e_load() with start or e_start() zeros memory?

[AndyC]

Hi Guys,

When you use e_load with start or e_start does it zero the epiphany memory?

If I write to the eram from the arm before calling these the epiphany code only sees 0. If I write to eram after calling these the epiphany sees the data.

Or is it yet another thing I'm confused about?

Thanks for any help

Andy

[sebraa]

e_load() only reads the SREC file into Epiphany memory. Check this file and you will see that it contains a memory dump of all sections referenced in your code (even if it is only zeros). It became a problem to me when a core would start running and writing results to shared memory, while the host was still loading cores (and happily overwrite anything in shared memory over and over again).

I now run e-objcopy -R .shared_dram --output-target srec --srec-forceS3 program.elf program.srec to leave out a specific section from the SREC file when creating it. This way, the memory will not be touched. Make sure this section does not contain important information (and remember that memory will not be zero-initialized either!).

[AndyC]

Hi Sebraa,

Many thanks for the info, very useful.

All the best

Andy

[notzed]

You have to look at crt0.s and the object file to see what is initialised at each start.

It does initialise BSS to zero, but the supplied linker script is kind of broken and properly set any bss for on-core data - so it's just treated as data and the loader will load 0 for it.

More important perhaps is that data sections are not re-initialised so they will retain their data. i.e. global initialisation; wont happen after the initial load.

There's no point in using srec unless you like making the build process more convoluted: the esdk supports elf as of months ago.

[sebraa]

[notzed]

easiest source of 'documentation':

https://github.com/adapteva/epiphany-li ... e-loader.c