[PATCH testsuite]XFAIL gcc.dg/tree-ssa/pr71347 on some targets

[PATCH testsuite]XFAIL gcc.dg/tree-ssa/pr71347 on some targets

[Bin Cheng]

[-- Attachment #1: Type: text/plain, Size: 753 bytes --]

Hi,
Test gcc.dg/tree-ssa/pr71347 failed on some targets if the two memory references are re-written into different forms by IVOPT.  This could be because of various reasons, for example, auto-increment addressing mode.  Since the address expressions are of different form, DOM fails to eliminate the redundant load at the moment.  I think DOM should be improved to handle address expressions appearing in different forms (at least for simple cases).  For now, I will XFAIL this test indicating a possible enhancement.

Test checked on affected targets.  Is it OK?

Thanks,
bin

gcc/testsuite/ChangeLog
2016-06-20  Bin Cheng  <bin.cheng@arm.com>

	PR tree-optimization/71347
	* gcc.dg/tree-ssa/pr71347.c: XFAIL on ia64, arm, m68k and sparc.

[-- Attachment #2: xfail-pr71347.txt --]
[-- Type: text/plain, Size: 525 bytes --]

diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr71347.c b/gcc/testsuite/gcc.dg/tree-ssa/pr71347.c
index c8f87a9..428e41b 100644
--- a/gcc/testsuite/gcc.dg/tree-ssa/pr71347.c
+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr71347.c
@@ -14,4 +14,4 @@ void foo (void)
 }
 
 /* Load of X[i - i] can be omitted by reusing X[i] in previous iteration.  */
-/* { dg-final { scan-tree-dump-not ".* = MEM.*;" "optimized"} } */
+/* { dg-final { scan-tree-dump-not ".* = MEM.*;" "optimized" { xfail { ia64-*-* arm*-*-* m68k*-*-* sparc*-*-* } } } } */

Re: [PATCH testsuite]XFAIL gcc.dg/tree-ssa/pr71347 on some targets

[Jeff Law]

On 07/14/2016 10:11 AM, Bin Cheng wrote:
> Hi, Test gcc.dg/tree-ssa/pr71347 failed on some targets if the two
> memory references are re-written into different forms by IVOPT.  This
> could be because of various reasons, for example, auto-increment
> addressing mode.  Since the address expressions are of different
> form, DOM fails to eliminate the redundant load at the moment.  I
> think DOM should be improved to handle address expressions appearing
> in different forms (at least for simple cases).  For now, I will
> XFAIL this test indicating a possible enhancement.
FWIW, Adam Lawrence did some work in this area a few months ago. 
Essentially enhancing DOM to be able to detect certain MEM_REF and 
ARRAY_REF accesses were in fact equivalent.


>
> Test checked on affected targets.  Is it OK?
Yes.  This is fine.
jeff

Re: [PATCH testsuite]XFAIL gcc.dg/tree-ssa/pr71347 on some targets

[Bin.Cheng]

On Thu, Jul 14, 2016 at 6:14 PM, Jeff Law <law@redhat.com> wrote:
> On 07/14/2016 10:11 AM, Bin Cheng wrote:
>>
>> Hi, Test gcc.dg/tree-ssa/pr71347 failed on some targets if the two
>> memory references are re-written into different forms by IVOPT.  This
>> could be because of various reasons, for example, auto-increment
>> addressing mode.  Since the address expressions are of different
>> form, DOM fails to eliminate the redundant load at the moment.  I
>> think DOM should be improved to handle address expressions appearing
>> in different forms (at least for simple cases).  For now, I will
>> XFAIL this test indicating a possible enhancement.
>
> FWIW, Adam Lawrence did some work in this area a few months ago. Essentially
> enhancing DOM to be able to detect certain MEM_REF and ARRAY_REF accesses
> were in fact equivalent.
Hi Jeff,
Do you recall any PR/patch of this?  I didn't find it in mail searching.

>
>
>>
>> Test checked on affected targets.  Is it OK?
Applied.

Thanks,
bin
>
> Yes.  This is fine.
> jeff
>

Re: [PATCH testsuite]XFAIL gcc.dg/tree-ssa/pr71347 on some targets

[Jeff Law]

On 07/15/2016 08:02 AM, Bin.Cheng wrote:
> On Thu, Jul 14, 2016 at 6:14 PM, Jeff Law <law@redhat.com> wrote:
>> On 07/14/2016 10:11 AM, Bin Cheng wrote:
>>>
>>> Hi, Test gcc.dg/tree-ssa/pr71347 failed on some targets if the two
>>> memory references are re-written into different forms by IVOPT.  This
>>> could be because of various reasons, for example, auto-increment
>>> addressing mode.  Since the address expressions are of different
>>> form, DOM fails to eliminate the redundant load at the moment.  I
>>> think DOM should be improved to handle address expressions appearing
>>> in different forms (at least for simple cases).  For now, I will
>>> XFAIL this test indicating a possible enhancement.
>>
>> FWIW, Adam Lawrence did some work in this area a few months ago. Essentially
>> enhancing DOM to be able to detect certain MEM_REF and ARRAY_REF accesses
>> were in fact equivalent.
> Hi Jeff,
> Do you recall any PR/patch of this?  I didn't find it in mail searching.
The commit references 63679.

Jeff