wide/
u32x8_.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 u32x8 { pub(crate) avx2: m256i }
8  } else {
9    #[derive(Default, Clone, Copy, PartialEq, Eq)]
10    #[repr(C, align(32))]
11    pub struct u32x8 { pub(crate) a : u32x4, pub(crate) b : u32x4 }
12  }
13}
14
15int_uint_consts!(u32, 8, u32x8, 256);
16
17unsafe impl Zeroable for u32x8 {}
18unsafe impl Pod for u32x8 {}
19
20impl Add for u32x8 {
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_i32_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 u32x8 {
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_i32_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 Mul for u32x8 {
57  type Output = Self;
58  #[inline]
59  #[must_use]
60  fn mul(self, rhs: Self) -> Self::Output {
61    pick! {
62      if #[cfg(target_feature="avx2")] {
63        Self { avx2: mul_i32_keep_low_m256i(self.avx2, rhs.avx2) }
64      } else {
65        Self {
66          a : self.a.mul(rhs.a),
67          b : self.b.mul(rhs.b),
68        }
69      }
70    }
71  }
72}
73
74impl BitAnd for u32x8 {
75  type Output = Self;
76  #[inline]
77  #[must_use]
78  fn bitand(self, rhs: Self) -> Self::Output {
79    pick! {
80      if #[cfg(target_feature="avx2")] {
81        Self { avx2: bitand_m256i(self.avx2, rhs.avx2) }
82      } else {
83        Self {
84          a : self.a.bitand(rhs.a),
85          b : self.b.bitand(rhs.b),
86        }
87      }
88    }
89  }
90}
91
92impl BitOr for u32x8 {
93  type Output = Self;
94  #[inline]
95  #[must_use]
96  fn bitor(self, rhs: Self) -> Self::Output {
97    pick! {
98      if #[cfg(target_feature="avx2")] {
99        Self { avx2: bitor_m256i(self.avx2, rhs.avx2) }
100      } else {
101        Self {
102          a : self.a.bitor(rhs.a),
103          b : self.b.bitor(rhs.b),
104        }
105      }
106    }
107  }
108}
109
110impl BitXor for u32x8 {
111  type Output = Self;
112  #[inline]
113  #[must_use]
114  fn bitxor(self, rhs: Self) -> Self::Output {
115    pick! {
116      if #[cfg(target_feature="avx2")] {
117        Self { avx2: bitxor_m256i(self.avx2, rhs.avx2) }
118      } else {
119        Self {
120          a : self.a.bitxor(rhs.a),
121          b : self.b.bitxor(rhs.b),
122        }
123      }
124    }
125  }
126}
127
128impl From<u16x8> for u32x8 {
129  /// widens and zero extends to u32x8
130  #[inline]
131  #[must_use]
132  fn from(v: u16x8) -> Self {
133    pick! {
134      if #[cfg(target_feature="avx2")] {
135        Self { avx2:convert_to_i32_m256i_from_u16_m128i(v.sse) }
136      } else if #[cfg(target_feature="sse2")] {
137        Self {
138          a: u32x4 { sse: shr_imm_u32_m128i::<16>( unpack_low_i16_m128i(v.sse, v.sse)) },
139          b: u32x4 { sse: shr_imm_u32_m128i::<16>( unpack_high_i16_m128i(v.sse, v.sse)) },
140        }
141      } else {
142        u32x8::new([
143          u32::from(v.as_array_ref()[0]),
144          u32::from(v.as_array_ref()[1]),
145          u32::from(v.as_array_ref()[2]),
146          u32::from(v.as_array_ref()[3]),
147          u32::from(v.as_array_ref()[4]),
148          u32::from(v.as_array_ref()[5]),
149          u32::from(v.as_array_ref()[6]),
150          u32::from(v.as_array_ref()[7]),
151        ])
152      }
153    }
154  }
155}
156
157macro_rules! impl_shl_t_for_u32x8 {
158  ($($shift_type:ty),+ $(,)?) => {
159    $(impl Shl<$shift_type> for u32x8 {
160      type Output = Self;
161      /// Shifts all lanes by the value given.
