Re: [Fwd: Re: compiling with icc]

2004-04-01 Thread Yonah Russ
sorry- typo- the gcc binary was 4.0.18
yonah
Yonah Russ wrote:

Hi all- Here is more info on compiling mysql with icc 8 from the mysql 
packagers list-
As for my attempts, I compiled successfully a couple times and the 
first benchmark I did was faster than the gcc compiled server by about 
10%-

unfortunately, I can't say if this is a conculsive test since the gcc 
binary was 3.23 and the icc binary 4.1.1 - I only managed one 
benchmark b/c something has broken the benchmark suite and I haven't 
had time to check into it...

I'll try the following and get back to you- hopefully with good news.
yonah
 Original Message 
Subject: Re: compiling with icc
Date: Thu, 1 Apr 2004 22:23:02 +0200 (CEST)
From: Lenz Grimmer <[EMAIL PROTECTED]>
To: Yonah Russ <[EMAIL PROTECTED]>
CC: [EMAIL PROTECTED]
References: <[EMAIL PROTECTED]>


-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
Hi Yonah,

sorry for the late reply. I have now been finally able to install icc 
8.0 on one our build hosts (Running SuSE Linux 8.2) and made some tests.

On Wed, 24 Mar 2004, Yonah Russ wrote:

I'm trying to compile mysql 4.0.18 with icc 8.0 on RH AS 3u1. I'm
including my last configuration for compiling and how it died- I've
tried about a million other ways- this seems to come the closest to
compiling normally but dies with an internal error(looks like intel's
fault not yours).
  Does anyone have a configuration where this compiles? What did I do 
wrong?

fyi- there is a problem with the configure/libtool script when you don't
compile -static and you use ipo... the compilation runs icc -shared to
link the libraries but it needs the -ipo option as well.
thanks in advance for any help!
yonah
here the environment variables:
AR=xiar
CC=icc
CFLAGS='-O3 -ipo -static -axWN -march=pentium4 -mcpu=pentium4'
CXX=icc
CXXFLAGS='-O3 -ipo -static -axWN -march=pentium4 -mcpu=pentium4'
LANG=C
LD=xild
PATH=/usr/local/intel/bin:/usr/kerberos/sbin:/usr/kerberos/bin:/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin:/usr/local/sbin:/usr/bin/X11:/usr/X11R6/bin:/root/bin:/usr/local/mysql/bin 

this is the configure line:

./configure --prefix=/usr/local/mysql --enable-thread-safe-client 
--without-innodb --without-debug --enable-assembler 
--with-named-curses-libs=/usr/lib/libncurses.so.5.3

it dies with this error:

IPO Warning: unresolved : rl_attempted_completion_function
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : rl_completion_entry_function
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : rl_readline_name
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : popen
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : pclose
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : readline
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : add_history
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : completion_matches
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : write_history
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : read_history
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : __xstat64
Referenced in libmysqlclient.a(my_lib_ipo.o)
IPO Warning: unresolved : __fxstat64
Referenced in libmysqlclient.a(my_lib_ipo.o)
IPO Warning: unresolved : readdir64
Referenced in libmysqlclient.a(my_lib_ipo.o)
IPO Warning: unresolved : fdopen
Referenced in libmysqlclient.a(my_fopen_ipo.o)
IPO Warning: unresolved : fclose
Referenced in libmysqlclient.a(my_fopen_ipo.o)
IPO Warning: unresolved : gethostbyname_r
Referenced in libmysqlclient.a(my_gethostbyname_ipo.o)
IPO Warning: unresolved : __lxstat64
Referenced in libmysqlclient.a(my_symlink_ipo.o)
IPO Warning: unresolved : realpath
Referenced in libmysqlclient.a(my_symlink_ipo.o)
libmysql.c(1808): internal error: 0_1345
icc: error: problem during multi-file optimization compilation (code 4)
make[2]: *** [mysql] Error 4
make[2]: Leaving directory `/moat-files/mysql-4.0.18/client'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/moat-files/mysql-4.0.18'
make: *** [all] Error 2


That really sounds like an icc bug. Here's the line we currently use 
for our binaries on icc:

CFLAGS="-O3 -unroll2 -ip -mp -no-gcc -restrict" CC=icc CXX=icc
CXXFLAGS="-O3 -unroll2 -ip -mp -no-gcc -restrict" ./configure
- --prefix=/usr/local/mysql --localstatedir=/usr/local/mysql/data
- --libexecdir=/usr/local/mysql/bin --with-extra-charsets=complex
- --enable-thread-safe-client --enable-local-infile --enable-assembler
- --disable-shared --with-client-ldflags=-all-static
- --with-mysqld-ldflags=-all-static --with-readline 
--with-embedded-server
- --with-innodb

You may be able to use higher optimization levels (e.g. by removing 
"-mp", which will however cause some loss in floatingpoint accuracy - 
some of the test suite tests will fail). I assume the "--no-gcc" is 
key here.

