This change should significantly improve the performance of single precision floating point math library functions. The vast majority of changes have to do with preventing the compiler from needlessly promoting floats to doubles, performing the calculation with doubles, only to demote the result to float. These changes only affect the math lib functions that return float.

VFS ioctl() again.  Per comments from David Sidrane, file_ioctl() should not return success if the ioctl method is not supported.  It probably should return ENOTTY in that case.

stm32l4_tickless; avoid Lyme, support timers, pwm, oneshot, freerun, and tickless scheduling

Feature/stm32f4discovery can

stm32_serial.c edited online with Bitbucket

Upstream_nucleo 144

stm32_serial.c edited online with Bitbucket

Stm32l4 USB support

update usb dev/host controller drivers to reflect new(ish) logging standards; augment device enpoint and fifo allocation #defines to do more sanity checking, and be automatically adaptive to size changes.  Update README.txt to reflect current status of the implementation.

Conflicts:
	configs/stm32l476vg-disco/nsh/defconfig

cdcacm_unbind leaks write request objects.  This arises due to freeing the

bulk IN endpoint before the loop that frees the requests via cdcasm_freereq.
That function checks the parameters and skips the freeing if either is NULL.
Freeing the bulk IN enpoint will cause the first param to be NULL, thereby
bypassing the free operation.  To fix, I moved the release of the bulk IN
endpoint until after to loop (much as was the case for the OUT and read
requests, which did not exhibit the problem).