Setting `nthreads` > 1 doesn't seem to work

[aufdenkampe]

When I initialize a PhreeqcRM reaction model with more than one thread for parallel computations (i.e. model.initialize(nthreads=4), I do not see a speedup in run times.

We can confirm that it isn't being set correctly by running:

model.phreeqc_rm.GetThreadCount()

I also noticed that the mf6rtm.yaml that is created lists nthreads: 1.

I think the issue is related to how we are creating the PhreeqcRM instance

Alternately, it is possible that the PhreeqcRM Python package does not compile the bundled PHREEQC executable with the directive to "USE_OPENMP" for parallelization. See https://usgs-coupled.github.io/phreeqcrm/namespacebmiphreeqcrm.html#ac799e2e2b4364722a16be26bb81f0825

I need to dig into this deeper, unless @p-ortega or anyone else has other ideas.

[aufdenkampe]

I've discovered that I can look in my local installation directory -- ~/miniconda3/envs/modflow/lib/python3.12/site-packages/phreeqcrm/phreeqcrm.py -- to get better insight into how C++ code is wrapped into Python files. The instructions for building this wrapper are in https://github.com/usgs-coupled/phreeqcrm/blob/master/swig/PhreeqcRM.i

[aufdenkampe]

It appears that PHREEQC packaged with https://pypi.org/project/phreeqcrm/ was not compiled for parallelization.

I did a simple test with this code:

nxyz = 40
nthreads = 4
rm = phreeqcrm.PhreeqcRM(nxyz,nthreads)
rm.GetThreadCount()

Output:

1

@scharlton2 & @dlparkhurst, is that right that nthreads cannot be increased above 1 if using the Python phreeqcrm package?

[dlparkhurst]

We are working on it, hopefully within a month.

…

On Wed, Mar 11, 2026, 4:38 PM Anthony Aufdenkampe ***@***.***> wrote: *aufdenkampe* left a comment (p-ortega/mf6rtm#54) <#54 (comment)> It appears that PHREEQC packaged with https://pypi.org/project/phreeqcrm/ was not compiled for parallelization. I did a simple test with this code: nxyz = 40nthreads = 4rm = phreeqcrm.PhreeqcRM(nxyz,nthreads)rm.GetThreadCount() Output: 1 @scharlton2 <https://github.com/scharlton2> & @dlparkhurst <https://github.com/dlparkhurst>, is that right that nthreads cannot be increased above 1 if using the Python phreeqcrm <https://pypi.org/project/phreeqcrm/>? — Reply to this email directly, view it on GitHub <#54 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/APZJL54VYA5LYHWLIW2RZYL4QHTEZAVCNFSM6AAAAACWOZ4XT2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DANBSGY4DGNBSGI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[scharlton2]

I just tested on my windows machine:

Python 3.13.6 (main, Aug  8 2025, 17:02:53) [MSC v.1944 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import phreeqcrm
>>> rm = phreeqcrm.PhreeqcRM(40, 4)
>>> print(rm.GetThreadCount())
4

We build all the wheels using cibuildwheel on github runners. By default it uses OpenMP if it's found.

@aufdenkampe:
I assume you're working on macOS. Is it ARM64 or x86-64? Which version of python are you using?

[scharlton2]

I confirmed that OpenMP is not enabled in any of the macOS builds on PyPI. We’ll work to address this in the next release.

The conda-forge builds do have OpenMP enabled on macOS (at least for ARM64) but they are currently one version behind the wheels available on PyPI.

[aufdenkampe]

@scharlton2 & @dlparkhurst, thanks for your quick, helpful, and encouraging responses! That's great to know that phreeqcrm supports nthreads >1 as long as it's compiled with OpenMP. Thanks for working on it for the next release!

@scharlton2, yes, I'm running on MacOS arm64. Most Mac users are using arm64 machines these days, as Intel-based Macs were discontinued in 2021. Like most scientists & engineers who use Mac, I can also run a Windows virtual machine, but that also needs to be compiled for win-arm64 architectures. This work is for USACE, and their plan is to run the operational model on a Linux supercomputer.

I'm currently using Python 3.12 but expect to update to 3.13 soon.