Anton's Research Ramblings

Inline Assembler

The only time that I've ever seriously used assembler code was for an assignment at university. That was 8086 assembly if I recall. It was quite fun writing the entire move/copy/add/ifjump sequence to replicate a for loop. Eventually you start to see your way through the logical process in the same way as you do with a more familiar higher-level language. It sits somewhere between digital electronics and full programming languages. It's a lot easier than designing electronics actually, because you have a limited set of instructions, and a limited set of resources (registers) to work with. Also you can't screw up soldering.

I had another little hobby bash recently using GCC directly as an x86 assembly compiler, and pulled apart generated assembly code, compiled in separate stages, and that kind of thing. It's not really as enlightening as some old salts would have you think.

Something that I never tried, but saw in a tweet today, was a reference that harks back to the mist-shrouded early days of video games programming: using assembly in-line with C. This was often done to optimise performance. I read that it's not so useful these days as the compilers are better at optimising code, and yours is probably going to be worse. Is that true? Do people still do this? Probably not - it's not going to help improve GPU performance anyway. Well I gave it a go anyway - it's quite easy. GCC has it's own style, based on the AT&T assembler. Apparently there is also Intel style which has operands around the other way, and ever so slightly different special characters. Here's my test programme:

So, you enclose your block of assembler with either asm() or __asm__(), which I guess implies that it's GCC-specific, and not part of the language. I'll see what the C book says about this tomorrow. The bit at the end of the enclosure has three lines beginning with a semi-colon. These are output C variables, input C variables, and a list of registers to "clobber". The two instructions are both movl (copy using 32-bit longs). The first copies the value in %1 (the input variable) into a register called ebx. The second instruction copies ebx into %0 (the output variable). I haven't quite figured out all of the special formatting thingies so I won't attempt to describe some of them here. That's it - compile as normal and it works!

Here I used cgdb (the GNU debugger + a text editor to set breakpoints in your files) to debug my in-line assembler.

You can use the GNU debugger (gdb) to debug your assembler if you compile with the -g flag. I prefer a programme called "cgdb" (pictured), which includes a text editor for interactively setting break-points on particular lines in your code. You press ESC to get into the text edtior and "i" to go back to the gdb console. In the text editor use cursor keys to scroll text, and press the space bar to toggle a break-point. If I remember correctly "o" opens a different file. With a break-point where my in-line assembler starts, you invoke "run" in the gdb console. Then you can type in "info registers" to get a list of the values in all the built-in registers. Pgup/pgdn to scroll. to step through assembler lines you can use "stepi" (also "si"). This may take a couple of steps before you see the registers flick over to the new values. Splendid! Well, that's easy enough to do, but is it actually useful? Hmmm...

Next - I had my first go at Arduino tinkering, have a couple of book reviews to make, and maybe some details about graphics stuff.