write to memory mapped registers

A subtle one-complement/two-complement misinterpretation?

Meindert

Try #define ADDR 0x80000000UL

Leo Havmøller.

no joy.

It is true that an 'int' is taken as 32-bit two's complement number but I don't see how this can lead to the described behavior. B.t.w. 0x80000002 is correctly interpreted.

You may need to do this in assembly, or with a scrap of assembly. It looks like your problem isn't in just the compiler, but is a mismatch between the compiler and the assembler.

Your compiler is emitting addresses as decimal, two's compliment numbers, so instead of getting 2147483648 (which is 0x80000000 in anyone's book) you're getting -2147483648 (which is only 0x80000000 in 32-bit two's complement when it's correctly interpreted).

I suspect that this bit of bad style in the compiler is being complimented by an assembler uses library code to read in numbers and somehow does the conversion from -2147483648 to 0x80000001 (if you're using saturated arithmetic 0x80000000 is an insidiously bad number; someone may have "fixed" a bug in library code).

Does g21k let you use assembler directives? Try sticking in your own code in there, either using 2147483648 in the assembly, or 0x80000000.

BTW: my preferred way to do memory-mapped registers is to declare them as extern in C (generally as structures of related registers, one for each peripheral), then define their locations either in the linker file (which is easy if you're using something derived from gcc) or in an assembly file (using a directive to absolutely locate stuff). Hopefully doing so would sidestep this nastiness.

But what looks strange to me is that the assembler is responsible for both the listing and the object file. How can it then produce a different result?

If you mean an in-line assembly instruction within the C code, yes. I can use the instruction 'asm()' to access to the register and the value here seems correct. Is just that I don't like that much to clutter my code with inline assembly functions and I didn't come out with a solution that please my 'sense of aesthetic' :-)

It's indeed an option I would also like to try. So far I have a header file where I defined all the memory-mapped registers. It shouldn't be any problem in moving everything to an assembler file and nail down those values. Thanks for the pointer.

(snip)

Are you sure that the listing is coming from the assembler? If the g21k is a gnu toolchain, and if you're making it cough up assembly with the '- s' directive, then it's coming from the compiler.

Even if the listing is coming from the assembler, it would be proper for it to just echo what goes into it. So if the problem is in how it interprets decimal numbers, you would see the symptoms you are seeing.