Especially these days. Current-gen x86 architecture has all kinds of insane optimizations and special instruction sets that the Pentium I never had (e.g. SSE). You really do need a higher-level compiler at your back to make the most of it these days. And even then, there are cases where you have to resort to inline ASM or processor-specific intrinsics to optimize to the level that Roller Coaster Tycoon is/was. (original system specs)
I might be wrong, but doesn’t SSE require you to explicitly use it in C/C++? Laying out your data as arrays and specifically calling the SIMD operations on them?
I think they meant the other way around, that if you wanted to use it in C/C++, you’d have to either use assembly or some specific SSE construct otherwise the compiler wouldn’t bother.
That probably was the case at one point, but I’d be surprised if it’s still the case. Though maybe that’s part of the reason why the Intel compiler can generate faster code. But I suspect it’s more of a case of better optimization by people who have a better understanding of how it works under the hood, and maybe better utilization of newer instruction set extensions.
SSE has been around for a long time and is present in most (all?) x86 chips these days and I’d be very surprised if gcc and other popular compilers don’t use it effectively today. Some of the other extensions might be different though.
If you want to use instructions from an extension (for example SIMD), you either: provide 2 versions of the function, or just won’t run in some CPUs. It would be weird for someone that doesn’t know about that to compile it for x86 and then have it not run on another x86 machine. I don’t think compilers use those instructions if you don’t tell them too.
Anyway, the SIMD the compilers will do is nowhere near the amount that it’s possible. If you manually use SIMD intrinsics/inline SIMD assembly, chances are that it will be faster than what the compiler would do. Especially because you are reducing the % of CPUs your program can run on.
Oh I see your point. Yeah, I think they meant that. And yes, there was a time you’d have to do trickery in C to force the use of SSE or whatever extensions you wanted to use.
Technically assembly is a human-readable, paper-thin abstraction of the machine code. It really only implements one additional feature over raw machine code and that’s labels, which prevents you from having to rewrite jump and goto instructions EVERY TIME you refactor upstream code to have a different number of instructions.
So not strictly the bunch of bits. But very close to it.
Honestly, I’m not 100% sure. I would bet that a modern compiler would just “do the right thing” but I’ve never written code in such a high performance fashion before.
Especially these days. Current-gen x86 architecture has all kinds of insane optimizations and special instruction sets that the Pentium I never had (e.g. SSE). You really do need a higher-level compiler at your back to make the most of it these days. And even then, there are cases where you have to resort to inline ASM or processor-specific intrinsics to optimize to the level that Roller Coaster Tycoon is/was. (original system specs)
I might be wrong, but doesn’t SSE require you to explicitly use it in C/C++? Laying out your data as arrays and specifically calling the SIMD operations on them?
There’s absolutely nothing you can do in C that you can’t also do in assembly. Because assembly is just the bunch of bits that the compiler generates.
That said, you’d have to be insane to write a game featuring SIMD instructions these days in assembly.
I think they meant the other way around, that if you wanted to use it in C/C++, you’d have to either use assembly or some specific SSE construct otherwise the compiler wouldn’t bother.
That probably was the case at one point, but I’d be surprised if it’s still the case. Though maybe that’s part of the reason why the Intel compiler can generate faster code. But I suspect it’s more of a case of better optimization by people who have a better understanding of how it works under the hood, and maybe better utilization of newer instruction set extensions.
SSE has been around for a long time and is present in most (all?) x86 chips these days and I’d be very surprised if gcc and other popular compilers don’t use it effectively today. Some of the other extensions might be different though.
If you want to use instructions from an extension (for example SIMD), you either: provide 2 versions of the function, or just won’t run in some CPUs. It would be weird for someone that doesn’t know about that to compile it for x86 and then have it not run on another x86 machine. I don’t think compilers use those instructions if you don’t tell them too.
Anyway, the SIMD the compilers will do is nowhere near the amount that it’s possible. If you manually use SIMD intrinsics/inline SIMD assembly, chances are that it will be faster than what the compiler would do. Especially because you are reducing the % of CPUs your program can run on.
Oh I see your point. Yeah, I think they meant that. And yes, there was a time you’d have to do trickery in C to force the use of SSE or whatever extensions you wanted to use.
Technically assembly is a human-readable, paper-thin abstraction of the machine code. It really only implements one additional feature over raw machine code and that’s labels, which prevents you from having to rewrite jump and goto instructions EVERY TIME you refactor upstream code to have a different number of instructions.
So not strictly the bunch of bits. But very close to it.
Technically correct. The best kind of correct.
Honestly, I’m not 100% sure. I would bet that a modern compiler would just “do the right thing” but I’ve never written code in such a high performance fashion before.