Host/Target specific installation notes for GCC

Please read this document carefully before installing the GNU Compiler Collection on your machine.

Note that this list of install notes is not a list of supported hosts or targets. Not all supported hosts and targets are listed here, only the ones that require host-specific or target-specific information are.


This section contains general configuration information for all alpha-based platforms using ELF (in particular, ignore this section for DEC OSF/1, Digital UNIX and Tru64 UNIX). In addition to reading this section, please read all other sections that match your target.

We require binutils 2.11.2 or newer. Previous binutils releases had a number of problems with DWARF 2 debugging information, not the least of which is incorrect linking of shared libraries.


Systems using processors that implement the DEC Alpha architecture and are running the DEC/Compaq Unix (DEC OSF/1, Digital UNIX, or Compaq Tru64 UNIX) operating system, for example the DEC Alpha AXP systems.

As of GCC 3.2, versions before alpha*-dec-osf4 are no longer supported. (These are the versions which identify themselves as DEC OSF/1.)

In Digital Unix V4.0, virtual memory exhausted bootstrap failures may be fixed by configuring with --with-gc=simple, reconfiguring Kernel Virtual Memory and Swap parameters per the /usr/sbin/sys_check Tuning Suggestions, or applying the patch in

In Tru64 UNIX V5.1, Compaq introduced a new assembler that does not currently (2001-06-13) work with mips-tfile. As a workaround, we need to use the old assembler, invoked via the barely documented -oldas option. To bootstrap GCC, you either need to use the Compaq C Compiler:

        % CC=cc srcdir/configure [options] [target]

or you can use a copy of GCC 2.95.3 or higher built on Tru64 UNIX V4.0:

        % CC=gcc -Wa,-oldas srcdir/configure [options] [target]

As of GNU binutils 2.11.2, neither GNU as nor GNU ld are supported on Tru64 UNIX, so you must not configure GCC with --with-gnu-as or --with-gnu-ld.

GCC writes a .verstamp directive to the assembler output file unless it is built as a cross-compiler. It gets the version to use from the system header file /usr/include/stamp.h. If you install a new version of DEC Unix, you should rebuild GCC to pick up the new version stamp.

make compare may fail on old versions of DEC Unix unless you add -save-temps to BOOT_CFLAGS. On these systems, the name of the assembler input file is stored in the object file, and that makes comparison fail if it differs between the stage1 and stage2 compilations. The option -save-temps forces a fixed name to be used for the assembler input file, instead of a randomly chosen name in /tmp. Do not add -save-temps unless the comparisons fail without that option. If you add -save-temps, you will have to manually delete the .i and .s files after each series of compilations.

GCC now supports both the native (ECOFF) debugging format used by DBX and GDB and an encapsulated STABS format for use only with GDB. See the discussion of the --with-stabs option of configure above for more information on these formats and how to select them.

There is a bug in DEC's assembler that produces incorrect line numbers for ECOFF format when the .align directive is used. To work around this problem, GCC will not emit such alignment directives while writing ECOFF format debugging information even if optimization is being performed. Unfortunately, this has the very undesirable side-effect that code addresses when -O is specified are different depending on whether or not -g is also specified.

To avoid this behavior, specify -gstabs+ and use GDB instead of DBX. DEC is now aware of this problem with the assembler and hopes to provide a fix shortly.


Argonaut ARC processor. This configuration is intended for embedded systems.


ARM-family processors. Subtargets that use the ELF object format require GNU binutils 2.13 or newer. Such subtargets include: arm-*-freebsd, arm-*-netbsdelf, arm-*-*linux and arm-*-rtems.


ARM-family processors. Note that there are two different varieties of PE format subtarget supported: arm-wince-pe and arm-pe as well as a standard COFF target arm-*-coff.


ARM-family processors. These targets support the AOUT file format: arm-*-aout, arm-*-netbsd.


ATMEL AVR-family micro controllers. These are used in embedded applications. There are no standard Unix configurations. See “AVR Options” in the main manual for the list of supported MCU types.

Use configure --target=avr --enable-languages="c" to configure GCC.

Further installation notes and other useful information about AVR tools can also be obtained from:

We strongly recommend using binutils 2.13 or newer.

The following error:

       Error: register required

indicates that you should upgrade to a newer version of the binutils.


The Blackfin processor, an Analog Devices DSP. See “Blackfin Options” in the main manual

More information, and a version of binutils with support for this processor, is available at


CRIS is the CPU architecture in Axis Communications ETRAX system-on-a-chip series. These are used in embedded applications.

See “CRIS Options” in the main manual for a list of CRIS-specific options.

There are a few different CRIS targets:

Mainly for monolithic embedded systems. Includes a multilib for the v10 core used in ETRAX 100 LX.
A GNU/Linux port for the CRIS architecture, currently targeting ETRAX 100 LX by default.

For cris-axis-elf you need binutils 2.11 or newer. For cris-axis-linux-gnu you need binutils 2.12 or newer.

Pre-packaged tools can be obtained from More information about this platform is available at


The CRX CompactRISC architecture is a low-power 32-bit architecture with fast context switching and architectural extensibility features.

See “CRX Options” in the main manual for a list of CRX-specific options.

Use configure --target=crx-elf --enable-languages=c,c++ to configure GCC for building a CRX cross-compiler. The option --target=crx-elf is also used to build the newlib C library for CRX.

It is also possible to build libstdc++-v3 for the CRX architecture. This needs to be done in a separate step with the following configure settings: gcc/libstdc++-v3/configure --host=crx-elf --with-newlib --enable-sjlj-exceptions --enable-cxx-flags='-fexceptions -frtti'


Please have a look at the binaries page.

