core/stdarch/crates/core_arch/src/x86/
sse41.rs

1//! Streaming SIMD Extensions 4.1 (SSE4.1)
2
3use crate::core_arch::{simd::*, x86::*};
4use crate::intrinsics::simd::*;
5
6#[cfg(test)]
7use stdarch_test::assert_instr;
8
9// SSE4 rounding constants
10/// round to nearest
11#[stable(feature = "simd_x86", since = "1.27.0")]
12pub const _MM_FROUND_TO_NEAREST_INT: i32 = 0x00;
13/// round down
14#[stable(feature = "simd_x86", since = "1.27.0")]
15pub const _MM_FROUND_TO_NEG_INF: i32 = 0x01;
16/// round up
17#[stable(feature = "simd_x86", since = "1.27.0")]
18pub const _MM_FROUND_TO_POS_INF: i32 = 0x02;
19/// truncate
20#[stable(feature = "simd_x86", since = "1.27.0")]
21pub const _MM_FROUND_TO_ZERO: i32 = 0x03;
22/// use MXCSR.RC; see `vendor::_MM_SET_ROUNDING_MODE`
23#[stable(feature = "simd_x86", since = "1.27.0")]
24pub const _MM_FROUND_CUR_DIRECTION: i32 = 0x04;
25/// do not suppress exceptions
26#[stable(feature = "simd_x86", since = "1.27.0")]
27pub const _MM_FROUND_RAISE_EXC: i32 = 0x00;
28/// suppress exceptions
29#[stable(feature = "simd_x86", since = "1.27.0")]
30pub const _MM_FROUND_NO_EXC: i32 = 0x08;
31/// round to nearest and do not suppress exceptions
32#[stable(feature = "simd_x86", since = "1.27.0")]
33pub const _MM_FROUND_NINT: i32 = 0x00;
34/// round down and do not suppress exceptions
35#[stable(feature = "simd_x86", since = "1.27.0")]
36pub const _MM_FROUND_FLOOR: i32 = _MM_FROUND_RAISE_EXC | _MM_FROUND_TO_NEG_INF;
37/// round up and do not suppress exceptions
38#[stable(feature = "simd_x86", since = "1.27.0")]
39pub const _MM_FROUND_CEIL: i32 = _MM_FROUND_RAISE_EXC | _MM_FROUND_TO_POS_INF;
40/// truncate and do not suppress exceptions
41#[stable(feature = "simd_x86", since = "1.27.0")]
42pub const _MM_FROUND_TRUNC: i32 = _MM_FROUND_RAISE_EXC | _MM_FROUND_TO_ZERO;
43/// use MXCSR.RC and do not suppress exceptions; see
44/// `vendor::_MM_SET_ROUNDING_MODE`
45#[stable(feature = "simd_x86", since = "1.27.0")]
46pub const _MM_FROUND_RINT: i32 = _MM_FROUND_RAISE_EXC | _MM_FROUND_CUR_DIRECTION;
47/// use MXCSR.RC and suppress exceptions; see `vendor::_MM_SET_ROUNDING_MODE`
48#[stable(feature = "simd_x86", since = "1.27.0")]
49pub const _MM_FROUND_NEARBYINT: i32 = _MM_FROUND_NO_EXC | _MM_FROUND_CUR_DIRECTION;
50
51/// Blend packed 8-bit integers from `a` and `b` using `mask`
52///
53/// The high bit of each corresponding mask byte determines the selection.
54/// If the high bit is set, the element of `b` is selected.
55/// Otherwise, the element of `a` is selected.
56///
57/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blendv_epi8)
58#[inline]
59#[target_feature(enable = "sse4.1")]
60#[cfg_attr(test, assert_instr(pblendvb))]
61#[stable(feature = "simd_x86", since = "1.27.0")]
62pub fn _mm_blendv_epi8(a: __m128i, b: __m128i, mask: __m128i) -> __m128i {
63    unsafe {
64        let mask: i8x16 = simd_lt(mask.as_i8x16(), i8x16::ZERO);
65        transmute(simd_select(mask, b.as_i8x16(), a.as_i8x16()))
66    }
67}
68
69/// Blend packed 16-bit integers from `a` and `b` using the mask `IMM8`.
70///
71/// The mask bits determine the selection. A clear bit selects the
72/// corresponding element of `a`, and a set bit the corresponding
73/// element of `b`.
74///
75/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blend_epi16)
76#[inline]
77#[target_feature(enable = "sse4.1")]
78#[cfg_attr(test, assert_instr(pblendw, IMM8 = 0xB1))]
79#[rustc_legacy_const_generics(2)]
80#[stable(feature = "simd_x86", since = "1.27.0")]
81pub fn _mm_blend_epi16<const IMM8: i32>(a: __m128i, b: __m128i) -> __m128i {
82    static_assert_uimm_bits!(IMM8, 8);
83    unsafe {
84        transmute::<i16x8, _>(simd_shuffle!(
85            a.as_i16x8(),
86            b.as_i16x8(),
87            [
88                [0, 8][IMM8 as usize & 1],
89                [1, 9][(IMM8 >> 1) as usize & 1],
90                [2, 10][(IMM8 >> 2) as usize & 1],
91                [3, 11][(IMM8 >> 3) as usize & 1],
92                [4, 12][(IMM8 >> 4) as usize & 1],
93                [5, 13][(IMM8 >> 5) as usize & 1],
94                [6, 14][(IMM8 >> 6) as usize & 1],
95                [7, 15][(IMM8 >> 7) as usize & 1],
96            ]
97        ))
98    }
99}
100
101/// Blend packed double-precision (64-bit) floating-point elements from `a`
102/// and `b` using `mask`
103///
104/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blendv_pd)
105#[inline]
106#[target_feature(enable = "sse4.1")]
107#[cfg_attr(test, assert_instr(blendvpd))]
108#[stable(feature = "simd_x86", since = "1.27.0")]
109pub fn _mm_blendv_pd(a: __m128d, b: __m128d, mask: __m128d) -> __m128d {
110    unsafe {
111        let mask: i64x2 = simd_lt(transmute::<_, i64x2>(mask), i64x2::ZERO);
112        transmute(simd_select(mask, b.as_f64x2(), a.as_f64x2()))
113    }
114}
115
116/// Blend packed single-precision (32-bit) floating-point elements from `a`
117/// and `b` using `mask`
118///
119/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blendv_ps)
120#[inline]
121#[target_feature(enable = "sse4.1")]
122#[cfg_attr(test, assert_instr(blendvps))]
123#[stable(feature = "simd_x86", since = "1.27.0")]
124pub fn _mm_blendv_ps(a: __m128, b: __m128, mask: __m128) -> __m128 {
125    unsafe {
126        let mask: i32x4 = simd_lt(transmute::<_, i32x4>(mask), i32x4::ZERO);
127        transmute(simd_select(mask, b.as_f32x4(), a.as_f32x4()))
128    }
129}
130
131/// Blend packed double-precision (64-bit) floating-point elements from `a`
132/// and `b` using control mask `IMM2`
133///
134/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blend_pd)
135#[inline]
136#[target_feature(enable = "sse4.1")]
137// Note: LLVM7 prefers the single-precision floating-point domain when possible
138// see https://bugs.llvm.org/show_bug.cgi?id=38195
139// #[cfg_attr(test, assert_instr(blendpd, IMM2 = 0b10))]
140#[cfg_attr(test, assert_instr(blendps, IMM2 = 0b10))]
141#[rustc_legacy_const_generics(2)]
142#[stable(feature = "simd_x86", since = "1.27.0")]
143pub fn _mm_blend_pd<const IMM2: i32>(a: __m128d, b: __m128d) -> __m128d {
144    static_assert_uimm_bits!(IMM2, 2);
145    unsafe {
146        transmute::<f64x2, _>(simd_shuffle!(
147            a.as_f64x2(),
148            b.as_f64x2(),
149            [[0, 2][IMM2 as usize & 1], [1, 3][(IMM2 >> 1) as usize & 1]]
150        ))
151    }
152}
153
154/// Blend packed single-precision (32-bit) floating-point elements from `a`
155/// and `b` using mask `IMM4`
156///
157/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_blend_ps)
158#[inline]
159#[target_feature(enable = "sse4.1")]
160#[cfg_attr(test, assert_instr(blendps, IMM4 = 0b0101))]
161#[rustc_legacy_const_generics(2)]
162#[stable(feature = "simd_x86", since = "1.27.0")]
163pub fn _mm_blend_ps<const IMM4: i32>(a: __m128, b: __m128) -> __m128 {
164    static_assert_uimm_bits!(IMM4, 4);
165    unsafe {
166        transmute::<f32x4, _>(simd_shuffle!(
167            a.as_f32x4(),
168            b.as_f32x4(),
169            [
170                [0, 4][IMM4 as usize & 1],
171                [1, 5][(IMM4 >> 1) as usize & 1],
172                [2, 6][(IMM4 >> 2) as usize & 1],
173                [3, 7][(IMM4 >> 3) as usize & 1],
174            ]
175        ))
176    }
177}
178
179/// Extracts a single-precision (32-bit) floating-point element from `a`,
180/// selected with `IMM8`. The returned `i32` stores the float's bit-pattern,
181/// and may be converted back to a floating point number via casting.
182///
183/// # Example
184/// ```rust
185/// # #[cfg(target_arch = "x86")]
186/// # use std::arch::x86::*;
187/// # #[cfg(target_arch = "x86_64")]
188/// # use std::arch::x86_64::*;
189/// # fn main() {
190/// #    if is_x86_feature_detected!("sse4.1") {
191/// #       #[target_feature(enable = "sse4.1")]
192/// #       #[allow(unused_unsafe)] // FIXME remove after stdarch bump in rustc
193/// #       unsafe fn worker() { unsafe {
194/// let mut float_store = vec![1.0, 1.0, 2.0, 3.0];
195/// let simd_floats = _mm_set_ps(2.5, 5.0, 7.5, 10.0);
196/// let x: i32 = _mm_extract_ps::<2>(simd_floats);
197/// float_store.push(f32::from_bits(x as u32));
198/// assert_eq!(float_store, vec![1.0, 1.0, 2.0, 3.0, 5.0]);
199/// #       }}
200/// #       unsafe { worker() }
201/// #   }
202/// # }
203/// ```
204/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_extract_ps)
205#[inline]
206#[target_feature(enable = "sse4.1")]
207#[cfg_attr(all(test, not(target_env = "msvc")), assert_instr(extractps, IMM8 = 0))]
208#[rustc_legacy_const_generics(1)]
209#[stable(feature = "simd_x86", since = "1.27.0")]
210pub fn _mm_extract_ps<const IMM8: i32>(a: __m128) -> i32 {
211    static_assert_uimm_bits!(IMM8, 2);
212    unsafe { simd_extract!(a, IMM8 as u32, f32).to_bits() as i32 }
213}
214
215/// Extracts an 8-bit integer from `a`, selected with `IMM8`. Returns a 32-bit
216/// integer containing the zero-extended integer data.
