tcg: Fix do_nonatomic_op_* vs signed operations

The smin/smax/umin/umax operations require the operands to be
properly sign extended.  Do not drop the MO_SIGN bit from the
load, and additionally extend the val input.

Reviewed-by: LIU Zhiwei <[email protected]>
Reported-by: LIU Zhiwei <[email protected]>
Signed-off-by: Richard Henderson <[email protected]>
Message-Id: <[email protected]>

Author: rth7680

tcg/tcg-op.c

@@ -3189,8 +3189,9 @@ static void do_nonatomic_op_i32(TCGv_i32 ret, TCGv addr, TCGv_i32 val,
 
     memop = tcg_canonicalize_memop(memop, 0, 0);
 
-    tcg_gen_qemu_ld_i32(t1, addr, idx, memop & ~MO_SIGN);
-    gen(t2, t1, val);
+    tcg_gen_qemu_ld_i32(t1, addr, idx, memop);
+    tcg_gen_ext_i32(t2, val, memop);
+    gen(t2, t1, t2);
     tcg_gen_qemu_st_i32(t2, addr, idx, memop);
 
     tcg_gen_ext_i32(ret, (new_val ? t2 : t1), memop);
@@ -3232,8 +3233,9 @@ static void do_nonatomic_op_i64(TCGv_i64 ret, TCGv addr, TCGv_i64 val,
 
     memop = tcg_canonicalize_memop(memop, 1, 0);
 
-    tcg_gen_qemu_ld_i64(t1, addr, idx, memop & ~MO_SIGN);
-    gen(t2, t1, val);
+    tcg_gen_qemu_ld_i64(t1, addr, idx, memop);
+    tcg_gen_ext_i64(t2, val, memop);
+    gen(t2, t1, t2);
     tcg_gen_qemu_st_i64(t2, addr, idx, memop);
 
     tcg_gen_ext_i64(ret, (new_val ? t2 : t1), memop);