Max Prokhorov
Max Prokhorov
> For further motivation, here are the results when substituting Delegate back in for std::function: dok-net:all_in_functional_w_delegate@80MHz: IRQs/s = 144092 Notable improvement is removal of if(functional), arg casts, which is mighty...
Could not reproduce with 10.2 & 11.1 x86_64. Try on current xtensa-toolchain gcc locally and run / `disassemble test` in gdb? And it *seems* only string literals are affected. Can't...
[Small update based on the gitter discussion and a compressed example](https://gist.github.com/mcspr/00971c5e1c5c5b6861db37db0f74de41). Using and not using `-fno-constants-merge` as a build flag has different results for the test case. For the function...
A step before linking, assembly has *slight* differences. Without the flag, strings have `.rodata.$origin$.str#.#` section ([name originates here](https://github.com/gcc-mirror/gcc/blob/f5e2f2d0ad1b293c534338a72094926313e12039/gcc/varasm.c#L860-L918)), while with the flag they are assigned a generic `.rodata` ```diff diff...
Another test, but now with some custom section that does not start with `.rodata`, *but* is set through code to be 'mergeable' (notice that both strings must be in the...
[(Unsurprisingly?), can also reproduce on the current arduino-esp32 2.0.1, just a matter of making strings appear in a certain order](https://gist.github.com/mcspr/8ee6e6d6f8565d1bb74f1abbdbdee05f). But, can't reproduce just with the xtensa toolchain or godbolt...
As I understood the original ArduinoOTA ide, we are dealing with local updates. And we already have HTTP Update server; no need to transfer IP, it is a direct connection...
I had this q bookmarked, but now I also noticed I've been looking at some of my own modifications... Right now, only umm_info() getHeapStats() implementations is accepting ctx pointer, and...
so, in general, since everything essentially happens through the context - have umm_info() to accept pointer as well, so there's no need to get 'current' context - public umm_*_core, so...
> This would become a problem with free-ing an allocation that was made from ISR and was forced to use DRAM. The safest approach for free is to stay in...