Implement PCG64 as extension by RAMitchell · Pull Request #6292 · NVIDIA/cccl

RAMitchell · 2025-10-20T14:18:37Z

Description

Part of #5679

Depends on #6109

Signed-off-by: Rory Mitchell <[email protected]>

copy-pr-bot · 2025-10-20T14:18:41Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

davebayer · 2025-10-20T16:37:26Z

libcudacxx/include/cuda/__random/pcg_engine.h

+#endif // !_CCCL_COMPILER(NVRTC)
+
+private:
+  using __uint128_type = unsigned __int128;


Just use __uint128_t

I still need to implement a fallback int128.

Do you need this on windows right now? I think it would be easier to guard it by #if _CCCL_HAS_INT128() for now and allow this feature on windows once we have our internal int128 fallback

From slack it sounded like we wanted to have a complete implementation. I don't think its too hard so I think we can do it here.

Please, do it in a separate PR

yeah lets do this as a followup

libcudacxx/include/cuda/__random/pcg_engine.h

Signed-off-by: Rory Mitchell <[email protected]>

RAMitchell · 2025-10-28T11:31:55Z

I am a bit unsure about the naming yet and if we would support variants of this engine.

I'm thinking as follows:

pcg64_engine - PCG XSL RR 128/64 algorithm templated on the multiplier/increment
pcg64 - named instance of the above engine with default multiplier/increment
pcg64_discard - a version of pcg64 that can discard a compile time or runtime number of values with each operator(), allowing higher performing "leapfrogging" PRNG in GPU kernels.

@fbusato maybe you have some thoughts?

libcudacxx/include/cuda/__random/pcg_engine.h

libcudacxx/test/libcudacxx/cuda/random/pcg64.pass.cpp

libcudacxx/test/libcudacxx/libcxx/random/pcg_engine.pass.cpp

fbusato · 2025-11-04T23:19:05Z

pcg64_engine - PCG XSL RR 128/64 algorithm templated on the multiplier/increment
pcg64 - named instance of the above engine with default multiplier/increment
pcg64_discard - a version of pcg64 that can discard a compile time or runtime number of values with each operator(), allowing higher performing "leapfrogging" PRNG in GPU kernels.

these names are perfectly fine. I'm thinking if there are better names for pcg64_discard but I'm not sure

fbusato · 2025-11-07T21:10:07Z

/ok to test a628a81

miscco · 2025-11-10T10:55:16Z

/ok to test 78e1f4a

davebayer · 2025-11-10T11:43:21Z

/ok to test a289174

miscco · 2025-11-10T13:56:22Z

/ok to test 5d9cdae

libcudacxx/include/cuda/__random/pcg_engine.h

RAMitchell · 2025-11-17T10:38:22Z

@davebayer @fbusato please re-review thanks :)

libcudacxx/include/cuda/__random/pcg_engine.h

libcudacxx/test/libcudacxx/cuda/random/pcg64.pass.cpp

libcudacxx/include/cuda/__random/pcg_engine.h

fbusato · 2025-11-17T23:17:20Z

libcudacxx/include/cuda/__random/pcg_engine.h

+public:
+  using result_type = ::cuda::std::uint64_t;
+
+private:


would be better to keep private members at the end of the class

Actually not the general advice is to move them to the front, because they are crucial to understand what is in the class

I normally belong to the "end of class" church, but we can do whatever here :)

The point being that usually a class implementation starts with constructors and other SMF, so I need to know what are the actual data members. In that case and others I have to jump around to the back of the potentially long definition to know what I am working with

interesting, I always thought about user perspective. Users are interested to the interface, not implementation details. Also, the implementation rarely changes.
Anyway, I'm fine with both approaches.

fbusato · 2025-11-17T23:18:10Z

libcudacxx/include/cuda/__random/pcg_engine.h

+  __uint128_t __x_{};
+
+public:
+  static constexpr result_type default_seed = 0xcafef00dd15ea5e5ULL;


Suggested change

static constexpr result_type default_seed = 0xcafef00dd15ea5e5ULL;

static constexpr result_type default_seed = 0xCAFEF00DD15EA5E5ull;

why? Is this a rule?

definitely not a rule, but suggested by some c++ secure coding guidelines, e.g. Autosar C++14 Rule A2-13-5 https://www.autosar.org/fileadmin/standards/R18-03_R1.4.0/AP/AUTOSAR_RS_CPP14Guidelines.pdf. ull lowercase to have a clear distinction with uppercase digits

libcudacxx/include/cuda/__random/pcg_engine.h

github-actions · 2025-11-18T10:47:16Z

🥳 CI Workflow Results

🟩 Finished in 2h 25m: Pass: 100%/88 | Total: 14h 21m | Max: 1h 35m | Hits: 99%/212945

See results here.

RAMitchell · 2025-11-21T08:33:56Z

@fbusato waiting on you :)

RAMitchell added 7 commits October 17, 2025 03:47

Basic impelementation

438fc34

Signed-off-by: Rory Mitchell <[email protected]>

First attempt

56d9811

Signed-off-by: Rory Mitchell <[email protected]>

Remove redundant file

f587dd2

Passes tests

baacf43

Signed-off-by: Rory Mitchell <[email protected]>

Style

0f82787

Add some docs

c67254e

Remove file

965e201

Signed-off-by: Rory Mitchell <[email protected]>

github-project-automation bot added this to CCCL Oct 20, 2025

github-project-automation bot moved this to Todo in CCCL Oct 20, 2025

cccl-authenticator-app bot moved this from Todo to In Progress in CCCL Oct 20, 2025

davebayer requested changes Oct 20, 2025

View reviewed changes

RAMitchell added 7 commits October 21, 2025 00:47

Review comments

d635d9d

Signed-off-by: Rory Mitchell <[email protected]>

Efficient discard

75adb20

Signed-off-by: Rory Mitchell <[email protected]>

Guard against int128

f0c34ad

Signed-off-by: Rory Mitchell <[email protected]>

Fix endif

03c2e23

Merge branch 'main' of github.com:NVIDIA/cccl into pcg

54275c6

Pass tests

b89e216

Test against reference values

f189363

RAMitchell marked this pull request as ready for review October 28, 2025 11:24

RAMitchell requested a review from a team as a code owner October 28, 2025 11:24

RAMitchell requested a review from wmaxey October 28, 2025 11:24

cccl-authenticator-app bot moved this from In Progress to In Review in CCCL Oct 28, 2025

fbusato requested changes Nov 4, 2025

View reviewed changes

github-project-automation bot moved this from In Review to In Progress in CCCL Nov 4, 2025

RAMitchell added 2 commits November 7, 2025 02:13

Review comments

009ada6

Create alias pcg64

a628a81

Fix tests

78e1f4a

Try again guarding msvc test

a289174

This comment has been minimized.

Sign in to view

Merge branch 'main' of github.com:NVIDIA/cccl into pcg

5d9cdae

This comment has been minimized.

Sign in to view

davebayer requested changes Nov 11, 2025

View reviewed changes

libcudacxx/include/cuda/__random/pcg_engine.h Outdated Show resolved Hide resolved

libcudacxx/include/cuda/__random/pcg_engine.h Outdated Show resolved Hide resolved

Review comments

bca96fb

This comment has been minimized.

Sign in to view

Merge branch 'main' of github.com:NVIDIA/cccl into pcg

57c5b9a

This comment has been minimized.

Sign in to view

Merge branch 'main' of github.com:NVIDIA/cccl into pcg

7f60a3b

This comment has been minimized.

Sign in to view

miscco reviewed Nov 17, 2025

View reviewed changes

RAMitchell added 2 commits November 17, 2025 03:28

Review comments

d2b51d8

Move test files

36c6a1e

This comment has been minimized.

Sign in to view

fbusato requested changes Nov 17, 2025

View reviewed changes

Review comments

a4cfa4f

davebayer approved these changes Nov 18, 2025

View reviewed changes

fbusato approved these changes Nov 21, 2025

View reviewed changes

github-project-automation bot moved this from In Progress to In Review in CCCL Nov 21, 2025

fbusato merged commit 6f1829d into NVIDIA:main Nov 21, 2025
108 checks passed

github-project-automation bot moved this from In Review to Done in CCCL Nov 21, 2025

	static constexpr result_type default_seed = 0xcafef00dd15ea5e5ULL;
	static constexpr result_type default_seed = 0xCAFEF00DD15EA5E5ull;

Conversation

RAMitchell commented Oct 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Uh oh!

copy-pr-bot bot commented Oct 20, 2025

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!

Uh oh!

Uh oh!

Uh oh!

RAMitchell commented Oct 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

fbusato commented Nov 4, 2025

Uh oh!

fbusato commented Nov 7, 2025

Uh oh!

miscco commented Nov 10, 2025

Uh oh!

davebayer commented Nov 10, 2025

Uh oh!

This comment has been minimized.

miscco commented Nov 10, 2025

Uh oh!

This comment has been minimized.

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

RAMitchell commented Nov 17, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

Uh oh!

Uh oh!

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!

Uh oh!

Uh oh!

RAMitchell commented Oct 20, 2025 •

edited

Loading

RAMitchell commented Oct 28, 2025 •

edited

Loading