[ICE]: autodiff fails compiling gemm-f32 microkernel

I have some code using the nightly autodiff feature, and compiling one of my dependencies (gemm-f32) fails with an ICE. With the help of an LLM I put together an MRE based on gemm-f32's fma x1x1 microkernel, which is below, compiled with:

CARGO_PROFILE_RELEASE_LTO=fat RUSTFLAGS='-Z autodiff=Enable' cargo +nightly test --release

Code

use core::arch::x86_64::{_mm256_fmadd_ps, _mm256_set1_ps};

const N: usize = 8;
type Pack = [f32; N];

unsafe fn splat(x: f32) -> Pack {
    unsafe { core::mem::transmute(_mm256_set1_ps(x)) }
}

unsafe fn mul_add(a: Pack, b: Pack, c: Pack) -> Pack {
    unsafe {
        core::mem::transmute(_mm256_fmadd_ps(
            core::mem::transmute(a),
            core::mem::transmute(b),
            core::mem::transmute(c),
        ))
    }
}

#[target_feature(enable = "fma")]
// Reduced from gemm-f32 0.19.0's FMA x1x1 microkernel.
// https://docs.rs/gemm-common/0.19.0/src/gemm_common/microkernel.rs.html#428-495
pub unsafe fn kernel(dst: *mut f32, lhs: *const f32, rhs: f32) {
    let mut storage = [[unsafe { splat(0.0) }; 1]; 1];
    let accum = storage.as_mut_ptr() as *mut Pack;
    #[derive(Copy, Clone)]
    struct KernelIter {
        lhs: *const f32,
        rhs: f32,
        accum: *mut Pack,
    }
    impl KernelIter {
        unsafe fn execute(self, i: usize) {
            unsafe {
                *self.accum = mul_add(
                    *(self.lhs.add(i * N) as *const Pack),
                    splat(self.rhs),
                    *self.accum,
                )
            }
        }
    }
    let iter = KernelIter { lhs, rhs, accum };
    unsafe {
        iter.execute(0);
        iter.execute(1)
    };
    unsafe { (dst as *mut Pack).write_unaligned(*accum) };
}

Meta

rustc +nightly --version --verbose:

rustc 1.99.0-nightly (7608eb7b0 2026-08-05)
binary: rustc
commit-hash: 7608eb7b07eaf93f16d7cf5bcb2098eca87503df
commit-date: 2026-08-05
host: x86_64-unknown-linux-gnu
release: 1.99.0-nightly
LLVM version: 23.1.0

