Re: [dpdk-dev] [PATCH] config: make AVX and AVX512 configurable

Re: [dpdk-dev] [PATCH] config: make AVX and AVX512 configurable

[Thomas Monjalon]

27/04/2017 18:34, Zhihong Wang:
> Making AVX and AVX512 configurable is useful for performance and power
> testing.
> 
> The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.
[...]
> +#
> +# Recognize/ignore the AVX/AVX512 CPU flags for performance/power testing
> +#
> +CONFIG_RTE_ENABLE_AVX=y
> +CONFIG_RTE_ENABLE_AVX512=n

It is disabling AVX512 in default configuration.
Please explain this behaviour change.

Re: [dpdk-dev] [PATCH] config: make AVX and AVX512 configurable

[Wang, Zhihong]

> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas@monjalon.net]
> Sent: Thursday, April 27, 2017 5:08 PM
> To: Wang, Zhihong <zhihong.wang@intel.com>; Richardson, Bruce
> <bruce.richardson@intel.com>
> Cc: dev@dpdk.org; yuanhan.liu@linux.intel.com
> Subject: Re: [dpdk-dev] [PATCH] config: make AVX and AVX512 configurable
> 
> 27/04/2017 18:34, Zhihong Wang:
> > Making AVX and AVX512 configurable is useful for performance and power
> > testing.
> >
> > The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.
> [...]
> > +#
> > +# Recognize/ignore the AVX/AVX512 CPU flags for performance/power
> testing
> > +#
> > +CONFIG_RTE_ENABLE_AVX=y
> > +CONFIG_RTE_ENABLE_AVX512=n
> 
> It is disabling AVX512 in default configuration.
> Please explain this behaviour change.

Though AVX512 rte_memcpy has been in DPDK for quite a while it's still
unproven in hardware with rich use cases. Mark it as experimental for
now, user can enable it for their own testing.

Will enable it with enough field tests and possible optimization.

Should I add the explanation in commit log, or comments in the source,
or both?

Re: [dpdk-dev] [PATCH] config: make AVX and AVX512 configurable

[Thomas Monjalon]

27/04/2017 11:18, Wang, Zhihong:
> From: Thomas Monjalon [mailto:thomas@monjalon.net]
> > 27/04/2017 18:34, Zhihong Wang:
> > > Making AVX and AVX512 configurable is useful for performance and power
> > > testing.
> > >
> > > The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.
> > [...]
> > > +#
> > > +# Recognize/ignore the AVX/AVX512 CPU flags for performance/power
> > testing
> > > +#
> > > +CONFIG_RTE_ENABLE_AVX=y
> > > +CONFIG_RTE_ENABLE_AVX512=n
> > 
> > It is disabling AVX512 in default configuration.
> > Please explain this behaviour change.
> 
> Though AVX512 rte_memcpy has been in DPDK for quite a while it's still
> unproven in hardware with rich use cases. Mark it as experimental for
> now, user can enable it for their own testing.
> 
> Will enable it with enough field tests and possible optimization.
> 
> Should I add the explanation in commit log, or comments in the source,
> or both?

Yes please, add the explanation in the commit log and experimental comment
in the config.

Re: [dpdk-dev] [PATCH] config: make AVX and AVX512 configurable

[Wang, Zhihong]

> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas@monjalon.net]
> Sent: Thursday, April 27, 2017 5:20 PM
> To: Wang, Zhihong <zhihong.wang@intel.com>
> Cc: Richardson, Bruce <bruce.richardson@intel.com>; dev@dpdk.org;
> yuanhan.liu@linux.intel.com
> Subject: Re: [dpdk-dev] [PATCH] config: make AVX and AVX512 configurable
> 
> 27/04/2017 11:18, Wang, Zhihong:
> > From: Thomas Monjalon [mailto:thomas@monjalon.net]
> > > 27/04/2017 18:34, Zhihong Wang:
> > > > Making AVX and AVX512 configurable is useful for performance and
> power
> > > > testing.
> > > >
> > > > The similar kernel patch at
> https://patchwork.kernel.org/patch/9618883/.
> > > [...]
> > > > +#
> > > > +# Recognize/ignore the AVX/AVX512 CPU flags for
> performance/power
> > > testing
> > > > +#
> > > > +CONFIG_RTE_ENABLE_AVX=y
> > > > +CONFIG_RTE_ENABLE_AVX512=n
> > >
> > > It is disabling AVX512 in default configuration.
> > > Please explain this behaviour change.
> >
> > Though AVX512 rte_memcpy has been in DPDK for quite a while it's still
> > unproven in hardware with rich use cases. Mark it as experimental for
> > now, user can enable it for their own testing.
> >
> > Will enable it with enough field tests and possible optimization.
> >
> > Should I add the explanation in commit log, or comments in the source,
> > or both?
> 
> Yes please, add the explanation in the commit log and experimental
> comment
> in the config.

