[dpdk-dev,v2,16/16] acl: check for SSE 4.1 support

  The main goal of this check is to avoid passing the -msse4.1
option to the GCC that does not support it (like arm toolchains).

Anyway, the ACL library does not compile on ARM.

Signed-off-by: Jan Viktorin <viktorin@rehivetech.com>
---
 lib/librte_acl/Makefile | 4 ++++
 1 file changed, 4 insertions(+)
  

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jan Viktorin
> Sent: Monday, October 26, 2015 4:38 PM
> To: Thomas Monjalon; Hunt, David; dev@dpdk.org
> Subject: [dpdk-dev] [PATCH v2 16/16] acl: check for SSE 4.1 support
> 
> The main goal of this check is to avoid passing the -msse4.1
> option to the GCC that does not support it (like arm toolchains).
> 
> Anyway, the ACL library does not compile on ARM.
> 
> Signed-off-by: Jan Viktorin <viktorin@rehivetech.com>
> ---
>  lib/librte_acl/Makefile | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/lib/librte_acl/Makefile b/lib/librte_acl/Makefile
> index 7a1cf8a..401fb8c 100644
> --- a/lib/librte_acl/Makefile
> +++ b/lib/librte_acl/Makefile
> @@ -50,7 +50,11 @@ SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_gen.c
>  SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_scalar.c
>  SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_sse.c
> 
> +CC_SSE4_1_SUPPORT := $(shell $(CC) -msse4.1 -dM -E - < /dev/null >/dev/null 2>&1 && echo 1)
> +
> +ifeq ($(CC_SSE4_1_SUPPORT),1)
>  CFLAGS_acl_run_sse.o += -msse4.1
> +endif

I don't think acl_run_sse.c would compile if SSE4_1 is not supported.
So, I think you need to do same thing, as is done for AVX2:
Compile in acl_run_sse.c  only if SSE41 is supported by the compiler:  

- SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_sse.c
-
-CFLAGS_acl_run_sse.o += -msse4.1

+CC_SSE41_SUPPORT=$(shell $(CC) -msse4.1 -dM -E - </dev/null 2>&1 | \
grep -q  && echo 1)
+ifeq ($(CC_SSE41_SUPPORT), 1)
+        SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_sse.c
+        CFLAGS_rte_acl.o += -DCC_SSE41_SUPPORT
+        CFLAGS_acl_run_sse.o += -msse4.1	
+endif

And then change rte_acl_init() accordingly.
Something like:

int __attribute__ ((weak))
rte_acl_classify_sse(__rte_unused const struct rte_acl_ctx *ctx,
        __rte_unused const uint8_t **data,
        __rte_unused uint32_t *results,
        __rte_unused uint32_t num,
        __rte_unused uint32_t categories)
{
        return -ENOTSUP;
}

....

static void __attribute__((constructor))
rte_acl_init(void)
{
        enum rte_acl_classify_alg alg = RTE_ACL_CLASSIFY_DEFAULT;

#if defined(CC_AVX2_SUPPORT)
        if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2))
                alg = RTE_ACL_CLASSIFY_AVX2;
        else if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_SSE4_1))
#elif defined (CC_SSE41_SUPPORT)
        if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_SSE4_1))
                alg = RTE_ACL_CLASSIFY_SSE;
#endif

        rte_acl_set_default_classify(alg);
}

After that, I suppose, you should be able to build and (probably use) librte_acl on arm.
Konstantin
  

Hello Konstantin,

thank you for those hints! I reworked the code a little bit -
refactored the LPM vector operations into rte_vect.h which helped to
the ACL library as well. I will sort it out a little and send the v3 of
the series. The LPM and ACL now seem to be much more sane on ARM.

Regards
Jan

On Tue, 27 Oct 2015 15:55:48 +0000
"Ananyev, Konstantin" <konstantin.ananyev@intel.com> wrote:

> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Jan Viktorin
> > Sent: Monday, October 26, 2015 4:38 PM
> > To: Thomas Monjalon; Hunt, David; dev@dpdk.org
> > Subject: [dpdk-dev] [PATCH v2 16/16] acl: check for SSE 4.1 support
> > 
> > The main goal of this check is to avoid passing the -msse4.1
> > option to the GCC that does not support it (like arm toolchains).
> > 
> > Anyway, the ACL library does not compile on ARM.
> > 
> > Signed-off-by: Jan Viktorin <viktorin@rehivetech.com>
> > ---
> >  lib/librte_acl/Makefile | 4 ++++
> >  1 file changed, 4 insertions(+)
> > 
> > diff --git a/lib/librte_acl/Makefile b/lib/librte_acl/Makefile
> > index 7a1cf8a..401fb8c 100644
> > --- a/lib/librte_acl/Makefile
> > +++ b/lib/librte_acl/Makefile
> > @@ -50,7 +50,11 @@ SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_gen.c
> >  SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_scalar.c
> >  SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_sse.c
> > 
> > +CC_SSE4_1_SUPPORT := $(shell $(CC) -msse4.1 -dM -E - < /dev/null >/dev/null 2>&1 && echo 1)
> > +
> > +ifeq ($(CC_SSE4_1_SUPPORT),1)
> >  CFLAGS_acl_run_sse.o += -msse4.1
> > +endif  
> 
> I don't think acl_run_sse.c would compile if SSE4_1 is not supported.
> So, I think you need to do same thing, as is done for AVX2:
> Compile in acl_run_sse.c  only if SSE41 is supported by the compiler:  
> 
> - SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_sse.c
> -
> -CFLAGS_acl_run_sse.o += -msse4.1
> 
> +CC_SSE41_SUPPORT=$(shell $(CC) -msse4.1 -dM -E - </dev/null 2>&1 | \
> grep -q  && echo 1)
> +ifeq ($(CC_SSE41_SUPPORT), 1)
> +        SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_sse.c
> +        CFLAGS_rte_acl.o += -DCC_SSE41_SUPPORT
> +        CFLAGS_acl_run_sse.o += -msse4.1	
> +endif
> 
> And then change rte_acl_init() accordingly.
> Something like:
> 
> int __attribute__ ((weak))
> rte_acl_classify_sse(__rte_unused const struct rte_acl_ctx *ctx,
>         __rte_unused const uint8_t **data,
>         __rte_unused uint32_t *results,
>         __rte_unused uint32_t num,
>         __rte_unused uint32_t categories)
> {
>         return -ENOTSUP;
> }
> 
> ....
> 
> static void __attribute__((constructor))
> rte_acl_init(void)
> {
>         enum rte_acl_classify_alg alg = RTE_ACL_CLASSIFY_DEFAULT;
> 
> #if defined(CC_AVX2_SUPPORT)
>         if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2))
>                 alg = RTE_ACL_CLASSIFY_AVX2;
>         else if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_SSE4_1))
> #elif defined (CC_SSE41_SUPPORT)
>         if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_SSE4_1))
>                 alg = RTE_ACL_CLASSIFY_SSE;
> #endif
> 
>         rte_acl_set_default_classify(alg);
> }
> 
> After that, I suppose, you should be able to build and (probably use) librte_acl on arm.
> Konstantin
> 
> 
>