In https://ceres-solver-review.git.corp.google.com/c/ceres-solver/+/23802
the computation of the norm of a quaternion
scale = 1/sqrt(q[0] * q[0] + q[1] * q[1] + q[2] * q[2] + q[3] * q[3]);
was replaced by
scale = 1/hypot(q[0], q[1], hypot(q[2], q[3]));
while this appear to be a more accurate computation because of the
use of hypot which can handle over and underflow it introduces a
bug for the case where q[2] = q[3] = 0.
While the hypot(q[2], q[3]) == 0 as scalars, if q[2] and q[3] are
jets, then the derivative will be NaN. Which means that even though
q[0] or q[1] is non-zero and the norm of the quaternion is non-zero,
and the resulting derivative is finite, this way of computing the
scale will produce nans in the derivative of scale.
The following quaternion will replicate the problem described above.
using Jet = ceres::Jet<double, 4>;
std::array<Jet, 4> quaternion = {Jet(1.0, 0), Jet(0.0, 1), Jet(0.0, 2), Jet(0.0, 3)};
This CL reverts the change to QuaternionRotatePoint and
adds a test for it.
Thanks to Jonathan Taylor for reproducing this bug.
Change-Id: I0fbbcc77d6945a38563d82efba4429f4b5278cd5
* Use relative error instead of absolute error.
* Update tolerance to account for embedded GPUs such as the Jetson TX2.
Change-Id: I05742ecbfd915e797fcbc4137acc5d1b513cd465
1. Add threading to all three subclasses of BlockRandomAccessMatrix.
i.e. BlockRandomAccessDenseMatrix, BlockRandomAccessSparseMatrix
and BlockRandomAccessDenseMatrix.
For BlockRandomAccessDenseMatrix and BlockRandomAccessSparseMatrix
this just means SetZero is parallelized. Which by itself is no
big deal, but by doing so, the constructor for all three subclasses
become uniform.
BlockRandomAccessSparseMatrix::SymmetricRightMultiplyAndAccumulate
maybe threaded in the future if needed.
BlockRandomAccessDiagonalMatrix is the biggest beneficiary. SetZero
Invert and RightMultiplyAndAccumulate are all threaded now.
2. Change the storage in BlockRandomAccessDiagonalMatrix from
TripletSparseMatrix to CompressedRowSparseMatrix. This has no
performance implications since we do not really use the capabilities
of the underlying matrix indexing representation. This is a forward
looking change when we decide to transfer this matrix to the GPU,
a CompressedRowSparseMatrix will save on a data conversion.
3. Use std::unique_ptr as needed and eliminate the need for custom
destructors.
4. Modify CompressedRowSparseMatrix::CreateBlockDiagonalMatrix to
take a nullptr as the data vector.
Fixes https://github.com/ceres-solver/ceres-solver/issues/936
Fixes https://github.com/ceres-solver/ceres-solver/issues/935
Change-Id: Ia6487f2d924fbe669835bdcc38abf2b451bda4ee
CoordinateDescentMinimizer optimizes one parameter block at a time.
To do this, it manipulates the parameter block object. It was doing
so inconsistently, where the tangent space offset was being set to
zero but the ambient state offset was not being set to zero. This
did not cause problems because these offsets were not really being
used inside the CoordinateDescentMinimizer. However the recent
change which parallelizes Program::Plus uncovered this bug.
The reason this bug was not caught was because, CoordinateDescentMinimizer
does not have any tests. I will fix this shortly, but in the interim
to unbreak inner iterations at head, this small change should go in.
Change-Id: I55d2698e8509f9cb5751e7a5180427129d86e720
Main focus of this change is to parallelize remaining operations (most of them
are operations on vectors) in code-path utilized with iterative Schur
complement.
Parallelization is handled using lazy evaluation of Eigen expressions.
On linux pc with intel 8176 processor parallelization of vector operations has
the following effect:
Running ./bin/parallel_vector_operations_benchmark
Run on (112 X 3200.32 MHz CPU s)
CPU Caches:
L1 Data 32 KiB (x56)
L1 Instruction 32 KiB (x56)
L2 Unified 1024 KiB (x56)
L3 Unified 39424 KiB (x2)
Load Average: 3.30, 8.41, 11.82
-----------------------------------
Benchmark Time
-----------------------------------
SetZero 10009532 ns
SetZeroParallel/1 10024139 ns
...
SetZeroParallel/16 877606 ns
Negate 4978856 ns
NegateParallel/1 5145413 ns
...
NegateParallel/16 721823 ns
Assign 10731408 ns
AssignParallel/1 10749944 ns
...
AssignParallel/16 1829381 ns
D2X 15214399 ns
D2XParallel/1 15623245 ns
...
D2XParallel/16 2687060 ns
DivideSqrt 8220050 ns
DivideSqrtParallel/1 9088467 ns
...
