[WIP] Add Branson Monte Carlo Kernel by michaelmckinsey1 · Pull Request #683 · llnl/RAJAPerf

michaelmckinsey1 · 2026-05-20T22:16:35Z

Summary

An intern (@RanivG) will be implementing https://github.com/lanl/branson/blob/3f538e31a51b144b6621d4665f465d8d44ee85b8/src/event_based_transport.h#L323 as a kernel in RAJAPerf.

Add Apps_MC_HISTORY_PARTICLE_TRANSPORT kernel
Add reference implementation to header file (see other kernels for examples)
CPU-only implementation
RAJA implementation
CUDA implementation
HIP implementation
Rename to MC_HISTORY_PARTICLE_TRANSPORT

MrBurmark · 2026-05-20T22:23:10Z

Is there a better name for this kernel?

michaelmckinsey1 · 2026-05-20T22:46:05Z

Is there a better name for this kernel?

I am open to suggestions, but it can be renamed easily.

rhornung67 · 2026-05-21T15:22:51Z

I'm fine with the name. @MrBurmark do you have a specific concern about the name?

MrBurmark · 2026-05-21T15:41:40Z

The name is just very generic. Is the kernel any specific part of the Monty Carlo algorithm?

rhornung67 · 2026-05-21T16:00:50Z

The name is just very generic. Is the kernel any specific part of the Monty Carlo algorithm?

That's a good point. A brief scan of the code at the link @michaelmckinsey1 provided above to the Branson code, perhaps something like MC_EVENT_PHOTON_TRANSPORT, if that's not too long.

rhornung67 · 2026-05-21T16:01:23Z

@michaelmckinsey1 please add a reference implementation to the kernel header file.

…l-photon-transport

artv3 · 2026-08-07T03:16:07Z

probably don't want this and would be good to double check on the other extra files

Yes, there are a bunch of unwanted files committed in this. Please refrain from using git add *.

MrBurmark · 2026-08-07T17:20:15Z

+      transport3dmc_lam_setup(vid, iend);
+      // Loop counter increment uses macro to quiet C++20 compiler warning
+      for (RepIndex_type irep = 0; irep < run_reps; RP_REPCOUNTINC(irep)) {
+        transport3dmc_lam_reset();


Only the loop body needs to be in a lambda.

Unless the reset is also a loop body.

MrBurmark · 2026-08-07T17:20:55Z

+    case Base_Seq : {
+
+      startTimer();
+      TRANSPORT3DMC_DATA_SETUP(vid, iend);


Setup should happen once above the variant switch statement.

MrBurmark · 2026-08-07T17:22:23Z

+      TRANSPORT3DMC_DATA_SETUP(vid, iend);
+      // Loop counter increment uses macro to quiet C++20 compiler warning
+      for (RepIndex_type irep = 0; irep < run_reps; RP_REPCOUNTINC(irep)) {
+        TRANSPORT3DMC_RESET


If reset involves a loop from begin to end, you probably want to express that loop outside of the reset macro.

No loop, just a deep copy of the reference set to another object actually conducting transport.

MrBurmark · 2026-08-07T17:23:22Z

+  setItsPerRep( getActualProblemSize() );
+  setKernelsPerRep(1);
+
+  setBytesAllocatedPerRep( 0 ); // in, out
+  setBytesReadPerRep( 0 ); // in
+  setBytesWrittenPerRep( 0 ); // out
+  setBytesModifyWrittenPerRep( 0 );
+  setBytesAtomicModifyWrittenPerRep( 0 );
+  setFLOPsPerRep( 0 );


Might want to add a TODO comment as a reminder.

MrBurmark · 2026-08-07T17:29:39Z

+
+void TRANSPORT3DMC::initMaterials(std::vector<TRANSPORT3DMC::Material> &mats) {
+  size_t matCt = 12;
+  mats.resize(matCt);


You'll probably want to use something like allocDataForInit to get memory here and allow you to manually set it up. Note that all setup for the data must be completed before the object returned by allocDataForInit goes out of scope, see examples by searching for ForInit.

implemented

MrBurmark · 2026-08-07T17:30:23Z

+{
+}
+
+int TRANSPORT3DMC::Cell::GLBL = 0;


Why are you using static class variables?

