Repository navigation
Add cmake as the default for building stan math's tests - #3463
Open
SteveBronder wants to merge 21 commits into
Open
SteveBronder wants to merge 21 commits into
SteveBronder wants to merge 21 commits into
Conversation
Import and modernize the CMake setup for current vendored dependencies. Use CMake by default in runTests.py, group directory tests into executables, reuse per-source objects and PCH, and cap build parallelism at six jobs. Migrate Jenkins test stages, document the workflow, and ignore CMake build trees. Fix shared test helper linkage, Eigen fixture initialization, and the Bernoulli GLM invalid-label fixture; suppress GCC column-tracking notes. Validation: 21 runner, selection, and CMake build regression checks passed. Previously validated reverse and primitive function groups, generated probability and expression tests, and object reuse across test selections.
Collaborator
Author
|
Note this is not ready for a review yet. I think once all of the jenkins passes it will be good to look at |
Member
|
Exciting! |
Collaborator
Author
|
Yes I was fingers crossed those would not be necessary, but it is much much less than the last time 😆 . So I'll make a PR for those once all the tests pass here |
Keep grouped thread-pool tests independent of singleton initialization and restore thread-count environments. Match Windows UCRT and ARM definitions across TBB and consumers, disable broken Clang 9/10 PCH use, and validate varmat build setup before classifying signatures. Scope the old GoogleTest policy minimum and allow OpenCL header checks without a loader while retaining executable link requirements. Move CMake runner commands into runTests.py and update Jenkins and documentation.
Remove unused declarations exposed by the Jenkins -Werror build and group conditional callback captures without changing the derivative branches. Forward empty softmax inputs and exercise reverse-mode segment exceptions. Remove discarded reduction copies that consumed temporary matrix inputs. Add regressions covering temporary inputs to all six affected factories.
Collaborator
Author
|
I'll have to make a stacked pr here for all of the math changes. There are a few little places that I missed. |
Remove all 23 tuple captures from the OpenCL and CPU reverse callbacks. Use direct value captures to avoid Clang 10 unused explicit-capture warnings while keeping the original derivative bodies and callback ownership. Verify trivial callback destruction across 23 affected test translation units using validation-only compile-time assertions.
Link ARKODE for the new ODE wrappers and mirror Make's Windows math constant definition. Pair symmetric finite differences before weighting to avoid FMA cancellation noise without relaxing test tolerances. Keep MPI command registration objects in every test executable and use Stan's test flags for its customized GoogleTest main. Validated on the exact CI merge: 11 numerical/ODE executables and five MPI executables pass. The gradient suite also passes with Clang 10 and -Werror; Windows definition propagation and CMake 4 configuration checked.
Remove the unused capture and two profiling-test variables rejected by Clang 10. Pair symmetric fixed-step finite differences before weighting to cancel constant terms, with constant-function and analytic-polynomial regressions. Use absolute bounds for zero-valued matrix sums and solver finite differences, and verify the degenerate solver's analytic gradient independently. Preserve public APIs and thread-pool initialization. Validated all 20 Jenkins threading executables with Clang 10 and -Werror, seven numerical suites with FMA enabled for C and C++, and ten shuffled grouped-core repetitions each with threading on and off.
runPod supplies PARALLEL in the container environment, while the Groovy env lookup fell back to one. Resolve it through sh and reuse that count for CMake builds and CTest. Apply the same lookup to header and doxygen builds, retaining the default of one and explicit serial test overrides. Validated the old failure and repaired commands with a Groovy harness that separates the controller binding from the container shell environment. Checked unset, empty, four, six, and nonpositive values, plus serial test overrides. All eight CMake command-planning tests pass.
Update standalone header checks, test executables, PCH producers, and compile-only PCH producers to use the clearer source filename. Verified before renaming that header targets compile ordinary ELF objects with PCH enabled on Clang 18, without producing PCH artifacts. After renaming, rebuilt three header checks, a test executable, and the unit/compile-only PCH paths. The test and CMake planning checks pass.
