Compile errors with Assembler messages

Question

I'm working with Point Cloud Library. It's mostly in C++ When I compile it, gives such error:

[  0%] Building CXX object common/CMakeFiles/pcl_common.dir/src/intersections.cpp.o
In file included from /home/lv/pcl-trunk/common/include/pcl/point_types.h:301:0,
                 from /home/lv/pcl-trunk/common/include/pcl/common/impl/common.hpp:41,
                 from /home/lv/pcl-trunk/common/include/pcl/common/common.h:186,
                 from /home/lv/pcl-trunk/common/include/pcl/common/intersections.h:41,
                 from /home/lv/pcl-trunk/common/src/intersections.cpp:38:
/home/lv/pcl-trunk/common/include/pcl/impl/point_types.hpp:1009:68: warning: ‘SHOT’ is deprecated [-Wdeprecated-declarations]
/tmp/ccRLy4Re.s: Assembler messages:
/tmp/ccRLy4Re.s:2488: Error: no such instruction: `vfmadd312ss (%r9),%xmm2,%xmm1'
/tmp/ccRLy4Re.s:2638: Error: no such instruction: `vfmadd312ss (%rdx),%xmm2,%xmm1'
/tmp/ccRLy4Re.s:3039: Error: no such instruction: `vfmadd312ss (%rax,%r11,4),%xmm5,%xmm1'
/tmp/ccRLy4Re.s:3402: Error: no such instruction: `vfmadd312ss (%rax,%r11,4),%xmm5,%xmm1'
/tmp/ccRLy4Re.s:3534: Error: no such instruction: `vfmadd312ss (%rax,%rdx,4),%xmm1,%xmm2'
/tmp/ccRLy4Re.s:3628: Error: no such instruction: `vfmadd312ss (%rax,%rdx,4),%xmm1,%xmm2'
/tmp/ccRLy4Re.s:6103: Error: no such instruction: `vfmadd312ss (%r11),%xmm0,%xmm4'
/tmp/ccRLy4Re.s:6121: Error: no such instruction: `vfmadd312ss (%r11,%rbx,4),%xmm0,%xmm3'
/tmp/ccRLy4Re.s:6131: Error: no such instruction: `vfmadd312ss (%r11,%rbp,4),%xmm0,%xmm2'
/tmp/ccRLy4Re.s:6135: Error: no such instruction: `vfmadd312ss (%r11,%r13,4),%xmm0,%xmm1'
/tmp/ccRLy4Re.s:6344: Error: no such instruction: `vfmadd312ss (%r10),%xmm0,%xmm1'
/tmp/ccRLy4Re.s:11760: Error: no such instruction: `vfnmadd312ss (%rdx),%xmm0,%xmm0'
/tmp/ccRLy4Re.s:13976: Error: no such instruction: `vfmadd312ss (%rdi),%xmm0,%xmm1'
/tmp/ccRLy4Re.s:14125: Error: no such instruction: `vfmadd312ss (%rdx),%xmm0,%xmm1'
make[2]: *** [common/CMakeFiles/pcl_common.dir/src/intersections.cpp.o] Error 1
make[1]: *** [common/CMakeFiles/pcl_common.dir/all] Error 2
make: *** [all] Error 2

Could anyone tell me how to read these errors and give me some suggestions to solve?

Contact

Answer 1

You can use the command below to see what is your CPU (for the compiler): gcc -march=native -Q --help=target | grep march

Then you can add your cpu type in the ccmake CMAKE_CXX_FLAGS option, for me: CMAKE_CXX_FLAGS = -march=corei7-avx

Georgianngeorgianna

How is that going to help? The problem is the assembler being too old to assemble the instructions gcc is emitting. If your CPU supports FMA, gcc -march=native or gcc -march=haswell will emit FMA instructions. Using a lower -march setting that doesn't use FMA instructions (or -mno-fma) will cost you run-time performance, but that's not even what this answer is suggesting. – Bernete

Answer 2

You can add -march=native -mno-avx This worked for me.