You cannot install GCC by itself on MSDOS; it will not compile under any MSDOS compiler except itself. You need to get the complete compilation package DJGPP, which includes binaries as well as sources, and includes all the necessary compilation tools and libraries.


The version of binutils installed in /usr/bin probably works with this release of GCC. However, on FreeBSD 4, bootstrapping against the latest FSF binutils is known to improve overall testsuite results; and, on FreeBSD/alpha, using binutils 2.14 or later is required to build libjava.

Support for FreeBSD 1 was discontinued in GCC 3.2.

Support for FreeBSD 2 will be discontinued after GCC 3.4. The following was true for GCC 3.1 but the current status is unknown. For FreeBSD 2 or any mutant a.out versions of FreeBSD 3: All configuration support and files as shipped with GCC 2.95 are still in place. FreeBSD 2.2.7 has been known to bootstrap completely; however, it is unknown which version of binutils was used (it is assumed that it was the system copy in /usr/bin) and C++ EH failures were noted.

For FreeBSD using the ELF file format: DWARF 2 debugging is now the default for all CPU architectures. It had been the default on FreeBSD/alpha since its inception. You may use -gstabs instead of -g, if you really want the old debugging format. There are no known issues with mixing object files and libraries with different debugging formats. Otherwise, this release of GCC should now match more of the configuration used in the stock FreeBSD configuration of GCC. In particular, --enable-threads is now configured by default. However, as a general user, do not attempt to replace the system compiler with this release. Known to bootstrap and check with good results on FreeBSD 4.9-STABLE and 5-CURRENT. In the past, known to bootstrap and check with good results on FreeBSD 3.0, 3.4, 4.0, 4.2, 4.3, 4.4, 4.5, 4.8-STABLE.

In principle, --enable-threads is now compatible with --enable-libgcj on FreeBSD. However, it has only been built and tested on i386-*-freebsd[45] and alpha-*-freebsd[45]. The static library may be incorrectly built (symbols are missing at link time). There is a rare timing-based startup hang (probably involves an assumption about the thread library). Multi-threaded boehm-gc (required for libjava) exposes severe threaded signal-handling bugs on FreeBSD before 4.5-RELEASE. Other CPU architectures supported by FreeBSD will require additional configuration tuning in, at the very least, both boehm-gc and libffi.

Shared is now built and installed by default.


Renesas H8/300 series of processors.

Please have a look at the binaries page.

The calling convention and structure layout has changed in release 2.6. All code must be recompiled. The calling convention now passes the first three arguments in function calls in registers. Structures are no longer a multiple of 2 bytes.


Support for HP-UX version 9 and older was discontinued in GCC 3.4.

We require using gas/binutils on all hppa platforms. Version 2.19 or later is recommended.

It may be helpful to configure GCC with the --with-gnu-as and --with-as=... options to ensure that GCC can find GAS.

The HP assembler should not be used with GCC. It is rarely tested and may not work. It shouldn't be used with any languages other than C due to its many limitations.

Specifically, -g does not work (HP-UX uses a peculiar debugging format which GCC does not know about). It also inserts timestamps into each object file it creates, causing the 3-stage comparison test to fail during a bootstrap. You should be able to continue by saying make all-host all-target after getting the failure from make.

Various GCC features are not supported. For example, it does not support weak symbols or alias definitions. As a result, explicit template instantiations are required when using C++. This makes it difficult if not impossible to build many C++ applications.

There are two default scheduling models for instructions. These are PROCESSOR_7100LC and PROCESSOR_8000. They are selected from the pa-risc architecture specified for the target machine when configuring. PROCESSOR_8000 is the default. PROCESSOR_7100LC is selected when the target is a hppa1* machine.

The PROCESSOR_8000 model is not well suited to older processors. Thus, it is important to completely specify the machine architecture when configuring if you want a model other than PROCESSOR_8000. The macro TARGET_SCHED_DEFAULT can be defined in BOOT_CFLAGS if a different default scheduling model is desired.

As of GCC 4.0, GCC uses the UNIX 95 namespace for HP-UX 10.10 through 11.00, and the UNIX 98 namespace for HP-UX 11.11 and later. This namespace change might cause problems when bootstrapping with an earlier version of GCC or the HP compiler as essentially the same namespace is required for an entire build. This problem can be avoided in a number of ways. With HP cc, UNIX_STD can be set to 95 or 98. Another way is to add an appropriate set of predefines to CC. The description for the munix= option contains a list of the predefines used with each standard.

More specific information to hppa*-hp-hpux* targets follows.


For hpux10.20, we highly recommend you pick up the latest sed patch PHCO_19798 from HP. HP has two sites which provide patches free of charge:

The C++ ABI has changed incompatibly in GCC 4.0. COMDAT subspaces are used for one-only code and data. This resolves many of the previous problems in using C++ on this target. However, the ABI is not compatible with the one implemented under HP-UX 11 using secondary definitions.


GCC 3.0 and up support HP-UX 11. GCC 2.95.x is not supported and cannot be used to compile GCC 3.0 and up.

The libffi and libjava libraries haven't been ported to 64-bit HP-UX and don't build.

Refer to binaries for information about obtaining precompiled GCC binaries for HP-UX. Precompiled binaries must be obtained to build the Ada language as it can't be bootstrapped using C. Ada is only available for the 32-bit PA-RISC runtime.

Starting with GCC 3.4 an ISO C compiler is required to bootstrap. The bundled compiler supports only traditional C; you will need either HP's unbundled compiler, or a binary distribution of GCC.

