| Age | Commit message (Collapse) | Author |
|---|
|
|
|
|
|
|
|
|
|
speed matters. (#64)
This commit adds the HBXX64 and HBXX128 classes for use as a drop-in
replacement for the slow LLMD5 hashing class, where speed matters and
backward compatibility (with standard hashing algorithms) and/or
cryptographic hashing qualities are not required.
It also replaces LLMD5 with HBXX* in a few existing hot (well, ok, just
"warm" for some) paths meeting the above requirements, while paving the way for
future use cases, such as in the DRTVWR-559 and sibling branches where the slow
LLMD5 is used (e.g. to hash materials and vertex buffer cache entries), and
could be use such a (way) faster algorithm with very significant benefits and
no negative impact.
Here is the comment I added in indra/llcommon/hbxx.h:
// HBXXH* classes are to be used where speed matters and cryptographic quality
// is not required (no "one-way" guarantee, though they are likely not worst in
// this respect than MD5 which got busted and is now considered too weak). The
// xxHash code they are built upon is vectorized and about 50 times faster than
// MD5. A 64 bits hash class is also provided for when 128 bits of entropy are
// not needed. The hashes collision rate is similar to MD5's.
// See https://github.com/Cyan4973/xxHash#readme for details.
|
|
|
|
|
|
speed matters. (#64)
This commit adds the HBXX64 and HBXX128 classes for use as a drop-in
replacement for the slow LLMD5 hashing class, where speed matters and
backward compatibility (with standard hashing algorithms) and/or
cryptographic hashing qualities are not required.
It also replaces LLMD5 with HBXX* in a few existing hot (well, ok, just
"warm" for some) paths meeting the above requirements, while paving the way for
future use cases, such as in the DRTVWR-559 and sibling branches where the slow
LLMD5 is used (e.g. to hash materials and vertex buffer cache entries), and
could be use such a (way) faster algorithm with very significant benefits and
no negative impact.
Here is the comment I added in indra/llcommon/hbxx.h:
// HBXXH* classes are to be used where speed matters and cryptographic quality
// is not required (no "one-way" guarantee, though they are likely not worst in
// this respect than MD5 which got busted and is now considered too weak). The
// xxHash code they are built upon is vectorized and about 50 times faster than
// MD5. A 64 bits hash class is also provided for when 128 bits of entropy are
// not needed. The hashes collision rate is similar to MD5's.
// See https://github.com/Cyan4973/xxHash#readme for details.
|
|
|
|
|
|
|
|
|
|
Fixes folders being invidible (missing arrange)
Fixes sroll to target not working reliably
|
|
D577 should have picked part of the changes from contribute branch, picking up the rest for the sake of branch specific crash fixes
|
|
|
|
# Conflicts:
# autobuild.xml
# indra/newview/llagent.cpp
# indra/newview/llimview.cpp
# indra/newview/llimview.h
# indra/newview/llinventoryfunctions.cpp
# indra/newview/llpanelmediasettingsgeneral.cpp
# indra/newview/pipeline.cpp
|
|
branches (#47)
Revert part of "DRTVWR-575: Address review comments on Xcode 14.1 type tweaks."
Crash was reproduced when assigning areastr to llsd, but likely present in other cases of assigning ui strings to llsd (instead of going for lluistring's result directly copy constructor was engaged and either copy or original crashed due to invalid pointers, copy shouldn't have been created).
|
|
|
|
|
|
One could argue that passing a negative index to an LLSD array should do
something other than shrug and reference element [0], but as that's legacy
behavior, it seems all too likely that the viewer sometimes relies on it.
This specific problem arises if the index passed to operator[]() is negative
-- either with the previous Integer parameter or with size_t (which of course
reinterprets the negative index as hugely positive). The non-const
ImplArray::ref() overload checks parameter 'i' and, if it appears negative,
sets internal 'index' to 0.
But in the next stanza, if (index >= existing size()), it calls resize() to
scale the internal array up to one more than the requested index. The trouble
is that it passed resize(i + 1), not the adjusted resize(index + 1).
With a requested index of exactly -1, that would pass resize(0), which would
result in the ensuing array[0] reference being invalid.
With a requested index less than -1, that would pass resize(hugely positive)
-- since, whether operator[]() accepts signed LLSD::Integer or size_t,
resize() accepts std::vector::size_type. Given that the footprint of an LLSD
array element is at least a pointer, the number of bytes required for
resize(hugely positive) is likely to exceed available heap storage.
Passing the adjusted resize(index + 1) should defend against that case.
|
|
The compiler was deducing an unsigned type for the difference (U64 desired
microseconds - half KERNEL_SLEEP_INTERVAL_US). When the desired sleep was less
than that constant, the difference went hugely positive, resulting in a very
long snooze.
Amusingly, forcing that U64 result into an S32 num_sleep_intervals worked only
*because* of integer truncation: the high-order bits were discarded, resulting
in a negative result as intended.
Ensuring that both integer operands are signed at the outset, though, produces
a more formally correct result.
|
|
|
|
Apparently Visual Studio and Xcode disagree on the intended lifespan of a
certain temporary expression. Capturing it in a named variable works.
|
|
# Conflicts:
# doc/contributions.txt
# indra/newview/llappviewer.cpp
# indra/newview/skins/default/colors.xml
|
|
|
|
|
|
# Conflicts:
# doc/contributions.txt
# indra/newview/app_settings/shaders/class1/deferred/materialF.glsl
# indra/newview/llfloater360capture.cpp
|
|
|
|
|
|
For work queues that don't need timestamped tasks, eliminate the overhead of a
priority queue ordered by timestamp. Timestamped task support moves to
WorkSchedule. WorkQueue is a simpler queue that just waits for work.
