* [PATCH] x86: set spincount 1 for x86 hybrid platform [PR109812]
@ 2023-10-10 4:59 Jun Zhang
2023-10-10 7:56 ` Jakub Jelinek
0 siblings, 1 reply; 4+ messages in thread
From: Jun Zhang @ 2023-10-10 4:59 UTC (permalink / raw)
To: gcc-patches; +Cc: ubizjak, jakub, hongtao.liu, jun.zhang
From: "Zhang, Jun" <jun.zhang@intel.com>
By test, we find in hybrid platform spincount 1 is better.
Use '-march=native -Ofast -funroll-loops -flto',
results as follows:
spec2017 speed RPL ADL
657.xz_s 0.00% 0.50%
603.bwaves_s 10.90% 26.20%
607.cactuBSSN_s 5.50% 72.50%
619.lbm_s 2.40% 2.50%
621.wrf_s -7.70% 2.40%
627.cam4_s 0.50% 0.70%
628.pop2_s 48.20% 153.00%
638.imagick_s -0.10% 0.20%
644.nab_s 2.30% 1.40%
649.fotonik3d_s 8.00% 13.80%
654.roms_s 1.20% 1.10%
Geomean-int 0.00% 0.50%
Geomean-fp 6.30% 21.10%
Geomean-all 5.70% 19.10%
omp2012 RPL ADL
350.md -1.81% -1.75%
351.bwaves 7.72% 12.50%
352.nab 14.63% 19.71%
357.bt331 -0.20% 1.77%
358.botsalgn 0.00% 0.00%
359.botsspar 0.00% 0.65%
360.ilbdc 0.00% 0.25%
362.fma3d 2.66% -0.51%
363.swim 10.44% 0.00%
367.imagick 0.00% 0.12%
370.mgrid331 2.49% 25.56%
371.applu331 1.06% 4.22%
372.smithwa 0.74% 3.34%
376.kdtree 10.67% 16.03%
GEOMEAN 3.34% 5.53%
include/ChangeLog:
* omphook.h: define RUNOMPHOOK macro.
libgomp/ChangeLog:
* env.c (initialize_env): add RUNOMPHOOK macro.
* config/linux/x86/omphook.h: define RUNOMPHOOK macro.
---
include/omphook.h | 1 +
libgomp/config/linux/x86/omphook.h | 19 +++++++++++++++++++
libgomp/env.c | 3 +++
3 files changed, 23 insertions(+)
create mode 100644 include/omphook.h
create mode 100644 libgomp/config/linux/x86/omphook.h
diff --git a/include/omphook.h b/include/omphook.h
new file mode 100644
index 00000000000..2ebe3ad57e6
--- /dev/null
+++ b/include/omphook.h
@@ -0,0 +1 @@
+#define RUNOMPHOOK()
diff --git a/libgomp/config/linux/x86/omphook.h b/libgomp/config/linux/x86/omphook.h
new file mode 100644
index 00000000000..aefb311cc07
--- /dev/null
+++ b/libgomp/config/linux/x86/omphook.h
@@ -0,0 +1,19 @@
+#ifdef __x86_64__
+#include "cpuid.h"
+
+/* only for x86 hybrid platform */
+#define RUNOMPHOOK() \
+ do \
+ { \
+ unsigned int eax, ebx, ecx, edx; \
+ if ((getenv ("GOMP_SPINCOUNT") == NULL) && (wait_policy < 0) \
+ && __get_cpuid_count (7, 0, &eax, &ebx, &ecx, &edx) \
+ && ((edx >> 15) & 1)) \
+ gomp_spin_count_var = 1LL; \
+ if (gomp_throttled_spin_count_var > gomp_spin_count_var) \
+ gomp_throttled_spin_count_var = gomp_spin_count_var; \
+ } \
+ while (0)
+#else
+# include "../../../../include/omphook.h"
+#endif
diff --git a/libgomp/env.c b/libgomp/env.c
index a21adb3fd4b..1f13a148694 100644
--- a/libgomp/env.c
+++ b/libgomp/env.c
@@ -61,6 +61,7 @@
#include "secure_getenv.h"
#include "environ.h"
+#include "omphook.h"
/* Default values of ICVs according to the OpenMP standard,
except for default-device-var. */
@@ -2496,5 +2497,7 @@ initialize_env (void)
goacc_runtime_initialize ();
goacc_profiling_initialize ();
+
+ RUNOMPHOOK ();
}
#endif /* LIBGOMP_OFFLOADED_ONLY */
--
2.31.1
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] x86: set spincount 1 for x86 hybrid platform [PR109812]
2023-10-10 4:59 [PATCH] x86: set spincount 1 for x86 hybrid platform [PR109812] Jun Zhang
@ 2023-10-10 7:56 ` Jakub Jelinek
2023-10-11 2:47 ` Zhang, Jun
0 siblings, 1 reply; 4+ messages in thread
From: Jakub Jelinek @ 2023-10-10 7:56 UTC (permalink / raw)
To: Jun Zhang; +Cc: gcc-patches, ubizjak, hongtao.liu
On Tue, Oct 10, 2023 at 12:59:52PM +0800, Jun Zhang wrote:
> include/ChangeLog:
>
> * omphook.h: define RUNOMPHOOK macro.
