jjppeters / cltem Goto Github PK

Changes to the way the simulation area is handled means the structure info on the main window is incorrect/irrelevant. This just needs to make sure the correct function are called

instead of needing the GUI, a separate console application should be created to use the simulation library. This makes it easier to run on servers, or to automate with scripting.

tested on windows doing cbed with 30 TDS configs. Suspected OpenCL issue not completing kernels before getting data/continueing.

Perhaps try spamming WaitForQueueFinish commands to see if it helps

Images need to be saved. Ideally as a full precision tiff and as an image format (probably bmp).

Check the CBED probe is in the correct position. It is fairly easy to export the generated incident wavefunction to measure. Check complications due to the reversal of the y-axis too.

In windows, the default install location is write protected, so users cannot easily edit the default. This should be moved to a new location so it can be changed, with a possible option to 'save as default' so the user does not need to know the location.

Instead of showing certain parts of a complex iamge (ie. the exit wave being a phase and amplitude image) show the complex image as one part with the ability to set how it is displayed.

Images with small dimensions (found in a 2x2 image) do not resize properly as the window is resized. Untested for even slightly larger images. May not be a problem as 2x2 simulations are going to be very rare.

I beleive a bug is caused where the bins get overladed with atoms. This main occurs when loading a structure that should be in nm as Angstroms. The simulation then faults at the potential generation stage (it completes the empty 'pre-structure' slices and fails as soon as it enters the structure).

I'm guessing this is due to the bins containing too many atoms but I'm not really sure...

On some systems, some of the time, the simulation produces blank results. Similar problems before have related to asynchronous read/write of opencl buffers before a kernel has completed.

Currently the slice thickness/position cannot be altered. This is an important feature to select how the structure is divided.

e.g. instead of kirkland's potentials, there are a few others that could be used. This requires modifying the way they are calculated (in the kernel) pseudo-dynamically (i.e. set it per simulation).

So multiple simulations can be performed and parallelised fully, an input of multiple files/settings should be accpeted and then run through.

Need some sort of ID so that the thread progress can be reported per thread. Currently they all compete and overwrite the same output.

Whilst batch simulating CBED patterns, an unexpected error occurred (program has stopped working) only sometimes. Running 400+ 10 TDS configuration simulations produced zero error, but with 30 TDS configurations the error happens sometimes (every 30 simulations or so?). No clue what caused it.

To avoid repeating simulations for structures of different thickness, an option to export the output of a multislice results at intermediate slices should be implemented

When tabbing through text boxes, the order in which they are selected is muddled. See the CBED frame on the main window.

The dose/CCD simulation doesn't work on some hardware (i.e. an intel i7-4600u) when it does on others (tested working on a gtx 670). This is likely caused by reading/writing buffers before kernels have finished, or similar.

Need to use a defined code style for function names etc...

Currently it is a complete mess. Sorry.

Currently, clTEM simply terminates if OpenCL calls fail. This should be handled appropriately (i.e. disable the simulation running) so the GUI can still be accessed.

A graphical interface for the exact slice dimensions and how this positioned in respect to the structure is important so users can easily see the slices/simulation area

When the mouse hovers over an image, the coordinates should be displayed

Get STEM working!

Currently dialogs/frames/panels all interface with the data in completely different ways. For ease of use, this should be changed to have a defined way of setting/getting data.

Currently, CTEM simulations remove the padded area around the desired simulation area. This should be an optional as the diffraction pattern always includes the padded area and it may be needed to show exactly what has been simulated.

Existing code for finite difference exists, though I'm not entirely familiar with how it works. This should be added and tested at some point.

Every variables needs to be tested to verify it's units. Currently it is not obvious and some things seem to have a massive effect on the simulation when they might not be expected to.

If there is no image shown, you can still right click and 'save.' This causes a crash

Currently, CTEM cropping appears to be correct (for the few test cases) This needs to be thoroughly checked for multiple cyrstals to be confident how the cropping is performed and defined.

Because some options are not obvious when setting them from the terminal (especially the first time) It would be nice to have some flag that can be set so the simulation confirms everything before it actually does the simulation.

The images are updated multiple times (at least after the first simulation). This is very noticeable for larger images. Possible causes are the image plot redrawing multiple times or the signal slot mechanism to return images is called multiple times.

Currently, only the exit wave part is simulated. Functions to simulate the actual image are present but not 'hooked up' and are untested in the current version.

For CTEM image simulations, the exit wave does not need to be recalculated when the aberrations change (only for voltage, etc...). It could be possible to preserve the exit wave simulation so it does not need to be recalculated (if possible), though this is made tricky by the use of threads and multi-device support.

currently, only xyz files are handled. Generating a structure from a .cif would also be useful (and maybe allow it to be exportable as a .xyz)

Previously, DQEs had been hard coded. This should be changed to external files so that users can add/remove CCD profiles as needed.

The simulation area has different purposes for different modes. Currently it works for the CTEM mode, but doesnt make much sense in STEM/CBED. In these modes, the simulation area is purely to determine the reciprocal scale. The dialog should depend changing on the mode, and for STEM/CBED, should be allowed to set a desired resolution, or desired maximum in reciprocal space (similar options could be used to achieve a set resolution in CTEM).

Error message:

terminate called after throwing an instance of 'nlohmann::detail::type_error'
what():  [json.exception.type_error.304] cannot use at() with null

i.e. give exact info on what the simulation is doing. Show the simulation parameters that are going to be used etc..

Might be hard due to the multiple threads...

See title...

See title. Only happens when the parallel pixels setting is not 1

If you start the simulation and quickly click cancel (before the simulation has properly started) then the program will crash.

Currently, not all settings are actually used. This is obviously a major flaw and should be easily fixed.

For windows, icon needs to be added at build

The current settings sould be exportable from the GUI (preferrable to something easily readable and editable, maybe JSON?) as well as importable.

Settings defaults should be editable by the user (perhaps from a file a la #3). Maybe multiple defaults can be loaded and selected from. This would help as many people only need to simulate for a select few microscopes in certain configurations.

The units used should be displayed to the user in some form.