Module activation 

Activation kernels: ReLU, GELU, SiLU.

Matches activation-kernel-v1.yaml.

Each function provides one of three backends:

• fn {name}_scalar(...) — Pure Rust scalar reference (ground truth)
• unsafe fn {name}_avx2(...) — AVX2 SIMD implementation
• fn {name}_ptx() -> &'static str — PTX assembly source string

Functions

gelu_avx2^⚠

AVX2 GELU — delegates to scalar (no hardware tanh in AVX2).

gelu_ptx

PTX assembly for the GELU kernel.

gelu_scalar

GELU: 0.5 * x * (1 + tanh(sqrt(2/pi) * (x + 0.044715 * x^3)))

relu_avx2^⚠

AVX2 ReLU: _mm256_max_ps(x, zero) with scalar tail.

relu_ptx

PTX assembly for the ReLU kernel (elementwise, 1 thread per element).

relu_scalar

ReLU: max(0, x)

silu_avx2^⚠

AVX2 SiLU — delegates to scalar (no hardware exp in AVX2).

silu_ptx

PTX assembly for the SiLU (Swish) kernel.

silu_scalar

SiLU (Swish): x / (1 + exp(-x))