testSubgraphPreconditioner bad_alloc exception under i386 #182
Labels
bug
Bug report
Comments
5 tasks
|
I tested on a i386 box with gcc 5.4.0 on DigitalOcean: One test hangs: |
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
|
Now everything points to |
|
(cont.d) whose reason is that This maps to that Note that @jlblancoc @dellaert void KeyInfo::initialize(const GaussianFactorGraph &fg) {
const map<Key, size_t> colspec = fg.getKeyDimMap();
const size_t n = ordering_.size();
size_t start = 0;
cout << "SIZE: " << colspec.size() << endl;
for (const auto &[k, v] : colspec)
cout << "[" << k << ", " << (size_t)k << ", " << v << "], ";
cout << endl;
for (size_t i = 0; i < n; ++i) {
const size_t key = ordering_[i];
cout << "KEY:" << key << endl;
auto item = colspec.find(key);
cout << (item == colspec.end()) << endl;
const size_t dim = item->second;
this->emplace(key, KeyInfoEntry(i, dim, start));
start += dim;
}
numCols_ = start;
} |
|
I see. Well, indeed we want Key to always be 64 bit, so we should track
down we’re this goes wrong....
On Sun, Dec 22, 2019 at 23:58 Fan Jiang ***@***.***> wrote:
(cont.d) whose reason is that const map<Key, size_t> colspec =
fg.getKeyDimMap(); does not have the key.
This maps to that fg.getKeyDimMap() is misbehaving.
SIZE: 9
[8646911284551353321, 1001, 2], [8646911284551353322, 1002, 2], [8646911284551353323, 1003, 2], [8646911284551354321, 2001, 2], [8646911284551354322, 2002, 2], [8646911284551354323, 2003, 2], [8646911284551355321, 3001, 2], [8646911284551355322, 3002, 2], [8646911284551355323, 3003, 2],
KEY:1001
1
Note that 8646911284551353321 is 0x78*2^{63-8}-1001, and (size_t) k is
indeed 1001.
@jlblancoc <https://github.com/jlblancoc> @dellaert
<https://github.com/dellaert>
In short, the issue is with the wrong assumption that size_t is the same
with Key (uint_64 always).
void KeyInfo::initialize(const GaussianFactorGraph &fg) {
const map<Key, size_t> colspec = fg.getKeyDimMap();
const size_t n = ordering_.size();
size_t start = 0;
cout << "SIZE: " << colspec.size() << endl;
for (const auto &[k, v] : colspec)
cout << "[" << k << ", " << (size_t)k << ", " << v << "], ";
cout << endl;
for (size_t i = 0; i < n; ++i) {
const size_t key = ordering_[i];
cout << "KEY:" << key << endl;
auto item = colspec.find(key);
cout << (item == colspec.end()) << endl;
const size_t dim = item->second;
this->emplace(key, KeyInfoEntry(i, dim, start));
start += dim;
}
numCols_ = start;
}
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#182?email_source=notifications&email_token=ACQHGSPSQTG6F5TZDVYKNYDQ2BAPVA5CNFSM4JXKFZQKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEHQGSRI#issuecomment-568355141>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ACQHGSM7RMXISGG35ZFT2VDQ2BAPVANCNFSM4JXKFZQA>
.
--
Best !
Frank Dellaert
http://frank.dellaert.com
|
|
Great debugging! 👍 I think the line to blame is 118 in IterativeSolver.cpp Will open a PR for you to test, @ProfFan . |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Another spin off of #168 .
Building for i386, the unit test testSubgraphPreconditioner fails.
Here's a gdb stack trace:
Will have to re-run with debug info to investigate what's wrong...
The text was updated successfully, but these errors were encountered: