Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 24 Aug 2018 15:50:10 -0700
From: Andre McCurdy <armccurdy@...il.com>
To: musl@...ts.openwall.com
Subject: Re: [PATCH] fix build failure on arm because of missing clz instruction

On Fri, Aug 24, 2018 at 3:05 PM, Szabolcs Nagy <nsz@...t70.net> wrote:
> * Andre McCurdy <armccurdy@...il.com> [2018-08-24 14:58:04 -0700]:
>> On Fri, Aug 24, 2018 at 12:30 PM, Szabolcs Nagy <nsz@...t70.net> wrote:
>> > another arm patch, clz usage (in fma) was broken with -mthumb -march=armv5t.
>>
>> That conditional was originally written under the assumption that musl
>> doesn't support thumb1 (so -mthumb -march=armv5t is not a supported
>> configuration).
>>
>> Was that assumption wrong?
>
> the asm code in musl does not support thumb1,
> but you can compile everything else with thumb1
> (the compiler does not pass -mthumb to the assembler
> by default so all asm code will be in arm mode)
>
> so with the changed condition the build succeeds
> with -mthumb -march=armv5t, but the resulting libc
> will not be competely thumb code, i think that's
> an improvement.

Yes, I agree it's an improvement. Mostly thumb1 with a little ARM is
still useful for a lot of use cases.

As an aside, it seems that glibc unconditionally includes some thumb2
when targeting ARMv7, so there's precedent for mixing ARM and
thumb[12] in libc.

  https://sourceware.org/bugzilla/show_bug.cgi?id=23031#c15

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.