907303 - AudioNodeEngineNEON.cpp doesn't build on GCC 4.4.3

Status: RESOLVED FIXED

(Core :: Web Audio, defect)

Description

[Reuben Morais [:reuben]]

/Users/Reuben/Development/B2G-master/gecko/content/media/AudioNodeEngineNEON.cpp: In function 'void mozilla::AudioBlockPanStereoToStereo_NEON(const float*, const float*, float, float, bool, float*, float*)':
/Users/Reuben/Development/B2G-master/gecko/content/media/AudioNodeEngineNEON.cpp:249: internal compiler error: in change_address_1, at emit-rtl.c:1954

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48561

Comment 1

[Paul Adenot (:padenot)]

jwwang, could you have a look?

Comment 2

[Dave Huseby [:huseby]]

CC'ing ryanvm because the Android build bustage is back.  

This was originally added in m-c revision a232e4df28e5 but was backed out in m-c revision 21dcb96139ee but has since been re-merged in revision 305facc074df and Android build bustage is back.

Comment 3

[Dave Huseby [:huseby]]

Attachment: bustage_fix.patch

Here's a patch for m-c tip that backs out 305facc074df and fixes the Android bustage.

Comment 4

[Ryan VanderMeulen [:RyanVM]]

(In reply to Dave Huseby [:huseby] from comment #2)
> CC'ing ryanvm because the Android build bustage is back.  

I just landed the patch. jwwang is the one who will need to look into this. That said, GCC 4.4.3? Bleh, isn't that really old?

Comment 5

[Dave Huseby [:huseby]]

(In reply to Ryan VanderMeulen [:RyanVM UTC-4] from comment #4)
> I just landed the patch. jwwang is the one who will need to look into this.

Thanks!

> That said, GCC 4.4.3? Bleh, isn't that really old?

Yes, you're right but that's the latest and greatest arm toolchain we use for b2g.

Comment 6

[JW Wang [:jwwang] [:jw_wang]]

I see no build fail on try server: https://tbpl.mozilla.org/?tree=Try&rev=811971fcc280

Btw, the default compiler for B2G is not GCC 4.4.3.
How can I config GCC 4.4.3 to build B2G tree?

Comment 7

[(no longer active)]

(In reply to comment #6)
> Btw, the default compiler for B2G is not GCC 4.4.3.
> How can I config GCC 4.4.3 to build B2G tree?

<https://developer.mozilla.org/en-US/docs/Mozilla/Firefox_OS/Customization_with_the_.userconfig_file#Changing_the_default_host_compiler>

Comment 8

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

This is about the target compiler (→ ARM), not the host compiler (as far as I know).

The FxOS target tools are controlled by TARGET_TOOLS_PREFIX, which is the string to prepend to "gcc" and "ld" and such to get the desired cross-tools.  And they are GCC 4.4.3 by default (unless you're on a Jelly Bean target, but as I understand it those aren't entirely ready for use yet).

Comment 9

[JW Wang [:jwwang] [:jw_wang]]

I am following the steps in https://wiki.mozilla.org/B2G/Profiling#Quick_Start to create a B2G build for profiling where GCC 4.4.3 is used. It turns out to build fine for me. Can you describe the build setup?

Comment 10

[Jed Davis [:jld] ⟨⏰|UTC-7⟩ ⟦he/him⟧]

It looks as if this bug needs optimization to be disabled (B2G_NOOPT=1 in .userconfig) to reproduce it.

There shouldn't be a difference between the toolchain I built for profiling and the official toolchain, for this purpose — they're both based on 4.4.3 with the same set of Android patches — and I'm told that the bug happens with both of them.

Comment 11

[(no longer active)]

I think someone who can reproduce this is probably the best person to fix it!  Please ping me on IRC if you can repro and need help fixing this.

Comment 12

[JW Wang [:jwwang] [:jw_wang]]

Attachment: builderrorfix.patch

Somehow, this patch happens to workaround the bug in GCC 4.4.3.

Comment 13

[JW Wang [:jwwang] [:jw_wang]]

Please check in the patch, builderrorfix.patch (attachment 797714 [details] [diff] [review]), thanks.

Comment 14

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/85ca556d2bcf

Comment 15

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/85ca556d2bcf

Comment 16

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/releases/mozilla-aurora/rev/40582abc4cd2