Pietro Delugas
4345997d23
removing references to deleted report_clock_pw from build systems
2021-05-28 11:16:18 +02:00
Pietro Delugas
5e171bf17b
fixing and cleaning last merge request
2021-05-28 10:31:51 +02:00
Pietro Delugas
3a9428d7dc
adding all clocks to the timing_info element of xml file
2021-05-27 00:16:02 +02:00
Ye Luo
1095cccbab
Fix two OpenMP bug.
2021-05-23 00:07:47 -05:00
Paolo Giannozzi
802256ff10
Clocks for DFT-D2 and D3 added
2021-05-13 08:52:13 +02:00
Paolo Giannozzi
cebb5c3670
Cleanup of unused variables and of free_unit
2021-05-13 08:51:19 +02:00
Paolo Giannozzi
abd245fe5a
Quick-and-dirty check on incompatible options
...
There is no real reason for such incompatibility, but it requires some thought
2021-05-11 16:32:56 +02:00
Victor Yu
11c8fe176b
Don't call atomic_wfc_gpu when natomwfc is zero
...
Otherwise atomic_wfc_gpu tries to allocate a zero-sized GPU array,
causing a crash at run time.
The same change was also made to the CPU version.
2021-05-06 11:26:57 -04:00
Paolo Giannozzi
4d1a7c9787
Atomic starting wavefunctions can be computed on GPU: the code already exists.
2021-05-01 14:30:38 +02:00
Paolo Giannozzi
02b661e19c
Minor correction
2021-05-01 14:12:20 +02:00
Victor Yu
97fa508ce4
Fix gg_d initialization in post_xml_init
2021-04-27 23:28:06 -04:00
giannozz
b0ce9de399
Merge branch 'cudaclean_again' into 'develop'
...
More CUDA cleanup: g, gg
See merge request QEF/q-e!1398
2021-04-27 19:19:34 +00:00
Paolo Giannozzi
3db123fd98
More cleanup of replicated CUDA variabes that are already present.
...
PS: previous commit sort of fixes issue #319
2021-04-27 14:00:34 +02:00
giannozz
ebf9611dac
Merge branch 'qe-emacs-mode' into 'develop'
...
Further improvements of *.def files and QE_emacs-modes
See merge request QEF/q-e!1396
2021-04-26 17:30:13 +00:00
Paolo Giannozzi
35b242c310
Some more cleaning
2021-04-26 18:43:18 +02:00
Paolo Giannozzi
312a692a50
Compilaztion glitch again
2021-04-24 11:40:54 +02:00
Paolo Giannozzi
dbab41c2bf
Cleaning the cleaning
...
Compilation glitches, make.depend spurious file, missing alignment added to
read_file_new; found the missing alignment in variable-cell CP (in newinit)
2021-04-23 20:36:18 +02:00
Paolo Giannozzi
88b5c8a05c
More CUDA cleanup: g, gg
...
Module "gvect_gpum" deleted, replaced with copy of g, gg onto g_d, gg_d after
initialization of G-vectors, analogously to what is done with mill and mill_d.
NOTA BENE 1: it is not yet possible to move module recvec_subs into recvec and
make g_d and gg_d (and possibly other variables as well) "protected" variables,
because ggen is called, only once, in exx_bands, with a different set of
variables, and the code that does that is totally obscure :-(
NOTA BENE 2: I suspect that somewhere in CP G-vectors have to be rescaled for
variable-cell calculations, but I don't know where this is done. If not fixed,
this might break CP on GPU for variable-cell calculations.
2021-04-23 19:07:27 +02:00
giannozz
1ff9b244f3
Merge branch 'bfgs-failure-1' into 'develop'
...
Catch and exit gracefully when BFGS optimizer fails
See merge request QEF/q-e!1383
2021-04-22 19:21:58 +00:00
Tone Kokalj
a5cc817334
improving description of ADDITIONAL_K_POINTS
2021-04-22 16:41:02 +02:00
Paolo Giannozzi
24c1c859a6
Missing mill_d update
2021-04-22 14:03:14 +02:00
Paolo Giannozzi
af4b1d5816
Some more CUDA cleaning:
...
- removal of old PGI workaround in stress that does not seem to be any longer relevant
- removal of using_mill*
2021-04-21 22:04:18 +02:00
Paolo Giannozzi
9419b5e8a8
More pseudopotential reorganization
...
Calculation of interpolation tables moved to separate routines init_tab_*.f90
2021-04-19 20:28:22 -04:00
Lorenzo Paulatto
c56463b5bf
Better name
2021-04-19 20:28:22 -04:00
Lorenzo Paulatto
25e99a60f5
lmdif -> lmdif0
2021-04-19 20:28:22 -04:00
Lorenzo Paulatto
b74a9f07ec
Let's not call it chi^2 if it is not
2021-04-19 20:28:22 -04:00
Lorenzo Paulatto
586ed6ceae
lmdif wants to minimize a function, when used to fit a function f over some data d
...
it works much better if provided with f-d than (f-d)**2.
I have updated ev.x and scan_ibrav.x to keep this into account.
Also new interface lmdif0 allocates workspace internally and uses a much larger number
of iterations, t can still fail but should not give up prematurely.
