On Wed, Dec 10, 2014 at 12:03 PM, Jakub Jelinek wrote:
> On Wed, Dec 10, 2014 at 02:37:13AM +0300, Evgeny Stupachenko wrote:
>> 2014-12-10 Evgeny Stupachenko
>
> I went ahead and filed a PR, so we have something to refer to in the
> ChangeLog and name the testcases.
Thanks!
>> --- a/gcc/confi
On Wed, Dec 10, 2014 at 02:37:13AM +0300, Evgeny Stupachenko wrote:
> 2014-12-10 Evgeny Stupachenko
I went ahead and filed a PR, so we have something to refer to in the
ChangeLog and name the testcases.
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -47546,6 +47546,7 @@ exp
On Wed, Dec 10, 2014 at 12:54 AM, Jakub Jelinek wrote:
> On Wed, Dec 10, 2014 at 12:33:52AM +0300, Evgeny Stupachenko wrote:
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.target/i386/blend.c
>> @@ -0,0 +1,61 @@
>> +/* Test correctness of size 3 store groups permutation. */
>> +/* { dg-do run } */
>
On Wed, Dec 10, 2014 at 12:33:52AM +0300, Evgeny Stupachenko wrote:
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/blend.c
> @@ -0,0 +1,61 @@
> +/* Test correctness of size 3 store groups permutation. */
> +/* { dg-do run } */
> +/* { dg-options "-O3" } */
> +
> +#define N 50
> +
> +enum nu
I've added the reproducer to the patch.
is it ok?
ChangeLog:
2014-12-10 Evgeny Stupachenko
gcc/testsuite
* gcc.target/i386/blend.c: New.
gcc/
* config/i386/i386.c (expand_vec_perm_pblendv): Gen new rtx for
expand_vec_perm_1 target.
diff --git a/gcc/config/i386/i386.
I mean that there are a lot of people tracking spec2006 stability and
therefore the issue should be on track in future.
And that I can create the test case, but it would be as big as several
GCC functions.
Will work on reducing the test case.
On Tue, Dec 9, 2014 at 7:20 PM, Richard Henderson wro
On Tue, Dec 09, 2014 at 06:59:27PM +0300, Evgeny Stupachenko wrote:
> I've tried to get smaller reproducer, however currently it is
> complicated as several functions in GCC.
Just add a gcc_assert where you are changing the code, testing for
what you want to avoid.
Then just delta reduce it or cre
On 12/09/2014 07:59 AM, Evgeny Stupachenko wrote:
> However patch is fixing spec2006 benchmark. Shouldn't that be enough
> for regression testing?
>
No. Spec is not free.
r~
I've tried to get smaller reproducer, however currently it is
complicated as several functions in GCC.
However patch is fixing spec2006 benchmark. Shouldn't that be enough
for regression testing?
Thanks,
Evgeny
On Tue, Dec 9, 2014 at 4:29 PM, Evgeny Stupachenko wrote:
> The case comes from spec2
The case comes from spec2006 403.gcc (or old GCC itself).
for (i = 0; i < FIRST_PSEUDO_REGISTER; ++i)
{
vd->e[i].mode = VOIDmode;
vd->e[i].oldest_regno = i;
vd->e[i].next_regno = INVALID_REGNUM;
}
It is vectorized and only then completely peeled.
Only after peeling all
On Tue, Dec 9, 2014 at 9:57 AM, Uros Bizjak wrote:
>> The patch fix pblendv expand.
>> The bug was uncovered when permutation operands are constants.
>> In this case we init target register for expand_vec_perm_1 with
>> constant and then rewrite the target with constant for
>> expand_vec_perm_pbl
On Tue, Dec 9, 2014 at 12:33 AM, Evgeny Stupachenko wrote:
> The patch fix pblendv expand.
> The bug was uncovered when permutation operands are constants.
> In this case we init target register for expand_vec_perm_1 with
> constant and then rewrite the target with constant for
> expand_vec_perm_
Hi,
The patch fix pblendv expand.
The bug was uncovered when permutation operands are constants.
In this case we init target register for expand_vec_perm_1 with
constant and then rewrite the target with constant for
expand_vec_perm_pblend.
The patch fixes 403.gcc execution, compiled with -Ofast -
13 matches
Mail list logo
