fix(simd): gate simd_avx512 tests behind target_feature = avx512f

AGENTS.md

@@ -18,9 +18,10 @@ This is a Rust library crate (ndarray fork with HPC extensions). No external ser
 
 ### Environment notes
 
-- **Rust 1.94.0** is pinned via `rust-toolchain.toml`; rustup auto-selects it in `/workspace`.
-- **No AVX-512 hardware** in Cloud Agent VMs — SIMD kernel tests using `#[target_feature(enable = "avx512f")]` are compile-gated and will be skipped at runtime. This is expected behavior.
+- **Rust 1.94.1** is pinned via `rust-toolchain.toml`; rustup auto-selects it in `/workspace`.
+- **No AVX-512 hardware** in Cloud Agent VMs — all test modules in `src/simd_avx512.rs` are gated with `#[cfg(all(test, target_feature = "avx512f"))]` and compile away entirely on non-AVX-512 targets. This is intentional: raw AVX-512 intrinsic tests must never run on CI/Cloud (x86-64-v3). The `simd.rs` LazyLock polyfill dispatches to `simd_avx2.rs` on these machines.
 - **Feature gates**: `intel-mkl` and `openblas` are mutually exclusive and require system libraries not installed by default. The default build uses `native` (pure Rust SIMD) which needs no extra libs.
-- **Build time**: ~18s cold, <1s incremental. Tests (~1819) take ~70s.
+- **Build time**: ~18s cold, <1s incremental. Tests (~1776 on non-AVX-512) take ~70s.
 - The workspace has sub-crates under `crates/` and `ndarray-rand/`. Default members exclude `blas-tests` and `blas-mock-tests` (they activate the `blas` feature which needs cblas-sys linking).
 - `libssl-dev` is needed as a build dependency for some transitive crates.
+- **`cargo fmt`**: `rustfmt.toml` uses 13+ nightly-only options (`brace_style`, `imports_granularity`, etc.). Stable rustfmt ignores them and reports massive diffs. This is a known pre-existing issue — do not attempt to fix formatting drift without coordinating with the project owner.

src/simd_avx512.rs

@@ -2619,7 +2619,7 @@ unsafe fn convert_f32_to_bf16_avx512f_rne(input: &[f32], output: &mut [u16]) {
     }
 }
 
-#[cfg(test)]
+#[cfg(all(test, target_feature = "avx512f"))]
 mod bf16_tests {
     use super::*;
 
@@ -3260,7 +3260,7 @@ pub fn f32_to_f16_batch_ieee754_rne(input: &[f32], output: &mut [u16]) {
     }
 }
 
-#[cfg(test)]
+#[cfg(all(test, target_feature = "avx512f"))]
 mod f16_tests {
     use super::*;
 
@@ -3351,7 +3351,7 @@ mod f16_tests {
     }
 }
 
-#[cfg(test)]
+#[cfg(all(test, target_feature = "avx512f"))]
 mod u8x64_rasterizer_tests {
     use super::U8x64;
 
@@ -3467,7 +3467,7 @@ mod u8x64_rasterizer_tests {
     }
 }
 
-#[cfg(test)]
+#[cfg(all(test, target_feature = "avx512f"))]
 mod tier3_tests {
     use super::{U8x64, U16x32};
 
@@ -3590,7 +3590,7 @@ mod tier3_tests {
 // whichever path the linker selected.
 // ────────────────────────────────────────────────────────────────────────
 
-#[cfg(test)]
+#[cfg(all(test, target_feature = "avx512f"))]
 mod int_simd_tests {
     use crate::simd::{I8x32, I8x64, I16x16, I16x32};