5 replies to this topic

[Pickle]

link: http://gcc.gnu.org/g....4/changes.html


ARM

• GCC now supports optimizing for the Cortex-A9, Cortex-R4 and Cortex-R4F processors and has many other improvements to optimization for ARM processors.
• GCC now supports the VFPv3 variant with 16 double-precision registers with -mfpu=vfpv3-d16. The option -mfpu=vfp3 has been renamed to -mfpu=vfpv3.
• GCC now supports the -mfix-cortex-m3-ldrd option to work around an erratum on Cortex-M3 processors.
• GCC now supports the __sync_* atomic operations for ARM EABI GNU/Linux.
• The section anchors optimization is now enabled by default when optimizing for ARM.
• GCC now uses a new EABI-compatible profiling interface for EABI targets. This requires a function __gnu_mcount_nc, which is provided by GNU libc versions 2.8 and later.

I find the first line item interesting.

(Maybe we can kick off a discussion to split from the codesourcery toolchain.)

Edited by Pickle, 23 April 2009 - 09:39 PM.

[Laurent]

I find the first line item interesting.

(Maybe we can kick off a discussion to split from the codesourcery toolchain.)

Don''t waste your time: all of these ARM improvements are contributions from CodeSourcery people

EDIT : But perhaps gcc 4.4.0 is better than 4.3.2, though I would not trust an x.y.0 release...

Edited by Laurent, 23 April 2009 - 09:51 PM.

[Archaemic]

Well, it mentions the A9, but not the A8, so it doesn''t sound like a lot of specific stuff to the A8 was changed/fixed.

[arrrgh]

Does this mean that NEON will work properly?

Edited by arrrgh, 23 April 2009 - 11:35 PM.

[darkblu]

good to hear codesourcery code found way upstream.

EDIT : But perhaps gcc 4.4.0 is better than 4.3.2, though I would not trust an x.y.0 release...

my impressions from the pre-release 4.4 (have been following it for about half a year) is that it is much more efficient at expression optimizations and at distinguishing expression invariance. actually, 4.4.0 was the first gcc to ever optimize out the artificial matrix mul test loop from my test suite (as seen in the efficiency cornered thread), so i was forced to introduce non-determinism to the loop.

otherwise, reliability wise, well, it suffices to say that it''s a brand new gcc.. : )

[cb88]

does graphite have anything to do with the improvements ... or is is only for multithreaded optimisations?