Hello all,

I'm little confused about the status of Floating-point support in the or1200.

I see single/double FP in the OR1000 specs, and only single FP in the OR1200 implementation, that's fine.

I see "or1200-fp.S" tests in trunk/orpsocv2/sw/tests/or1200/sim which does HW (apparent in the rtl simulation) computation.

I see other tests (like or1200-float.c) not using the HW FPU, even if the Makefile is modified as:

# All hardware flags
MARCH_FLAGS ?=-mhard-mul -mhard-div -mhard-float

Is the problem with the GCC not emitting the FPU opcodes?
What other problem sources can lead to this behaviour?

Thanks in advance,
Tarek

Hi Tarek,

There was another thread on this recently. It seems that GCC does not correctly issue floating point hardware instructions (at least for compare operations). The release was regression tested with software floating point only (which does work correctly).

You might like to try running GCC regression wiht -mhard-float to see what tests fail. That will give us something to work on.

HTH

Jeremy

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

Hello,

I don't know how to run GCC regression tests,
But the orpsoc2 test (or1200-fp.S) report the following single-precision floating-point failures:


1.0 * 2^127 + 1.0 * 2^127 = +inf: Failed
report(0x7f7fffff);
-1.0 * 2^127 + -1.0 * 2^127 = -inf: Failed
report(0xff7fffff);
+inf - +inf = -NaN: Failed
report(0x7fc00000);
-NaN - -NaN = -NaN: Failed
report(0x7fc00000);
1.0 - +sNaN = +qNaN: Failed
report(0x7fc00000);
1.0 - -sNaN = -qNaN: Failed
report(0xffc00000);
+inf / +inf = -NaN: Failed
report(0x7fc00000);
-inf / -inf = -NaN: Failed
report(0x7fc00000);
-1.0 / +NaN = +NaN: Failed
report(0xffc00000);
-1.0 / -NaN = -NaN: Failed
report(0x7fc00000);
+NaN / -1.0 = +NaN: Failed
report(0xffc00000);
-NaN / -1.0 = -NaN: Failed
report(0x7fc00000);
+NaN / -NaN = +NaN: Failed
report(0xffc00000);
-NaN / -NaN = -NaN: Failed
report(0x7fc00000);
1.0 * 2^127 / 1.0 * 2^-129 = +inf: Failed
report(0x7f7fffff);
-1.0 * 2^127 / -1.0 * 2^-129 = +inf: Failed
report(0x7f7fffff);
1.0 * 2^-129 / 1.0 * 2^127 = +0.0: Failed
report(0x00800000);
-1.0 * 2^-129 / 1.0 * 2^127 = -0.0: Failed
report(0x80800000);
1.0 * 2^-129 / -1.0 * 2^127 = -0.0: Failed
report(0x80800000);
-1.0 * 2^-129 / -1.0 * 2^127 = +0.0: Failed

Hi Tarek

Thanks for this. Would you file a bug report in Bugzilla against GCC (http://bugzilla.opencores.org/). That way the issue won't be forgotten.

Best wishes,

Jeremy