Who uses AMD CPUs

[Christoph Hart]

No thanks, I think I've figured it out - AMD has a CPU instruction set which is called SSE4a, and some parts of HLAC require SSE4.1 and the detection method didn't work correctly.

FYI, there's a compiler flag called HI_ENABLE_LEGACY_CPU_SUPPORT that bypasses all SSE instructions which should make it work on the most simple systems.

[d.healey]

FYI, there's a compiler flag called HI_ENABLE_LEGACY_CPU_SUPPORT that bypasses all SSE instructions which should make it work on the most simple systems.

Any reason not to enable it?

[Christoph Hart]

You lose about 10% - 20% streaming decoding performance.

[d.healey]

Ok, I'll use it if someone complains :)

[Christoph Hart]

Yeah, there were maybe less than 1% of all HEXERACT users with this issue, so there's no reason to deprive the other 99% of the performance of vectorised audio processing :)

[d.healey]

Could this be added as a CLI option? then I can build a legacy AMD build automatically alongside my normal builds.

[d.healey]

@d-healey said in Who uses AMD CPUs:

Could this be added as a CLI option? then I can build a legacy AMD build automatically alongside my normal builds.

Bump bump

[d.healey]

@d-healey said in Who uses AMD CPUs:

Could this be added as a CLI option? then I can build a legacy AMD build automatically alongside my normal builds.

I've added this as a CLI option -l. It also shows up as a checkbox in project preferences.

The CLI option is implemented in the project templates for GNU/Linux builds, I think Windows and Mac work differently (?) but I'll check next time I build on those systems.

[andioak]

@Christoph-Hart @d-healey @dustbro

I´m a bit confused. Why is it a problem using the HI_ENABLE_LEGACY_CPU_SUPPORT and then having a user run that plugin in a new cpu? I mean one that handles SSE4.1 fine. Is it not pre-programmed that one cpu-type not able to process that type just falls back on other less optimized code?

Why has that got anything to do with performance o that newer system? (that would not use the same code, right?)

[Christoph Hart]

I don't to runtime CPU dispatch. Either you want the SSE optimizations or you care about old computers.

The reason is that the possible performance overhead of a runtime branching to the most suitable SSE code might be higher than the actual improvement. It might be possible that the performance overhead is neglible on most systems, but the effort of profiling every single configuration is not worth it.

With SNEX it's another thing. It will compile the code on the fly and thus can find the best SIMD instruction set for every particular machine without branching.

[Natan]

Hey Anyone
Going To Cop A Microsoft Surface Laptop 3, Which Has An Amd Ryzen Built Into It.

Is Hise / Ipp / Vs2017 Fully Supported With AMD?
Or I'm Going To Have Difficulties?

Edit: Here Is The Details: https://www.microsoft.com/en-us/p/surface-laptop-3/8vfggh1r94tm?activetab=techspecs

13 Inch Model Got Intel Core i7.
15 Got Amd Processor.

Thanks

[Christoph Hart]

The AMD issue does only affect a certain generation of CPUs which came out around 2009. Newer CPUs work fine.

[Natan]

@Christoph-Hart Thanks Christoph
So What's Up With The Ipp Thing?
It Works And Integrates With VS2017 On AMD Based Laptops?

Thanks Man

[Christoph Hart]

Yes, but I wouldn‘t be surprised if these Intel crooks make their routines slower on AMD CPUs lololol

[Natan]

@Christoph-Hart Thank You Christoph
Sounds Like It's A Little Risky To Spending Over 1K $ On That AMD Machine.
Let's Stay Away From Fire, And Looking For A Intel Based Model :)

[Christoph Hart]

That‘s exactly what a sales person from Intel would say :)

[A Former User]

Pity it's a hard compile and no runtime check.