Use pinned memory for GPU version

Description

Add a special allocation function in C that will allocate pinned memory when running on GPUs for the wavefunctions. This increases the transfer speed by more than a factor of 2.

This function could in the future also be used to make the allocated memory aligned for CPU runs which would improve the usage of vectorization.

News snippet

Use pinned memory for the GPU version to improve transfer speeds.

Checklist

• I have checked that my code follows the Octopus coding standards

Closes #223 (closed). Relates to #162 (closed).

added Core GPU labels

changed milestone to %10.0

added 1 commit

• c47f3fdb - Rename allocation functions and add C implementation

Compare with previous version

Codecov Report

Merging #669 into develop will decrease coverage by 0.15%. The diff coverage is 81.94%.

@@             Coverage Diff             @@
##           develop     #669      +/-   ##
===========================================
- Coverage    70.83%   70.68%   -0.16%     
===========================================
  Files          472      473       +1     
  Lines        89541    89590      +49     
===========================================
- Hits         63429    63323     -106     
- Misses       26112    26267     +155

Impacted Files	Coverage Δ
src/grid/batch_inc.F90	100% <100%> (ø)
src/states/states_elec.F90	84.04% <100%> (+0.03%)
src/grid/allocate_special.c	66.66% <66.66%> (ø)
src/grid/batch.F90	86.48% <81.81%> (-0.94%)
src/grid/subarray_inc.F90	53.65% <0%> (-43.91%)
src/basic/accel_inc.F90	71.11% <0%> (-28.89%)
src/grid/boundaries_inc.F90	79.73% <0%> (-14.06%)
src/grid/boundaries.F90	84.87% <0%> (-10.09%)
src/grid/nl_operator.F90	65.97% <0%> (-5.68%)
src/grid/nl_operator_inc.F90	64.25% <0%> (-4.26%)
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 1f1f919...5f4cf9c. Read the comment docs.

added 2 commits

• 127a294f - Revert "Improve GPU packing/unpacking"
• 63d85ad0 - Put debug print statements into preprocessor macros

Compare with previous version

marked as a Work In Progress

changed the description

added 2 commits

• 2a7a5b38 - Add license header to allocate_special.c
• c3a11ea4 - Move Fortran interface to separate module

Compare with previous version

unmarked as a Work In Progress

mentioned in merge request !673 (merged)

added 1 commit

• 5f4cf9cd - Fix missing brace in debug code

Compare with previous version

resolved all threads

approved this merge request

merged

mentioned in commit 15ad4bcb

mentioned in issue #223 (closed)

changed the description