On 10/26/2020 7:12 AM, Rong, Leyi wrote:
-----Original Message-----
From: David Marchand <david.march...@redhat.com>
Sent: Monday, October 26, 2020 12:24 AM
To: Rong, Leyi <leyi.r...@intel.com>; Zhang, Qi Z <qi.z.zh...@intel.com>; Yigit,
Ferruh <ferruh.yi...@intel.com>
Cc: Lu, Wenzhuo <wenzhuo...@intel.com>; dev <dev@dpdk.org>; Richardson,
Bruce <bruce.richard...@intel.com>; Thomas Monjalon
<tho...@monjalon.net>
Subject: Re: [dpdk-dev] [PATCH v4 1/3] net/ice: add AVX512 vector path
Hello Leyi, Qi, Ferruh,
On Fri, Oct 23, 2020 at 6:37 AM Leyi Rong <leyi.r...@intel.com> wrote:
diff --git a/drivers/net/ice/meson.build b/drivers/net/ice/meson.build
index 254595af85..85e8baf912 100644
--- a/drivers/net/ice/meson.build
+++ b/drivers/net/ice/meson.build
@@ -34,6 +34,17 @@ if arch_subdir == 'x86'
c_args: [cflags, '-mavx2'])
objs += ice_avx2_lib.extract_objects('ice_rxtx_vec_avx2.c')
endif
+
+ if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX512F') or (not
machine_args.contains('-mno-avx512f') and cc.has_argument('-mavx512f'))
+ cflags += ['-DCC_AVX512_SUPPORT']
+ ice_avx512_lib = static_library('ice_avx512_lib',
+ 'ice_rxtx_vec_avx512.c',
+ dependencies: [static_rte_ethdev,
+ static_rte_kvargs, static_rte_hash],
+ include_directories: includes,
+ c_args: [cflags, '-march=skylake-avx512',
'-mavx512f'])
+ objs += ice_avx512_lib.extract_objects('ice_rxtx_vec_avx512.c')
+ endif
endif
sources += files('ice_dcf.c',
--
2.17.1
RTE_MACHINE_CPUFLAG_AVX512F can be removed.
Worth fixing before hitting the main tree.
Thanks.
--
David Marchand
Hello David,
Would prefer using __AVX512F__ instead of RTE_MACHINE_CPUFLAG_AVX512F here
rather than remove the RTE_MACHINE_CPUFLAG_ macro directly to check the CPU
capability.
So the judgment statement will be
if cc.get_define('__AVX512F__', args: machine_args) != '' or (not
machine_args.contains('-mno-avx512f') and cc.has_argument('-mavx512f'))
what do you think?
Hello Ferruh,
As the patchset is already merged into dpdk-next-net, I'm going to make another
patch for this if it's accepted?
Hi Leyi,
Please make another patch, I can squash it in the next-net.
Thanks,
ferruh