wide/

i16x16_.rs

use super::*;

pick! {
  if #[cfg(target_feature="avx2")] {
    /// A SIMD vector with 16 elements of type [`i16`].
    ///
    /// See the [crate level documentation] for more information about SIMD
    /// vectors.
    ///
    /// [crate level documentation]: crate
    #[derive(Default, Clone, Copy, PartialEq, Eq)]
    #[repr(C, align(32))]
    pub struct i16x16 { pub(crate) avx2: m256i }
  } else {
    /// A SIMD vector with 16 elements of type [`i16`].
    ///
    /// See the [crate level documentation] for more information about SIMD
    /// vectors.
    ///
    /// [crate level documentation]: crate
    #[derive(Default, Clone, Copy, PartialEq, Eq)]
    #[repr(C, align(32))]
    pub struct i16x16 { pub(crate) a : i16x8, pub(crate) b : i16x8 }
  }
}

impl_simd_int! {
  unsafe {
    T = i16,
    N = 16,
    Simd = i16x16,
    UintSimd = u16x16,
    T_BITS = 16,
    T_BITS_MUL_2 = 32,
    BitmaskType = u32,
    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
    optional_type_x86_inner { X86Inner = __m256i },
    optional_type_arm_inner {},
    optional_type_wasm_inner {},
  }

  #[inline]
  fn simd_lt(self, rhs: Self) -> Self::Output {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: !cmp_gt_mask_i16_m256i(self.avx2, rhs.avx2) ^ cmp_eq_mask_i16_m256i(self.avx2,rhs.avx2) }
      } else {
        Self {
          a : self.a.simd_lt(rhs.a),
          b : self.b.simd_lt(rhs.b),
        }
      }
    }
  }

  #[inline]
  fn simd_gt(self, rhs: Self) -> Self::Output {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: cmp_gt_mask_i16_m256i(self.avx2, rhs.avx2) }
      } else {
        Self {
          a : self.a.simd_gt(rhs.a),
          b : self.b.simd_gt(rhs.b),
        }
      }
    }
  }

  #[inline]
  fn simd_le(self, rhs: Self) -> Self::Output {
    pick! {
      if #[cfg(target_feature="avx2")] {
        !self.simd_gt(rhs)
      } else {
        Self {
          a : self.a.simd_le(rhs.a),
          b : self.b.simd_le(rhs.b),
        }
      }
    }
  }

  #[inline]
  fn simd_ge(self, rhs: Self) -> Self::Output {
    pick! {
      if #[cfg(target_feature="avx2")] {
        !self.simd_lt(rhs)
      } else {
        Self {
          a : self.a.simd_ge(rhs.a),
          b : self.b.simd_ge(rhs.b),
        }
      }
    }
  }

  #[inline]
  fn shr(self, rhs: u16x16) -> Self::Output {
    pick! {
      if #[cfg(all(target_feature="avx512bw", target_feature="avx512vl"))] {
        #[cfg(target_arch = "x86")]
        use core::arch::x86::_mm256_srav_epi16;
        #[cfg(target_arch = "x86_64")]
        use core::arch::x86_64::_mm256_srav_epi16;

        // Mask `rhs` to 15 to match `wrapping_shr`.
        let rhs = bitand_m256i(rhs.avx2, set_splat_i16_m256i(15));
        // TODO(safe_arch): Add `_mm256_srav_epi16`.
        cast(unsafe { _mm256_srav_epi16(self.avx2.0, rhs.0) })
      } else {
        let [self_a, self_b]: [i16x8; 2] = cast(self);
        let [rhs_a, rhs_b]: [u16x8; 2] = cast(rhs);

        cast([self_a >> rhs_a, self_b >> rhs_b])
      }
    }
  }

  #[inline]
  fn shr(self, rhs: u32) -> Self::Output {
    pick! {
      if #[cfg(target_feature="avx2")] {
        // Use `rhs % 16` to perform wrapping shift and not unbounded shift.
        #[expect(clippy::suspicious_arithmetic_impl)]
        let shift = cast([rhs as u64 & 15, 0]);
        Self { avx2: shr_all_i16_m256i(self.avx2, shift) }
      } else {
        Self {
          a : self.a.shr(rhs),
          b : self.b.shr(rhs),
        }
      }
    }
  }

  #[inline]
  pub fn max(self, rhs: Self) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: max_i16_m256i(self.avx2, rhs.avx2) }
      } else {
        Self {
          a : self.a.max(rhs.a),
          b : self.b.max(rhs.b),
        }
      }
    }
  }

  #[inline]
  pub fn min(self, rhs: Self) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: min_i16_m256i(self.avx2, rhs.avx2) }
      } else {
        Self {
          a : self.a.min(rhs.a),
          b : self.b.min(rhs.b),
        }
      }
    }
  }