162      #[inline]
163      #[must_use]
164      fn shl(self, rhs: $shift_type) -> Self::Output {
165        pick! {
166          if #[cfg(target_feature="avx2")] {
167            let shift = cast([rhs as u64, 0]);
168            Self { avx2: shl_all_u32_m256i(self.avx2, shift) }
169          } else {
170            Self {
171              a : self.a.shl(rhs),
172              b : self.b.shl(rhs),
173            }
174          }
175        }
176      }
177    })+
178  };
179}
180impl_shl_t_for_u32x8!(i8, u8, i16, u16, i32, u32, i64, u64, i128, u128);
181
182macro_rules! impl_shr_t_for_u32x8 {
183  ($($shift_type:ty),+ $(,)?) => {
184    $(impl Shr<$shift_type> for u32x8 {
185      type Output = Self;
186      /// Shifts all lanes by the value given.
187      #[inline]
188      #[must_use]
189      fn shr(self, rhs: $shift_type) -> Self::Output {
190        pick! {
191          if #[cfg(target_feature="avx2")] {
192            let shift = cast([rhs as u64, 0]);
193            Self { avx2: shr_all_u32_m256i(self.avx2, shift) }
194          } else {
195            Self {
196              a : self.a.shr(rhs),
197              b : self.b.shr(rhs),
198            }
199          }
200        }
201      }
202    })+
203  };
204}
205
206impl_shr_t_for_u32x8!(i8, u8, i16, u16, i32, u32, i64, u64, i128, u128);
207
208/// Shifts lanes by the corresponding lane.
209///
210/// Bitwise shift-right; yields `self >> mask(rhs)`, where mask removes any
211/// high-order bits of `rhs` that would cause the shift to exceed the bitwidth
212/// of the type. (same as `wrapping_shr`)
213impl Shr<u32x8> for u32x8 {
214  type Output = Self;
215
216  #[inline]
217  #[must_use]
218  fn shr(self, rhs: u32x8) -> Self::Output {
219    pick! {
220      if #[cfg(target_feature="avx2")] {
221        // ensure same behavior as scalar wrapping_shr
222        let shift_by = bitand_m256i(rhs.avx2, set_splat_i32_m256i(31));
223        Self { avx2: shr_each_u32_m256i(self.avx2, shift_by ) }
224      } else {
225        Self {
226          a : self.a.shr(rhs.a),
227          b : self.b.shr(rhs.b),
228        }
229      }
230    }
231  }
232}
233
234/// Shifts lanes by the corresponding lane.
235///
236/// Bitwise shift-left; yields `self << mask(rhs)`, where mask removes any
237/// high-order bits of `rhs` that would cause the shift to exceed the bitwidth
238/// of the type. (same as `wrapping_shl`)
239impl Shl<u32x8> for u32x8 {
240  type Output = Self;
241
242  #[inline]
243  #[must_use]
244  fn shl(self, rhs: u32x8) -> Self::Output {
245    pick! {
246      if #[cfg(target_feature="avx2")] {
247        // ensure same behavior as scalar wrapping_shl
248        let shift_by = bitand_m256i(rhs.avx2, set_splat_i32_m256i(31));
249        Self { avx2: shl_each_u32_m256i(self.avx2, shift_by) }
250      } else {
251        Self {
252          a : self.a.shl(rhs.a),
253          b : self.b.shl(rhs.b),
254        }
255      }
256    }
257  }
258}
259
260impl u32x8 {
261  #[inline]
262  #[must_use]
263  pub const fn new(array: [u32; 8]) -> Self {
264    unsafe { core::intrinsics::transmute(array) }
265  }
266  #[inline]
267  #[must_use]
268  pub fn cmp_eq(self, rhs: Self) -> Self {
269    pick! {
270      if #[cfg(target_feature="avx2")] {
271        Self { avx2: cmp_eq_mask_i32_m256i(self.avx2, rhs.avx2 ) }
272      } else {
273        Self {
274          a : self.a.cmp_eq(rhs.a),
275          b : self.b.cmp_eq(rhs.b),
276        }
277      }
278    }
279  }
280  #[inline]
281  #[must_use]
282  pub fn cmp_gt(self, rhs: Self) -> Self {
283    pick! {
284      if #[cfg(target_feature="avx2")] {
285        // no unsigned gt than so inverting the high bit will get the correct result
286        let highbit = u32x8::splat(1 << 31);
287        Self { avx2: cmp_gt_mask_i32_m256i((self ^ highbit).avx2, (rhs ^ highbit).avx2 ) }
