//! 流式 SIMD 扩展 4.1 (SSE4.1)

use crate::{
    core_arch::{simd::*, simd_llvm::*, x86::*},
    mem::transmute,
};

#[cfg(test)]
use stdarch_test::assert_instr;

// SSE4 舍入常数
/// 四舍五入到最接近的
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_TO_NEAREST_INT: i32 = 0x00;
/// 四舍五入
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_TO_NEG_INF: i32 = 0x01;
/// 围捕
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_TO_POS_INF: i32 = 0x02;
/// truncate
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_TO_ZERO: i32 = 0x03;
/// 使用 MXCSR.RC； 请参见 `vendor::_MM_SET_ROUNDING_MODE`
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_CUR_DIRECTION: i32 = 0x04;
/// 不要抑制异常
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_RAISE_EXC: i32 = 0x00;
/// 抑制异常
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_NO_EXC: i32 = 0x08;
/// 四舍五入到最接近，不排除异常
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_NINT: i32 = 0x00;
/// 四舍五入，不抑制异常
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_FLOOR: i32 = _MM_FROUND_RAISE_EXC | _MM_FROUND_TO_NEG_INF;
/// 向上舍入并且不抑制异常
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_CEIL: i32 = _MM_FROUND_RAISE_EXC | _MM_FROUND_TO_POS_INF;
/// 截断并且不抑制异常
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_TRUNC: i32 = _MM_FROUND_RAISE_EXC | _MM_FROUND_TO_ZERO;
/// 使用 MXCSR.RC 且不抑制异常； 请参见 `vendor::_MM_SET_ROUNDING_MODE`
///
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_RINT: i32 = _MM_FROUND_RAISE_EXC | _MM_FROUND_CUR_DIRECTION;
/// 使用 MXCSR.RC 并抑制异常； 请参见 `vendor::_MM_SET_ROUNDING_MODE`
#[stable(feature = "simd_x86", since = "1.27.0")]
pub const _MM_FROUND_NEARBYINT: i32 = _MM_FROUND_NO_EXC | _MM_FROUND_CUR_DIRECTION;

/// 使用 `mask` 混合 `a` 和 `b` 中的包装 8 位整数
///
/// 每个对应的屏蔽字节的高位确定选择。
/// 如果设置了高位，则选择 `a` 的元素。否则选择 `b` 的元素。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blendv_epi8)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pblendvb))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_blendv_epi8(a: __m128i, b: __m128i, mask: __m128i) -> __m128i {
    transmute(pblendvb(a.as_i8x16(), b.as_i8x16(), mask.as_i8x16()))
}

/// 使用掩码 `IMM8` 混合来自 `a` 和 `b` 的 16 位整数包装。
///
/// 掩码位确定选择。
/// 清除位选择 `a` 的相应元素，置 1 位选择 `b` 的相应元素。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blend_epi16)
///
#[inline]
#[target_feature(enable = "sse4.1")]
// Note: LLVM7 尽可能使用单精度浮点域
// see https://bugs.llvm.org/show_bug.cgi?id=38195
// #[cfg_attr(test, assert_instr(pblendw, IMM8 = 0xF0))]
#[cfg_attr(test, assert_instr(blendps, IMM8 = 0xF0))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_blend_epi16<const IMM8: i32>(a: __m128i, b: __m128i) -> __m128i {
    static_assert_uimm_bits!(IMM8, 8);
    transmute(pblendw(a.as_i16x8(), b.as_i16x8(), IMM8 as u8))
}

/// 使用 `mask` 混合来自 `a` 和 `b` 的包装的双精度 (64-bit) 浮点元素
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blendv_pd)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(blendvpd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_blendv_pd(a: __m128d, b: __m128d, mask: __m128d) -> __m128d {
    blendvpd(a, b, mask)
}

/// 使用 `mask` 混合来自 `a` 和 `b` 的包装的单精度 (32-bit) 浮点元素
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blendv_ps)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(blendvps))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_blendv_ps(a: __m128, b: __m128, mask: __m128) -> __m128 {
    blendvps(a, b, mask)
}

/// 使用控制掩码 `IMM2` 从 `a` 和 `b` 混合包装的双精度 (64-bit) 浮点元素
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blend_pd)
///
#[inline]
#[target_feature(enable = "sse4.1")]
// Note: LLVM7 尽可能使用单精度浮点域
// see https://bugs.llvm.org/show_bug.cgi?id=38195
// #[cfg_attr(test, assert_instr(blendpd, IMM2 = 0b10))]
#[cfg_attr(test, assert_instr(blendps, IMM2 = 0b10))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_blend_pd<const IMM2: i32>(a: __m128d, b: __m128d) -> __m128d {
    static_assert_uimm_bits!(IMM2, 2);
    blendpd(a, b, IMM2 as u8)
}

/// 使用掩码 `IMM4` 从 `a` 和 `b` 混合包装的单精度 (32-bit) 浮点元素
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blend_ps)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(blendps, IMM4 = 0b0101))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_blend_ps<const IMM4: i32>(a: __m128, b: __m128) -> __m128 {
    static_assert_uimm_bits!(IMM4, 4);
    blendps(a, b, IMM4 as u8)
}

/// 从用 `IMM8` 选择的 `a` 中提取单精度 (32-bit) 浮点元素。
/// 返回的 `i32` 存储浮点数的位模式，并且可以通过转换将其转换回浮点数。
///
/// # Example
///
/// ```rust
/// # #[cfg(target_arch = "x86")]
/// # use std::arch::x86::*;
/// # #[cfg(target_arch = "x86_64")]
/// # use std::arch::x86_64::*;
/// # fn main() {
/// #    if is_x86_feature_detected!("sse4.1") {
/// #       #[target_feature(enable = "sse4.1")]
/// #       unsafe fn worker() {
/// let mut float_store = vec![1.0, 1.0, 2.0, 3.0];
/// let simd_floats = _mm_set_ps(2.5, 5.0, 7.5, 10.0);
/// let x: i32 = _mm_extract_ps::<2>(simd_floats);
/// float_store.push(f32::from_bits(x as u32));
/// assert_eq!(float_store, vec![1.0, 1.0, 2.0, 3.0, 5.0]);
/// #       }
/// #       unsafe { worker() }
/// #   }
/// # }
/// ```
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_extract_ps)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(
    all(test, not(target_os = "windows")),
    assert_instr(extractps, IMM8 = 0)
)]
#[rustc_legacy_const_generics(1)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_extract_ps<const IMM8: i32>(a: __m128) -> i32 {
    static_assert_uimm_bits!(IMM8, 2);
    transmute(simd_extract::<_, f32>(a, IMM8 as u32))
}

/// 从 `a` 中提取一个 8 位整数，用 `IMM8` 选择。
/// 返回包含零扩展整数数据的 32 位整数。
///  See [LLVM commit D20468](https://reviews.llvm.org/D20468).
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_extract_epi8)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pextrb, IMM8 = 0))]
#[rustc_legacy_const_generics(1)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_extract_epi8<const IMM8: i32>(a: __m128i) -> i32 {
    static_assert_uimm_bits!(IMM8, 4);
    simd_extract::<_, u8>(a.as_u8x16(), IMM8 as u32) as i32
}

/// 从 `IMM8` 选择的 `a` 中提取 32 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_extract_epi32)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(
    all(test, not(target_os = "windows")),
    assert_instr(extractps, IMM8 = 1)
)]
#[rustc_legacy_const_generics(1)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_extract_epi32<const IMM8: i32>(a: __m128i) -> i32 {
    static_assert_uimm_bits!(IMM8, 2);
    simd_extract::<_, i32>(a.as_i32x4(), IMM8 as u32)
}

