Fix arm compile problem on Lucid.

Fix arm compile problem on Lucid.

Patch from Jeff Bailey <jeffbailey@google.com>

These functions on ARM both are doing something weird where, when
optimization is not enable, there aren't enough low registers available for
the compiler to handle the asm.

My ARM-fu is pretty weak, so I have forced the functions to -O1, which
allows them to compile.  It would be reasonable to set them to O2 as I
expect anyone debugging this will just tweak it as appropriate.

The error that this is solving:

third_party/skia/src/opts/SkBitmapProcState_opts_arm.cpp: In function 'void
SI8_D16_nofilter_DX_arm(const SkBitmapProcState&, const uint32_t*, int,
uint16_t*)':
third_party/skia/src/opts/SkBitmapProcState_opts_arm.cpp:101:24: error:
can't find a register in class 'LO_REGS' while reloading 'asm'
third_party/skia/src/opts/SkBitmapProcState_opts_arm.cpp:101:24: error:
'asm' operand has impossible constraints

However, it has to be set on both functions, as after the first error is
cleared, it triggers a second time.

Committed: http://code.google.com/p/skia/source/detail?r=1228

[Steve VanDeBogart, 2011-04-23 00:43:21]

Anyone with arm assembly-foo want to comment?

[reed1, 2011-04-25 13:16:14]

Are we seeing any problems building on other arm environments? i.e. can I
reproduce this?

[jeffbailey, 2011-04-25 13:45:25]

Use g++ 4.5 for ARM with optimization disabled.
On Apr 25, 2011 6:16 AM, <reed@google.com> wrote:
> Are we seeing any problems building on other arm environments? i.e. can
> I reproduce this?
>
> http://codereview.appspot.com/4452044/

[Steve VanDeBogart, 2011-04-26 16:34:26]

Any other comments on this?  I think Jeff is waiting on a resolution to this
issue to update a bot.

[reed1, 2011-04-26 17:46:22]

pending testing on android builds... LGTM

[jeffbailey, 2011-04-26 22:07:32]

I don't know how to do that.  Do you have a setup that it can be patched
into, or can you tell me the best way to emulate that setup?

Thanks!

Jeff Bailey >|< Google, Inc.


On Tue, Apr 26, 2011 at 10:46 AM, <reed@google.com> wrote:

> pending testing on android builds... LGTM
>
>
> http://codereview.appspot.com/4452044/
>

[reed1, 2011-04-27 11:51:07]

Derek, can you chime in?

[DerekS, 2011-04-27 14:23:13]

I didn't realize that we were even compiling the ARM files using the standard
Lucid toolchain, the only consumer of those files that I was aware of was
Android.

The only way that I know to test ARM code is by running it on an android device
compiled with the android toolchain. To further complicate the matter android
hasn't been taking the most recent ARM opts files from Skia due to some
rendering issues with the latest code.

This type of testing is impossible without access to the android source and a
flashable device.  As I'm probably the only one with those tools I can do the
tests and LGTM it if everything works.

[jeffbailey, 2011-04-27 15:56:04]

If the Android build is compiling at -O2, then setting the opt level to 2
for those functions should result in a bit-for-bit identical function.

This isn't the standard Lucid toolchain, it's a code sourcery cross-compiler
in use on the bot.  However, this bug will come up as soon as the g++-4.6
compiler lands anyway, so it's worth fixing.

If you're considering taking this patch, I would guess it to be safe (once
the opt level is matched with Android) and would be inclined to toss it in
and wait for bug reports.

Jeff Bailey >|< Google, Inc.


On Wed, Apr 27, 2011 at 7:23 AM, <djsollen@google.com> wrote:

> I didn't realize that we were even compiling the ARM files using the
> standard Lucid toolchain, the only consumer of those files that I was
> aware of was Android.
>
> The only way that I know to test ARM code is by running it on an android
> device compiled with the android toolchain. To further complicate the
> matter android hasn't been taking the most recent ARM opts files from
> Skia due to some rendering issues with the latest code.
>
> This type of testing is impossible without access to the android source
> and a flashable device.  As I'm probably the only one with those tools I
> can do the tests and LGTM it if everything works.
>
>
>
> http://codereview.appspot.com/4452044/
>

[Steve VanDeBogart, 2011-05-02 17:17:52]

ping... any progress on this?

djsollen, I way I read things, we are waiting for your ok.  Please correct me if
I'm wrong.

[DerekS, 2011-05-03 02:18:00]

I'm out of the office this week, but I'm ok if you want to go ahead and
submit this. I'll validate it's benefit the next time I do a android merge.

~ Derek

On Mon, May 2, 2011 at 5:17 PM, <vandebo@chromium.org> wrote:

> ping... any progress on this?
>
> djsollen, I way I read things, we are waiting for your ok.  Please
> correct me if I'm wrong.
>
>
> http://codereview.appspot.com/4452044/
>

[Steve VanDeBogart, 2011-05-03 02:23:34]

On 2011/05/03 02:18:00, djsollen wrote:
> I'm out of the office this week, but I'm ok if you want to go ahead and
> submit this. I'll validate it's benefit the next time I do a android merge.
> 
> ~ Derek

Ok, committed.