Since the mesh is an array, the static GLBL helps to identify/assign cells their ID/index into the mesh vector. Can probably be removed once all vectors have been migrated to pointers.

MrBurmark · 2026-08-07T17:31:37Z

+  #define dT 0.001
+  #define N_ISOTOPE 68
+  //#define GROUPS 10
+  #define cutoff 1e-5


Defs should be above.

Though you may be able to make these static constexpr class variables instead.

MrBurmark · 2026-08-07T17:33:15Z

+    public:
+      static int GLBL;
+      int ID;
+      std::array<double, 6> planes; //-x, +x, -y, +y, -z, +z


You might want to use camp::array, its the same as std::array but works on gpus.

MrBurmark · 2026-08-07T17:33:44Z

+    SOURCE,
+  };
+
+  struct XS {


Rename to CrossSection for clarity?

MrBurmark · 2026-08-07T17:34:08Z

+    XS operator*(const double c) const;
+    XS operator+(const XS &sigma) const;


Define inline for performance?

MrBurmark · 2026-08-07T17:35:18Z

+
+  struct Material {
+    int isotopeCt;
+    std::vector<double> conc;


Prefer allocData and deallocData and related helper functions to get memory.

implemented.

MrBurmark · 2026-08-07T17:36:52Z

+   transport3dMC_particles(10000),
+   transport3dMC_cube_dim(3),
+   transport3dMC_groups(10),


Nice, we'll have to have to think about what are reasonable defaults later.

MrBurmark · 2026-08-07T17:38:05Z

 /install_*/
 /install-*/
 /Debug/
+*.cmake


This seems a bit broad, don't we use cmake?

removed, checked out the gitignore from develop

MrBurmark · 2026-08-07T17:39:27Z

+  setDefaultProblemSize(1);
+  setDefaultReps(1000);


This is fine for now, long term I'm going to guess this is going to move toward values like this.

Suggested change

setDefaultProblemSize(1);

setDefaultReps(1000);

setDefaultProblemSize(1000000);

setDefaultReps(10);

MrBurmark · 2026-08-07T17:40:30Z

+  setSize(params.getTargetSize(getDefaultProblemSize()),
+          params.getReps(getDefaultReps()));
+
+  setChecksumConsistency(ChecksumConsistency::ConsistentPerVariantTuning);


Hopefully this stays consistent as we'll have consistent pseudo random numbers.

MrBurmark · 2026-08-07T17:42:52Z

+TRANSPORT3DMC::Material::Material() {
+  isotopeCt = 1;
+  conc = std::vector<double>();
+  nucIDs = std::vector<int>();
+}


This is fine for now, but c++ has a syntax for initializing class members https://en.cppreference.com/cpp/language/constructor.

Ended up just deleting it and letting the compiler define a default constructor.

RanivG · 2026-08-08T01:20:50Z

I was reading through earlier comments, I disagree with the name MC_EVENT_PHOTON_TRANSPORT; the kernel emulates history-based transport of neutrons (or at least that's how I designed it).
Might I propose "HISTORY_TRANSPORT3DMC" ?

michaelmckinsey1 · 2026-08-10T18:29:15Z

I was reading through earlier comments, I disagree with the name MC_EVENT_PHOTON_TRANSPORT; the kernel emulates history-based transport of neutrons (or at least that's how I designed it). Might I propose "HISTORY_TRANSPORT3DMC" ?

If the difference is history vs event, why not MC_HISTORY_PHOTON_TRANSPORT then?

