testsuite: Adjust possibly fragile slp-perm-9.c [PR104015]

  Hi,

As discussed in PR104015, the test case slp-perm-9.c can be
fragile when vectorizer tries to use different vectorisation
strategies.

As Richard suggested, this patch tries to make the check not
sensitive on the re-trying times by removing the times checking.
To still retain the test coverage on unnecessary re-trying, for
example this exposed PR104015 on Power9, I added two test cases
to powerpc test bucket.

Tested on x86_64-redhat-linux, aarch64-linux-gnu and
powerpc64-linux-gnu Power8 and powerpc64le-linux-gnu
Power9/Power10.

Is it ok for trunk?

BR,
Kewen
-----
gcc/testsuite/ChangeLog:

	PR tree-optimization/104015
	* gcc.dg/vect/slp-perm-9.c: Adjust.
	* gcc.target/powerpc/pr104015-1.c: New test.
	* gcc.target/powerpc/pr104015-2.c: New test.
---
 gcc/testsuite/gcc.dg/vect/slp-perm-9.c        |  4 +--
 gcc/testsuite/gcc.target/powerpc/pr104015-1.c | 28 +++++++++++++++++++
 gcc/testsuite/gcc.target/powerpc/pr104015-2.c | 28 +++++++++++++++++++
 3 files changed, 57 insertions(+), 3 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/powerpc/pr104015-1.c
 create mode 100644 gcc/testsuite/gcc.target/powerpc/pr104015-2.c
  

on 2022/1/18 上午11:06, Kewen.Lin via Gcc-patches wrote:
> Hi,
> 
> As discussed in PR104015, the test case slp-perm-9.c can be
> fragile when vectorizer tries to use different vectorisation
> strategies.
> 
> As Richard suggested, this patch tries to make the check not
> sensitive on the re-trying times by removing the times checking.
> To still retain the test coverage on unnecessary re-trying, for
> example this exposed PR104015 on Power9, I added two test cases
> to powerpc test bucket.
> 
> Tested on x86_64-redhat-linux, aarch64-linux-gnu and
> powerpc64-linux-gnu Power8 and powerpc64le-linux-gnu
> Power9/Power10.
> 
> Is it ok for trunk?
> 
> BR,
> Kewen
> -----
> gcc/testsuite/ChangeLog:
> 
> 	PR tree-optimization/104015
> 	* gcc.dg/vect/slp-perm-9.c: Adjust.
> 	* gcc.target/powerpc/pr104015-1.c: New test.
> 	* gcc.target/powerpc/pr104015-2.c: New test.

One updated version is attached to modify pr104015-2.c slightly by
using more clear required effective target lp64.

Tested as before.

BR,
Kewen
gcc/testsuite/gcc.dg/vect/slp-perm-9.c        |  4 +--
 gcc/testsuite/gcc.target/powerpc/pr104015-1.c | 28 ++++++++++++++++++
 gcc/testsuite/gcc.target/powerpc/pr104015-2.c | 29 +++++++++++++++++++
 3 files changed, 58 insertions(+), 3 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/powerpc/pr104015-1.c
 create mode 100644 gcc/testsuite/gcc.target/powerpc/pr104015-2.c