217///
218/// See [LLVM commit D20468](https://reviews.llvm.org/D20468).
219///
220/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_extract_epi8)
221#[inline]
222#[target_feature(enable = "sse4.1")]
223#[cfg_attr(test, assert_instr(pextrb, IMM8 = 0))]
224#[rustc_legacy_const_generics(1)]
225#[stable(feature = "simd_x86", since = "1.27.0")]
226pub fn _mm_extract_epi8<const IMM8: i32>(a: __m128i) -> i32 {
227    static_assert_uimm_bits!(IMM8, 4);
228    unsafe { simd_extract!(a.as_u8x16(), IMM8 as u32, u8) as i32 }
229}
230
231/// Extracts an 32-bit integer from `a` selected with `IMM8`
232///
233/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_extract_epi32)
234#[inline]
235#[target_feature(enable = "sse4.1")]
236#[cfg_attr(all(test, not(target_env = "msvc")), assert_instr(extractps, IMM8 = 1))]
237#[rustc_legacy_const_generics(1)]
238#[stable(feature = "simd_x86", since = "1.27.0")]
239pub fn _mm_extract_epi32<const IMM8: i32>(a: __m128i) -> i32 {
240    static_assert_uimm_bits!(IMM8, 2);
241    unsafe { simd_extract!(a.as_i32x4(), IMM8 as u32, i32) }
242}
243
244/// Select a single value in `b` to store at some position in `a`,
245/// Then zero elements according to `IMM8`.
246///
247/// `IMM8` specifies which bits from operand `b` will be copied, which bits in
248/// the result they will be copied to, and which bits in the result will be
249/// cleared. The following assignments are made:
250///
251/// * Bits `[7:6]` specify the bits to copy from operand `b`:
252///     - `00`: Selects bits `[31:0]` from operand `b`.
253///     - `01`: Selects bits `[63:32]` from operand `b`.
254///     - `10`: Selects bits `[95:64]` from operand `b`.
255///     - `11`: Selects bits `[127:96]` from operand `b`.
256///
257/// * Bits `[5:4]` specify the bits in the result to which the selected bits
258///   from operand `b` are copied:
259///     - `00`: Copies the selected bits from `b` to result bits `[31:0]`.
260///     - `01`: Copies the selected bits from `b` to result bits `[63:32]`.
261///     - `10`: Copies the selected bits from `b` to result bits `[95:64]`.
262///     - `11`: Copies the selected bits from `b` to result bits `[127:96]`.
263///
264/// * Bits `[3:0]`: If any of these bits are set, the corresponding result
265///   element is cleared.
266///
267/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_insert_ps)
268#[inline]
269#[target_feature(enable = "sse4.1")]
270#[cfg_attr(test, assert_instr(insertps, IMM8 = 0b1010))]
271#[rustc_legacy_const_generics(2)]
272#[stable(feature = "simd_x86", since = "1.27.0")]
273pub fn _mm_insert_ps<const IMM8: i32>(a: __m128, b: __m128) -> __m128 {
274    static_assert_uimm_bits!(IMM8, 8);
275    unsafe { insertps(a, b, IMM8 as u8) }
276}
277
278/// Returns a copy of `a` with the 8-bit integer from `i` inserted at a
279/// location specified by `IMM8`.
280///
281/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_insert_epi8)
282#[inline]
283#[target_feature(enable = "sse4.1")]
284#[cfg_attr(test, assert_instr(pinsrb, IMM8 = 0))]
285#[rustc_legacy_const_generics(2)]
286#[stable(feature = "simd_x86", since = "1.27.0")]
287pub fn _mm_insert_epi8<const IMM8: i32>(a: __m128i, i: i32) -> __m128i {
288    static_assert_uimm_bits!(IMM8, 4);
289    unsafe { transmute(simd_insert!(a.as_i8x16(), IMM8 as u32, i as i8)) }
290}
291
292/// Returns a copy of `a` with the 32-bit integer from `i` inserted at a
293/// location specified by `IMM8`.
294///
295/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_insert_epi32)
296#[inline]
297#[target_feature(enable = "sse4.1")]
298#[cfg_attr(test, assert_instr(pinsrd, IMM8 = 0))]
299#[rustc_legacy_const_generics(2)]
300#[stable(feature = "simd_x86", since = "1.27.0")]
301pub fn _mm_insert_epi32<const IMM8: i32>(a: __m128i, i: i32) -> __m128i {
302    static_assert_uimm_bits!(IMM8, 2);
303    unsafe { transmute(simd_insert!(a.as_i32x4(), IMM8 as u32, i)) }
304}
305
306/// Compares packed 8-bit integers in `a` and `b` and returns packed maximum
307/// values in dst.
308///
309/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_max_epi8)
310#[inline]
311#[target_feature(enable = "sse4.1")]
312#[cfg_attr(test, assert_instr(pmaxsb))]
313#[stable(feature = "simd_x86", since = "1.27.0")]
314pub fn _mm_max_epi8(a: __m128i, b: __m128i) -> __m128i {
315    unsafe {
316        let a = a.as_i8x16();
317        let b = b.as_i8x16();
318        transmute(simd_select::<i8x16, _>(simd_gt(a, b), a, b))
319    }
320}
321
322/// Compares packed unsigned 16-bit integers in `a` and `b`, and returns packed
323/// maximum.
324///
325/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_max_epu16)
326#[inline]
327#[target_feature(enable = "sse4.1")]
328#[cfg_attr(test, assert_instr(pmaxuw))]
329#[stable(feature = "simd_x86", since = "1.27.0")]
330pub fn _mm_max_epu16(a: __m128i, b: __m128i) -> __m128i {
331    unsafe {
332        let a = a.as_u16x8();
333        let b = b.as_u16x8();
334        transmute(simd_select::<i16x8, _>(simd_gt(a, b), a, b))
335    }
336}
337
338/// Compares packed 32-bit integers in `a` and `b`, and returns packed maximum
339/// values.
340///
341/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_max_epi32)
342#[inline]
343#[target_feature(enable = "sse4.1")]
344#[cfg_attr(test, assert_instr(pmaxsd))]
345#[stable(feature = "simd_x86", since = "1.27.0")]
346pub fn _mm_max_epi32(a: __m128i, b: __m128i) -> __m128i {
347    unsafe {
348        let a = a.as_i32x4();
349        let b = b.as_i32x4();
350        transmute(simd_select::<i32x4, _>(simd_gt(a, b), a, b))
351    }
352}
353
354/// Compares packed unsigned 32-bit integers in `a` and `b`, and returns packed
355/// maximum values.
356///
357/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_max_epu32)
358#[inline]
359#[target_feature(enable = "sse4.1")]
360#[cfg_attr(test, assert_instr(pmaxud))]
361#[stable(feature = "simd_x86", since = "1.27.0")]
362pub fn _mm_max_epu32(a: __m128i, b: __m128i) -> __m128i {
363    unsafe {
364        let a = a.as_u32x4();
365        let b = b.as_u32x4();
366        transmute(simd_select::<i32x4, _>(simd_gt(a, b), a, b))
367    }
368}
369
370/// Compares packed 8-bit integers in `a` and `b` and returns packed minimum
371/// values in dst.
372///
373/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_min_epi8)
374#[inline]
375#[target_feature(enable = "sse4.1")]
376#[cfg_attr(test, assert_instr(pminsb))]
377#[stable(feature = "simd_x86", since = "1.27.0")]
378pub fn _mm_min_epi8(a: __m128i, b: __m128i) -> __m128i {
379    unsafe {
380        let a = a.as_i8x16();
381        let b = b.as_i8x16();
382        transmute(simd_select::<i8x16, _>(simd_lt(a, b), a, b))
383    }
384}
385
386/// Compares packed unsigned 16-bit integers in `a` and `b`, and returns packed
387/// minimum.
388///
389/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_min_epu16)
390#[inline]
391#[target_feature(enable = "sse4.1")]
392#[cfg_attr(test, assert_instr(pminuw))]
393#[stable(feature = "simd_x86", since = "1.27.0")]
394pub fn _mm_min_epu16(a: __m128i, b: __m128i) -> __m128i {
395    unsafe {
396        let a = a.as_u16x8();
397        let b = b.as_u16x8();
398        transmute(simd_select::<i16x8, _>(simd_lt(a, b), a, b))
399    }
400}
401
402/// Compares packed 32-bit integers in `a` and `b`, and returns packed minimum
403/// values.
404///
405/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_min_epi32)
406#[inline]
407#[target_feature(enable = "sse4.1")]
408#[cfg_attr(test, assert_instr(pminsd))]
409#[stable(feature = "simd_x86", since = "1.27.0")]
410pub fn _mm_min_epi32(a: __m128i, b: __m128i) -> __m128i {
411    unsafe {
412        let a = a.as_i32x4();
413        let b = b.as_i32x4();
414        transmute(simd_select::<i32x4, _>(simd_lt(a, b), a, b))
415    }
416}
417
418/// Compares packed unsigned 32-bit integers in `a` and `b`, and returns packed
419/// minimum values.
420///
421/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_min_epu32)
422#[inline]
423#[target_feature(enable = "sse4.1")]
424#[cfg_attr(test, assert_instr(pminud))]
425#[stable(feature = "simd_x86", since = "1.27.0")]
426pub fn _mm_min_epu32(a: __m128i, b: __m128i) -> __m128i {
427    unsafe {
428        let a = a.as_u32x4();
429        let b = b.as_u32x4();
430        transmute(simd_select::<i32x4, _>(simd_lt(a, b), a, b))
431    }
432}
433
434/// Converts packed 32-bit integers from `a` and `b` to packed 16-bit integers
435/// using unsigned saturation
436///
437/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_packus_epi32)
438#[inline]
439#[target_feature(enable = "sse4.1")]
440#[cfg_attr(test, assert_instr(packusdw))]
441#[stable(feature = "simd_x86", since = "1.27.0")]
442pub fn _mm_packus_epi32(a: __m128i, b: __m128i) -> __m128i {
443    unsafe { transmute(packusdw(a.as_i32x4(), b.as_i32x4())) }
444}
445
446/// Compares packed 64-bit integers in `a` and `b` for equality
447///
448/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cmpeq_epi64)
449#[inline]
450#[target_feature(enable = "sse4.1")]
451#[cfg_attr(test, assert_instr(pcmpeqq))]
452#[stable(feature = "simd_x86", since = "1.27.0")]
453pub fn _mm_cmpeq_epi64(a: __m128i, b: __m128i) -> __m128i {
454    unsafe { transmute(simd_eq::<_, i64x2>(a.as_i64x2(), b.as_i64x2())) }
455}
456
457/// Sign extend packed 8-bit integers in `a` to packed 16-bit integers
458///
459/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi8_epi16)
460#[inline]
461#[target_feature(enable = "sse4.1")]
462#[cfg_attr(test, assert_instr(pmovsxbw))]
463#[stable(feature = "simd_x86", since = "1.27.0")]
464pub fn _mm_cvtepi8_epi16(a: __m128i) -> __m128i {
465    unsafe {
466        let a = a.as_i8x16();
467        let a: i8x8 = simd_shuffle!(a, a, [0, 1, 2, 3, 4, 5, 6, 7]);
468        transmute(simd_cast::<_, i16x8>(a))
469    }
470}
471
472/// Sign extend packed 8-bit integers in `a` to packed 32-bit integers
473///
474/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi8_epi32)
475#[inline]
476#[target_feature(enable = "sse4.1")]
477#[cfg_attr(test, assert_instr(pmovsxbd))]
478#[stable(feature = "simd_x86", since = "1.27.0")]
479pub fn _mm_cvtepi8_epi32(a: __m128i) -> __m128i {
480    unsafe {
481        let a = a.as_i8x16();
482        let a: i8x4 = simd_shuffle!(a, a, [0, 1, 2, 3]);
483        transmute(simd_cast::<_, i32x4>(a))
484    }
485}
486
487/// Sign extend packed 8-bit integers in the low 8 bytes of `a` to packed
488/// 64-bit integers
489///
490/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi8_epi64)
491#[inline]
492#[target_feature(enable = "sse4.1")]
493#[cfg_attr(test, assert_instr(pmovsxbq))]
494#[stable(feature = "simd_x86", since = "1.27.0")]
495pub fn _mm_cvtepi8_epi64(a: __m128i) -> __m128i {
496    unsafe {
497        let a = a.as_i8x16();
498        let a: i8x2 = simd_shuffle!(a, a, [0, 1]);
499        transmute(simd_cast::<_, i64x2>(a))
500    }
501}
502
503/// Sign extend packed 16-bit integers in `a` to packed 32-bit integers
504///
505/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi16_epi32)
506#[inline]
507#[target_feature(enable = "sse4.1")]
508#[cfg_attr(test, assert_instr(pmovsxwd))]
509#[stable(feature = "simd_x86", since = "1.27.0")]
510pub fn _mm_cvtepi16_epi32(a: __m128i) -> __m128i {
511    unsafe {
512        let a = a.as_i16x8();
513        let a: i16x4 = simd_shuffle!(a, a, [0, 1, 2, 3]);
514        transmute(simd_cast::<_, i32x4>(a))
515    }
516}
517
518/// Sign extend packed 16-bit integers in `a` to packed 64-bit integers
519///
520/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi16_epi64)
521#[inline]
522#[target_feature(enable = "sse4.1")]
523#[cfg_attr(test, assert_instr(pmovsxwq))]
524#[stable(feature = "simd_x86", since = "1.27.0")]
525pub fn _mm_cvtepi16_epi64(a: __m128i) -> __m128i {
526    unsafe {
527        let a = a.as_i16x8();
528        let a: i16x2 = simd_shuffle!(a, a, [0, 1]);
529        transmute(simd_cast::<_, i64x2>(a))
530    }
531}
532
533/// Sign extend packed 32-bit integers in `a` to packed 64-bit integers
534///
535/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepi32_epi64)
536#[inline]
537#[target_feature(enable = "sse4.1")]
538#[cfg_attr(test, assert_instr(pmovsxdq))]
539#[stable(feature = "simd_x86", since = "1.27.0")]
540pub fn _mm_cvtepi32_epi64(a: __m128i) -> __m128i {
541    unsafe {
542        let a = a.as_i32x4();
543        let a: i32x2 = simd_shuffle!(a, a, [0, 1]);
544        transmute(simd_cast::<_, i64x2>(a))
545    }
546}
547
548/// Zeroes extend packed unsigned 8-bit integers in `a` to packed 16-bit integers
549///
550/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu8_epi16)
551#[inline]
552#[target_feature(enable = "sse4.1")]
553#[cfg_attr(test, assert_instr(pmovzxbw))]
554#[stable(feature = "simd_x86", since = "1.27.0")]
555pub fn _mm_cvtepu8_epi16(a: __m128i) -> __m128i {
556    unsafe {
557        let a = a.as_u8x16();
558        let a: u8x8 = simd_shuffle!(a, a, [0, 1, 2, 3, 4, 5, 6, 7]);
559        transmute(simd_cast::<_, i16x8>(a))
560    }
561}
562
563/// Zeroes extend packed unsigned 8-bit integers in `a` to packed 32-bit integers
564///
565/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu8_epi32)
566#[inline]
567#[target_feature(enable = "sse4.1")]
568#[cfg_attr(test, assert_instr(pmovzxbd))]
569#[stable(feature = "simd_x86", since = "1.27.0")]
570pub fn _mm_cvtepu8_epi32(a: __m128i) -> __m128i {
571    unsafe {
572        let a = a.as_u8x16();
573        let a: u8x4 = simd_shuffle!(a, a, [0, 1, 2, 3]);
574        transmute(simd_cast::<_, i32x4>(a))
575    }
576}
577
578/// Zeroes extend packed unsigned 8-bit integers in `a` to packed 64-bit integers
579///
580/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu8_epi64)
581#[inline]
582#[target_feature(enable = "sse4.1")]
583#[cfg_attr(test, assert_instr(pmovzxbq))]
584#[stable(feature = "simd_x86", since = "1.27.0")]
585pub fn _mm_cvtepu8_epi64(a: __m128i) -> __m128i {
586    unsafe {
587        let a = a.as_u8x16();
588        let a: u8x2 = simd_shuffle!(a, a, [0, 1]);
589        transmute(simd_cast::<_, i64x2>(a))
590    }
591}
592
593/// Zeroes extend packed unsigned 16-bit integers in `a`
594/// to packed 32-bit integers
595///
596/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu16_epi32)
597#[inline]
598#[target_feature(enable = "sse4.1")]
599#[cfg_attr(test, assert_instr(pmovzxwd))]
600#[stable(feature = "simd_x86", since = "1.27.0")]
601pub fn _mm_cvtepu16_epi32(a: __m128i) -> __m128i {
602    unsafe {
603        let a = a.as_u16x8();
604        let a: u16x4 = simd_shuffle!(a, a, [0, 1, 2, 3]);
605        transmute(simd_cast::<_, i32x4>(a))
606    }
607}
608
609/// Zeroes extend packed unsigned 16-bit integers in `a`
610/// to packed 64-bit integers
611///
612/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu16_epi64)
613#[inline]
614#[target_feature(enable = "sse4.1")]
615#[cfg_attr(test, assert_instr(pmovzxwq))]
616#[stable(feature = "simd_x86", since = "1.27.0")]
617pub fn _mm_cvtepu16_epi64(a: __m128i) -> __m128i {
618    unsafe {
619        let a = a.as_u16x8();
620        let a: u16x2 = simd_shuffle!(a, a, [0, 1]);
621        transmute(simd_cast::<_, i64x2>(a))
622    }
623}
624
625/// Zeroes extend packed unsigned 32-bit integers in `a`
626/// to packed 64-bit integers
627///
628/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtepu32_epi64)
629#[inline]
630#[target_feature(enable = "sse4.1")]
631#[cfg_attr(test, assert_instr(pmovzxdq))]
632#[stable(feature = "simd_x86", since = "1.27.0")]
633pub fn _mm_cvtepu32_epi64(a: __m128i) -> __m128i {
634    unsafe {
635        let a = a.as_u32x4();
636        let a: u32x2 = simd_shuffle!(a, a, [0, 1]);
637        transmute(simd_cast::<_, i64x2>(a))
638    }
639}
640
641/// Returns the dot product of two __m128d vectors.
642///
643/// `IMM8[1:0]` is the broadcast mask, and `IMM8[5:4]` is the condition mask.
644/// If a condition mask bit is zero, the corresponding multiplication is
645/// replaced by a value of `0.0`. If a broadcast mask bit is one, the result of
646/// the dot product will be stored in the return value component. Otherwise if
647/// the broadcast mask bit is zero then the return component will be zero.
648///
649/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_dp_pd)
650#[inline]
651#[target_feature(enable = "sse4.1")]
652#[cfg_attr(test, assert_instr(dppd, IMM8 = 0))]
653#[rustc_legacy_const_generics(2)]
654#[stable(feature = "simd_x86", since = "1.27.0")]
655pub fn _mm_dp_pd<const IMM8: i32>(a: __m128d, b: __m128d) -> __m128d {
656    unsafe {
657        static_assert_uimm_bits!(IMM8, 8);
658        dppd(a, b, IMM8 as u8)
659    }
660}
661
662/// Returns the dot product of two __m128 vectors.
663///
664/// `IMM8[3:0]` is the broadcast mask, and `IMM8[7:4]` is the condition mask.
665/// If a condition mask bit is zero, the corresponding multiplication is
666/// replaced by a value of `0.0`. If a broadcast mask bit is one, the result of
667/// the dot product will be stored in the return value component. Otherwise if
668/// the broadcast mask bit is zero then the return component will be zero.
