Re: [PATCH] xfail and improve some failing libgomp tests
On 10/22/20 3:19 PM, Jakub Jelinek wrote:
> On Tue, Oct 06, 2020 at 05:45:31PM +0200, Tom de Vries wrote:
>> I've updated the patch accordingly.
>>
>> FWIW, I now run into an ICE which looks like PR96680:
>
> With the patch I've posted today to fix up declare variant LTO handling,
> Tobias reported the patch still doesn't work, and there are two
> reasons for that.
> One is that when the base function is marked implicitly as declare target,
> we don't mark also implicitly the variants. I'll need to ask on omp-lang
> about details for that, but generally the compiler should do it some way.
> The other one is that the way base_delay is written, it will always
> call the usleep function, which is undesirable for nvptx. While the
> compiler will replace all direct calls to base_delay to nvptx_delay,
> the base_delay definition which calls usleep stays.
>
> The following should work instead (I've tested it without offloading and
> Tobias with offloading):
>
I've tested this patch in combination with:
- "[PATCH] lto: LTO cgraph support for late declare variant resolution"
https://gcc.gnu.org/pipermail/gcc-patches/2020-October/556793.html
- "[omp, simt] Handle alternative IV"
https://gcc.gnu.org/pipermail/gcc-patches/2020-October/555352.html
on top of commit c26d7df1031 "OpenMP: Fortran - support omp flush's
memorder clauses".
The only FAILs I see are for PR97532 (
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97532 ), 10 in total.
So, LGTM.
Thanks,
- Tom
> 2020-10-22 Jakub Jelinek
> Tom de Vries
>
> PR testsuite/81690
> * testsuite/libgomp.c/usleep.h: New file.
> * testsuite/libgomp.c/target-32.c: Include usleep.h.
> (main): Use tgt_usleep instead of usleep.
> * testsuite/libgomp.c/thread-limit-2.c: Include usleep.h.
> (main): Use tgt_usleep instead of usleep.
>
> --- gcc/libgomp/testsuite/libgomp.c/usleep.h.jj 2020-10-22
> 14:45:14.034196695 +0200
> +++ gcc/libgomp/testsuite/libgomp.c/usleep.h 2020-10-22 14:48:05.186719495
> +0200
> @@ -0,0 +1,24 @@
> +#include
> +
> +int
> +nvptx_usleep (useconds_t d)
> +{
> + /* This function serves as a replacement for usleep in
> + this test case. It does not even attempt to be functionally
> + equivalent - we just want some sort of delay. */
> + int i;
> + int N = d * 2000;
> + for (i = 0; i < N; i++)
> +asm volatile ("" : : : "memory");
> + return 0;
> +}
> +
> +#pragma omp declare variant (nvptx_usleep)
> match(construct={target},device={arch(nvptx)})
> +#pragma omp declare variant (usleep) match(user={condition(1)})
> +int
> +tgt_usleep (useconds_t d)
> +{
> + return 0;
> +}
> +
> +#pragma omp declare target to (nvptx_usleep, tgt_usleep)
> --- gcc/libgomp/testsuite/libgomp.c/target-32.c.jj2020-01-12
> 11:54:39.037373820 +0100
> +++ gcc/libgomp/testsuite/libgomp.c/target-32.c 2020-10-22
> 14:46:23.211195456 +0200
> @@ -1,5 +1,6 @@
> #include
> #include
> +#include "usleep.h"
>
> int main ()
> {
> @@ -18,28 +19,28 @@ int main ()
>
> #pragma omp target nowait map(alloc: b) depend(in: d[2]) depend(out:
> d[3])
> {
> - usleep (1000);
> + tgt_usleep (1000);
>#pragma omp atomic update
>b |= 4;
> }
>
> #pragma omp target nowait map(alloc: b) depend(in: d[2]) depend(out:
> d[4])
> {
> - usleep (5000);
> + tgt_usleep (5000);
>#pragma omp atomic update
>b |= 1;
> }
>
> #pragma omp target nowait map(alloc: c) depend(in: d[3], d[4])
> depend(out: d[5])
> {
> - usleep (5000);
> + tgt_usleep (5000);
>#pragma omp atomic update
>c |= 8;
> }
>
> #pragma omp target nowait map(alloc: c) depend(in: d[3], d[4])
> depend(out: d[6])
> {
> - usleep (1000);
> + tgt_usleep (1000);
>#pragma omp atomic update
>c |= 2;
> }
> --- gcc/libgomp/testsuite/libgomp.c/thread-limit-2.c.jj 2020-01-12
> 11:54:39.037373820 +0100
> +++ gcc/libgomp/testsuite/libgomp.c/thread-limit-2.c 2020-10-22
> 14:57:31.957516284 +0200
> @@ -4,6 +4,7 @@
> #include
> #include
> #include
> +#include "usleep.h"
>
> int
> main ()
> @@ -48,7 +49,7 @@ main ()
> v = ++cnt;
> if (v > 6)
> abort ();
> - usleep (1);
> + tgt_usleep (1);
> #pragma omp atomic
> --cnt;
> }
>
>
> Jakub
>
Re: [PATCH] xfail and improve some failing libgomp tests
On Tue, Oct 06, 2020 at 05:45:31PM +0200, Tom de Vries wrote:
> I've updated the patch accordingly.