ChangeLog formatting. The description should start with
capital letter. If you add a new file, you should just
mention : New file. or something similar.
But much more importantly, we don't do hooks like that in libgomp,
it should be better a static inline function where the name shows
what it is doing (e.g. do_adjust_default_spincount), best in some
already existing header (e.g. wait.h), and best at the end of the
if (!parse_spincount ("GOMP_SPINCOUNT", &gomp_spin_count_var))
block in env.c.
>
> libgomp/ChangeLog:
>
> * env.c (initialize_env): add RUNOMPHOOK macro.
> * config/linux/x86/omphook.h: define RUNOMPHOOK macro.
> ---
> include/omphook.h | 1 +
> libgomp/config/linux/x86/omphook.h | 19 +++++++++++++++++++
> libgomp/env.c | 3 +++
> 3 files changed, 23 insertions(+)
> create mode 100644 include/omphook.h
> create mode 100644 libgomp/config/linux/x86/omphook.h
>
> diff --git a/include/omphook.h b/include/omphook.h
> new file mode 100644
> index 00000000000..2ebe3ad57e6
> --- /dev/null
> +++ b/include/omphook.h
> @@ -0,0 +1 @@
> +#define RUNOMPHOOK()
> diff --git a/libgomp/config/linux/x86/omphook.h b/libgomp/config/linux/x86/omphook.h
> new file mode 100644
> index 00000000000..aefb311cc07
> --- /dev/null
> +++ b/libgomp/config/linux/x86/omphook.h
> @@ -0,0 +1,19 @@
> +#ifdef __x86_64__
> +#include "cpuid.h"
> +
> +/* only for x86 hybrid platform */
Comments should again start with capital letter and end with dot and 2
spaces before */
> +#define RUNOMPHOOK() \
> + do \
> + { \
> + unsigned int eax, ebx, ecx, edx; \
> + if ((getenv ("GOMP_SPINCOUNT") == NULL) && (wait_policy < 0) \
Spurious ()s around the subcondition, the first shouldn't be tested when
placed right, if condition doesn't fit on one line, each subcondition
should be on separate line.
> + && __get_cpuid_count (7, 0, &eax, &ebx, &ecx, &edx) \
If we don't have a macro for the CPUID.07H.0H:EDX[15] bit, there should
be a comment which says what that bit is.
> + && ((edx >> 15) & 1)) \
> + gomp_spin_count_var = 1LL; \
> + if (gomp_throttled_spin_count_var > gomp_spin_count_var) \
> + gomp_throttled_spin_count_var = gomp_spin_count_var; \
The above 2 lines won't be needed if placed right.
Jakub
^ permalink raw reply [flat|nested] 4+ messages in thread
* RE: [PATCH] x86: set spincount 1 for x86 hybrid platform [PR109812]
2023-10-10 7:56 ` Jakub Jelinek
@ 2023-10-11 2:47 ` Zhang, Jun
2023-10-11 6:27 ` Jakub Jelinek
0 siblings, 1 reply; 4+ messages in thread
From: Zhang, Jun @ 2023-10-11 2:47 UTC (permalink / raw)
To: Jakub Jelinek; +Cc: gcc-patches, ubizjak, Liu, Hongtao
> -----Original Message-----
> From: Jakub Jelinek <jakub@redhat.com>
> Sent: Tuesday, October 10, 2023 3:57 PM
> To: Zhang, Jun <jun.zhang@intel.com>
> Cc: gcc-patches@gcc.gnu.org; ubizjak@gmail.com; Liu, Hongtao
> <hongtao.liu@intel.com>
> Subject: Re: [PATCH] x86: set spincount 1 for x86 hybrid platform [PR109812]
>
> On Tue, Oct 10, 2023 at 12:59:52PM +0800, Jun Zhang wrote:
> > include/ChangeLog:
> >
> > * omphook.h: define RUNOMPHOOK macro.
>
> ChangeLog formatting. The description should start with capital letter. If you add
> a new file, you should just mention : New file. or something similar.
