Re: [PATCH] Fix compile-time hog in MPX boundary checking (PR target/84988).

[Richard Biener <richard.guenther@gmail.com>]

On Tue, Mar 20, 2018 at 10:44 PM, Richard Sandiford
<richard.sandiford@linaro.org> wrote:
> Jeff Law <law@redhat.com> writes:
>> On 03/20/2018 01:36 PM, Martin Liška wrote:
>>> Hi.
>>>
>>> This is a work-around to not iterate all members of array that can be huge.
>>> As MPX will be removed in GCC 9.x, I hope it's acceptable. I don't want
>>> to come
>>> up with a new param for it.
>>>
>>> Survives tests&bootstrap on x86_64-linux-gnu.
>>>
>>> Martin
>>>
>>> gcc/ChangeLog:
>>>
>>> 2018-03-20  Martin Liska  <mliska@suse.cz>
>>>
>>>     PR target/84988
>>>     * tree-chkp.c (CHKP_ARRAY_MAX_CHECK_STEPS): Define a new macro.
>>>     (chkp_find_bound_slots_1): Limit number of iterations.
>> Or just CLOSE/WONTFIX :-)
>>
>> I've got no objections here -- we want to  minimize the effort put into
>> CHKP given its going to be deprecated.
>
> The problem is that this affects normal configs, not just ones with
> MPX enabled.

Indeed.  It get's called via

#0  chkp_find_bound_slots_1 (type=0x7ffff69ee9d8, have_bound=0x2ed3868, offs=0)
    at /space/rguenther/src/svn/early-lto-debug/gcc/tree-chkp.c:1708
#1  0x0000000001379a13 in chkp_find_bound_slots (type=0x7ffff69ee9d8,
    res=0x2ed3868)
    at /space/rguenther/src/svn/early-lto-debug/gcc/tree-chkp.c:1754
#2  0x0000000001377054 in chkp_type_bounds_count (type=0x7ffff69ee9d8)
    at /space/rguenther/src/svn/early-lto-debug/gcc/tree-chkp.c:1009
#3  0x00000000016c664f in ix86_function_arg_advance (cum_v=...,
    mode=E_BLKmode, type=0x7ffff69ee9d8, named=true)
    at /space/rguenther/src/svn/early-lto-debug/gcc/config/i386/i386.c:8621
8616        {
8617          /* Track if there are outgoing arguments on stack.  */
8618          if (cum->caller)
8619            cfun->machine->outgoing_args_on_stack = true;
8620
8621          cum->bnds_in_bt = chkp_type_bounds_count (type);
8622        }
8623    }
8624
8625    /* Define where to put the arguments to a function.

but I think we know POINTER_BOUNDS_TYPE_P etc. never return
true if -fcheck-pointer-* or -mmpx is not enabled, right?  So we can
guard the above call appropriately and save some compile-time
for all of us?

Richard.

> Thanks,
> Richard