RanivG · 2026-08-10T19:56:18Z

I was reading through earlier comments, I disagree with the name MC_EVENT_PHOTON_TRANSPORT; the kernel emulates history-based transport of neutrons (or at least that's how I designed it). Might I propose "HISTORY_TRANSPORT3DMC" ?

If the difference is history vs event, why not MC_HISTORY_PHOTON_TRANSPORT then?

Well it's not really photons, at least I don't think. Also I was more trying to keep with the 3D kernel's naming scheme (MASS3D, CONVECTION3D, VOL3D, DIFFUSION3D, etc). If this convention isn't desired for this kernel then we can stay with the MC_HISTORY_PARTICLE_TRANSPORT convention

MrBurmark · 2026-08-10T20:03:30Z

Well it's not really photons, at least I don't think. Also I was more trying to keep with the 3D kernel's naming scheme (MASS3D, CONVECTION3D, VOL3D, DIFFUSION3D, etc). If this convention isn't desired for this kernel then we can stay with the MC_HISTORY_PARTICLE_TRANSPORT convention

I don't think there is much reason to limit yourself to that 3D naming convention as this is a different type of kernel.

michaelmckinsey1 · 2026-08-11T16:54:52Z

@RanivG Rm this

michaelmckinsey1 · 2026-08-11T16:54:56Z

@RanivG Rm this

…l-photon-transport Update to develop

Add placeholder for new kernel

9d9a04a

michaelmckinsey1 self-assigned this May 20, 2026

RanivG added 9 commits July 27, 2026 17:13

adding basic data structs

742f249

first steps(work moved to private repo, move back later)

36a922b

Migrate code from private repo, untested in RPerf

c56782b

builds and runs

ea4fea5

add temp particles

daced47

Merge branch 'develop' of github.com:llnl/RAJAPerf into branson-kerne…

5083c02

…l-photon-transport

caliper fixes + particle distribution

05fdc28

change logging to build conditional

b2efc30

add runtime params for parts, zones, groups

2864998

artv3 reviewed Aug 7, 2026

View reviewed changes

RanivG added 2 commits August 7, 2026 08:53

Remove extra files

2eb61f5

Better Caliper Logging

0713050

MrBurmark reviewed Aug 7, 2026

View reviewed changes

setup fixes, migrate to pointer-arrays

ac23c9c

inline XS operators

8276e4f

remove static var

4633311

michaelmckinsey1 commented Aug 11, 2026

View reviewed changes

Comment thread build-linux-rhel8-sapphirerapids-5szrqd5 Outdated

michaelmckinsey1 Aug 11, 2026

Copy link
Copy Markdown

Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@RanivG Rm this

michaelmckinsey1 commented Aug 11, 2026

View reviewed changes

RanivG added 7 commits August 11, 2026 09:58

remove extraneous build files

97b7e01

Merge branch 'develop' of github.com:llnl/RAJAPerf into branson-kerne…

4c13399

…l-photon-transport Update to develop

guard Cali annots, rename kernel

9505f9d

more annotation fixes

55f6a3e

silence print logs

cd843af

attempting analytic count of writes

63f4c20

remove logs, add write/rep formula

20c4bdc

michaelmckinsey1 assigned RanivG Aug 14, 2026

add reference implementation, clean stale and add new comments

a20b008

		XS operator*(const double c) const;
		XS operator+(const XS &sigma) const;

Uh oh!

Conversation

michaelmckinsey1 commented May 20, 2026 • edited by RanivG Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Uh oh!

MrBurmark commented May 20, 2026

Uh oh!

michaelmckinsey1 commented May 20, 2026

Uh oh!

rhornung67 commented May 21, 2026

Uh oh!

MrBurmark commented May 21, 2026

Uh oh!

rhornung67 commented May 21, 2026

Uh oh!

rhornung67 commented May 21, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

MrBurmark Aug 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

michaelmckinsey1 commented May 20, 2026 •

edited by RanivG

Loading

MrBurmark Aug 7, 2026 •

edited

Loading

RanivG commented Aug 8, 2026 •

edited

Loading