288      } else {
289        Self {
290          a : self.a.cmp_gt(rhs.a),
291          b : self.b.cmp_gt(rhs.b),
292        }
293      }
294    }
295  }
296
297  #[inline]
298  #[must_use]
299  pub fn cmp_lt(self, rhs: Self) -> Self {
300    // lt is just gt the other way around
301    rhs.cmp_gt(self)
302  }
303
304  /// Multiplies 32x32 bit to 64 bit and then only keeps the high 32 bits of the
305  /// result. Useful for implementing divide constant value (see t_usefulness
306  /// example)
307  #[inline]
308  #[must_use]
309  pub fn mul_keep_high(self, rhs: u32x8) -> u32x8 {
310    pick! {
311      if #[cfg(target_feature="avx2")] {
312        let a : [u32;8]= cast(self);
313        let b : [u32;8]= cast(rhs);
314
315        // let the compiler shuffle the values around, it does the right thing
316        let r1 : [u32;8] = cast(mul_u64_low_bits_m256i(cast([a[0], 0, a[1], 0, a[2], 0, a[3], 0]), cast([b[0], 0, b[1], 0, b[2], 0, b[3], 0])));
317        let r2 : [u32;8] = cast(mul_u64_low_bits_m256i(cast([a[4], 0, a[5], 0, a[6], 0, a[7], 0]), cast([b[4], 0, b[5], 0, b[6], 0, b[7], 0])));
318
319        cast([r1[1], r1[3], r1[5], r1[7], r2[1], r2[3], r2[5], r2[7]])
320      } else {
321        Self {
322          a : self.a.mul_keep_high(rhs.a),
323          b : self.b.mul_keep_high(rhs.b),
324        }
325      }
326    }
327  }
328
329  #[inline]
330  #[must_use]
331  pub fn blend(self, t: Self, f: Self) -> Self {
332    pick! {
333      if #[cfg(target_feature="avx2")] {
334        Self { avx2: blend_varying_i8_m256i(f.avx2, t.avx2, self.avx2) }
335      } else {
336        Self {
337          a : self.a.blend(t.a, f.a),
338          b : self.b.blend(t.b, f.b),
339        }
340      }
341    }
342  }
343
344  #[inline]
345  #[must_use]
346  pub fn max(self, rhs: Self) -> Self {
347    pick! {
348      if #[cfg(target_feature="avx2")] {
349        Self { avx2: max_u32_m256i(self.avx2, rhs.avx2 ) }
350      } else {
351        Self {
352          a : self.a.max(rhs.a),
353          b : self.b.max(rhs.b),
354        }
355      }
356    }
357  }
358  #[inline]
359  #[must_use]
360  pub fn min(self, rhs: Self) -> Self {
361    pick! {
362      if #[cfg(target_feature="avx2")] {
363        Self { avx2: min_u32_m256i(self.avx2, rhs.avx2 ) }
364      } else {
365        Self {
366          a : self.a.min(rhs.a),
367          b : self.b.min(rhs.b),
368        }
369      }
370    }
371  }
372
373  #[inline]
374  #[must_use]
375  pub fn any(self) -> bool {
376    pick! {
377      if #[cfg(target_feature="avx2")] {
378        ((move_mask_i8_m256i(self.avx2) as u32) & 0b10001000100010001000100010001000) != 0
379      } else {
380        (self.a | self.b).any()
381      }
382    }
383  }
384
385  #[inline]
386  #[must_use]
387  pub fn all(self) -> bool {
388    pick! {
389      if #[cfg(target_feature="avx2")] {
390        ((move_mask_i8_m256i(self.avx2) as u32) & 0b10001000100010001000100010001000) == 0b10001000100010001000100010001000
391      } else {
392        (self.a & self.b).all()
393      }
394    }
395  }
396
397  #[inline]
398  #[must_use]
399  pub fn none(self) -> bool {
400    !self.any()
401  }
402
403  #[inline]
404  pub fn to_array(self) -> [u32; 8] {
405    cast(self)
406  }
407
408  #[inline]
409  pub fn as_array_ref(&self) -> &[u32; 8] {
410    cast_ref(self)
411  }
412
413  #[inline]
414  pub fn as_array_mut(&mut self) -> &mut [u32; 8] {
415    cast_mut(self)
416  }
417}
418
419impl Not for u32x8 {
420  type Output = Self;
421  #[inline]
422  fn not(self) -> Self {
423    pick! {
424      if #[cfg(target_feature="avx2")] {
425        Self { avx2: self.avx2.not()  }
426      } else {
427        Self {
428          a : self.a.not(),
429          b : self.b.not(),
430        }
431      }
432    }
433  }
434}
wide/u32x8_.rs

wide/
u32x8_.rs
