ASM vs C

[losbellos]

thanks,

not directly connected just a question:
What do you think Assembly is still ahead of C in terms of speed?
Like for example wiriting a ray tracer in assembly or a sampler would be beneficial?

[SATtva]

This is an off-topic here, so please restrain from further discussion, but yes, this is beneficial if you're able to write and debug a complex SIMD-optimized multithreaded program in a plain asm.

[losbellos]

Hej Sattva thanks, but I asked him not you.. Hope you dont mind.

[SATtva]

I do mind off-topic discussions. I've split the topic, so it's okay now.

[tomb]

I agee with SATTva, lets try to keep the forum threads nice and tidy

On topic: I did write a simple raytracer in Motorola 68000 once upon a time (on the Amiga in the late 80s)- it had only sphere primitives, but it was several orders of magnitude faster than the Amiga Basic version. These days, writing efficient assembly code is no walk in the park - what with RISC microarchitectures, out of order execution etc. Pure legacy x86 is pretty straight forward though, though not portable obviously. I think it's better to focus on making the algorithms friendly towards modern microarchitectures and optimizing compilers.

[losbellos]

well, on amiga everybody used assembly even kids, since the basic is very very slow.
But assembly and c compared.... There is not that much difference I think.
If let say the code would run 50% faster then it migght be that it would be interesting to do some more important parts in assembly.

[jeanphi]

Hi,

Modern processors are extremely complex beasts with out of order scheduling, instruction/data caches management on die, huge pipelines, ... which makes writing optimized assembly by hand a daunting task. If you add to the mix that the algorithms we're dealing with are extremely complex to expres even in high level languages, then it's often better to write in a high level language and when the code is mature enough, you can try to hand optimize some critical spots.
However knowing ASM and the basics of processors and compilers helps a lot writing efficient code.

Jeanphi

[Lord Crc]

I wrote quite a lot of asm code for the 486 and pentium. Some years ago I wanted to optimize a graphics routine and wrote some 300 lines of careful assembly code by all the rules I knew, even having read the intel optimization guide to get the latest tricks.

The result was 10% slower than what the Delphi compiler produced...

So yeah, as jeanphi said, it's very difficult to write significantly better asm code than what the modern compilers can produce. Especially if you have learned your compiler and know how to write "optimizable" code, which I had with the Delphi compiler.

[dougal2]

[moved to off-topic forum too]