General

Profile

Berk Hess

Issues

Projects

Activity

04/25/2017

09:39 AM GROMACS Bug #2136: mdrun -noconfout checkpointing issue
My request was rather the other way around. I'd rather not have a confout.gro at all. What's the point of having a se...

04/24/2017

11:02 AM GROMACS Bug #2165 (Rejected): GROMACS-2016 gives the domain decomposition error when doing free energy c...
Have you looked at the error message?
Fatal error:
The initial cell size (0.715508) is smaller than the cell size...
09:17 AM GROMACS Feature #2163: Use better input/output value choices for SIMD tests
Changed the title to reflect the change to feature.
Many test values are such that we only test 3 mantissa bits. Thi...

04/22/2017

10:34 AM GROMACS Bug #2164: SIMD sqrt in double-precision build does not work correctly
Also ignore my other suggestions (I need more coffee, I think).
The clamping to zero for float_min will switch to ...
10:22 AM GROMACS Bug #2164: SIMD sqrt in double-precision build does not work correctly
Ignore the last paragraph in my previous comment. There is no rsqrt in double on x86.
10:17 AM GROMACS Bug #2164: SIMD sqrt in double-precision build does not work correctly
The clamping you suggest is not nice, as you often take the sqrt of a square, you get issues at values of float_min (...
09:20 AM GROMACS Bug #2162 (Resolved): AVX(2) double precision reduce is actual single precision
Applied in changeset commit:8b1ee2835b80e6c6f9eee64f8d0b0170ea46b83c.

04/21/2017

06:37 PM GROMACS Revision 8b1ee283: Fix bugs in double AVX(2)_256 Simd4
The double precision version of reduce() and dotProduct()
returned a float with AVX_256 and AVX2_256.
Only reduce() i...
04:05 PM GROMACS Feature #2163 (Accepted): Use better input/output value choices for SIMD tests
Many of the SIMD unit tests use input and output values that have 0 for all bits beyond single precision. Thus they w...
03:59 PM GROMACS Bug #2162 (Resolved): AVX(2) double precision reduce is actual single precision
The SIMD reduce() function in double precision returns a float instead of a double with AVX_256 and AVX2_256. This is...

Also available in: Atom