Richard Sandiford via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> Jakub Jelinek via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
>> Hi!
>>
>> This is a sequel to the PR85022 changes, inline-asm can (unfortunately)
>> introduce VOIDmode MEMs and in PR85022 they have been changed so that
>> we don't pretend we know their size (as opposed to assuming they have
>> zero size).
>>
>> This time we ICE in rtx_addr_can_trap_p_1 because it assumes that
>> all memory but BLKmode has known size. The patch just treats VOIDmode
>> MEMs like BLKmode in that regard. And, the STRICT_ALIGNMENT change
>> is needed because VOIDmode has GET_MODE_SIZE of 0 and we don't want to
>> check if something is a multiple of 0.
>
> Sorry if this is reopening an old discussion, but could we instead
> canonicalise these MEMs to BLKmode when expanding the inline asm?
> I think quite a lot of RTL code assumes that MEMs can be VOIDmode,
...I meant “can't” of course
> so I'd be surprised if these two patches are enough.
>
> Thanks,
> Richard
>
>>
>> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>>
>> 2021-04-10 Jakub Jelinek <ja...@redhat.com>
>>
>> PR rtl-optimization/98601
>> * rtlanal.c (rtx_addr_can_trap_p_1): Allow in assert unknown size
>> not just for BLKmode, but also for VOIDmode. For STRICT_ALIGNMENT
>> unaligned_mems handle VOIDmode like BLKmode.
>>
>> * gcc.dg/torture/pr98601.c: New test.
>>
>> --- gcc/rtlanal.c.jj 2021-03-05 21:51:48.689185518 +0100
>> +++ gcc/rtlanal.c 2021-04-09 11:15:20.341906331 +0200
>> @@ -464,12 +464,17 @@ rtx_addr_can_trap_p_1 (const_rtx x, poly
>> machine_mode mode, bool unaligned_mems)
>> {
>> enum rtx_code code = GET_CODE (x);
>> - gcc_checking_assert (mode == BLKmode || known_size_p (size));
>> + gcc_checking_assert (mode == BLKmode
>> + || mode == VOIDmode
>> + || known_size_p (size));
>> poly_int64 const_x1;
>>
>> /* The offset must be a multiple of the mode size if we are considering
>> unaligned memory references on strict alignment machines. */
>> - if (STRICT_ALIGNMENT && unaligned_mems && mode != BLKmode)
>> + if (STRICT_ALIGNMENT
>> + && unaligned_mems
>> + && mode != BLKmode
>> + && mode != VOIDmode)
>> {
>> poly_int64 actual_offset = offset;
>>
>> --- gcc/testsuite/gcc.dg/torture/pr98601.c.jj 2021-04-09
>> 11:18:26.964817854 +0200
>> +++ gcc/testsuite/gcc.dg/torture/pr98601.c 2021-04-09 11:18:16.758932053
>> +0200
>> @@ -0,0 +1,14 @@
>> +/* PR rtl-optimization/98601 */
>> +/* { dg-do compile } */
>> +
>> +void
>> +foo (void *p)
>> +{
>> + asm ("" : "=m" (*p)); /* { dg-warning "dereferencing
>> 'void \\*' pointer" } */
>> +}
>> +
>> +void
>> +bar (void *p)
>> +{
>> + asm volatile ("" : : "m" (*p)); /* { dg-warning "dereferencing 'void
>> \\*' pointer" } */
>> +}
>>
>> Jakub
