Directx 什么'；当索引=0时，XMVectorSetByIndex和XMVectorSetX之间的差异是多少？

我阅读了DirectXMath库的源代码，发现

xmvectorstebyindex

和

XMVectorSetX

的实现完全不同。为什么不

XMVectorSetX

简单地返回

xmvectorsetbyndex

（index=0）？

XMVectorSetX

实际上能够使用SSE或ARM-NEON内部函数，而

xmvectorsetbyndex

必须“溢出到内存”（即它根本不是SIMD）

查看

XMVectorSetY

案例，以及使用

/arch:AVX

或

/arch:AVX2

它能够使用SSE4指令

\u mm\u insert\u ps

否则它必须做相当多的工作来获取SIMD代码生成器，而不必“溢出到内存”

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
}

请注意，DirectXMath现在在上可用

// 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
}

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
}