It is possible to build GCC 3.3 starting with the bundled HP compiler, but the process requires several steps. GCC 3.3 can then be used to build later versions. The fastjar program contains ISO C code and can't be built with the HP bundled compiler. This problem can be avoided by not building the Java language. For example, use the --enable-languages="c,c++,f77,objc" option in your configure command.

There are several possible approaches to building the distribution. Binutils can be built first using the HP tools. Then, the GCC distribution can be built. The second approach is to build GCC first using the HP tools, then build binutils, then rebuild GCC. There have been problems with various binary distributions, so it is best not to start from a binary distribution.

On 64-bit capable systems, there are two distinct targets. Different installation prefixes must be used if both are to be installed on the same system. The hppa[1-2]*-hp-hpux11* target generates code for the 32-bit PA-RISC runtime architecture and uses the HP linker. The hppa64-hp-hpux11* target generates 64-bit code for the PA-RISC 2.0 architecture.

The script config.guess now selects the target type based on the compiler detected during configuration. You must define PATH or CC so that configure finds an appropriate compiler for the initial bootstrap. When CC is used, the definition should contain the options that are needed whenever CC is used.

Specifically, options that determine the runtime architecture must be in CC to correctly select the target for the build. It is also convenient to place many other compiler options in CC. For example, CC="cc -Ac +DA2.0W -Wp,-H16376 -D_CLASSIC_TYPES -D_HPUX_SOURCE" can be used to bootstrap the GCC 3.3 branch with the HP compiler in 64-bit K&R/bundled mode. The +DA2.0W option will result in the automatic selection of the hppa64-hp-hpux11* target. The macro definition table of cpp needs to be increased for a successful build with the HP compiler. _CLASSIC_TYPES and _HPUX_SOURCE need to be defined when building with the bundled compiler, or when using the -Ac option. These defines aren't necessary with -Ae.

It is best to explicitly configure the hppa64-hp-hpux11* target with the --with-ld=... option. This overrides the standard search for ld. The two linkers supported on this target require different commands. The default linker is determined during configuration. As a result, it's not possible to switch linkers in the middle of a GCC build. This has been reported to sometimes occur in unified builds of binutils and GCC.

A recent linker patch must be installed for the correct operation of GCC 3.3 and later. PHSS_26559 and PHSS_24304 are the oldest linker patches that are known to work. They are for HP-UX 11.00 and 11.11, respectively. PHSS_24303, the companion to PHSS_24304, might be usable but it hasn't been tested. These patches have been superseded. Consult the HP patch database to obtain the currently recommended linker patch for your system.

The patches are necessary for the support of weak symbols on the 32-bit port, and for the running of initializers and finalizers. Weak symbols are implemented using SOM secondary definition symbols. Prior to HP-UX 11, there are bugs in the linker support for secondary symbols. The patches correct a problem of linker core dumps creating shared libraries containing secondary symbols, as well as various other linking issues involving secondary symbols.

GCC 3.3 uses the ELF DT_INIT_ARRAY and DT_FINI_ARRAY capabilities to run initializers and finalizers on the 64-bit port. The 32-bit port uses the linker +init and +fini options for the same purpose. The patches correct various problems with the +init/+fini options, including program core dumps. Binutils 2.14 corrects a problem on the 64-bit port resulting from HP's non-standard use of the .init and .fini sections for array initializers and finalizers.

Although the HP and GNU linkers are both supported for the hppa64-hp-hpux11* target, it is strongly recommended that the HP linker be used for link editing on this target.

At this time, the GNU linker does not support the creation of long branch stubs. As a result, it can't successfully link binaries containing branch offsets larger than 8 megabytes. In addition, there are problems linking shared libraries, linking executables with -static, and with dwarf2 unwind and exception support. It also doesn't provide stubs for internal calls to global functions in shared libraries, so these calls can't be overloaded.

The HP dynamic loader does not support GNU symbol versioning, so symbol versioning is not supported. It may be necessary to disable symbol versioning with --disable-symvers when using GNU ld.

POSIX threads are the default. The optional DCE thread library is not supported, so --enable-threads=dce does not work.


Versions of libstdc++-v3 starting with 3.2.1 require bug fixes present in glibc 2.2.5 and later. More information is available in the libstdc++-v3 documentation.


As of GCC 3.3, binutils 2.13.1 or later is required for this platform. See bug 10877 for more information.

If you receive Signal 11 errors when building on GNU/Linux, then it is possible you have a hardware problem. Further information on this can be found on


The Sun assembler in Solaris 8 and 9 has several bugs and limitations. While GCC works around them, several features are missing, so it is recommended to use the GNU assembler instead. There is no bundled version, but the current version, from GNU binutils 2.20.1, is known to work.

Solaris~2/x86 doesn't support the execution of SSE/SSE2 instructions before Solaris~9 4/04, even if the CPU supports them. Programs will receive SIGILL if they try. The fix is available both in Solaris~9 Update~6 and kernel patch 112234-12 or newer. There is no corresponding patch for Solaris 8. To avoid this problem, -march defaults to pentiumpro on Solaris 8 and 9. If you have the patch installed, you can configure GCC with an appropriate --with-arch option, but need GNU as for SSE2 support.


Use this for Solaris 10 or later on x86 and x86-64 systems. This configuration is supported by GCC 4.0 and later versions only.

It is recommended that you configure GCC to use the GNU assembler in /usr/sfw/bin/gas but the Sun linker, using the options --with-gnu-as --with-as=/usr/sfw/bin/gas --without-gnu-ld --with-ld=/usr/ccs/bin/ld.


IA-64 processor (also known as IPF, or Itanium Processor Family) running GNU/Linux.

If you are using the installed system libunwind library with --with-system-libunwind, then you must use libunwind 0.98 or later.

