[HOWTO] Intel C++ Compiler and per-package CFLAGS

[mattst88]

I've rewritten Gentoo-Wiki's HOWTO on using the Intel C++ Compiler.

http://gentoo-wiki.com/HOWTO_Intel_C_Compiler

The old guide was terribly written and horribly messy.

It also made Portage compile with icc unless you explicitly told it to use gcc. This is bad. Far too many packages don't work with icc to use this behavior. My guide makes Portage compile only specified packages with icc and all others with gcc which will save a few headaches in the long run. It also will give your system the ability to use per-package CFLAGS for gcc and icc!

Need feedback on the guide. Suggestions are welcome.

[Etal]

Thanks, that looks interesting.

Is the change in performance noticeable, though? For example, if I bzip an archive or encode a video in x264, how much faster would it be than if I use gcc 4.1 on a Core2 processor?

[mattst88]

Code: Select all

decompression linux-2.6.23.tar.bz2
 CC    CFLAGS                     time 1  time 2  time 3  time avg
 icc   "-O2 -xT -parallel -gcc"   20.970  20.917  21.037 20.975
 icc   "-O2 -xT -gcc"             21.384  22.115  21.346 21.615
 icc   "-O3 -xT -parallel -gcc"   22.298  21.682  22.643 22.208
 icc   "-O3 -xT -gcc"             21.543  21.465  21.938 21.649
 icc   "-fast -xT -parallel -gcc" 21.583  21.818  21.584 21.662
 icc   "-fast -xT -gcc"           21.645  21.383  22.042 21.690
 gcc   "-O2 -march=nocona"        21.221  21.620  21.376 21.406
 gcc   "-O3 -march=nocona"        21.076  21.170  20.956 21.067
 gcc   "-Os -march=nocona"        21.712  21.492  21.848 21.684

As far as bzip2 decompression is concerned, "icc -xT -parallel -gcc" beats "gcc -O3 -march=nocona" 20.975 seconds to 21.067. Hardly at all in other words.

Code: Select all

compression linux-2.6.23.tar.bz2
 CC  CFLAGS                     time 1 time 2 time 3 time avg
 icc "-O2 -xT -parallel -gcc"   48.751 48.191 48.592 48.511
 icc "-O2 -xT -gcc"             47.452 48.208 47.969 47.876
 icc "-O3 -xT -parallel -gcc"   48.633 48.822 47.827 48.427
 icc "-O3 -xT -gcc"             46.851 48.087 48.295 47.744
 icc "-fast -xT -parallel -gcc" 49.289 49.789 49.199 49.426
 icc "-fast -xT -gcc"           49.200 49.467 49.441 49.369
 gcc "-O2 -march=nocona"        51.830 53.395 52.037 52.421
 gcc "-O3 -march=nocona"        52.073 53.426 52.331 52.610
 gcc "-Os -march=nocona"        50.256 51.365 51.464 51.028

In the case of compression, "icc -O3 -xT -gcc" beats "gcc -Os -march=nocona" 47.744 to 51.028 seconds. This is a significant difference.

In this thread a user states that xvid is 10-15% faster on an Athlon XP (this is without SSE2,3 optimizations)

Also see:
PHP gcc/icc benchmarks

[easy target]

Thanks, that looks interesting.

Is the change in performance noticeable, though? For example, if I bzip an archive or encode a video in x264, how much faster would it be than if I use gcc 4.1 on a Core2 processor?

This article says something about icc and gcc.

[deno]

Here faac fails with error:

Code: Select all

icc -O2 -xP -gcc -o .libs/faac main.o input.o  ../libfaac/.libs/libfaac.so -lm /usr/lib/libmp4v2.so -ldl
ld: .libs/faac: hidden symbol `__svml_sin2' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_dsin2.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [faac] Error 1
make[2]: Leaving directory `/var/tmp/portage/media-libs/faac-1.26-r1/work/faac/frontend'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/var/tmp/portage/media-libs/faac-1.26-r1/work/faac'
make: *** [all] Error 2

Any suggestions on how could i solve this?

edit:

same(?) happens witg libtheora. error:

Code: Select all

ld: .libs/decoder_example: hidden symbol `__svml_cosf4' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_scos4.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [decoder_example] Error 1
ld: .libs/chaining_example: hidden symbol `__svml_cosf4' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_scos4.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [chaining_example] Error 1
make[2]: Leaving directory `/var/tmp/portage/media-libs/libvorbis-1.2.0/work/libvorbis-1.2.0/examples'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/var/tmp/portage/media-libs/libvorbis-1.2.0/work/libvorbis-1.2.0'
make: *** [all] Error 2

[paluszak]

I have just compiled x11-themes/qtcurve, x11-themes/gtk-engines-curve and app-mobilephone/obex-data-server, they all seem to work fine. Is there a place where we could make a list of packages that compile fine with ICC?

