Compiling with g++ using multiple cores

Question

Quick question: what is the compiler flag to allow g++ to spawn multiple instances of itself in order to compile large projects quicker (for example 4 source files at a time for a multi-core CPU)?

Solution 1

You can do this with make - with gnu make it is the -j flag (this will also help on a uniprocessor machine).

For example if you want 4 parallel jobs from make:

make -j 4

You can also run gcc in a pipe with

gcc -pipe

This will pipeline the compile stages, which will also help keep the cores busy.

If you have additional machines available too, you might check out distcc, which will farm compiles out to those as well.

Solution 2

There is no such flag, and having one runs against the Unix philosophy of having each tool perform just one function and perform it well. Spawning compiler processes is conceptually the job of the build system. What you are probably looking for is the -j (jobs) flag to GNU make, a la

make -j4

Or you can use pmake or similar parallel make systems.

Solution 3

People have mentioned make but bjam also supports a similar concept. Using bjam -jx instructs bjam to build up to x concurrent commands.

We use the same build scripts on Windows and Linux and using this option halves our build times on both platforms. Nice.

Solution 4

If using make, issue with -j. From man make:

  -j [jobs], --jobs[=jobs]
       Specifies the number of jobs (commands) to run simultaneously.  
       If there is more than one -j option, the last one is effective.
       If the -j option is given without an argument, make will not limit the
       number of jobs that can run simultaneously.

And most notably, if you want to script or identify the number of cores you have available (depending on your environment, and if you run in many environments, this can change a lot) you may use ubiquitous Python function cpu_count():

https://docs.python.org/3/library/multiprocessing.html#multiprocessing.cpu_count

Like this:

make -j $(python3 -c 'import multiprocessing as mp; print(int(mp.cpu_count() * 1.5))')

If you're asking why 1.5 I'll quote user artless-noise in a comment above:

The 1.5 number is because of the noted I/O bound problem. It is a rule of thumb. About 1/3 of the jobs will be waiting for I/O, so the remaining jobs will be using the available cores. A number greater than the cores is better and you could even go as high as 2x.

Solution 5

make will do this for you. Investigate the -j and -l switches in the man page. I don't think g++ is parallelizable.

View more solutions

Comments

[user1568901]

Will it really help? All my compile jobs are I/O bound rather than CPU bound.

[Flexo]

Even if they are I/O bound you can probably keep the I/O load higher when the CPU heavy bits are happening (with just one g++ instance there will be lulls) and possibly gain I/O efficiencies if the scheduler has more choice about what to read from disk next. My experience has been that judicious use of make -j almost always results in some improvement.

[大宝剑]

@BrianKnoblauch But on my machine(real one or in VirtualBox), it's CPU bound, I found that the CPU is busy through 'top' command when compiling.

[大宝剑]

Even if they are I/O bound, we can use gcc's flag '-pipe' to reduce pain.

[Jim Michaels]

just saw this in google: gcc.gnu.org/onlinedocs/libstdc++/manual/…

[underscore_d]

@JimMichaels That's completely unrelated: it's about parallel processing as part of the program at runtime, i.e. after compilation. The question is about compiling in parallel using multiple jobs.

[tofutim]

david-smith.org/blog/2011/07/27/is-compilation-cpu-bound this fellow thinks cpu bound

[Mark Beckwith]

You're -j number should be 1.5x the number of cores you have.

[frankodwyer]

yes, something like that makes sense given there is I/O as well - although may need some tuning if using -pipe as well

[Flexo]

+1, distcc is a useful tool to have in one's arsenal for large builds.

[chriv]

Thanks. I kept trying to pass "-j#" to gcc via CFLAGS/CPPFLAGS/CXXFLAGS. I had completely forgotten that "-j#" was a parameter for GNU make (and not for GCC).

[Alex Bitek]

Why does the -j option for GNU Make needs to be 1.5 x the number of CPU cores?

[artless noise]

The 1.5 number is because of the noted I/O bound problem. It is a rule of thumb. About 1/3 of the jobs will be waiting for I/O, so the remaining jobs will be using the available cores. A number greater than the cores is better and you could even go as high as 2x. See also: Gnu make -j arguments

[rogerdpack]

Looks like there are a few that work "like" distcc as well: stackoverflow.com/questions/5374106/distributed-make/…

[Jim Michaels]

make -j is broken, with mingw-w64 it will cause a long list of compilation errors whereas without project will compile fine. don't recommend it. I recommend submitting a bug report to the gnu make folk.

[Jim Michaels]

gnu.org/software/make/manual/html_node/Parallel.html also gnu.org/software/make/manual/html_node/…

[Piotr Kula]

Will the Raspberry Pi 2 benefit from this flag? Will it be able to compile faster thanks to the new 4 core processor?

[Antonio]

@JimMichaels It could be because dependencies are badly set within your project, (a target starts building even if its dependencies are not ready yet) so that only a sequential build ends up being successful.

[Sebi2020]

no N ist not the number of threads! Many people misunderstand that, but -j N tells make how many processes at once should be spawned, not threads. That's the reason why it is not as performant as MS cl -MT(really multithreaded).

[Lightness Races in Orbit]

"Unix pedantry is not helpful" Good thing it wasn't pedantry then, anonymous editor. Rolled back. Reviewers please pay more attention to what you're doing.

[Nick]

-pipe is not pipeline, its to use pipes instead of temporary files. Uses more memory, but in some cases is faster. If your project is large, it might be worth trying.

[jjpe]

Ok so this is quite a while after the original discussion, BUT: compiling Emacs git master on an AMD Threadripper 1950x with 16 cores using Fedora 27 completely breaks this 1.5x rule of thumb. There are 32 threads available, and 16 physical cores, yet make -j 12 is roughly fastest with approximately 1m15 to 1m20 user time. Increasing the arg to -j only increases build times. So in effect it's more like 3/4 x or 3/8 x depending on whether you want to count SMT "cores" or not. Of course it's entirely possible here that TR just rips through the parallellizable parts quickly...

[NGI]

+1 for mentioning -l option ( does not start a new job unless all previous jobs did terminate ). Otherwise it seems that the linker job begins with not all object files built (as some compilations are still ongoing), so that the linker job fails.

[Ciro Santilli OurBigBook.com]

Most Linux users will likely prefer the shorter: make -j`nproc` with nproc in GNU Coreutils.

[Ed K]

If you're using an SSD, I/O isn't going to be as much of an issue. Just to build on Ciro's comment above, you can do this: make -j $(( $(nproc) + 1 )) (make sure you put spaces where I have them).

[hoefling]

Nice suggestion using python, on systems where nproc isn't available, e.g. in manylinux1 containers, it saves additional time by avoiding running yum update/yum install.

[mercury0114]

what happens if N is too large? E.g. can -j 100 break the system or is N merely an upper bound that is not required to achieve?

[Spike0xff]

despite the claim of non-pedantry, gcc is getting a flag -fparallel-jobs=N Better tell the GCC devs they're doing it wrong.