build error

[Thomas Wuerthinger]

The particular information we would be interested in:
- Is there a specific limit on the number of registers, and if so, can we assume it is in the range of PTX (i.e., thousands of registers available)?
- Does the performance improve if less registers are used in a kernel?
- Is there a performance cost for register to register moves?

Thanks, thomas

On Feb 20, 2013, at 2:38 PM, "Venkatachalam, Vasanth" <Vasanth.Venkatachalam at amd.com> wrote:

> Hi Thomas,
> 
> The HSAIL spec isn't public yet, so unfortunately I can't talk about specific constraints impacting HSAIL register allocation.
> You may want to look over Norm Reuben's ASDF talk last year at:
> 
> 
> http://www.youtube.com/watch?v=PbkA_GPLJq4
> 
> 
> 
> and another publically available deck giving an early preview
> 
> 
> 
> http://www.slideshare.net/hsafoundation/hsail-final-11junepptx
> 
> However, keep in mind that the spec may have evolved since then.
> 
> Meanwhile, if you can clarify the kinds of detail you would find useful, we can make a note to provide you the information once the spec  goes public.
> 
> Vasanth
> 
> 
> 
> From: Thomas Wuerthinger [mailto:thomas.wuerthinger at oracle.com]
> Sent: Wednesday, February 20, 2013 3:57 PM
> To: Venkatachalam, Vasanth
> Cc: graal-dev at openjdk.java.net
> Subject: Re: build error
> 
> Vasanth,
> 
> I'll provide you with exactly what you describe below. However, I'll still need another 24 hours - will let you know when it is pushed.
> 
> I guess that you'll later want to add a HSAIL implementation. Could you please advise as to what would be your requirements and constraints regarding register allocation for the HSAIL code?
> 
> - thomas
> 
> On Feb 20, 2013, at 9:14 AM, "Venkatachalam, Vasanth" <Vasanth.Venkatachalam at amd.com<mailto:Vasanth.Venkatachalam at amd.com>> wrote:
> 
> 
> Hi Thomas,
> 
> Thanks for the fast response. I checked out the latest copy of the repo and am able to build the code.
> 
> What I would like to do now is run a small Java program (say one that just adds two integers) and see PTX code being emitted for the addition.
> 
> You mention that you're refactoring the backend. Is the code that I've checked out currently in a state where I will be able to do the above?
> If so, can you provide some details on what I need to do to get there?
> 
> Thanks,
> Vasanth
> 
> From: Thomas Wuerthinger [mailto:thomas.wuerthinger at oracle.com<http://oracle.com>]
> Sent: Tuesday, February 19, 2013 5:51 PM
> To: Venkatachalam, Vasanth
> Cc: graal-dev at openjdk.java.net<mailto:graal-dev at openjdk.java.net>
> Subject: Re: build error
> 
> Vasanth,
> 
> I've just pushed a new version to http://hg.openjdk.java.net/graal/graal/ that contains Christian's changes provisionally integrated. You should be able to pull and build. I'm currently working on refactoring our backend to fit the needs of PTX in terms of register allocation and assembly output.
> 
> - thomas
> 
> On Feb 19, 2013, at 3:28 PM, "Venkatachalam, Vasanth" <Vasanth.Venkatachalam at amd.com<mailto:Vasanth.Venkatachalam at amd.com>> wrote:
> 
> 
> 
> Hi,
> 
> I'm trying to build the following patch which adds PTX ISA support to GRAAL.
> 
> http://cr.openjdk.java.net/~twisti/0000000/
> 
> I applied the patch successfully, but when I tried to build it (using mx build) I got errors such as:
> 
> 
> 
> 
> 
> AbstractPTXAssembler.java: 27: error: package com.oracle.graal.lir.ptx
> 
> 
> 
> does not exist
> 
> 
> 
> Import com.oracle.graal.lir.ptx.*
> 
> 
> 
> 
> 
> 
> 
> 
> Apparently the build process is not able to find the above package in question, even though it has been added to the source tree (as a result of applying this patch). It looks like the build scripts may need to be updated so that they know where to find the source file for the above newly added package.
> 
> Can someone explain what configuration change needs to be made here so that I will  be able to do an mx build?
> 
> Vasanth
>