Commit Graph

22 Commits

Author SHA1 Message Date
Yuval Tassa 772326e9eb Fix potential out-of-range write in solver statistics saving.
PiperOrigin-RevId: 569094007
Change-Id: I2e596a7c03d76523359cae7ae7f4501b06256a38
2023-09-28 00:58:13 -07:00
Yuval Tassa 2cc1205498 Add island support to CG solver.
BEGIN_PUBLIC

Add island support to CG solver.

END_PUBLIC

PiperOrigin-RevId: 565653886
Change-Id: Ib971ae37dd011b8f2cbe1f88d8875e0ed657c7af
2023-09-15 05:47:24 -07:00
Yuval Tassa 86d8b912f4 Rejigged solver statistics in preparation for islanding.
- See public description below.
- Stopped incrementing the iteration count in saveStats().
- `sizeof(mjSolverStat) == 40`, so this ends up costing 160KB, up from 40KB.

BEGIN_PUBLIC

Changed the size of `mjData.solver`, the structure used to collect solver diagnostic information. The array is now of length `mjNISLAND * mjNSOLVER`, where each row of length `mjNSOLVER` contains separate solver statistics for each constraint island. Until solver islanding is implemented, only row 0 is used.

- The new constant `mjNISLAND` is set to 20.
- `mjNSOLVER` is reduced from 1000 to 200.
- Added `mjData.solver_nisland`, the number of islands for which the solver ran.
- `mjData.solver_niter` (renamed from mjData.solver_iter) and `mjData.solver_nnz` are now integer vectors of length `mjNISLAND`.

END_PUBLIC

PiperOrigin-RevId: 565030093
Change-Id: I773e918805c6ced79f0dab5f19ea23956760c8c5
2023-09-13 06:18:37 -07:00
Google DeepMind 040d3efce2 Add options for controlling early termination criteria of CG/Newton linesearch.
Useful for performance tuning models.

PiperOrigin-RevId: 563175078
Change-Id: Id64a510d6080038ea8d5876dabbf0b2b311435bb
2023-09-06 12:02:58 -07:00
Saran Tunyasuvunakool 4929077211 Remove mjMARKSTACK and mjFREESTACK macros.
PiperOrigin-RevId: 562844184
Change-Id: Id2f57f5d132c47094dad75c6dab7297d7aa73458
2023-09-05 11:41:43 -07:00
Yuval Tassa d5292976de Add internal utility for zeroing int vectors.
Remove unnecessary branches in related functions.

PiperOrigin-RevId: 562768873
Change-Id: I3964a7a953d95bcd4e29137a7b590be5ffa6e311
2023-09-05 07:15:14 -07:00
Yuval Tassa 29aa5e4a41 Allow dot products with sparse vectors to specify that one vector uses uncompressed memory.
PiperOrigin-RevId: 561951337
Change-Id: I1310d7c09d9a85ad9b43877155844a9d0ce6edab
2023-09-01 07:40:16 -07:00
Saran Tunyasuvunakool b979a26b10 Rename mj_stackAlloc to mj_stackAllocNum.
PiperOrigin-RevId: 561040193
Change-Id: I49ea21ffb5b596471d2aacefb1961c11c5c905d0
2023-08-29 08:43:03 -07:00
Yuval Tassa 5f3e22a45f Refactor a loop in engine_solver.c
PiperOrigin-RevId: 560997065
Change-Id: I26540e4f750eb18d9497bf35396adbb15cc3a122
2023-08-29 05:17:14 -07:00
Matthew Bennice 59138af10b Refactor stack allocations to have a clean byte allocation interface.
PiperOrigin-RevId: 558881470
Change-Id: I08c45cb5fa25a1a9c39560e598e805cfa5141c3e
2023-08-21 13:18:29 -07:00
Kyle Bayes f34cb22bd3 Reuse sparse Dof-Dof matrix to make compressed sparse M in mj_addM.
PiperOrigin-RevId: 556823353
Change-Id: I650ac83767d60c69859aae6e38aa331f43814378
2023-08-14 09:49:34 -07:00
Kyle Bayes 2d7d5319f7 Rework sparse addM function to support compressed sparse matrices.
Note:

- M is still a nv x nv uncompressed sparse matrix.
- NNZ precounting still needs to be implemented using the combineSparseCount helper function.

PiperOrigin-RevId: 554452710
Change-Id: I14adf58fc3acd4ed82d864d7c2e9e96ee6b0377d
2023-08-07 06:07:24 -07:00
Kyle Bayes ac1f41160d Add engine internal error macro mjERROR to prepend calling function to error function.
PiperOrigin-RevId: 541584847
Change-Id: I57447e64375413ee07aa96372a402e1ced59ff4a
2023-06-19 03:29:31 -07:00
Kyle Bayes 81d74c56e1 Separate mj_addM into dense and sparse functions.
PiperOrigin-RevId: 539066339
Change-Id: Ide3f9644fee46a5fb1a2e79a25be162b929e2421
2023-06-09 06:44:19 -07:00
Yuval Tassa 455b1cd2e2 Add spaces around comparison operators in engine source files.
PiperOrigin-RevId: 535989348
Change-Id: I883f7e82351299933c49b35a31842b5d8d6aea04
2023-05-28 05:02:34 -07:00
Saran Tunyasuvunakool f887c1e928 Add ASAN and MSAN instrumentation for arena and mjData stack.
PiperOrigin-RevId: 535635490
Change-Id: Ia1e58771bae25bf72c861f57452fb8faa2b918fb
2023-05-26 09:17:26 -07:00
Kyle Bayes 9924cce4b7 Add nnz per row precount method for mju_sqrMatTDSparse.
PiperOrigin-RevId: 521744893
Change-Id: Ie0fbfab552a680127f4acf9041b07916f8a2a490
2023-04-04 06:21:22 -07:00
Nimrod Gileadi 145fe7f354 Declare loop variables inside for loop declaration where possible.
PiperOrigin-RevId: 518846098
Change-Id: Icdb2e5894bd80110552e29270472eeb242e6cf2c
2023-03-23 06:49:09 -07:00
Kyle Bayes 8696247f9d Add mj_stackAllocInt to get correct size for allocating ints on mjData stack. Reducing stack memory usage by 10% - 15%.
PiperOrigin-RevId: 517672195
Change-Id: Ie14e6c9c99e8b9e1c5d3686b77631f2df0651c67
2023-03-18 13:04:14 -07:00
Kyle Bayes 056e849273 Provide improved mju_sqrMatTDSparse implementation that doesn't require dense memory allocation for sparse matrices.
PiperOrigin-RevId: 516812783
Change-Id: Ieb43337831d8d3b2c7f18a7facd8e0a0f2b0eff5
2023-03-15 06:56:16 -07:00
Yuval Tassa f4e7fa97af Add mju_mulVecMatVec, mutiplies a square matrix M by a vector x on both sides. Returns x^T * M * x.
PiperOrigin-RevId: 474292806
Change-Id: I3432469dbe1f02ccf5a13241c7aa12d824cbe034
2022-09-14 08:05:28 -07:00
Saran Tunyasuvunakool 1913a02b40 Initial open sourcing of MuJoCo.
PiperOrigin-RevId: 450374687
Change-Id: Ie3225a46ce095fc28ae8e63c326a640261f562bb
2022-05-23 01:08:49 -07:00