[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 --- Comment #12 from paolo at gcc dot gnu.org --- Author: paolo Date: Tue Oct 17 14:17:17 2017 New Revision: 253815 URL: https://gcc.gnu.org/viewcvs?rev=253815&root=gcc&view=rev Log: gcc/ChangeLog 2017-10-17 Qing Zhao Wilco Dijkstra * builtins.c (expand_builtin_update_setjmp_buf): Add a converstion to Pmode from the buf_addr. gcc/testsuite/ChangeLog 2017-10-17 Qing Zhao Wilco Dijkstra PR middle-end/80295 * gcc.target/aarch64/pr80295.c: New test. Modified: trunk/gcc/testsuite/ChangeLog
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 --- Comment #11 from paolo at gcc dot gnu.org --- Author: paolo Date: Tue Oct 17 14:16:35 2017 New Revision: 253814 URL: https://gcc.gnu.org/viewcvs?rev=253814&root=gcc&view=rev Log: gcc/ChangeLog 2017-10-17 Qing Zhao Wilco Dijkstra * builtins.c (expand_builtin_update_setjmp_buf): Add a converstion to Pmode from the buf_addr. gcc/testsuite/ChangeLog 2017-10-17 Qing Zhao Wilco Dijkstra PR middle-end/80295 * gcc.target/aarch64/pr80295.c: New test. Added: trunk/gcc/testsuite/gcc.target/aarch64/pr80295.c Modified: trunk/gcc/ChangeLog trunk/gcc/builtins.c
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 --- Comment #10 from Qing Zhao --- Comment on attachment 42319 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42319 proposed patch The implementation of __builtin_update_setjmp_buf is incorrect. It takes a pointer as an operand and treats the Mode of the pointer as Pmode, which is not correct. a conversion from ptr_mode to Pmode is needed.
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 --- Comment #9 from Qing Zhao --- Created attachment 42319 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42319&action=edit proposed patch
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295
--- Comment #8 from Wilco ---
(In reply to Qing Zhao from comment #7)
> However, I am still not very sure about the current implementation of
> -mabi=ilp32 on aarch64
> for example:
>
> qinzhao@gcc116:~/Bugs/80295$ cat t1.c
> void f (void *b) { __builtin_return(b); }
> qinzhao@gcc116:~/Bugs/80295$ sh t1
> /home/qinzhao/Install/latest/bin/gcc -mabi=ilp32 -fdump-rtl-all
> -fdump-tree-all t1.c
>
> in t1.c.231r.expand, I see the following stack pushing RTL insn:
>
> (insn 2 4 3 2 (set (mem/f/c:SI (plus:DI (reg/f:DI 68 virtual-stack-vars)
> (const_int -4 [0xfffc])) [1 b+0 S4 A32])
> (reg:SI 0 x0 [ b ])) "t1.c":1 -1
> (nil))
>
> the above looks like that ilp32 treats the stack pushing as 32 bit (not 64
> bits as aarch64 LP64 mode).
> is this right?
You're compiling without optimization, so the above instruction is spilling b
to the stack (not a push, just a store - there are no callee-saves for this
function). And b is SImode in ILP32, so it's correct.
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295
--- Comment #7 from Qing Zhao ---
thank you, Wilco.
> Wilco changed:
>
> What|Removed |Added
>
> CC||wilco at gcc dot gnu.org
>
> --- Comment #6 from Wilco ---
> The implementation of __builtin_update_setjmp_buf is incorrect, it takes a
> pointer (ie. SImode in ILP32) and treats the result of the add as Pmode
> (DImode). So it's missing a cast to Pmode.
Yes, this was what I thought too after more study.
However, I am still not very sure about the current implementation of
-mabi=ilp32 on aarch64
for example:
qinzhao@gcc116:~/Bugs/80295$ cat t1.c
void f (void *b) { __builtin_return(b); }
qinzhao@gcc116:~/Bugs/80295$ sh t1
/home/qinzhao/Install/latest/bin/gcc -mabi=ilp32 -fdump-rtl-all -fdump-tree-all
t1.c
in t1.c.231r.expand, I see the following stack pushing RTL insn:
(insn 2 4 3 2 (set (mem/f/c:SI (plus:DI (reg/f:DI 68 virtual-stack-vars)
(const_int -4 [0xfffc])) [1 b+0 S4 A32])
(reg:SI 0 x0 [ b ])) "t1.c":1 -1
(nil))
the above looks like that ilp32 treats the stack pushing as 32 bit (not 64 bits
as aarch64 LP64 mode).
is this right?
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 Wilco changed: What|Removed |Added CC||wilco at gcc dot gnu.org --- Comment #6 from Wilco --- The implementation of __builtin_update_setjmp_buf is incorrect, it takes a pointer (ie. SImode in ILP32) and treats the result of the add as Pmode (DImode). So it's missing a cast to Pmode.
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 Paolo Carlini changed: What|Removed |Added Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |qing.zhao at oracle dot com --- Comment #5 from Paolo Carlini --- Working on it.
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 --- Comment #4 from Qing Zhao --- working on a fix to this bug. let me know if you are working on it too.
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 Qing Zhao changed: What|Removed |Added CC||qing.zhao at oracle dot com --- Comment #3 from Qing Zhao --- confirmed with the latest upstream GCC. -mabi=ilp32 triggers the failure.
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 Richard Biener changed: What|Removed |Added Priority|P3 |P2 Target Milestone|--- |7.3
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 Richard Biener changed: What|Removed |Added Target Milestone|--- |7.3
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 --- Comment #2 from Andreas Schwab --- No, it doesn't. It only made it possible to call the builtin from the frontend.
[Bug middle-end/80295] [7/8 Regression] ICE in __builtin_update_setjmp_buf expander
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80295 Martin Liška changed: What|Removed |Added Status|UNCONFIRMED |NEW Last reconfirmed||2017-08-09 CC||marxin at gcc dot gnu.org Known to work||6.4.0 Summary|ICE in |[7/8 Regression] ICE in |__builtin_update_setjmp_buf |__builtin_update_setjmp_buf |expander|expander Ever confirmed|0 |1 Known to fail||7.1.0, 8.0 --- Comment #1 from Martin Liška --- Confirmed. Started with your commit r246720.