Both WorkQueue and WorkSchedule can be accessed via new WorkQueueBase API. Of
course the WorkQueueBase API doesn't deal with timestamps, but a WorkSchedule
can be accessed directly to post timestamped tasks and then handled normally
(e.g. by ThreadPool) to run them.
Most ThreadPool functionality migrates to new ThreadPoolBase class, with
template subclass ThreadPoolUsing<WorkQueue> or ThreadPoolUsing<WorkSchedule>
depending on need. ThreadPool is now an alias for ThreadPoolUsing<WorkQueue>.
Importantly, ThreadPoolUsing::getQueue() delivers a reference to the specific
queue subclass type, so you can post timestamped tasks on a queue retrieved
from ThreadPoolUsing<WorkSchedule>::getQueue().
Since ThreadPool is no longer a simple class but an alias for a particular
template specialization, introduce threadpool_fwd.h to forward-declare it.
Recast workqueue_test.cpp to exercise WorkSchedule, since some of the tests
are time-based. A future todo would be to exercise each applicable test with
both WorkQueue and WorkSchedule.
|
|
Deriving your tracked class T from LLInstanceTracker<T> gives you
T::getInstance() et al. But what about a subclass S derived from T?
S::getInstance() still delivers a pointer to T, requiring explicit downcast.
And so on for other LLInstanceTracker methods.
Instead, derive S from LLInstanceTrackerSubclass<S, T>. This implies that S is
a grandchild class of T, but it also recasts the LLInstanceTracker methods to
deliver results for S rather than for T.
|
|
|
|
Co-authored-by: Nat Goodspeed <nat@lindenlab.com>
|
|
Looks like pollTick tried to call an already dead process
|
|
|
|
|
|
The unsigned index arithmetic was problematic in that case.
|
|
|
|
Since LLSDSerialize::SIZE_UNLIMITED is negative, passing that through unsigned
size_t parameters could result in peculiar behavior.
|
|
and use it to replace dubious loops in asLLSD() and trimEmpty().
|
|
|
|
When sending multiple LEAP packets in the same file (for testing convenience),
use a length prefix instead of delimiting with '\n'. Now that we allow a
serialization format that includes an LLSD format header (e.g.
"<?llsd/binary?>"), '\n' is part of the packet content. But in fact, testing
binary LLSD means we can't pick any delimiter guaranteed not to appear in the
packet content.
Using a length prefix also lets us pass a specific max_bytes to the subject
C++ LLSD parser.
Make llleap_test.cpp use new freestanding Python llsd package when available.
Update Python-side LEAP protocol code to work directly with encoded bytes
stream, avoiding bytes<->str encoding and decoding, which breaks binary LLSD.
Make LLSDSerialize::deserialize() recognize LLSD format header case-
insensitively. Python emits and checks for "llsd/binary", while LLSDSerialize
emits and checks for "LLSD/Binary". Once any of the headers is recognized,
pass corrected max_bytes to the specific parser.
Make deserialize() more careful about the no-header case: preserve '\n' in
content. Introduce debugging code (disabled) because it's a little tricky to
recreate.
Revert LLLeap child process stdout parser from LLSDSerialize::deserialize() to
the specific LLSDNotationParser(), as at present: the generic parser fails one
of LLLeap's integration tests for reasons that remain mysterious.
|
|
reduce CPU usage of background threads.
|
|
Since parsing binary LLSD is faster than parsing notation LLSD, send data from
the viewer to the LEAP plugin child process's stdin in binary instead of
notation.
Similarly, instead of parsing the child process's stdout using specifically a
notation parser, use the generic LLSDSerialize::deserialize() LLSD parser.
Add more LLSDSerialize Python compatibility tests.
|
|
Absent a header from LLSDSerialize::serialize(), make deserialize()
distinguish between XML or notation by recognizing an initial '<'.
|
|
LLSDSerialize::serialize() emits a header string, e.g. "<? llsd/notation ?>"
for notation format. Until now, LLSDSerialize::deserialize() has required that
header to properly decode the input stream.
But none of LLSDBinaryFormatter, LLSDXMLFormatter or LLSDNotationFormatter
emit that header themselves. Nor do any of the Python llsd.format_binary(),
format_xml() or format_notation() functions. Until now, you could not use
LLSD::deserialize() to parse an arbitrary-format LLSD stream serialized by
anything but LLSDSerialize::serialize().
Change LLSDSerialize::deserialize() so that if no header is recognized,
instead of failing, it attempts to parse as notation. Add tests to exercise
this case.
The tricky part about this processing is that deserialize() necessarily reads
some number of bytes from the input stream first, to try to recognize the
header. If it fails to do so, it must prepend the bytes it has already read to
the rest of the input stream since they're probably the beginning of the
serialized data.
To support this use case, introduce cat_streambuf, a std::streambuf subclass
that (virtually) concatenates other std::streambuf instances. When read by a
std::istream, the sequence of underlying std::streambufs appears to the
consumer as a single continuous stream.
|
|
|
|
abilities and remove some more fast timers.
|
|
|
|
|