In my CMakeLists.txt, I added the above to the CMAKE_CXX_FLAGS :

SET(CMAKE_CXX_FLAGS "-ggdb -O3 -fPIC -std=c++0x -march=native -mno-avx")

[Source can be found in this page.]

Unbalance

This was the solution to successfully compile rgbdslam w ROS via ubuntu 12.04.05 i386 on a core i7 processor. – Daveen

If you want to make a slower executable to work around an old assembler, you should first try -mno-fma and/or -mno-avx2 to just disable FMA instructions but leave AVX enabled. – Bernete

Answer 3

I faced this issue when trying to install glove.

The problem I faced was resolved by upgrading the g++ version installed in my ubuntu 14.04 machine.

My original g++ version was 4.6.4 and I upgraded it to the g++-5 (5.4.1 when I upgraded it).

Ameliaamelie

Answer 4

There are error message from assembler, which means assembler doesn't know these assembly code, for example vfmadd312ss.

This happens when compiler generate some CPU only assembly code, for example intel E5 with arch core-avx2 have extra instruction set advanced vector extensions, but if "as" is tool old to know the new CPUs, this error would happen.

Try a new version Binutils, the latest is 2.29 now.

If you compile code from another tool, for example bazel, make sure it will call the new "as".

Kolk

This is the only good answer: upgrade your assembler to understand AVX2 / FMA instructions so you can create an executable that takes advantage of everything your CPU can do. (Preferably use a recent compiler version, too, that was released after your hardware so it's likely to know how to tune for it. e.g. gcc6.4 is a good choice these days, or the latest gcc7.x release if you're feeling adventurous.) – Bernete

Also worth trying clang, which has its own built-in assembler so doesn't depend on binutils. In many cases it makes better code than gcc, especially for number-crunching. (It unrolls tiny loops by default.) – Bernete

Answer 5

Have you tried using a different compiler?

I had assembler errors in a project using Point Cloud Library. It was possibly due to using Ubuntu 12.04 on an core i7 processor (see https://github.com/uzh-rpg/rpg_svo/issues/7).

I was able to compile by installing the 'clang' compiler

sudo apt-get install clang

and then running cmake with:

cmake -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ .

Orvas

This is a good suggestion: clang has its own built-in assembler, and sometimes makes faster code than gcc. Especially a new version of clang should be better than an old gcc, especially if your gcc install is gimped by not being able to use FMA instructions. – Bernete

Answer 6

I have met the same error messages, I removed the -march=native from CXXFLAGS and CFLAGS in my makefile, it works in my case. But since you didn't put out your makefile, I am not sure this will works in your case, any way you can have a try.

Bonspiel

That works if you want to make a slower binary that doesn't use FMA instructions. This is the worst solution, worse than upgrading your assembler, or using -mno-fma to just disable FMA without removing the AVX and -mtune=native effects of -march=native. – Bernete

@PeterCordes Hi, thanks for the advice, but how to upgrading assembler? I am on a Linux server(Redhat). And in my case, my error message is shown as "no such instruction: 'shlx %edi, %edx,%edx' ", what's the best solution for my case in your opinion? Thanks in advance! – Bonspiel

Install a new binutils to match the newer compiler that's emitting BMI2 instructions, or cross-compile binaries on a different machine. Or just live with your software running slower than it could. Or install clang if you don't want to touch an existing gcc + binutils install: clang has a built-in assembler and doesn't use /usr/bin/as. – Bernete

Answer 7

You probably need to enable advanced vector extensions when compiling (and use toolchain that supports this). Try -march=core-avx2 or something similar.

If you compiled library yourself, it is possible you need to configure it to not use AVX.

Fluellen

Sorry, I am not a experienced programmer. I don't quite catch your idea. Could you explain it a bit more? – Contact

I don't know how this is built, but try to get compiler/assembler to use -march=core-avx2. You will need to pass the switches when configuring (CPPFLAGS="-march=core-avx2) or just edit you buil scripts and add it. – Fluellen

Nope, the problem is that the assembler is too old to support the FMA instructions the compiler is emitting. The only thing you could do with -march is disable new instructions and make a slower binary. – Bernete