Capture the nested arena count explicitly across GCC and Clang. Preserve value categories when returning unit_vector_free results and remove the remaining unused OpenCL and normal-sufficient variables. Match Make's ARM floating-point options for Stan and SUNDIALS, honor OpenCL header overrides, and select existing complete test aggregates so Make can use the requested parallelism. Split Jenkins expression generation into 24 batches to fit Clang 10 compilation within its 64 GiB pod.
Collaborator
|
Would it be worth updating the vendored TBB instead of cmake-ing around it? |
Member
Doing this eventually is one of the main reasons we’d like CMake, but it will be hard to do for the period of time where we’d like to still support the make-only build system. I think it’s better off doing the upgrade separately (and, unfortunately, a good bit later) |
Collaborator
Author
|
@andrjohns yes. To clarify the steps here we will be doing the following
So it is a bit of a long road, but there is a clear path. We can also start the cmake stuff in the other repos now. There is not really a stan math dependency to using cmake in the cmdstan and stan repos. |
Add cmake/StanTest.cmake with the shared test flags, PCH creation and test configuration, and use it for the compile test. Replace the duplicated per-test and grouping logic in test/ and test/unit/ with three functions: stan_add_test_binary, stan_add_gtest and stan_add_test_group. Umbrella unit tests now get the unit_subtests target and stan_build label, and use mpiexec only when STAN_MPI is on. Jenkins now tests whole directories through their <prefix>_<dir>_all_tests executables instead of building every per-file executable: expressions, the threading unit tests on the main branch, and each test/prob distribution directory. runTests.py gains --test-jobs to run several executables at once, and a trailing slash in a glob selects only directories. Linking each distribution directory into one executable exposed name collisions. Move the plain TESTs in twelve prob fixture headers into the matching prim unit tests, and give four fixture classes unique names (neg_binomial_2 ccdf_log, poisson_log, loglogistic cdf and std_normal cdf_log). The neg_binomial_2 collision made the cdf_log tests run the ccdf_log fixture.
Start from develop's Jenkinsfile, keeping its stages, pod sizes, stanorg/ci:v1 image, changed-test selection and GNU parallel distribution runs, and replace each make call and make/local setting with CMake options passed through runTests.py: - Add STAN_SANITIZE_ADDRESS and cpplint/test-math-dependencies targets, and print the build settings at configure time in place of make print-compiler-flags. - Build test objects as PIC and link executables as PIE. - runTests.py --fresh deletes CMakeCache.txt and the top-level CMakeFiles directory, as cmake --fresh does, so it works with the image's CMake 3.16. Run ctest from the build directory instead of using --test-dir (3.20). - Build selected files and distribution fixtures through one stan_selected_tests target, since CMake's top-level Makefile builds separate targets one at a time. - Accept distribution fixtures (test/prob/*/*_test.hpp) in runTests.py and add --configure-only. getDependencies.py lists each fixture's includes with the configured build's compile command instead of make. - Keep the distribution fixtures out of prob_pch. Clang 7 runs static initializers from a PCH out of order, so REGISTER_TYPED_TEST_SUITE_P checked the suites before TYPED_TEST_P had added any tests and every generated test aborted.
- cpplint: pass explicit file lists like make/cpplint, since --recursive needs Python 3.9 and stanorg/ci:v1 has 3.8. - varmat compatibility checks: return the result from non-static functions so -Wall -Werror builds have no unused variables or functions. - Expression tests: split into 4 files per job like develop; one file per job ran clang out of memory. - Add a ReleaseTesting configuration (-O1 without NDEBUG, so assertions stay on as in make) and make it the runTests.py default.
Contributor
Jenkins Console Log Machine informationDistributor ID: Ubuntu Description: Ubuntu 20.04.3 LTS Release: 20.04 Codename: focal CPU: Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian Address sizes: 52 bits physical, 57 bits virtual CPU(s): 192 On-line CPU(s) list: 0-191 Thread(s) per core: 2 Core(s) per socket: 48 Socket(s): 2 NUMA node(s): 2 Vendor ID: AuthenticAMD CPU family: 25 Model: 17 Model name: AMD EPYC 9474F 48-Core Processor Stepping: 1 Frequency boost: enabled CPU MHz: 1497.678 CPU max MHz: 4114.4229 CPU min MHz: 1500.0000 BogoMIPS: 7189.39 Virtualization: AMD-V L1d cache: 3 MiB L1i cache: 3 MiB L2 cache: 96 MiB L3 cache: 512 MiB NUMA node0 CPU(s): 0-47,96-143 NUMA node1 CPU(s): 48-95,144-191 Vulnerability Gather data sampling: Not affected Vulnerability Indirect target selection: Not affected Vulnerability Itlb multihit: Not affected Vulnerability L1tf: Not affected Vulnerability Mds: Not affected Vulnerability Meltdown: Not affected Vulnerability Mmio stale data: Not affected Vulnerability Old microcode: Not affected Vulnerability Reg file data sampling: Not affected Vulnerability Retbleed: Not affected Vulnerability Spec rstack overflow: Mitigation; Safe RET Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; STIBP always-on; PBRSB-eIBRS Not affected; BHI Not affected Vulnerability Srbds: Not affected Vulnerability Tsa: Mitigation; Clear CPU buffers Vulnerability Tsx async abort: Not affected Vulnerability Vmscape: Mitigation; IBPB before exit to userspace Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl xtopology nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local user_shstk avx512_bf16 clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d debug_swap G++: g++ (Ubuntu 9.4.0-1ubuntu1~20.04) 9.4.0 Copyright (C) 2019 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. Clang: clang version 10.0.0-4ubuntu1 Target: x86_64-pc-linux-gnu Thread model: posix InstalledDir: /usr/bin |
Member
|
Latest jenkins run looks pretty good, do you want to split off the c++ changes necessary for this and maybe look into breaking up the |
Collaborator
Author
|
Yes I'm going to need to make a good few number of stacked PRs. I'd like this PR to be just cmake and nothing else |
This branch has not been deployed
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
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.
Summary
This is my new take on having a cmake scheme. It imports a lot of what I did in #3068. This is the nth time we have attempted to make a cmake scheme for our build setup, but this time, I feel good about it! I've made many smaller PRs to try to get the math library ready for this, so now instead of 100k+ changes we have about 1K+ changes.
Since this is rather new for stan developers the readme.md has a big intro for how to use the new cmake backend. After a few releases I will put that in the normal docs page. But while we are moving to using this new tool I think it is better we are very verbose in showing people how to use it.
Tests
./runTests.pyand jenkins now use the cmake backend.The cmake creates folder level test executables and individual test executables.
The folder level tests are much faster to compile now. For example the
test/unit/math/mix/funtests all compile into one object file which took about 30 minutes on my machine using 24 cores (and at max about 65 gb of memory). On develop these take about an hour and a half on my machine.Side Effects
Yes. I mean this is a pretty big change.
One really nice change is that we have precompiled headers with this. That should speed up test compilation. In another PR I am planning to do some tricks with the distribution tests which should speed compilation for those by a good bit.
Release notes
Add cmake as the build backend
Checklist
Copyright holder: Simons Foundation
The copyright holder is typically you or your assignee, such as a university or company. By submitting this pull request, the copyright holder is agreeing to the license the submitted work under the following licenses:
- Code: BSD 3-clause (https://opensource.org/licenses/BSD-3-Clause)
- Documentation: CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
the basic tests are passing
./runTests.py test/unit)make test-headers)make test-math-dependencies)make doxygen)make cpplint)the code is written in idiomatic C++ and changes are documented in the doxygen
the new changes are tested