[llvm] r239071 - [bpf] add big- and host- endian support

[Alexei Starovoitov]

On Thu, Jun 4, 2015 at 2:17 PM, Quentin Colombet <qcolombet at apple.com> wrote:
>
>> On Jun 4, 2015, at 2:12 PM, Alexei Starovoitov <alexei.starovoitov at gmail.com> wrote:
>>
>> On Thu, Jun 4, 2015 at 2:09 PM, Quentin Colombet <qcolombet at apple.com> wrote:
>>>
>>> On Jun 4, 2015, at 1:53 PM, Alexei Starovoitov
>>> <alexei.starovoitov at gmail.com> wrote:
>>>
>>> On Thu, Jun 4, 2015 at 12:24 PM, Eric Christopher <echristo at gmail.com>
>>> wrote:
>>>
>>> It doesn't look like anyone actually approved this? (I know, it was in my
>>> list of things to look at :)
>>>
>>>
>>> nope. It started to bit rot too quickly, so I pushed it.
>>>
>>>
>>> Please don’t do that. Unless I am mistaken, the review was out only for a
>>> couple of days.
>>> Here is the review policy, if you want more information.
>>> http://llvm.org/docs/DeveloperPolicy.html#code-reviews
>>
>> hmm, I'm exactly doing the following part:
>> "(or changes where the developer owns the component) can be reviewed
>> after commit."
>
> Well, in that case, don’t post a review request in the first place :).

:)
well I mainly wanted to get feedback for logic that
-march=bpf does _host_ endian.
I don't think any other backend does things like that.
What the patch does:
-march=bpf -> host endian
-march=bpf_le -> little endian
-march=bpf_be -> big endian

the typical scenario is to compile it on the host,
load into the kernel and run it there.
So from ease-of-use point of view it's the best
(this behavior is specifically what s390 folks have asked)
but since no other backend does host-endian,
I wanted to double check and/or argue about it :)

For llvm+bpf backend users it's easier to git pull from main
tree instead of applying my one-of patches that I'm sending
them over email and they bit-rot too quickly and fail to apply.