DivideSqrtParallel/16 905569 ns
Clamp 3502010 ns
ClampParallel/1 4507897 ns
...
ClampParallel/16 759576 ns
Norm 4426782 ns
NormParallel/1 4442805 ns
...
NormParallel/16 430290 ns
Dot 9023276 ns
DotParallel/1 9031304 ns
...
DotParallel/16 1157267 ns
Axpby 14608289 ns
AxpbyParallel/1 14570825 ns
...
AxpbyParallel/16 2672220 ns
-----------------------------------
Multi-threading of vector operations in ISC and program evaluation results into
the following improvement:
Running ./bin/evaluation_benchmark
--------------------------------------------------------------------------------------
Benchmark this 2fd81de
--------------------------------------------------------------------------------------
Residuals<problem-13682-4456117-pre.txt>/1 4136 ms 4292 ms
Residuals<problem-13682-4456117-pre.txt>/2 2919 ms 2670 ms
Residuals<problem-13682-4456117-pre.txt>/4 2065 ms 2198 ms
Residuals<problem-13682-4456117-pre.txt>/8 1458 ms 1609 ms
Residuals<problem-13682-4456117-pre.txt>/16 1152 ms 1227 ms
ResidualsAndJacobian<problem-13682-4456117-pre.txt>/1 19759 ms 20084 ms
ResidualsAndJacobian<problem-13682-4456117-pre.txt>/2 10921 ms 10977 ms
ResidualsAndJacobian<problem-13682-4456117-pre.txt>/4 6220 ms 6941 ms
ResidualsAndJacobian<problem-13682-4456117-pre.txt>/8 3490 ms 4398 ms
ResidualsAndJacobian<problem-13682-4456117-pre.txt>/16 2277 ms 3172 ms
Plus<problem-13682-4456117-pre.txt>/1 339 ms 322 ms
Plus<problem-13682-4456117-pre.txt>/2 220 ms
Plus<problem-13682-4456117-pre.txt>/4 128 ms
Plus<problem-13682-4456117-pre.txt>/8 78.0 ms
Plus<problem-13682-4456117-pre.txt>/16 49.8 ms
ISCRightMultiplyAndAccumulate<problem-13682-4456117-pre.txt>/1 2434 ms 2478 ms
ISCRightMultiplyAndAccumulate<problem-13682-4456117-pre.txt>/2 2706 ms 2688 ms
ISCRightMultiplyAndAccumulate<problem-13682-4456117-pre.txt>/4 1430 ms 1548 ms
ISCRightMultiplyAndAccumulate<problem-13682-4456117-pre.txt>/8 742 ms 883 ms
ISCRightMultiplyAndAccumulate<problem-13682-4456117-pre.txt>/16 438 ms 555 ms
ISCRightMultiplyAndAccumulateDiag<problem-13682-4456117-pre.txt>/1 2438 ms 2481 ms
ISCRightMultiplyAndAccumulateDiag<problem-13682-4456117-pre.txt>/2 2565 ms 2790 ms
ISCRightMultiplyAndAccumulateDiag<problem-13682-4456117-pre.txt>/4 1434 ms 1551 ms
ISCRightMultiplyAndAccumulateDiag<problem-13682-4456117-pre.txt>/8 765 ms 892 ms
ISCRightMultiplyAndAccumulateDiag<problem-13682-4456117-pre.txt>/16 435 ms 559 ms
JacobianSquaredColumnNorm<problem-13682-4456117-pre.txt>/1 1278 ms
JacobianSquaredColumnNorm<problem-13682-4456117-pre.txt>/2 1555 ms
JacobianSquaredColumnNorm<problem-13682-4456117-pre.txt>/4 833 ms
JacobianSquaredColumnNorm<problem-13682-4456117-pre.txt>/8 459 ms
JacobianSquaredColumnNorm<problem-13682-4456117-pre.txt>/16 250 ms
JacobianScaleColumns<problem-13682-4456117-pre.txt>/1 1468 ms
JacobianScaleColumns<problem-13682-4456117-pre.txt>/2 1871 ms
JacobianScaleColumns<problem-13682-4456117-pre.txt>/4 957 ms
JacobianScaleColumns<problem-13682-4456117-pre.txt>/8 528 ms
JacobianScaleColumns<problem-13682-4456117-pre.txt>/16 294 ms
End-to-end improvements with bundle_adjuster invoked with
./bin/bundle_adjuster --num_threads 28 --num_iterations 40 \
--linear_solver iterative_schur \
--preconditioner jacobi --input
---------------------------------------------
Problem this 2fd81de
---------------------------------------------
problem-13682-4456117-pre.txt 508.6 892.7
problem-1778-993923-pre.txt 763.8 1129.9
problem-1723-156502-pre.txt 6.3 14.4
problem-356-226730-pre.txt 76.3 116.2
problem-257-65132-pre.txt 38.6 52.0
Change-Id: Ie31cc5015f13fa479c16ffb5ce48c9b880990d49
Since c++11, we can depend on C++ threads always being available.