669///
670/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_dp_ps)
671#[inline]
672#[target_feature(enable = "sse4.1")]
673#[cfg_attr(test, assert_instr(dpps, IMM8 = 0))]
674#[rustc_legacy_const_generics(2)]
675#[stable(feature = "simd_x86", since = "1.27.0")]
676pub fn _mm_dp_ps<const IMM8: i32>(a: __m128, b: __m128) -> __m128 {
677    static_assert_uimm_bits!(IMM8, 8);
678    unsafe { dpps(a, b, IMM8 as u8) }
679}
680
681/// Round the packed double-precision (64-bit) floating-point elements in `a`
682/// down to an integer value, and stores the results as packed double-precision
683/// floating-point elements.
684///
685/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_floor_pd)
686#[inline]
687#[target_feature(enable = "sse4.1")]
688#[cfg_attr(test, assert_instr(roundpd))]
689#[stable(feature = "simd_x86", since = "1.27.0")]
690pub fn _mm_floor_pd(a: __m128d) -> __m128d {
691    unsafe { simd_floor(a) }
692}
693
694/// Round the packed single-precision (32-bit) floating-point elements in `a`
695/// down to an integer value, and stores the results as packed single-precision
696/// floating-point elements.
697///
698/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_floor_ps)
699#[inline]
700#[target_feature(enable = "sse4.1")]
701#[cfg_attr(test, assert_instr(roundps))]
702#[stable(feature = "simd_x86", since = "1.27.0")]
703pub fn _mm_floor_ps(a: __m128) -> __m128 {
704    unsafe { simd_floor(a) }
705}
706
707/// Round the lower double-precision (64-bit) floating-point element in `b`
708/// down to an integer value, store the result as a double-precision
709/// floating-point element in the lower element of the intrinsic result,
710/// and copies the upper element from `a` to the upper element of the intrinsic
711/// result.
712///
713/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_floor_sd)
714#[inline]
715#[target_feature(enable = "sse4.1")]
716#[cfg_attr(test, assert_instr(roundsd))]
717#[stable(feature = "simd_x86", since = "1.27.0")]
718pub fn _mm_floor_sd(a: __m128d, b: __m128d) -> __m128d {
719    unsafe { roundsd(a, b, _MM_FROUND_FLOOR) }
720}
721
722/// Round the lower single-precision (32-bit) floating-point element in `b`
723/// down to an integer value, store the result as a single-precision
724/// floating-point element in the lower element of the intrinsic result,
725/// and copies the upper 3 packed elements from `a` to the upper elements
726/// of the intrinsic result.
727///
728/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_floor_ss)
729#[inline]
730#[target_feature(enable = "sse4.1")]
731#[cfg_attr(test, assert_instr(roundss))]
732#[stable(feature = "simd_x86", since = "1.27.0")]
733pub fn _mm_floor_ss(a: __m128, b: __m128) -> __m128 {
734    unsafe { roundss(a, b, _MM_FROUND_FLOOR) }
735}
736
737/// Round the packed double-precision (64-bit) floating-point elements in `a`
738/// up to an integer value, and stores the results as packed double-precision
739/// floating-point elements.
740///
741/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_ceil_pd)
742#[inline]
743#[target_feature(enable = "sse4.1")]
744#[cfg_attr(test, assert_instr(roundpd))]
745#[stable(feature = "simd_x86", since = "1.27.0")]
746pub fn _mm_ceil_pd(a: __m128d) -> __m128d {
747    unsafe { simd_ceil(a) }
748}
749
750/// Round the packed single-precision (32-bit) floating-point elements in `a`
751/// up to an integer value, and stores the results as packed single-precision
752/// floating-point elements.
753///
754/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_ceil_ps)
755#[inline]
756#[target_feature(enable = "sse4.1")]
757#[cfg_attr(test, assert_instr(roundps))]
758#[stable(feature = "simd_x86", since = "1.27.0")]
759pub fn _mm_ceil_ps(a: __m128) -> __m128 {
760    unsafe { simd_ceil(a) }
761}
762
763/// Round the lower double-precision (64-bit) floating-point element in `b`
764/// up to an integer value, store the result as a double-precision
765/// floating-point element in the lower element of the intrinsic result,
766/// and copies the upper element from `a` to the upper element
767/// of the intrinsic result.
768///
769/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_ceil_sd)
770#[inline]
771#[target_feature(enable = "sse4.1")]
772#[cfg_attr(test, assert_instr(roundsd))]
773#[stable(feature = "simd_x86", since = "1.27.0")]
774pub fn _mm_ceil_sd(a: __m128d, b: __m128d) -> __m128d {
775    unsafe { roundsd(a, b, _MM_FROUND_CEIL) }
776}
777
778/// Round the lower single-precision (32-bit) floating-point element in `b`
779/// up to an integer value, store the result as a single-precision
780/// floating-point element in the lower element of the intrinsic result,
781/// and copies the upper 3 packed elements from `a` to the upper elements
782/// of the intrinsic result.
783///
784/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_ceil_ss)
785#[inline]
786#[target_feature(enable = "sse4.1")]
787#[cfg_attr(test, assert_instr(roundss))]
788#[stable(feature = "simd_x86", since = "1.27.0")]
789pub fn _mm_ceil_ss(a: __m128, b: __m128) -> __m128 {
790    unsafe { roundss(a, b, _MM_FROUND_CEIL) }
791}
792
793/// Round the packed double-precision (64-bit) floating-point elements in `a`
794/// using the `ROUNDING` parameter, and stores the results as packed
795/// double-precision floating-point elements.
796/// Rounding is done according to the rounding parameter, which can be one of:
797///
798/// * [`_MM_FROUND_TO_NEAREST_INT`] | [`_MM_FROUND_NO_EXC`] : round to nearest and suppress exceptions
799/// * [`_MM_FROUND_TO_NEG_INF`] | [`_MM_FROUND_NO_EXC`] : round down and suppress exceptions
800/// * [`_MM_FROUND_TO_POS_INF`] | [`_MM_FROUND_NO_EXC`] : round up and suppress exceptions
801/// * [`_MM_FROUND_TO_ZERO`] | [`_MM_FROUND_NO_EXC`] : truncate and suppress exceptions
802/// * [`_MM_FROUND_CUR_DIRECTION`] : use `MXCSR.RC` - see [`_MM_SET_ROUNDING_MODE`]
803///
804/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_round_pd)
805#[inline]
806#[target_feature(enable = "sse4.1")]
807#[cfg_attr(test, assert_instr(roundpd, ROUNDING = 0))]
808#[rustc_legacy_const_generics(1)]
809#[stable(feature = "simd_x86", since = "1.27.0")]
810pub fn _mm_round_pd<const ROUNDING: i32>(a: __m128d) -> __m128d {
811    static_assert_uimm_bits!(ROUNDING, 4);
812    unsafe { roundpd(a, ROUNDING) }
813}
814
815/// Round the packed single-precision (32-bit) floating-point elements in `a`
816/// using the `ROUNDING` parameter, and stores the results as packed
817/// single-precision floating-point elements.
818/// Rounding is done according to the rounding parameter, which can be one of:
819///
820/// * [`_MM_FROUND_TO_NEAREST_INT`] | [`_MM_FROUND_NO_EXC`] : round to nearest and suppress exceptions
821/// * [`_MM_FROUND_TO_NEG_INF`] | [`_MM_FROUND_NO_EXC`] : round down and suppress exceptions
822/// * [`_MM_FROUND_TO_POS_INF`] | [`_MM_FROUND_NO_EXC`] : round up and suppress exceptions
823/// * [`_MM_FROUND_TO_ZERO`] | [`_MM_FROUND_NO_EXC`] : truncate and suppress exceptions
824/// * [`_MM_FROUND_CUR_DIRECTION`] : use `MXCSR.RC` - see [`_MM_SET_ROUNDING_MODE`]
825///
826/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_round_ps)
827#[inline]
828#[target_feature(enable = "sse4.1")]
829#[cfg_attr(test, assert_instr(roundps, ROUNDING = 0))]
830#[rustc_legacy_const_generics(1)]
831#[stable(feature = "simd_x86", since = "1.27.0")]
832pub fn _mm_round_ps<const ROUNDING: i32>(a: __m128) -> __m128 {
833    static_assert_uimm_bits!(ROUNDING, 4);
834    unsafe { roundps(a, ROUNDING) }
835}
836
837/// Round the lower double-precision (64-bit) floating-point element in `b`
838/// using the `ROUNDING` parameter, store the result as a double-precision
839/// floating-point element in the lower element of the intrinsic result,
840/// and copies the upper element from `a` to the upper element of the intrinsic
841/// result.
842/// Rounding is done according to the rounding parameter, which can be one of:
843///
844/// * [`_MM_FROUND_TO_NEAREST_INT`] | [`_MM_FROUND_NO_EXC`] : round to nearest and suppress exceptions
845/// * [`_MM_FROUND_TO_NEG_INF`] | [`_MM_FROUND_NO_EXC`] : round down and suppress exceptions
846/// * [`_MM_FROUND_TO_POS_INF`] | [`_MM_FROUND_NO_EXC`] : round up and suppress exceptions
847/// * [`_MM_FROUND_TO_ZERO`] | [`_MM_FROUND_NO_EXC`] : truncate and suppress exceptions
848/// * [`_MM_FROUND_CUR_DIRECTION`] : use `MXCSR.RC` - see [`_MM_SET_ROUNDING_MODE`]
849///
850/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_round_sd)
851#[inline]
852#[target_feature(enable = "sse4.1")]
853#[cfg_attr(test, assert_instr(roundsd, ROUNDING = 0))]
854#[rustc_legacy_const_generics(2)]
855#[stable(feature = "simd_x86", since = "1.27.0")]
856pub fn _mm_round_sd<const ROUNDING: i32>(a: __m128d, b: __m128d) -> __m128d {
857    static_assert_uimm_bits!(ROUNDING, 4);
858    unsafe { roundsd(a, b, ROUNDING) }
859}
860
861/// Round the lower single-precision (32-bit) floating-point element in `b`
862/// using the `ROUNDING` parameter, store the result as a single-precision
863/// floating-point element in the lower element of the intrinsic result,
864/// and copies the upper 3 packed elements from `a` to the upper elements
865/// of the intrinsic result.
866/// Rounding is done according to the rounding parameter, which can be one of:
867///
868/// * [`_MM_FROUND_TO_NEAREST_INT`] | [`_MM_FROUND_NO_EXC`] : round to nearest and suppress exceptions
869/// * [`_MM_FROUND_TO_NEG_INF`] | [`_MM_FROUND_NO_EXC`] : round down and suppress exceptions
870/// * [`_MM_FROUND_TO_POS_INF`] | [`_MM_FROUND_NO_EXC`] : round up and suppress exceptions
871/// * [`_MM_FROUND_TO_ZERO`] | [`_MM_FROUND_NO_EXC`] : truncate and suppress exceptions
872/// * [`_MM_FROUND_CUR_DIRECTION`] : use `MXCSR.RC` - see [`_MM_SET_ROUNDING_MODE`]
873///
874/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_round_ss)
875#[inline]
876#[target_feature(enable = "sse4.1")]
877#[cfg_attr(test, assert_instr(roundss, ROUNDING = 0))]
878#[rustc_legacy_const_generics(2)]
879#[stable(feature = "simd_x86", since = "1.27.0")]
880pub fn _mm_round_ss<const ROUNDING: i32>(a: __m128, b: __m128) -> __m128 {
881    static_assert_uimm_bits!(ROUNDING, 4);
882    unsafe { roundss(a, b, ROUNDING) }
883}
884
885/// Finds the minimum unsigned 16-bit element in the 128-bit __m128i vector,
886/// returning a vector containing its value in its first position, and its
887/// index
888/// in its second position; all other elements are set to zero.
889///
890/// This intrinsic corresponds to the `VPHMINPOSUW` / `PHMINPOSUW`
891/// instruction.
892///
893/// Arguments:
894///
895/// * `a` - A 128-bit vector of type `__m128i`.
896///
897/// Returns:
898///
899/// A 128-bit value where:
900///
901/// * bits `[15:0]` - contain the minimum value found in parameter `a`,
902/// * bits `[18:16]` - contain the index of the minimum value
903/// * remaining bits are set to `0`.
904///
905/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_minpos_epu16)
906#[inline]
907#[target_feature(enable = "sse4.1")]
908#[cfg_attr(test, assert_instr(phminposuw))]
909#[stable(feature = "simd_x86", since = "1.27.0")]
910pub fn _mm_minpos_epu16(a: __m128i) -> __m128i {
911    unsafe { transmute(phminposuw(a.as_u16x8())) }
912}
913
914/// Multiplies the low 32-bit integers from each packed 64-bit
915/// element in `a` and `b`, and returns the signed 64-bit result.
916///
917/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mul_epi32)
918#[inline]
919#[target_feature(enable = "sse4.1")]
920#[cfg_attr(test, assert_instr(pmuldq))]
921#[stable(feature = "simd_x86", since = "1.27.0")]
922pub fn _mm_mul_epi32(a: __m128i, b: __m128i) -> __m128i {
923    unsafe {
924        let a = simd_cast::<_, i64x2>(simd_cast::<_, i32x2>(a.as_i64x2()));
925        let b = simd_cast::<_, i64x2>(simd_cast::<_, i32x2>(b.as_i64x2()));
926        transmute(simd_mul(a, b))
927    }
928}
929
930/// Multiplies the packed 32-bit integers in `a` and `b`, producing intermediate
931/// 64-bit integers, and returns the lowest 32-bit, whatever they might be,
932/// reinterpreted as a signed integer. While `pmulld __m128i::splat(2),
933/// __m128i::splat(2)` returns the obvious `__m128i::splat(4)`, due to wrapping
934/// arithmetic `pmulld __m128i::splat(i32::MAX), __m128i::splat(2)` would
935/// return a negative number.
936///
937/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mullo_epi32)
938#[inline]
939#[target_feature(enable = "sse4.1")]
940#[cfg_attr(test, assert_instr(pmulld))]
941#[stable(feature = "simd_x86", since = "1.27.0")]
942pub fn _mm_mullo_epi32(a: __m128i, b: __m128i) -> __m128i {
943    unsafe { transmute(simd_mul(a.as_i32x4(), b.as_i32x4())) }
944}
945
946/// Subtracts 8-bit unsigned integer values and computes the absolute
947/// values of the differences to the corresponding bits in the destination.
948/// Then sums of the absolute differences are returned according to the bit
949/// fields in the immediate operand.
950///
951/// The following algorithm is performed:
952///
953/// ```ignore
954/// i = IMM8[2] * 4
955/// j = IMM8[1:0] * 4
956/// for k := 0 to 7
957///     d0 = abs(a[i + k + 0] - b[j + 0])
958///     d1 = abs(a[i + k + 1] - b[j + 1])
959///     d2 = abs(a[i + k + 2] - b[j + 2])
960///     d3 = abs(a[i + k + 3] - b[j + 3])
961///     r[k] = d0 + d1 + d2 + d3
962/// ```
963///
964/// Arguments:
965///
966/// * `a` - A 128-bit vector of type `__m128i`.
967/// * `b` - A 128-bit vector of type `__m128i`.
968/// * `IMM8` - An 8-bit immediate operand specifying how the absolute
969///   differences are to be calculated
970///     * Bit `[2]` specify the offset for operand `a`
971///     * Bits `[1:0]` specify the offset for operand `b`
972///
973/// Returns:
974///
975/// * A `__m128i` vector containing the sums of the sets of   absolute
976///   differences between both operands.
977///
978/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mpsadbw_epu8)
979#[inline]
980#[target_feature(enable = "sse4.1")]
981#[cfg_attr(test, assert_instr(mpsadbw, IMM8 = 0))]
982#[rustc_legacy_const_generics(2)]
983#[stable(feature = "simd_x86", since = "1.27.0")]
984pub fn _mm_mpsadbw_epu8<const IMM8: i32>(a: __m128i, b: __m128i) -> __m128i {
985    static_assert_uimm_bits!(IMM8, 3);
986    unsafe { transmute(mpsadbw(a.as_u8x16(), b.as_u8x16(), IMM8 as u8)) }
987}
988
989/// Tests whether the specified bits in a 128-bit integer vector are all
990/// zeros.
991///
992/// Arguments:
993///
994/// * `a` - A 128-bit integer vector containing the bits to be tested.
995/// * `mask` - A 128-bit integer vector selecting which bits to test in
996///   operand `a`.
997///
998/// Returns:
999///
1000/// * `1` - if the specified bits are all zeros,
1001/// * `0` - otherwise.
1002///
1003/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_testz_si128)
1004#[inline]
1005#[target_feature(enable = "sse4.1")]
1006#[cfg_attr(test, assert_instr(ptest))]
1007#[stable(feature = "simd_x86", since = "1.27.0")]
1008pub fn _mm_testz_si128(a: __m128i, mask: __m128i) -> i32 {
1009    unsafe { ptestz(a.as_i64x2(), mask.as_i64x2()) }
1010}
1011
1012/// Tests whether the specified bits in a 128-bit integer vector are all
1013/// ones.
1014///
1015/// Arguments:
1016///
1017/// * `a` - A 128-bit integer vector containing the bits to be tested.
1018/// * `mask` - A 128-bit integer vector selecting which bits to test in
1019///   operand `a`.
1020///
1021/// Returns:
1022///
1023/// * `1` - if the specified bits are all ones,
1024/// * `0` - otherwise.
1025///
1026/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_testc_si128)
1027#[inline]
1028#[target_feature(enable = "sse4.1")]
1029#[cfg_attr(test, assert_instr(ptest))]
1030#[stable(feature = "simd_x86", since = "1.27.0")]
1031pub fn _mm_testc_si128(a: __m128i, mask: __m128i) -> i32 {
1032    unsafe { ptestc(a.as_i64x2(), mask.as_i64x2()) }
1033}
1034
1035/// Tests whether the specified bits in a 128-bit integer vector are
1036/// neither all zeros nor all ones.
1037///
1038/// Arguments:
1039///
1040/// * `a` - A 128-bit integer vector containing the bits to be tested.
1041/// * `mask` - A 128-bit integer vector selecting which bits to test in
1042///   operand `a`.
1043///
1044/// Returns:
1045///
1046/// * `1` - if the specified bits are neither all zeros nor all ones,
1047/// * `0` - otherwise.
1048///
1049/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_testnzc_si128)
1050#[inline]
1051#[target_feature(enable = "sse4.1")]
1052#[cfg_attr(test, assert_instr(ptest))]
1053#[stable(feature = "simd_x86", since = "1.27.0")]
1054pub fn _mm_testnzc_si128(a: __m128i, mask: __m128i) -> i32 {
1055    unsafe { ptestnzc(a.as_i64x2(), mask.as_i64x2()) }
1056}
1057
1058/// Tests whether the specified bits in a 128-bit integer vector are all
1059/// zeros.
1060///
1061/// Arguments:
1062///
1063/// * `a` - A 128-bit integer vector containing the bits to be tested.
1064/// * `mask` - A 128-bit integer vector selecting which bits to test in
1065///   operand `a`.
1066///
1067/// Returns:
1068///
1069/// * `1` - if the specified bits are all zeros,
1070/// * `0` - otherwise.
1071///
1072/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_test_all_zeros)
1073#[inline]
1074#[target_feature(enable = "sse4.1")]
1075#[cfg_attr(test, assert_instr(ptest))]
1076#[stable(feature = "simd_x86", since = "1.27.0")]
1077pub fn _mm_test_all_zeros(a: __m128i, mask: __m128i) -> i32 {
1078    _mm_testz_si128(a, mask)
1079}
1080
1081/// Tests whether the specified bits in `a` 128-bit integer vector are all
1082/// ones.
1083///
1084/// Argument:
1085///
1086/// * `a` - A 128-bit integer vector containing the bits to be tested.
1087///
1088/// Returns:
1089///
1090/// * `1` - if the bits specified in the operand are all set to 1,
1091/// * `0` - otherwise.
1092///
1093/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_test_all_ones)
1094#[inline]
1095#[target_feature(enable = "sse4.1")]
1096#[cfg_attr(test, assert_instr(pcmpeqd))]
1097#[cfg_attr(test, assert_instr(ptest))]
1098#[stable(feature = "simd_x86", since = "1.27.0")]
1099pub fn _mm_test_all_ones(a: __m128i) -> i32 {
1100    _mm_testc_si128(a, _mm_cmpeq_epi32(a, a))
1101}
1102
1103/// Tests whether the specified bits in a 128-bit integer vector are
1104/// neither all zeros nor all ones.
