fix Issue 19861 - core.cpuid reports the wrong number of threads

[rainers]

do not use i7 detection on AMD processors
use cpuid 0x8000_001E to detect the number of threads per core

It seems AMD uses the term "core" for "logical cores" or threads, too.

[dlang-bot]

Thanks for your pull request, @rainers!

Bugzilla references

Auto-close	Bugzilla	Severity	Description
✓	19861	minor	core.cpuid reports the wrong number of threads

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub fetch digger
dub run digger -- build "stable + druntime#2620"

[rainers]

@FraMecca This worked for an AMD Ryzen, could you check it on your system, too?

[thewilsonator]

Enforce whitespace before opening parenthesis
--
  | grep -nrE "\<(for\|foreach\|foreach_reverse\|if\|while\|switch\|catch\|version)\(" $(find src -name '*.d') ; test $? -eq 1
  | src/core/cpuid.d:949:            version(GNU) asm pure nothrow @nogc {
  | posix.mak:383: recipe for target 'style_lint' failed

[rainers]

Anyone else with an AMD processor who can try this? Running and verifying this should be enough:

import core.cpuid;
import core.stdc.stdio;

void main()
{
    printf("coresPerCPU: %d\n", coresPerCPU);
    printf("threadsPerCPU: %d\n", threadsPerCPU);
}

Do we know whether there is an AMD used by any CI?

Parallel marking relies on this information, so would be good to have this in before the next release.

[JinShil]

Anyone else with an AMD processor who can try this?

cc @baziotis

[baziotis]

I have an AMD Ryzen 7 1700. It is 8-core processor, with 2 threads per core.
But I get:

coresPerCPU: 16
threadsPerCPU: 1

lscpu data:

CPU(s):              16
Thread(s) per core:  2
Core(s) per socket:  8

[rainers]

have an AMD Ryzen 7 1700. It is 8-core processor, with 2 threads per core.
But I get:
coresPerCPU: 16
threadsPerCPU: 1

Thanks for testing. I guess this is without this patch? With the patch it should be correct (8/16), but unfortunately your CPU is exactly the same I've been trying it on.

[baziotis]

but unfortunately your CPU is exactly the same I've been trying it on.

What are the odds..

I guess this is without this patch? With the patch it should be correct (8/16)

Yes and yes.

[rainers]

Rebased on stable as this is important for parallel marking to work on AMD systems.
I added a test to print out cpuid values, mostly to see if there is an AMD processor in our test fleet.

[rainers]

auto tester redmond reports dubious values for Intel Core2, too:

processor: Intel(R) Core(TM)2 Quad CPU    Q9400  @ 2.66GHz
threadsPerCPU:    1
coresPerCPU:      4

[FraMecca]

Sorry for the delay. It works on my machines.

[rainers]

Sorry for the delay. It works on my machines.

Thanks for testing. I'll remove the verbose output of all the flags from the test.

[thewilsonator]

this good to go?

[rainers]

this good to go?

I think so. The bad result for Intel Core2 is another issue, but it is a pretty old CPU (released 2008).