[cfe-commits] Patch for bug 14744

Chandler Carruth chandlerc at google.com
Tue Jan 1 19:29:30 PST 2013


On Tue, Jan 1, 2013 at 4:55 PM, Krzysztof Parzyszek <kparzysz at codeaurora.org
> wrote:

> On 1/1/2013 6:49 PM, David Blaikie wrote:
>
>>
>> Is the hexagon backend not enabled by default? Does it fail even when
>> the Hexagon backend is built in but the test suite is run on
>> non-Hexagon?
>>
>>
> I think it is enabled by default (i.e. it will be built if no specific
> targets were given at the configure step).  I believe that you actually
> have to try to compile something using it for this problem to happen.  I
> usually only build the Hexagon backend, so I don't know if it would show
> after merely having it built (I suspect not).  I can try it tomorrow.


The standard development practice is to build all of the backends. This is
(in theory) to allow developers to refactor common parts of the toolchain
while building and testing the various backends so that they don't break
something.

For this to work, we need regression tests in the suite which will exercise
all of the various backends and targets such that developers actually have
this confidence. It is very likely that non-hexagon developers will have to
change the hexagon backend or parts of either Clang or LLVM that impact the
hexagon backend. Those developers need to have some reasonable confidence
they haven't broken things.

-Chandler
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130101/222aaa8d/attachment.html>


More information about the cfe-commits mailing list