What's the difference between XMVectorSetByIndex when index = 0 and XMVectorSetX?

I read the source code of the DirectXMath Library and found that the implementation of XMVectorSetByIndex and XMVectorSetX are completely different. Why not XMVectorSetX simply returns XMVectorSetByIndex (index = 0) ?

XMVectorSetX is actually able to use SSE or ARM-NEON intrinsics, while XMVectorSetByIndex has to 'spill to memory' (i.e. it's not SIMD at all).

// Set a single indexed floating point component
inline XMVECTOR XM_CALLCONV XMVectorSetByIndex(FXMVECTOR V, float f, size_t i)
{
    assert( i < 4 );
    _Analysis_assume_( i < 4 );
#if defined(_XM_NO_INTRINSICS_)
    XMVECTOR U;
    U = V;
    U.vector4_f32[i] = f;
    return U;
#elif defined(_XM_ARM_NEON_INTRINSICS_)
    XMVECTOR U = V;
    U.n128_f32[i] = f;
    return U;
#elif defined(_XM_SSE_INTRINSICS_)
    XMVECTOR U = V;
    U.m128_f32[i] = f;
    return U;
#endif
}       

vs.

// Sets the X component of a vector to a passed floating point value
inline XMVECTOR XM_CALLCONV XMVectorSetX(FXMVECTOR V, float x)
{
#if defined(_XM_NO_INTRINSICS_)
    XMVECTOR U;
    U.vector4_f32[0] = x;
    U.vector4_f32[1] = V.vector4_f32[1];
    U.vector4_f32[2] = V.vector4_f32[2];
    U.vector4_f32[3] = V.vector4_f32[3];
    return U;
#elif defined(_XM_ARM_NEON_INTRINSICS_)
    return vsetq_lane_f32(x,V,0);
#elif defined(_XM_SSE_INTRINSICS_)
    XMVECTOR vResult = _mm_set_ss(x);
    vResult = _mm_move_ss(V,vResult);
    return vResult;
#endif
}

It's informative to look at the XMVectorSetY case as well where with /arch:AVX or /arch:AVX2 it's able to use the SSE4 instruction _mm_insert_ps otherwise it has to do a fair bit of work get SIMD code-gen instead of having to 'spill to memory'.

inline XMVECTOR XM_CALLCONV XMVectorSetY(FXMVECTOR V, float y)
{
#if defined(_XM_NO_INTRINSICS_)
    XMVECTOR U;
    U.vector4_f32[0] = V.vector4_f32[0];
    U.vector4_f32[1] = y;
    U.vector4_f32[2] = V.vector4_f32[2];
    U.vector4_f32[3] = V.vector4_f32[3];
    return U;
#elif defined(_XM_ARM_NEON_INTRINSICS_)
    return vsetq_lane_f32(y,V,1);
#elif defined(_XM_SSE4_INTRINSICS_)
    XMVECTOR vResult = _mm_set_ss(y);
    vResult = _mm_insert_ps( V, vResult, 0x10 );
    return vResult;
#elif defined(_XM_SSE_INTRINSICS_)
    // Swap y and x
    XMVECTOR vResult = XM_PERMUTE_PS(V,_MM_SHUFFLE(3,2,0,1));
    // Convert input to vector
    XMVECTOR vTemp = _mm_set_ss(y);
    // Replace the x component
    vResult = _mm_move_ss(vResult,vTemp);
    // Swap y and x again
    vResult = XM_PERMUTE_PS(vResult,_MM_SHUFFLE(3,2,0,1));
    return vResult;
#endif
}

Note that DirectXMath is now available on GitHub.