GCC 4.4.0 released!

[cyberfish]

GCC, the GNU Compiler Collection - GNU Project - Free Software Foundation (FSF)

Current release series: GCC 4.4.0
Status: 2009-03-13 (regression fixes and docs only).
Serious regressions. All regressions.
Previous release series: GCC 4.3.3
Status: 2009-01-24 (regression fixes and docs only).
Serious regressions. All regressions.
Active development: GCC 4.5.0 (changes)
Status: Stage 1, open for development.
Serious regressions. All regressions.

I will try building a MinGW distribution some time later (and possibly x86-64 cross-compiler, too).

[brewbuck]

Now if only the GNU ld guys would get on the ball and fix the core dump I've been seeing for three years...

[cyberfish]

Hmm for some reason 4.4.0 is not up on the GNU FTP and mirrors yet. Guess we still have to wait a bit, or SVN it.

[cpjust]

So what's new in 4.4?

[cyberfish]

GCC 4.4 Release Series — Changes, New Features, and Fixes - GNU Project - Free Software Foundation (FSF)

Mostly optimizations from what I heard, with a new register allocator (IRA).

And more C++0x stuff of course (including <thread>!).

[abachler]

yeah, no direct download though, which is one thign that always ........es me off about open source, as if everyone is going to go through the trouble of setting up an SVN directory for every project they download. I mean for sciences sake, at least rar the ........ers up and give people a direct link, not a link to a mirror that doesnt respond.

[laserlight]

Originally Posted by abachler

yeah, no direct download though, which is one thign that always ........es me off about open source, as if everyone is going to go through the trouble of setting up an SVN directory for every project they download.

The keyword is "yet". If you are going to hack on the code or use the bleeding edge, then yes, you should go through the trouble of downloading the version control repository for the project (unless the project also provides nightly snapshots to download or something). Otherwise, wait for the release to be properly available.

[Tonto]

I keep trying to build this under cygwin and then mfpr fails damnit

[cyberfish]

I will try it with MSYS tonight.

[Tonto]

i have it compiling because i couldn't find any binary distribution for cygwin. there's a binary distribution for mingw and msys though TDM's GCC/mingw32 Builds

i've had it compiling for like 10 hours now @_@

[cyberfish]

Hmm my computer decided to die on me tonight, so no building GCC for me . Building all that stuff on my 12" AMD tablet that doesn't need any help to overheat is probably not a good idea.

I don't think it took that long when I last tried it, though. More like 30 mins - 1 hr (on my Core 2). Maybe it's Cygwin. Apparently Cygwin has a very slow implementation of fork(), so configure scripts all run forever. Not sure if make is affected by that, too.

[cyberfish]

Done!

MSYS version, no cygwin requirement.

C:\Documents and Settings\Administrator\My Documents>gcc -v
Using built-in specs.
Target: i686-pc-mingw32
Configured with: ../configure --enable-languages=c,c++ --target=i686-pc-mingw32
--prefix=/mingw32 --with-sysroot=/mingw32 --disable-shared --disable-nls
Thread model: win32
gcc version 4.4.0 20090404 (prerelease) (GCC)

Built on a new 64-bit XP virtual machine (since I'm going to build the mingw-w64 cross compiler next, and presumably want to test it).

Took me a whole afternoon + night hunting for the bits and pieces, compiling, troubleshooting, etc. GCC itself took about 2 hours to compile.

This version uses the CRT and headers from the mingw-w64 project (NOT the original mingw). It seems to be better maintained, and more feature-rich, and supports 64-bit (as well as 32-bit). No multilib for now, though (-m32/-m64).

I only tested it briefly, and it seems to work (-std=gnu++0x is too nice ). Didn't test any Windows API code, though, since I don't have any...

If anyone wants to try it out -
http://cyberfish.wecheer.com/mingw32.7z (~27MB, decompresses to 250MB)

how I compiled everything (I'm really inexperienced at this... anyone see anything wrong?)

CFLAGS="-pipe -O2 -mms-bitfields -march=i686"
SourceForge.net Repository - [mingw-w64] View of /trunk/mingw-w64-doc/howto-build/mingw-w64-howto-build.txt

binutils 2.19
mkdir build-32
cd build-32
../configure --target=i686-pc-mingw32 --prefix=/mingw32 --with-sysroot=/mingw32 --disable-nls --disable-shared
csmake -j 2
csmake install

GMP 4.2.4 (Nabble - MinGW - User - help to compile gmp-4.2.2 fixes the Makefile)
mkdir build-32
cd build-32
ABI=32 ../configure --prefix=/mingw --disable-shared
csmake -j 2
csmake check #all tests passed
csmake install

mpfr 2.4.1
mkdir build-32
cd build-32
../configure --prefix=/mingw --disable-shared
csmake -j 2
csmake install

gcc 4.4 SVN branch (checked out Apr 4, 2009)
mkdir build-32
cd build-32
../configure --enable-languages=c,c++ --target=i686-pc-mingw32 --prefix=/mingw32 --with-sysroot=/mingw32 --disable-shared --disable-nls
csmake -j 2 all-gcc
csmake install-gcc

mingw-w64-crt
export PATH=$PATH:/mingw32/bin
mkdir build-32
cd build-32
../configure --host=i686-pc-mingw32 --enable-lib32 --disable-lib64 --prefix=/mingw32 --with-sysroot=/mingw32
csmake -j 2
csmake install

rest of GCC
rm -rf /mingw32/mingw
cp -r /mingw32/i686-pc-mingw32 /mingw32/mingw
cd build-32
csmake -j 2
csmake install

[cyberfish]

For my chess program, GCC 4.4 (on Linux) doesn't make it measurably faster than with GCC 4.3, but "-O3" compilation time is about 30% longer.

cyberfish@cyberfish-desktop:/data/cpp/Brainless$ time g++ -march=native -g -O3 -Wall Brainless.cpp -o bin/Release/Brainless -lboost_thread

real 0m6.711s
user 0m5.616s
sys 0m0.348s
cyberfish@cyberfish-desktop:/data/cpp/Brainless$ time g++-4.4 -march=native -g -O3 -Wall Brainless.cpp -o bin/Release/Brainless -lboost_thread

real 0m8.929s
user 0m8.569s
sys 0m0.172s
cyberfish@cyberfish-desktop:/data/cpp/Brainless$ time g++-4.4 -march=native -g -O0 -Wall Brainless.cpp -o bin/Release/Brainless -lboost_thread

real 0m3.502s
user 0m2.904s
sys 0m0.200s

At least I can switch to std::thread from boost::thread now .

[brewbuck]

Originally Posted by cyberfish

For my chess program, GCC 4.4 (on Linux) doesn't make it measurably faster than with GCC 4.3, but "-O3" compilation time is about 30% longer.

But did you build GCC the same way the older one was built?

[cyberfish]

Hmm good point.

I didn't.

4.3 is from Ubuntu APT. I only built 4.4 myself.

And I just realized I actually checked out trunk (4.5.0) instead of 4.4 release branch. Not sure what difference that makes, since it was just branched a few days ago. I am checking out 4.4 branch now.

GCC trunk was built with -O2, though, and I suspect the Ubuntu people build with -O2, too.