>
> FWIW, I now run into an ICE which looks like PR96680:
With the patch I've posted today to fix up declare variant LTO handling,
Tobias reported the patch still doesn't work, and there are two
reasons for that.
One is that when the base function is marked implicitly as declare target,
we don't mark also implicitly the variants. I'll need to ask on omp-lang
about details for that, but generally the compiler should do it some way.
The other one is that the way base_delay is written, it will always
call the usleep function, which is undesirable for nvptx. While the
compiler will replace all direct calls to base_delay to nvptx_delay,
the base_delay definition which calls usleep stays.
The following should work instead (I've tested it without offloading and
Tobias with offloading):
2020-10-22 Jakub Jelinek
Tom de Vries
PR testsuite/81690
* testsuite/libgomp.c/usleep.h: New file.
* testsuite/libgomp.c/target-32.c: Include usleep.h.
(main): Use tgt_usleep instead of usleep.
* testsuite/libgomp.c/thread-limit-2.c: Include usleep.h.
(main): Use tgt_usleep instead of usleep.
--- gcc/libgomp/testsuite/libgomp.c/usleep.h.jj 2020-10-22 14:45:14.034196695
+0200
+++ gcc/libgomp/testsuite/libgomp.c/usleep.h2020-10-22 14:48:05.186719495
+0200
@@ -0,0 +1,24 @@
+#include
+
+int
+nvptx_usleep (useconds_t d)
+{
+ /* This function serves as a replacement for usleep in
+ this test case. It does not even attempt to be functionally
+ equivalent - we just want some sort of delay. */
+ int i;
+ int N = d * 2000;
+ for (i = 0; i < N; i++)
+asm volatile ("" : : : "memory");
+ return 0;
+}
+
+#pragma omp declare variant (nvptx_usleep)
match(construct={target},device={arch(nvptx)})
+#pragma omp declare variant (usleep) match(user={condition(1)})
+int
+tgt_usleep (useconds_t d)
+{
+ return 0;
+}
+
+#pragma omp declare target to (nvptx_usleep, tgt_usleep)
--- gcc/libgomp/testsuite/libgomp.c/target-32.c.jj 2020-01-12
11:54:39.037373820 +0100
+++ gcc/libgomp/testsuite/libgomp.c/target-32.c 2020-10-22 14:46:23.211195456
+0200
@@ -1,5 +1,6 @@
#include
#include
+#include "usleep.h"
int main ()
{
@@ -18,28 +19,28 @@ int main ()
#pragma omp target nowait map(alloc: b) depend(in: d[2]) depend(out: d[3])
{
- usleep (1000);
+ tgt_usleep (1000);
#pragma omp atomic update
b |= 4;
}
#pragma omp target nowait map(alloc: b) depend(in: d[2]) depend(out: d[4])
{
- usleep (5000);
+ tgt_usleep (5000);
#pragma omp atomic update
b |= 1;
}
#pragma omp target nowait map(alloc: c) depend(in: d[3], d[4]) depend(out:
d[5])
{
- usleep (5000);
+ tgt_usleep (5000);
#pragma omp atomic update
c |= 8;
}
#pragma omp target nowait map(alloc: c) depend(in: d[3], d[4]) depend(out:
d[6])
{
- usleep (1000);
+ tgt_usleep (1000);
#pragma omp atomic update
c |= 2;
}
--- gcc/libgomp/testsuite/libgomp.c/thread-limit-2.c.jj 2020-01-12
11:54:39.037373820 +0100
+++ gcc/libgomp/testsuite/libgomp.c/thread-limit-2.c2020-10-22
14:57:31.957516284 +0200
@@ -4,6 +4,7 @@
#include
#include
#include
+#include "usleep.h"
int
main ()
@@ -48,7 +49,7 @@ main ()
v = ++cnt;
if (v > 6)
abort ();
- usleep (1);
+ tgt_usleep (1);
#pragma omp atomic
--cnt;
}
Jakub
Re: [PATCH] xfail and improve some failing libgomp tests
On 10/6/20 5:02 PM, Jakub Jelinek wrote:
> On Tue, Oct 06, 2020 at 04:48:40PM +0200, Tom de Vries wrote:
>> On 10/5/20 3:15 PM, Tom de Vries wrote:
>>> On 2/7/20 4:29 PM, Jakub Jelinek wrote:
On Fri, Feb 07, 2020 at 09:56:38AM +0100, Harwath, Frederik wrote:
> * {target-32.c, thread-limit-2.c}:
> no "usleep" implemented for nvptx. Cf. https://gcc.gnu.org/PR81690
Please don't, I want to deal with that using declare variant, just didn't
get yet around to finishing the last patch needed for that. Will try next
week.