  #[inline]
  pub fn reduce_max(self) -> i16 {
    let arr: [i16x8; 2] = cast(self);

    arr[0].max(arr[1]).reduce_max()
  }

  #[inline]
  pub fn reduce_min(self) -> i16 {
    let arr: [i16x8; 2] = cast(self);

    arr[0].min(arr[1]).reduce_min()
  }

  #[inline]
  pub fn unbounded_shr(self, rhs: u16x16) -> Self {
    pick! {
      if #[cfg(all(target_feature="avx512bw", target_feature="avx512vl"))] {
        #[cfg(target_arch = "x86")]
        use core::arch::x86::_mm256_srav_epi16;
        #[cfg(target_arch = "x86_64")]
        use core::arch::x86_64::_mm256_srav_epi16;

        // TODO(safe_arch): Add `_mm256_srav_epi16`.
        cast(unsafe { _mm256_srav_epi16(self.avx2.0, rhs.avx2.0) })
      } else {
        let [self_a, self_b] = cast::<i16x16, [i16x8; 2]>(self);
        let [rhs_a, rhs_b] = cast::<u16x16, [u16x8; 2]>(rhs);

        cast([self_a.unbounded_shr(rhs_a), self_b.unbounded_shr(rhs_b)])
      }
    }
  }

  #[inline]
  pub fn unbounded_shr_scalar(self, rhs: u32) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: shr_all_i16_m256i(self.avx2, cast([rhs as u64, 0])) }
      } else {
        Self {
          a: self.a.unbounded_shr_scalar(rhs),
          b: self.b.unbounded_shr_scalar(rhs),
        }
      }
    }
  }

  #[inline]
  pub fn saturating_add(self, rhs: Self) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: add_saturating_i16_m256i(self.avx2, rhs.avx2) }
      } else {
        Self {
          a : self.a.saturating_add(rhs.a),
          b : self.b.saturating_add(rhs.b),
        }
      }
    }
  }

  #[inline]
  pub fn saturating_sub(self, rhs: Self) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: sub_saturating_i16_m256i(self.avx2, rhs.avx2) }
      } else {
        Self {
          a : self.a.saturating_sub(rhs.a),
          b : self.b.saturating_sub(rhs.b),
        }
      }
    }
  }

  #[inline]
  pub fn overflowing_mul(self, rhs: Self) -> (Self, Self) {
    let (low, high) = self.mul_keep_low_high(rhs);
    let low = cast::<u16x16, i16x16>(low);

    let overflow = high.simd_ne(low.is_negative());
    (low, overflow)
  }

  optional_fn_widening_mul {
    #[inline]
    pub fn widening_mul(self, rhs: Self) -> i32x16 {
      // x86 has no `_mm256_mul_epi16` intrinsic so there is no `avx2`
      // optimization.

      let [self_a, self_b] = cast::<i16x16, [i16x8; 2]>(self);
      let [rhs_a, rhs_b] = cast::<i16x16, [i16x8; 2]>(rhs);

      cast([self_a.widening_mul(rhs_a), self_b.widening_mul(rhs_b)])
    }
  }

  #[inline]
  pub fn mul_keep_low_high(self, rhs: Self) -> (u16x16, i16x16) {
    // x86 has no `_mm256_mul_epi16` intrinsic so there is no `avx2`
    // optimization.

    let [self_a, self_b] = cast::<i16x16, [i16x8; 2]>(self);
    let [rhs_a, rhs_b] = cast::<i16x16, [i16x8; 2]>(rhs);

    let result_a = self_a.mul_keep_low_high(rhs_a);
    let result_b = self_b.mul_keep_low_high(rhs_b);
    (cast([result_a.0, result_b.0]), cast([result_a.1, result_b.1]))
  }

  #[inline]
  pub fn mul_keep_high(self, rhs: Self) -> Self {
    // x86 has no `_mm256_mul_epi16` intrinsic so there is no `avx2`
    // optimization.

    let [self_a, self_b] = cast::<i16x16, [i16x8; 2]>(self);
    let [rhs_a, rhs_b] = cast::<i16x16, [i16x8; 2]>(rhs);

    cast([self_a.mul_keep_high(rhs_a), self_b.mul_keep_high(rhs_b)])
  }

  #[inline]
  pub fn abs(self) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: abs_i16_m256i(self.avx2) }
      } else {
        Self {
          a : self.a.abs(),
          b : self.b.abs(),
        }
      }
    }
  }

  #[inline]
  pub fn is_positive(self) -> Self {
    pick! {
      if #[cfg(all(target_feature="neon", target_arch="aarch64"))] {
        // `neon` has dedicated greater-than-zero intrinsics.
        Self {
          a: self.a.is_positive(),
          b: self.b.is_positive(),
        }
      } else {
        self.simd_gt(Self::ZERO)
      }
    }
  }

