Numerical noise in devel/sage/sage/symbolic/pynac.pyx

[sagetrac-drkirkby]

== The computer hardware & software ==

• Sun Blade 1000
• 2 x 900 MHz UltraSPARC III+ CPUs
• 2 GB RAM
• Solaris 10 03/2005 - the first release of Solaris 10.

== The Sage software ==
Sage 4.3.3 with various patches to get it to compile on Solaris. (The notebook is not working properly though).

== The test failure ==

**********************************************************************
File "/export/home/drkirkby/sage-4.3.3/devel/sage/sage/symbolic/pynac.pyx", line 1282:
    sage: py_exp(float(1))
Expected:
    2.7182818284590451
Got:
    2.7182818284590455
**********************************************************************

Component: porting: Solaris

Author: David Kirkby

Reviewer: Mitesh Patel, Jaap Spies

Merged: sage-4.3.4.alpha1

Issue created by migration from https://trac.sagemath.org/ticket/8375

[sagetrac-drkirkby]

comment:1

The attached patch fixes this. It also has some notes showing

• A high precision value of e
• The IEEE 754 value
• The correctly rounded result
• The value given on the SPARC processor.

drkirkby@redstart:~/sage-4.3.3$ ./sage -t devel/sage/sage/symbolic/pynac.pyx
sage -t  "devel/sage/sage/symbolic/pynac.pyx"
         [156.7 s]

----------------------------------------------------------------------
All tests passed!
Total time for all tests: 156.7 seconds
sage/sage/symbolic/pynac.pyx.3$ drkirkby@redstart:~/sage-4.3.3$ ./sage -t devel/
-bash: drkirkby@redstart:~/sage-4.3.3$: No such file or directory
sage -t  "devel/sage/sage/symbolic/pynac.pyx"
         [156.7 s]

[qed777]

comment:2

The patches at #8374 and here fix the corresponding doctest failures in 4.3.0.1 on t2 and still pass in 4.3.3 on sage.math.

Just to check: Did "sage-devel" agree that this was the best approach to the problem?

I'm not sure if Minh is already reviewing these tickets. To the extent it counts, my review is positive, provided that you fix the [minor, admittedly] spelling / grammatical errors.

[qed777]

Reviewer: Mitesh Patel

[qed777]

Author: David Kirkby

[sagetrac-drkirkby]

Attachment: numerical-noise-on-SPARC.patch.gz

With grammer/spelling corrections.

[sagetrac-drkirkby]

comment:3

There was only one comment from Robert as to whether there was another way to get the SPARC processor to produce the same output as Intel. I mentioned there was, but it would require very significant changes.

The exact same correction for 'e' was made once before for the SPARC processor. It has been implemented numerous time for other processors.

I've corrected the grammar/spelling.

[jaapspies]

comment:4

Replying to @sagetrac-drkirkby:

There was only one comment from Robert as to whether there was another way to get the SPARC processor to produce the same output as Intel. I mentioned there was, but it would require very significant changes.

The exact same correction for 'e' was made once before for the SPARC processor. It has been implemented numerous time for other processors.

I've corrected the grammar/spelling.

From me a positive review.

Jaap

[jaapspies]

Changed reviewer from Mitesh Patel to Mitesh Patel, Jaap Spies

[mwhansen]

Merged: sage-4.3.4.alpha1