1105///
1106/// Arguments:
1107///
1108/// * `a` - A 128-bit integer vector containing the bits to be tested.
1109/// * `mask` - A 128-bit integer vector selecting which bits to test in
1110///   operand `a`.
1111///
1112/// Returns:
1113///
1114/// * `1` - if the specified bits are neither all zeros nor all ones,
1115/// * `0` - otherwise.
1116///
1117/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_test_mix_ones_zeros)
1118#[inline]
1119#[target_feature(enable = "sse4.1")]
1120#[cfg_attr(test, assert_instr(ptest))]
1121#[stable(feature = "simd_x86", since = "1.27.0")]
1122pub fn _mm_test_mix_ones_zeros(a: __m128i, mask: __m128i) -> i32 {
1123    _mm_testnzc_si128(a, mask)
1124}
1125
1126/// Load 128-bits of integer data from memory into dst. mem_addr must be aligned on a 16-byte
1127/// boundary or a general-protection exception may be generated. To minimize caching, the data
1128/// is flagged as non-temporal (unlikely to be used again soon)
1129///
1130/// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_stream_load_si128)
1131#[inline]
1132#[target_feature(enable = "sse4.1")]
1133#[cfg_attr(test, assert_instr(movntdqa))]
1134#[stable(feature = "simd_x86_updates", since = "1.82.0")]
1135pub unsafe fn _mm_stream_load_si128(mem_addr: *const __m128i) -> __m128i {
1136    let dst: __m128i;
1137    crate::arch::asm!(
1138        vpl!("movntdqa {a}"),
1139        a = out(xmm_reg) dst,
1140        p = in(reg) mem_addr,
1141        options(pure, readonly, nostack, preserves_flags),
1142    );
1143    dst
1144}
1145
1146#[allow(improper_ctypes)]
1147unsafe extern "C" {
1148    #[link_name = "llvm.x86.sse41.insertps"]
1149    fn insertps(a: __m128, b: __m128, imm8: u8) -> __m128;
1150    #[link_name = "llvm.x86.sse41.packusdw"]
1151    fn packusdw(a: i32x4, b: i32x4) -> u16x8;
1152    #[link_name = "llvm.x86.sse41.dppd"]
1153    fn dppd(a: __m128d, b: __m128d, imm8: u8) -> __m128d;
1154    #[link_name = "llvm.x86.sse41.dpps"]
1155    fn dpps(a: __m128, b: __m128, imm8: u8) -> __m128;
1156    #[link_name = "llvm.x86.sse41.round.pd"]
1157    fn roundpd(a: __m128d, rounding: i32) -> __m128d;
1158    #[link_name = "llvm.x86.sse41.round.ps"]
1159    fn roundps(a: __m128, rounding: i32) -> __m128;
1160    #[link_name = "llvm.x86.sse41.round.sd"]
1161    fn roundsd(a: __m128d, b: __m128d, rounding: i32) -> __m128d;
1162    #[link_name = "llvm.x86.sse41.round.ss"]
1163    fn roundss(a: __m128, b: __m128, rounding: i32) -> __m128;
1164    #[link_name = "llvm.x86.sse41.phminposuw"]
1165    fn phminposuw(a: u16x8) -> u16x8;
1166    #[link_name = "llvm.x86.sse41.mpsadbw"]
1167    fn mpsadbw(a: u8x16, b: u8x16, imm8: u8) -> u16x8;
1168    #[link_name = "llvm.x86.sse41.ptestz"]
1169    fn ptestz(a: i64x2, mask: i64x2) -> i32;
1170    #[link_name = "llvm.x86.sse41.ptestc"]
1171    fn ptestc(a: i64x2, mask: i64x2) -> i32;
1172    #[link_name = "llvm.x86.sse41.ptestnzc"]
1173    fn ptestnzc(a: i64x2, mask: i64x2) -> i32;
1174}
1175
1176#[cfg(test)]
1177mod tests {
1178    use crate::core_arch::x86::*;
1179    use std::mem;
1180    use stdarch_test::simd_test;
1181
1182    #[simd_test(enable = "sse4.1")]
1183    unsafe fn test_mm_blendv_epi8() {
1184        #[rustfmt::skip]
1185        let a = _mm_setr_epi8(
1186            0, 1, 2, 3, 4, 5, 6, 7,
1187            8, 9, 10, 11, 12, 13, 14, 15,
1188        );
1189        #[rustfmt::skip]
1190        let b = _mm_setr_epi8(
1191            16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
1192        );
1193        #[rustfmt::skip]
1194        let mask = _mm_setr_epi8(
1195            0, -1, 0, -1, 0, -1, 0, -1,
1196            0, -1, 0, -1, 0, -1, 0, -1,
1197        );
1198        #[rustfmt::skip]
1199        let e = _mm_setr_epi8(
1200            0, 17, 2, 19, 4, 21, 6, 23, 8, 25, 10, 27, 12, 29, 14, 31,
1201        );
1202        assert_eq_m128i(_mm_blendv_epi8(a, b, mask), e);
1203    }
1204
1205    #[simd_test(enable = "sse4.1")]
1206    unsafe fn test_mm_blendv_pd() {
1207        let a = _mm_set1_pd(0.0);
1208        let b = _mm_set1_pd(1.0);
1209        let mask = transmute(_mm_setr_epi64x(0, -1));
1210        let r = _mm_blendv_pd(a, b, mask);
1211        let e = _mm_setr_pd(0.0, 1.0);
1212        assert_eq_m128d(r, e);
1213    }
1214
1215    #[simd_test(enable = "sse4.1")]
1216    unsafe fn test_mm_blendv_ps() {
1217        let a = _mm_set1_ps(0.0);
1218        let b = _mm_set1_ps(1.0);
1219        let mask = transmute(_mm_setr_epi32(0, -1, 0, -1));
1220        let r = _mm_blendv_ps(a, b, mask);
1221        let e = _mm_setr_ps(0.0, 1.0, 0.0, 1.0);
1222        assert_eq_m128(r, e);
1223    }
1224
1225    #[simd_test(enable = "sse4.1")]
1226    unsafe fn test_mm_blend_pd() {
1227        let a = _mm_set1_pd(0.0);
1228        let b = _mm_set1_pd(1.0);
1229        let r = _mm_blend_pd::<0b10>(a, b);
1230        let e = _mm_setr_pd(0.0, 1.0);
1231        assert_eq_m128d(r, e);
1232    }
1233
1234    #[simd_test(enable = "sse4.1")]
1235    unsafe fn test_mm_blend_ps() {
1236        let a = _mm_set1_ps(0.0);
1237        let b = _mm_set1_ps(1.0);
1238        let r = _mm_blend_ps::<0b1010>(a, b);
1239        let e = _mm_setr_ps(0.0, 1.0, 0.0, 1.0);
1240        assert_eq_m128(r, e);
1241    }
1242
1243    #[simd_test(enable = "sse4.1")]
1244    unsafe fn test_mm_blend_epi16() {
1245        let a = _mm_set1_epi16(0);
1246        let b = _mm_set1_epi16(1);
1247        let r = _mm_blend_epi16::<0b1010_1100>(a, b);
1248        let e = _mm_setr_epi16(0, 0, 1, 1, 0, 1, 0, 1);
1249        assert_eq_m128i(r, e);
1250    }
1251
1252    #[simd_test(enable = "sse4.1")]
1253    unsafe fn test_mm_extract_ps() {
1254        let a = _mm_setr_ps(0.0, 1.0, 2.0, 3.0);
1255        let r: f32 = f32::from_bits(_mm_extract_ps::<1>(a) as u32);
1256        assert_eq!(r, 1.0);
1257        let r: f32 = f32::from_bits(_mm_extract_ps::<3>(a) as u32);
1258        assert_eq!(r, 3.0);
1259    }
1260
1261    #[simd_test(enable = "sse4.1")]
1262    unsafe fn test_mm_extract_epi8() {
1263        #[rustfmt::skip]
1264        let a = _mm_setr_epi8(
1265            -1, 1, 2, 3, 4, 5, 6, 7,
1266            8, 9, 10, 11, 12, 13, 14, 15
1267        );
1268        let r1 = _mm_extract_epi8::<0>(a);
1269        let r2 = _mm_extract_epi8::<3>(a);
1270        assert_eq!(r1, 0xFF);
1271        assert_eq!(r2, 3);
1272    }
1273
1274    #[simd_test(enable = "sse4.1")]
1275    unsafe fn test_mm_extract_epi32() {
1276        let a = _mm_setr_epi32(0, 1, 2, 3);
1277        let r = _mm_extract_epi32::<1>(a);
1278        assert_eq!(r, 1);
1279        let r = _mm_extract_epi32::<3>(a);
1280        assert_eq!(r, 3);
1281    }
1282
1283    #[simd_test(enable = "sse4.1")]
1284    unsafe fn test_mm_insert_ps() {
1285        let a = _mm_set1_ps(1.0);
1286        let b = _mm_setr_ps(1.0, 2.0, 3.0, 4.0);
1287        let r = _mm_insert_ps::<0b11_00_1100>(a, b);
1288        let e = _mm_setr_ps(4.0, 1.0, 0.0, 0.0);
1289        assert_eq_m128(r, e);
1290
1291        // Zeroing takes precedence over copied value
1292        let a = _mm_set1_ps(1.0);
1293        let b = _mm_setr_ps(1.0, 2.0, 3.0, 4.0);
1294        let r = _mm_insert_ps::<0b11_00_0001>(a, b);
1295        let e = _mm_setr_ps(0.0, 1.0, 1.0, 1.0);
1296        assert_eq_m128(r, e);
1297    }
1298
1299    #[simd_test(enable = "sse4.1")]
1300    unsafe fn test_mm_insert_epi8() {
1301        let a = _mm_set1_epi8(0);
1302        let e = _mm_setr_epi8(0, 32, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
1303        let r = _mm_insert_epi8::<1>(a, 32);
1304        assert_eq_m128i(r, e);
1305        let e = _mm_setr_epi8(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 32, 0);