/// 在 `a` 中选择单个值存储在 `b` 中的某个位置，然后根据 `IMM8` 将元素归零。
///
/// `IMM8` 指定操作数 `a` 中的哪些位将被复制，它们将复制到结果中的哪些位，以及结果中的哪些位将被清除。
/// 进行以下分配：
///
/// * `[7:6]` 位指定要从操作数 `a` 复制的位：
///     - `00`: 从操作数 `a` 中选择位 `[31:0]`。
///     - `01`: 从操作数 `a` 中选择位 `[63:32]`。
///     - `10`: 从操作数 `a` 中选择位 `[95:64]`。
///     - `11`: 从操作数 `a` 中选择位 `[127:96]`。
///
/// * `[5:4]` 位指定结果中的位，将从操作数 `a` 中选择的位复制到该位：
///     - `00`: 将选定的位从 `a` 复制到结果位 `[31:0]`。
///     - `01`: 将选定的位从 `a` 复制到结果位 `[63:32]`。
///     - `10`: 将选定的位从 `a` 复制到结果位 `[95:64]`。
///     - `11`: 将选定的位从 `a` 复制到结果位 `[127:96]`。
///
/// * `[3:0]` 位：如果这些位中的任何一位被置位，则相应的结果元素将被清除。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_insert_ps)
///
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(insertps, IMM8 = 0b1010))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_insert_ps<const IMM8: i32>(a: __m128, b: __m128) -> __m128 {
    static_assert_uimm_bits!(IMM8, 8);
    insertps(a, b, IMM8 as u8)
}

/// 返回 `a` 的副本，其中 `i` 的 8 位整数插入到 `IMM8` 指定的位置。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_insert_epi8)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pinsrb, IMM8 = 0))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_insert_epi8<const IMM8: i32>(a: __m128i, i: i32) -> __m128i {
    static_assert_uimm_bits!(IMM8, 4);
    transmute(simd_insert(a.as_i8x16(), IMM8 as u32, i as i8))
}

/// 返回 `a` 的副本，其中 `i` 的 32 位整数插入到 `IMM8` 指定的位置。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_insert_epi32)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pinsrd, IMM8 = 0))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_insert_epi32<const IMM8: i32>(a: __m128i, i: i32) -> __m128i {
    static_assert_uimm_bits!(IMM8, 2);
    transmute(simd_insert(a.as_i32x4(), IMM8 as u32, i))
}

/// 比较 `a` 和 `b` 中的包装的 8 位整数，并以 dst 返回包装的最大值。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_max_epi8)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmaxsb))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_max_epi8(a: __m128i, b: __m128i) -> __m128i {
    let a = a.as_i8x16();
    let b = b.as_i8x16();
    transmute(simd_select::<i8x16, _>(simd_gt(a, b), a, b))
}

/// 比较 `a` 和 `b` 中包装的无符号 16 位整数，并返回包装的最大值。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_max_epu16)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmaxuw))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_max_epu16(a: __m128i, b: __m128i) -> __m128i {
    let a = a.as_u16x8();
    let b = b.as_u16x8();
    transmute(simd_select::<i16x8, _>(simd_gt(a, b), a, b))
}

/// 比较 `a` 和 `b` 中的包装的 32 位整数，并返回包装的最大值。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_max_epi32)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmaxsd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_max_epi32(a: __m128i, b: __m128i) -> __m128i {
    let a = a.as_i32x4();
    let b = b.as_i32x4();
    transmute(simd_select::<i32x4, _>(simd_gt(a, b), a, b))
}

/// 比较 `a` 和 `b` 中的包装无符号 32 位整数，并返回包装最大值。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_max_epu32)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmaxud))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_max_epu32(a: __m128i, b: __m128i) -> __m128i {
    let a = a.as_u32x4();
    let b = b.as_u32x4();
    transmute(simd_select::<i32x4, _>(simd_gt(a, b), a, b))
}

/// 比较 `a` 和 `b` 中的包装的 8 位整数，并以 dst 返回包装的最小值。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_min_epi8)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pminsb))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_min_epi8(a: __m128i, b: __m128i) -> __m128i {
    let a = a.as_i8x16();
    let b = b.as_i8x16();
    transmute(simd_select::<i8x16, _>(simd_lt(a, b), a, b))
}

/// 比较 `a` 和 `b` 中的包装的无符号 16 位整数，并返回包装的最小值。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_min_epu16)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pminuw))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_min_epu16(a: __m128i, b: __m128i) -> __m128i {
    let a = a.as_u16x8();
    let b = b.as_u16x8();
    transmute(simd_select::<i16x8, _>(simd_lt(a, b), a, b))
}

/// 比较 `a` 和 `b` 中的包装的 32 位整数，并返回包装的最小值。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_min_epi32)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pminsd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_min_epi32(a: __m128i, b: __m128i) -> __m128i {
    let a = a.as_i32x4();
    let b = b.as_i32x4();
    transmute(simd_select::<i32x4, _>(simd_lt(a, b), a, b))
}

/// 比较 `a` 和 `b` 中的包装的无符号 32 位整数，并返回包装的最小值。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_min_epu32)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pminud))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_min_epu32(a: __m128i, b: __m128i) -> __m128i {
    let a = a.as_u32x4();
    let b = b.as_u32x4();
    transmute(simd_select::<i32x4, _>(simd_lt(a, b), a, b))
}

/// 使用无符号饱和度将包装的 32 位整数从 `a` 和 `b` 转换为包装的 16 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_packus_epi32)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(packusdw))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_packus_epi32(a: __m128i, b: __m128i) -> __m128i {
    transmute(packusdw(a.as_i32x4(), b.as_i32x4()))
}

/// 比较 `a` 和 `b` 中的包装 64 位整数是否相等
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cmpeq_epi64)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pcmpeqq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cmpeq_epi64(a: __m128i, b: __m128i) -> __m128i {
    transmute(simd_eq::<_, i64x2>(a.as_i64x2(), b.as_i64x2()))
}

/// 符号将 `a` 中的包装 8 位整数扩展为包装 16 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi8_epi16)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovsxbw))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepi8_epi16(a: __m128i) -> __m128i {
    let a = a.as_i8x16();
    let a: i8x8 = simd_shuffle!(a, a, [0, 1, 2, 3, 4, 5, 6, 7]);
    transmute(simd_cast::<_, i16x8>(a))
}

/// 符号将 `a` 中的包装 8 位整数扩展为包装 32 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi8_epi32)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovsxbd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepi8_epi32(a: __m128i) -> __m128i {
    let a = a.as_i8x16();
    let a: i8x4 = simd_shuffle!(a, a, [0, 1, 2, 3]);
    transmute(simd_cast::<_, i32x4>(a))
}

/// 符号将 `a` 的低 8 字节中的包装的 8 位整数扩展为包装的
/// 64 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi8_epi64)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovsxbq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepi8_epi64(a: __m128i) -> __m128i {
    let a = a.as_i8x16();
    let a: i8x2 = simd_shuffle!(a, a, [0, 1]);
    transmute(simd_cast::<_, i64x2>(a))
}

/// 符号将 `a` 中的包装的 16 位整数扩展为包装的 32 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi16_epi32)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovsxwd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepi16_epi32(a: __m128i) -> __m128i {
    let a = a.as_i16x8();
    let a: i16x4 = simd_shuffle!(a, a, [0, 1, 2, 3]);
    transmute(simd_cast::<_, i32x4>(a))
}

/// 符号将 `a` 中的包装 16 位整数扩展为包装 64 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi16_epi64)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovsxwq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepi16_epi64(a: __m128i) -> __m128i {
    let a = a.as_i16x8();
    let a: i16x2 = simd_shuffle!(a, a, [0, 1]);
    transmute(simd_cast::<_, i64x2>(a))
}

