Update TWOPI and PIOVER2 to 39 significant digits (issue #428). Added make.inc for gfortran in quadruple precision #433
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
to prevent accuracy loss when REAL or DOUBLE PRECISION types correspond to an extended precision real kind. These constants should be sufficient to work with 128-bit extended-IEEE data type. Rebuilding the modified source using the default gfortran make.inc (with gfortran 7.5.0) does not add any new test failures. The only test failure, which is also present in the master baranch, is: ZDRGEV3: ZGGEV31 returned INFO= 7. N= 12, JTYPE= 16, ISEED=( 2855, 3551, 2406, 2393) Additionally, included a sample build file for quadruple-precision compilation using gfortran (version 4.7 or later). With this config file, both REAL and DOUBLE PRECISION are mapped onto a 128-bit REAL(16) type, with the corresponding change to the COMPLEX and COMPLEX*16. There are numerical failure 6 failures for each type, all in *GBSVX type 6 matrix, with S/D routines losing 4 significant digits, and C/Z routines 5 digits relative to the expected result. I am not really qualified to debug this failure ...
|
Thanks @serguei-patchkovskii ! |
christoph-conrads
pushed a commit
to christoph-conrads/lapack
that referenced
this pull request
May 23, 2021
…ue428 Update TWOPI and PIOVER2 to 39 significant digits (issue Reference-LAPACK#428). Added make.inc for gfortran in quadruple precision
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This resolves #428
Updated values of TWOPI and PIOVER2 to 39 significant digits,
to prevent accuracy loss when REAL or DOUBLE PRECISION types
correspond to an extended precision real kind. These constants
should be sufficient to work with 128-bit extended-IEEE data
type.
In order to fit the constants on a single line, the indent is less than
the default. At least to my eye, this is less visually jarring than splitting
the constant across two lines.
Rebuilding the modified source using the default gfortran make.inc
(with gfortran 7.5.0) does not add any new test failures. The
only test failure, which is also present in the master branch, is:
ZDRGEV3: ZGGEV31 returned INFO= 7.
N= 12, JTYPE= 16, ISEED=( 2855, 3551, 2406, 2393)
Additionally, included a sample build file for quadruple-precision
compilation using gfortran (version 4.7 or later). With this
config file, both REAL and DOUBLE PRECISION are mapped onto
a 128-bit REAL(16) type, with the corresponding change to the
COMPLEX and COMPLEX*16.
With gfortran-7.5.0, there are numerical failure 6 failures for each
type, all in *GBSVX type 6 matrix, with S/D routines losing 4 significant
digits, and C/Z routines 5 digits relative to the expected result.
I am not really qualified to debug this failure ...