With the recent work on the performance of CXX threading, the
additional complexity of maintaining multiple backends for some
minor performance delta is not worth it
https://github.com/ceres-solver/ceres-solver/issues/886
Change-Id: Idee480b22a498daec9c4366da8589aa58eaf36a1
The version component macro names are delimited by multiple spaces in
the new release resulting in a failure to parse the version.
An additional guard ensures that if the version cannot be correctly
parsed it is discarded and a user warning is issued.
Fixes#919
Change-Id: I630f30dba0fd23979b6fe5d854e59701c22c3469
The constructor now takes a LinearSolver::Options as input
and uses that to compute the partitioning once and uses it
for its lifetime.
Change-Id: I9ef30df0b60f8fa91c8b5601c397b2d9314a2cc7
Parallel left products for PartitionedMatrixView using parallel for
loops with fair partitioning.
Updates for evaluation benchmark including replacing program with a
preprocessed one
Change-Id: Ia1cc3293f106cec6b7d933675cea4e7c5a6b71e4
The new SuiteSparse deployment bundles METIS 5.1.0 instead of 5.1.1 to
avoid heap corruption.
Fixes#918
Change-Id: Ie779dc2015c60e928c237529675e21d4635716a2
Parallel for with user-supplied [cumulative] iteration costs allows to
get performance improvements on problems with significantly different
time requirements per parallel loop iteration.
One of those problems is left multiplication with block-sparse matrix.
Using number of non-zero values per column block, we partition column
blocks into contiguous sets with approximately equal number of
operations to be performed.
Change-Id: I4a862a10a586cdfbec22e8168a3423537039abc2
- Enables relocatable installs if the CUDA libraries are not installed
in a location on the LD_LIBRARY_PATH.
- Also bump the minimum CMake version to 3.11 to reflect the issue
reported in #903.
Change-Id: I333882b7238c76104d739c7054f29cc35cc4e919
When AnalyzeCholesky is called on a Jacobian which has been
pre-ordered in the preprocessor, we assume a NATURAL ordering.
In this case, asking CHOLMOD to do a postordering is detrimental
to the performance. This regression was introduced in
https://github.com/ceres-solver/ceres-solver/commit/d09f7e9d5e3bfab2d7ec7e81fd6a55786edca17a
based on an email exchange with Prof. Tim Davis the author of CHOLMOD
who suggested that this should be innocous. Unfortunately this is not
the case. The following table shows the performance of bundle_adjuster
on a variety of problems. This matches the performance of the
bundle adjuster before the CL that introduced the regression.
Before After
problem-49-7776-pre.txt
SPARSE_NORMAL_CHOLESKY + AMD 0.8 0.6
SPARSE_NORMAL_CHOLESKY + NESDIS 0.8 0.6
SPARSE_SCHUR + AMD 0.1 0.1
SPARSE_SCHUR + NESDIS 0.1 0.1
problem-1778-993923-pre.txt
SPARSE_NORMAL_CHOLESKY + AMD 247.1 197.0
SPARSE_NORMAL_CHOLESKY + NESDIS 234.0 193.2
SPARSE_SCHUR + AMD 78.7 71.8
SPARSE_SCHUR + NESDIS 78.7 71.4
problem-1031-110968-pre.txt
SPARSE_NORMAL_CHOLESKY + AMD 26.1 22.4
SPARSE_NORMAL_CHOLESKY + NESDIS 26.0 23.0
SPARSE_SCHUR + AMD 14.9 13.1
SPARSE_SCHUR + NESDIS 14.9 13.1
problem-356-226730-pre.txt
SPARSE_NORMAL_CHOLESKY + AMD 43.9 34.7
SPARSE_NORMAL_CHOLESKY + NESDIS 42.5 34.8
SPARSE_SCHUR + AMD 6.8 6.1
SPARSE_SCHUR + NESDIS 6.8 6.1
problem-951-708276-pre.txt
SPARSE_NORMAL_CHOLESKY + AMD 170.3 139.0
SPARSE_NORMAL_CHOLESKY + NESDIS 168.9 139.8
SPARSE_SCHUR + AMD 52.3 47.9
SPARSE_SCHUR + NESDIS 52.0 47.7
Change-Id: I6421060dd835ab2fc81d270238f6f67e29f3ecd5
Add structure of transposed matrix to BlockSparseMatrix
Number of non-zero values per row block and cumulative non-zero
values count are maintained for transposed structure
Change-Id: Icf38bb7a734ca695c788579eece1c92d36d78e54
Sameer Agarwal