[Qemu-devel] [ARM] Contributing tests for Neon

[Qemu-devel] [ARM] Contributing tests for Neon

[Christophe Lyon]

Hi all,

I have developed some tests for ARM-Neon in the form of C sources files calling ARM Neon intrinsics, and comparing the results of the resulting program with a known reference (eg execution on actual CPU) shows if the execution engine is follows the spec.

These tests currently represent 750KB of sources (150 files, 12000 lines), and I would like to contribute them to the qemu testing infrastructure.

I have a few questions on how to proceed:
- we wish to release the files under the MIT license, is it OK to deliver them as-is in the 'tests' qemu subdir?
- given the size of the whole stuff, I don't think it's suitable that I send it on the list for review, what should I do?
- at the time of writing, GCC/ARM fails to compile those tests correctly, and I rely on ARM's compiler to generate my reference executable and output. Is it acceptable that the Makefile references armcc?

Thanks,

Christophe.

Re: [Qemu-devel] [ARM] Contributing tests for Neon

[Peter Maydell]

On 21 January 2011 10:07, Christophe Lyon <christophe.lyon@st.com> wrote:
> I have developed some tests for ARM-Neon in the form of C sources files
> calling ARM Neon intrinsics, and comparing the results of the resulting
> program with a known reference (eg execution on actual CPU) shows
> if the execution engine is follows the spec.
>
> These tests currently represent 750KB of sources (150 files, 12000 lines),

That's a larger and more comprehensive test suite than I was
expecting :-)  (which is fantastic)

> I have a few questions on how to proceed:
> - we wish to release the files under the MIT license, is it OK to deliver
> them as-is in the 'tests' qemu subdir?
> - given the size of the whole stuff, I don't think it's suitable that I
> send it on the list for review, what should I do?

How about you make it available somewhere (tarball via http, git
repository on gitorious, or other method of your choice) for the
moment? Then we can take a look at it and proceed from there.

> - at the time of writing, GCC/ARM fails to compile those tests correctly,
> and I rely on ARM's compiler to generate my reference executable and
> output. Is it acceptable that the Makefile references armcc?

This is gcc bugs rather than the test sources using armcc-specific
syntax, right?

-- PMM

Re: [Qemu-devel] [ARM] Contributing tests for Neon

[Edgar E. Iglesias]

On Fri, Jan 21, 2011 at 11:25:05PM +0000, Peter Maydell wrote:
> On 21 January 2011 10:07, Christophe Lyon <christophe.lyon@st.com> wrote:
> > I have developed some tests for ARM-Neon in the form of C sources files
> > calling ARM Neon intrinsics, and comparing the results of the resulting
> > program with a known reference (eg execution on actual CPU) shows
> > if the execution engine is follows the spec.
> >
> > These tests currently represent 750KB of sources (150 files, 12000 lines),
> 
> That's a larger and more comprehensive test suite than I was
> expecting :-)  (which is fantastic)
> 
> > I have a few questions on how to proceed:
> > - we wish to release the files under the MIT license, is it OK to deliver
> > them as-is in the 'tests' qemu subdir?
> > - given the size of the whole stuff, I don't think it's suitable that I
> > send it on the list for review, what should I do?
> 
> How about you make it available somewhere (tarball via http, git
> repository on gitorious, or other method of your choice) for the
> moment? Then we can take a look at it and proceed from there.

I agree, this is a good first step. If possible, its good if you
can publish source and binaries so people with the appropriate
tools can rebuild the tests. Also, when possible, binaries with
debug info can be helpful when analyzing test failures.

Cheers

Re: [Qemu-devel] [ARM] Contributing tests for Neon

[Christophe Lyon]

>> How about you make it available somewhere (tarball via http, git
>> repository on gitorious, or other method of your choice) for the
>> moment? Then we can take a look at it and proceed from there.
> 
> I agree, this is a good first step. If possible, its good if you
> can publish source and binaries so people with the appropriate
> tools can rebuild the tests. Also, when possible, binaries with
> debug info can be helpful when analyzing test failures.
> 

OK, I am going to have a look at gitorious. In the mean time, I have a bit of cleanup and minimal documentation to perform.

Christophe.

Re: [Qemu-devel] [ARM] Contributing tests for Neon

[Christophe Lyon]

On 24.01.2011 11:23, Christophe Lyon wrote:
> OK, I am going to have a look at gitorious. In the mean time, I have a bit of cleanup and minimal documentation to perform.
> 

Hi,
You can have a look at http://gitorious.org/arm-neon-tests/arm-neon-tests.
For sure it will need some polishing, but it's already functional and handy enough for me to continue chasing bugs in the ARM/Neon emulation of QEmu.

Comments welcome.

Christophe.