wide/
u16x16_.rs

1use super::*;
2
3pick! {
4  if #[cfg(target_feature="avx2")] {
5    #[derive(Default, Clone, Copy, PartialEq, Eq)]
6    #[repr(C, align(32))]
7    pub struct u16x16 { pub(crate) avx2: m256i }
8  } else {
9    #[derive(Default, Clone, Copy, PartialEq, Eq)]
10    #[repr(C, align(32))]
11    pub struct u16x16 { pub(crate) a : u16x8, pub(crate) b : u16x8 }
12  }
13}
14
15int_uint_consts!(u16, 16, u16x16, 256);
16
17unsafe impl Zeroable for u16x16 {}
18unsafe impl Pod for u16x16 {}
19
20impl Add for u16x16 {
21  type Output = Self;
22  #[inline]
23  #[must_use]
24  fn add(self, rhs: Self) -> Self::Output {
25    pick! {
26      if #[cfg(target_feature="avx2")] {
27        Self { avx2: add_i16_m256i(self.avx2, rhs.avx2) }
28      } else {
29        Self {
30          a : self.a.add(rhs.a),
31          b : self.b.add(rhs.b),
32        }
33      }
34    }
35  }
36}
37
38impl Sub for u16x16 {
39  type Output = Self;
40  #[inline]
41  #[must_use]
42  fn sub(self, rhs: Self) -> Self::Output {
43    pick! {
44      if #[cfg(target_feature="avx2")] {
45        Self { avx2: sub_i16_m256i(self.avx2, rhs.avx2) }
46      } else {
47        Self {
48          a : self.a.sub(rhs.a),
49          b : self.b.sub(rhs.b),
50        }
51      }
52    }
53  }
54}
55
56impl Add<u16> for u16x16 {
57  type Output = Self;
58  #[inline]
59  #[must_use]
60  fn add(self, rhs: u16) -> Self::Output {
61    self.add(Self::splat(rhs))
62  }
63}
64
65impl Sub<u16> for u16x16 {
66  type Output = Self;
67  #[inline]
68  #[must_use]
69  fn sub(self, rhs: u16) -> Self::Output {
70    self.sub(Self::splat(rhs))
71  }
72}
73
74impl Add<u16x16> for u16 {
75  type Output = u16x16;
76  #[inline]
77  #[must_use]
78  fn add(self, rhs: u16x16) -> Self::Output {
79    u16x16::splat(self).add(rhs)
80  }
81}
82
83impl Sub<u16x16> for u16 {
84  type Output = u16x16;
85  #[inline]
86  #[must_use]
87  fn sub(self, rhs: u16x16) -> Self::Output {
88    u16x16::splat(self).sub(rhs)
89  }
90}
91
92impl BitAnd for u16x16 {
93  type Output = Self;
94  #[inline]
95  #[must_use]
96  fn bitand(self, rhs: Self) -> Self::Output {
97    pick! {
98      if #[cfg(target_feature="avx2")] {
99        Self { avx2: bitand_m256i(self.avx2, rhs.avx2) }
100      } else {
101        Self {
102          a : self.a.bitand(rhs.a),
103          b : self.b.bitand(rhs.b),
104        }
105      }
106    }
107  }
108}
109
110impl BitOr for u16x16 {
111  type Output = Self;
112  #[inline]
113  #[must_use]
114  fn bitor(self, rhs: Self) -> Self::Output {
115    pick! {
116      if #[cfg(target_feature="avx2")] {
117        Self { avx2: bitor_m256i(self.avx2, rhs.avx2) }
118      } else {
119        Self {
120          a : self.a.bitor(rhs.a),
121          b : self.b.bitor(rhs.b),
122        }
123      }
124    }
125  }
126}
127
128impl BitXor for u16x16 {
129  type Output = Self;
130  #[inline]
131  #[must_use]
132  fn bitxor(self, rhs: Self) -> Self::Output {
133    pick! {
134      if #[cfg(target_feature="avx2")] {
135        Self { avx2: bitxor_m256i(self.avx2, rhs.avx2) }
136      } else {
137        Self {
138          a : self.a.bitxor(rhs.a),
139          b : self.b.bitxor(rhs.b),
140        }
141      }
142    }
143  }
144}
145
146macro_rules! impl_shl_t_for_u16x16 {
147  ($($shift_type:ty),+ $(,)?) => {
148    $(impl Shl<$shift_type> for u16x16 {
149      type Output = Self;
150      /// Shifts all lanes by the value given.
151      #[inline]
152      #[must_use]
153      fn shl(self, rhs: $shift_type) -> Self::Output {
154        pick! {
155          if #[cfg(target_feature="avx2")] {
156            let shift = cast([rhs as u64, 0]);
157            Self { avx2: shl_all_u16_m256i(self.avx2, shift) }
158          } else {
159            Self {
160              a : self.a.shl(rhs),
161              b : self.b.shl(rhs),
162            }
163          }
164       }
165     }
166    })+
167  };
168}
169impl_shl_t_for_u16x16!(i8, u8, i16, u16, i32, u32, i64, u64, i128, u128);
170
171macro_rules! impl_shr_t_for_u16x16 {
172  ($($shift_type:ty),+ $(,)?) => {
173    $(impl Shr<$shift_type> for u16x16 {
174      type Output = Self;
175      /// Shifts all lanes by the value given.