Error output

thread 'rustc' (1155723) panicked at /rustc-dev/7608eb7b07eaf93f16d7cf5bcb2098eca87503df/compiler/rustc_middle/src/ty/normalize_erasing_regions.rs:76:13:
assertion `left == right` failed
  left: [f32; AliasConst(No, Alias { kind: Anon { def_id: DefId(0:8 ~ diffsol_gemm_ice[9671]::Pack::{constant#0}) }, args: [], .. })]
 right: [f32; 8_usize]
stack backtrace:
   0:     0x76420d302f76 - <<std[1d42cb0d866aa333]::sys::backtrace::BacktraceLock>::print::DisplayBacktrace as core[4e7ceada952a6ea0]::fmt::Display>::fmt
   1:     0x76420da1114f - core[4e7ceada952a6ea0]::fmt::write
   2:     0x76420d317e9c - <std[1d42cb0d866aa333]::sys::stdio::unix::Stderr as core[4e7ceada952a6ea0]::io::write::Write>::write_fmt
   3:     0x76420d2d5d5a - std[1d42cb0d866aa333]::panicking::default_hook::{closure#0}
   4:     0x76420d2f6af3 - std[1d42cb0d866aa333]::panicking::default_hook
   5:     0x76420c240b9a - std[1d42cb0d866aa333]::panicking::update_hook::<alloc[571f99e20346f6ba]::boxed::Box<rustc_driver_impl[f766a19ff37a9559]::install_ice_hook::{closure#1}>>::{closure#0}
   6:     0x76420d2f6f92 - std[1d42cb0d866aa333]::panicking::panic_with_hook
   7:     0x76420d2d5e12 - std[1d42cb0d866aa333]::panicking::panic_handler::{closure#0}
   8:     0x76420d2ccf79 - std[1d42cb0d866aa333]::sys::backtrace::__rust_end_short_backtrace::<std[1d42cb0d866aa333]::panicking::panic_handler::{closure#0}, !>
   9:     0x76420d2d77ed - __rustc[e79ffd21e4789ee5]::rust_begin_unwind
  10:     0x764209f40c7c - core[4e7ceada952a6ea0]::panicking::panic_fmt
  11:     0x76420bbd3e73 - core[4e7ceada952a6ea0]::panicking::assert_failed_inner
  12:     0x76420c809396 - core[4e7ceada952a6ea0]::panicking::assert_failed::<rustc_middle[760f81c9f8f0fdc5]::ty::Ty, rustc_middle[760f81c9f8f0fdc5]::ty::Ty>
  13:     0x76420e2d96df - <rustc_middle[760f81c9f8f0fdc5]::ty::context::TyCtxt>::struct_tail_for_codegen
  14:     0x76420c85b7a5 - rustc_middle[760f81c9f8f0fdc5]::ty::typetree::handle_indirection
  15:     0x76420c85c40e - rustc_middle[760f81c9f8f0fdc5]::ty::typetree::typetree_from_ty_impl_inner
  16:     0x76420c85c547 - rustc_middle[760f81c9f8f0fdc5]::ty::typetree::typetree_from_ty_impl_inner
  17:     0x76420df80f94 - <rustc_codegen_ssa[cfbee05c377dfbef]::mir::operand::OperandValue<&rustc_codegen_llvm[c56b079047732da9]::llvm::ffi::Value>>::store_with_flags::<rustc_codegen_llvm[c56b079047732da9]::builder::GenericBuilder<rustc_codegen_llvm[c56b079047732da9]::context::FullCx>>
  18:     0x76420df7fbd3 - <rustc_codegen_ssa[cfbee05c377dfbef]::mir::operand::OperandRef<&rustc_codegen_llvm[c56b079047732da9]::llvm::ffi::Value>>::store_with_annotation_and_flags::<rustc_codegen_llvm[c56b079047732da9]::builder::GenericBuilder<rustc_codegen_llvm[c56b079047732da9]::context::FullCx>>
  19:     0x76420ee78f65 - rustc_codegen_ssa[cfbee05c377dfbef]::mir::codegen_mir::<rustc_codegen_llvm[c56b079047732da9]::builder::GenericBuilder<rustc_codegen_llvm[c56b079047732da9]::context::FullCx>>
  20:     0x76420dfa190b - rustc_codegen_llvm[c56b079047732da9]::base::compile_codegen_unit::module_codegen
  21:     0x76420ea9a578 - <rustc_codegen_llvm[c56b079047732da9]::LlvmCodegenBackend as rustc_codegen_ssa[cfbee05c377dfbef]::traits::backend::ExtraBackendMethods>::compile_codegen_unit
  22:     0x76420ea9649a - rustc_codegen_ssa[cfbee05c377dfbef]::base::codegen_crate::<rustc_codegen_llvm[c56b079047732da9]::LlvmCodegenBackend, rustc_codegen_llvm[c56b079047732da9]::ModuleLlvm>
  23:     0x76420ea94eeb - <rustc_codegen_llvm[c56b079047732da9]::LlvmCodegenBackend as rustc_codegen_ssa[cfbee05c377dfbef]::traits::backend::CodegenBackend>::codegen_crate
  24:     0x76420ec70d25 - <rustc_interface[fcf9d59e9eac587f]::queries::Linker>::codegen_and_build_linker
  25:     0x76420ec686d5 - rustc_interface[fcf9d59e9eac587f]::interface::run_compiler::<(), rustc_driver_impl[f766a19ff37a9559]::run_compiler::{closure#0}>::{closure#2}
  26:     0x76420ec9a424 - std[1d42cb0d866aa333]::sys::backtrace::__rust_begin_short_backtrace::<rustc_interface[fcf9d59e9eac587f]::util::run_in_thread_with_globals<rustc_interface[fcf9d59e9eac587f]::util::run_in_thread_pool_with_globals<rustc_interface[fcf9d59e9eac587f]::interface::run_compiler<(), rustc_driver_impl[f766a19ff37a9559]::run_compiler::{closure#0}>::{closure#2}, ()>::{closure#0}, ()>::{closure#0}::{closure#0}, ()>
  27:     0x76420ec9a1ed - <std[1d42cb0d866aa333]::thread::lifecycle::spawn_unchecked<rustc_interface[fcf9d59e9eac587f]::util::run_in_thread_with_globals<rustc_interface[fcf9d59e9eac587f]::util::run_in_thread_pool_with_globals<rustc_interface[fcf9d59e9eac587f]::interface::run_compiler<(), rustc_driver_impl[f766a19ff37a9559]::run_compiler::{closure#0}>::{closure#2}, ()>::{closure#0}, ()>::{closure#0}::{closure#0}, ()>::{closure#1} as core[4e7ceada952a6ea0]::ops::function::FnOnce<()>>::call_once::{shim:vtable#0}
  28:     0x76420ec9733a - <std[1d42cb0d866aa333]::sys::thread::unix::Thread>::new::thread_start
  29:     0x76420828307a - start_thread
                               at ./nptl/pthread_create.c:454:8
  30:     0x764208316534 - clone
                               at ./misc/../sysdeps/unix/sysv/linux/x86_64/clone.S:100:0
  31:                0x0 - <unknown>