Another thing is that it pays to read the ICC docs, there are some optimization options that could be considered, for instance, my flags on Pentium4 look like this:

Code: Select all

-march=pentium4 -xW -O2 -ip -gcc

It seems that -ipo could give a huge boost cause it enables multi-file optimizations, but, unfortunately, it's not really compatible with the build system of most Linux packages as it causes the compiler not to generate standard object files during compilation phase.

[deno]

I have just compiled firefox, but it wont run. I get:

Code: Select all

No running windows found
/usr/libexec/mozilla-launcher: line 119:  2513 Segmentation fault      $(type -P aoss) "$mozbin" $xulparams "$@"
firefox-bin exited with non-zero status (139)

btw I am just compiling e17 packages. for now only thing i can say is that a lot of them compiles fine. enterminus runs, and it is slow as always .

[deno]

my specs: core due t 2250, icc version 10.1

e17 (everything below) and all modules build fine

• dev-libs/eet
  dev-db/edb
  x11-libs/evas
  x11-libs/ecore
  dev-libs/efreet
  dev-libs/embryo
  media-libs/edje
  media-libs/emotion
  media-libs/imlib2
  media-libs/epeg
  media-libs/epsilon
  x11-libs/ewl
  x11-libs/e_dbus
  x11-wm/e
  x11-libs/esmart
  x11-misc/entrance
  x11-terms/enterminus
  app-misc/entropy

from other apps i compiled those also build and run fine:

• x11-terms/xterm
  x11-plugins/itask-ng
  dev-libs/nss
  dev-libs/nspr
  app-text/hunspell
  xfce-base/thunar
  app-editors/vim-core
  app-editors/gvim

all these are ~x86 at the moment.

edit:

I am writing this with firefox beta 3 compiled with -xP -Os -gcc-version=420 -ipo -fomit-frame-pointer -parallel.

[deno]

@paluszak

From icc man page:

-xW -- Can generate SSE2 and SSE instructions, and it can optimize for Intel(R) Pentium(R) 4 processors and
Intel(R) Xeon(R) processors with SSE2. This is the default on Linux systems using Intel(R) 64 architecture. This
option is the same as specifying -march=pentium4.

Since -xW is the same thing as as -march=pentium4 I don't see the point in setting both flags.

[paluszak]

@paluszak

From icc man page:

-xW -- Can generate SSE2 and SSE instructions, and it can optimize for Intel(R) Pentium(R) 4 processors and
Intel(R) Xeon(R) processors with SSE2. This is the default on Linux systems using Intel(R) 64 architecture. This
option is the same as specifying -march=pentium4.

Since -xW is the same thing as -march as -march=pentium4 I don't the point in setting both flags.

Oops, that's all about reading the docs. Anyway, it seems that using -ipo or -ip if -ipo fails gives a huge boost. I generally try to compile packages with -ipo, and if it fails I switch back to -ip.

[mattst88]

Here faac fails with error:

Code: Select all

icc -O2 -xP -gcc -o .libs/faac main.o input.o  ../libfaac/.libs/libfaac.so -lm /usr/lib/libmp4v2.so -ldl
ld: .libs/faac: hidden symbol `__svml_sin2' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_dsin2.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [faac] Error 1
make[2]: Leaving directory `/var/tmp/portage/media-libs/faac-1.26-r1/work/faac/frontend'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/var/tmp/portage/media-libs/faac-1.26-r1/work/faac'
make: *** [all] Error 2

Any suggestions on how could i solve this?

edit:

same(?) happens witg libtheora. error:

Code: Select all

ld: .libs/decoder_example: hidden symbol `__svml_cosf4' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_scos4.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [decoder_example] Error 1
ld: .libs/chaining_example: hidden symbol `__svml_cosf4' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_scos4.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [chaining_example] Error 1
make[2]: Leaving directory `/var/tmp/portage/media-libs/libvorbis-1.2.0/work/libvorbis-1.2.0/examples'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/var/tmp/portage/media-libs/libvorbis-1.2.0/work/libvorbis-1.2.0'
make: *** [all] Error 2

What are your CFLAGS? Both of these packages compile cleanly with "-O3 -ipo -xT -gcc" on my AMD64 system.

