Develop stream 2024-07-01#512
Merged
stanleytsang-amd merged 69 commits intoROCm:developfrom Aug 8, 2024
Merged
Conversation
Member
Author
|
@samjwu The clang format CI step seems to not handle external PRs properly, I think this might also be happening in other repositories. |
Naraenda
requested review from
a team,
RobsonRLemos,
lawruble13,
stanleytsang-amd and
umfranzw
as code owners
Member
Author
|
@stanleytsang-amd Let me know if I should squash the commits into their respective merge commits. |
Collaborator
Run git fetch origin develop
|
Naraenda
force-pushed
the
develop_stream_20240701
branch
from
4ccc1c5 to
b6c8731
Compare
Member
Author
|
Rebased to get the clang-format ci fixes. |
rocm-docs-core distributes headers and stylesheets for doxygen for embedding its HTML output into sphinx. These mostly fix dark-theme and other minor visual issues when doxygen output is used this way.
docs(api reference): rocm-docs-core headers and stylesheets in doxyfile See merge request amd/libraries/rocRAND!326
…ream' Improve accuracy of Poisson histogram test Closes ROCm#240 See merge request amd/libraries/rocRAND!327
…ns.hpp and src/rng/device_engines.hpp
…tream' Resolve "Remove deprecated internal headers" Closes ROCm#341 See merge request amd/libraries/rocRAND!330
AMDGPU_TARGETS doesn't pick up updates correctly (needs cache clean) whereas GPU_TARGETS does. Every other doc and CI too refers to GPU_TARGETS.
Resolve "Some host generators might not support large sizes due to min / max" See merge request amd/libraries/rocRAND!329
Recent changes required for HIP graph support added a new path with approximation of Poisson with normal distribution when lambda is large. However, the decision whether to use the alias/CDF methods or the approximation is made in the kernel for every generated value even though lambda is the same. This change moves it to host side: depending on lambda the kernel is launched with one of two distributions (poisson_distribution or poisson_distribution_huge).
Resolve "Document HIP Graph support" Closes ROCm#360 See merge request amd/libraries/rocRAND!335
…p_stream' Fix performance regression of Poisson distribution introduced by HIP graph support Closes ROCm#366 See merge request amd/libraries/rocRAND!336
hipcc from ROCm 6.2 does not add `-mllvm -amdgpu-early-inline-all=true -mllvm -amdgpu-function-calls=false` by default.
Improve performance on ROCm 6.2 See merge request amd/libraries/rocRAND!337
…as the former is unintuitive
ci(.gitlab-ci.yml): replace 'ROCM_PATH' variable with 'env:HIP_PATH' as the former is unintuitive See merge request amd/libraries/rocRAND!339
Remove unused FindTestU01.cmake See merge request amd/libraries/rocRAND!342
…te' into 'develop_stream' Resolve "Add check for nullptr data when calling host generate." Closes ROCm#369 See merge request amd/libraries/rocRAND!341
operator >> has higher precedence than operator &. This bug causes very low quality in crush tests.
…ream' Fix threefry2x64 and threefry4x64 Closes ROCm#371 See merge request amd/libraries/rocRAND!343
Naraenda
force-pushed
the
develop_stream_20240701
branch
from
b6c8731 to
d4f972c
Compare
Member
Author
|
Rebased and added threefry fixes by @ex-rzr |
umfranzw
approved these changes
Aug 7, 2024
Collaborator
umfranzw
left a comment
umfranzw
left a comment
There was a problem hiding this comment.
I've finished working my way through this one, and I think it looks good. All our CI checks are also passing now.
@stanleytsang-amd if you have no objections, I'm ok with this one going in.
stanleytsang-amd
requested changes
Aug 7, 2024
Collaborator
stanleytsang-amd
left a comment
stanleytsang-amd
left a comment
There was a problem hiding this comment.
LGTM once the packaging version is updated.
stanleytsang-amd
approved these changes
Aug 8, 2024
ammallya
pushed a commit
that referenced
this pull request
Oct 28, 2025
* Removed accidentally included #include "hip/amd_detail/host_defines.h" * chore(gitignore): ignore python venvs * docs(api reference): rocm-docs-core headers and stylesheets in doxyfile rocm-docs-core distributes headers and stylesheets for doxygen for embedding its HTML output into sphinx. These mostly fix dark-theme and other minor visual issues when doxygen output is used this way. * improve accuracy of poisson histogram test * fix format and copyright dates * feat(test): Added CMake option RUN_EXTRA_TESTS * Removed deprecated internal headers, src/rng/distribution/distributions.hpp and src/rng/device_engines.hpp * Using .lint:clang-format * feat(test): Added large size tests for host generators * fix(generator): Fixed the usage of min in host generators * docs(dyn_ordering): Use GPU_TARGETS instead of AMDGPU_TARGETS AMDGPU_TARGETS doesn't pick up updates correctly (needs cache clean) whereas GPU_TARGETS does. Every other doc and CI too refers to GPU_TARGETS. * Use alias method in rocrand_discrete for MTGP32, LFSR113 and ThreeFry discrete_alias is faster than discrete_cdf. Though discrete_cdf can be used with PRNGs, it is supposed to be used with QRNGs (Sobol generators) as it maintains quasi-randomness. * refactor mt19937 to support host version as well * update test * move jump_ahead_thread_count back to template param * implement memcpy in host and device systems * fix compilation issues and segfault * Removed 'apt-get install flang' * fix jump_ahead on host * create host implementation of some functions refactor generate_long_mt19937 to work on host as well * fix remaining inconsistencies in host mt19937 generator * fix format end compile errors * fix missing gen_next_n calls fix clang-format issues * fix format issues and missing __host__s * fix messed up host/device allocations * fix merge conflicts fix format * fix format issues and compile error * fix format issues * fix format issue * disable most mt19937 host tests for normal run (enabled for slow test run) * fix review comments * remove synchronization from host_system::launch function add synchronization to host_system alloc, free and memcpy * Implement asynchronous initialization of poisson distribution * use ROCRAND_HIP_FATAL_ASSERT for hipDeviceSynchronize call * generate_poisson test with many lambdas * Testing poisson with hipGraphs * Test [blocking] host_generator with non-blocking stream * Fixing poisson distribution selection in benchmark_tuning * Updated changelog * fix(docs): Added links to unaccessible doc pages * fix(docs): Removed duplicated CUDA Compatibility section from Programmer's guide * Added hipGraphs doc and sample * Fix performance regression of Poisson distribution Recent changes required for HIP graph support added a new path with approximation of Poisson with normal distribution when lambda is large. However, the decision whether to use the alias/CDF methods or the approximation is made in the kernel for every generated value even though lambda is the same. This change moves it to host side: depending on lambda the kernel is launched with one of two distributions (poisson_distribution or poisson_distribution_huge). * clang-format: Break after attributes * Add missing __forceinline__ to improve performance on ROCm 6.2 hipcc from ROCm 6.2 does not add `-mllvm -amdgpu-early-inline-all=true -mllvm -amdgpu-function-calls=false` by default. * Remove meaningless code in xorwow introduced during rebase/merge * style: update formatting * ci(.gitlab-ci.yml): replace 'ROCM_PATH' variable with 'env:HIP_PATH' as the former is unintuitive * ci(.gitlab-ci.yml): do not force download deps on windows * ci(.gitlab-ci.yml): pass amdclang filepath properly to windows package build test * Remove unused FindTestU01.cmake * Added checks for nullptr data with tests * Fix bit rotation for threefry2x64 and threefry4x64 operator >> has higher precedence than operator &. This bug causes very low quality in crush tests. * chore: bump version --------- Co-authored-by: Lőrinc Serfőző <[email protected]> Co-authored-by: Gergely Meszaros <[email protected]> Co-authored-by: Nol Moonen <[email protected]> Co-authored-by: Bence Parajdi <[email protected]> Co-authored-by: Nick Breed <[email protected]> Co-authored-by: Anton Gorenko <[email protected]> [ROCm/rocRAND commit: a296589]
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
8 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR brings various updates, intended for ROCm 6.3.
It contains the following merge commits:
GPU_TARGETSinstead ofAMDGPU_TARGETThis PR does not contain fixes for the recent performance regressions, we'll either add them here or create a new PR for those.