Giovanni Di Siena authored 4 years ago and Gigg, Martyn Anthony committed 4 years ago

In places other substitutions have been made, e.g
Clang does not yet specialize std::shared_ptr for T[]. Vector
has been used instead. The operator[] methods were incorrectly
marked const but returning a non-const reference - this has been fixed.
Refs #25842

Also operator!=
Operators provided for LogManager, PropertyManager, Geometry and Run alongside testing

Previously, if an attenuation factor was requested for a lambda that
is outside the range of an explicit attenuation profile the code
extrapolated based on the gradient of the two most extreme points.
If the data was noisy this gave some poor results. Improve this by
falling back on the tabulated linear absorption coefficient of an
optional ExtrapolationMaterial

Miscellaneous code review comments to add comments, remove redundant brackets
Also added comment to the sample environment concept page to explain where
Mantid will search for the attenuation profile file

Signed-off-by: Steven Hahn <hahnse@ornl.gov>

Runs clang-tidy pass by ref across the codebase to fix-up various
warnings we are seeing in cppcheck

Add feature where an externally generated attenuation profile can be loaded into
the material definition from a .DAT text file. This will override the normal
exp(-alpha*thickness) calculation of the attenuation. The new feature was requested
as part of the absorption calculation for the Pearl instrument where some parts
of the environment are crystalline

Found the lookup\interpolation on the loaded profile was slowing down the absorption
Monte Carlo calculation.
So improved performance of Kernel::Interpolation. Changed it to use vector of pairs
instead of two separate vectors to reduce amount of navigation around vectors

Considered adding feature into AttenuationProfile where I add any interpolated points
into the override attenuation profile to prevent interpolation next iteration of
Monte Carlo simulation. However wasn't convinced this was going to be thread-safe
during the parallel part of the absorption calculation - may revisit this. While doing
this investigation I fixed the following issues with the function Interpolation::AddPoint:

a) it doesn't add the same point twice
b) it reuses the lower_bound function rather than stepping through the vector sequentially

I removed the exception from findIndexOfNextLargerValue because the end()
iterator is a valid result if the x value supplied equals the last x value
already in the vector. Could have left check in for begin() but didn't seem
much point

Finally - the interpolation outside the range is not v good. The data at the extremes
is noisy so we end up extrapolating based on random gradient in extreme 2 points. Need
to either fit or use more points to determine gradient

Also updated class_maker.py

This commit removes the final header guard that had
been missed by the auto replacer.

re: #28200

This commit replaces all header guards in Kernel with #pragma once

Git only converts CRLF to LF and not CR so to produce
the same SHA as git we must do the same.

A new boolean parameter SimulateTracksForEachWavelength has been added to
the MonteCarloAbsorption algorithm. This exposes a switch on the MCAbsorptionStrategy
that was present before this change but was hardcoded to False.

The parameter NumberOfWavelengthPoints is only required if SimulateTracksForEachWavelength
is set to True. This change means that an attenuation factor is calculated for all wavelength
bins if SimulateTracksForEachWavelength=false. To facilitate this a performance enhancement
has been made to the Material class to calculate the total scatter cross section and the
absorption cross section once when the Material rather than every time these cross sections
are retrieved by the MonteCarloAbsorption calculation.

This change required some unit and system tests to be updated eg ILLDirectGeometryReductionTest
has been changed to remove the NumberOfWavelengthPoints parameter and the reference file to be
updated. To make this test more robust I have increased the number of scenarios used in the MC
simulation from 5000 to 20000

Co-Authored-By: Conor Finn <47181718+ConorMFinn@users.noreply.github.com>

This commit adds atomic distance measured in units of angstroms to the unit factory

re: #27769

Newer boost bind seems to be causing
compiler warnings on Windows. It's no
longer required now bind is in the STL.
Refs #16578

Refs #16578

Add const

Co-Authored-By: Dan Nixon <dan@dan-nixon.com>

Add const

Co-Authored-By: Dan Nixon <dan@dan-nixon.com>

Refs #16578

Windows devices have been running into issues in places where calling
the functions created to fix an issue on Unix machines in parallel
regions was causing error messages to  appear.

This is due to changes in implementation of omp_set_num_threads between
OpenMP v2 and v2.5+. As MSVC uses the older version of OpenMP I have
added a check that does not bother with the changes from the previous
fix if compiled with MSVC, as the original issue only appeared on non
MSVC compiled versions anyway.

TimeSeriesProperty::SplitByTimeVector has a potential bug were
exploring, however it is currently very difficult to understand. This
commit refactors the method to make it readable

The main declareProperty that accepted a pointer to a new Property
was public so having the others protected makes little sense
as you could always achieve the same end result from the public
method but with more verbosity at the call site.

It is not needed as each class can only find it's own conversions.
Also it wiped all quick conversions from all classes.

re #27458

Added done alongside setting the number of threads to ensure that
subsequent parallel regions do not change the number of threads they are
working with from the value set in the config.

Before each of the OpenMP parallel macros, the max threads value is
updated for the current scope to match the value in the config. This is
a fix that can be generally applied to replace the solution from #27343.