2021-04-19 20:28:21 -04:00
Paolo Giannozzi
438a490c2d
Renaming of some routines for atomic wavefunctions:
...
tab_atwfc[_d] is the interpolation table;
It is computed in init_tab_atwfc;
interpolation takes place in interp_atwfc[_gpu].
2021-04-19 20:28:21 -04:00
Paolo Giannozzi
ee6a75a215
No longer using using_eigts
...
The "using" machinery is rather pointless for variables eigts[123]: they are
allocated at the very beginning, updated by a single routine (struc_fact) that
is called in very few places, take nothing to compute. We just compute on CPU
and copy to GPU the few times they are updated.
2021-04-19 22:46:33 +02:00
giannozz
7db3b572fa
Merge branch 'devel-upflib' into 'develop'
...
More pseudopotential reorganization
See merge request QEF/q-e!1385
2021-04-19 18:57:48 +00:00
Lorenzo Paulatto
92746c08b9
Better name
2021-04-19 12:03:27 +02:00
Lorenzo Paulatto
27cdf1f99e
lmdif -> lmdif0
2021-04-19 10:29:21 +02:00
Lorenzo Paulatto
e406a00fe7
Let's not call it chi^2 if it is not
2021-04-19 10:26:45 +02:00
Lorenzo Paulatto
363818f45a
lmdif wants to minimize a function, when used to fit a function f over some data d
...
it works much better if provided with f-d than (f-d)**2.
I have updated ev.x and scan_ibrav.x to keep this into account.
Also new interface lmdif0 allocates workspace internally and uses a much larger number
of iterations, t can still fail but should not give up prematurely.
2021-04-19 10:15:57 +02:00
Paolo Giannozzi
bb870ed1cb
More pseudopotential reorganization
...
Calculation of interpolation tables moved to separate routines init_tab_*.f90
2021-04-17 08:25:38 +02:00
Paolo Giannozzi
7329909ad9
Renaming of some routines for atomic wavefunctions:
...
tab_atwfc[_d] is the interpolation table;
It is computed in init_tab_atwfc;
interpolation takes place in interp_atwfc[_gpu].
2021-04-15 09:40:53 +02:00
Sasha Fonari
cd6e88f56e
Catch and exit gracefully when BFGS optimizer fails
2021-04-14 15:05:02 -04:00
Paolo Giannozzi
b7ba7d59df
Merge branch 'develop' of https://gitlab.com/max-centre/components/q-e into devel-upflib
2021-04-07 15:06:49 +02:00
Paolo Giannozzi
9f4d481929
Removal of duplicated routine
2021-04-07 15:06:22 +02:00
giannozz
3e455ca625
Merge branch 'devel-upflib' into 'develop'
...
More upflib merge - PLEASE CHECK BERRY/EL.FIELDS WITH USPP/PAW
See merge request QEF/q-e!1376
2021-04-07 06:18:26 +00:00
Daniele Cesarini
042a8ca1a0
CMake fixes for shared lib builds and added a CI build for NVHPC
2021-04-06 19:07:49 +00:00
Paolo Giannozzi
017d78707a
Why PW/src/CMakelist.txt is in PW/ ... ?
2021-04-05 22:23:25 +02:00
Paolo Giannozzi
437f630e80
Removal of duplicated PP code
...
*** PLEASE TEST MACROSCOPIC ELECTRIC FIELDS AND BERRY PHASES WITH US-PP ***
2021-04-05 22:21:48 +02:00
Jae-Mo Lihm
62ae9cda59
In [H,x] commutator routines, make shape(dpsi,2) nbnd_calc(=nbnd_occ), not nbnd
2021-04-05 13:23:11 +09:00
Jae-Mo Lihm
9f3f6384b9
Make [H,x] commutator work for arbitrary direction
...
Previously, it assumed crystal coordinate at(:,1) or at(:,2) or at(:,3).
Now, it accepts vpol(1:3) as input.
2021-04-05 13:22:23 +09:00
Paolo Giannozzi
074fe831a3
Calculation of dj_l(x)/dx moved to upflib. There are two routines, one used
...
in a single place (bp_calc_btq) but never actually called, one used in CPV.
In upflib instead dj_l(x)/dx is never used because the derivation is done
numerically directtly on the interpolation table. Cleanup and merge needed,
as well as a check on the quality of the numerical differentiation.
Minor fixes to timing printout.
2021-04-04 19:01:26 +02:00
Paolo Giannozzi
3196087a73
make.depend again
2021-04-03 17:52:03 +02:00
Paolo Giannozzi
710bb524e2
Functions sph_ind and spinor moved to upflib
...
Removes circular dependency between upflib and PW/src
Ideally these functions should go into upf_spinorb module
2021-04-03 17:25:15 +02:00
Jae-Mo Lihm
e59e176fcd
Revert "Merge branch 'revert-c395d46c' into 'develop'"
...
This reverts commit 194acab48e
, reversing
changes made to c395d46c8d
.
2021-04-03 12:26:30 +09:00
Pietro Delugas
78fc480a78
Revert "Merge branch 'commutator' into 'develop'"
...
This reverts merge request !1368
2021-04-02 16:03:58 +00:00