error: the compiler unexpectedly panicked. This is a bug

note: we would appreciate a bug report: https://github.com/rust-lang/rust/issues/new?labels=C-bug%2C+I-ICE%2C+T-compiler&template=ice.md

note: please make sure that you have updated to the latest nightly

note: please attach the file at `/home/mrobins/git/diffsol-gemm-ice/rustc-ice-2026-08-06T12_28_40-1155718.txt` to your bug report

note: rustc 1.99.0-nightly (7608eb7b0 2026-08-05) running on x86_64-unknown-linux-gnu

note: compiler flags: --crate-type lib -C opt-level=3 -C linker-plugin-lto -C strip=debuginfo -Z autodiff=Enable

note: some of the compiler flags provided by cargo are hidden

query stack during panic:
end of query stack
error: could not compile `diffsol-gemm-ice` (lib)

Caused by:
  process didn't exit successfully: `/home/mrobins/.rustup/toolchains/nightly-x86_64-unknown-linux-gnu/bin/rustc --crate-name diffsol_gemm_ice --edition=2021 src/lib.rs --error-format=json --json=diagnostic-rendered-ansi,artifacts,future-incompat --diagnostic-width=135 --crate-type lib --emit=dep-info,metadata,link -C opt-level=3 -C linker-plugin-lto --check-cfg 'cfg(docsrs,test)' --check-cfg 'cfg(feature, values())' -C metadata=b07a6737c808e529 -C extra-filename=-3e15a5b336437cad --out-dir /home/mrobins/git/diffsol-gemm-ice/target/release/build/diffsol-gemm-ice/3e15a5b336437cad/out -C strip=debuginfo -Z autodiff=Enable` (exit status: 101)

Start by reproducing the ICE from the provided MRE in src/lib.rs with the nightly toolchain and the shown cargo command. Inspect the panic at compiler/rustc_middle/src/ty/normalize_erasing_regions.rs:76 and the attached rustc ICE log; done means the reproduction no longer causes rustc to panic and the test compiles successfully.