[scronkey]

Good effort on your clean and well written HOWTO.

I installed icc a little while ago, not using portage though. I've just been experimenting with it but I was wondering if the set up described in your article could still be used, without me re-installing the compiler through portage.

From the looks of it I'd say it would work but I'm not sure whether the /etc/portage/bashrc file and the ICCCFLAGS would still be adhered to by portage. I really like the clean separation of what is compiled with what and I want to get Python compiled with icc but considering how central Python is to Gentoo I don't want to go in blindly (Been there before. You're not really a Gentoo until you've borked your Python install!).

Any tips or caveats you're aware of? Is the /etc/portage/bashrc used for other parts of portage or is it something you've implemented?

[deno]

@|mattst88|

Thanks for your replay. Have just compiled libtheora, but faac still fails. I have experimented with flags and it always fails. I have tried:

• -xP -O2 -gcc-version=420 -fomit-frame-pointer -ipo
  -xP -O2 -gcc-version=420
  -xP -O3 -gcc-version=420 -fomit-frame-pointer -ipo
  -xP -O3 -gcc-version=420
  -xP -Os -gcc-version=420 -fomit-frame-pointer -ipo
  -xP -Os -gcc-version=420

edit:

faac compiles with 'media-libs/faac -mtune=pentium4 -O2 -gcc-version=420'. the same is a must here for libvorbis.

[deno]

splashutils now fails (with gcc-4.2.3, probably because of freetype been compiled with icc) error:

gcc -march=prescott -O2 -pipe -fomit-frame-pointer -I/usr/include/freetype2 -L/usr/lib -o fbtruetype -L/usr/lib fbtruetype.o messages.o console.o ttf.o luxisri.o -lfreetype -lm
gcc -march=prescott -O2 -pipe -fomit-frame-pointer -I/usr/include/freetype2 -L/usr/lib -static -o fbtruetype.static -L/usr/lib fbtruetype.o messages.o console.o ttf.o luxisri.o /usr/lib/libfreetype.a /usr/lib/libz.a -lm
ttf.o: In function `Find_Glyph':
ttf.c:(.text+0x13c): undefined reference to `FT_Load_Glyph'
ttf.c:(.text+0x278): undefined reference to `FT_Render_Glyph'
ttf.c:(.text+0x3e9): undefined reference to `FT_Get_Char_Index'
ttf.c:(.text+0x409): undefined reference to `FT_Render_Glyph'
ttf.c:(.text+0x456): undefined reference to `FT_Outline_Transform'
ttf.o: In function `TTF_CloseFont':
ttf.c:(.text+0xc29): undefined reference to `FT_Done_Face'
ttf.o: In function `TTF_OpenFontIndex':
ttf.c:(.text+0xc96): undefined reference to `FT_New_Face'
ttf.c:(.text+0xccc): undefined reference to `FT_New_Memory_Face'
ttf.c:(.text+0xd2b): undefined reference to `FT_Set_Char_Size'
ttf.c:(.text+0xd48): undefined reference to `FT_MulFix'
ttf.c:(.text+0xd69): undefined reference to `FT_MulFix'
ttf.c:(.text+0xd98): undefined reference to `FT_MulFix'
ttf.c:(.text+0xdba): undefined reference to `FT_MulFix'
ttf.c:(.text+0xdd9): undefined reference to `FT_MulFix'
ttf.c:(.text+0xe44): undefined reference to `FT_Done_Face'
ttf.c:(.text+0xe61): undefined reference to `FT_New_Face'
ttf.o: In function `TTF_Quit':
ttf.c:(.text+0x103d): undefined reference to `FT_Done_FreeType'
ttf.o: In function `TTF_Init':
ttf.c:(.text+0x105a): undefined reference to `FT_Init_FreeType'
collect2: ld returned 1 exit status
make[1]: *** [fbtruetype.static] Error 1
make[1]: Leaving directory `/var/tmp/portage/media-gfx/splashutils-1.5.3.4/work/miscsplashutils-0.1.8/fbtruetype'
make: *** [all] Error 2

now even if recompile freetype with gcc i get this error compiling splashutils:

Making all in src
CREATE fbsplash.h
Making all in .
In file included from common.h:4,
from mng_callbacks.c:15:
../config.h:51:1: warning: "HAVE_STDLIB_H" redefined
In file included from /usr/include/jpeglib.h:28,
from /usr/include/libmng_types.h:192,
from /usr/include/libmng.h:379,
from mng_callbacks.c:12:
/usr/include/jconfig.h:12:1: warning: this is the location of the previous definition
In file included from common.h:4,
from mng_render.c:17:
../config.h:51:1: warning: "HAVE_STDLIB_H" redefined
In file included from /usr/include/jpeglib.h:28,
from /usr/include/libmng_types.h:192,
from /usr/include/libmng.h:379,
from mng_render.c:15:
/usr/include/jconfig.h:12:1: warning: this is the location of the previous definition
/usr/lib/libpng12.a(libpng12_la-png.o): In function `png_info_destroy':
(.text+0x267): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-png.o): In function `png_destroy_info_struct':
(.text+0x7ac): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-png.o): In function `png_create_info_struct':
(.text+0x800): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-png.o): In function `png_sig_cmp':
(.text+0x945): undefined reference to `_intel_fast_memcmp'
/usr/lib/libpng12.a(libpng12_la-png.o): In function `png_check_sig':
(.text+0x9ce): undefined reference to `_intel_fast_memcmp'
/usr/lib/libpng12.a(libpng12_la-png.o): In function `png_info_init':
(.text+0xa46): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-png.o): In function `png_info_init_3':
(.text+0xa82): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_destroy.':
(.text+0x1642): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_destroy.':
(.text+0x166e): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_destroy.':
(.text+0x169d): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init':
(.text+0x207d): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init':
(.text+0x2099): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init':
(.text+0x20a9): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init_2':
(.text+0x21f7): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init_2':
(.text+0x2221): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init_2':
(.text+0x2231): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init_3':
(.text+0x240c): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init_3':
(.text+0x243e): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngread.o): In function `png_read_init_3':
(.text+0x244e): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngrtran.o): In function `png_set_dither':
(.text+0x4448): undefined reference to `__intel_VEC_memzero'
/usr/lib/libpng12.a(libpng12_la-pngrtran.o): In function `png_set_dither':
(.text+0x4467): undefined reference to `__intel_VEC_memset'
/usr/lib/libpng12.a(libpng12_la-pngrtran.o): In function `png_build_gamma_table.':
(.text+0x5069): undefined reference to `__libm_sse2_pow'
/usr/lib/libpng12.a(libpng12_la-pngrtran.o): In function `png_build_gamma_table.':
(.text+0x5248): undefined reference to `__libm_sse2_pow'
/usr/lib/libpng12.a(libpng12_la-pngrtran.o): In function `png_build_gamma_table.':
(.text+0x538b): undefined reference to `__libm_sse2_pow'
/usr/lib/libpng12.a(libpng12_la-pngrtran.o): In function `png_build_gamma_table.':
(.text+0x54af): undefined reference to `__libm_sse2_pow'
/usr/lib/libpng12.a(libpng12_la-pngrtran.o): In function `png_build_gamma_table.':
(.text+0x55dc): undefined reference to `__libm_sse2_pow'
/usr/lib/libpng12.a(libpng12_la-pngrtran.o):(.text+0x56a2): more undefined references to `__libm_sse2_pow' follow
/usr/lib/libpng12.a(libpng12_la-pngmem.o): In function `png_memset_check':
(.text+0x47): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngmem.o): In function `png_memcpy_check':
(.text+0x5d): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngmem.o): In function `png_create_struct_2':
(.text+0x197): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngmem.o): In function `png_create_struct_2':
(.text+0x1c6): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngmem.o): In function `png_create_struct':
(.text+0x206): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngrutil.o): In function `png_do_read_interlace':
(.text+0x10c0): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngrutil.o): In function `png_do_read_interlace':
(.text+0x10e3): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngrutil.o): In function `png_combine_row':
(.text+0x14aa): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngrutil.o): In function `png_combine_row':
(.text+0x15ee): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngrutil.o): In function `png_decompress_chunk.':
(.text+0x1d67): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngrutil.o):(.text+0x1d8f): more undefined references to `_intel_fast_memcpy' follow
/usr/lib/libpng12.a(libpng12_la-pngset.o): In function `png_set_PLTE':
(.text+0x135c): undefined reference to `_intel_fast_memset'
/usr/lib/libpng12.a(libpng12_la-pngset.o): In function `png_set_PLTE':
(.text+0x1370): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngset.o): In function `png_set_pCAL':
(.text+0x147d): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngset.o): In function `png_set_pCAL':
(.text+0x14eb): undefined reference to `_intel_fast_memcpy'
/usr/lib/libpng12.a(libpng12_la-pngset.o): In function `png_set_pCAL':
(.text+0x156c): undefined reference to `_intel_fast_memcpy'
collect2: ld returned 1 exit status
make[4]: *** [splash_util.static] Error 1
make[3]: *** [all-recursive] Error 1
make[2]: *** [all] Error 2
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2

...what is here strange to me is that i see intel mentioned again... could ccache have something to do with it?

[mattst88]

deno,

I experienced this also. My solution was to compile libpng with gcc. Alternatively, compile freetype, libpng, and splashutils with gcc, and following, recompile libpng and freetype with icc. That's a bit of a headache though.

[kaethorn]

Hello everyone and thanks for the HOWTO. I've followed the steps mentioned and it seems to work fine for app-arch/bzip2, but that's about it. As soon as I try any package which includes limits.h, compilation fails:

Code: Select all

>>> Emerging (1 of 1) app-arch/cpio-2.9-r1 to /

[snip]

if icc -c99 -DHAVE_CONFIG_H -I. -I. -I..      -MT argp-parse.o -MD -MP -MF ".deps/argp-parse.Tpo" -c -o argp-parse.o argp-parse.c; \
	then mv -f ".deps/argp-parse.Tpo" ".deps/argp-parse.Po"; else rm -f ".deps/argp-parse.Tpo"; exit 1; fi
icc: command line remark #10010: option '-c99' is deprecated and will be removed in a future release. See '-help deprecated'
icc: command line remark #10010: option '-c99' is deprecated and will be removed in a future release. See '-help deprecated'
/usr/include/gentoo-multilib/amd64/limits.h(125): catastrophic error: could not open source file "limits.h"
  # include_next <limits.h>
                           ^

compilation aborted for argp-eexst.c (code 4)
/usr/include/gentoo-multilib/amd64/limits.h(125): catastrophic error: could not open source file "limits.h"
  # include_next <limits.h>
                           ^

compilation aborted for argp-help.c (code 4)
make[3]: *** [argp-eexst.o] Error 1
make[3]: *** Waiting for unfinished jobs....
make[3]: *** [argp-help.o] Error 1
/usr/include/gentoo-multilib/amd64/limits.h(125): catastrophic error: could not open source file "limits.h"
  # include_next <limits.h>
                           ^

compilation aborted for argp-parse.c (code 4)
make[3]: *** [argp-parse.o] Error 1
make[3]: Leaving directory `/var/tmp/portage/app-arch/cpio-2.9-r1/work/cpio-2.9/lib'
make[2]: *** [all] Error 2
make[2]: Leaving directory `/var/tmp/portage/app-arch/cpio-2.9-r1/work/cpio-2.9/lib'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/var/tmp/portage/app-arch/cpio-2.9-r1/work/cpio-2.9'
make: *** [all] Error 2

It seems as if the headers from sys-libs/glibc are used instead of the ones found in /opt/intel/cce/10.1.012/include/. I've tried compiling with different combination of flags, such as '-O2 -xT -gcc' which all yield the same result. I'm using amd64.

Any hint/help[/quote] would be greatly appreciated!

[mattst88]

I experienced this problem for the first time today on my laptop. I can compile the same ebuild on my server with no problems.

I'll investigate and see if I can find a solution.

[kaethorn]

OK, I found out how to get around the error. I've recently compiled gcc-4.3 (4.3.0-pre20080302 to be precise) from the dirtyepic overlay and set it as the default compiler. Setting it back to 4.2.3 did the trick:

Code: Select all

gcc-config x86_64-pc-linux-gnu-4.2.3

But there's another issue I noticed regarding your /etc/portage/bashrc. It seems to break compilation of app-crypt/gnupg (using gcc), most similar to what this [post=4938398]post[/post] reports. Temporarily replacing the icc bashrc with the original one seems to take care of the problem.

[mattst88]

This is odd.

My laptop cannot compile php which includes limits.h with icc. I'm using gcc-4.2.3 (no multislot), icc-10.1.012. No other versions of either compiler are installed.

