Re: PATCH: PR target/52146: [x32] - Wrong code to access addresses 0x80000000 to 0xFFFFFFFF

[Uros Bizjak <ubizjak at gmail dot com>]

On Fri, Feb 10, 2012 at 6:42 PM, H.J. Lu <hongjiu.lu@intel.com> wrote:

>> Since constant address in x32 is signed extended to 64bit, negative
>> displacement without base nor index is out of range. ?OK for trunk?
>>
>
> Here is a different patch.
>
> H.J.
> ---
> gcc/
>
> 2012-02-10 ?Uros Bizjak ?<ubizjak@gmail.com>
>
> ? ? ? ?PR target/52146
> ? ? ? ?* config/i386/i386.c (ix86_legitimate_address_p): Disallow
> ? ? ? ?negative constant address for x32.
>
> gcc/testsuite/
>
> 2012-02-10 ?H.J. Lu ?<hongjiu.lu@intel.com>
>
> ? ? ? ?PR target/52146
> ? ? ? ?* gcc.target/i386/pr52146.c: New.
>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index 009dd53..8f4e72e 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -11932,6 +11932,13 @@ ix86_legitimate_address_p (enum machine_mode mode ATTRIBUTE_UNUSED,
> ? rtx base, index, disp;
> ? HOST_WIDE_INT scale;
>
> + ?/* Since constant address in x32 is signed extended to 64bit,
> + ? ? we have to prevent addresses from 0x80000000 to 0xffffffff. ?*/
> + ?if (TARGET_X32
> + ? ? ?&& CONST_INT_P (addr)
> + ? ? ?&& val_signbit_known_set_p (SImode, INTVAL (addr)))

As said in the PR, val_signbit_known_set_p is a bit overkill. Please
use INTVAL (addr) < 0, it works as well.

> +++ b/gcc/testsuite/gcc.target/i386/pr52146.c
> @@ -0,0 +1,17 @@
> +/* { dg-do compile { target { { i?86-*-linux* x86_64-*-linux* } && { ! { ia32 } } } } } */

we _are_ in x86 directory, so:

/* { dg-do compile { target { ! { ia32 } } } }  */

> +/* { dg-options "-O2 -mx32" } */
> +
> +void test1() {
> + ?int* apic_tpr_addr = (int *)0xfee00080;
> + ?*apic_tpr_addr += 4;
> +}
> +void test2() {
> + ?volatile int* apic_tpr_addr = (int *)0xfee00080;
> + ?*apic_tpr_addr = 0;

No need for volatile.

> +}
> +void test3() {
> + ?volatile int* apic_tpr_addr = (int *)0x7fffffff;
> + ?*apic_tpr_addr = 0;
> +}

test2 is enough. No need to test what worked OK.

> +
> +/* { dg-final { scan-assembler-not "-18874240" } } */

Please also reformat the test to GNU coding standards.

Patch is OK for 4.7 and 4.6 after bootstrap and regression test on x32 target.

Thanks,
Uros.