Gentoo Virtual Machine in the UTM App on Macbook Pro M1 Max

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

*(tldr edit): -march=native is not presently working for me
8-25-24, I am now leaving -march=native in make.conf to have it globally, and using per-package fixes for the broken packages.

I have a Gentoo virtual machine, ~arm64, running on an M1 Max Macbook Pro in UTM Version 4.0.9. I first built it in August and everything worked.

Since then some updates are now failing. One type of failure is

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

media-libs/xvid-1.3.7-r1 was also failing with

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

Did a sync, now I'm up to 9 broken packages.

This install started and worked with gcc-12.1, no longer in the tree.

edit to add context: The nine failed packages are with march declared as implied by gcc -march=native -E -v - </dev/null 2>&1 | grep cc1 results.

NeddySeagoon · Posted: Tue Dec 13, 2022 10:11 pm Post subject:

flysideways,

Is COMMON_FLAGS march=native any help?
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

@Neddy, it is. I was rebuilding stuff and didn't make the previous post as detailed as I should have.

With native for march and mcpu, I get the best results.

Built with gcc-12.2.0, only these two have errors that cause failures

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

Oh yeah, gentoo-kernel-6.0.12 also failed with the declared options. I did not bother checking the errors in that build log as I had already looked through all of the others and they all had some different, and shared some common errors. They were all "processor does not support" errors.

gentoo-kernel-6.0.12 built fine with native.

Also, python was not building with the declared options.

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

Updating to sys-devel/gcc-12.2.1_p20221210 both media-libs/openh264-2.3.1-r1 and media-libs/libaom-3.5.0 continue with the previous errors and media-libs/libaom-3.5.0 has added

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

I ran an emerge -ve @world, 38 of the installed packages will no longer build, including my kernels. I'll look at the errors later.

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

With

Hu · Administrator Joined: 06 Mar 2007 Posts: 22754

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

Hi @Hu, thanks for looking.

I did eventually switch to binutils 2.39 with similar results.

Using resolve-march-native output, I added each Feature Modifier, one at a time to the -march= statement until the build of libaom failed. What I found was +sha3 creates the bcax related failures and +dotprod creates the udot failures.

This is the working emerge --info for libaom with all -march= Feature Modifiers from resolve-march-native, except +sha3 and +dotprod that cause the Errors and failures to build.

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

Running --emptytree with native currently shows these failing packages;

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

Using the M1 MacBook to maintain my Raspberry Pi looks promising.

gluker · n00b Joined: 06 Jan 2023 Posts: 1

I've found the problem related to GCC LTO optimization for GMP for ARM aarch64
under QEMU with ARM Neoverse-N1 virt-4.2 (Oracle Cloud) and ARM virt-7.2 (Apple M1)
both running the latest Gentoo with gmp-6.2.1-r5 compiled by both
gcc version 11.3.1 20221209 and gcc version 12.2.1 20221231.

To reproduce:
1. Compile GMP with LTO optimization -- compiles OK
2. Try to compile iproute2-6.1.0 or cmake-3.25.1 -- compilation hangs during the process

Disabling LTO optimization for GMP solves the problem.

P.S.: i've reported the issue to gmp-bugs and Gentoo's bugzilla

chin39 · n00b Joined: 29 Jan 2023 Posts: 1

Hi, I'm using m2 max with UTM, I found some packages like openh264 could build by clang with -march=native.

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

Tried another --emptytree with march=native.

costel78 · Guru Joined: 20 Apr 2007 Posts: 407

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

I have just tried VMWare's Fusion 13. It too has the same problems in gcc with -march=native.

UTM has an issue recognizing many external drives that Fusion does not. With Fusion, I can mount and update the Samsung T7 drives that I had been using as the boot disks for a couple of Pi's. Using UTM, I had to return to sd cards, I could mount them.

I am going to use Fusion and again use the ssd drives with my Pi's.

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

emptytree with -march=native and gcc 14 has the following that do not compile in a VMWare Fusion VM

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

An emerge emptytree -march=native in a UTM VM with gcc 14, this one is XFCE with no Ruby:

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

The packages that will not complete with -march=native in the above two posts do complete with -march=armv8-a.

Both UTM and VMWare VM's had previously been successfully compiling everything for about 1 1/2 years with -march=armv8-a+crypto+crc+lse+rcpc+rdma+fp16fml+sb+flagm+pauth. It has only been the last few months that things began to fail.

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

I have been able to build all of the packages that currently do not build with -march=native with some reduced combination of cpu extensions.

flysideways · Guru Joined: 29 Jan 2005 Posts: 496

glibc stopped building in late August.

It was failing complaining about a lack of SVE ISA extensions.

After chasing that for a while, I discovered that removing the -mcpu=native declaration from make.conf fixed the problem. Doing that has not appeared to break anything new, but has fixed glibc.