target/hppa: Fix trans_ds for hppa64

This instruction always uses the input carry from bit 32,
but produces all 16 output carry bits.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

target/hppa/translate.c

@@ -803,6 +803,12 @@ static bool cond_need_cb(int c)
     return c == 4 || c == 5;
 }
 
+/* Need extensions from TCGv_i32 to TCGv_reg. */
+static bool cond_need_ext(DisasContext *ctx, bool d)
+{
+    return TARGET_REGISTER_BITS == 64 && !d;
+}
+
 /*
  * Compute conditional for arithmetic.  See Page 5-3, Table 5-1, of
  * the Parisc 1.1 Architecture Reference Manual for details.
@@ -1040,6 +1046,22 @@ static DisasCond do_unit_cond(unsigned cf, TCGv_reg res,
     return cond;
 }
 
+static TCGv_reg get_carry(DisasContext *ctx, bool d,
+                          TCGv_reg cb, TCGv_reg cb_msb)
+{
+    if (cond_need_ext(ctx, d)) {
+        TCGv_reg t = tcg_temp_new();
+        tcg_gen_extract_reg(t, cb, 32, 1);
+        return t;
+    }
+    return cb_msb;
+}
+
+static TCGv_reg get_psw_carry(DisasContext *ctx, bool d)
+{
+    return get_carry(ctx, d, cpu_psw_cb, cpu_psw_cb_msb);
+}
+
 /* Compute signed overflow for addition.  */
 static TCGv_reg do_add_sv(DisasContext *ctx, TCGv_reg res,
                           TCGv_reg in1, TCGv_reg in2)
@@ -2712,6 +2734,7 @@ static bool trans_dcor_i(DisasContext *ctx, arg_rr_cf *a)
 static bool trans_ds(DisasContext *ctx, arg_rrr_cf *a)
 {
     TCGv_reg dest, add1, add2, addc, zero, in1, in2;
+    TCGv_reg cout;
 
     nullify_over(ctx);
 
@@ -2726,18 +2749,20 @@ static bool trans_ds(DisasContext *ctx, arg_rrr_cf *a)
 
     /* Form R1 << 1 | PSW[CB]{8}.  */
     tcg_gen_add_reg(add1, in1, in1);
-    tcg_gen_add_reg(add1, add1, cpu_psw_cb_msb);
+    tcg_gen_add_reg(add1, add1, get_psw_carry(ctx, false));
 
-    /* Add or subtract R2, depending on PSW[V].  Proper computation of
-       carry{8} requires that we subtract via + ~R2 + 1, as described in
-       the manual.  By extracting and masking V, we can produce the
-       proper inputs to the addition without movcond.  */
-    tcg_gen_sari_reg(addc, cpu_psw_v, TARGET_REGISTER_BITS - 1);
+    /*
+     * Add or subtract R2, depending on PSW[V].  Proper computation of
+     * carry requires that we subtract via + ~R2 + 1, as described in
+     * the manual.  By extracting and masking V, we can produce the
+     * proper inputs to the addition without movcond.
+     */
+    tcg_gen_sextract_reg(addc, cpu_psw_v, 31, 1);
     tcg_gen_xor_reg(add2, in2, addc);
     tcg_gen_andi_reg(addc, addc, 1);
-    /* ??? This is only correct for 32-bit.  */
-    tcg_gen_add2_i32(dest, cpu_psw_cb_msb, add1, zero, add2, zero);
-    tcg_gen_add2_i32(dest, cpu_psw_cb_msb, dest, cpu_psw_cb_msb, addc, zero);
+
+    tcg_gen_add2_reg(dest, cpu_psw_cb_msb, add1, zero, add2, zero);
+    tcg_gen_add2_reg(dest, cpu_psw_cb_msb, dest, cpu_psw_cb_msb, addc, zero);
 
     /* Write back the result register.  */
     save_gpr(ctx, a->t, dest);
@@ -2747,7 +2772,8 @@ static bool trans_ds(DisasContext *ctx, arg_rrr_cf *a)
     tcg_gen_xor_reg(cpu_psw_cb, cpu_psw_cb, dest);
 
     /* Write back PSW[V] for the division step.  */
-    tcg_gen_neg_reg(cpu_psw_v, cpu_psw_cb_msb);
+    cout = get_psw_carry(ctx, false);
+    tcg_gen_neg_reg(cpu_psw_v, cout);
     tcg_gen_xor_reg(cpu_psw_v, cpu_psw_v, in2);
 
     /* Install the new nullification.  */
@@ -2757,7 +2783,7 @@ static bool trans_ds(DisasContext *ctx, arg_rrr_cf *a)
             /* ??? The lshift is supposed to contribute to overflow.  */
             sv = do_add_sv(ctx, dest, add1, add2);
         }
-        ctx->null_cond = do_cond(a->cf, dest, cpu_psw_cb_msb, sv);
+        ctx->null_cond = do_cond(a->cf, dest, cout, sv);
     }
 
     return nullify_end(ctx);