Re: [boost] [coroutine] new location

Subject: Re: [boost] [coroutine] new location
From: Christopher Kormanyos (e_float_at_[hidden])
Date: 2012-09-05 17:52:10

>> 1) How can I set the desired stack size to a small amount

>> such as 256 bytes?

> with a stack of 256 bytes I assume you are not interested
> in a stack with a guard page :) you could simply create your
> own stack allocator - it needs only satisfy following requirements:

> - member functions void * allocate( std::size_t) and void deallocate( void *, std::size_t)
> - most important the pointer returned by allocate() must be the
> start address of the stack and because ARM uses descending stacks
> (I know the instruction set supports ascending instructions too) it must
> be the highest address
> - simply calloc()/free() should do it

OK. I have a variety of custom allocators in my collections that will work
for these simple first prototyping requirements. I have, for example,
a custom allocator based on a chunk of re-usable static RAM
allocated as a static array that I can size to anything (such as 256 byte)
up front.

>> 2) I do not want cross-build boost right now.
>> Which asm/cpp files should I extract from context
>> for ARM(R) to include as sources in my project?
>> Please don't worry about me or my sanity here,
>> I roll with cross-compilers and builds all the time!

> - files fcontext.cpp and fcontext_arm_aapcs_elf_gas.S (headers?)

I will split up the build with the asm/cpp files extracted to my project
workspace, whereas the Boost.Context headers will simply remain
in boost 1.51. It is unconventional, but it should work for prototyping.

>> 3) How can I disable saving/restoring of the
>> floating-point context in Boost.Context?
>> My target micro does not have an FPU?
> 

> If the Cortex-Mx has no FPU so that you do not use the compiler flag --hard-float (or so) to compile your code.
> The asm preserving the FPU registers is wrapped by:

> #if (defined(__VFP_FP__) && !defined(__SOFTFP__))
> In your case the FPU preserving instructions will not be compiled.
> For sanity you could explictly disable FPU preserving
> via the 4th argument of jump_fcontext().

> intptr_t jump_fcontext( fcontext_t * ofc, fcontext_t const* nfc, intptr_t vp, bool preserve_fpu = true);

> regards, Oliver

OK, excellent! That all makes sense.

Microcontrollers are tricky and sometimes you need to use
an in-circuit emulator (ICE) when assembly programming is involved.
I will give it my best try in the short time I have and see if it works.
It sounds more like a multi-day endeavor. But I only have a day or two.

I will report back with the results.

Thank you. Sincerely, Chris.