1306        let r = _mm_insert_epi8::<14>(a, 32);
1307        assert_eq_m128i(r, e);
1308    }
1309
1310    #[simd_test(enable = "sse4.1")]
1311    unsafe fn test_mm_insert_epi32() {
1312        let a = _mm_set1_epi32(0);
1313        let e = _mm_setr_epi32(0, 32, 0, 0);
1314        let r = _mm_insert_epi32::<1>(a, 32);
1315        assert_eq_m128i(r, e);
1316        let e = _mm_setr_epi32(0, 0, 0, 32);
1317        let r = _mm_insert_epi32::<3>(a, 32);
1318        assert_eq_m128i(r, e);
1319    }
1320
1321    #[simd_test(enable = "sse4.1")]
1322    unsafe fn test_mm_max_epi8() {
1323        #[rustfmt::skip]
1324        let a = _mm_setr_epi8(
1325            1, 4, 5, 8, 9, 12, 13, 16,
1326            17, 20, 21, 24, 25, 28, 29, 32,
1327        );
1328        #[rustfmt::skip]
1329        let b = _mm_setr_epi8(
1330            2, 3, 6, 7, 10, 11, 14, 15,
1331            18, 19, 22, 23, 26, 27, 30, 31,
1332        );
1333        let r = _mm_max_epi8(a, b);
1334        #[rustfmt::skip]
1335        let e = _mm_setr_epi8(
1336            2, 4, 6, 8, 10, 12, 14, 16,
1337            18, 20, 22, 24, 26, 28, 30, 32,
1338        );
1339        assert_eq_m128i(r, e);
1340    }
1341
1342    #[simd_test(enable = "sse4.1")]
1343    unsafe fn test_mm_max_epu16() {
1344        let a = _mm_setr_epi16(1, 4, 5, 8, 9, 12, 13, 16);
1345        let b = _mm_setr_epi16(2, 3, 6, 7, 10, 11, 14, 15);
1346        let r = _mm_max_epu16(a, b);
1347        let e = _mm_setr_epi16(2, 4, 6, 8, 10, 12, 14, 16);
1348        assert_eq_m128i(r, e);
1349    }
1350
1351    #[simd_test(enable = "sse4.1")]
1352    unsafe fn test_mm_max_epi32() {
1353        let a = _mm_setr_epi32(1, 4, 5, 8);
1354        let b = _mm_setr_epi32(2, 3, 6, 7);
1355        let r = _mm_max_epi32(a, b);
1356        let e = _mm_setr_epi32(2, 4, 6, 8);
1357        assert_eq_m128i(r, e);
1358    }
1359
1360    #[simd_test(enable = "sse4.1")]
1361    unsafe fn test_mm_max_epu32() {
1362        let a = _mm_setr_epi32(1, 4, 5, 8);
1363        let b = _mm_setr_epi32(2, 3, 6, 7);
1364        let r = _mm_max_epu32(a, b);
1365        let e = _mm_setr_epi32(2, 4, 6, 8);
1366        assert_eq_m128i(r, e);
1367    }
1368
1369    #[simd_test(enable = "sse4.1")]
1370    unsafe fn test_mm_min_epi8() {
1371        #[rustfmt::skip]
1372        let a = _mm_setr_epi8(
1373            1, 4, 5, 8, 9, 12, 13, 16,
1374            17, 20, 21, 24, 25, 28, 29, 32,
1375        );
1376        #[rustfmt::skip]
1377        let b = _mm_setr_epi8(
1378            2, 3, 6, 7, 10, 11, 14, 15,
1379            18, 19, 22, 23, 26, 27, 30, 31,
1380        );
1381        let r = _mm_min_epi8(a, b);
1382        #[rustfmt::skip]
1383        let e = _mm_setr_epi8(
1384            1, 3, 5, 7, 9, 11, 13, 15,
1385            17, 19, 21, 23, 25, 27, 29, 31,
1386        );
1387        assert_eq_m128i(r, e);
1388
1389        #[rustfmt::skip]
1390        let a = _mm_setr_epi8(
1391            1, -4, -5, 8, -9, -12, 13, -16,
1392            17, 20, 21, 24, 25, 28, 29, 32,
1393        );
1394        #[rustfmt::skip]
1395        let b = _mm_setr_epi8(
1396            2, -3, -6, 7, -10, -11, 14, -15,
1397            18, 19, 22, 23, 26, 27, 30, 31,
1398        );
1399        let r = _mm_min_epi8(a, b);
1400        #[rustfmt::skip]
1401        let e = _mm_setr_epi8(
1402            1, -4, -6, 7, -10, -12, 13, -16,
1403            17, 19, 21, 23, 25, 27, 29, 31,
1404        );
1405        assert_eq_m128i(r, e);
1406    }
1407
1408    #[simd_test(enable = "sse4.1")]
1409    unsafe fn test_mm_min_epu16() {
1410        let a = _mm_setr_epi16(1, 4, 5, 8, 9, 12, 13, 16);
1411        let b = _mm_setr_epi16(2, 3, 6, 7, 10, 11, 14, 15);
1412        let r = _mm_min_epu16(a, b);
1413        let e = _mm_setr_epi16(1, 3, 5, 7, 9, 11, 13, 15);
1414        assert_eq_m128i(r, e);
1415    }
1416
1417    #[simd_test(enable = "sse4.1")]
1418    unsafe fn test_mm_min_epi32() {
1419        let a = _mm_setr_epi32(1, 4, 5, 8);
1420        let b = _mm_setr_epi32(2, 3, 6, 7);
1421        let r = _mm_min_epi32(a, b);
1422        let e = _mm_setr_epi32(1, 3, 5, 7);
1423        assert_eq_m128i(r, e);
1424
1425        let a = _mm_setr_epi32(-1, 4, 5, -7);
1426        let b = _mm_setr_epi32(-2, 3, -6, 8);
1427        let r = _mm_min_epi32(a, b);
1428        let e = _mm_setr_epi32(-2, 3, -6, -7);
1429        assert_eq_m128i(r, e);
1430    }
1431
1432    #[simd_test(enable = "sse4.1")]
1433    unsafe fn test_mm_min_epu32() {
1434        let a = _mm_setr_epi32(1, 4, 5, 8);
1435        let b = _mm_setr_epi32(2, 3, 6, 7);
1436        let r = _mm_min_epu32(a, b);
1437        let e = _mm_setr_epi32(1, 3, 5, 7);
1438        assert_eq_m128i(r, e);
1439    }
1440
1441    #[simd_test(enable = "sse4.1")]
1442    unsafe fn test_mm_packus_epi32() {
1443        let a = _mm_setr_epi32(1, 2, 3, 4);
1444        let b = _mm_setr_epi32(-1, -2, -3, -4);
1445        let r = _mm_packus_epi32(a, b);
1446        let e = _mm_setr_epi16(1, 2, 3, 4, 0, 0, 0, 0);
1447        assert_eq_m128i(r, e);
1448    }
1449
1450    #[simd_test(enable = "sse4.1")]
1451    unsafe fn test_mm_cmpeq_epi64() {
1452        let a = _mm_setr_epi64x(0, 1);
1453        let b = _mm_setr_epi64x(0, 0);
1454        let r = _mm_cmpeq_epi64(a, b);
1455        let e = _mm_setr_epi64x(-1, 0);
1456        assert_eq_m128i(r, e);
1457    }
1458
1459    #[simd_test(enable = "sse4.1")]
1460    unsafe fn test_mm_cvtepi8_epi16() {
1461        let a = _mm_set1_epi8(10);
1462        let r = _mm_cvtepi8_epi16(a);
1463        let e = _mm_set1_epi16(10);
1464        assert_eq_m128i(r, e);
1465        let a = _mm_set1_epi8(-10);
1466        let r = _mm_cvtepi8_epi16(a);
1467        let e = _mm_set1_epi16(-10);
1468        assert_eq_m128i(r, e);
1469    }
1470
1471    #[simd_test(enable = "sse4.1")]
1472    unsafe fn test_mm_cvtepi8_epi32() {
1473        let a = _mm_set1_epi8(10);
1474        let r = _mm_cvtepi8_epi32(a);
1475        let e = _mm_set1_epi32(10);
1476        assert_eq_m128i(r, e);
1477        let a = _mm_set1_epi8(-10);
1478        let r = _mm_cvtepi8_epi32(a);
1479        let e = _mm_set1_epi32(-10);
1480        assert_eq_m128i(r, e);
1481    }
1482
1483    #[simd_test(enable = "sse4.1")]
1484    unsafe fn test_mm_cvtepi8_epi64() {
1485        let a = _mm_set1_epi8(10);
1486        let r = _mm_cvtepi8_epi64(a);
1487        let e = _mm_set1_epi64x(10);
1488        assert_eq_m128i(r, e);
1489        let a = _mm_set1_epi8(-10);
1490        let r = _mm_cvtepi8_epi64(a);
1491        let e = _mm_set1_epi64x(-10);
1492        assert_eq_m128i(r, e);
1493    }
1494
1495    #[simd_test(enable = "sse4.1")]
1496    unsafe fn test_mm_cvtepi16_epi32() {
1497        let a = _mm_set1_epi16(10);
1498        let r = _mm_cvtepi16_epi32(a);
1499        let e = _mm_set1_epi32(10);
1500        assert_eq_m128i(r, e);
1501        let a = _mm_set1_epi16(-10);
1502        let r = _mm_cvtepi16_epi32(a);
1503        let e = _mm_set1_epi32(-10);
1504        assert_eq_m128i(r, e);
1505    }
1506
1507    #[simd_test(enable = "sse4.1")]
1508    unsafe fn test_mm_cvtepi16_epi64() {
1509        let a = _mm_set1_epi16(10);
1510        let r = _mm_cvtepi16_epi64(a);
1511        let e = _mm_set1_epi64x(10);
1512        assert_eq_m128i(r, e);
1513        let a = _mm_set1_epi16(-10);
1514        let r = _mm_cvtepi16_epi64(a);
1515        let e = _mm_set1_epi64x(-10);
1516        assert_eq_m128i(r, e);
1517    }
1518
1519    #[simd_test(enable = "sse4.1")]
1520    unsafe fn test_mm_cvtepi32_epi64() {
1521        let a = _mm_set1_epi32(10);
1522        let r = _mm_cvtepi32_epi64(a);
1523        let e = _mm_set1_epi64x(10);
1524        assert_eq_m128i(r, e);
