514624 - UTF8 to Unicode conversion of SSE2 optimization for x86_64 platform

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Makoto Kato [:m_kato]]

Although bug 506430 is landed, this is for x86 only.  We should enable it for x86_64 (x86_64 supports SSE2 as default!)

Comment 1

[Makoto Kato [:m_kato]]

Attachment: patch v1

in testing...

Comment 2

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Does OSX actually specify -msse2 by default?

Comment 3

[Makoto Kato [:m_kato]]

(In reply to comment #2)
> Does OSX actually specify -msse2 by default?

Yes.

In http://developer.apple.com/hardwaredrivers/ve/sse.html

SSE2 is enabled using the GCC compiler flag -msse2. SSE2 is enabled by default on gcc-4.0. If SSE2 is enabled, the C preprocessor symbol __SSE2__ is defined.

Comment 4

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Hmm.  So I've been wondering about that for a while.  Here's gcc -v (gcc 4.0.1):

 /usr/libexec/gcc/i686-apple-darwin9/4.0.1/cc1 -quiet -v -D__DYNAMIC__ f.c -fPIC -quiet -dumpbase f.c -mmacosx-version-min=10.5.8 -mtune=generic -march=apple -auxbase f -O2 -version -o /var/folders/5M/5MNK73rKEWemghU5EyInjU+++TI/-Tmp-//ccbbwavM.s

and gcc -dumpspecs | grep sse does't show anything.  So I guess they're enabling sse via the 'apple' arch?

Comment 5

[Makoto Kato [:m_kato]]

(In reply to comment #4)
> Hmm.  So I've been wondering about that for a while.  Here's gcc -v (gcc
> 4.0.1):
> 
>  /usr/libexec/gcc/i686-apple-darwin9/4.0.1/cc1 -quiet -v -D__DYNAMIC__ f.c
> -fPIC -quiet -dumpbase f.c -mmacosx-version-min=10.5.8 -mtune=generic
> -march=apple -auxbase f -O2 -version -o
> /var/folders/5M/5MNK73rKEWemghU5EyInjU+++TI/-Tmp-//ccbbwavM.s
> 
> and gcc -dumpspecs | grep sse does't show anything.  So I guess they're
> enabling sse via the 'apple' arch?

Maybe, you are right.  But, I don't know gcc source code that apple modifies.  Since apple hardware is Core Duo or later, I think sse2 is turned by default on their gcc.

Also, if __SSE2__ isn't defined, gcc cannot compile built-in functions for SSE2.  emmintrin.h on gcc (even if apple version) expects that __SSE2__ is defined.

Comment 6

[Makoto Kato [:m_kato]]

http://hg.mozilla.org/mozilla-central/rev/d6e21bdfd014

Comment 7

[Chris Lord [:cwiiis]]

It seems this change enabled this code on Intel Atom, where it segfaults on the line 'out2 = _mm_unpackhi_epi8(in, zeroes);'

SSE2 is supported on Atom, should this work or is it relying on other features too?

Comment 8

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Hmm, I'm running these builds fine on an Atom, though on Windows -- are you on Linux?  If so, are you on a kernel that has SSE2 support disabled or something?

Comment 9

[Chris Lord [:cwiiis]]

I am on Linux, but SSE2 is enabled (or should be). I'll try and look into this further and report back.

Comment 10

[Chris Lord [:cwiiis]]

A little bit of extra info (though I've still not looked into this properly yet), we're using the -m flags "-m32 -march=core2 -msse3 -mtune=generic -mfpmath=sse". That -march part doesn't look right to me...

Comment 11

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

I -think- arch=core2 is valid for Atom... it's the same ISA, no?  We should probably move this to a new bug.

Comment 12

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

(In reply to comment #11)
> I -think- arch=core2 is valid for Atom... it's the same ISA, no?  We should
> probably move this to a new bug.

Yeah Intel's literature says they have the same ISA.

Comment 13

[Makoto Kato [:m_kato]]

(In reply to comment #10)
> A little bit of extra info (though I've still not looked into this properly
> yet), we're using the -m flags "-m32 -march=core2 -msse3 -mtune=generic
> -mfpmath=sse". That -march part doesn't look right to me...

Chris, What's version of gcc?