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

sagemath / sage

Main repository of SageMath

https://www.sagemath.org

Other

1.44k stars 479 forks source link

Numerical noise in devel/sage/sage/symbolic/pynac.pyx #8375

Closed bac7d3ea-3f1b-4826-8464-f0b53d5e12d2 closed 14 years ago

bac7d3ea-3f1b-4826-8464-f0b53d5e12d2 commented 14 years ago

== 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

bac7d3ea-3f1b-4826-8464-f0b53d5e12d2 commented 14 years ago

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]

e14f4152-4982-4ace-8c95-73a0599b109b commented 14 years ago

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.

e14f4152-4982-4ace-8c95-73a0599b109b commented 14 years ago

Reviewer: Mitesh Patel

e14f4152-4982-4ace-8c95-73a0599b109b commented 14 years ago

Author: David Kirkby

bac7d3ea-3f1b-4826-8464-f0b53d5e12d2 commented 14 years ago

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

With grammer/spelling corrections.

bac7d3ea-3f1b-4826-8464-f0b53d5e12d2 commented 14 years ago

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 commented 14 years ago

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 commented 14 years ago

Changed reviewer from Mitesh Patel to Mitesh Patel, Jaap Spies

mwhansen commented 14 years ago

Merged: sage-4.3.4.alpha1