>>>
>>> Hi Jakub,
>>>
>>> Ping, any update on this?
>
> Not finished the last step, I run into LTO issues. Will need to return to
> that soon.
> Last progress in "[RFH] LTO cgraph support for late declare variant
> resolution"
> mail from May on gcc-patches.
>
Ack, thanks for the update.
>> --- a/libgomp/testsuite/libgomp.c/target-32.c
>> +++ b/libgomp/testsuite/libgomp.c/target-32.c
>> @@ -1,6 +1,26 @@
>> #include
>> #include
>>
>> +extern void base_delay(int);
>
> No need to declare this one early.
>
>> +extern void nvptx_delay(int);
>
> Space before (, and the definition could go here instead of
> the declaration.
>
>> +#pragma omp declare variant( nvptx_delay ) match( construct={target},
>> implementation={vendor(nvidia)} )
>
> This isn't the right declare variant for what we want though,
> we only provide gnu as accepted vendor, it is implementation's vendor,
> not vendor of one of the hw components.
> So, it ought to be instead
> #pragma omp declare variant (nvptx_delay)
> match(construct={target},device={arch(nvptx)})
>
>> +void base_delay(int d)
>> +{
>> + usleep (d);
>> +}
I've updated the patch accordingly.
FWIW, I now run into an ICE which looks like PR96680:
...
lto1: internal compiler error: in lto_fixup_prevailing_decls, at
lto/lto-common.c:2595^M
0x93afcd lto_fixup_prevailing_decls^M
/home/vries/oacc/trunk/source-gcc/gcc/lto/lto-common.c:2595^M
0x93b1d6 lto_fixup_decls^M
/home/vries/oacc/trunk/source-gcc/gcc/lto/lto-common.c:2645^M
0x93bcc4 read_cgraph_and_symbols(unsigned int, char const**)^M
/home/vries/oacc/trunk/source-gcc/gcc/lto/lto-common.c:2897^M
0x910358 lto_main()^M
/home/vries/oacc/trunk/source-gcc/gcc/lto/lto.c:625^M
...