/// 符号将 `a` 中的包装 32 位整数扩展为包装 64 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi32_epi64)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovsxdq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepi32_epi64(a: __m128i) -> __m128i {
    let a = a.as_i32x4();
    let a: i32x2 = simd_shuffle!(a, a, [0, 1]);
    transmute(simd_cast::<_, i64x2>(a))
}

/// 零将 `a` 中的包装的无符号 8 位整数扩展为包装的 16 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu8_epi16)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovzxbw))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepu8_epi16(a: __m128i) -> __m128i {
    let a = a.as_u8x16();
    let a: u8x8 = simd_shuffle!(a, a, [0, 1, 2, 3, 4, 5, 6, 7]);
    transmute(simd_cast::<_, i16x8>(a))
}

/// 零将 `a` 中的包装无符号 8 位整数扩展为包装 32 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu8_epi32)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovzxbd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepu8_epi32(a: __m128i) -> __m128i {
    let a = a.as_u8x16();
    let a: u8x4 = simd_shuffle!(a, a, [0, 1, 2, 3]);
    transmute(simd_cast::<_, i32x4>(a))
}

/// 零将 `a` 中的包装无符号 8 位整数扩展为包装 64 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu8_epi64)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovzxbq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepu8_epi64(a: __m128i) -> __m128i {
    let a = a.as_u8x16();
    let a: u8x2 = simd_shuffle!(a, a, [0, 1]);
    transmute(simd_cast::<_, i64x2>(a))
}

/// 零将 `a` 中的包装的无符号 16 位整数扩展为包装的 32 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu16_epi32)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovzxwd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepu16_epi32(a: __m128i) -> __m128i {
    let a = a.as_u16x8();
    let a: u16x4 = simd_shuffle!(a, a, [0, 1, 2, 3]);
    transmute(simd_cast::<_, i32x4>(a))
}

/// 零将 `a` 中的包装的无符号 16 位整数扩展为包装的 64 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu16_epi64)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovzxwq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepu16_epi64(a: __m128i) -> __m128i {
    let a = a.as_u16x8();
    let a: u16x2 = simd_shuffle!(a, a, [0, 1]);
    transmute(simd_cast::<_, i64x2>(a))
}

/// 零将 `a` 中的包装无符号 32 位整数扩展为包装 64 位整数
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu32_epi64)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmovzxdq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_cvtepu32_epi64(a: __m128i) -> __m128i {
    let a = a.as_u32x4();
    let a: u32x2 = simd_shuffle!(a, a, [0, 1]);
    transmute(simd_cast::<_, i64x2>(a))
}

/// 返回两个 __m128d vectors 的点积。
///
/// `IMM8[1:0]` 是广播掩码，`IMM8[5:4]` 是条件掩码。
/// 如果条件掩码位为零，则将相应的乘法替换为 `0.0` 的值。
/// 如果广播掩码位为 1，则点积的结果将存储在返回值组件中。
/// 否则，如果广播掩码位为零，则返回分量将为零。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_dp_pd)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(dppd, IMM8 = 0))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_dp_pd<const IMM8: i32>(a: __m128d, b: __m128d) -> __m128d {
    static_assert_uimm_bits!(IMM8, 8);
    dppd(a, b, IMM8 as u8)
}

/// 返回两个 __m128 vectors 的点积。
///
/// `IMM8[3:0]` 是广播掩码，`IMM8[7:4]` 是条件掩码。
/// 如果条件掩码位为零，则将相应的乘法替换为 `0.0` 的值。
/// 如果广播掩码位为 1，则点积的结果将存储在返回值组件中。
/// 否则，如果广播掩码位为零，则返回分量将为零。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_dp_ps)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(dpps, IMM8 = 0))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_dp_ps<const IMM8: i32>(a: __m128, b: __m128) -> __m128 {
    static_assert_uimm_bits!(IMM8, 8);
    dpps(a, b, IMM8 as u8)
}

/// 将 `a` 中的包装的双精度 (64-bit) 浮点元素四舍五入为整数，并将结果存储为包装的双精度浮点元素。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_floor_pd)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundpd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_floor_pd(a: __m128d) -> __m128d {
    simd_floor(a)
}

/// 将 `a` 中的包装的单精度 (32-bit) 浮点元素四舍五入为整数，并将结果存储为包装的单精度浮点元素。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_floor_ps)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundps))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_floor_ps(a: __m128) -> __m128 {
    simd_floor(a)
}

/// 将 `b` 中的下部双精度 (64-bit) 浮点元素四舍五入为整数值，将结果作为双精度浮点元素存储在内部结果的下部元素中，并将上部元素从 `a` 复制到上部内在结果的元素。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_floor_sd)
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundsd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_floor_sd(a: __m128d, b: __m128d) -> __m128d {
    roundsd(a, b, _MM_FROUND_FLOOR)
}

/// 将 `b` 中的下部单精度 (32-bit) 浮点元素四舍五入为整数值，将结果作为单精度浮点元素存储在内部结果的下部元素中，然后将 `a` 的上部 3 个包装的元素复制到内在结果的上层要素。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_floor_ss)
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundss))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_floor_ss(a: __m128, b: __m128) -> __m128 {
    roundss(a, b, _MM_FROUND_FLOOR)
}

/// 将 `a` 中的包装的双精度 (64-bit) 浮点元素四舍五入为整数，并将结果存储为包装的双精度浮点元素。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_ceil_pd)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundpd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_ceil_pd(a: __m128d) -> __m128d {
    simd_ceil(a)
}

/// 将 `a` 中的包装的单精度 (32-bit) 浮点元素四舍五入为整数，并将结果存储为包装的单精度浮点元素。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_ceil_ps)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundps))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_ceil_ps(a: __m128) -> __m128 {
    simd_ceil(a)
}

/// 将 `b` 中的低位双精度 (64-bit) 浮点型元素四舍五入为整数，将结果作为双精度浮点型元素存入内部函数结果的低位元素，并将 `a` 中的高位元素复制到内部函数结果的上部元素。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_ceil_sd)
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundsd))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_ceil_sd(a: __m128d, b: __m128d) -> __m128d {
    roundsd(a, b, _MM_FROUND_CEIL)
}

/// 将 `b` 中的下部单精度 (32-bit) 浮点元素四舍五入为整数，将结果作为单精度浮点元素存储在内部结果的下部元素中，并将高 3 个包装的元素从 `a` 复制到内在结果的上层要素。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_ceil_ss)
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundss))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_ceil_ss(a: __m128, b: __m128) -> __m128 {
    roundss(a, b, _MM_FROUND_CEIL)
}

/// 使用 `ROUNDING` 参数将 `a` 中的包装的双精度 (64-bit) 浮点元素舍入，并将结果存储为包装的双精度浮点元素。
///
/// 根据舍入参数进行舍入，该参数可以是以下之一：
///
/// ```
/// #[cfg(target_arch = "x86")]
/// use std::arch::x86::*;
/// #[cfg(target_arch = "x86_64")]
/// use std::arch::x86_64::*;
///
/// # fn main() {
/// // 四舍五入到最接近的值，并排除异常：
/// # let _x =
/// _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC;
/// // 四舍五入并抑制异常：
/// # let _x =
/// _MM_FROUND_TO_NEG_INF | _MM_FROUND_NO_EXC;
/// // 汇总并排除异常：
/// # let _x =
/// _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC;
/// // 截断并抑制异常：
/// # let _x =
/// _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC;
/// // 使用 MXCSR.RC； 参见 `_MM_SET_ROUNDING_MODE`：
/// # let _x =
/// _MM_FROUND_CUR_DIRECTION;
/// # }
/// ```
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_round_pd)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundpd, ROUNDING = 0))]
#[rustc_legacy_const_generics(1)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_round_pd<const ROUNDING: i32>(a: __m128d) -> __m128d {
    static_assert_uimm_bits!(ROUNDING, 4);
    roundpd(a, ROUNDING)
}

