New armv6 hard float cross compiler tool chain for x86 windows (gcc tool chain revision 5.2 from July 2015)

[Rabia CHEBAH <chebahrabia at yahoo dot fr>]

Hello,

I successfully built a cross compiler tool chain using the last ct-ng
from the git repo.
As i did not found any cross compiler for windows with a gcc revision >
4.7 (2012) for armv6 (raspberry pi), i prefered to build my own tool chain.
It was not easier as the host is cygwin and the environment offered is
not compatible with ct-ng.
So need to update the binutils for host script and the glibc source
taken, to build without errors.

The details:
host: x86 cygwin
target: armv6-rpi-linux-gnueabi with hardfloat
gcc tool chain revision 5.2 (July 2015) for last standards (c++11 and
14) including c/c++/gfortran compilers
gdb revision 7.9 for debugging
glibc revision: 2.21
sysroot build for headers and libs.
Dependancies: Cygwin1.dll. You need to install Cygwin and add this dll
path into $PATH.

Attached an example with codeblocks under windows using a remote
debugging of linpack benchmark code build with the crosscompiler.
I added this example to show users that it is possible as it was my
first need.

Comparison of the linpack code build using native compiler from
raspberry pi (gcc 4.6) and cross compiler (gcc 5.2 using hard float from
the raspberry pi cpu):

native:  file linpack
linpack: ELF 32-bit LSB executable, ARM, version 1 (SYSV), dynamically
linked (uses shared libs), for GNU/Linux 2.6.26,
BuildID[sha1]=0xab70ae18ea0d66d5ab5784bba66ec462d75c8be3, not stripped

cross-compiler: file Linpack-dynamic-release
Linpack-dynamic-release: ELF 32-bit LSB executable, ARM, version 1
(SYSV), dynamically linked (uses shared libs), for GNU/Linux 3.2.27,
BuildID[sha1]=0x4d92b05e1326911042eb7d30deb0a2bdc6cadacd, not stripped

native build execution: ./linpack
Enter array size (q to quit) [200]: 
Memory required:  315K.


LINPACK benchmark, Double precision.
Machine precision:  15 digits.
Array size 200 X 200.
Average rolled and unrolled performance:

    Reps Time(s) DGEFA   DGESL  OVERHEAD    KFLOPS
----------------------------------------------------
      16   0.50  92.00%   4.00%   4.00%  45777.778
      32   1.01  90.10%   1.98%   7.92%  47254.480
      64   2.03  90.64%   1.48%   7.88%  47001.783
     128   4.04  91.34%   4.21%   4.46%  45540.587
     256   8.07  86.99%   3.84%   9.17%  47963.620
     512  16.14  85.56%   4.89%   9.54%  48160.731

cross-compiler build execution: ./Linpack-dynamic-release
Enter array size (q to quit) [200]: 
Memory required:  315K.


LINPACK benchmark, Double precision.
Machine precision:  15 digits.
Array size 200 X 200.
Average rolled and unrolled performance:

    Reps Time(s) DGEFA   DGESL  OVERHEAD    KFLOPS
----------------------------------------------------
      16   0.50  84.00%   8.00%   8.00%  47768.116
      32   0.99  88.89%   1.01%  10.10%  49378.277
      64   1.98  87.88%   4.04%   8.08%  48293.040
     128   3.95  87.59%   3.54%   8.86%  48829.630
     256   7.88  88.20%   2.66%   9.14%  49102.421
     512  15.92  88.00%   4.08%   7.91%  47963.620

The cross compiler for windows is available on git at this link:
https://github.com/Cheb57/armv6-rpi-linux-gnueabi

Regards,
R. Chebah.

--
For unsubscribe information see http://sourceware.org/lists.html#faq