176      #[inline]
177      #[must_use]
178      fn shr(self, rhs: $shift_type) -> Self::Output {
179        pick! {
180          if #[cfg(target_feature="avx2")] {
181            let shift = cast([rhs as u64, 0]);
182            Self { avx2: shr_all_u16_m256i(self.avx2, shift) }
183          } else {
184            Self {
185              a : self.a.shr(rhs),
186              b : self.b.shr(rhs),
187            }
188          }
189        }
190      }
191    })+
192  };
193}
194impl_shr_t_for_u16x16!(i8, u8, i16, u16, i32, u32, i64, u64, i128, u128);
195
196impl CmpEq for u16x16 {
197  type Output = Self;
198  #[inline]
199  #[must_use]
200  fn cmp_eq(self, rhs: Self) -> Self::Output {
201    pick! {
202      if #[cfg(target_feature="avx2")] {
203        Self { avx2: cmp_eq_mask_i16_m256i(self.avx2, rhs.avx2) }
204      } else {
205        Self {
206          a : self.a.cmp_eq(rhs.a),
207          b : self.b.cmp_eq(rhs.b),
208        }
209      }
210    }
211  }
212}
213
214impl Mul for u16x16 {
215  type Output = Self;
216  #[inline]
217  #[must_use]
218  fn mul(self, rhs: Self) -> Self::Output {
219    pick! {
220      if #[cfg(target_feature="avx2")] {
221        // non-widening multiplication is the same for unsigned and signed
222        Self { avx2: mul_i16_keep_low_m256i(self.avx2, rhs.avx2) }
223      } else {
224        Self {
225          a : self.a.mul(rhs.a),
226          b : self.b.mul(rhs.b),
227        }
228      }
229    }
230  }
231}
232
233impl From<u8x16> for u16x16 {
234  /// widens and sign extends to u16x16
235  #[inline]
236  #[must_use]
237  fn from(v: u8x16) -> Self {
238    pick! {
239      if #[cfg(target_feature="avx2")] {
240        u16x16 { avx2:convert_to_i16_m256i_from_u8_m128i(v.sse) }
241      } else if #[cfg(target_feature="sse2")] {
242        u16x16 {
243          a: u16x8 { sse: shr_imm_u16_m128i::<8>( unpack_low_i8_m128i(v.sse, v.sse)) },
244          b: u16x8 { sse: shr_imm_u16_m128i::<8>( unpack_high_i8_m128i(v.sse, v.sse)) },
245        }
246      } else {
247
248        u16x16::new([
249          v.as_array_ref()[0] as u16,
250          v.as_array_ref()[1] as u16,
251          v.as_array_ref()[2] as u16,
252          v.as_array_ref()[3] as u16,
253          v.as_array_ref()[4] as u16,
254          v.as_array_ref()[5] as u16,
255          v.as_array_ref()[6] as u16,
256          v.as_array_ref()[7] as u16,
257          v.as_array_ref()[8] as u16,
258          v.as_array_ref()[9] as u16,
259          v.as_array_ref()[10] as u16,
260          v.as_array_ref()[11] as u16,
261          v.as_array_ref()[12] as u16,
262          v.as_array_ref()[13] as u16,
263          v.as_array_ref()[14] as u16,
264          v.as_array_ref()[15] as u16,
265          ])
266      }
267    }
268  }
269}
270
271impl u16x16 {
272  #[inline]
273  #[must_use]
274  pub const fn new(array: [u16; 16]) -> Self {
275    unsafe { core::intrinsics::transmute(array) }
276  }
277
278  #[inline]
279  #[must_use]
280  pub fn blend(self, t: Self, f: Self) -> Self {
281    pick! {
282      if #[cfg(target_feature="avx2")] {
283        Self { avx2: blend_varying_i8_m256i(f.avx2, t.avx2, self.avx2) }
284      } else {
285        Self {
286          a : self.a.blend(t.a, f.a),
287          b : self.b.blend(t.b, f.b),
288        }
289      }
290    }
291  }
292
293  #[inline]
294  #[must_use]
295  pub fn max(self, rhs: Self) -> Self {
296    pick! {
297      if #[cfg(target_feature="avx2")] {
298        Self { avx2: max_u16_m256i(self.avx2, rhs.avx2) }
299      } else {
300        Self {
301          a : self.a.max(rhs.a),
302          b : self.b.max(rhs.b),
303        }
304      }
305    }
306  }
307  #[inline]
308  #[must_use]
309  pub fn min(self, rhs: Self) -> Self {
310    pick! {
311      if #[cfg(target_feature="avx2")] {
312        Self { avx2: min_u16_m256i(self.avx2, rhs.avx2) }
313      } else {
314        Self {
315          a : self.a.min(rhs.a),
316          b : self.b.min(rhs.b),
317        }
318      }
319    }
320  }
321
322  #[inline]
323  #[must_use]
324  pub fn saturating_add(self, rhs: Self) -> Self {
325    pick! {
326      if #[cfg(target_feature="avx2")] {
327        Self { avx2: add_saturating_u16_m256i(self.avx2, rhs.avx2) }
328      } else {
329        Self {
330          a : self.a.saturating_add(rhs.a),
331          b : self.b.saturating_add(rhs.b),
332        }
333      }
334    }
335  }
336  #[inline]
337  #[must_use]
338  pub fn saturating_sub(self, rhs: Self) -> Self {
339    pick! {
340      if #[cfg(target_feature="avx2")] {
341        Self { avx2: sub_saturating_u16_m256i(self.avx2, rhs.avx2) }
342      } else {
343        Self {
344          a : self.a.saturating_sub(rhs.a),
345          b : self.b.saturating_sub(rhs.b),
346        }
347      }
348    }
349  }
350
351  #[inline]
352  pub fn to_array(self) -> [u16; 16] {
353    cast(self)
354  }
355
356  #[inline]
357  pub fn as_array_ref(&self) -> &[u16; 16] {
358    cast_ref(self)
359  }
360
361  #[inline]
362  pub fn as_array_mut(&mut self) -> &mut [u16; 16] {
363    cast_mut(self)
364  }
365}
wide/u16x16_.rs

wide/
u16x16_.rs