/// 使用 `ROUNDING` 参数将 `a` 中的包装的单精度 (32-bit) 浮点元素舍入，并将结果存储为包装的单精度浮点元素。
///
/// 根据舍入参数进行舍入，该参数可以是以下之一：
///
/// ```
/// #[cfg(target_arch = "x86")]
/// use std::arch::x86::*;
/// #[cfg(target_arch = "x86_64")]
/// use std::arch::x86_64::*;
///
/// # fn main() {
/// // 四舍五入到最接近的值，并排除异常：
/// # let _x =
/// _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC;
/// // 四舍五入并抑制异常：
/// # let _x =
/// _MM_FROUND_TO_NEG_INF | _MM_FROUND_NO_EXC;
/// // 汇总并排除异常：
/// # let _x =
/// _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC;
/// // 截断并抑制异常：
/// # let _x =
/// _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC;
/// // 使用 MXCSR.RC； 参见 `_MM_SET_ROUNDING_MODE`：
/// # let _x =
/// _MM_FROUND_CUR_DIRECTION;
/// # }
/// ```
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_round_ps)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundps, ROUNDING = 0))]
#[rustc_legacy_const_generics(1)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_round_ps<const ROUNDING: i32>(a: __m128) -> __m128 {
    static_assert_uimm_bits!(ROUNDING, 4);
    roundps(a, ROUNDING)
}

/// 使用 `ROUNDING` 参数将 `b` 中的低位双精度浮点元素 (64-bit) 舍入，将结果作为双精度浮点元素存储在内部结果的低位元素中，并将高位元素从 `a` 复制到高位元素的内在结果。
///
/// 根据舍入参数进行舍入，该参数可以是以下之一：
///
/// ```
/// #[cfg(target_arch = "x86")]
/// use std::arch::x86::*;
/// #[cfg(target_arch = "x86_64")]
/// use std::arch::x86_64::*;
///
/// # fn main() {
/// // 四舍五入到最接近的值，并排除异常：
/// # let _x =
/// _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC;
/// // 四舍五入并抑制异常：
/// # let _x =
/// _MM_FROUND_TO_NEG_INF | _MM_FROUND_NO_EXC;
/// // 汇总并排除异常：
/// # let _x =
/// _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC;
/// // 截断并抑制异常：
/// # let _x =
/// _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC;
/// // 使用 MXCSR.RC； 参见 `_MM_SET_ROUNDING_MODE`：
/// # let _x =
/// _MM_FROUND_CUR_DIRECTION;
/// # }
/// ```
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_round_sd)
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundsd, ROUNDING = 0))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_round_sd<const ROUNDING: i32>(a: __m128d, b: __m128d) -> __m128d {
    static_assert_uimm_bits!(ROUNDING, 4);
    roundsd(a, b, ROUNDING)
}

/// 使用 `ROUNDING` 参数对 `b` 中的低位单精度 (32-bit) 浮点元素进行舍入，将结果作为单精度浮点元素存储在内部函数结果的低位元素中，并从 `a` 复制高位 3 个包装元素到内部函数结果的上层元素。
///
/// 根据舍入参数进行舍入，该参数可以是以下之一：
///
/// ```
/// #[cfg(target_arch = "x86")]
/// use std::arch::x86::*;
/// #[cfg(target_arch = "x86_64")]
/// use std::arch::x86_64::*;
///
/// # fn main() {
/// // 四舍五入到最接近的值，并排除异常：
/// # let _x =
/// _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC;
/// // 四舍五入并抑制异常：
/// # let _x =
/// _MM_FROUND_TO_NEG_INF | _MM_FROUND_NO_EXC;
/// // 汇总并排除异常：
/// # let _x =
/// _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC;
/// // 截断并抑制异常：
/// # let _x =
/// _MM_FROUND_TO_ZERO | _MM_FROUND_NO_EXC;
/// // 使用 MXCSR.RC； 参见 `_MM_SET_ROUNDING_MODE`：
/// # let _x =
/// _MM_FROUND_CUR_DIRECTION;
/// # }
/// ```
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_round_ss)
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(roundss, ROUNDING = 0))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_round_ss<const ROUNDING: i32>(a: __m128, b: __m128) -> __m128 {
    static_assert_uimm_bits!(ROUNDING, 4);
    roundss(a, b, ROUNDING)
}

/// 在 128 位 __m128i vector 中查找最小的无符号 16 位元素，并返回一个 vector，其中在第一个位置包含其值，在第二个位置包含其索引； 所有其他元素都设置为零。
///
///
/// 该内部函数对应于 `VPHMINPOSUW`/`PHMINPOSUW` 指令。
///
/// Arguments:
///
/// * `a`-`__m128i` 类型的 128 位 vector。
///
/// Returns:
///
/// 128 位值，其中：
///
/// * `[15:0]` 位 - 包含在参数 `a` 中找到的最小值，
/// * `[18:16]` 位 - 包含最小值的索引
/// * 其余位设置为 `0`。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_minpos_epu16)
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(phminposuw))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_minpos_epu16(a: __m128i) -> __m128i {
    transmute(phminposuw(a.as_u16x8()))
}

/// 将 `a` 和 `b` 中每个包装的 64 位元素的低 32 位整数相乘，并返回带符号的 64 位结果。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mul_epi32)
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmuldq))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_mul_epi32(a: __m128i, b: __m128i) -> __m128i {
    transmute(pmuldq(a.as_i32x4(), b.as_i32x4()))
}

/// 将 `a` 和 `b` 中包装的 32 位整数相乘，产生中间值
/// 64 位整数，并返回最低的 32 位 (无论它们是什么)，都将重新解释为有符号整数。
/// 虽然 `pmulld __m128i::splat(2), __m128i::splat(2)` 返回明显的 `__m128i::splat(4)`，但由于包装算术，`pmulld __m128i::splat(i32::MAX), __m128i::splat(2)` 将返回负数。
///
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mullo_epi32)
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pmulld))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_mullo_epi32(a: __m128i, b: __m128i) -> __m128i {
    transmute(simd_mul(a.as_i32x4(), b.as_i32x4()))
}

/// 减去 8 位无符号整数值，然后将差的绝对值计算为目标中的相应位。
///
/// 然后，根据立即操作数中的位字段返回绝对差之和。
///
/// 执行以下算法：
///
/// ```ignore
/// i = IMM8[2] * 4
/// j = IMM8[1:0] * 4
/// for k := 0 to 7
///     d0 = abs(a[i + k + 0] - b[j + 0])
///     d1 = abs(a[i + k + 1] - b[j + 1])
///     d2 = abs(a[i + k + 2] - b[j + 2])
///     d3 = abs(a[i + k + 3] - b[j + 3])
///     r[k] = d0 + d1 + d2 + d3
/// ```
///
/// Arguments:
///
/// * `a`-`__m128i` 类型的 128 位 vector。
/// * `b`-`__m128i` 类型的 128 位 vector。
/// * `IMM8` - 一个 8 位 immediate 操作数，指定如何计算绝对差
///     * `[2]` 位指定操作数 `a` 的偏移量
///     * `[1:0]` 位指定操作数 `b` 的偏移量
///
/// Returns:
///
/// * `__m128i` vector 包含两个操作数之间的绝对差集的总和。
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mpsadbw_epu8)
///
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(mpsadbw, IMM8 = 0))]
#[rustc_legacy_const_generics(2)]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_mpsadbw_epu8<const IMM8: i32>(a: __m128i, b: __m128i) -> __m128i {
    static_assert_uimm_bits!(IMM8, 3);
    transmute(mpsadbw(a.as_u8x16(), b.as_u8x16(), IMM8 as u8))
}

