[llvm-dev] Memory overflow during cmake/ninja build

[Ron Brender via llvm-dev]

I'm trying to do a simple build from the git 8.0.0 sources. The sources
seem to build OK but a link step fails from running out of memory. I
need some clues how to figure out where the bottleneck might be.

The cmake command is:

cmake -G Ninja                                          \
    -DLLVM_TARGETS_TO_BUILD=X86                         \
    -DCMAKE_INSTALL_PREFIX=/home/ron/bin/llvm_project/  \
    ~/llvm-project/llvm

All seems to go well up until this link step:

[8/323] Linking CXX executable bin/llvm-lto
FAILED: bin/llvm-lto
: && /usr/bin/c++  -fPIC -fvisibility-inlines-hidden -Werror=date-time
-std=c++11 -Wall -Wextra -Wno-unused-parameter -Wwrite-strings
-Wcast-qual -Wno-missing-field-initializers -pedantic -Wno-long-long
-Wimplicit-fallthrough -Wno-maybe-uninitialized -Wno-class-memaccess
-Wno-noexcept-type -Wdelete-non-virtual-dtor -Wno-comment
-fdiagnostics-color -g -Wl,-allow-shlib-undefined
-Wl,-rpath-link,/home/ron/llvm-project/build-try3/./lib
tools/llvm-lto/CMakeFiles/llvm-lto.dir/llvm-lto.cpp.o  -o bin/llvm-lto 
-Wl,-rpath,"\$ORIGIN/../lib" lib/libLLVMX86CodeGen.a
lib/libLLVMX86AsmParser.a lib/libLLVMX86AsmPrinter.a
lib/libLLVMX86Desc.a lib/libLLVMX86Disassembler.a lib/libLLVMX86Info.a
lib/libLLVMX86Utils.a lib/libLLVMBitReader.a lib/libLLVMBitWriter.a
lib/libLLVMCore.a lib/libLLVMIRReader.a lib/libLLVMLTO.a lib/libLLVMMC.a
lib/libLLVMObject.a lib/libLLVMSupport.a lib/libLLVMTarget.a -lpthread
lib/libLLVMAsmPrinter.a lib/libLLVMDebugInfoDWARF.a
lib/libLLVMGlobalISel.a lib/libLLVMSelectionDAG.a
lib/libLLVMX86AsmPrinter.a lib/libLLVMX86Utils.a
lib/libLLVMMCDisassembler.a lib/libLLVMObjCARCOpts.a lib/libLLVMPasses.a
lib/libLLVMCodeGen.a lib/libLLVMTarget.a lib/libLLVMipo.a
lib/libLLVMBitWriter.a lib/libLLVMIRReader.a lib/libLLVMAsmParser.a
lib/libLLVMLinker.a lib/libLLVMScalarOpts.a
lib/libLLVMAggressiveInstCombine.a lib/libLLVMInstCombine.a
lib/libLLVMInstrumentation.a lib/libLLVMVectorize.a
lib/libLLVMTransformUtils.a lib/libLLVMAnalysis.a lib/libLLVMObject.a
lib/libLLVMBitReader.a lib/libLLVMMCParser.a lib/libLLVMMC.a
lib/libLLVMDebugInfoCodeView.a lib/libLLVMDebugInfoMSF.a
lib/libLLVMProfileData.a lib/libLLVMCore.a lib/libLLVMBinaryFormat.a
lib/libLLVMSupport.a -lrt -ldl -lpthread -lm lib/libLLVMDemangle.a && :
/usr/bin/ld:
/usr/lib/gcc/x86_64-linux-gnu/8/../../../x86_64-linux-gnu/librt.so:
error adding symbols: memory exhausted
collect2: error: ld returned 1 exit status
[9/323] Building CXX object
tools/bugpoint/CMakeFiles/bugpoint.dir/bugpoint.cpp.o
[10/323] Linking CXX shared library lib/libLTO.so.9svn
ninja: build stopped: subcommand failed.

My hardware is a rather ancient HP a6745f (AMD) with 4GB memory, 200 GB
HDD. OS is Kubuntu 18.10.

Is my system just too limited or is there something else I should look
for? What other info can I provide?

Ron

--
Ron Brender
Whose favorite airplane is N6119A, a 1979 Cessna T210.

_______________________________________________
LLVM Developers mailing list
llvm...@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

[Peter Smith via llvm-dev]

By default the build system will do a static link including debug, and
will use up all available cores for parallelsim. If you happen to have
a small amount of memory but multiple cores there can be several
multi-gigabyte links going on in parallel. Some cmake flag
suggestions:
Prevent concurrent links:
-DLLVM_PARALLEL_LINK_JOBS=1
If you don't need debug information:
-DCMAKE_BUILD_TYPE=Release
If you do:
-DBUILD_SHARED_LIBS=True

It has been mentioned that ld.bfd can use significantly more memory
than gold or LLD on debug builds, may be worth trying a different
linker. The -DLLVM_USE_LINKER=linker can be used
(https://llvm.org/docs/CMake.html).

Peter

[Stephan Bergmann via llvm-dev]

On 21/06/2019 16:00, Ron Brender via llvm-dev wrote:
> cmake -G Ninja                                          \
>     -DLLVM_TARGETS_TO_BUILD=X86                         \
>     -DCMAKE_INSTALL_PREFIX=/home/ron/bin/llvm_project/  \
>     ~/llvm-project/llvm

try -DLLVM_PARALLEL_LINK_JOBS=1

[Alex Brachet-Mialot via llvm-dev]

The above solutions are all great ones. +1 to using ld.gold or LLD if you have them. No matter what though linkers just use a ton of memory! If you have the storage to spare you could increase your swap space.

-Alex

[Neil Nelson via llvm-dev]

It has also been suggested to try compiling with clang and linking with
lld. The gold linker should be already installed and is also recommended
to reduce memory use.

apt install clang

apt install lld

4G is very tight. You will want to run just the terminal emulator for
the compile and another running top to see when you go into swap and
what process is doing that. You will need to avoid running anything else
that is not absolutely necessary to make the most memory available.

You can control-c out of the compile sequence and restart at any time.
The sequence will resume.

If you have more than one core and it goes into swap, you might stop the
compile and use

ninja -j 1

to try to squeak through. You can stop the compile, go back to more
cores, and restart in more easy stretches.

I used

cmake -G Ninja -DLLVM_ENABLE_PROJECTS="clang" -DLLVM_USE_LINKER=lld
-DCMAKE_BUILD_TYPE="Release" ../llvm

this morning and it appeared in general to stay below 4G and perhaps
with enough swap you will make it. But when it goes into heavy swap, you
are doing a number on your disk with not much forward gain.

If you change compilers such as from GNU cpp to clang after running the
above cmake you will need to delete the build directory for the new
compiler to be used.

Change the compiler (after apt above) by adding the following lines to
/etc/environment and reboot.
export CC=clang
export CXX=clang++

[Ron Brender via llvm-dev]

Peter, Steve, Alex, Neil,

Thanks for the many suggestions. I will try again this weekend...

Ron

--

Ron Brender
Whose favorite airplane is N6119A, a 1979 Cessna T210.

_______________________________________________