DirectX 12 support on pre SSE4.2 CPUs

Might as well ask, "Why do AMD& nVidia force me to buy a dx12-supporting GPU just to run d3d12 in hardware?"  For that matter, I've never been able to understand why people would spend hundreds of dollars to upgrade a GPU but get stingy when it comes to their cpu, core logic, and system ram, etc.?  SSE4.2 has been a *standard* instruction for *several years* now, so why shouldn't AMD support it?  Even games like the recently released No Man's Sky were released requiring SSE 4.2--the game had to be recompiled and patched just so that it could support SSE2 as well as SSE 4.2.  Oh, and nVidia doesn't actually support the same features in D3d12 that AMD supports, either (although nVidia claims to support *more* D3d12 features than AMD--the fact is that nVidia doesn't support things like Async Compute even though the company really wants to convince people that their GPUs do.) But perhaps you'd be better off with more backwards-looking hardware--like a nVidia GPU.

Requiring SSE 4.2 in a CPU is no different than requiring a d3d12-level GPU to support d3d12.  As cheap as SSE4.2-supporting CPUs are these days it is silly to object to a GPU driver requiring SSE4.2 in the cpu--you can buy an AMD FX-xxxx CPU for a *quarter* of the price of an R9 390x, for instance.   Why try and blame a GPU manufacturer for your reluctance to keep your CPU and system core-logic support updated *during the last ten years?*

tapek wrote:

 

I have check NVIDIA drivers, they use only SSE2 instructions in DX12 driver, so why you (AMD) enforces SSE4.2 ? I think driver should not enforce instruction set that OS is not enforcing, and Windows 10 x64 requires only SSE2 + CMPXCHG16B + LAHF/SAHF + PrefetchW.

16.5.1 works here on my C2D.. or hey, at least Forza doesn't crash before even creating the window.

I tested 16.5.2, 16.5.2.1, 16.6.1 and latest 16.7.3 and those are indeed broken.

Very nice catch with those Radeon Pro drivers then.

Oddly theirs amdxc64.dll has more or less the same pdb info than 16.5.2 one (and about the same size), but 16.5.1 build number.

Use "expand -r amdxc64.dl_" command

Did you merged the old dll functions or did you implemented a workaround yourself? Will I get performance loss?

Thank you very much for your efforts! Everything works now.

I do have one consideration though, I realize that we are lucky to have you around but you know what they say about teaching a man to fish... Could you maybe describe to us how to perform this task so that you are not burdened with requests to be releasing something like this all the time? It's up to you, of course, it is your work after all.

Thanks agian.

If Quantum Break has popcnt calls in it's own .exe or .dll's, my files will not help. And i won't be able to modify them, since it uses Denuvo.

every step:

1) clean uninstall amd radeon drivers

2) download official 16.9.1 drivers for win10 x64 (whql-win10-64bit-radeon-software-crimson-16.9.1-sep13.exe)

3) run and extract it(folder by default: C:\AMD\WHQL-Win10-64Bit-Radeon-Software-Crimson-16.9.1-Sep13\, close "amd installer" window, when it shows.

4) update the 16.9.1 install kit with the two custom files (overwrite the originals) using your path: "replacing the original ones used by installer in folder "C:\AMD\WHQL-Win10-64Bit-Radeon-Software-Crimson-16.9.1-Sep13\Packages\Drivers\Display\WT6A_INF\B306709"(path by default)"

5) run C:\AMD\WHQL-Win10-64Bit-Radeon-Software-Crimson-16.9.1-Sep13\Setup.exe and install everything as usual

6) if(when) there will be warning about non-original files, click "install anyway"(or something like that - i don't have english windows to check what exactly it warns about)

I agree. This matter has not been resolved to a satisfactory degree, nor have we had any input from AMD themselves.

It's bad enough we have people showing up just to misrepresent our arguments and bring us down in this thread now.

EDIT:Looks like the thread got cleaned. Thank you mods!

Who's this guy?

We're doing the best we can. I can't speak for everyone, but I would imagine most if not all of us are interested in learning as much as we can about the issue. We have to start somewhere, regardless of whether or not any of us are classically trained in such topics; we have to comment on them.

wow. this is great news! tx4sharing. Maybe now amd fixes also the amd hd3d issues.

Thank you, Ray!