cpu/sam3: reworked timer driver

[haukepetersen]

While searching for the bug that was failing the periph_timer test, I decided to go for a hard cleanup. I decided to make only 3 channels per timer available for the following reasons:

• not using 3 channels saves energy
• this simplifies the code quite a bit
• removes some error-prone code (e.g. the disabling of interrupts in two steps)
• the high-level timer system in RIOT (xtimer) needs only a single channel anyway

Further I reduced the amount of defined timers for the udoo and arudino-due to 2, as normally 1 is used and 2 are idle, so now only 1 being unused should be enough for use cases where one wants to use an additional timer...

So let the code size speak:

tests/periph_timer, master:
   text    data     bss     dec     hex 
   8908     132    2700   11740    2ddc

this branch:
   text    data     bss     dec     hex 
   8320     132    2708   11160    2b98

-> saves 588 byte and ~300 lines of code...

[kYc0o]

I like the code, periph/timer test passed, all periph/xtimer_* tests passed, so just wait for travis and go?

[haukepetersen]

yes, once you give your explicit ACK and Travis is green this can be merged :-)

[haukepetersen]

Found a last-minute unclean line in the driver, the timer_irq_enable/disable implementations were broken: once you would define the used timers in another order, you would not enable/disable the correct interrupts. Fixed this -> @kYc0o: would you look at this and re-ack? Thx

[kYc0o]

wow! yes, you're right, I really missed that. Yes, so travis should restart?

[haukepetersen]

it already went over the latest commit and is happy. Do you ack?

[kYc0o]

yes! ACK and merge ;)

[dondonprad]

may i ask for Makefile this application?