Re: [dpdk-dev] [PATCH v1 1/4] x86: change cpuflag macros to compiler macros

[Bruce Richardson <bruce.richardson@intel.com>]

On Wed, Sep 02, 2020 at 10:43:40AM +0000, Radu Nicolau wrote:
> Replace use of RTE_MACHINE_CPUFLAG macros with regular compiler
> macros.
> 

I think it's worth noting in the commit log that the set of macros provided
by the compilers are more complete than those provided by DPDK, and by not
having our own it allows new instruction sets to be leveraged without
having to do extra work to set them up in DPDK.

> Signed-off-by: Sean Morrissey <sean.morrissey@intel.com>
> Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
> ---
>  app/test/test_memcpy_perf.c             |  8 ++++----
>  config/x86/meson.build                  |  2 --
>  drivers/net/enic/Makefile               |  2 +-
>  drivers/net/enic/meson.build            |  2 +-
>  drivers/net/i40e/Makefile               |  2 +-
>  drivers/net/i40e/meson.build            |  2 +-
>  drivers/net/iavf/Makefile               |  2 +-
>  drivers/net/iavf/meson.build            |  2 +-
>  drivers/net/ice/Makefile                |  2 +-
>  drivers/net/ice/meson.build             |  2 +-
>  examples/l3fwd/l3fwd_em.c               |  4 ++--
>  lib/librte_acl/Makefile                 |  2 +-
>  lib/librte_acl/meson.build              |  2 +-
>  lib/librte_eal/common/rte_random.c      |  4 ++--
>  lib/librte_eal/x86/include/rte_memcpy.h |  8 ++++----
>  lib/librte_efd/rte_efd_x86.h            |  2 +-
>  lib/librte_hash/rte_cuckoo_hash.c       |  2 +-
>  lib/librte_member/rte_member_ht.c       | 10 +++++-----
>  lib/librte_member/rte_member_x86.h      |  2 +-
>  lib/librte_net/rte_net_crc.c            |  2 +-
>  mk/rte.cpuflags.mk                      |  1 -
>  21 files changed, 31 insertions(+), 34 deletions(-)
> 
<snip>
> @@ -231,7 +231,7 @@ em_mask_key(void *key, xmm_t mask)
>  
>  	return vandq_s32(data, mask);
>  }
> -#elif defined(RTE_MACHINE_CPUFLAG_ALTIVEC)
> +#elif defined(__ALTIVEC__)

Not an x86 flag. Belongs in patch 3.

>  static inline xmm_t
>  em_mask_key(void *key, xmm_t mask)
>  {
> diff --git a/lib/librte_acl/Makefile b/lib/librte_acl/Makefile
> index f4332b044..3b591c2ed 100644
> --- a/lib/librte_acl/Makefile
> +++ b/lib/librte_acl/Makefile
> @@ -38,7 +38,7 @@ endif
>  #
>  
>  #check if flag for AVX2 is already on, if not set it up manually
> -ifeq ($(findstring RTE_MACHINE_CPUFLAG_AVX2,$(CFLAGS)),RTE_MACHINE_CPUFLAG_AVX2)
> +ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),)
>  	CC_AVX2_SUPPORT=1
>  else
>  	CC_AVX2_SUPPORT=\
> diff --git a/lib/librte_acl/meson.build b/lib/librte_acl/meson.build
> index d1e2c184c..b31a3f798 100644
> --- a/lib/librte_acl/meson.build
> +++ b/lib/librte_acl/meson.build
> @@ -15,7 +15,7 @@ if dpdk_conf.has('RTE_ARCH_X86')
>  	# in former case, just add avx2 C file to files list
>  	# in latter case, compile c file to static lib, using correct compiler
>  	# flags, and then have the .o file from static lib linked into main lib.
> -	if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2')
> +	if cc.get_define('__AVX2__', args: machine_args) != ''

Since this is used in a number of places, we probably should just get the
result in a variable in config/x86/meson.build.

>  		sources += files('acl_run_avx2.c')
>  		cflags += '-DCC_AVX2_SUPPORT'
>  	elif cc.has_argument('-mavx2')
<snip>