* [PATCH] Fix TYPE_EMPTY_P handling - x86_64 ABI issue (PR target/84502)
@ 2018-02-21 22:49 Jakub Jelinek
2018-02-22 7:40 ` Richard Biener
0 siblings, 1 reply; 2+ messages in thread
From: Jakub Jelinek @ 2018-02-21 22:49 UTC (permalink / raw)
To: Richard Biener, Jeff Law, Jason Merrill, Marek Polacek; +Cc: gcc-patches
Hi!
The following testcase shows that we set TYPE_EMPTY_P just one of the
possibly many variants of an aggregate type.
On the testcase, in one case (in the callee) we check TYPE_EMPTY_P on X which is
the type variant for the using type, and in another case we check it on the
A<int> type instead, which has TYPE_EMPTY_P set.
Bootstrap/regtest pending on x86_64-linux and i686-linux, ok if it passes?
2018-02-21 Jakub Jelinek <jakub@redhat.com>
PR target/84502
* stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
to all type variants.
* g++.dg/torture/pr84502.C: New test.
--- gcc/stor-layout.c.jj 2018-02-13 21:23:29.187981310 +0100
+++ gcc/stor-layout.c 2018-02-21 21:43:24.783522853 +0100
@@ -1883,6 +1883,9 @@ finalize_type_size (tree type)
&& TREE_CODE (TYPE_SIZE_UNIT (type)) != INTEGER_CST)
TYPE_SIZE_UNIT (type) = variable_size (TYPE_SIZE_UNIT (type));
+ /* Handle empty records as per the x86-64 psABI. */
+ TYPE_EMPTY_P (type) = targetm.calls.empty_record_p (type);
+
/* Also layout any other variants of the type. */
if (TYPE_NEXT_VARIANT (type)
|| type != TYPE_MAIN_VARIANT (type))
@@ -1895,6 +1898,7 @@ finalize_type_size (tree type)
unsigned int precision = TYPE_PRECISION (type);
unsigned int user_align = TYPE_USER_ALIGN (type);
machine_mode mode = TYPE_MODE (type);
+ bool empty_p = TYPE_EMPTY_P (type);
/* Copy it into all variants. */
for (variant = TYPE_MAIN_VARIANT (type);
@@ -1911,11 +1915,9 @@ finalize_type_size (tree type)
SET_TYPE_ALIGN (variant, valign);
TYPE_PRECISION (variant) = precision;
SET_TYPE_MODE (variant, mode);
+ TYPE_EMPTY_P (variant) = empty_p;
}
}
-
- /* Handle empty records as per the x86-64 psABI. */
- TYPE_EMPTY_P (type) = targetm.calls.empty_record_p (type);
}
/* Return a new underlying object for a bitfield started with FIELD. */
--- gcc/testsuite/g++.dg/torture/pr84502.C.jj 2018-02-21 22:13:22.279420017 +0100
+++ gcc/testsuite/g++.dg/torture/pr84502.C 2018-02-21 22:13:03.727424362 +0100
@@ -0,0 +1,20 @@
+// PR target/84502
+// { dg-do run }
+
+template<typename T>
+struct A { };
+using X = A<int>;
+
+void
+foo (X, int a1, int a2, int a3, int a4, int a5, int a6, int a7, int a8)
+{
+ if (a1 != 0 || a2 != 1 || a3 != 2 || a4 != 3
+ || a5 != 4 || a6 != 5 || a7 != 6 || a8 != 7)
+ __builtin_abort ();
+}
+
+int
+main ()
+{
+ foo (X{}, 0, 1, 2, 3, 4, 5, 6, 7);
+}
Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Fix TYPE_EMPTY_P handling - x86_64 ABI issue (PR target/84502)
2018-02-21 22:49 [PATCH] Fix TYPE_EMPTY_P handling - x86_64 ABI issue (PR target/84502) Jakub Jelinek
@ 2018-02-22 7:40 ` Richard Biener
0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2018-02-22 7:40 UTC (permalink / raw)
To: Jakub Jelinek, Jeff Law, Jason Merrill, Marek Polacek; +Cc: gcc-patches
On February 21, 2018 11:37:20 PM GMT+01:00, Jakub Jelinek <jakub@redhat.com> wrote:
>Hi!
