Multithread a for loop

The Integrate1D operation is listed as automatically multithreaded, and also threadsafe. Automatic threading may depend on your integrand function being threadsafe- you need to qualify the function declaration with the Threadsafe keyword.

But if you are doing a number of separate integrations for each cell of a matrix, you may get a better boost by running integrations in threads. If you can write the integration as a function returning a value to assign to the matrix wave, you may be able to do it with the Multithread keyword.

To learn more about threading in Igor, read our help starting with DisplayHelpTopic "ThreadSafe Functions and Multitasking"

johnweeks wrote:

The Integrate1D operation is listed as automatically multithreaded, and also threadsafe. Automatic threading may depend on your integrand function being threadsafe- you need to qualify the function declaration with the Threadsafe keyword.

But if you are doing a number of separate integrations for each cell of a matrix, you may get a better boost by running integrations in threads. If you can write the integration as a function returning a value to assign to the matrix wave, you may be able to do it with the Multithread keyword.

To learn more about threading in Igor, read our help starting with DisplayHelpTopic "ThreadSafe Functions and Multitasking"

Thanks for the ideas, it clears a few problem I have. I didn't know Integrate1D is automatically multithreaded.

Naively, I tried to just flag them threadsafe but it crashed Igor. I then found that I have been doing things that Threadsafe disallowed. I will go through the help file in more details.

Update, going through the code, I think one problem I have (from the example as well) is that I need to update the globalY coordinate in the outerloop, which is updated by the outer Integrate1D. Here is the code:

do2DintForWF is the first function which perfoms a 2D integration for one coordinate.
It does the first Integrate1d and call the outerInt.

In outerInt, the globalY is being updated by the inY produced when outerInt is being called. This globalY is then used when doing the second Integrate1D with innerInt. This prevents me from flagging the outerInt function as Threadsafe (or it will crash).

How should I solve this?

Hello Sandbo,

Any reason for not using Integrate2D?

AG

Maybe it didn't work with complex number?

I think you gave me this suggestion last time, but I haven't really tried to make it work that way yet.
If I want to use it with complex number, how should I start?

In order to deal with the complex number you would have to express your integrand in terms of real and imaginary parts.  Not so easy unless you are considering a limited range of m,n.  Also, before trying any of these it would help to know what are alpha and alpha*.

If you wouldn't mind taking a look, I am trying to implement equ (6) in this paper:

https://arxiv.org/pdf/1011.6668.pdf

alpha = x+i*p is a complex number as the input argument to equation (6) (sorry I forgot to include a definition for the pwave, the parameter wave). Alpha represent the coordinate (x,p) on a complex plane.

Then, equ (6) integrates for a given coordinate (as x and p, with alpha=x+i*p) for a certain xrange and yrange.
My bad to use the x again, let's called it x' here. Then xrange (x') and yrange (y') represent the integration variable of the complex number lambda (lambda = x'+i*y').

In short, my Igor integration will take the parameter alpha=x+i*p which is a constant in the integration, then integrate over a complex variable lambda=x'+i*y' with equation (6).
Let me know if there is anything else I can clarify, thank you.

1.  Please start by changing x and p in innerInt() to different variable names.  This may save you from unintended consequences.

2.  It may be trivial but I think it is important to re-write equation (6) so that all the factors that are independent of the integration are out of the integral.  You do not want even a single multiplication inside any of the integration functions if it can be computed outside.  You are also potentially setting twice both inB=cmplx(1,0), inC=cmplx(1,0).

3.  Before actually attempting the integration, it may be worth your time to check in some tables if the integration can't be done in closed form for some values of m, n and alpha.  Even if you can find a closed form solution for one point, it can be used to check your integration and IMO worth the trouble.