Re: [PATCH 1/1] gcc-multilib-config: make aarch64 support multilib

[Mark Hatle <mark.hatle@windriver.com>]

On 11/5/15 11:35 PM, Khem Raj wrote:
> 
>> On Nov 5, 2015, at 5:43 AM, Mark Hatle <mark.hatle@windriver.com> wrote:
>>
>> On 11/5/15 2:20 AM, Robert Yang wrote:
>>> Fixed:
>>> MACHINE = qemuarm64
>>> require conf/multilib.conf
>>> MULTILIBS = "multilib:lib32"
>>> DEFAULTTUNE_virtclass-multilib-lib32 = "armv7at-neon"
>>>
>>> $ bitbake core-image-minimal -cpopulate_sdk
>>>
>>> WARNING: gcc multilib setup is not supported for TARGET_ARCH=aarch64
>>> WARNING: gcc multilib setup is not supported for TARGET_ARCH=aarch64
>>>
>>> [YOCTO #8638]
>>>
>>> Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
>>> ---
>>> meta/recipes-devtools/gcc/gcc-multilib-config.inc |    7 +++++++
>>> 1 file changed, 7 insertions(+)
>>>
>>> diff --git a/meta/recipes-devtools/gcc/gcc-multilib-config.inc b/meta/recipes-devtools/gcc/gcc-multilib-config.inc
>>> index 1c0a45a..a0a2ac0 100644
>>> --- a/meta/recipes-devtools/gcc/gcc-multilib-config.inc
>>> +++ b/meta/recipes-devtools/gcc/gcc-multilib-config.inc
>>> @@ -29,6 +29,9 @@ python gcc_multilib_setup() {
>>>     bb.utils.remove(build_conf_dir, True)
>>>     ml_globs = ('%s/*/t-linux64' % src_conf_dir,
>>>                 '%s/*/linux64.h' % src_conf_dir,
>>> +                '%s/aarch64/t-aarch64' % src_conf_dir,
>>> +                '%s/aarch64/aarch64.h' % src_conf_dir,
>>> +                '%s/aarch64/aarch64-cores.def' % src_conf_dir,
>>>                 '%s/*/linux.h' % src_conf_dir,
>>>                 '%s/linux.h' % src_conf_dir)
>>>
>>> @@ -130,6 +133,8 @@ python gcc_multilib_setup() {
>>>         'mips64'    : ['gcc/config/mips/t-linux64'],
>>>         'powerpc'   : ['gcc/config/rs6000/t-linux64'],
>>>         'powerpc64' : ['gcc/config/rs6000/t-linux64'],
>>> +        'aarch64'   : ['gcc/config/aarch64/t-aarch64'],
>>> +        'arm'       : ['gcc/config/aarch64/t-aarch64'],
>>>     }
>>>
>>>     gcc_header_config_files = {
>>> @@ -140,6 +145,8 @@ python gcc_multilib_setup() {
>>>         'mips64'    : ['gcc/config/mips/linux.h', 'gcc/config/mips/linux64.h'],
>>>         'powerpc'   : ['gcc/config/rs6000/linux64.h'],
>>>         'powerpc64' : ['gcc/config/rs6000/linux64.h'],
>>> +        'aarch64'   : ['gcc/config/aarch64/aarch64.h'],
>>> +        'arm'       : ['gcc/config/aarch64/aarch64.h'],
>>>     }
>>
>> I'm not sure the above is correct in this case.  As I believe GCC treats aarch64
>> and arm as different architectures unlike MIPS, Power and IA32.
>>
>> In this case, I would expect two specific cross compilers to be generated, one
>> for armv7 and one for aarch64, instead of a combination single compiler that
>> understand both.
>>
> 
> Well not entirely so, gcc can support ilp32 as a mutlilib variant for aarch64
> so this patch is fine although if we should support ilp32 as a variant is another
> question, which folks can chime in and provide feedback if they have use of it.
> but nevertheless this patch should be ok

I've seen no real world need for ilp32 at this point.. but I have seen many
requests to run armv7 code on an aarch64 system.  (Compatibility multilib.)

--Mark

>> So in this case, it might be better to have an exception to the warning
>> message...  (unless I'm wrong, in which case this IS likely a correct fix....)
>>
>> Hopefully Khem and/or Richard will be able to comment as they are more familiar
>> with this code path.
>>
>>>
>>>     libdir32 = 'SYSTEMLIBS_DIR'
>>>
>>
>