Add sequence Cholesky solver (#1396)#1396
Closed
cdtwigg wants to merge 1 commit into
Closed
Conversation
meta-cla
Bot
added
the
CLA Signed
Contributor
|
@cdtwigg has exported this pull request. If you are a Meta employee, you can view the originating Diff in D105035576. |
meta-codesync
Bot
changed the title
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 14, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync
Bot
force-pushed
the
export-D105035576
branch
from
db53809 to
8622ab2
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 14, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync
Bot
force-pushed
the
export-D105035576
branch
from
8622ab2 to
e66aea0
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 14, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 14, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync
Bot
force-pushed
the
export-D105035576
branch
2 times, most recently
from
6bcccba to
f2eea7a
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 15, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync
Bot
force-pushed
the
export-D105035576
branch
from
f2eea7a to
5e30ec4
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 15, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 15, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync
Bot
force-pushed
the
export-D105035576
branch
2 times, most recently
from
3415ae4 to
4dbf529
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 15, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync
Bot
force-pushed
the
export-D105035576
branch
2 times, most recently
from
a82137d to
125a1aa
Compare
meta-codesync Bot
pushed a commit
that referenced
this pull request
May 15, 2026
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
Summary: Add a separate `SequenceCholeskySolver` for sequence problems where the existing online-QR solver is too expensive. The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before `J^T J`, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve. This diff also extracts shared sequence-solver helpers into `SequenceSolverBase`, exposes read-only `SequenceSolverFunction` accessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size, `J^T J` packing, double normal-equation accumulation, and 93-joint larger-character cases. Reviewed By: cstollmeta Differential Revision: D105035576
meta-codesync
Bot
force-pushed
the
export-D105035576
branch
from
125a1aa to
6d42147
Compare
Contributor
|
This pull request has been merged in 879e212. |
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
1 participant
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:
Add a separate
SequenceCholeskySolverfor sequence problems where the existing online-QR solver is too expensive.The new solver accumulates compact banded normal equations online, packs frame-local and universal columns contiguously before
J^T J, supports chunked multithreaded accumulation with deterministic boundary and sequence-error handling, and solves the banded system with block LDLT by default. A scalar banded LDLT path remains user-switchable for debugging and benchmarking. Compact normal-equation accumulation defaults to double precision so float error functions can still evaluate residuals and Jacobians in float without losing precision before the double LDLT solve.This diff also extracts shared sequence-solver helpers into
SequenceSolverBase, exposes read-onlySequenceSolverFunctionaccessors needed by the solver, wires the new solver into Momentum build/fwd generation, adds solver timing counters for normal-equation accumulation and linear solve, and adds a benchmark target that compares QR, block LDLT, scalar LDLT behavior, chunk size,J^T Jpacking, double normal-equation accumulation, and 93-joint larger-character cases.Reviewed By: cstollmeta
Differential Revision: D105035576