None of the following versions of GCC has an ABI that is compatible with any of the other versions in this list, with the exception that Red Hat 2.96 and Trillian 000171 are compatible with each other: 3.1, 3.0.2, 3.0.1, 3.0, Red Hat 2.96, and Trillian 000717. This primarily affects C++ programs and programs that create shared libraries. GCC 3.1 or later is recommended for compiling linux, the kernel. As of version 3.1 GCC is believed to be fully ABI compliant, and hence no more major ABI changes are expected.


Building GCC on this target requires the GNU Assembler. The bundled HP assembler will not work. To prevent GCC from using the wrong assembler, the option --with-gnu-as may be necessary.

The GCC libunwind library has not been ported to HPUX. This means that for GCC versions 3.2.3 and earlier, --enable-libunwind-exceptions is required to build GCC. For GCC 3.3 and later, this is the default. For gcc 3.4.3 and later, --enable-libunwind-exceptions is removed and the system libunwind library will always be used.


Support for AIX version 3 and older was discontinued in GCC 3.4.

“out of memory” bootstrap failures may indicate a problem with process resource limits (ulimit). Hard limits are configured in the /etc/security/limits system configuration file.

To speed up the configuration phases of bootstrapping and installing GCC, one may use GNU Bash instead of AIX /bin/sh, e.g.,

        % CONFIG_SHELL=/opt/freeware/bin/bash
        % export CONFIG_SHELL

and then proceed as described in the build instructions, where we strongly recommend specifying an absolute path to invoke srcdir/configure.

Because GCC on AIX is built as a 32-bit executable by default, (although it can generate 64-bit programs) the GMP and MPFR libraries required by gfortran must be 32-bit libraries. Building GMP and MPFR as static archive libraries works better than shared libraries.

Errors involving alloca when building GCC generally are due to an incorrect definition of CC in the Makefile or mixing files compiled with the native C compiler and GCC. During the stage1 phase of the build, the native AIX compiler must be invoked as cc (not xlc). Once configure has been informed of xlc, one needs to use make distclean to remove the configure cache files and ensure that CC environment variable does not provide a definition that will confuse configure. If this error occurs during stage2 or later, then the problem most likely is the version of Make (see above).

The native as and ld are recommended for bootstrapping on AIX 4 and required for bootstrapping on AIX 5L. The GNU Assembler reports that it supports WEAK symbols on AIX 4, which causes GCC to try to utilize weak symbol functionality although it is not supported. The GNU Assembler and Linker do not support AIX 5L sufficiently to bootstrap GCC. The native AIX tools do interoperate with GCC.

Building libstdc++.a requires a fix for an AIX Assembler bug APAR IY26685 (AIX 4.3) or APAR IY25528 (AIX 5.1). It also requires a fix for another AIX Assembler bug and a co-dependent AIX Archiver fix referenced as APAR IY53606 (AIX 5.2) or a APAR IY54774 (AIX 5.1)

libstdc++ in GCC 3.4 increments the major version number of the shared object and GCC installation places the libstdc++.a shared library in a common location which will overwrite the and GCC 3.3 version of the shared library. Applications either need to be re-linked against the new shared library or the GCC 3.1 and GCC 3.3 versions of the libstdc++ shared object needs to be available to the AIX runtime loader. The GCC 3.1, if present, and GCC 3.3 shared objects can be installed for runtime dynamic loading using the following steps to set the F_LOADONLY flag in the shared object for each multilib libstdc++.a installed:

Extract the shared objects from the currently installed libstdc++.a archive:

        % ar -x libstdc++.a

Enable the F_LOADONLY flag so that the shared object will be available for runtime dynamic loading, but not linking:

        % strip -e

Archive the runtime-only shared object in the GCC 3.4 libstdc++.a archive:

        % ar -q libstdc++.a

Linking executables and shared libraries may produce warnings of duplicate symbols. The assembly files generated by GCC for AIX always have included multiple symbol definitions for certain global variable and function declarations in the original program. The warnings should not prevent the linker from producing a correct library or runnable executable.

AIX 4.3 utilizes a “large format” archive to support both 32-bit and 64-bit object modules. The routines provided in AIX 4.3.0 and AIX 4.3.1 to parse archive libraries did not handle the new format correctly. These routines are used by GCC and result in error messages during linking such as “not a COFF file”. The version of the routines shipped with AIX 4.3.1 should work for a 32-bit environment. The -g option of the archive command may be used to create archives of 32-bit objects using the original “small format”. A correct version of the routines is shipped with AIX 4.3.2 and above.

Some versions of the AIX binder (linker) can fail with a relocation overflow severe error when the -bbigtoc option is used to link GCC-produced object files into an executable that overflows the TOC. A fix for APAR IX75823 (OVERFLOW DURING LINK WHEN USING GCC AND -BBIGTOC) is available from IBM Customer Support and from its website as PTF U455193.

The AIX linker (bos.rte.bind_cmds Level will dump core with a segmentation fault when invoked by any version of GCC. A fix for APAR IX87327 is available from IBM Customer Support and from its website as PTF U461879. This fix is incorporated in AIX 4.3.3 and above.

The initial assembler shipped with AIX 4.3.0 generates incorrect object files. A fix for APAR IX74254 (64BIT DISASSEMBLED OUTPUT FROM COMPILER FAILS TO ASSEMBLE/BIND) is available from IBM Customer Support and from its website as PTF U453956. This fix is incorporated in AIX 4.3.1 and above.

AIX provides National Language Support (NLS). Compilers and assemblers use NLS to support locale-specific representations of various data formats including floating-point numbers (e.g., . vs , for separating decimal fractions). There have been problems reported where GCC does not produce the same floating-point formats that the assembler expects. If one encounters this problem, set the LANG environment variable to C or En_US.

By default, GCC for AIX 4.1 and above produces code that can be used on both Power or PowerPC processors.

A default can be specified with the -mcpu=cpu_type switch and using the configure option --with-cpu-cpu_type.


Vitesse IQ2000 processors. These are used in embedded applications. There are no standard Unix configurations.


Renesas M32C processor. This configuration is intended for embedded systems.


Renesas M32R processor. This configuration is intended for embedded systems.


Motorola 68HC11 family micro controllers. These are used in embedded applications. There are no standard Unix configurations.


Motorola 68HC12 family micro controllers. These are used in embedded applications. There are no standard Unix configurations.


By default, m68k-*-aout, m68k-*-coff*, m68k-*-elf*, m68k-*-rtems, m68k-*-uclinux and m68k-*-linux build libraries for both M680x0 and ColdFire processors. If you only need the M680x0 libraries, you can omit the ColdFire ones by passing --with-arch=m68k to configure. Alternatively, you can omit the M680x0 libraries by passing --with-arch=cf to configure. These targets default to 5206 or 5475 code as appropriate for the target system when configured with --with-arch=cf and 68020 code otherwise.

The m68k-*-netbsd and m68k-*-openbsd targets also support the --with-arch option. They will generate ColdFire CFV4e code when configured with --with-arch=cf and 68020 code otherwise.

You can override the default processors listed above by configuring with --with-cpu=target. This target can either be a -mcpu argument or one of the following values: m68000, m68010, m68020, m68030, m68040, m68060, m68020-40 and m68020-60.


GCC 4.3 changed the uClinux configuration so that it uses the m68k-linux-gnu ABI rather than the m68k-elf ABI. It also added improved support for C++ and flat shared libraries, both of which were ABI changes. However, you can still use the original ABI by configuring for m68k-uclinuxoldabi or m68k-vendor-uclinuxoldabi.


If on a MIPS system you get an error message saying “does not have gp sections for all it's [sic] sectons [sic]”, don't worry about it. This happens whenever you use GAS with the MIPS linker, but there is not really anything wrong, and it is okay to use the output file. You can stop such warnings by installing the GNU linker.

It would be nice to extend GAS to produce the gp tables, but they are optional, and there should not be a warning about their absence.

The libstdc++ atomic locking routines for MIPS targets requires MIPS II and later. A patch went in just after the GCC 3.3 release to make mips*-*-* use the generic implementation instead. You can also configure for mipsel-elf as a workaround. The mips*-*-linux* target continues to use the MIPS II routines. More work on this is expected in future releases.

The built-in __sync_* functions are available on MIPS II and later systems and others that support the ll, sc and sync instructions. This can be overridden by passing --with-llsc or --without-llsc when configuring GCC. Since the Linux kernel emulates these instructions if they are missing, the default for mips*-*-linux* targets is --with-llsc. The --with-llsc and --without-llsc configure options may be overridden at compile time by passing the -mllsc or -mno-llsc options to the compiler.

MIPS systems check for division by zero (unless -mno-check-zero-division is passed to the compiler) by generating either a conditional trap or a break instruction. Using trap results in smaller code, but is only supported on MIPS II and later. Also, some versions of the Linux kernel have a bug that prevents trap from generating the proper signal (SIGFPE). To enable the use of break, use the --with-divide=breaks configure option when configuring GCC. The default is to use traps on systems that support them.

Cross-compilers for the MIPS as target using the MIPS assembler currently do not work, because the auxiliary programs mips-tdump.c and mips-tfile.c can't be compiled on anything but a MIPS. It does work to cross compile for a MIPS if you use the GNU assembler and linker.

The assembler from GNU binutils 2.17 and earlier has a bug in the way it sorts relocations for REL targets (o32, o64, EABI). This can cause bad code to be generated for simple C++ programs. Also the linker from GNU binutils versions prior to 2.17 has a bug which causes the runtime linker stubs in very large programs, like, to be incorrectly generated. GNU Binutils 2.18 and later (and snapshots made after Nov. 9, 2006) should be free from both of these problems.


In order to compile GCC on an SGI running IRIX 5, the compiler_dev.hdr subsystem must be installed from the IDO CD-ROM supplied by SGI. It is also available for download from

If you use the MIPS C compiler to bootstrap, it may be necessary to increase its table size for switch statements with the -Wf,-XNg1500 option. If you use the -O2 optimization option, you also need to use -Olimit 3000.

To enable debugging under IRIX 5, you must use GNU binutils 2.15 or later, and use the --with-gnu-ld configure option when configuring GCC. You need to use GNU ar and nm, also distributed with GNU binutils.

Some users have reported that /bin/sh will hang during bootstrap. This problem can be avoided by running the commands:

        % CONFIG_SHELL=/bin/ksh
        % export CONFIG_SHELL

before starting the build.


If you are using SGI's MIPSpro cc as your bootstrap compiler, you must ensure that the N32 ABI is in use. To test this, compile a simple C file with cc and then run file on the resulting object file. The output should look like:

     test.o: ELF N32 MSB ...

If you see:

     test.o: ELF 32-bit MSB ...


     test.o: ELF 64-bit MSB ...

then your version of cc uses the O32 or N64 ABI by default. You should set the environment variable CC to cc -n32 before configuring GCC.