Thanks,
- Tom
diff --git a/libgomp/testsuite/libgomp.c/target-32.c b/libgomp/testsuite/libgomp.c/target-32.c
index 233877b702b..b8deae72b08 100644
--- a/libgomp/testsuite/libgomp.c/target-32.c
+++ b/libgomp/testsuite/libgomp.c/target-32.c
@@ -1,6 +1,25 @@
#include
#include
+void
+nvptx_delay (int d)
+{
+ /* This function serves as a replacement for usleep in
+ this test case. It does not even attempt to be functionally
+ equivalent - we just want some sort of delay. */
+ int i;
+ int N = d * 2000;
+ for (i = 0; i < N; i++)
+asm volatile ("" : : : "memory");
+}
+
+#pragma omp declare variant (nvptx_delay) match(construct={target},device={arch(nvptx)})
+void
+base_delay(int d)
+{
+ usleep (d);
+}
+
int main ()
{
int a = 0, b = 0, c = 0, d[7];
@@ -18,28 +37,28 @@ int main ()
#pragma omp target nowait map(alloc: b) depend(in: d[2]) depend(out: d[3])
{
- usleep (1000);
+ base_delay (1000);
#pragma omp atomic update
b |= 4;
}
#pragma omp target nowait map(alloc: b) depend(in: d[2]) depend(out: d[4])
{
- usleep (5000);
+ base_delay (5000);
#pragma omp atomic update
b |= 1;
}
#pragma omp target nowait map(alloc: c) depend(in: d[3], d[4]) depend(out: d[5])
{
- usleep (5000);
+ base_delay (5000);
#pragma omp atomic update
c |= 8;
}
#pragma omp target nowait map(alloc: c) depend(in: d[3], d[4]) depend(out: d[6])
{
- usleep (1000);
+ base_delay (1000);
#pragma omp atomic update
c |= 2;
}
Re: [PATCH] xfail and improve some failing libgomp tests
On Tue, Oct 06, 2020 at 04:48:40PM +0200, Tom de Vries wrote:
> On 10/5/20 3:15 PM, Tom de Vries wrote:
> > On 2/7/20 4:29 PM, Jakub Jelinek wrote:
> >> On Fri, Feb 07, 2020 at 09:56:38AM +0100, Harwath, Frederik wrote:
> >>> * {target-32.c, thread-limit-2.c}:
> >>> no "usleep" implemented for nvptx. Cf. https://gcc.gnu.org/PR81690
> >>
> >> Please don't, I want to deal with that using declare variant, just didn't
> >> get yet around to finishing the last patch needed for that. Will try next
> >> week.
> >>
> >
> > Hi Jakub,
> >
> > Ping, any update on this?
Not finished the last step, I run into LTO issues. Will need to return to
that soon.
Last progress in "[RFH] LTO cgraph support for late declare variant resolution"
mail from May on gcc-patches.
> --- a/libgomp/testsuite/libgomp.c/target-32.c
> +++ b/libgomp/testsuite/libgomp.c/target-32.c
> @@ -1,6 +1,26 @@
> #include
> #include
>
> +extern void base_delay(int);
No need to declare this one early.
> +extern void nvptx_delay(int);
Space before (, and the definition could go here instead of
the declaration.
> +#pragma omp declare variant( nvptx_delay ) match( construct={target},
> implementation={vendor(nvidia)} )
This isn't the right declare variant for what we want though,
we only provide gnu as accepted vendor, it is implementation's vendor,
not vendor of one of the hw components.
So, it ought to be instead
#pragma omp declare variant (nvptx_delay)
match(construct={target},device={arch(nvptx)})
> +void base_delay(int d)
> +{
> + usleep (d);
> +}
Jakub
Re: [PATCH] xfail and improve some failing libgomp tests
On 10/5/20 3:15 PM, Tom de Vries wrote:
> On 2/7/20 4:29 PM, Jakub Jelinek wrote:
>> On Fri, Feb 07, 2020 at 09:56:38AM +0100, Harwath, Frederik wrote:
>>> * {target-32.c, thread-limit-2.c}:
>>> no "usleep" implemented for nvptx. Cf. https://gcc.gnu.org/PR81690
>>
>> Please don't, I want to deal with that using declare variant, just didn't
>> get yet around to finishing the last patch needed for that. Will try next
>> week.
>>
>
> Hi Jakub,
>
> Ping, any update on this?
FWIW, I've tried as in patch attached below, but I didn't get it
compiling, I still got:
...
FAIL: libgomp.c/target-32.c (test for excess errors)
Excess errors:
unresolved symbol usleep
...
Jakub, is this already supposed to work?
Thanks,
- Tom
diff --git a/libgomp/testsuite/libgomp.c/target-32.c b/libgomp/testsuite/libgomp.c/target-32.c
index 233877b702b..7ddf8721ed3 100644
--- a/libgomp/testsuite/libgomp.c/target-32.c
+++ b/libgomp/testsuite/libgomp.c/target-32.c
@@ -1,6 +1,26 @@
#include
#include
+extern void base_delay(int);
+extern void nvptx_delay(int);
+
+#pragma omp declare variant( nvptx_delay ) match( construct={target}, implementation={vendor(nvidia)} )
+void base_delay(int d)
+{
+ usleep (d);
+}
+
+void nvptx_delay(int d)
+{
+ /* This function serves as a replacement for usleep in
+ this test case. It does not even attempt to be functionally
+ equivalent - we just want some sort of delay. */
+ int i;
+ int N = d * 2000;
+ for (i = 0; i < N; i++)
+asm volatile ("" : : : "memory");
+}
+
int main ()
{
int a = 0, b = 0, c = 0, d[7];
@@ -18,28 +38,28 @@ int main ()
#pragma omp target nowait map(alloc: b) depend(in: d[2]) depend(out: d[3])
{
- usleep (1000);
+ base_delay (1000);
#pragma omp atomic update
b |= 4;
}
#pragma omp target nowait map(alloc: b) depend(in: d[2]) depend(out: d[4])
{
- usleep (5000);
+ base_delay (5000);
#pragma omp atomic update
b |= 1;
}
#pragma omp target nowait map(alloc: c) depend(in: d[3], d[4]) depend(out: d[5])
{
- usleep (5000);
+ base_delay (5000);
#pragma omp atomic update
c |= 8;
}
#pragma omp target nowait map(alloc: c) depend(in: d[3], d[4]) depend(out: d[6])
{
- usleep (1000);
+ base_delay (1000);
#pragma omp atomic update
c |= 2;
}
Re: [PATCH] xfail and improve some failing libgomp tests
On 2/7/20 4:29 PM, Jakub Jelinek wrote:
> On Fri, Feb 07, 2020 at 09:56:38AM +0100, Harwath, Frederik wrote:
>> * {target-32.c, thread-limit-2.c}:
>> no "usleep" implemented for nvptx. Cf. https://gcc.gnu.org/PR81690
>
> Please don't, I want to deal with that using declare variant, just didn't
> get yet around to finishing the last patch needed for that. Will try next
> week.
>
Hi Jakub,
Ping, any update on this?
Thanks,
- Tom
Re: [PATCH] xfail and improve some failing libgomp tests
On Mon, Feb 10, 2020 at 08:49:47AM +0100, Harwath, Frederik wrote: > Add xfails for nvptx offloading because > "no GOMP_OFFLOAD_async_run implemented in plugin-nvptx.c" > (https://gcc.gnu.org/PR81688) and because > "omp target link not implemented for nvptx" > (https://gcc.gnu.org/PR81689). > > libgomp/ > * testsuite/libgomp.c/target-33.c: Add xfail for execution on > offload_target_nvptx, cf. https://gcc.gnu.org/PR81688. > * testsuite/libgomp.c/target-34.c: Likewise. > * testsuite/libgomp.c/target-link-1.c: Add xfail for > offload_target_nvptx, cf. https://gcc.gnu.org/PR81689. Ok, thanks. Jakub
Re: [PATCH] xfail and improve some failing libgomp tests
Hi Jakub,
On 07.02.20 16:29, Jakub Jelinek wrote:
> On Fri, Feb 07, 2020 at 09:56:38AM +0100, Harwath, Frederik wrote:
>> * {target-32.c, thread-limit-2.c}:
>> no "usleep" implemented for nvptx. Cf. https://gcc.gnu.org/PR81690
>
> Please don't, I want to deal with that using declare variant, just didn't
> get yet around to finishing the last patch needed for that. Will try next
> week.
Ok, great! looking forward to see a better solution.
>> * target-{33,34}.c:
>> no "GOMP_OFFLOAD_async_run" implemented in plugin-nvptx.c. Cf.
>> https://gcc.gnu.org/PR81688
>>
>> * target-link-1.c:
>> omp "target link" not implemented for nvptx. Cf. https://gcc.gnu.org/PR81689
>
> I guess this is ok, though of course the right thing would be to implement
> both
Ok, this means that I can commit the attached patch which contains only the
changes to
target-{33,43}.c and target-link-1.c? Of course, I agree that those features
should be
implemented.
> There has been even in some PR a suggestion that instead of failing
> in nvptx async_run we should just ignore the nowait clause if the plugin
> doesn't implement it properly.
This must be https://gcc.gnu.org/PR93481.
Best regards,
Frederik
From e5165ccb143022614920dbd208f6f368b84b4382 Mon Sep 17 00:00:00 2001
From: Frederik Harwath
Date: Mon, 10 Feb 2020 08:08:00 +0100
Subject: [PATCH] Add xfails to libgomp tests target-{33,34}.c, target-link-1.c
Add xfails for nvptx offloading because
"no GOMP_OFFLOAD_async_run implemented in plugin-nvptx.c"
(https://gcc.gnu.org/PR81688) and because
"omp target link not implemented for nvptx"
(https://gcc.gnu.org/PR81689).
libgomp/
* testsuite/libgomp.c/target-33.c: Add xfail for execution on
offload_target_nvptx, cf. https://gcc.gnu.org/PR81688.
