Re: Mono 4.6.1 FTBFS on mipsel

[Jo Shields <joshield@microsoft.com>]

It seems L-prefix instructions should never be emitted on 32 bit platforms, so it's likely a problem with 32 bit userland on 64 bit kernel, on the new builders. Ppc had a similar issue 

Sent from my iPhone

> On 8 Oct 2016, at 12:42, Aurelien Jarno <aurelien@aurel32.net> wrote:
> 
>> On 2016-10-07 21:04, Jo Shields wrote:
>> Dear MIPS team,
>> 
>> 
>> It seems that between Mono 4.2 and 4.6, Mono stopped building properly on little-endian 32-bit MIPS.
>> 
>> 
>> https://buildd.debian.org/status/fetch.php?pkg=mono&arch=mipsel&ver=4.6.1.3%2Bdfsg-2&stamp=1475869442
> 
> I have just got a quick look at the code triggering the assert:
> 
> |        case OP_LMUL:
> |        case OP_LDIV:
> |        case OP_LDIV_UN:
> |        case OP_LREM:
> |        case OP_LREM_UN:
> |        case OP_LSHL:
> |        case OP_LSHR:
> |        case OP_LSHR_UN:
> |                mono_print_ins (ins);
> |                g_assert_not_reached ();
> 
> For what I understand, it means that one of the above "internal
> intermediate instruction" (sorry don't know the mono name for that) has
> been emitted, and is not yet implemented for the MIPS backend.
> 
> Could you confirm? If it is the case and the only problem, that should
> be relatively easy to implement. I'll give a try in the next days.
> 
> Aurelien
> 
> -- 
> Aurelien Jarno                          GPG: 4096R/1DDD8C9B
> aurelien@aurel32.net                 http://www.aurel32.net