This is my patch to support little endian.

I test the RTL and toolchains by following sequence. It passes all tests on little endian and big endian.

Kuoping

================================================================
cd orpsocv2/sim/run
# define OR1200_IMPL_LITTLEENDIAN on or1200_defines.v
make clean
make rtl-tests # test big endian with define OR1200_IMPL_LITTLEENDIAN
make clean
make OR32LE=1 rtl-tests # test little endian with define OR1200_IMPL_LITTLEENDIAN

# undefined OR1200_IMPL_LITTLEENDIAN on or1200_defines.v
make clean
make rtl-tests # test big endian without define OR1200_IMPL_LITTLEENDIAN

patch file for newlib to fix the endian issue of floating point

I only took a quick look, but from what I saw it looks pretty good.
Patches are more suited for the mailing lists than the forums though,
do you think you could re-post them (unpacked) there?

Stefan

Sure, here is patch for binutils-2.20.1

patch for gcc-4.5.1

patch for orpsocv2

Hi Kuoping,

Thanks for all this. Have you run regression against this tool chain? If that looks good, I'll merge it into the main SVN.

Best wishes,

Jeremy

--
Tel: +44 (1590) 610184
Cell: +44 (7970) 676050
SkypeID: jeremybennett
Email: jeremy.bennett@embecosm.com
Web: www.embecosm.com

Hi Jeremy,

Yes, it passes the regression test by following procedure.

================================================================
cd orpsocv2/sim/run
# define OR1200_IMPL_LITTLEENDIAN on or1200_defines.v
make clean
make rtl-tests # test big endian with define OR1200_IMPL_LITTLEENDIAN
make clean
make OR32LE=1 rtl-tests # test little endian with define OR1200_IMPL_LITTLEENDIAN

# undefined OR1200_IMPL_LITTLEENDIAN on or1200_defines.v
make clean
make rtl-tests # test big endian without define OR1200_IMPL_LITTLEENDIAN ================================================================

There are four patch files:
orpsocv2.patch
gcc-4.5.1.patch
binutils-2.20.1.patch
newlib-1.18.0.patch

By the way, please ignore the previous patch of newlib. Use the new one.

Kuoping

Hi Kuoping

That's not the regression I mean. There are only a handful of tests in rtl-tests, and they are very simple, testing only a few instructions and the very simplest use of the compiler

I mean setting your design up as a target for GNU tool chain regression, and running the tool regressions (make check-binutils, make check-gcc etc). There are about 100,000 tests which check every corner of the tool chain's implementation.

You can see the current regression results for the bare metal tool chain here: http://opencores.org/or1k/Newlib_tool_chain_test_results.

Best wishes,

Jeremy

Hi Jeremy,

OK. I'll try it. Currently, I do not run the regression yet.

Kuoping