diff --git a/gcc/testsuite/gcc.dg/vect/slp-perm-9.c b/gcc/testsuite/gcc.dg/vect/slp-perm-9.c
index 873eddf223e..154c00af598 100644
--- a/gcc/testsuite/gcc.dg/vect/slp-perm-9.c
+++ b/gcc/testsuite/gcc.dg/vect/slp-perm-9.c
@@ -61,9 +61,7 @@ int main (int argc, const char* argv[])
 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target { { vect_perm_short || vect32 } || vect_load_lanes } } } } */
 /* We don't try permutes with a group size of 3 for variable-length
    vectors.  */
-/* { dg-final { scan-tree-dump-times "permutation requires at least three vectors" 1 "vect" { target { vect_perm_short && { { ! vect_perm3_short } && { ! vect_partial_vectors_usage_1 } } } xfail vect_variable_length } } } */
-/* Try to vectorize the epilogue using partial vectors.  */
-/* { dg-final { scan-tree-dump-times "permutation requires at least three vectors" 2 "vect" { target { vect_perm_short && { { ! vect_perm3_short } && vect_partial_vectors_usage_1 } } xfail vect_variable_length } } } */
+/* { dg-final { scan-tree-dump "permutation requires at least three vectors" "vect" { target { vect_perm_short && { ! vect_perm3_short } } xfail vect_variable_length } } } */
 /* { dg-final { scan-tree-dump-not "permutation requires at least three vectors" "vect" { target vect_perm3_short } } } */
 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 0 "vect" { target { { ! { vect_perm3_short || vect32 } } || vect_load_lanes } } } } */
 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { target { { vect_perm3_short || vect32 } && { ! vect_load_lanes } } } } } */
diff --git a/gcc/testsuite/gcc.target/powerpc/pr104015-1.c b/gcc/testsuite/gcc.target/powerpc/pr104015-1.c
new file mode 100644
index 00000000000..895c243aaf8
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/pr104015-1.c
@@ -0,0 +1,28 @@
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mdejagnu-cpu=power9 -O2 -ftree-vectorize -fno-vect-cost-model -fdump-tree-vect-details" } */
+
+/* As PR104015, we don't expect vectorizer will re-try some vector modes
+   for epilogues on Power9, since Power9 doesn't support partial vector
+   by defaut.  */
+
+#include <stdarg.h>
+#define N 200
+
+void __attribute__((noinline))
+foo (unsigned short *__restrict__ pInput, unsigned short *__restrict__ pOutput)
+{
+  unsigned short i, a, b, c;
+
+  for (i = 0; i < N / 3; i++)
+    {
+       a = *pInput++;
+       b = *pInput++;
+       c = *pInput++;
+
+       *pOutput++ = a + b + c + 3;
+       *pOutput++ = a + b + c + 12;
+       *pOutput++ = a + b + c + 1;
+    }
+}
+
+/* { dg-final { scan-tree-dump-not "Re-trying epilogue analysis with vector mode" "vect" } } */
diff --git a/gcc/testsuite/gcc.target/powerpc/pr104015-2.c b/gcc/testsuite/gcc.target/powerpc/pr104015-2.c
new file mode 100644
index 00000000000..ab482b11629
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/pr104015-2.c
@@ -0,0 +1,29 @@
+/* { dg-require-effective-target power10_ok } */
+/* Vector with length instructions lxvl/stxvl are only enabled for 64 bit.  */
+/* { dg-require-effective-target lp64 } */
+/* { dg-options "-mdejagnu-cpu=power10 -O2 -ftree-vectorize -fno-vect-cost-model -fdump-tree-vect-details" } */
+
+/* Power10 support partial vector for epilogue by default, it's expected
+   vectorizer would re-try for it once.  */
+
+#include <stdarg.h>
+#define N 200
+
+void __attribute__((noinline))
+foo (unsigned short *__restrict__ pInput, unsigned short *__restrict__ pOutput)
+{
+  unsigned short i, a, b, c;
+
+  for (i = 0; i < N / 3; i++)
+    {
+       a = *pInput++;
+       b = *pInput++;
+       c = *pInput++;
+
+       *pOutput++ = a + b + c + 3;
+       *pOutput++ = a + b + c + 12;
+       *pOutput++ = a + b + c + 1;
+    }
+}
+
+/* { dg-final { scan-tree-dump-times "Re-trying epilogue analysis with vector mode" 1 "vect" } } */
  