  #[inline]
  pub fn is_negative(self) -> Self {
    pick! {
      if #[cfg(all(target_feature="neon", target_arch="aarch64"))] {
        // `neon` has dedicated less-than-zero intrinsics.
        Self {
          a: self.a.is_negative(),
          b: self.b.is_negative(),
        }
      } else {
        self.simd_lt(Self::ZERO)
      }
    }
  }

  optional_fn_deserialize {}
}

impl From<i8x16> for i16x16 {
  /// widen with sign extend from i8 to i16
  #[inline]
  fn from(i: i8x16) -> Self {
    i16x16::from_i8x16(i)
  }
}

impl From<u8x16> for i16x16 {
  /// widen with zero extend from u8 to i16
  #[inline]
  fn from(i: u8x16) -> Self {
    cast(u16x16::from(i))
  }
}

/// The following functionality exists only for [`i16x16`], or only for
/// particular types inconsistently.
impl i16x16 {
  /// Converts each element from [`i8`] to [`i16`].
  #[inline]
  #[must_use]
  pub fn from_i8x16(v: i8x16) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        i16x16 { avx2:convert_to_i16_m256i_from_i8_m128i(v.sse) }
      } else if #[cfg(target_feature="sse4.1")] {
        i16x16 {
          a: i16x8 { sse: convert_to_i16_m128i_from_lower8_i8_m128i(v.sse) },
          b: i16x8 { sse: convert_to_i16_m128i_from_lower8_i8_m128i(unpack_high_i64_m128i(v.sse, v.sse)) }
        }
      } else if #[cfg(target_feature="sse2")] {
        i16x16 {
          a: i16x8 { sse: shr_imm_i16_m128i::<8>( unpack_low_i8_m128i(v.sse, v.sse)) },
          b: i16x8 { sse: shr_imm_i16_m128i::<8>( unpack_high_i8_m128i(v.sse, v.sse)) },
        }
      } else {

        i16x16::new([
          v.as_array()[0] as i16,
          v.as_array()[1] as i16,
          v.as_array()[2] as i16,
          v.as_array()[3] as i16,
          v.as_array()[4] as i16,
          v.as_array()[5] as i16,
          v.as_array()[6] as i16,
          v.as_array()[7] as i16,
          v.as_array()[8] as i16,
          v.as_array()[9] as i16,
          v.as_array()[10] as i16,
          v.as_array()[11] as i16,
          v.as_array()[12] as i16,
          v.as_array()[13] as i16,
          v.as_array()[14] as i16,
          v.as_array()[15] as i16,
          ])
      }
    }
  }

  /// Partially computes the dot product.
  ///
  /// First this multiplies the input 16-bit integers, producing intermediate
  /// 32-bit integers. Then this horizontally adds adjacent pairs, resulting in
  /// eight 32-bit integers.
  #[inline]
  #[must_use]
  pub fn dot(self, rhs: Self) -> i32x8 {
    pick! {
      if #[cfg(target_feature="avx2")] {
        i32x8 { avx2:  mul_i16_horizontal_add_m256i(self.avx2, rhs.avx2) }
      } else {
        i32x8 {
          a : self.a.dot(rhs.a),
          b : self.b.dot(rhs.b),
        }
      }
    }
  }

  /// Multiply and scale equivalent to `((self * rhs) + 0x4000) >> 15` on each
  /// lane, effectively multiplying by a 16 bit fixed point number between `-1`
  /// and `1`. This corresponds to the following instructions:
  /// - `vqrdmulhq_n_s16` instruction on neon
  /// - `i16x8_q15mulr_sat` on simd128
  /// - `_mm256_mulhrs_epi16` on avx2
  /// - emulated via `mul_i16_*` on sse2
  #[inline]
  #[must_use]
  pub fn mul_scale_round(self, rhs: Self) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: mul_i16_scale_round_m256i(self.avx2, rhs.avx2) }
      } else {
        Self {
          a : self.a.mul_scale_round(rhs.a),
          b : self.b.mul_scale_round(rhs.b),
        }
      }
    }
  }

  /// Multiply and scale equivalent to `((self * rhs) + 0x4000) >> 15` on each
  /// lane, effectively multiplying by a 16 bit fixed point number between `-1`
  /// and `1`. This corresponds to the following instructions:
  /// - `vqrdmulhq_n_s16` instruction on neon
  /// - `i16x8_q15mulr_sat` on simd128
  /// - `_mm256_mulhrs_epi16` on avx2
  /// - emulated via `mul_i16_*` on sse2
  #[inline]
  #[must_use]
  pub fn mul_scale_round_n(self, rhs: i16) -> Self {
    pick! {
      if #[cfg(target_feature="avx2")] {
        Self { avx2: mul_i16_scale_round_m256i(self.avx2, set_splat_i16_m256i(rhs)) }
      } else {
        Self {
          a : self.a.mul_scale_round_n(rhs),
          b : self.b.mul_scale_round_n(rhs),
        }
      }
    }
  }
}