1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
|
High Priority (we would need it for IPE Camera)
=============
1. DMA without ring buffer, just send data in the newly configured buffers while there is the space...
* We need a way to add (remove?) pages into the same use. Some times in the middle of existing pages, i.e. tructuring seems complicated.
* vmsplice them out trough the device?
2. Allow DMA into the user pages
* Dangerous as hardware accessed memory is only valid during life-time of one application. Instead the kernel pages can be re-assembled as big buffer using vmmngr_map_page (?)
3. Use huge-pages + dynamicaly configure DMA page size
4. DirectGMA/GPUDirect support
5. High-speed interface to Infiniband
Normal Priority (it would make just few things a bit easier)
===============
1. Integrate base streaming model into the pcitool
2. Implement pcilib_configure_autotrigger
3. Really check the specified min, max values while setting registers
4. Provide OR and AND operations on registers in cli
5. Support writting a data from a binary file in cli
Low Priority (only as generalization for other projects)
============
1. We managed kmem performance using next kmem prediction, but it is still wise to provide additionally a binary tree for faster search
2. Support for Network Registers and Network DMA
3. Define a syntax for register dependencies / delays (?)
4. Use pthread_condition_t instead of polling
5. Support FIFO reads/writes from/to registers
6. OPC UA interface to the registers
7. Generate XML models from SystemRDL descriptions
Performance
===========
1. Even with fully algined data, glibc defaults to __memcpy_sse2_unaligned (called from ipecamera_data_callback and fastwriter_push). Can we do something?
|