If you want the resulting gcc to run on old 32-bit systems with the MIPS R4400 CPU, you need to ensure that only code for the mips3 instruction set architecture (ISA) is generated. While GCC 3.x does this correctly, both GCC 2.95 and SGI's MIPSpro cc may change the ISA depending on the machine where GCC is built. Using one of them as the bootstrap compiler may result in mips4 code, which won't run at all on mips3-only systems. For the test program above, you should see:

     test.o: ELF N32 MSB mips-3 ...

If you get:

     test.o: ELF N32 MSB mips-4 ...

instead, you should set the environment variable CC to cc -n32 -mips3 or gcc -mips3 respectively before configuring GCC.

MIPSpro C 7.4 may cause bootstrap failures, due to a bug when inlining memcmp. Either add -U__INLINE_INTRINSICS to the CC environment variable as a workaround or upgrade to MIPSpro C 7.4.1m.

GCC on IRIX 6 is usually built to support the N32, O32 and N64 ABIs. If you build GCC on a system that doesn't have the N64 libraries installed or cannot run 64-bit binaries, you need to configure with --disable-multilib so GCC doesn't try to use them. This will disable building the O32 libraries, too. Look for /usr/lib64/ to see if you have the 64-bit libraries installed.

To enable debugging for the O32 ABI, you must use GNU as from GNU binutils 2.15 or later. You may also use GNU ld, but this is not required and currently causes some problems with Ada.

The --enable-libgcj option is disabled by default: IRIX 6 uses a very low default limit (20480) for the command line length. Although libtool contains a workaround for this problem, at least the N64 libgcj is known not to build despite this, running into an internal error of the native ld. A sure fix is to increase this limit (ncargs) to its maximum of 262144 bytes. If you have root access, you can use the systune command to do this.

wchar_t support in libstdc++ is not available for old IRIX 6.5.x releases, x < 19. The problem cannot be autodetected and in order to build GCC for such targets you need to configure with --disable-wchar_t.

See for more information about using GCC on IRIX platforms.


You can specify a default version for the -mcpu=cpu_type switch by using the configure option --with-cpu-cpu_type.

You will need binutils 2.15 or newer for a working GCC.


PowerPC running Darwin (Mac OS X kernel).

Pre-installed versions of Mac OS X may not include any developer tools, meaning that you will not be able to build GCC from source. Tool binaries are available at (free registration required).

This version of GCC requires at least cctools-590.36. The cctools-590.36 package referenced from will not work on systems older than 10.3.9 (aka darwin7.9.0).


PowerPC system in big endian mode, running System V.4.


PowerPC system in big endian mode running Linux.


PowerPC system in big endian mode running NetBSD.


Embedded PowerPC system in big endian mode for use in running under the PSIM simulator.


Embedded PowerPC system in big endian mode.


PowerPC system in little endian mode, running System V.4.


Embedded PowerPC system in little endian mode for use in running under the PSIM simulator.


Embedded PowerPC system in little endian mode.


S/390 system running GNU/Linux for S/390.


zSeries system (64-bit) running GNU/Linux for zSeries.


zSeries system (64-bit) running TPF. This platform is supported as cross-compilation target only.


Sun does not ship a C compiler with Solaris 2. To bootstrap and install GCC you first have to install a pre-built compiler, see the binaries page for details.

The Solaris 2 /bin/sh will often fail to configure libstdc++-v3, boehm-gc or libjava. We therefore recommend using the following initial sequence of commands

        % CONFIG_SHELL=/bin/ksh
        % export CONFIG_SHELL

and proceed as described in the configure instructions. In addition we strongly recommend specifying an absolute path to invoke srcdir/configure.

Solaris 2 comes with a number of optional OS packages. Some of these are needed to use GCC fully, namely SUNWarc, SUNWbtool, SUNWesu, SUNWhea, SUNWlibm, SUNWsprot, and SUNWtoo. If you did not install all optional packages when installing Solaris 2, you will need to verify that the packages that GCC needs are installed.

To check whether an optional package is installed, use the pkginfo command. To add an optional package, use the pkgadd command. For further details, see the Solaris 2 documentation.

Trying to use the linker and other tools in /usr/ucb to install GCC has been observed to cause trouble. For example, the linker may hang indefinitely. The fix is to remove /usr/ucb from your PATH.

The build process works more smoothly with the legacy Sun tools so, if you have /usr/xpg4/bin in your PATH, we recommend that you place /usr/bin before /usr/xpg4/bin for the duration of the build.

We recommend the use of GNU binutils 2.14 or later, or the vendor tools (Sun as, Sun ld). Note that your mileage may vary if you use a combination of the GNU tools and the Sun tools: while the combination GNU as + Sun ld should reasonably work, the reverse combination Sun as + GNU ld is known to cause memory corruption at runtime in some cases for C++ programs.

The stock GNU binutils 2.15 release is broken on this platform because of a single bug. It has been fixed on the 2.15 branch in the CVS repository. You can obtain a working version by checking out the binutils-2_15-branch from the CVS repository or applying the patch to the release.

We recommend the use of GNU binutils 2.16 or later in conjunction with GCC 4.x, or the vendor tools (Sun as, Sun ld). However, for Solaris 10 and above, an additional patch is required in order for the GNU linker to be able to cope with a new flavor of shared libraries. You can obtain a working version by checking out the binutils-2_16-branch from the CVS repository or applying the patch to the release.

Sun bug 4296832 turns up when compiling X11 headers with GCC 2.95 or newer: g++ will complain that types are missing. These headers assume that omitting the type means int; this assumption worked for C89 but is wrong for C++, and is now wrong for C99 also.

g++ accepts such (invalid) constructs with the option -fpermissive; it will assume that any missing type is int (as defined by C89).

There are patches for Solaris 7 (108376-21 or newer for SPARC, 108377-20 for Intel), and Solaris 8 (108652-24 or newer for SPARC, 108653-22 for Intel) that fix this bug.

Sun bug 4927647 sometimes causes random spurious testsuite failures related to missing diagnostic output. This bug doesn't affect GCC itself, rather it is a kernel bug triggered by the expect program which is used only by the GCC testsuite driver. When the bug causes the expect program to miss anticipated output, extra testsuite failures appear.

There are patches for Solaris 8 (117350-12 or newer for SPARC, 117351-12 or newer for Intel) and Solaris 9 (117171-11 or newer for SPARC, 117172-11 or newer for Intel) that address this problem.


When GCC is configured to use binutils 2.14 or later the binaries produced are smaller than the ones produced using Sun's native tools; this difference is quite significant for binaries containing debugging information.

Starting with Solaris 7, the operating system is capable of executing 64-bit SPARC V9 binaries. GCC 3.1 and later properly supports this; the -m64 option enables 64-bit code generation. However, if all you want is code tuned for the UltraSPARC CPU, you should try the -mtune=ultrasparc option instead, which produces code that, unlike full 64-bit code, can still run on non-UltraSPARC machines.

When configuring on a Solaris 7 or later system that is running a kernel that supports only 32-bit binaries, one must configure with --disable-multilib, since we will not be able to build the 64-bit target libraries.

GCC 3.3 and GCC 3.4 trigger code generation bugs in earlier versions of the GNU compiler (especially GCC 3.0.x versions), which lead to the miscompilation of the stage1 compiler and the subsequent failure of the bootstrap process. A workaround is to use GCC 3.2.3 as an intermediary stage, i.e. to bootstrap that compiler with the base compiler and then use it to bootstrap the final compiler.

GCC 3.4 triggers a code generation bug in versions 5.4 (Sun ONE Studio 7) and 5.5 (Sun ONE Studio 8) of the Sun compiler, which causes a bootstrap failure in form of a miscompilation of the stage1 compiler by the Sun compiler. This is Sun bug 4974440. This is fixed with patch 112760-07.

GCC 3.4 changed the default debugging format from STABS to DWARF-2 for 32-bit code on Solaris 7 and later. If you use the Sun assembler, this change apparently runs afoul of Sun bug 4910101 (which is referenced as a x86-only problem by Sun, probably because they do not use DWARF-2). A symptom of the problem is that you cannot compile C++ programs like groff 1.19.1 without getting messages similar to the following:

     ld: warning: relocation error: R_SPARC_UA32: ...
       external symbolic relocation against non-allocatable section
       .debug_info cannot be processed at runtime: relocation ignored.

To work around this problem, compile with -gstabs+ instead of plain -g.

When configuring the GNU Multiple Precision Library (GMP) or the MPFR library on a Solaris 7 or later system, the canonical target triplet must be specified as the build parameter on the configure line. This triplet can be obtained by invoking ./config.guess in the toplevel source directory of GCC (and not that of GMP or MPFR). For example on a Solaris 7 system:

        % ./configure --build=sparc-sun-solaris2.7 --prefix=xxx


Sun patch 107058-01 (1999-01-13) for Solaris 7/SPARC triggers a bug in the dynamic linker. This problem (Sun bug 4210064) affects GCC 2.8 and later, including all EGCS releases. Sun formerly recommended 107058-01 for all Solaris 7 users, but around 1999-09-01 it started to recommend it only for people who use Sun's compilers.

Here are some workarounds to this problem:

GCC 3.3 triggers a bug in version 5.0 Alpha 03/27/98 of the Sun assembler, which causes a bootstrap failure when linking the 64-bit shared version of libgcc. A typical error message is:

     ld: fatal: relocation error: R_SPARC_32: file libgcc/sparcv9/_muldi3.o:
       symbol <unknown>:  offset 0xffffffff7ec133e7 is non-aligned.

This bug has been fixed in the final 5.0 version of the assembler.

A similar problem was reported for version Sun WorkShop 6 99/08/18 of the Sun assembler, which causes a bootstrap failure with GCC 4.0.0:

     ld: fatal: relocation error: R_SPARC_DISP32:
       file .libs/libstdc++.lax/libsupc++convenience.a/vterminate.o:
         symbol <unknown>: offset 0xfccd33ad is non-aligned

This bug has been fixed in more recent revisions of the assembler.


GCC versions 3.0 and higher require binutils 2.11.2 and glibc 2.2.4 or newer on this platform. All earlier binutils and glibc releases mishandled unaligned relocations on sparc-*-* targets.


When configuring the GNU Multiple Precision Library (GMP) or the MPFR library, the canonical target triplet must be specified as the build parameter on the configure line. For example on a Solaris 7 system:

        % ./configure --build=sparc64-sun-solaris2.7 --prefix=xxx

The following compiler flags must be specified in the configure step in order to bootstrap this target with the Sun compiler:

        % CC="cc -xarch=v9 -xildoff" srcdir/configure [options] [target]

-xarch=v9 specifies the SPARC-V9 architecture to the Sun toolchain and -xildoff turns off the incremental linker.


This is a synonym for sparc64-*-solaris2*.


Support for VxWorks is in flux. At present GCC supports only the very recent VxWorks 5.5 (aka Tornado 2.2) release, and only on PowerPC. We welcome patches for other architectures supported by VxWorks 5.5. Support for VxWorks AE would also be welcome; we believe this is merely a matter of writing an appropriate “configlette” (see below). We are not interested in supporting older, a.out or COFF-based, versions of VxWorks in GCC 3.

