assert!(cfg!(target_feature = "sse") && cfg!(target_feature = "sse2")) panic at build on i386, i586, FreeBSD, or many target_os=none x86 targets.

[girgen]

Hi!

I get this error when building an app using ring-0.17.8

error[E0080]: evaluation of constant value failed
 --> /wrkdirs/usr/ports/www/sqlpage/work/SQLpage-0.20.0/cargo-crates/ring-0.17.8/src/cpu/[intel.rs:28](http://intel.rs:28/):9
  |
28 |         assert!(cfg!(target_feature = "sse") && cfg!(target_feature = "sse2"));
  |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the evaluated program panicked at 'assertion failed: cfg!(target_feature = "sse") && cfg!(target_feature = "sse2")', /wrkdirs/usr/ports/www/sqlpage/work/SQLpage-0.20.0/cargo-crates/ring-0.17.8/src/cpu/[intel.rs:28](http://intel.rs:28/):9
  |
  = note: this error originates in the macro `assert` (in Nightly builds, run with -Z macro-backtrace for more info)

For more information about this error, try `rustc --explain E0080`.
error: could not compile `ring` (lib) due to 1 previous error

I tried to find info using the three issue links in the code, but I don't have enough experience with rust, to be honest, I'm just packaging an application. It seems to me that the problem is with the assert call itself, so perhaps this is actually a bug?

Removing the check will make the code build.

Do you need more information, just specify what is needed. Build log attached.

sqlpage-0.20.0.log

[briansmith]

What are the steps to reproduce? cargo build --target=????

[girgen]

See the attached log above ☝️

Looks like

--target i686-unknown-freebsd

[briansmith]

What do you get when you do this?

$ rustc --print=cfg --target i686-unknown-freebsd | grep sse
debug_assertions
target_feature="sse"
target_feature="sse2"

For me, rustc indicates that sse and sse2 are there.

I see in your build log that there are warnings that CARGO_CFG_TARGET_FEATURE not set. THat might be a clue.

Make sure you're not disabling SSE/SSSE2 in .cargo/config.toml or otherwise.

[girgen]

% rustc --print=cfg --target i686-unknown-freebsd | grep sse
debug_assertions

Seems like rust is configured for pentiumpro (no SSEx) in the FreeBSD ports tree?

See https://github.com/freebsd/freebsd-ports/blob/main/lang/rust/files/patch-compiler_rustc__target_src_spec_targets_i686__unknown__freebsd.rs

[briansmith]

At https://www.freebsd.org/releases/14.0R/hardware/#proc-i386, it is written:

2.3. i386 Architecture Support
FreeBSD maintains support for i386 (x86) as a Tier 2 architecture. It is not recommended for new installations.

Support for this architecture will be removed in FreeBSD 15.0.

[dch]

FreeBSD 15.0 hasn't been released yet, so this is in the future.
We still have FreeBSD 13.3-RELEASE and 14.1-RELEASE active, which
will run for several more years yet.

For me, this blocks e.g. lang/gleam on i386, which admittedly is neither
a popular port, nor a very common platform yet.

Do we amend the i386 target on FreeBSD, for rust? @MikaelUrankar I know you're familiar with the FreeBSD rust porting, what do you suggest here?

Perhaps we should discuss this in bugzilla?

[briansmith]

Note that you can build ring with RUSTFLAGS="-C target-feature=+sse2" (or whatever) to get it to work on those targets. Note that it might have built before I added these assertions, but if you tried to run it it would execute SSE2 instructions without checking at runtime whether the system actually supported SSE2 instructions.

If you actually want ring to support x86 systems that don't have SSE2, then we'd need somebody to submit a PR that adds the runtime checking for SSE2 to every assembly language function called. They are easy to find; look for target_arch="x86" in the source code. To see how to add the check for SSE2, look at how the check for SSSE3 is done by searching for SSSE3.

[he32]

I'll just note that I'm observing the same problem for the i586-unknown-netbsd target. It also does not support the SSE or SSE2 instructions, and targets Pentium and above, for maximal backward compatibility:

$ rustc --print=cfg --target i586-unknown-netbsd | grep sse
debug_assertions
$ rustc --print=cfg | grep sse
debug_assertions
$ uname -rps
NetBSD 9.3 i386
$ rustc --version
rustc 1.78.0 (9b00956e5 2024-04-29) (built from a source tarball)
$

This is seen for ring-0.17.8:

error[E0080]: evaluation of constant value failed
  --> /usr/pkgsrc/net/routinator/work/vendor/ring-0.17.8/src/cpu/intel.rs:28:9
   |
28 |         assert!(cfg!(target_feature = "sse") && cfg!(target_feature = "sse2"));
   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the evaluated program panicked at 'assertion failed: cfg!(target_feature = "sse") && cfg!(target_feature = "sse2")', /usr/pkgsrc/net/routinator/work/vendor/ring-0.17.8/src/cpu/intel.rs:28:9

[briansmith]

Yes, the also applies to that target.

[guenhter]

@briansmith I'd like to help out supporting the i586 target. After looking into the code, I have no clue what exactly I should do. Also looking at SSSE3 didn't really help.
Do I only have to adapt the rust code or also the .pl stuff? Would it be possible to show me one example what I have to do, so that I can do it then on all other places where target_arch="x86" is set.