/// 测试 128 位整数 vector 中的指定位是否全部为零。
///
/// Arguments:
///
/// * `a` - 包含要测试的位的 128 位整数 vector。
/// * `mask` - 一个 128 位整数 vector，用于选择要在操作数 `a` 中测试的位。
///
/// Returns:
///
/// * `1` - 如果指定的位都是零，
/// * `0` - otherwise.
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_testz_si128)
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(ptest))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_testz_si128(a: __m128i, mask: __m128i) -> i32 {
    ptestz(a.as_i64x2(), mask.as_i64x2())
}

/// 测试 128 位整数 vector 中的指定位是否全部为 1。
///
/// Arguments:
///
/// * `a` - 包含要测试的位的 128 位整数 vector。
/// * `mask` - 一个 128 位整数 vector，用于选择要在操作数 `a` 中测试的位。
///
/// Returns:
///
/// * `1` - 如果指定的位都是 1，
/// * `0` - otherwise.
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_testc_si128)
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(ptest))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_testc_si128(a: __m128i, mask: __m128i) -> i32 {
    ptestc(a.as_i64x2(), mask.as_i64x2())
}

/// 测试 128 位整数 vector 中的指定位是否既不是全零也不是全 1。
///
/// Arguments:
///
/// * `a` - 包含要测试的位的 128 位整数 vector。
/// * `mask` - 一个 128 位整数 vector，用于选择要在操作数 `a` 中测试的位。
///
/// Returns:
///
/// * `1` - 如果指定的位既不是全零也不是全 1，
/// * `0` - otherwise.
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_testnzc_si128)
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(ptest))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_testnzc_si128(a: __m128i, mask: __m128i) -> i32 {
    ptestnzc(a.as_i64x2(), mask.as_i64x2())
}

/// 测试 128 位整数 vector 中的指定位是否全部为零。
///
/// Arguments:
///
/// * `a` - 包含要测试的位的 128 位整数 vector。
/// * `mask` - 一个 128 位整数 vector，用于选择要在操作数 `a` 中测试的位。
///
/// Returns:
///
/// * `1` - 如果指定的位都是零，
/// * `0` - otherwise.
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_test_all_zeros)
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(ptest))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_test_all_zeros(a: __m128i, mask: __m128i) -> i32 {
    _mm_testz_si128(a, mask)
}

/// 测试 `a` 128 位整数 vector 中的指定位是否全部为 1。
///
/// Argument:
///
/// * `a` - 包含要测试的位的 128 位整数 vector。
///
/// Returns:
///
/// * `1` - 如果操作数中指定的位都设置为 1，
/// * `0` - otherwise.
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_test_all_ones)
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(pcmpeqd))]
#[cfg_attr(test, assert_instr(ptest))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_test_all_ones(a: __m128i) -> i32 {
    _mm_testc_si128(a, _mm_cmpeq_epi32(a, a))
}

/// 测试 128 位整数 vector 中的指定位是否既不是全零也不是全 1。
///
/// Arguments:
///
/// * `a` - 包含要测试的位的 128 位整数 vector。
/// * `mask` - 一个 128 位整数 vector，用于选择要在操作数 `a` 中测试的位。
///
/// Returns:
///
/// * `1` - 如果指定的位既不是全零也不是全 1，
/// * `0` - otherwise.
///
/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_test_mix_ones_zeros)
///
///
#[inline]
#[target_feature(enable = "sse4.1")]
#[cfg_attr(test, assert_instr(ptest))]
#[stable(feature = "simd_x86", since = "1.27.0")]
pub unsafe fn _mm_test_mix_ones_zeros(a: __m128i, mask: __m128i) -> i32 {
    _mm_testnzc_si128(a, mask)
}

#[allow(improper_ctypes)]
extern "C" {
    #[link_name = "llvm.x86.sse41.pblendvb"]
    fn pblendvb(a: i8x16, b: i8x16, mask: i8x16) -> i8x16;
    #[link_name = "llvm.x86.sse41.blendvpd"]
    fn blendvpd(a: __m128d, b: __m128d, mask: __m128d) -> __m128d;
    #[link_name = "llvm.x86.sse41.blendvps"]
    fn blendvps(a: __m128, b: __m128, mask: __m128) -> __m128;
    #[link_name = "llvm.x86.sse41.blendpd"]
    fn blendpd(a: __m128d, b: __m128d, imm2: u8) -> __m128d;
    #[link_name = "llvm.x86.sse41.blendps"]
    fn blendps(a: __m128, b: __m128, imm4: u8) -> __m128;
    #[link_name = "llvm.x86.sse41.pblendw"]
    fn pblendw(a: i16x8, b: i16x8, imm8: u8) -> i16x8;
    #[link_name = "llvm.x86.sse41.insertps"]
    fn insertps(a: __m128, b: __m128, imm8: u8) -> __m128;
    #[link_name = "llvm.x86.sse41.packusdw"]
    fn packusdw(a: i32x4, b: i32x4) -> u16x8;
    #[link_name = "llvm.x86.sse41.dppd"]
    fn dppd(a: __m128d, b: __m128d, imm8: u8) -> __m128d;
    #[link_name = "llvm.x86.sse41.dpps"]
    fn dpps(a: __m128, b: __m128, imm8: u8) -> __m128;
    #[link_name = "llvm.x86.sse41.round.pd"]
    fn roundpd(a: __m128d, rounding: i32) -> __m128d;
    #[link_name = "llvm.x86.sse41.round.ps"]
    fn roundps(a: __m128, rounding: i32) -> __m128;
    #[link_name = "llvm.x86.sse41.round.sd"]
    fn roundsd(a: __m128d, b: __m128d, rounding: i32) -> __m128d;
    #[link_name = "llvm.x86.sse41.round.ss"]
    fn roundss(a: __m128, b: __m128, rounding: i32) -> __m128;
    #[link_name = "llvm.x86.sse41.phminposuw"]
    fn phminposuw(a: u16x8) -> u16x8;
    #[link_name = "llvm.x86.sse41.pmuldq"]
    fn pmuldq(a: i32x4, b: i32x4) -> i64x2;
    #[link_name = "llvm.x86.sse41.mpsadbw"]
    fn mpsadbw(a: u8x16, b: u8x16, imm8: u8) -> u16x8;
    #[link_name = "llvm.x86.sse41.ptestz"]
    fn ptestz(a: i64x2, mask: i64x2) -> i32;
    #[link_name = "llvm.x86.sse41.ptestc"]
    fn ptestc(a: i64x2, mask: i64x2) -> i32;
    #[link_name = "llvm.x86.sse41.ptestnzc"]
    fn ptestnzc(a: i64x2, mask: i64x2) -> i32;
}

#[cfg(test)]
mod tests {
    use crate::core_arch::x86::*;
    use std::mem;
    use stdarch_test::simd_test;

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_blendv_epi8() {
        #[rustfmt::skip]
        let a = _mm_setr_epi8(
            0, 1, 2, 3, 4, 5, 6, 7,
            8, 9, 10, 11, 12, 13, 14, 15,
        );
        #[rustfmt::skip]
        let b = _mm_setr_epi8(
            16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
        );
        #[rustfmt::skip]
        let mask = _mm_setr_epi8(
            0, -1, 0, -1, 0, -1, 0, -1,
            0, -1, 0, -1, 0, -1, 0, -1,
        );
        #[rustfmt::skip]
        let e = _mm_setr_epi8(
            0, 17, 2, 19, 4, 21, 6, 23, 8, 25, 10, 27, 12, 29, 14, 31,
        );
        assert_eq_m128i(_mm_blendv_epi8(a, b, mask), e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_blendv_pd() {
        let a = _mm_set1_pd(0.0);
        let b = _mm_set1_pd(1.0);
        let mask = transmute(_mm_setr_epi64x(0, -1));
        let r = _mm_blendv_pd(a, b, mask);
        let e = _mm_setr_pd(0.0, 1.0);
        assert_eq_m128d(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_blendv_ps() {
        let a = _mm_set1_ps(0.0);
        let b = _mm_set1_ps(1.0);
        let mask = transmute(_mm_setr_epi32(0, -1, 0, -1));
        let r = _mm_blendv_ps(a, b, mask);
        let e = _mm_setr_ps(0.0, 1.0, 0.0, 1.0);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_blend_pd() {
        let a = _mm_set1_pd(0.0);
        let b = _mm_set1_pd(1.0);
        let r = _mm_blend_pd::<0b10>(a, b);
        let e = _mm_setr_pd(0.0, 1.0);
        assert_eq_m128d(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_blend_ps() {
        let a = _mm_set1_ps(0.0);
        let b = _mm_set1_ps(1.0);
        let r = _mm_blend_ps::<0b1010>(a, b);
        let e = _mm_setr_ps(0.0, 1.0, 0.0, 1.0);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_blend_epi16() {
        let a = _mm_set1_epi16(0);
        let b = _mm_set1_epi16(1);
        let r = _mm_blend_epi16::<0b1010_1100>(a, b);
        let e = _mm_setr_epi16(0, 0, 1, 1, 0, 1, 0, 1);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_extract_ps() {
        let a = _mm_setr_ps(0.0, 1.0, 2.0, 3.0);
        let r: f32 = transmute(_mm_extract_ps::<1>(a));
        assert_eq!(r, 1.0);
        let r: f32 = transmute(_mm_extract_ps::<3>(a));
        assert_eq!(r, 3.0);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_extract_epi8() {
        #[rustfmt::skip]
        let a = _mm_setr_epi8(
            -1, 1, 2, 3, 4, 5, 6, 7,
            8, 9, 10, 11, 12, 13, 14, 15
        );
        let r1 = _mm_extract_epi8::<0>(a);
        let r2 = _mm_extract_epi8::<3>(a);
        assert_eq!(r1, 0xFF);
        assert_eq!(r2, 3);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_extract_epi32() {
        let a = _mm_setr_epi32(0, 1, 2, 3);
        let r = _mm_extract_epi32::<1>(a);
        assert_eq!(r, 1);
        let r = _mm_extract_epi32::<3>(a);
        assert_eq!(r, 3);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_insert_ps() {
        let a = _mm_set1_ps(1.0);
        let b = _mm_setr_ps(1.0, 2.0, 3.0, 4.0);
        let r = _mm_insert_ps::<0b11_00_1100>(a, b);
        let e = _mm_setr_ps(4.0, 1.0, 0.0, 0.0);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_insert_epi8() {
        let a = _mm_set1_epi8(0);
        let e = _mm_setr_epi8(0, 32, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
        let r = _mm_insert_epi8::<1>(a, 32);
        assert_eq_m128i(r, e);
        let e = _mm_setr_epi8(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 32, 0);
        let r = _mm_insert_epi8::<14>(a, 32);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_insert_epi32() {
        let a = _mm_set1_epi32(0);
        let e = _mm_setr_epi32(0, 32, 0, 0);
        let r = _mm_insert_epi32::<1>(a, 32);
        assert_eq_m128i(r, e);
        let e = _mm_setr_epi32(0, 0, 0, 32);
        let r = _mm_insert_epi32::<3>(a, 32);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_max_epi8() {
        #[rustfmt::skip]
        let a = _mm_setr_epi8(
            1, 4, 5, 8, 9, 12, 13, 16,
            17, 20, 21, 24, 25, 28, 29, 32,
        );
        #[rustfmt::skip]
        let b = _mm_setr_epi8(
            2, 3, 6, 7, 10, 11, 14, 15,
            18, 19, 22, 23, 26, 27, 30, 31,
        );
        let r = _mm_max_epi8(a, b);
        #[rustfmt::skip]
        let e = _mm_setr_epi8(
            2, 4, 6, 8, 10, 12, 14, 16,
            18, 20, 22, 24, 26, 28, 30, 32,
        );
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_max_epu16() {
        let a = _mm_setr_epi16(1, 4, 5, 8, 9, 12, 13, 16);
        let b = _mm_setr_epi16(2, 3, 6, 7, 10, 11, 14, 15);
        let r = _mm_max_epu16(a, b);
        let e = _mm_setr_epi16(2, 4, 6, 8, 10, 12, 14, 16);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_max_epi32() {
        let a = _mm_setr_epi32(1, 4, 5, 8);
        let b = _mm_setr_epi32(2, 3, 6, 7);
        let r = _mm_max_epi32(a, b);
        let e = _mm_setr_epi32(2, 4, 6, 8);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_max_epu32() {
        let a = _mm_setr_epi32(1, 4, 5, 8);
        let b = _mm_setr_epi32(2, 3, 6, 7);
        let r = _mm_max_epu32(a, b);
        let e = _mm_setr_epi32(2, 4, 6, 8);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_min_epi8_1() {
        #[rustfmt::skip]
        let a = _mm_setr_epi8(
            1, 4, 5, 8, 9, 12, 13, 16,
            17, 20, 21, 24, 25, 28, 29, 32,
        );
        #[rustfmt::skip]
        let b = _mm_setr_epi8(
            2, 3, 6, 7, 10, 11, 14, 15,
            18, 19, 22, 23, 26, 27, 30, 31,
        );
        let r = _mm_min_epi8(a, b);
        #[rustfmt::skip]
        let e = _mm_setr_epi8(
            1, 3, 5, 7, 9, 11, 13, 15,
            17, 19, 21, 23, 25, 27, 29, 31,
        );
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_min_epi8_2() {
        #[rustfmt::skip]
        let a = _mm_setr_epi8(
            1, -4, -5, 8, -9, -12, 13, -16,
            17, 20, 21, 24, 25, 28, 29, 32,
        );
        #[rustfmt::skip]
        let b = _mm_setr_epi8(
            2, -3, -6, 7, -10, -11, 14, -15,
            18, 19, 22, 23, 26, 27, 30, 31,
        );
        let r = _mm_min_epi8(a, b);
        #[rustfmt::skip]
        let e = _mm_setr_epi8(
            1, -4, -6, 7, -10, -12, 13, -16,
            17, 19, 21, 23, 25, 27, 29, 31,
        );
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_min_epu16() {
        let a = _mm_setr_epi16(1, 4, 5, 8, 9, 12, 13, 16);
        let b = _mm_setr_epi16(2, 3, 6, 7, 10, 11, 14, 15);
        let r = _mm_min_epu16(a, b);
        let e = _mm_setr_epi16(1, 3, 5, 7, 9, 11, 13, 15);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_min_epi32_1() {
        let a = _mm_setr_epi32(1, 4, 5, 8);
        let b = _mm_setr_epi32(2, 3, 6, 7);
        let r = _mm_min_epi32(a, b);
        let e = _mm_setr_epi32(1, 3, 5, 7);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_min_epi32_2() {
        let a = _mm_setr_epi32(-1, 4, 5, -7);
        let b = _mm_setr_epi32(-2, 3, -6, 8);
        let r = _mm_min_epi32(a, b);
        let e = _mm_setr_epi32(-2, 3, -6, -7);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_min_epu32() {
        let a = _mm_setr_epi32(1, 4, 5, 8);
        let b = _mm_setr_epi32(2, 3, 6, 7);
        let r = _mm_min_epu32(a, b);
        let e = _mm_setr_epi32(1, 3, 5, 7);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_packus_epi32() {
        let a = _mm_setr_epi32(1, 2, 3, 4);
        let b = _mm_setr_epi32(-1, -2, -3, -4);
        let r = _mm_packus_epi32(a, b);
        let e = _mm_setr_epi16(1, 2, 3, 4, 0, 0, 0, 0);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cmpeq_epi64() {
        let a = _mm_setr_epi64x(0, 1);
        let b = _mm_setr_epi64x(0, 0);
        let r = _mm_cmpeq_epi64(a, b);
        let e = _mm_setr_epi64x(-1, 0);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepi8_epi16() {
        let a = _mm_set1_epi8(10);
        let r = _mm_cvtepi8_epi16(a);
        let e = _mm_set1_epi16(10);
        assert_eq_m128i(r, e);
        let a = _mm_set1_epi8(-10);
        let r = _mm_cvtepi8_epi16(a);
        let e = _mm_set1_epi16(-10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepi8_epi32() {
        let a = _mm_set1_epi8(10);
        let r = _mm_cvtepi8_epi32(a);
        let e = _mm_set1_epi32(10);
        assert_eq_m128i(r, e);
        let a = _mm_set1_epi8(-10);
