123456789101112131415161718192021222324252627282930313233 |
- 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?
-
|