summaryrefslogtreecommitdiff
path: root/lib
AgeCommit message (Collapse)Author
2013-05-07gras: serialization loop for sbufferJosh Blum
2013-05-06gras: increase default number of buffersJosh Blum
Think about benchmark implications later, this only helps parallelism. And also those blocks that inline buffers.
2013-05-06Merge branch 'theron6_work'Josh Blum
2013-05-06gras: leave assertions in, can help more than hurtJosh Blum
2013-05-06gras: dont need this is automaticJosh Blum
2013-05-05query: added bytes copied statJosh Blum
2013-05-05query: added actor depth statJosh Blum
2013-05-05Merge branch 'master' into theron6_workJosh Blum
2013-05-05gras: added post input buffer/message/tag api callsJosh Blum
This allows the user to post stuff to the input port from an external entity aka outside of the scheduler. resolves #86
2013-05-05gras: fixed wrong integer type for item indexJosh Blum
2013-05-05gras: name of block in error messages helpsJosh Blum
2013-05-04gras: rename to serialize_types.cppJosh Blum
2013-05-04gras: added serialization support for gras typesJosh Blum
sbuffer, streamtag, and packetmsg get registered into the polymorphic archive w/ unit tests resolves #65
2013-05-04gras: fix nasty little error possibility when not assertingJosh Blum
2013-05-04gras: implement force done input port optionJosh Blum
fixes #28
2013-05-02Merge branch 'master' into theron6_workJosh Blum
2013-05-02gras: inline the sbuffer deref codeJosh Blum
2013-04-30query: added allocator debug elementJosh Blum
2013-04-30benchmarks: removed keep_m_in_n block from many ratesJosh Blum
This block is optimized for laziness and not performance since it returns before finishing its input based on whatever M is. This means that m=1, n=1, this block produces 1000's of outputs per input buffer, and its not really what I am looking to benchmark... ironically, keep_m_in_n crappy implementation makes for big wins on GRAS either due to lower scheduler overhead or locking contentions easier. However, its crazy number of tiny outputs really rails on the mailbox queue of the next block in the chain and causes the caching allocators to get very exited. I guess thats all fine since the block is meant to be lazy... But I cant measure the effectiveness of a typical allocator situation. Signing off...
2013-04-29quick stashiesJosh Blum
2013-04-28Merge branch 'master' into theron6_workJosh Blum
2013-04-27gras: fixed buffer inlining with explicit support in buffer queueJosh Blum
Rather than push the inlined buffer into the BufferQueue, the output buffer queues wrapper explicitly supports holding onto the inlined buffer. This is because BufferQueue overloads generally do a token check so they only accept buffers that they are in charge of. fixes #79
2013-04-27gras: moved flush output to consume function in output queuesJosh Blum
This cleans up some code in block actor. The message sending code is now in task_main, and the consume routine is now the shared one.
2013-04-24gras: work on benchmarks w/ theron6 supportJosh Blum
2013-04-23gras: use framework send, prepare for theron v6Josh Blum
2013-04-16gras: enable half consumed metric on pool buffersJosh Blum
2013-04-16gras: isolate this callback to only the source fileJosh Blum
2013-04-15gras: added vec call to work bufferJosh Blum
2013-04-15gras: forgotten change from last commitJosh Blum
2013-04-15gras: assign a few more missing UNLIKELYsJosh Blum
2013-04-14gras: warning prints for block implementations that hangJosh Blum
2013-04-14Merge branch 'benchmark_work'Josh Blum
2013-04-14gras: move code into component filesJosh Blum
2013-04-14gras: increase chunk size for fair benchmarksJosh Blum
2013-04-14gras: yield policy env, and throw on bad valueJosh Blum
2013-04-14gras: added more GRAS_UNLIKELY in handlersJosh Blum
2013-04-13gras: minor tweaks while benchmarkingJosh Blum
2013-04-10gras: switched to using token for prio messagesJosh Blum
See #70 for more details
2013-04-10gras: formalize the weak container api w/ docsJosh Blum
address issue #69, we now have a set_container call, which is now formally part of the api. Anyone can call set_container for custom container types. GRAS supports shared ptr and pyobject out of the box.
2013-04-06gras: cleaner way for msgs to override item reserveJosh Blum
2013-04-06gras: split register property into two callsJosh Blum
2013-04-06gras: fixed a bug in swapping out buffer queuesJosh Blum
The default pool buffer was getting circ buffers when they went out from the deref in this circumstance: Basically, the circ buffs were created and deref'd before the circ queue was set into the output queue. This seems to fix the pfb clock sync issue, but not others is #42
2013-04-04gras: revisit container storage after yesterdays learningJosh Blum
The mechanisms of connect are now resonsible for grabbing the container ref. When the object is a shared ptr, the element contructor overload sets weakself. When the object is in python, the python reference is held, even if its shared ptr underneath that. * removed the need for shared_from_this * removed the ref stuff in python Block
2013-04-03gras: code cleanup and comments from last commitJosh Blum
2013-04-03gras: figured out how to do the python referenceJosh Blum
2013-04-01gras: fix from previous commit: swig element equalsJosh Blum
Just put an API call into element so its always there, and call it from swig.
2013-03-31Merge branch 'query_work'Josh Blum
2013-03-31gras: fix #60 by checking done logic after handlerJosh Blum
2013-03-30query: filter query_stats to only blocks requestedJosh Blum
2013-03-29gras: reuse inputs_available for done logicJosh Blum
generated by cgit (git 2.34.1) at 2025-03-13 16:24:37 +0000