[llvm] r183238 - ARM: Fix crash in ARM backend inside of ARMConstantIslandPass

[James Molloy]

> That being said 8-bit constant pool entries should be supported in some
fashion, not cause a crash. I just think it shouldn't get emitted :-)

The architecture in the test case provided was armv4t, which does not have
MOVW/MOVT.

Also, neither does v6m/v7m, so we do need to support LDRB.

Cheers,

James

On 5 June 2013 05:05, JF Bastien <jfb at google.com> wrote:

> > Presumably for the same reason why we use ldr for 'ptrtoint (i32* @a to
> i32)' instead of movw/movt. Why do you ask? Do you think using ldrb is
> incorrect?
>
> I do think that it is incorrect unless the critical path is extremely
> integer execution unit bound, and the memory side is completely free. I
> think that it's better to always use immediate mov for 8-bit values.
>
> Things are different for 16, 32 and 64 bit constants, it varies quite a
> bit depending in which processor is measured and on which workload.
>
> That being said 8-bit constant pool entries should be supported in some
> fashion, not cause a crash. I just think it shouldn't get emitted :-)
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130605/80f70f87/attachment.html>