On 2021/5/19 20:34, David Marchand wrote:
> On Wed, May 19, 2021 at 1:57 PM Chengwen Feng <fengcheng...@huawei.com> wrote:
>>
>> Compile error with gcc8.3 and crossfile arm64_thunderx2_linux_gcc:
>> ../drivers/net/hns3/hns3_rxtx_vec_sve.c
>> cc1: error: switch ‘-mcpu=armv8.1-a’ conflicts with
>> ‘-march=armv8.2-a’ switch [-Werror]
>> ../drivers/net/hns3/hns3_rxtx_vec_sve.c:5:10: fatal error:
>> arm_sve.h: No such file or directory
>> 5 | #include <arm_sve.h>
>>
>> The root cause is that gcc8.3 supports SVE, but it doesn't support
>> compile ACLE[1] SVE code, and the hns3_rxtx_vec_sve.c was written
>> by ACLE SVE code.
>
> Quite confusing.
> What is the difference between the two?
>
Only SVE assembly instructions are supported in gcc8/9
And the ACLE SVE code like normal C code.
There are also ACLE NEON code which many driver already supports.
>
>>
>> This patch also filters out '-march=' '-mcpu' '-mtune' when compile
>> with hns3_rxtx_vec_sve.c.
>>
>> [1] ACLE: Arm C Language Extensions, user should include <arm_sve.h>
>> when writing ACLE SVE code.
>>
>> Fixes: 203fbaf8813d ("net/hns3: refactor SVE code compile method")
>>
>> Signed-off-by: Chengwen Feng <fengcheng...@huawei.com>
>> ---
>> v2:
>> * fix check tool error
>> ---
>> drivers/net/hns3/meson.build | 14 ++++++++++----
>> 1 file changed, 10 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/net/hns3/meson.build b/drivers/net/hns3/meson.build
>> index 8563d70..5f9af9b 100644
>> --- a/drivers/net/hns3/meson.build
>> +++ b/drivers/net/hns3/meson.build
>> @@ -39,16 +39,22 @@ if arch_subdir == 'arm' and dpdk_conf.get('RTE_ARCH_64')
>> # compile SVE when:
>> # a. support SVE in minimum instruction set baseline
>> # b. it's not minimum instruction set, but compiler support
>> - if cc.get_define('__ARM_FEATURE_SVE', args: machine_args) != ''
>> + if cc.get_define('__ARM_FEATURE_SVE', args: machine_args) != '' and
>> cc.check_header('arm_sve.h')
>
> The rte_vect.h arm header unconditionally includes arm_sve.h if
> __ARM_FEATURE_SVE is defined [1].
> So it is surprising to see such a check being added.
>
> 1: https://git.dpdk.org/dpdk/tree/lib/eal/arm/include/rte_vect.h#n12
>
Yes, it is, I will fix in 21.08
>