My server can compile php with icc. Here, I'm using gcc-4.1.2 (no multislot), and icc-10.1.012. No other versions of either compiler are installed.

What could be going on?

[mattst88]

It appears to me that AMD64 systems are effected by the limits.h mess. My Core 2 Duo laptop, my AMD64 Sempron desktop, and your AMD64 box exhibit this behaviour while my 32-bit x86 server does not.

[richardpku]

Add '-fno-builtin' to CFLAGS and they will compile.

Probably it's a bug of ICC. I have also encountered packages that don't compile unless ipo is disabled.

Here faac fails with error:

Code: Select all

icc -O2 -xP -gcc -o .libs/faac main.o input.o  ../libfaac/.libs/libfaac.so -lm /usr/lib/libmp4v2.so -ldl
ld: .libs/faac: hidden symbol `__svml_sin2' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_dsin2.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [faac] Error 1
make[2]: Leaving directory `/var/tmp/portage/media-libs/faac-1.26-r1/work/faac/frontend'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/var/tmp/portage/media-libs/faac-1.26-r1/work/faac'
make: *** [all] Error 2

Any suggestions on how could i solve this?

edit:

same(?) happens witg libtheora. error:

Code: Select all

ld: .libs/decoder_example: hidden symbol `__svml_cosf4' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_scos4.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [decoder_example] Error 1
ld: .libs/chaining_example: hidden symbol `__svml_cosf4' in /opt/intel/cc/10.1.012/lib/libsvml.a(svml_scos4.o) is referenced by DSO
ld: final link failed: Nonrepresentable section on output
make[2]: *** [chaining_example] Error 1
make[2]: Leaving directory `/var/tmp/portage/media-libs/libvorbis-1.2.0/work/libvorbis-1.2.0/examples'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/var/tmp/portage/media-libs/libvorbis-1.2.0/work/libvorbis-1.2.0'
make: *** [all] Error 2

[deno]

You meant ICCCFLAGS, thanks for reply, I'll try it.

[AzraelNewtype]

I'm having a particularly bizarre problem with icc since installing using the wiki. Occasionally, programs not listed in /etc/portage/package.icc are getting compiled with icc, sometimes only certain lines of the make process. What's worse though is that it isn't picking up the ICCCFLAGS with the change of compiler, so these compilations break as soon as the switch happens. The last one I noticed it with is subversion, but the error is a little burried so I'll show the output of samba instead:

Code: Select all

icc -pthread -fno-strict-aliasing -O2 -march=native -fomit-frame-pointer -pipe -D_SAMBA_BUILD_=3 -fPIC -I/usr/include/python2.5 -c -I/usr/include/python2.5 -c python/py_spoolss.c -o build/temp.linux-i686-2.5/python/py_spoolss.o -O2 -march=native -fomit-frame-pointer -pipe -D_SAMBA_BUILD_=3 -DHAVE_CONFIG_H -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -DLDAP_DEPRECATED -O2 -march=native -fomit-frame-pointer -pipe -D_SAMBA_BUILD_=3 -I/var/tmp/portage/net-fs/samba-3.0.30/work/samba-3.0.30/source/iniparser/src -Iinclude -I./include -I. -I. -I./lib/replace -I./lib/talloc -I./tdb/include -I./libaddns -I./librpc -DHAVE_CONFIG_H -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -DLDAP_DEPRECATED -I/var/tmp/portage/net-fs/samba-3.0.30/work/samba-3.0.30/source/lib -D_SAMBA_BUILD_=3
icc: command line error: invalid argument for option '-m'
error: command 'icc' failed with exit status 1
make: *** [python_ext] Error 1

And just to confirm:

Code: Select all

$ grep CFLA /etc/make.conf 
CFLAGS="-O2 -march=native -fomit-frame-pointer -pipe"
CXXFLAGS="${CFLAGS}"
ICCCFLAGS="-O2 -xT -ip -gcc"
ICCCXXFLAGS="${ICCCFLAGS}"

Clearly this is mixing and matching for no reason I can readily discern. Has anyone else encountered this?

[deno]

Yup. It's because of python. I have recompiled it with gcc and subversion compiled then.

[mattst88]

Add '-fno-builtin' to CFLAGS and they will compile.

Thank you very much for finding this. I'll add it to the wiki.

Edit: upon testing with php-5.2.6-r1, this does not appear to work.