SSE4.2you have to remember generally speaking both fma and AVX for games is only going to be an incremental improvement over SSE. In would be interesting to see what PC devs actually target today? SSE4? SSSE3?
Oh so that's the reason why all games of the last 2-3 years run so well on any CPU... (not)Game developers aren't going to want to lose sales to people who don't have AVX2 CPUs. This is one of the reasons why game engines aren't being made that fully take advantage of eight threads. They need the games to run well on an i3.
Asking devs to make everything for the Anniversary Pentium is a bit of a stretch. As for 2-3 years of i5s and i7s, I fail to see the issue. Some of the slowest i3s may have difficulty but those with high enough clocks should still be viable even if they're two or three years old.Oh so that's the reason why all games of the last 2-3 years run so well on any CPU... (not)
SSE4.2
The problem with AVX is that the majority of the best selling processors are not support it. For example every Celeron and Pentium. Even if it might be a useful way to improve the application performance the publishers don't finance the research. They want an alternative optimization strategy to improve the performance for the Celeron/Pentium users also.
The consoles might help for AVX, but AVX2 is still a not really useful option.
I personally think that HSA runtime is the best way to get AVX2 and even AVX512 support for the applications. That platform is cheap and we can target a lot of extensions/accelerators with the same codebase. I really think that SYCL 2.1 will be also a revolutionary step for the programers.
SSE4.2
The problem with AVX is that the majority of the best selling processors are not support it. For example every Celeron and Pentium. Even if it might be a useful way to improve the application performance the publishers don't finance the research. They want an alternative optimization strategy to improve the performance for the Celeron/Pentium users also.
The consoles might help for AVX, but AVX2 is still a not really useful option.
I personally think that HSA runtime is the best way to get AVX2 and even AVX512 support for the applications. That platform is cheap and we can target a lot of extensions/accelerators with the same codebase. I really think that SYCL 2.1 will be also a revolutionary step for the programers.
Simple question, but do any games use AVX2 and FMA3 yet? I know that some are using AVX, especially since the PS4 and Xbox One both support it.. And several physics engines also use AVX..
I did a google search and the only thing I could find was that Serious Sam 3 might use AVX2, and this was speculative more than anything.
Phenom is not a target now. It's too old.Even SSE4.2 is a bit risky. Phenom didn't have it, and there are a lot of those still out there.
Phenom is not a target now. It's too old.
SIMD is not implemented well in x86. It is forcing you to deal with multithreading, prefetching, and small registers. MMX, SSE and AVX is very inefficient compared to other SIMD models. Also every new SIMD register sizes will just make your original code outdated. The best way to do SIMD is to generate a specialized code for each CPU model, and this will be relatively fast, but most publishers just don't finance it. For today it is much more logical to use one or more IR, and from there you can compile the code to a lot of hardware targets. The real question is that which IRs and compiling paths gives the best overall results.I'm no programmer, but I was under the impression that extensions such as AVX2 were backward compatible with older extensions. For example, a new CPU like Haswell or Skylake would run the fastest codepath with AVX2, while a CPU like Sandy Bridge would use the same codepath but with less throughput/performance due to lacking AVX2..
I really have to wonder though at some of the massive performance gains on the CPU side seen in recent games, such as Dying Light for instance. They went from this at the game's launch:
To this 11 months later, a more than doubling of performance for many CPUs on that list..
And the game definitely was CPU limited when it first shipped no doubt, but now it performs very well. So I wonder, did they get these gains by exploiting more vectorization, or was it all due to better multithreading?
It seems more the latter, as CPUs with more threads/cores gained more performance.
That's why SSE2 is mainstream now, but SSE4.2 is much more logical choice for the actual projects.Phenom II is still listed as minimum for plenty of recent games. (Llano also lacked SSE4.2.) SSE4.1/2 isn't that essential, to be honest- the blend instructions are nice, but you can get the same effect with an (and)|(andnot) sequence.
I'm no programmer, but I was under the impression that extensions such as AVX2 were backward compatible with older extensions. For example, a new CPU like Haswell or Skylake would run the fastest codepath with AVX2, while a CPU like Sandy Bridge would use the same codepath but with less throughput/performance due to lacking AVX2..
That's not how it works. If a program is compiled to only use AVX2, it won't run on a processor without AVX2 support (eg on Linux it will seg fault). If it's compiled to only use AVX, then processors that support AVX2 will also be able to run it.
And then there's dispatching that can be done where the code detects CPU flags and runs different code paths depending on what's found.
I'm no programmer, but I was under the impression that extensions such as AVX2 were backward compatible with older extensions.
I'm no programmer, but I was under the impression that extensions such as AVX2 were backward compatible with older extensions. For example, a new CPU like Haswell or Skylake would run the fastest codepath with AVX2, while a CPU like Sandy Bridge would use the same codepath but with less throughput/performance due to lacking AVX2..
That still means an increase in validation effortsas you have to ensure all paths are correct. And not everyone uses icc, the vast majority of the industry relies on MS compiler.Its not an issue to have AVX2 support and still run it on CPUs without AVX2. This is essentially what all the "intel compiler cheats" is about in the old days.
Everything performance oriented uses ICC.