Bye,
LenZ
- -- Le

[Fwd: Re: compiling with icc]

2004-04-01 Thread Yonah Russ
Hi all- Here is more info on compiling mysql with icc 8 from the mysql 
packagers list-
As for my attempts, I compiled successfully a couple times and the first 
benchmark I did was faster than the gcc compiled server by about 10%-

unfortunately, I can't say if this is a conculsive test since the gcc 
binary was 3.23 and the icc binary 4.1.1 - I only managed one benchmark 
b/c something has broken the benchmark suite and I haven't had time to 
check into it...

I'll try the following and get back to you- hopefully with good news.
yonah
 Original Message 
Subject:Re: compiling with icc
Date:   Thu, 1 Apr 2004 22:23:02 +0200 (CEST)
From:   Lenz Grimmer <[EMAIL PROTECTED]>
To: Yonah Russ <[EMAIL PROTECTED]>
CC: [EMAIL PROTECTED]
References: <[EMAIL PROTECTED]>


-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
Hi Yonah,

sorry for the late reply. I have now been finally able to install icc 8.0 
on one our build hosts (Running SuSE Linux 8.2) and made some tests.

On Wed, 24 Mar 2004, Yonah Russ wrote:

I'm trying to compile mysql 4.0.18 with icc 8.0 on RH AS 3u1. I'm
including my last configuration for compiling and how it died- I've
tried about a million other ways- this seems to come the closest to
compiling normally but dies with an internal error(looks like intel's
fault not yours).
  Does anyone have a configuration where this compiles? What did I do wrong?
fyi- there is a problem with the configure/libtool script when you don't
compile -static and you use ipo... the compilation runs icc -shared to
link the libraries but it needs the -ipo option as well.
thanks in advance for any help!
yonah
here the environment variables:
AR=xiar
CC=icc
CFLAGS='-O3 -ipo -static -axWN -march=pentium4 -mcpu=pentium4'
CXX=icc
CXXFLAGS='-O3 -ipo -static -axWN -march=pentium4 -mcpu=pentium4'
LANG=C
LD=xild
PATH=/usr/local/intel/bin:/usr/kerberos/sbin:/usr/kerberos/bin:/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin:/usr/local/sbin:/usr/bin/X11:/usr/X11R6/bin:/root/bin:/usr/local/mysql/bin
this is the configure line:

./configure --prefix=/usr/local/mysql --enable-thread-safe-client 
--without-innodb --without-debug --enable-assembler 
--with-named-curses-libs=/usr/lib/libncurses.so.5.3

it dies with this error:

IPO Warning: unresolved : rl_attempted_completion_function
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : rl_completion_entry_function
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : rl_readline_name
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : popen
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : pclose
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : readline
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : add_history
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : completion_matches
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : write_history
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : read_history
Referenced in ipo_icc5X7oE4.o
IPO Warning: unresolved : __xstat64
Referenced in libmysqlclient.a(my_lib_ipo.o)
IPO Warning: unresolved : __fxstat64
Referenced in libmysqlclient.a(my_lib_ipo.o)
IPO Warning: unresolved : readdir64
Referenced in libmysqlclient.a(my_lib_ipo.o)
IPO Warning: unresolved : fdopen
Referenced in libmysqlclient.a(my_fopen_ipo.o)
IPO Warning: unresolved : fclose
Referenced in libmysqlclient.a(my_fopen_ipo.o)
IPO Warning: unresolved : gethostbyname_r
Referenced in libmysqlclient.a(my_gethostbyname_ipo.o)
IPO Warning: unresolved : __lxstat64
Referenced in libmysqlclient.a(my_symlink_ipo.o)
IPO Warning: unresolved : realpath
Referenced in libmysqlclient.a(my_symlink_ipo.o)
libmysql.c(1808): internal error: 0_1345
icc: error: problem during multi-file optimization compilation (code 4)
make[2]: *** [mysql] Error 4
make[2]: Leaving directory `/moat-files/mysql-4.0.18/client'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/moat-files/mysql-4.0.18'
make: *** [all] Error 2


That really sounds like an icc bug. Here's the line we currently use for 
our binaries on icc:

CFLAGS="-O3 -unroll2 -ip -mp -no-gcc -restrict" CC=icc CXX=icc
CXXFLAGS="-O3 -unroll2 -ip -mp -no-gcc -restrict" ./configure
- --prefix=/usr/local/mysql --localstatedir=/usr/local/mysql/data
- --libexecdir=/usr/local/mysql/bin --with-extra-charsets=complex
- --enable-thread-safe-client --enable-local-infile --enable-assembler
- --disable-shared --with-client-ldflags=-all-static
- --with-mysqld-ldflags=-all-static --with-readline --with-embedded-server
- --with-innodb
You may be able to use higher optimization levels (e.g. by removing "-mp", 
which will however cause some loss in floatingpoint accuracy - some of the 
test suite tests will fail). I assume the "--no-gcc" is key here.

Bye,
	LenZ
- -- 
Lenz Grimmer <[EMAIL PROTECTED]>
Senior Production Engineer
MySQL GmbH, http: