Thanks a lot for Wilco’s help on this bug. 

Yes, Aarch64 does NOT do anything wrong.  

The implementation of __builtin_update_setjmp_buf is not correct. 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 for this pointer.

bootstrapped on aarch64-unknown-linux-gnu.
testing on aarch64-unknown-linux-gnu is running.

the new patch is:

=====
gcc/ChangeLog

        * builtins.c (expand_builtin_update_setjmp_buf): Add a
        converstion to Pmode from the buf_addr.

gcc/testsuite/ChangeLog

        PR middle-end/80295
        * gcc.target/aarch64/pr80295.c: New test.

---
 gcc/builtins.c                             | 1 +
 gcc/testsuite/gcc.target/aarch64/pr80295.c | 8 ++++++++
 2 files changed, 9 insertions(+)
 create mode 100644 gcc/testsuite/gcc.target/aarch64/pr80295.c

diff --git a/gcc/builtins.c b/gcc/builtins.c
index c8a5ea6..01fb08b 100644
--- a/gcc/builtins.c
+++ b/gcc/builtins.c
@@ -1199,6 +1199,7 @@ void
 expand_builtin_update_setjmp_buf (rtx buf_addr)
 {
   machine_mode sa_mode = STACK_SAVEAREA_MODE (SAVE_NONLOCAL);
+  buf_addr = convert_memory_address (Pmode, buf_addr);
   rtx stack_save
     = gen_rtx_MEM (sa_mode,
                   memory_address
diff --git a/gcc/testsuite/gcc.target/aarch64/pr80295.c 
b/gcc/testsuite/gcc.target/aarch64/pr80295.c
new file mode 100644
index 0000000..b3866d8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/pr80295.c
@@ -0,0 +1,8 @@
+/* { dg-do compile } */
+/* { dg-options "-mabi=ilp32" } */
+
+void f (void *b) 
+{ 
+  __builtin_update_setjmp_buf (b); 
+}
+
-- 
1.9.1


> 
>> On Oct 5, 2017, at 11:50 AM, Richard Earnshaw (lists) 
>> <richard.earns...@arm.com> wrote:
>> 
>> On 25/09/17 17:35, Qing Zhao wrote:
>>> 
>>> --- a/gcc/config/aarch64/aarch64.h
>>> +++ b/gcc/config/aarch64/aarch64.h
>>> @@ -782,7 +782,7 @@ typedef struct
>>> /* Specify the machine mode that the hardware addresses have.
>>>   After generation of rtl, the compiler makes no further distinction
>>>   between pointers and any other objects of this machine mode.  */
>>> -#define Pmode              DImode
>>> +#define Pmode              (TARGET_ILP32 ? SImode : DImode)
>> 
>> This is wrong.  AArch64 has both ptr_mode and Pmode.  Pmode must always
>> be DImode as (internally) all addresses must be 64-bit.  ptr_mode
>> reflects the ABI choice of 32/64-bit language level addresses.  The
>> register SP must always be a 64-bit value, even when all the top 32 bits
>> are zero.
> 

Reply via email to