        let r = _mm_cvtepi8_epi32(a);
        let e = _mm_set1_epi32(-10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepi8_epi64() {
        let a = _mm_set1_epi8(10);
        let r = _mm_cvtepi8_epi64(a);
        let e = _mm_set1_epi64x(10);
        assert_eq_m128i(r, e);
        let a = _mm_set1_epi8(-10);
        let r = _mm_cvtepi8_epi64(a);
        let e = _mm_set1_epi64x(-10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepi16_epi32() {
        let a = _mm_set1_epi16(10);
        let r = _mm_cvtepi16_epi32(a);
        let e = _mm_set1_epi32(10);
        assert_eq_m128i(r, e);
        let a = _mm_set1_epi16(-10);
        let r = _mm_cvtepi16_epi32(a);
        let e = _mm_set1_epi32(-10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepi16_epi64() {
        let a = _mm_set1_epi16(10);
        let r = _mm_cvtepi16_epi64(a);
        let e = _mm_set1_epi64x(10);
        assert_eq_m128i(r, e);
        let a = _mm_set1_epi16(-10);
        let r = _mm_cvtepi16_epi64(a);
        let e = _mm_set1_epi64x(-10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepi32_epi64() {
        let a = _mm_set1_epi32(10);
        let r = _mm_cvtepi32_epi64(a);
        let e = _mm_set1_epi64x(10);
        assert_eq_m128i(r, e);
        let a = _mm_set1_epi32(-10);
        let r = _mm_cvtepi32_epi64(a);
        let e = _mm_set1_epi64x(-10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepu8_epi16() {
        let a = _mm_set1_epi8(10);
        let r = _mm_cvtepu8_epi16(a);
        let e = _mm_set1_epi16(10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepu8_epi32() {
        let a = _mm_set1_epi8(10);
        let r = _mm_cvtepu8_epi32(a);
        let e = _mm_set1_epi32(10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepu8_epi64() {
        let a = _mm_set1_epi8(10);
        let r = _mm_cvtepu8_epi64(a);
        let e = _mm_set1_epi64x(10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepu16_epi32() {
        let a = _mm_set1_epi16(10);
        let r = _mm_cvtepu16_epi32(a);
        let e = _mm_set1_epi32(10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepu16_epi64() {
        let a = _mm_set1_epi16(10);
        let r = _mm_cvtepu16_epi64(a);
        let e = _mm_set1_epi64x(10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_cvtepu32_epi64() {
        let a = _mm_set1_epi32(10);
        let r = _mm_cvtepu32_epi64(a);
        let e = _mm_set1_epi64x(10);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_dp_pd() {
        let a = _mm_setr_pd(2.0, 3.0);
        let b = _mm_setr_pd(1.0, 4.0);
        let e = _mm_setr_pd(14.0, 0.0);
        assert_eq_m128d(_mm_dp_pd::<0b00110001>(a, b), e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_dp_ps() {
        let a = _mm_setr_ps(2.0, 3.0, 1.0, 10.0);
        let b = _mm_setr_ps(1.0, 4.0, 0.5, 10.0);
        let e = _mm_setr_ps(14.5, 0.0, 14.5, 0.0);
        assert_eq_m128(_mm_dp_ps::<0b01110101>(a, b), e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_floor_pd() {
        let a = _mm_setr_pd(2.5, 4.5);
        let r = _mm_floor_pd(a);
        let e = _mm_setr_pd(2.0, 4.0);
        assert_eq_m128d(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_floor_ps() {
        let a = _mm_setr_ps(2.5, 4.5, 8.5, 16.5);
        let r = _mm_floor_ps(a);
        let e = _mm_setr_ps(2.0, 4.0, 8.0, 16.0);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_floor_sd() {
        let a = _mm_setr_pd(2.5, 4.5);
        let b = _mm_setr_pd(-1.5, -3.5);
        let r = _mm_floor_sd(a, b);
        let e = _mm_setr_pd(-2.0, 4.5);
        assert_eq_m128d(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_floor_ss() {
        let a = _mm_setr_ps(2.5, 4.5, 8.5, 16.5);
        let b = _mm_setr_ps(-1.5, -3.5, -7.5, -15.5);
        let r = _mm_floor_ss(a, b);
        let e = _mm_setr_ps(-2.0, 4.5, 8.5, 16.5);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_ceil_pd() {
        let a = _mm_setr_pd(1.5, 3.5);
        let r = _mm_ceil_pd(a);
        let e = _mm_setr_pd(2.0, 4.0);
        assert_eq_m128d(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_ceil_ps() {
        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
        let r = _mm_ceil_ps(a);
        let e = _mm_setr_ps(2.0, 4.0, 8.0, 16.0);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_ceil_sd() {
        let a = _mm_setr_pd(1.5, 3.5);
        let b = _mm_setr_pd(-2.5, -4.5);
        let r = _mm_ceil_sd(a, b);
        let e = _mm_setr_pd(-2.0, 3.5);
        assert_eq_m128d(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_ceil_ss() {
        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
        let b = _mm_setr_ps(-2.5, -4.5, -8.5, -16.5);
        let r = _mm_ceil_ss(a, b);
        let e = _mm_setr_ps(-2.0, 3.5, 7.5, 15.5);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_round_pd() {
        let a = _mm_setr_pd(1.25, 3.75);
        let r = _mm_round_pd::<_MM_FROUND_TO_NEAREST_INT>(a);
        let e = _mm_setr_pd(1.0, 4.0);
        assert_eq_m128d(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_round_ps() {
        let a = _mm_setr_ps(2.25, 4.75, -1.75, -4.25);
        let r = _mm_round_ps::<_MM_FROUND_TO_ZERO>(a);
        let e = _mm_setr_ps(2.0, 4.0, -1.0, -4.0);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_round_sd() {
        let a = _mm_setr_pd(1.5, 3.5);
        let b = _mm_setr_pd(-2.5, -4.5);
        let old_mode = _MM_GET_ROUNDING_MODE();
        _MM_SET_ROUNDING_MODE(_MM_ROUND_TOWARD_ZERO);
        let r = _mm_round_sd::<_MM_FROUND_CUR_DIRECTION>(a, b);
        _MM_SET_ROUNDING_MODE(old_mode);
        let e = _mm_setr_pd(-2.0, 3.5);
        assert_eq_m128d(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_round_ss() {
        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
        let b = _mm_setr_ps(-1.75, -4.5, -8.5, -16.5);
        let old_mode = _MM_GET_ROUNDING_MODE();
        _MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);
        let r = _mm_round_ss::<_MM_FROUND_CUR_DIRECTION>(a, b);
        _MM_SET_ROUNDING_MODE(old_mode);
        let e = _mm_setr_ps(-2.0, 3.5, 7.5, 15.5);
        assert_eq_m128(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_minpos_epu16_1() {
        let a = _mm_setr_epi16(23, 18, 44, 97, 50, 13, 67, 66);
        let r = _mm_minpos_epu16(a);
        let e = _mm_setr_epi16(13, 5, 0, 0, 0, 0, 0, 0);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_minpos_epu16_2() {
        let a = _mm_setr_epi16(0, 18, 44, 97, 50, 13, 67, 66);
        let r = _mm_minpos_epu16(a);
        let e = _mm_setr_epi16(0, 0, 0, 0, 0, 0, 0, 0);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_mul_epi32() {
        {
            let a = _mm_setr_epi32(1, 1, 1, 1);
            let b = _mm_setr_epi32(1, 2, 3, 4);
            let r = _mm_mul_epi32(a, b);
            let e = _mm_setr_epi64x(1, 3);
            assert_eq_m128i(r, e);
        }
        {
            let a = _mm_setr_epi32(15, 2 /* ignored */, 1234567, 4 /* ignored */);
            let b = _mm_setr_epi32(
                -20, -256, /* ignored */
                666666, 666666, /* ignored */
            );
            let r = _mm_mul_epi32(a, b);
            let e = _mm_setr_epi64x(-300, 823043843622);
            assert_eq_m128i(r, e);
        }
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_mullo_epi32() {
        {
            let a = _mm_setr_epi32(1, 1, 1, 1);
            let b = _mm_setr_epi32(1, 2, 3, 4);
            let r = _mm_mullo_epi32(a, b);
            let e = _mm_setr_epi32(1, 2, 3, 4);
            assert_eq_m128i(r, e);
        }
        {
            let a = _mm_setr_epi32(15, -2, 1234567, 99999);
            let b = _mm_setr_epi32(-20, -256, 666666, -99999);
            let r = _mm_mullo_epi32(a, b);
            // 注意，r[2] 中的最高有效位被视为符号位：
            //
            // 1234567 * 666666 = -1589877210
            let e = _mm_setr_epi32(-300, 512, -1589877210, -1409865409);
            assert_eq_m128i(r, e);
        }
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_minpos_epu16() {
        let a = _mm_setr_epi16(8, 7, 6, 5, 4, 1, 2, 3);
        let r = _mm_minpos_epu16(a);
        let e = _mm_setr_epi16(1, 5, 0, 0, 0, 0, 0, 0);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_mpsadbw_epu8() {
        #[rustfmt::skip]
        let a = _mm_setr_epi8(
            0, 1, 2, 3, 4, 5, 6, 7,
            8, 9, 10, 11, 12, 13, 14, 15,
        );

