pichenettes / stmlib Goto Github PK

At this line:
https://github.com/pichenettes/stmlib/blob/master/dsp/dsp.h#L113
sqrtf is used math.h is not included; my compiler complains.

I have a question about the ring buffer implementation in stmlib.

As I understand it is expected to be thread-safe with a single producer and a single consumer (for example for comunicating between an ISR and application code). However from how I understand it there is no guarantee that a compiler will not reorder the access to the buffer content and the the access to the pointer indices. In particular, this could lead to a situation where an element is expected to be read (pointer indices have been changed) but it hasn't actually been written to the buffer so the receiver would read an old/invalid element if a context switch happened right in that moment.

The reason for that is that (how I understand it) volatile does not imply a memory fence and the safe way to handle it would be through std::atomic.

Please note that I am not saying this implementation does not work on arm. The compiler might very well emit working code. I am just wondering to what extend this is guaranteed.

Thanks for taking the time!

compiling with gcc-arm-none-eabi-9-2019-q4-major fails with the following error:

./stmlib/system/page_storage.h: In member function 'bool plaits::Settings::Init()':
./stmlib/system/page_storage.h:185:24: error: '*((void*)& h +4)' may be used uninitialized in this function [-Werror=maybe-uninitialized]
  185 |       FLASH_ProgramWord(address, *words);
      |       ~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~
./stmlib/system/page_storage.h:171:17: note: '*((void*)& h +4)' was declared here
  171 |     ChunkHeader h;
      |                 ^

The compiler gets confused because ChunkHeader doesn't default intiialize its varibles and fails to see the assignments next to it.
The fix is to use the initializer_list constructor instead of manually assigning the variables