# Multithread filter how to write to the same buffer
**URL:** https://discourse.itk.org/t/multithread-filter-how-to-write-to-the-same-buffer/1029
**Category:** Beginner Questions
**Created:** [June 24, 2018, 6:24pm UTC](https://discourse.itk.org/t/multithread-filter-how-to-write-to-the-same-buffer/1029 "2018-06-24T18:24:22Z")
**Posts on this page:** 5
**Page:** 1
### Author:  [@rolof](https://discourse.itk.org/u/rolof)
#### Post date: [June 24, 2018, 6:24pm UTC](https://discourse.itk.org/t/multithread-filter-how-to-write-to-the-same-buffer/1029/1 "2018-06-24T18:24:22Z")
Hi all,
I know how to multi-thread works a bit, and I know that the regions of images are passed to different CPU’s core. This process is useful to fill an output image and very convenient. However, I need now to process with multi-thread an image input and check its pixel value and depending on the conditions, store that pixel position in a itk::pointset buffer.
How can I avoid the concurrence to access to pointset buffer from different cores? or If you know a better way to do it, recommendations are welcome 🙂
Thanks in advance.
---
### Author:  [@blowekamp](https://discourse.itk.org/u/blowekamp)
#### Post date: [June 25, 2018, 1:20pm UTC](https://discourse.itk.org/t/multithread-filter-how-to-write-to-the-same-buffer/1029/2 "2018-06-25T13:20:57Z")
Are you working with ITKv5?
This pattern would fall into the “map-reduce” pattern.
The recommendation for now, would be in the threaded method, allocate a points buffer, and fill it during while looking over the threads chunk. At the end of the method obtain a lock on a globa or class level point set buffs, and merge the results into it.
---
### Author:  [@rolof](https://discourse.itk.org/u/rolof)
#### Post date: [June 25, 2018, 5:22pm UTC](https://discourse.itk.org/t/multithread-filter-how-to-write-to-the-same-buffer/1029/3 "2018-06-25T17:22:30Z")
Thanks for your response.
No, I am working with ITKv4. If I understood your solution is to make a global variable for example a member of the class filter and the end of the multi-thread function make fill it with the solution of the thread executed with a lock method to avoid write at the same position. However, does Itk provide method to do the lock or I should go to low level C/C++ to do it?
Thanks.
---
### Author:  [@blowekamp](https://discourse.itk.org/u/blowekamp)
#### Post date: [June 25, 2018, 5:27pm UTC](https://discourse.itk.org/t/multithread-filter-how-to-write-to-the-same-buffer/1029/4 "2018-06-25T17:27:58Z")
You are find a patch which adds a lock here:
>
>
> Coarsely fix threading issue with IsoContourDistanceImageFilter,
> utilized by the… ApproximateSignedDistance, and NarrowBand LevelSet
> framework. This patch address a periodic failing test on the ITK
> dashboard: itkApproximateSignedDistanceMapImageFilterTest1. And likely
> address NarrowBand predictability issues previously encountered.
>
> This patch oafishly utilizes a mutex lock when reading and writing to
> the output image. Given, that this filter only updates the values
> around the iso-contour the locking does not occur for every pixel.
>
> The algorithm should be re-witten with out the concurrent read/write
> problems. Potential solutions include, having each thread allocate a separate
> output buffer, then merging. Another is to crop the regions and handle
> the overlapping boundaries separately. But this would heavily depend
> on how the output regions are split, and there is a trend to have less
> control over the splitting methods.
And another similar discussion to this type of issue here:
> [@threadId in ParallelizeImageRegion](https://discourse.itk.org/t/threadid-in-parallelizeimageregion/1008/7):
>
> There are two components to be aware of: [SimpleFastMutexLock](https://itk.org/Doxygen/html/classitk_1_1SimpleFastMutexLock.html)[MutexLockHolder](https://itk.org/Doxygen/html/classitk_1_1MutexLockHolder.html) The mutex is variable scoped to provide serialized access. For your case, the need is to control access to threads in an instance of the class. So you are looking at a lock with object scope, a member variable. The holder provides exception safe lock/unlock handling of the mutex via [RAII](https://en.cppreference.com/w/cpp/language/raii). Here is an example which controls initialization and access to a global variable: [https://github.com/InsightSoftwareConsortium/…](https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Core/Common/src/itkMersenneTwisterRandomVariateGenerator.cxx#L62-L72)
---
### Author:  [@rolof](https://discourse.itk.org/u/rolof)
#### Post date: [June 25, 2018, 5:30pm UTC](https://discourse.itk.org/t/multithread-filter-how-to-write-to-the-same-buffer/1029/5 "2018-06-25T17:30:37Z")
Thanks I will see that.