VxWorks comes with an older version of GCC installed in $WIND_BASE/host; we recommend you do not overwrite it. Choose an installation prefix entirely outside $WIND_BASE. Before running configure, create the directories prefix and prefix/bin. Link or copy the appropriate assembler, linker, etc. into prefix/bin, and set your PATH to include that directory while running both configure and make.

You must give configure the --with-headers=$WIND_BASE/target/h switch so that it can find the VxWorks system headers. Since VxWorks is a cross compilation target only, you must also specify --target=target. configure will attempt to create the directory prefix/target/sys-include and copy files into it; make sure the user running configure has sufficient privilege to do so.

GCC's exception handling runtime requires a special “configlette” module, contrib/gthr_supp_vxw_5x.c. Follow the instructions in that file to add the module to your kernel build. (Future versions of VxWorks will incorporate this module.)

x86_64-*-*, amd64-*-*

GCC supports the x86-64 architecture implemented by the AMD64 processor (amd64-*-* is an alias for x86_64-*-*) on GNU/Linux, FreeBSD and NetBSD. On GNU/Linux the default is a bi-arch compiler which is able to generate both 64-bit x86-64 and 32-bit x86 code (via the -m32 switch).


This target is intended for embedded Xtensa systems using the newlib C library. It uses ELF but does not support shared objects. Designed-defined instructions specified via the Tensilica Instruction Extension (TIE) language are only supported through inline assembly.

The Xtensa configuration information must be specified prior to building GCC. The include/xtensa-config.h header file contains the configuration information. If you created your own Xtensa configuration with the Xtensa Processor Generator, the downloaded files include a customized copy of this header file, which you can use to replace the default header file.


This target is for Xtensa systems running GNU/Linux. It supports ELF shared objects and the GNU C library (glibc). It also generates position-independent code (PIC) regardless of whether the -fpic or -fPIC options are used. In other respects, this target is the same as the xtensa*-*-elf target.

Microsoft Windows

Intel 16-bit versions

The 16-bit versions of Microsoft Windows, such as Windows 3.1, are not supported.

However, the 32-bit port has limited support for Microsoft Windows 3.11 in the Win32s environment, as a target only. See below.

Intel 32-bit versions

The 32-bit versions of Windows, including Windows 95, Windows NT, Windows XP, and Windows Vista, are supported by several different target platforms. These targets differ in which Windows subsystem they target and which C libraries are used.

Intel 64-bit versions

GCC contains support for x86-64 using the mingw-w64 runtime library, available from This library should be used with the target triple x86_64-pc-mingw32.

Presently Windows for Itanium is not supported.

Windows CE

Windows CE is supported as a target only on ARM (arm-wince-pe), Hitachi SuperH (sh-wince-pe), and MIPS (mips-wince-pe).

Other Windows Platforms

GCC no longer supports Windows NT on the Alpha or PowerPC.

GCC no longer supports the Windows POSIX subsystem. However, it does support the Interix subsystem. See above.

Old target names including *-*-winnt and *-*-windowsnt are no longer used.

PW32 (i386-pc-pw32) support was never completed, and the project seems to be inactive. See for more information.

UWIN support has been removed due to a lack of maintenance.


Ports of GCC are included with the Cygwin environment.

GCC will build under Cygwin without modification; it does not build with Microsoft's C++ compiler and there are no plans to make it do so.

Cygwin can be compiled with i?86-pc-cygwin.


The Interix target is used by OpenNT, Interix, Services For UNIX (SFU), and Subsystem for UNIX-based Applications (SUA). Applications compiled with this target run in the Interix subsystem, which is separate from the Win32 subsystem. This target was last known to work in GCC 3.3.

For more information, see


GCC will build with and support only MinGW runtime 3.12 and later. Earlier versions of headers are incompatible with the new default semantics of extern inline in -std=c99 and -std=gnu99 modes.


GCC does not currently support OS/2. However, Andrew Zabolotny has been working on a generic OS/2 port with pgcc. The current code can be found at

Older systems

GCC contains support files for many older (1980s and early 1990s) Unix variants. For the most part, support for these systems has not been deliberately removed, but it has not been maintained for several years and may suffer from bitrot.

Starting with GCC 3.1, each release has a list of “obsoleted” systems. Support for these systems is still present in that release, but configure will fail unless the --enable-obsolete option is given. Unless a maintainer steps forward, support for these systems will be removed from the next release of GCC.

Support for old systems as hosts for GCC can cause problems if the workarounds for compiler, library and operating system bugs affect the cleanliness or maintainability of the rest of GCC. In some cases, to bring GCC up on such a system, if still possible with current GCC, may require first installing an old version of GCC which did work on that system, and using it to compile a more recent GCC, to avoid bugs in the vendor compiler. Old releases of GCC 1 and GCC 2 are available in the old-releases directory on the GCC mirror sites. Header bugs may generally be avoided using fixincludes, but bugs or deficiencies in libraries and the operating system may still cause problems.

Support for older systems as targets for cross-compilation is less problematic than support for them as hosts for GCC; if an enthusiast wishes to make such a target work again (including resurrecting any of the targets that never worked with GCC 2, starting from the last version before they were removed), patches following the usual requirements would be likely to be accepted, since they should not affect the support for more modern targets.

For some systems, old versions of GNU binutils may also be useful, and are available from pub/binutils/old-releases on mirror sites.

Some of the information on specific systems above relates to such older systems, but much of the information about GCC on such systems (which may no longer be applicable to current GCC) is to be found in the GCC texinfo manual.

all ELF targets (SVR4, Solaris 2, etc.)

C++ support is significantly better on ELF targets if you use the GNU linker; duplicate copies of inlines, vtables and template instantiations will be discarded automatically.

Return to the GCC Installation page