Instruction Set Translation

[K6]

My starting point is a software that MUST have a SSE enable CPU and many Athlon Classic CPU (no SSE but fast enought for running the software)

back in the old days there was a MMX emulator for running MMX software on standard CPU,
today I cant find any SSE emulator .

on the souceforge there are some CPU emulator but are complete emulator (like VMware) , so I think that is legal to emulate the SSE instruction set .

I dont know esactly how much is the difference between 3DNow! and SSE but are both floating point SIMD , I think a software can fill the gap.

But this software dont exist ..... WHY?

 

[K6]

Update ... I also found a X86-64 emulator , looks like the only extension without an emulator is just the SSE.

there is some thechical reason about it?????

 

[User1001]

look at this

Welcome to the Highly Technical Forum! This is a place to discuss the latest issues and technology in the fields of computer architecture and engineering, computer science, electrical engineering, and other sciences. It is meant as a forum where people who are serious about technology can share ideas and thoughts without concern for product and brand loyalties. We welcome a wide range of technical topics, from "How does a CPU work?" to "Which cache mapping policy offers the best trade-off in terms of hit ratio and access time?"

Questions regarding troubleshooting, technical support, and product advice belong in other appropriate forums such as General Hardware, Networking, Video, etc. They have more traffic and, they are visited by knowledgable, enthusiastic members eager to share good advice and help troubleshoot problems.

 

[K6]

Athlon machine are already upgraded , my question is just a thec question about Instruction Set.

Why there are emulator for every extension (3Dnow!,MMX,X86-64) exept for SSE?

 

[uart]

Originally posted by: User1001
look at this

Questions regarding troubleshooting, technical support, and product advice belong in other appropriate forums such as General Hardware, Networking, Video, etc. They have more traffic and, they are visited by knowledgable, enthusiastic members eager to share good advice and help troubleshoot problems.

I dont really see a problem with k6 posting this question here.


BTW k6, what particular software is it that wont run without SSE yet also does not require a lot of CPU horsepower?

Normally the use of SSE would indicate that the software is very cpu instensive, otherwise you wouldn't go to the trouble of SSE coding.


3dnow and SSE are indeed very similar, both in terms of their speed and efficiency and also in terms of what they actually achieve. But the implementation is quite different and I dont think 1-1 translation would be possible.

Run time emulation (like a software interpreter) would be very slow. It could actually slow down the code by maybe a factor of 100. So maybe you would only need a 100GHz Athlon to do it with emulation!

On the other hand if the actual code could be reworked to replace the SSE with 3dNOW then you should get no performance loss. Unfortunately I think it would be no easy task to write a program that could automatically strip out SSE and replace it with 3dnow. I think you need the program's developer to do it at the source code level.

 

[K6]

Originally posted by: uart

BTW k6, what particular software is it that wont run without SSE yet also does not require a lot of CPU horsepower?

Normally the use of SSE would indicate that the software is very cpu instensive, otherwise you wouldn't go to the trouble of SSE coding.

Adobe Premiere Pro ....
the software require SSE , but I think that a Tbird 1400 can run it without any problem for light work.
Adobe say that the entry configuration for the software is a PIII 850

Originally posted by: uart
3dnow and SSE are indeed very similar, both in terms of their speed and efficiency and also in terms of what they actually achieve. But the implementation is quite different and I dont think 1-1 translation would be possible.

both are SIMD floating point instruction they can do the work (on the Athlon XP core (as far I know) the SSE are microprogrammed on the 3DNow!)

Originally posted by: uart
Run time emulation (like a software interpreter) would be very slow. It could actually slow down the code by maybe a factor of 100. So maybe you would only need a 100GHz Athlon to do it with emulation!

Emulating a complete computer is hard , but not so much ... think abouti the VirtualPC on MAC or the TRANSMETA CPU , a pure Risc CPU emulating a X86 CPU with a software.

Emulating just some instruction is light and easy , think about the MMX wrapper , the performace between a Pentium MMX or a Pentium Classic + Wrapper are very close (the MMX is faster because have a doubled L1)

The performace Gap is also present ONLY when the programm use this LIMITED set of instruction , floating point or integer instruction are not slowed down


With Windows 2000 and later there is a nice "CPU driver" ... putting a wrapper inside the driver can be very useful.

