Unfortunately, libc++'s 3-argument std::hypot implementation is
numerically unstable until LLVM 19.x. Therefore, checking the arguments
for zeros is insufficient since an underflow can still occur resulting
in a zero norm which requires another check. As such, division by zero
cannot be reliably avoided.
Change-Id: I189c8dc722aaec1ebc3ec8b1a177e1d8ac3b36db
Some of the benchmark functions use the same name as other functions
in the ceres namespace. For example Axpby defines both benchmark but
also an utility function in eigen_vector_ops.h. It seems to confuse
some compilers and leads to a compilation error rooting deeper into
the benchmark header itself: it seems that the compiler can not
deduct which of the instances of such functions to use.
Wrapping the file into an anonymous namespace solves the problem.
Alternative could be to use benchmark namespace to make thins more
explicit, for example ceres::internal::benchmark.
Tested on the following configuration:
- macOS 15.4
- Xcode 16.3
- Apple M3 CPU
- google-benchmark 1.9.2 installed via homebrew
Change-Id: Id127015dd22de99c6c3da88e71f255736e0bed82
- Add methods to aceess the cached residuals and jacobian computed in
the optimization process in TinySolver. Usage of such methods will
retrieve the corresponding values associated with the converged
parameter.
- Reorder the Update() call to ensure that the jacobian/residuals
associated with the converged parameter are computed and cached.
Change-Id: If82e19d67d28b057833357f2c9a75b2d0fd139af
Using preprocessor directives in a macro expansion is a (non-standard)
language extension that is not supported by MSVC.
Change-Id: I6e158b108a9c13cd277afb96cab2017ac192e3a2
Previously these classes in analogy with ceres::Problem's interface
had interfaces to allow bare pointers as well as unique_ptrs. This
CL changes the API to always use unique_ptr, this is less error prone
and makes the default ownership semantics clearer.
Change-Id: I7577a90761f341c7e009c248c820f0fec2e6f32d
Starting with SuiteSparse version 7.4.0 CHOLMOD has support for single
precision matrices. This allows us to have single precision and mixed
precision solves when using the SUITE_SPARSE backend.
This CL also fixes sparse_cholesky_test which was completely broken for
single precision testing.
Sample performance on my Mac.
/usr/bin/time -l ./bin/bundle_adjuster --input=../../Downloads/problem-3068-310854-pre.txt
<SNIP>
Cost:
Initial 9.099334e+07
Final 4.161838e+06
Change 8.683150e+07
Minimizer iterations 6
Successful steps 4
Unsuccessful steps 2
Time (in seconds):
Preprocessor 2.528222
Residual only evaluation 0.142804 (5)
Jacobian & residual evaluation 0.424014 (4)
Linear solver 54.083396 (5)
Minimizer 54.895752
Postprocessor 0.024564
Total 57.448539
Termination: NO_CONVERGENCE (Maximum number of iterations reached. Number of iterations: 5.)
59.04 real 341.24 user 5.49 sys
5776375808 maximum resident set size
<SNIP>
616329634071 instructions retired
929475980510 cycles elapsed
5375034560 peak memory footprint
/usr/bin/time -l ./bin/bundle_adjuster --input=../../Downloads/problem-3068-310854-pre.txt -mixed_precision_solves
<SNIP>
Cost:
Initial 9.099334e+07
Final 4.148930e+06
Change 8.684441e+07
Minimizer iterations 6
Successful steps 4
Unsuccessful steps 2
Time (in seconds):
Preprocessor 2.580217
Residual only evaluation 0.144098 (5)
Jacobian & residual evaluation 0.396723 (4)
Linear solver 23.636074 (5)
Minimizer 24.427163
Postprocessor 0.023790
Total 27.031170
Termination: NO_CONVERGENCE (Maximum number of iterations reached. Number of iterations: 5.)
28.58 real 128.53 user 2.37 sys
4818386944 maximum resident set size
<SNIP>
395186936091 instructions retired
368802808856 cycles elapsed
4327029824 peak memory footprint
Change-Id: I1f137b0dd12da8da7f9ced338dd8f20f4bbdf99d
If the user has checked out the submodules in third_party, they will be
used, otherwise we will try and find the system installed versions of
these dependencies and use them if they are modern enough.
Change-Id: I52164bc48a6ea804b85cdda05fee9cb94632f6c0
Generator expressions were introduced in CMake version 3.24. Without
them including absl::log_flags as an internal dependency causes
linking problems on some platforms. So for platforms with an older
CMake version we make absl::log_flags a public dependency.
Change-Id: I686f41e76b0ac17a03f1cd1d614372a7f130dfd9
Previously this header was included in absl/log/log.h
but going forward this header will not be included.
Change-Id: Ie663b221d90e32a852fb00dfe20f731dcd1dcf6b
Replace WallTimeInSeconds with absl::Now and use
absl::Time and absl::Duration objects instead of doubles.
wall_time.h/cc -> event_logger.h/cc
Change-Id: I41279961368840fbdf6bb3456ffdbdf2f9bfb85b
Replace ceres::String* with their more modern and performant
absl strings library equivalent and delete our string
manipulation library.
Change-Id: Iecbdba9864e0abf329778f81fdc0708f78f7594f
Ceres Solver was using an old forked version of FixedArray,
now that we are using absl, we can use the official version
that ships with it.
Change-Id: Ic88d7f6e8a49b928d611f7cbb04172452b322b01
1. Add abseil-cpp as a submodule. We are tracking the latest LTS
release, which is lts_2024_01_16.
2. Replace glog/gflags with absl::log and absl::flags.
3. Remove miniglog
4. Also take a whack at making the bazel build work with
abseil-cpp and gtest.
There are a number of TODOs in this CL that still need to be resolved.
Change-Id: I39355ed7d61375be4ebcbc8596d9cc70acc1c678
cuDSS could be used as an alternative for SuiteSparse and EigenSparse
in case if CUDA capable GPU is available.
Change-Id: I7a567093ce91363478118153e181134ed5804573
Sergiu Deitsch