Thanks a lot! It's included in v2.

[dpdk-dev] [PATCH] config: make AVX and AVX512 configurable

[Zhihong Wang]

Making AVX and AVX512 configurable is useful for performance and power
testing.

The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.

Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>
---
 config/common_base | 6 ++++++
 mk/rte.cpuflags.mk | 6 ++++++
 2 files changed, 12 insertions(+)

diff --git a/config/common_base b/config/common_base
index 0b4297c..9ca94a8 100644
--- a/config/common_base
+++ b/config/common_base
@@ -103,6 +103,12 @@ CONFIG_RTE_EAL_IGB_UIO=n
 CONFIG_RTE_EAL_VFIO=n
 CONFIG_RTE_MALLOC_DEBUG=n
 
+#
+# Recognize/ignore the AVX/AVX512 CPU flags for performance/power testing
+#
+CONFIG_RTE_ENABLE_AVX=y
+CONFIG_RTE_ENABLE_AVX512=n
+
 # Default driver path (or "" to disable)
 CONFIG_RTE_EAL_PMD_PATH=""
 
diff --git a/mk/rte.cpuflags.mk b/mk/rte.cpuflags.mk
index e634abc..4288c14 100644
--- a/mk/rte.cpuflags.mk
+++ b/mk/rte.cpuflags.mk
@@ -70,8 +70,10 @@ CPUFLAGS += PCLMULQDQ
 endif
 
 ifneq ($(filter $(AUTO_CPUFLAGS),__AVX__),)
+ifeq ($(CONFIG_RTE_ENABLE_AVX),y)
 CPUFLAGS += AVX
 endif
+endif
 
 ifneq ($(filter $(AUTO_CPUFLAGS),__RDRND__),)
 CPUFLAGS += RDRAND
@@ -86,12 +88,16 @@ CPUFLAGS += F16C
 endif
 
 ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),)
+ifeq ($(CONFIG_RTE_ENABLE_AVX),y)
 CPUFLAGS += AVX2
 endif
+endif
 
 ifneq ($(filter $(AUTO_CPUFLAGS),__AVX512F__),)
+ifeq ($(CONFIG_RTE_ENABLE_AVX512),y)
 CPUFLAGS += AVX512F
 endif
+endif
 
 # IBM Power CPU flags
 ifneq ($(filter $(AUTO_CPUFLAGS),__PPC64__),)
-- 
2.7.4

[dpdk-dev] [PATCH v2] config: make AVX and AVX512 configurable

[Zhihong Wang]

Making AVX and AVX512 configurable is useful for performance and power
testing.

The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.

AVX512 support like in rte_memcpy has been in DPDK since 16.04, but it's
still unproven in rich use cases in hardware. Therefore it's marked as
experimental for now, will enable it after enough field test and possible
optimization.

Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>
---
Changes in v2:

 1. Add comments and explanation.

---
 config/common_base | 8 ++++++++
 mk/rte.cpuflags.mk | 6 ++++++
 2 files changed, 14 insertions(+)

diff --git a/config/common_base b/config/common_base
index 0b4297c..93e9235 100644
--- a/config/common_base
+++ b/config/common_base
@@ -103,6 +103,14 @@ CONFIG_RTE_EAL_IGB_UIO=n
 CONFIG_RTE_EAL_VFIO=n
 CONFIG_RTE_MALLOC_DEBUG=n
 
+#
+# Recognize/ignore the AVX/AVX512 CPU flags for performance/power testing.
+# AVX512 is marked as experimental for now, will enable it after enough
+# field test and possible optimization.
+#
+CONFIG_RTE_ENABLE_AVX=y
+CONFIG_RTE_ENABLE_AVX512=n
+
 # Default driver path (or "" to disable)
 CONFIG_RTE_EAL_PMD_PATH=""
 
diff --git a/mk/rte.cpuflags.mk b/mk/rte.cpuflags.mk
index e634abc..4288c14 100644
--- a/mk/rte.cpuflags.mk
+++ b/mk/rte.cpuflags.mk
@@ -70,8 +70,10 @@ CPUFLAGS += PCLMULQDQ
 endif
 
 ifneq ($(filter $(AUTO_CPUFLAGS),__AVX__),)
+ifeq ($(CONFIG_RTE_ENABLE_AVX),y)
 CPUFLAGS += AVX
 endif
+endif
 
 ifneq ($(filter $(AUTO_CPUFLAGS),__RDRND__),)
 CPUFLAGS += RDRAND
@@ -86,12 +88,16 @@ CPUFLAGS += F16C
 endif
 
 ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),)
+ifeq ($(CONFIG_RTE_ENABLE_AVX),y)
 CPUFLAGS += AVX2
 endif
+endif
 
 ifneq ($(filter $(AUTO_CPUFLAGS),__AVX512F__),)
+ifeq ($(CONFIG_RTE_ENABLE_AVX512),y)
 CPUFLAGS += AVX512F
 endif
+endif
 
 # IBM Power CPU flags
 ifneq ($(filter $(AUTO_CPUFLAGS),__PPC64__),)
-- 
2.7.4

Re: [dpdk-dev] [PATCH v2] config: make AVX and AVX512 configurable

[Yang, Zhiyong]

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Zhihong Wang
> Sent: Friday, April 28, 2017 7:00 AM
> To: dev@dpdk.org
> Cc: yuanhan.liu@linux.intel.com; thomas@monjalon.net; Richardson, Bruce
> <bruce.richardson@intel.com>; Wang, Zhihong <zhihong.wang@intel.com>
> Subject: [dpdk-dev] [PATCH v2] config: make AVX and AVX512 configurable
> 
> Making AVX and AVX512 configurable is useful for performance and power
> testing.
> 
> The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.
> 
> AVX512 support like in rte_memcpy has been in DPDK since 16.04, but it's still
> unproven in rich use cases in hardware. Therefore it's marked as experimental
> for now, will enable it after enough field test and possible optimization.
> 
> Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>
> ---

Reviewed-by: Zhiyong Yang <zhiyong.yang@intel.com>

Thanks
Zhiyong

Re: [dpdk-dev] [PATCH v2] config: make AVX and AVX512 configurable

[Yuanhan Liu]

On Thu, Apr 27, 2017 at 07:00:14PM -0400, Zhihong Wang wrote:
> Making AVX and AVX512 configurable is useful for performance and power
> testing.
> 
> The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.
> 
> AVX512 support like in rte_memcpy has been in DPDK since 16.04, but it's
> still unproven in rich use cases in hardware. Therefore it's marked as
> experimental for now, will enable it after enough field test and possible
> optimization.
> 
> Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>

Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>

	--yliu

Re: [dpdk-dev] [PATCH v2] config: make AVX and AVX512 configurable

[Thomas Monjalon]

28/04/2017 04:08, Yang, Zhiyong:
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Zhihong Wang
> > 
> > Making AVX and AVX512 configurable is useful for performance and power
> > testing.
> > 
> > The similar kernel patch at https://patchwork.kernel.org/patch/9618883/.
> > 
> > AVX512 support like in rte_memcpy has been in DPDK since 16.04, but it's still
> > unproven in rich use cases in hardware. Therefore it's marked as experimental
> > for now, will enable it after enough field test and possible optimization.
> > 
> > Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>
> 
> Reviewed-by: Zhiyong Yang <zhiyong.yang@intel.com>

Applied, thanks