LogicError setting upper Q range limit

[smk78]

Win v4.0 release

Whilst trying to reduce the Max Q fit setting on a dataset ranging from 0.007 - 0.25 to 0.007 - 0.07 using the sphere model with polydispersity on I got the following error:

Traceback (most recent call last):

File "sas\sascalc\data_util\calcthread.pyc", line 268, in _run

LogicError: clEnqueueReadBuffer failed: invalid value

Setting the Max Q to 0.1 instead of 0.07 did not generate the error.

The logic test generating this is probably fine, but the error is a bit alarming for a user!

[smk78]

On further investigation, this seems not to have anything to do with turning polydispersity on.

With the attached example data (some typically noisy, slightly oversubtracted user data):
Load it
Send it to fitting
Select sphere category
Select sphere model
Try and change the upper Q bound for fitting to a number between 0.068 - 0.073

[smk78]

some_users_data.txt

[smk78]

We have introduced this error in 4.0. The limits work fine in 3.1.2.

[Piotr Rozyczko <rozyczko@…]

In 96c2b985b33de934a49212309128f99ca063692b/sasmodels:

Allow for (rare) cases where the size of q_input is smaller than results. Fixes #794

[GitHub <noreply@…>]

In 11052863de57607ed36c5a18837874af1265ee7f/sasmodels:

Merge pull request #26 from SasView?/ticket-794

Allow for (rare) cases where the size of q_input is smaller than result. Fixes #794.

[pkienzle]

This error came about because vectors passed to the GPU are extended to the nearest 16 or 32 value boundary, with some extra space to store the normalization constant. Some parts of the code were reserving 2 slots while others were requesting 3; only 1 is needed. To check the fix, sascomp was run with -nq=100 … -nq=132 on both single and double precision.