dma: don't swap buffers #367
Conversation
* This uses a new closure-based method to the DMA HAL implementation which gives access to the inactive buffer directly. * It removes changing addresses, the third buffer for DBM, the inactive address poisoning, and allows the cancellation of the redundant repeat memory barriers and compiler fences. * This is now around 20 instructions per buffer down from about 100 cycles before. * Also introduces a new `SampleBuffer` type alias. * The required unpacking of the resources structure is a bit annoying but could probably abstraced away. TODO: * Test * Adapt `lockin`
* master: adapt to new heapless/serde-json-core after const-generics Bump serde-json-core from 0.3.0 to 0.4.0 build(deps): bump heapless from 0.6.1 to 0.7.1 setup: cleanup itcm: add some comments, make it safe build(deps): bump ndarray from 0.15.1 to 0.15.2 Updating dependencies Updating the embedded-nal Removing spurious settings updates deps: add rationales for git dependencies itcm: implement in rust and execute during setup() remove duplicate linker option gha: install gcc fmt dependencies: align with master bump cortex-m-rt to 0.6.13+git memory.x: remove comment about old cortex-m-rt enable itcm/dtcm explicitly load process into itcm
* master: pounder: clippy pounder: add comment on channel enum ad9959: refactor pad() pounder: enum for gpio ext pins pounder: fix attenuator indices (latch and shiftreg) pounder io extender: hack around some bug rf_power: fix measurement attenuators: use robust latching sequence deps: use mcp23017 release pounder: simplify attenuator spi interface
* One sample per batch is typical and sufficient. * DMA has more overhead than direct read for one sample.
This reverts commit 26b2613. First needs to reduce capture rate to batch interval. Otherwise it's jittery due to polling alignment.
Yes. |
|
bors merge |
|
Build succeeded: |
While we can still listen for these errors, we don't have to bind an ISR handler in the app. If we don't bind an ISR, they will use the default hardfault handler, which would clean up the apps a bit. In any case, it's a cosmetic difference imo. |
|
Ack. I'll leave them in dual-iir as the prime debugging app and remove them from lockin and fls. |
gives access to the inactive buffer directly.
address poisoning, and allows the cancellation of the redundant repeat
memory barriers and compiler fences.
before.
SampleBuffertype alias.but could probably abstraced away.
TODO:
lockin