /* Constraints have already placed both 32-bit inputs in xmm regs.  */
-        insn = OPC_PUNPCKLDQ;
-        goto gen_simd;
+        /* First merge the two 32-bit inputs to a single 64-bit element. */
+        tcg_out_vex_modrm(s, OPC_PUNPCKLDQ, a0, a1, a2);
+        /* Then replicate the 64-bit elements across the rest of the vector. */
+        if (type != TCG_TYPE_V64) {
+            tcg_out_dup_vec(s, type, MO_64, a0, a0);
+        }
+        break;
 #endif
     case INDEX_op_abs_vec:
         insn = abs_insn[vece];
-- 
2.20.1