* testsuite/libgomp.c/target-34.c: Likewise.
* testsuite/libgomp.c/target-link-1.c: Add xfail for
offload_target_nvptx, cf. https://gcc.gnu.org/PR81689.
---
libgomp/testsuite/libgomp.c/target-33.c | 3 +++
libgomp/testsuite/libgomp.c/target-34.c | 3 +++
libgomp/testsuite/libgomp.c/target-link-1.c | 3 +++
3 files changed, 9 insertions(+)
diff --git a/libgomp/testsuite/libgomp.c/target-33.c b/libgomp/testsuite/libgomp.c/target-33.c
index 1bed4b6bc67..15d2d7e38ab 100644
--- a/libgomp/testsuite/libgomp.c/target-33.c
+++ b/libgomp/testsuite/libgomp.c/target-33.c
@@ -1,3 +1,6 @@
+/* { dg-xfail-run-if "GOMP_OFFLOAD_async_run not implemented" { offload_target_nvptx } }
+ Cf. https://gcc.gnu.org/PR81688. */
+
extern void abort (void);
int
diff --git a/libgomp/testsuite/libgomp.c/target-34.c b/libgomp/testsuite/libgomp.c/target-34.c
index 66d9f54202b..5a3596424d8 100644
--- a/libgomp/testsuite/libgomp.c/target-34.c
+++ b/libgomp/testsuite/libgomp.c/target-34.c
@@ -1,3 +1,6 @@
+/* { dg-xfail-run-if "GOMP_OFFLOAD_async_run not implemented" { offload_target_nvptx } }
+ Cf. https://gcc.gnu.org/PR81688. */
+
extern void abort (void);
int
diff --git a/libgomp/testsuite/libgomp.c/target-link-1.c b/libgomp/testsuite/libgomp.c/target-link-1.c
index 681677cc2aa..99ce33bc9b4 100644
--- a/libgomp/testsuite/libgomp.c/target-link-1.c
+++ b/libgomp/testsuite/libgomp.c/target-link-1.c
@@ -1,3 +1,6 @@
+/* { dg-xfail-if "#pragma omp target link not implemented" { offload_target_nvptx } }
+ Cf. https://gcc.gnu.org/PR81689. */
+
struct S { int s, t; };
int a = 1, b = 1;
--
2.17.1
Re: [PATCH] xfail and improve some failing libgomp tests
On Fri, Feb 07, 2020 at 09:56:38AM +0100, Harwath, Frederik wrote:
> * {target-32.c, thread-limit-2.c}:
> no "usleep" implemented for nvptx. Cf. https://gcc.gnu.org/PR81690
Please don't, I want to deal with that using declare variant, just didn't
get yet around to finishing the last patch needed for that. Will try next week.
> * target-{33,34}.c:
> no "GOMP_OFFLOAD_async_run" implemented in plugin-nvptx.c. Cf.
> https://gcc.gnu.org/PR81688
>
> * target-link-1.c:
> omp "target link" not implemented for nvptx. Cf. https://gcc.gnu.org/PR81689
I guess this is ok, though of course the right thing would be to implement
both. There has been even in some PR a suggestion that instead of failing
in nvptx async_run we should just ignore the nowait clause if the plugin
doesn't implement it properly.
Jakub
[PATCH] xfail and improve some failing libgomp tests
Hi,
the libgomp testsuite contains some test cases (all in
/libgomp/testsuite/libgomp.c/)
which fail with nvptx offloading because of some long standing issues:
* {target-32.c, thread-limit-2.c}:
no "usleep" implemented for nvptx. Cf. https://gcc.gnu.org/PR81690
* target-{33,34}.c:
no "GOMP_OFFLOAD_async_run" implemented in plugin-nvptx.c. Cf.
https://gcc.gnu.org/PR81688
* target-link-1.c:
omp "target link" not implemented for nvptx. Cf. https://gcc.gnu.org/PR81689
All these issues have been known, at least, since 2016:
https://gcc.gnu.org/ml/gcc-patches/2016-11/msg00972.html
As suggested in this mail:
"Short term, it should be possible to implement something like -foffload=^nvptx
to skip PTX (and only PTX) offloading on those tests."
Well, we can now skip/xfail tests for nvptx offloading using the effective
target
"offload_target_nvptx" and the present patch uses this to xfail the tests for
which
no short-term solution is in sight, i.e. the GOMP_OFFLOAD_async_run and the
"target link"
related failures.
Regarding the "usleep" issue, I have decided to follow Jakub's suggestion
(cf. https://gcc.gnu.org/ml/gcc-patches/2016-11/msg01026.html) to
replace usleep by busy waiting. As noted by Tobias
(https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81690#c4), this involves creating
separate
test files for the cases with and without usleep. This solution is a bit
cumbersome but I
think we can live with it, in particular, since the actual test case
implementations do not
get duplicated (they have been moved into auxiliary header files which are
shared by both
variants of the corresponding tests).
Since the "usleep" issue also concerns amdgcn, I have introduced an effective
target
"offload_target_amdgcn" to add xfails for this offloading target, too. This
behaves like
"offload_target_nvptx" but for amdgcn. Note that the existing amdgcn effective
targets
cannot be used for our purpose since they are OpenACC-specific.
The new thread-limit-2-nosleep.c should now pass for both nvptx and amdgcn
offloading
whereas thread-limit-2.c should xfail. The new target-32-nosleep.c passes with
amdgcn
offloading, but xfails with nvptx offloading, because it also needs the
unimplemented
GOMP_OFFLOAD_async_run.
With the patch, the detailed test summary now looks as follows for me:
nvptx offloading:
// Expected execution failures due to missing usleep
PASS: libgomp.c/target-32-nosleep.c (test for excess errors)
XFAIL: libgomp.c/target-32-nosleep.c execution test// missing
GOMP_OFFLOAD_async_run
XFAIL: libgomp.c/target-32.c (test for excess errors)
UNRESOLVED: libgomp.c/target-32.c compilation failed to produce executable
PASS: libgomp.c/thread-limit-2-nosleep.c (test for excess errors)
PASS: libgomp.c/thread-limit-2-nosleep.c execution test
XFAIL: libgomp.c/thread-limit-2.c (test for excess errors)
UNRESOLVED: libgomp.c/thread-limit-2.c compilation failed to produce executable
// Expected execution failures due to missing GOMP_OFFLOAD_async_run
PASS: libgomp.c/target-33.c (test for excess errors)
XFAIL: libgomp.c/target-33.c execution test
PASS: libgomp.c/target-34.c (test for excess errors)
XFAIL: libgomp.c/target-34.c execution test
// Expected compilation failures due to missing target link
XFAIL: libgomp.c/target-link-1.c (test for excess errors)
UNRESOLVED: libgomp.c/target-link-1.c compilation failed to produce executable
amdgcn offloading:
// Tests using usleep
PASS: libgomp.c/target-32-nosleep.c (test for excess errors)
PASS: libgomp.c/target-32-nosleep.c execution test
XFAIL: libgomp.c/target-32.c 7 blank line(s) in output
XFAIL: libgomp.c/target-32.c (test for excess errors)
UNRESOLVED: libgomp.c/target-32.c compilation failed to produce executable
PASS: libgomp.c/thread-limit-2-nosleep.c (test for excess errors)
PASS: libgomp.c/thread-limit-2-nosleep.c execution test
XFAIL: libgomp.c/thread-limit-2.c 1 blank line(s) in output
XFAIL: libgomp.c/thread-limit-2.c (test for excess errors)
// No failures since GOMP_OFFLOAD_async_run works on amdgcn
PASS: libgomp.c/target-33.c (test for excess errors)
PASS: libgomp.c/target-33.c execution test
PASS: libgomp.c/target-34.c (test for excess errors)
PASS: libgomp.c/target-34.c execution test
// No xfail here
PASS: libgomp.c/target-link-1.c (test for excess errors)
FAIL: libgomp.c/target-link-1.c execution test
Note that target-link-1.c execution does also fail on amdgcn.
Since - in contrast to nvptx - it seems that the cause of this failure
has not yet been investigated and discussed, I have not added an xfail
for amdgcn to this test.
All testing has been done with a x86_64-linux-gnu host and target.
Ok to commit this patch?
Best regards,
Frederik
From 6e5e2d45f02235a0f72e6130dcd8d52f88f7b126 Mon Sep 17 00:00:00 2001
From: Frederik Harwath
Date: Fri, 7 Feb 2020 08:03:00 +0100
Subject: [PATCH] xfail and improve some failing libgomp
