top button
Flag Notify
    Connect to us
      Facebook Login
      Site Registration Why to Join

Facebook Login
Site Registration

Problem build gcc 5.1 with openacc support for nvptx

+3 votes
277 views

I am following instructions from
https://gcc.gnu.org/wiki/Offloading

I'm using the configure line for Nvidia ptx. The build stops in configuring nvptx-none/libgfortran with error:

checking whether symbol versioning is supported... configure: error: Link tests are not allowed after GCC_NO_EXECUTABLES.

My host OS is CentOS 6.5. I'm using gcc 4.7 (built with system gcc 4.4) to build gcc 5.1

Can someone give me hint to look where the problem could come from ?

posted May 22, 2015 by anonymous

Looking for an answer?  Promote on:
Facebook Share Button Twitter Share Button LinkedIn Share Button

Similar Questions
0 votes

We are trying to build our code base for CentOS 7 machine with x86_64 architecture on CentOS6 with x86_64.

For that we are trying to build a centos7 Toolchain. All the gcc/ld/ar should be statically linked, so that we are ensuring that it will not uses any native centos6 stuff.

In this process, we are able to generate statically linked GCC/AR/LD. But problem was this GCC not generating a statically linked binaries and fails while checking flags during configure at "checking static
flag -static ". Due to this while running those binaries it refers to native libgcc and fails.

We are using the below command for configure.

../gcc../configure -prefix=/mnt/data0/toolchain-vm-temp2 --build=x86_64-CentOS7-linux-gnu-with sysroot=/mnt/data0/tools/gnutools/toolchain-vm --disable-nls --disable-multilib --enable-languages=c,c++ --disable-sim --enable-symvers=gnu --enable__cxa_atexit --enable-lto --with-gnu-ld --enable-static

Can Anyone provide solution or how to achieve this task.

0 votes

Currently I compile gcc version 6.2.1 20161117 and I'm running very simple code that use 32 gangs,

I compile it with gcc -fopenacc -o3 -o piexample.x piexample.c with the following env.

export GOMP_DEBUG=1
export ACC_DEVICE_TYPE=nvidia

The problem is that anyone number of gangs that i use, the result is always the same: only 1 gang.

GOACC_parallel_keyed: mapnum=1, hostaddrs=0x7fffc872dff0, size=0x6012c0, kinds=0x6012b8
nvptx_exec: prepare mappings
nvptx_exec: kernel main$_omp_fn$0: launch gangs=1, workers=1, vectors=32
nvptx_exec: kernel main$_omp_fn$0: finished
GOACC_data_end: restore mappings
GOACC_data_end: mappings restored

If compile it with PGI compiler the number of gangs is setter correctly,

main:
12, Generating copy(pi)
14, Loop is parallelizable
Accelerator kernel generated
Generating Tesla code
14, #pragma acc loop gang(32), worker(8), vector(8) /*blockIdx.x threadIdx.y threadIdx.x */
16, Generating implicit reduction(+:pi)

Some ideas about the problem ?

+2 votes

I am building gcc 4.8.1 in a Red Hat 6.4 64 bit machine. I am building gnat first and it is build as 32 bit i686 architecture. I used gnat-gpl-2013-i686-pc-linux-gnu-bin for gnat and it created a 32 bit gcc in the directory, I set the path to that so, gcc is now gcc 4.7.4 from /gnat/bin/gcc. When I checked that is 32 bit gcc.

I used this configuration

./gcc-4.8.1/configure --disable-multilib --disable-bootstrap
--disable-install-libiberty --with-system-zlib --enable-clocale=gnu
--enable-shared --enable-lto --enable-threads=posix --enable-__cxa_atexit
--enable-languages=c,c++,fortran,java,ada --prefix=/usr/local/ CFLAGS="-pipe
-march=native -mtune=native -g -O2" CXXFLAGS="-pipe -march=native
-mtune=native -g -O2"

I was able to build the compiler with this configuration and arch is x86_64.

when I tried to compile a file with -m32 option it didn't work and got the error message

crtbegin.o: could not read symbols: File in wrong format.
Collect2: kd returned 1 exit status
+1 vote

In attempts to build the gomp_4.0 branch from svn (on CentOS 6.2), I find that I must remove the -Werror option in STRICT_WARN. I don't find documented how this should be done (in configure ?). I can get past it by repeatedly modifying gcc/Makefile, but this seems wrong.
I have little interest at this stage in finding out why it triggers -Werror=maybe-uninitialized in typeck.c; I'd only like to build the gcc to see if it does anything interesting with OpenMP 4.

+2 votes

we are trying to compile Android Jellybean (JB43) and It uses Toolchain GCC 4.7 with this toolchain we are seeing this error:

Error: selected processor does not support requested special purpose register -- `mrs r5,FPEXC

Is there a way to fix this. I use arm v7-a

Contact Us
+91 9880187415
sales@queryhome.net
support@queryhome.net
#280, 3rd floor, 5th Main
6th Sector, HSR Layout
Bangalore-560102
Karnataka INDIA.
QUERY HOME
...