        let r = _mm_mpsadbw_epu8::<0b000>(a, a);
        let e = _mm_setr_epi16(0, 4, 8, 12, 16, 20, 24, 28);
        assert_eq_m128i(r, e);

        let r = _mm_mpsadbw_epu8::<0b001>(a, a);
        let e = _mm_setr_epi16(16, 12, 8, 4, 0, 4, 8, 12);
        assert_eq_m128i(r, e);

        let r = _mm_mpsadbw_epu8::<0b100>(a, a);
        let e = _mm_setr_epi16(16, 20, 24, 28, 32, 36, 40, 44);
        assert_eq_m128i(r, e);

        let r = _mm_mpsadbw_epu8::<0b101>(a, a);
        let e = _mm_setr_epi16(0, 4, 8, 12, 16, 20, 24, 28);
        assert_eq_m128i(r, e);

        let r = _mm_mpsadbw_epu8::<0b111>(a, a);
        let e = _mm_setr_epi16(32, 28, 24, 20, 16, 12, 8, 4);
        assert_eq_m128i(r, e);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_testz_si128() {
        let a = _mm_set1_epi8(1);
        let mask = _mm_set1_epi8(0);
        let r = _mm_testz_si128(a, mask);
        assert_eq!(r, 1);
        let a = _mm_set1_epi8(0b101);
        let mask = _mm_set1_epi8(0b110);
        let r = _mm_testz_si128(a, mask);
        assert_eq!(r, 0);
        let a = _mm_set1_epi8(0b011);
        let mask = _mm_set1_epi8(0b100);
        let r = _mm_testz_si128(a, mask);
        assert_eq!(r, 1);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_testc_si128() {
        let a = _mm_set1_epi8(-1);
        let mask = _mm_set1_epi8(0);
        let r = _mm_testc_si128(a, mask);
        assert_eq!(r, 1);
        let a = _mm_set1_epi8(0b101);
        let mask = _mm_set1_epi8(0b110);
        let r = _mm_testc_si128(a, mask);
        assert_eq!(r, 0);
        let a = _mm_set1_epi8(0b101);
        let mask = _mm_set1_epi8(0b100);
        let r = _mm_testc_si128(a, mask);
        assert_eq!(r, 1);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_testnzc_si128() {
        let a = _mm_set1_epi8(0);
        let mask = _mm_set1_epi8(1);
        let r = _mm_testnzc_si128(a, mask);
        assert_eq!(r, 0);
        let a = _mm_set1_epi8(-1);
        let mask = _mm_set1_epi8(0);
        let r = _mm_testnzc_si128(a, mask);
        assert_eq!(r, 0);
        let a = _mm_set1_epi8(0b101);
        let mask = _mm_set1_epi8(0b110);
        let r = _mm_testnzc_si128(a, mask);
        assert_eq!(r, 1);
        let a = _mm_set1_epi8(0b101);
        let mask = _mm_set1_epi8(0b101);
        let r = _mm_testnzc_si128(a, mask);
        assert_eq!(r, 0);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_test_all_zeros() {
        let a = _mm_set1_epi8(1);
        let mask = _mm_set1_epi8(0);
        let r = _mm_test_all_zeros(a, mask);
        assert_eq!(r, 1);
        let a = _mm_set1_epi8(0b101);
        let mask = _mm_set1_epi8(0b110);
        let r = _mm_test_all_zeros(a, mask);
        assert_eq!(r, 0);
        let a = _mm_set1_epi8(0b011);
        let mask = _mm_set1_epi8(0b100);
        let r = _mm_test_all_zeros(a, mask);
        assert_eq!(r, 1);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_test_all_ones() {
        let a = _mm_set1_epi8(-1);
        let r = _mm_test_all_ones(a);
        assert_eq!(r, 1);
        let a = _mm_set1_epi8(0b101);
        let r = _mm_test_all_ones(a);
        assert_eq!(r, 0);
    }

    #[simd_test(enable = "sse4.1")]
    unsafe fn test_mm_test_mix_ones_zeros() {
        let a = _mm_set1_epi8(0);
        let mask = _mm_set1_epi8(1);
        let r = _mm_test_mix_ones_zeros(a, mask);
        assert_eq!(r, 0);
        let a = _mm_set1_epi8(-1);
        let mask = _mm_set1_epi8(0);
        let r = _mm_test_mix_ones_zeros(a, mask);
        assert_eq!(r, 0);
        let a = _mm_set1_epi8(0b101);
        let mask = _mm_set1_epi8(0b110);
        let r = _mm_test_mix_ones_zeros(a, mask);
        assert_eq!(r, 1);
        let a = _mm_set1_epi8(0b101);
        let mask = _mm_set1_epi8(0b101);
        let r = _mm_test_mix_ones_zeros(a, mask);
        assert_eq!(r, 0);
    }
}