>
>The following testcase shows that we set TYPE_EMPTY_P just one of the
>possibly many variants of an aggregate type.
>On the testcase, in one case (in the callee) we check TYPE_EMPTY_P on X
>which is
>the type variant for the using type, and in another case we check it on
>the
>A<int> type instead, which has TYPE_EMPTY_P set.
>
>Bootstrap/regtest pending on x86_64-linux and i686-linux, ok if it
>passes?
OK.
Richard.
>2018-02-21 Jakub Jelinek <jakub@redhat.com>
>
> PR target/84502
> * stor-layout.c (finalize_type_size): Propagate TYPE_EMPTY_P flag
> to all type variants.
>
> * g++.dg/torture/pr84502.C: New test.
>
>--- gcc/stor-layout.c.jj 2018-02-13 21:23:29.187981310 +0100
>+++ gcc/stor-layout.c 2018-02-21 21:43:24.783522853 +0100
>@@ -1883,6 +1883,9 @@ finalize_type_size (tree type)
> && TREE_CODE (TYPE_SIZE_UNIT (type)) != INTEGER_CST)
> TYPE_SIZE_UNIT (type) = variable_size (TYPE_SIZE_UNIT (type));
>
>+ /* Handle empty records as per the x86-64 psABI. */
>+ TYPE_EMPTY_P (type) = targetm.calls.empty_record_p (type);
>+
> /* Also layout any other variants of the type. */
> if (TYPE_NEXT_VARIANT (type)
> || type != TYPE_MAIN_VARIANT (type))
>@@ -1895,6 +1898,7 @@ finalize_type_size (tree type)
> unsigned int precision = TYPE_PRECISION (type);
> unsigned int user_align = TYPE_USER_ALIGN (type);
> machine_mode mode = TYPE_MODE (type);
>+ bool empty_p = TYPE_EMPTY_P (type);
>
> /* Copy it into all variants. */
> for (variant = TYPE_MAIN_VARIANT (type);
>@@ -1911,11 +1915,9 @@ finalize_type_size (tree type)
> SET_TYPE_ALIGN (variant, valign);
> TYPE_PRECISION (variant) = precision;
> SET_TYPE_MODE (variant, mode);
>+ TYPE_EMPTY_P (variant) = empty_p;
> }
> }
>-
>- /* Handle empty records as per the x86-64 psABI. */
>- TYPE_EMPTY_P (type) = targetm.calls.empty_record_p (type);
> }
>
>/* Return a new underlying object for a bitfield started with FIELD.
>*/
>--- gcc/testsuite/g++.dg/torture/pr84502.C.jj 2018-02-21
>22:13:22.279420017 +0100
>+++ gcc/testsuite/g++.dg/torture/pr84502.C 2018-02-21
>22:13:03.727424362 +0100
>@@ -0,0 +1,20 @@
>+// PR target/84502
>+// { dg-do run }
>+
>+template<typename T>
>+struct A { };
>+using X = A<int>;
>+
>+void
>+foo (X, int a1, int a2, int a3, int a4, int a5, int a6, int a7, int
>a8)
>+{
>+ if (a1 != 0 || a2 != 1 || a3 != 2 || a4 != 3
>+ || a5 != 4 || a6 != 5 || a7 != 6 || a8 != 7)
>+ __builtin_abort ();
>+}
>+
>+int
>+main ()
>+{
>+ foo (X{}, 0, 1, 2, 3, 4, 5, 6, 7);
>+}
>
> Jakub
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2018-02-22 7:40 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-02-21 22:49 [PATCH] Fix TYPE_EMPTY_P handling - x86_64 ABI issue (PR target/84502) Jakub Jelinek
2018-02-22 7:40 ` Richard Biener
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).