[M3devel] question re LONGINT

[Tony Hosking]

On Mar 14, 2011, at 6:33 PM, Jay K wrote:

>  - That error is likely from the frontend and will likely occur for any target. Tony's reply implies this as well.
>  
>  - LONGINT is believed to fully work on NT386. But more testing is welcome.
>  
>  - (atomics ditto -- believed to work, testing welcome -- and we should probably use them some, e.g. in MUTEX initialization)

They are still slightly broken.  I have some fixes still to check in.  If only I could find some time to test everything...
But, they will be essential if we ever do cooperative thread suspension.

>   - I would be curious to see if threadtest only fails on 64bit Windows. I haven't tried 32bit yet but I will soon.
>    - I was thinking more about this WOW64 GetThreadContext problem and ways to mitigate it, other than cooperative suspend.
>      I had already suggested and started to implement scanning the entire stack -- i.e. not depending on an ability to get ESP from a suspended thread.
>      Like how FreeBSD and OpenBSD work, albeit it is a significant deoptimization.
>      Regarding the rest of the context, perhaps it is a reasonable assumption for now to declare that it doesn't matter? e.g. to know/assume
>      that m3back never leaves a value only in registers, that anything in a register will also be on the stack? And to disable optimization on the
>      hypothetical MINGW platforms. ?
>      But still to confirm if there are no problems on 32bit, and to mull over implementing cooperative suspend..in the little time available...
>      (Oh, and also, I wonder if SuspendThread can suspend a thread while in the kernel, probably, and if GetThreadContext returns
>      the kernel context in that case; should be easy to determine via some experimentation -- which would imply a problem on 32bit Windows also.)
>  
>  
>  - Jay
>  
> From: hosking at cs.purdue.edu
> Date: Mon, 14 Mar 2011 17:21:01 -0400
> To: rcolebur at SCIRES.COM
> CC: m3devel at elegosoft.com
> Subject: Re: [M3devel] question re LONGINT
> 
> This is a known problem which I am still to fix.  I know what the issue is...
> You can probably make do by reading counts[i] to a variable, incrementing, and assigning.
> 
> On Mar 14, 2011, at 5:14 PM, Coleburn, Randy wrote:
> 
> I was making some local mods to Mika’s threadtest program to help me debug things a bit more on Windows.
>  
> I am trying to do:
>    INC(counts[i]);
> Where counts is REF ARRAY OF LONGINT
>  
> The compiler is complaining:
> "..\src\Main.m3", line 189: ** INTERNAL CG ERROR *** unable to find integer type?  type=Int.64  s/o/a=64/0/32
>  
> I’m not sure what the status of LONGINT is on Windows.  If it is not fully implemented, let me know and I’ll switch to REF ARRAY OF INTEGER, but I really need LONGINT.
>  
> Regards,
> Randy Coleburn
>  
> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://m3lists.elegosoft.com/pipermail/m3devel/attachments/20110314/552f4ee8/attachment-0002.html>