Opened 3 months ago
Closed 3 months ago
#31262 closed enhancement (fixed)
Implement non zero chunks for sparse bitsets
| Reported by: | gh-kliem | Owned by: | |
|---|---|---|---|
| Priority: | major | Milestone: | sage-9.3 |
| Component: | geometry | Keywords: | combinatorial polyhedron, sparse bitsets |
| Cc: | stumpc5, tscrim | Merged in: | |
| Authors: | Jonathan Kliem | Reviewers: | Travis Scrimshaw |
| Report Upstream: | N/A | Work issues: | |
| Branch: | b3212d0 (Commits, GitHub, GitLab) | Commit: | b3212d0bb74109461114b23d63fc26e604c121f2 |
| Dependencies: | #27103 | Stopgaps: |
Status badges
Description (last modified by )
For very large and challenging computations, most of the elements of the face lattice contain rather few atoms/bits. With little changes to the code, we can just collect the non zero chunks (64,128 or 256 bits).
RoaringBitmap? performs better (starting with maybe 100,000 bits, but not with less, as container contain 64k bits), but that would add an extra dependency and make things more complicated.
To avoid code duplications, we also gather some functions in bitset_intrinsics.h that work just the same (e.g. intersection and union).
(This actually accounts for most changes by this ticket.)
Before (on #27103):
sage: P = polytopes.Birkhoff_polytope(5) sage: C = CombinatorialPolyhedron(P) sage: %time _ = C.f_vector() CPU times: user 468 ms, sys: 0 ns, total: 468 ms Wall time: 467 ms sage: P = polytopes.Birkhoff_polytope(5) sage: C = CombinatorialPolyhedron(P) sage: %time _ = C.f_vector() CPU times: user 440 ms, sys: 0 ns, total: 440 ms Wall time: 439 ms sage: P = polytopes.hypercube(14) sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time _ = C.f_vector() CPU times: user 5.13 s, sys: 4.03 ms, total: 5.14 s Wall time: 5.13 s sage: P = polytopes.hypercube(15) sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time _ = C.f_vector() CPU times: user 30 s, sys: 32 ms, total: 30 s Wall time: 30 s sage: P = polytopes.permutahedron(6) sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time _ = C.f_vector() CPU times: user 1.68 ms, sys: 6 µs, total: 1.68 ms Wall time: 1.69 ms sage: P = polytopes.permutahedron(7) sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time _ = C.f_vector() CPU times: user 89.1 ms, sys: 4 µs, total: 89.1 ms Wall time: 88.9 ms sage: P = polytopes.permutahedron(8, backend='normaliz') sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time _ = C.f_vector() CPU times: user 14.2 s, sys: 8 ms, total: 14.2 s Wall time: 14.2 s sage: P = polytopes.associahedron(['A', 11], backend='normaliz') sage: C = CombinatorialPolyhedron(P) sage: %time _ = C.f_vector() CPU times: user 24.3 s, sys: 8.01 ms, total: 24.3 s Wall time: 24.3 s
After:
# Slight slowdown for few atoms. sage: P = polytopes.Birkhoff_polytope(5) sage: C = CombinatorialPolyhedron(P) sage: %time C.f_vector() CPU times: user 475 ms, sys: 0 ns, total: 475 ms Wall time: 474 ms (1, 120, 5040, 50250, 233400, 631700, 1113700, 1367040, 1220550, 817150, 419225, 167200, 52120, 12600, 2300, 300, 25, 1) sage: P = polytopes.hypercube(14) sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time C.f_vector() CPU times: user 1.67 s, sys: 0 ns, total: 1.67 s Wall time: 1.66 s (1, 16385, 131069, 487383, 1117948, 1769482, 2047331, 1788501, 1200342, 622908, 248963, 75361, 16640, 2470, 197, 1) sage: P = polytopes.hypercube(15) sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time C.f_vector() CPU times: user 7.28 s, sys: 8 ms, total: 7.28 s Wall time: 7.28 s (1, 32769, 278525, 1105876, 2723539, 4657926, 5866861, 5629624, 4196907, 2454738, 1128127, 404404, 110929, 22386, 3059, 226, 1) sage: P = polytopes.permutahedron(6) sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time C.f_vector() CPU times: user 1.44 ms, sys: 79 µs, total: 1.52 ms Wall time: 1.52 ms (1, 721, 1987, 1956, 808, 120, 1) sage: P = polytopes.permutahedron(7) sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time C.f_vector() CPU times: user 20.8 ms, sys: 0 ns, total: 20.8 ms Wall time: 20.9 ms (1, 5041, 16251, 19761, 11144, 2860, 267, 1) sage: P = polytopes.permutahedron(8, backend='normaliz') sage: P1 = P.stack(next(P.face_generator(1))) sage: C = CombinatorialPolyhedron(P1) sage: %time C.f_vector() CPU times: user 616 ms, sys: 8 ms, total: 624 ms Wall time: 623 ms (1, 40321, 148899, 215690, 154215, 56022, 9489, 572, 1) # No change for simplicial/simple polytopes. sage: P = polytopes.associahedron(['A', 11], backend='normaliz') sage: C = CombinatorialPolyhedron(P) sage: %time C.f_vector() CPU times: user 24.3 s, sys: 16.3 ms, total: 24.3 s Wall time: 24.3 s (1, 208012, 1144066, 2735810, 3730650, 3197700, 1790712, 659736, 157080, 23100, 1925, 77, 1)
This is the last subsequent improvement in comparison to sage 8.9 mentioned in https://arxiv.org/abs/1905.01945.
Follow ups:
- Move things from
geometry/polyhedron/combinatorial_polyhedronto a more general location. - Some renamings for lattices in general
face->element. - Expose to simplical complexes.
- Expose to lattice of flats of matroids.
Change History (13)
comment:1 Changed 3 months ago by
- Status changed from new to needs_review
comment:2 Changed 3 months ago by
- Status changed from needs_review to needs_work
comment:3 Changed 3 months ago by
- Commit changed from 24ff7ccf96a970655e2ce4c4db37cce343bd9c40 to eb6046e8eed53d2511f9a3d4fcf95f034df39108
Branch pushed to git repo; I updated commit sha1. New commits:
| eb6046e | improved documentation and fixed error
|
comment:4 Changed 3 months ago by
- Status changed from needs_work to needs_review
comment:5 Changed 3 months ago by
Possibly it might make sense to factor this ticket in two:
- One that removes duplicates in
bitset_intrinsics.h. - One that actually adds the slight modifications for sparse bitsets.
comment:6 Changed 3 months ago by
- Commit changed from eb6046e8eed53d2511f9a3d4fcf95f034df39108 to 4362e2ce0a52f8480793d34667a582d90163bc78
Branch pushed to git repo; I updated commit sha1. New commits:
| 4362e2c | use enum to make code a bit more canonical
|
comment:7 Changed 3 months ago by
- Commit changed from 4362e2ce0a52f8480793d34667a582d90163bc78 to fc0f1d96ff897f1897125956d9495e5536888dac
Branch pushed to git repo; I updated commit sha1. New commits:
| fc0f1d9 | resolve slow down comparison
|
comment:8 Changed 3 months ago by
- Description modified (diff)
Cleaner code makes things a bit faster, I suppose.
comment:9 Changed 3 months ago by
- Commit changed from fc0f1d96ff897f1897125956d9495e5536888dac to b3212d0bb74109461114b23d63fc26e604c121f2
Branch pushed to git repo; I updated commit sha1. New commits:
| b3212d0 | two mistakes with no intrinsics
|
comment:10 Changed 3 months ago by
All doctests pass on my side. With the current setup:
- AVX2, AVX, SSE4.1
- AVX, SSE4.1 (only
geometry/polyhedronanddata_structure/tested) - SSE4.1
- no intrinsics.
comment:11 Changed 3 months ago by
- Reviewers set to Travis Scrimshaw
- Status changed from needs_review to positive_review
LGTM.
comment:12 Changed 3 months ago by
Thank you.
comment:13 Changed 3 months ago by
- Branch changed from u/gh-kliem/small_improvements_for_sparse_bitsets to b3212d0bb74109461114b23d63fc26e604c121f2
- Resolution set to fixed
- Status changed from positive_review to closed
Note: See
TracTickets for help on using
tickets.
Failing doctests.