Re: [Gems-users] Sparc V9 workloads in LogTM (GEMS 2.1)


Date: Wed, 4 Jun 2008 10:21:30 +0200
From: "Enrique Vallejo" <enrique@xxxxxxxxxxxxx>
Subject: Re: [Gems-users] Sparc V9 workloads in LogTM (GEMS 2.1)

Thank you Jayaram!

 

      Thanks to your good pointing, I found that effectively v8plusa supports the V9 ISA, but there are some restrictions (see http://sunsolve.sun.com/search/document.do?assetkey=1-31-41020-1&attach=yes, section 2.1.1) regarding the V9 instructions allowed in v8plus. Specifically, my problem was that CASXA only works on global or out registers. Using any other register generates an "Invalid instruction" crash at runtime. I'll try to find an alternative on the hand-tuned assembler.

 

      Best regards,

 

Enrique Vallejo

University of Cantabria, Spain

 

-----Mensaje original-----
De: gems-users-bounces@xxxxxxxxxxx [mailto:gems-users-bounces@xxxxxxxxxxx] En nombre de Jayaram Bobba
Enviado el: miércoles, 04 de junio de 2008 6:50
Para: Gems Users
Asunto: Re: [Gems-users] Sparc V9 workloads in LogTM (GEMS 2.1)

 

 

I believe v8plusa supports all 32-bit instructions within V9 ISA.

http://docs.sun.com/app/docs/doc/819-5265/bjass?a=view

 

Are you trying to compile a 64-bit executable? I dont think there is

a quick easy way to port LogTM implementation in GEMS 2.1 to support

64-bit executables. You would have to change the undo logging in

TransactionVersionManager.C and might need to modify the trap interfaces

in TransactionInterfaceManager.C. In addition, you will have to alter

the address masks in transaction.h.

 

Jayaram

 

 

Enrique Vallejo wrote:

> 

> Hi all,

> 

> I would like to run some workloads in LogTM (GEMS 2.1) using specific

> Sparc-V9 instructions (such as CAS). However, the compilation Makefile

> generates, by default, V8 code (-xarch=v8plusa flag in

> /common/Makefile) and LogTM structures, such as threadTransContext in

> transaction.h, have to be compiled with the specific type sizes of V8,

> so that the LogTM code in the Ruby uses the same offsets, etc.

> 

> I don’t know if this issue might be solved with some compilation

> options, or the LogTM code has to be modified. However, I am afraid of

> doing that, because debugging the changes in Ruby without deep

> knowledge of LogTM in GEMS 2.1 seems a difficult task for me. Does

> anyone have a changes patch for LogTM to use V9 code, or any other

> solution for these issues? Any help would be greatly appreciated.

> 

> Best regards,

> 

> Enrique Vallejo

> 

> University of Cantabria, Spain

> 

> ------------------------------------------------------------------------

> 

> _______________________________________________

> Gems-users mailing list

> Gems-users@xxxxxxxxxxx

> https://lists.cs.wisc.edu/mailman/listinfo/gems-users

> Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.

> 

>  

_______________________________________________

Gems-users mailing list

Gems-users@xxxxxxxxxxx

https://lists.cs.wisc.edu/mailman/listinfo/gems-users

Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.

 

[← Prev in Thread] Current Thread [Next in Thread→]