// Linear search for an int in an int array
// with NEON SIMD vector instructions,
// because GCC's vector_size extensions
// to C don't support any horizontal
// operations.  Does appear to work in
// minimal smoke tests.
// Cleaned up o4-mini and GPT-5 mini code
#include <arm_neon.h>
#include <stddef.h>

size_t
find(const int32_t *a, size_t n, int32_t k)
{
    // Duplicate needle into a NEON vector
    int32x4_t kv = vdupq_n_s32(k);
    size_t i = 0;
    
    for (; i + 4 <= n; i += 4) {
        // Load 4 integers from the array
        // GPT-5 says, "vld1q_s32 accepts
        // unaligned pointers on ARMv8; for
        // older ARM or strict alignment, use
        // memcpy or aligned loads."
        int32x4_t v = vld1q_s32(&a[i]);
        
        // Compare them to k
        uint32x4_t eq = vceqq_s32(v, kv);
        
        // Sum lanes and bail out if any match
        if (vaddvq_u32(eq)) break;
    }

    // If we are here, either we found a
    // match (but don't yet know in which
    // element) or we have finished all
    // but the last 0–3 elememts.
    while (i < n && a[i] != k) i++;
    return i; // n if not found
}