Originally posted by: uart
On the other hand if the actual code could be reworked to replace the SSE with 3dNOW then you should get no performance loss. Unfortunately I think it would be no easy task to write a program that could automatically strip out SSE and replace it with 3dnow. I think you need the program's developer to do it at the source code level.

Adobe like Intel ... no hope

 

[Mday]

first:

premiere can use 3dnow.

second:

the tbird has sse. it doesnt have sse2. gg licensing.

--

3dnow and sse are different implementations that are meant to do the same thing. using emulation with premiere, if possible, will make it work horribly.

--

both are SIMD floating point instruction they can do the work (on the Athlon XP core (as far I know) the SSE are microprogrammed on the 3DNow!)

now that just makes no sense (the end of it).

 

[uart]

Originally posted by: Mday
first:

premiere can use 3dnow.

No Mday, i was also a little skeptical about this so I checked adobe's site. Premere-Pro definitely NEEDS to have SSE, it simply wont even start without it. I even read some posts from users who had AthlonXP's (which has SSE) but where the bios hadn't enabled SSE properly and Premier-Pro would not even start (apparently it just displays an error message about not being able to run without SSE and then exits).

3dnow and sse are different implementations that are meant to do the same thing. using emulation with premiere, if possible, will make it work horribly.

Agreed

both are SIMD floating point instruction they can do the work (on the Athlon XP core (as far I know) the SSE are microprogrammed on the 3DNow!)

now that just makes no sense (the end of it).

I dont know the details and haven't looked into this stuff lately. But I see no reason why a large part of SSE might not have been implemented with only microprogramming additions as opposed to actually adding a completely new exection unit.

After all, the existing 3dnow/mmx unit already had all the execution units to handle the vector processing. In particular 3dnow already had two seperate 64 bit 3dnow/mmx execution units (a superscalar design that allowed two simultaneous 3dnow vector operations to be retired per clock cycle). I'd have thought it would make perfect sense to implement the 128 bit SSE by means of these existing execution units and changes to the microprogramming.

 

[zephyrprime]

the tbird has sse. it doesnt have sse2. gg licensing.

The tbird doesn't have sse. Only the Tbred has that.

And I'm sure the reason that there's no SSE emulator is simply because there's very little interest. I don't think there's much interest in any software of this sort. People simply get new computers.

 

[K6]

Originally posted by: Mday
first:

premiere can use 3dnow.

no ...this is the problem

second:

the tbird has sse. it doesnt have sse2. gg licensing.

Athlon classic (Slot A) and Tbird (Socket A) have only 3Dnow! extension

3dnow and sse are different implementations that are meant to do the same thing. using emulation with premiere, if possible, will make it work horribly.

maybe will be a little slower but ONLY using the emulated instruction ... 99% of the code is just ALU or FPU instruction

Emulation will be not much slower than original , in this case is not an architecture emulation (like the virtualPC) but just a instruction renaming

both are SIMD floating point instruction they can do the work (on the Athlon XP core (as far I know) the SSE are microprogrammed on the 3DNow!)

now that just makes no sense (the end of it).

The Pentium was the last CISC x86 CPU.
starting from the P6 core all the x86 CPU use a internal RISC core and a decode unit.

X86 instruction are translated into fixed lenght risc-like instruction and executed by the RISC core inside the CPU.

the conversion table is microprogrammed and for each X86 instruction there is a short subroutine of risc-like instruction that make the job.
(on the P4 also L1 store decoded micro-ops )

adding the SSE support on AthlonXP is made just by adding some entry in the conversion table , and in most case this new entry just recall the corresponding 3Dnow! subroutine with the right parameter.

If there is a 1 to 1 correlation between a SSE ad a 3DNow! instruction the conversion can be made with a software device driver almost without performance penalty.

Some SSE instruction can be replaced with standard FPU assembly routine , using FPU is not a problem because also
SEE and 3Dnow! use the FPU register for executing , software emulation is not using more resource than hardware execution

 

[K6]

No News ????

 

[K6]

Anyone ?????

If you have a SSE emulator just show it to the world .... please