>
> But much more importantly, we don't do hooks like that in libgomp, it should be
> better a static inline function where the name shows what it is doing (e.g.
> do_adjust_default_spincount), best in some already existing header (e.g. wait.h),
Sorry, I don't know how to place wait.h. This patch is only for x86.
Wait.h is in libgomp/config/linux/.
if I directly place do_adjust_default_spincount in wait.h,
it maybe need #ifdef __x86_64__, and also need a wait.h in include/
for other os platform.
if I create a new wait.h in libgomp/config/linux/x86/,
It need copy whole wait.h from libgomp/config/linux/,
also need a wait.h in include/ for other os platform.
Could you help me?
> and best at the end of the if (!parse_spincount ("GOMP_SPINCOUNT",
> &gomp_spin_count_var)) block in env.c.
> >
> > libgomp/ChangeLog:
> >
> > * env.c (initialize_env): add RUNOMPHOOK macro.
> > * config/linux/x86/omphook.h: define RUNOMPHOOK macro.
> > ---
> > include/omphook.h | 1 +
> > libgomp/config/linux/x86/omphook.h | 19 +++++++++++++++++++
> > libgomp/env.c | 3 +++
> > 3 files changed, 23 insertions(+)
> > create mode 100644 include/omphook.h
> > create mode 100644 libgomp/config/linux/x86/omphook.h
> >
> > diff --git a/include/omphook.h b/include/omphook.h new file mode
> > 100644 index 00000000000..2ebe3ad57e6
> > --- /dev/null
> > +++ b/include/omphook.h
> > @@ -0,0 +1 @@
> > +#define RUNOMPHOOK()
> > diff --git a/libgomp/config/linux/x86/omphook.h
> > b/libgomp/config/linux/x86/omphook.h
> > new file mode 100644
> > index 00000000000..aefb311cc07
> > --- /dev/null
> > +++ b/libgomp/config/linux/x86/omphook.h
> > @@ -0,0 +1,19 @@
> > +#ifdef __x86_64__
> > +#include "cpuid.h"
> > +
> > +/* only for x86 hybrid platform */
>
> Comments should again start with capital letter and end with dot and 2 spaces
> before */
>
> > +#define RUNOMPHOOK() \
> > + do \
> > + { \
> > + unsigned int eax, ebx, ecx, edx; \
> > + if ((getenv ("GOMP_SPINCOUNT") == NULL) && (wait_policy < 0) \
>
> Spurious ()s around the subcondition, the first shouldn't be tested when placed
> right, if condition doesn't fit on one line, each subcondition should be on separate
> line.
>
> > + && __get_cpuid_count (7, 0, &eax, &ebx, &ecx, &edx) \
>
> If we don't have a macro for the CPUID.07H.0H:EDX[15] bit, there should be a
> comment which says what that bit is.
>
> > + && ((edx >> 15) & 1)) \
> > + gomp_spin_count_var = 1LL; \
> > + if (gomp_throttled_spin_count_var > gomp_spin_count_var) \
> > + gomp_throttled_spin_count_var = gomp_spin_count_var; \
>
> The above 2 lines won't be needed if placed right.
>
> Jakub
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] x86: set spincount 1 for x86 hybrid platform [PR109812]
2023-10-11 2:47 ` Zhang, Jun
@ 2023-10-11 6:27 ` Jakub Jelinek
0 siblings, 0 replies; 4+ messages in thread
From: Jakub Jelinek @ 2023-10-11 6:27 UTC (permalink / raw)
To: Zhang, Jun; +Cc: gcc-patches, ubizjak, Liu, Hongtao
On Wed, Oct 11, 2023 at 02:47:56AM +0000, Zhang, Jun wrote:
> Sorry, I don't know how to place wait.h. This patch is only for x86.
> Wait.h is in libgomp/config/linux/.
> if I directly place do_adjust_default_spincount in wait.h,
> it maybe need #ifdef __x86_64__, and also need a wait.h in include/
> for other os platform.
> if I create a new wait.h in libgomp/config/linux/x86/,
> It need copy whole wait.h from libgomp/config/linux/,
> also need a wait.h in include/ for other os platform.
> Could you help me?
Put it into a new spincount.h then.
Sorry, missed wait.h isn't CPU specific.
Jakub
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2023-10-11 6:27 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-10-10 4:59 [PATCH] x86: set spincount 1 for x86 hybrid platform [PR109812] Jun Zhang
2023-10-10 7:56 ` Jakub Jelinek
2023-10-11 2:47 ` Zhang, Jun
2023-10-11 6:27 ` Jakub Jelinek
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).