[ECOS] Eabi error or Compiler Bug?

[Privat]

Hi Max, Hi Edwards,  hi everyone

i ve bumped into some strange behaviour again lately during some code
cleanup.
I can work around it, but i wondered if other before me ran into this
and wether this is UB, a compiler bug or something ecos specific.

My Compiler:
Im compiling under ubuntu 14.04 for the MPC5668x powerpc architecture
with ppc-eabi-gcc 4.8.2 toolchain (if i found the correct compiler there) .

The problem occurs when casting pointers to a smaller size;

Example:
cyg_uint32 testValue = 0x12345678;
cyg_uint8  *ptrToLowestByte;
cyg_uint32 *ptrToWholeValue;

ptrToLowestByte = (cyg_uint8 *) (&testValue);
ptrToWholeValue = (cyg_uint32 *) ptrToLowestByte;

I can work with the ptrToTheWholeValue as long i do not use any binary
operations with a diffrent size
For example

cyg_uint32 container32 = ((*ptrToWholeValue) & 0xFF00 0000)
will return valid Value into the corresponding 32uint container;

cyg_uint8 container8 = (cyg_uint8) ((ptrToWholeValue & 0xFF00 0000 ) >>
24);

However will not.
My assumption is that the endianess of the dereferenced Value is wrong,
a cast pointer does no longer point to the lowest byte, but to the
highest instead.
I worked around it by not using bitops that shift into a diffrently
sized variable from a pointer.

If requested i could upload some dissasembly.

I also experienced some problems using memcpy and strncpy, when using
casted down uint8* ptr to uint8 arrays
- where the values to copy where copyied in at a offset off the original
ptrs data type (uint32*).

This all may be perfectly normal behaviour and i m just lacking
experience- in that case i would be very glad for any explenation.
Anybody experienced something similar?
There seem to be some todos regarding assembly binary operations in the
vle.h still open.
Should i report this to the gnu or will i get trampled by a stampede?

Regards Pica






-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss