armv7 vs neon

[Antoine Labour]

If you don't care about ARM, you can stop reading now.

I checked in a change that enables separation between ARMv7 and NEON. NEON is not supported by all ARMv7 chips. By default, NEON is still turned on (if armv7=1 is set), but it can be turned off by setting arm_neon=0 in your GYP_DEFINES.

I updated http://code.google.com/p/chromium/wiki/LinuxChromiumArm

Antoine

[Antoine Labour]

Follow-up: if NEON is off, it defaults to vfpv3, but some architectures don't support that, so I added a way to override it with the arm_fpu gyp define. Updated the doc.

Antoine

[Antoine Labour]

On Fri, Mar 5, 2010 at 7:48 AM, Stephen White <senor...@google.com> wrote:

Hi Antoine,

Seems like this is going to mean a lot of different ARM binaries.  Out of curiosity, would it be worthwhile to make some of these run-time checks?  I had to do that for SSE2, so it shouldn't be too hard to do it for neon too.

Stephen

For Chromium OS, we're already committed to having different binaries for different ARM platforms (we do it at the distribution level), so it's not an extra burden. We could use the extra registers and performance boost of using vector instructions when the compiler can do it, even for generic code.

I totally see the advantage of doing run-time checks for distributions, so that they can compile with safe flags, and only enable NEON for specifically optimized code. Mainly skia and ffmpeg are concerned I believe. I haven't done any work to do that yet though.

Antoine

 

Antoine

--
Chromium Developers mailing list: chromi...@chromium.org
View archives, change email options, or unsubscribe:
http://groups.google.com/a/chromium.org/group/chromium-dev

--
All truth passes through three stages. First, it is ridiculed. Second, it is violently opposed. Third, it is accepted as being self-evident. -- Schopenhauer