minimize register reads during firmware transfer

[mvduin]

• Don't bother updating the 'get' variable if the last known value already allows for starting the next block.
• Don't check NCP status unless a timeout happens.

These two small changes result in a 76% reduction of the number of register accesses per loop iteration (from 13 to 3.17) or a 70% reduction of the total number of cmd53 sdio transfers. This roughly halves the time needed to transfer the firmware, which itself is the dominant part of the wf200 initialization time.

Reduction in firmware download time measured on WGM160P:

• bare_metal example: 88 ms -> 48 ms
• micriumos example: 619 ms -> 207 ms

[jerome-pouiller]

Hello @mvduin,

For information, I have ported your idea to the linux version of the driver, I now load the firmware:

• in 64ms instead of 130ms using SDIO bus
• in 78ms instead of 115ms using SPI bus

[mvduin]

Yeah I noticed it had the same inefficiency, although I think it doesn't matter quite as much there since it's not going to be a significant contributor to your overall system performance, whereas for mcu applications it can be the single most time-consuming part of going from a low-power state to being able to send wifi packets.

[silabs-JulienT]

Hello @mvduin,

Thanks for the contribution, I am reviewing the commit impacts on other configurations (different hosts, SPI...) and will merge it if there is no regression.