• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD Claims Arm ISA Doesn't Offer Efficiency Advantage Over x86

DukenukemX said:
ARM and X86 are nearly identical but ARM is a complete mess.
Click to expand...
messy by how similar to x86 it becamse (and x86 over time also completletly change once it started to translated evreything in uCode), a bit like how messy avx512 went and was removed with Intel... same on arm with their equivalent stuff.

x86 become over time pseudo RISC with the pentium, they do not execute variable lenght instruction on the hardware, they decode everything into ARM Like single uop and ARM has so many extension that it is not some limited set of simple core one.

ARM still easier to decode on the chip too.

By the end of the video, he ended up with about twice the efficacy of a steam deck and achieved to be gpu limited a lot of the time and took him only a few months too close the gap with x86 having many years dedicated to it.

DukenukemX said:
Also LLVM for ARM is really bad. So bad that he had to do it in assembly to fix some bugs.
Click to expand...
I doubt that very much, Apple is incredibly involved with LLVM (LLVM creator went to apple and they did finance a lot of the work) and they are really good with arm cpu, same with Google.

Having some issues come up when you do something like this:
The SPU recompiler is typically the heaviest part of emulating the PlayStation 3 on x86 machines. Despite the x86 back and having been relentlessly optimized, I dove straight into the assembly that LLVM produced. And this is what I saw.
Just as a recap, RPCS3 uses the LVM project to translate PlayStation 3 game code to x86 or ARM. We take PlayStation 3 assembly, translate that to LLVM, then LVM takes that and outputs native code for your machine. It's not dissimilar to how games like Unleash Reompiled work, only they're translating to C rather than LLVM IR. Unleash Recompiled even requires use of the LLVMbased Clang compiler to recompile the C code it produces.

That quite unusual and a high bar to use, I am sure there was a long list of issue and manual work involved and years going with the x86 version.

Sometime it did better for ARM, sometime it did better for x86, if you compare GCC with LLVM/clang result on arm you will not see some massive gap between the 2. Or more bugs/performance if you target x86 vs arm with LLVM/clang, specially not for Apple ARM cpus, the qualcomm used here with 4 different cores type was a bit of an hard case.
 
Last edited:
LukeTbk said:
By the end of the video, he ended up with about twice the efficacy of a steam deck and achieved to be gpu limited a lot of the time and took him only a few months too close the gap with x86 having many years dedicated to it.
Click to expand...
Keep in mind that most of the x86 optimizations he has already implemented were just adopted over to ARM, so of course it only took him a few months. The AYN Odin 2 is also using a more advanced 4nm manufacturing process for the Qualcomm chip. Also, that handheld is no longer $300.
I doubt that very much, Apple is incredibly involved with LLVM (LLVM creator went to apple and they did finance a lot of the work) and they are really good with arm cpu, same with Google.
Click to expand...
He mentions the problems with LLVM often. This part here is what I was thinking. I can't find the assembly solution part as the video is an hour long.

View: https://youtu.be/-aI_XEwmKFk?t=1636
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
DukenukemX said:
He mentions the problems with LLVM often. This part here is what I was thinking. I can't find the assembly solution part as the video is an hour long.
Click to expand...
Yes there is some problem with LLVM, but what he is doing is quite a special use case (just in time versus static compilation...), we do not know how many issues he had doing it on a x86 target, I am sure he had to do manual work has well.

3 of the biggest contributor to LLVM are Apple, Google and ARM, giant proportion of the world is running on compiled for ARM by LLVM (amazon graviton stuff, google arm cpu, all the mobile world) it is really not some second citizen/second thought.

MIcrosoft/Intel/Google do work hard on the x86 side of things and it is also really good has well.
 
LukeTbk said:
Yes there is some problem with LLVM, but what he is doing is quite a special use case (just in time versus static compilation...), we do not know how many issues he had doing it on a x86 target, I am sure he had to do manual work has well.

3 of the biggest contributor to LLVM are Apple, Google and ARM, giant proportion of the world is running on compiled for ARM by LLVM (amazon graviton stuff, google arm cpu, all the mobile world) it is really not some second citizen/second thought.

MIcrosoft/Intel/Google do work hard on the x86 side of things and it is also really good has well.
Click to expand...
I'm just pointing out how LLVM isn't entirely ready for ARM. I'm sure LLVM had problems on x86 as well. It's not the first time an emulator author has found bugs and reported them. Delroth had to work with MESA to get some bugs fixed for Dolphin.
 
You must log in or register to reply here.
Back
Top