"Kewen.Lin" <linkw@linux.ibm.com> writes:
> on 2022/1/18 锟斤拷锟斤拷11:06, Kewen.Lin via Gcc-patches wrote:
>> Hi,
>> 
>> As discussed in PR104015, the test case slp-perm-9.c can be
>> fragile when vectorizer tries to use different vectorisation
>> strategies.
>> 
>> As Richard suggested, this patch tries to make the check not
>> sensitive on the re-trying times by removing the times checking.
>> To still retain the test coverage on unnecessary re-trying, for
>> example this exposed PR104015 on Power9, I added two test cases
>> to powerpc test bucket.
>> 
>> Tested on x86_64-redhat-linux, aarch64-linux-gnu and
>> powerpc64-linux-gnu Power8 and powerpc64le-linux-gnu
>> Power9/Power10.
>> 
>> Is it ok for trunk?
>> 
>> BR,
>> Kewen
>> -----
>> gcc/testsuite/ChangeLog:
>> 
>> 	PR tree-optimization/104015
>> 	* gcc.dg/vect/slp-perm-9.c: Adjust.
>> 	* gcc.target/powerpc/pr104015-1.c: New test.
>> 	* gcc.target/powerpc/pr104015-2.c: New test.
>
> One updated version is attached to modify pr104015-2.c slightly by
> using more clear required effective target lp64.
>
> Tested as before.
>
> BR,
> Kewen

OK for the target-independent part, thanks.  IMO it's OK independently
of the rs6000 tests.

Richard

>  gcc/testsuite/gcc.dg/vect/slp-perm-9.c        |  4 +--
>  gcc/testsuite/gcc.target/powerpc/pr104015-1.c | 28 ++++++++++++++++++
>  gcc/testsuite/gcc.target/powerpc/pr104015-2.c | 29 +++++++++++++++++++
>  3 files changed, 58 insertions(+), 3 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.target/powerpc/pr104015-1.c
>  create mode 100644 gcc/testsuite/gcc.target/powerpc/pr104015-2.c
>
> diff --git a/gcc/testsuite/gcc.dg/vect/slp-perm-9.c b/gcc/testsuite/gcc.dg/vect/slp-perm-9.c
> index 873eddf223e..154c00af598 100644
> --- a/gcc/testsuite/gcc.dg/vect/slp-perm-9.c
> +++ b/gcc/testsuite/gcc.dg/vect/slp-perm-9.c
> @@ -61,9 +61,7 @@ int main (int argc, const char* argv[])
>  /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target { { vect_perm_short || vect32 } || vect_load_lanes } } } } */
>  /* We don't try permutes with a group size of 3 for variable-length
>     vectors.  */
> -/* { dg-final { scan-tree-dump-times "permutation requires at least three vectors" 1 "vect" { target { vect_perm_short && { { ! vect_perm3_short } && { ! vect_partial_vectors_usage_1 } } } xfail vect_variable_length } } } */
> -/* Try to vectorize the epilogue using partial vectors.  */
> -/* { dg-final { scan-tree-dump-times "permutation requires at least three vectors" 2 "vect" { target { vect_perm_short && { { ! vect_perm3_short } && vect_partial_vectors_usage_1 } } xfail vect_variable_length } } } */
> +/* { dg-final { scan-tree-dump "permutation requires at least three vectors" "vect" { target { vect_perm_short && { ! vect_perm3_short } } xfail vect_variable_length } } } */
>  /* { dg-final { scan-tree-dump-not "permutation requires at least three vectors" "vect" { target vect_perm3_short } } } */
>  /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 0 "vect" { target { { ! { vect_perm3_short || vect32 } } || vect_load_lanes } } } } */
>  /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { target { { vect_perm3_short || vect32 } && { ! vect_load_lanes } } } } } */
> diff --git a/gcc/testsuite/gcc.target/powerpc/pr104015-1.c b/gcc/testsuite/gcc.target/powerpc/pr104015-1.c
> new file mode 100644
> index 00000000000..895c243aaf8
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/powerpc/pr104015-1.c
> @@ -0,0 +1,28 @@
> +/* { dg-require-effective-target powerpc_p9vector_ok } */
> +/* { dg-options "-mdejagnu-cpu=power9 -O2 -ftree-vectorize -fno-vect-cost-model -fdump-tree-vect-details" } */
> +
> +/* As PR104015, we don't expect vectorizer will re-try some vector modes
> +   for epilogues on Power9, since Power9 doesn't support partial vector
> +   by defaut.  */
> +
> +#include <stdarg.h>
> +#define N 200
> +
> +void __attribute__((noinline))
> +foo (unsigned short *__restrict__ pInput, unsigned short *__restrict__ pOutput)
> +{
> +  unsigned short i, a, b, c;
> +
> +  for (i = 0; i < N / 3; i++)
> +    {
> +       a = *pInput++;
> +       b = *pInput++;
> +       c = *pInput++;
> +
> +       *pOutput++ = a + b + c + 3;
> +       *pOutput++ = a + b + c + 12;
> +       *pOutput++ = a + b + c + 1;
> +    }
> +}
> +
> +/* { dg-final { scan-tree-dump-not "Re-trying epilogue analysis with vector mode" "vect" } } */
> diff --git a/gcc/testsuite/gcc.target/powerpc/pr104015-2.c b/gcc/testsuite/gcc.target/powerpc/pr104015-2.c
> new file mode 100644
> index 00000000000..ab482b11629
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/powerpc/pr104015-2.c
> @@ -0,0 +1,29 @@
> +/* { dg-require-effective-target power10_ok } */
> +/* Vector with length instructions lxvl/stxvl are only enabled for 64 bit.  */
> +/* { dg-require-effective-target lp64 } */
> +/* { dg-options "-mdejagnu-cpu=power10 -O2 -ftree-vectorize -fno-vect-cost-model -fdump-tree-vect-details" } */
> +
> +/* Power10 support partial vector for epilogue by default, it's expected
> +   vectorizer would re-try for it once.  */
> +
> +#include <stdarg.h>
> +#define N 200
> +
> +void __attribute__((noinline))
> +foo (unsigned short *__restrict__ pInput, unsigned short *__restrict__ pOutput)
> +{
> +  unsigned short i, a, b, c;
> +
> +  for (i = 0; i < N / 3; i++)
> +    {
> +       a = *pInput++;
> +       b = *pInput++;
> +       c = *pInput++;
> +
> +       *pOutput++ = a + b + c + 3;
> +       *pOutput++ = a + b + c + 12;
> +       *pOutput++ = a + b + c + 1;
> +    }
> +}
> +
> +/* { dg-final { scan-tree-dump-times "Re-trying epilogue analysis with vector mode" 1 "vect" } } */
  

On Tue, Jan 18, 2022 at 11:57:32AM +0000, Richard Sandiford wrote:
> "Kewen.Lin" <linkw@linux.ibm.com> writes:
> >> 	PR tree-optimization/104015
> >> 	* gcc.dg/vect/slp-perm-9.c: Adjust.
> >> 	* gcc.target/powerpc/pr104015-1.c: New test.
> >> 	* gcc.target/powerpc/pr104015-2.c: New test.

> OK for the target-independent part, thanks.  IMO it's OK independently
> of the rs6000 tests.

The rs6000 parts are fine as well.  Thanks!

I see you got rid of the ilp32 tests, I was going to holler about that,
there is no reason this should only work (or only be tested) on 64-bit
systems :-)


Segher
  

on 2022/1/19 上午5:34, Segher Boessenkool wrote:
> On Tue, Jan 18, 2022 at 11:57:32AM +0000, Richard Sandiford wrote:
>> "Kewen.Lin" <linkw@linux.ibm.com> writes:
>>>> 	PR tree-optimization/104015
>>>> 	* gcc.dg/vect/slp-perm-9.c: Adjust.
>>>> 	* gcc.target/powerpc/pr104015-1.c: New test.
>>>> 	* gcc.target/powerpc/pr104015-2.c: New test.
> 
>> OK for the target-independent part, thanks.  IMO it's OK independently
>> of the rs6000 tests.
> 
> The rs6000 parts are fine as well.  Thanks!
> 
> I see you got rid of the ilp32 tests, I was going to holler about that,
> there is no reason this should only work (or only be tested) on 64-bit
> systems :-)
> 

Thanks Richard and Segher, committed as r12-6717.

BR,
Kewen