1525        let a = _mm_set1_epi32(-10);
1526        let r = _mm_cvtepi32_epi64(a);
1527        let e = _mm_set1_epi64x(-10);
1528        assert_eq_m128i(r, e);
1529    }
1530
1531    #[simd_test(enable = "sse4.1")]
1532    unsafe fn test_mm_cvtepu8_epi16() {
1533        let a = _mm_set1_epi8(10);
1534        let r = _mm_cvtepu8_epi16(a);
1535        let e = _mm_set1_epi16(10);
1536        assert_eq_m128i(r, e);
1537    }
1538
1539    #[simd_test(enable = "sse4.1")]
1540    unsafe fn test_mm_cvtepu8_epi32() {
1541        let a = _mm_set1_epi8(10);
1542        let r = _mm_cvtepu8_epi32(a);
1543        let e = _mm_set1_epi32(10);
1544        assert_eq_m128i(r, e);
1545    }
1546
1547    #[simd_test(enable = "sse4.1")]
1548    unsafe fn test_mm_cvtepu8_epi64() {
1549        let a = _mm_set1_epi8(10);
1550        let r = _mm_cvtepu8_epi64(a);
1551        let e = _mm_set1_epi64x(10);
1552        assert_eq_m128i(r, e);
1553    }
1554
1555    #[simd_test(enable = "sse4.1")]
1556    unsafe fn test_mm_cvtepu16_epi32() {
1557        let a = _mm_set1_epi16(10);
1558        let r = _mm_cvtepu16_epi32(a);
1559        let e = _mm_set1_epi32(10);
1560        assert_eq_m128i(r, e);
1561    }
1562
1563    #[simd_test(enable = "sse4.1")]
1564    unsafe fn test_mm_cvtepu16_epi64() {
1565        let a = _mm_set1_epi16(10);
1566        let r = _mm_cvtepu16_epi64(a);
1567        let e = _mm_set1_epi64x(10);
1568        assert_eq_m128i(r, e);
1569    }
1570
1571    #[simd_test(enable = "sse4.1")]
1572    unsafe fn test_mm_cvtepu32_epi64() {
1573        let a = _mm_set1_epi32(10);
1574        let r = _mm_cvtepu32_epi64(a);
1575        let e = _mm_set1_epi64x(10);
1576        assert_eq_m128i(r, e);
1577    }
1578
1579    #[simd_test(enable = "sse4.1")]
1580    unsafe fn test_mm_dp_pd() {
1581        let a = _mm_setr_pd(2.0, 3.0);
1582        let b = _mm_setr_pd(1.0, 4.0);
1583        let e = _mm_setr_pd(14.0, 0.0);
1584        assert_eq_m128d(_mm_dp_pd::<0b00110001>(a, b), e);
1585    }
1586
1587    #[simd_test(enable = "sse4.1")]
1588    unsafe fn test_mm_dp_ps() {
1589        let a = _mm_setr_ps(2.0, 3.0, 1.0, 10.0);
1590        let b = _mm_setr_ps(1.0, 4.0, 0.5, 10.0);
1591        let e = _mm_setr_ps(14.5, 0.0, 14.5, 0.0);
1592        assert_eq_m128(_mm_dp_ps::<0b01110101>(a, b), e);
1593    }
1594
1595    #[simd_test(enable = "sse4.1")]
1596    unsafe fn test_mm_floor_pd() {
1597        let a = _mm_setr_pd(2.5, 4.5);
1598        let r = _mm_floor_pd(a);
1599        let e = _mm_setr_pd(2.0, 4.0);
1600        assert_eq_m128d(r, e);
1601    }
1602
1603    #[simd_test(enable = "sse4.1")]
1604    unsafe fn test_mm_floor_ps() {
1605        let a = _mm_setr_ps(2.5, 4.5, 8.5, 16.5);
1606        let r = _mm_floor_ps(a);
1607        let e = _mm_setr_ps(2.0, 4.0, 8.0, 16.0);
1608        assert_eq_m128(r, e);
1609    }
1610
1611    #[simd_test(enable = "sse4.1")]
1612    unsafe fn test_mm_floor_sd() {
1613        let a = _mm_setr_pd(2.5, 4.5);
1614        let b = _mm_setr_pd(-1.5, -3.5);
1615        let r = _mm_floor_sd(a, b);
1616        let e = _mm_setr_pd(-2.0, 4.5);
1617        assert_eq_m128d(r, e);
1618    }
1619
1620    #[simd_test(enable = "sse4.1")]
1621    unsafe fn test_mm_floor_ss() {
1622        let a = _mm_setr_ps(2.5, 4.5, 8.5, 16.5);
1623        let b = _mm_setr_ps(-1.5, -3.5, -7.5, -15.5);
1624        let r = _mm_floor_ss(a, b);
1625        let e = _mm_setr_ps(-2.0, 4.5, 8.5, 16.5);
1626        assert_eq_m128(r, e);
1627    }
1628
1629    #[simd_test(enable = "sse4.1")]
1630    unsafe fn test_mm_ceil_pd() {
1631        let a = _mm_setr_pd(1.5, 3.5);
1632        let r = _mm_ceil_pd(a);
1633        let e = _mm_setr_pd(2.0, 4.0);
1634        assert_eq_m128d(r, e);
1635    }
1636
1637    #[simd_test(enable = "sse4.1")]
1638    unsafe fn test_mm_ceil_ps() {
1639        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
1640        let r = _mm_ceil_ps(a);
1641        let e = _mm_setr_ps(2.0, 4.0, 8.0, 16.0);
1642        assert_eq_m128(r, e);
1643    }
1644
1645    #[simd_test(enable = "sse4.1")]
1646    unsafe fn test_mm_ceil_sd() {
1647        let a = _mm_setr_pd(1.5, 3.5);
1648        let b = _mm_setr_pd(-2.5, -4.5);
1649        let r = _mm_ceil_sd(a, b);
1650        let e = _mm_setr_pd(-2.0, 3.5);
1651        assert_eq_m128d(r, e);
1652    }
1653
1654    #[simd_test(enable = "sse4.1")]
1655    unsafe fn test_mm_ceil_ss() {
1656        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
1657        let b = _mm_setr_ps(-2.5, -4.5, -8.5, -16.5);
1658        let r = _mm_ceil_ss(a, b);
1659        let e = _mm_setr_ps(-2.0, 3.5, 7.5, 15.5);
1660        assert_eq_m128(r, e);
1661    }
1662
1663    #[simd_test(enable = "sse4.1")]
1664    unsafe fn test_mm_round_pd() {
1665        let a = _mm_setr_pd(1.25, 3.75);
1666        let r = _mm_round_pd::<_MM_FROUND_TO_NEAREST_INT>(a);
1667        let e = _mm_setr_pd(1.0, 4.0);
1668        assert_eq_m128d(r, e);
1669    }
1670
1671    #[simd_test(enable = "sse4.1")]
1672    unsafe fn test_mm_round_ps() {
1673        let a = _mm_setr_ps(2.25, 4.75, -1.75, -4.25);
1674        let r = _mm_round_ps::<_MM_FROUND_TO_ZERO>(a);
1675        let e = _mm_setr_ps(2.0, 4.0, -1.0, -4.0);
1676        assert_eq_m128(r, e);
1677    }
1678
1679    #[simd_test(enable = "sse4.1")]
1680    unsafe fn test_mm_round_sd() {
1681        let a = _mm_setr_pd(1.5, 3.5);
1682        let b = _mm_setr_pd(-2.5, -4.5);
1683        let r = _mm_round_sd::<_MM_FROUND_TO_NEAREST_INT>(a, b);
1684        let e = _mm_setr_pd(-2.0, 3.5);
1685        assert_eq_m128d(r, e);
1686
1687        let a = _mm_setr_pd(1.5, 3.5);
1688        let b = _mm_setr_pd(-2.5, -4.5);
1689        let r = _mm_round_sd::<_MM_FROUND_TO_NEG_INF>(a, b);
1690        let e = _mm_setr_pd(-3.0, 3.5);
1691        assert_eq_m128d(r, e);
1692
1693        let a = _mm_setr_pd(1.5, 3.5);
1694        let b = _mm_setr_pd(-2.5, -4.5);
1695        let r = _mm_round_sd::<_MM_FROUND_TO_POS_INF>(a, b);
1696        let e = _mm_setr_pd(-2.0, 3.5);
1697        assert_eq_m128d(r, e);
1698
1699        let a = _mm_setr_pd(1.5, 3.5);
1700        let b = _mm_setr_pd(-2.5, -4.5);
1701        let r = _mm_round_sd::<_MM_FROUND_TO_ZERO>(a, b);
1702        let e = _mm_setr_pd(-2.0, 3.5);
1703        assert_eq_m128d(r, e);
1704    }
1705
1706    #[simd_test(enable = "sse4.1")]
1707    unsafe fn test_mm_round_ss() {
1708        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
1709        let b = _mm_setr_ps(-1.75, -4.5, -8.5, -16.5);
1710        let r = _mm_round_ss::<_MM_FROUND_TO_NEAREST_INT>(a, b);
1711        let e = _mm_setr_ps(-2.0, 3.5, 7.5, 15.5);
1712        assert_eq_m128(r, e);
1713
1714        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
1715        let b = _mm_setr_ps(-1.75, -4.5, -8.5, -16.5);
1716        let r = _mm_round_ss::<_MM_FROUND_TO_NEG_INF>(a, b);
1717        let e = _mm_setr_ps(-2.0, 3.5, 7.5, 15.5);
1718        assert_eq_m128(r, e);
1719
1720        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
1721        let b = _mm_setr_ps(-1.75, -4.5, -8.5, -16.5);
1722        let r = _mm_round_ss::<_MM_FROUND_TO_POS_INF>(a, b);
1723        let e = _mm_setr_ps(-1.0, 3.5, 7.5, 15.5);
1724        assert_eq_m128(r, e);
1725
1726        let a = _mm_setr_ps(1.5, 3.5, 7.5, 15.5);
1727        let b = _mm_setr_ps(-1.75, -4.5, -8.5, -16.5);
1728        let r = _mm_round_ss::<_MM_FROUND_TO_ZERO>(a, b);
1729        let e = _mm_setr_ps(-1.0, 3.5, 7.5, 15.5);
1730        assert_eq_m128(r, e);
1731    }
1732
1733    #[simd_test(enable = "sse4.1")]
1734    unsafe fn test_mm_minpos_epu16_1() {
1735        let a = _mm_setr_epi16(23, 18, 44, 97, 50, 13, 67, 66);
1736        let r = _mm_minpos_epu16(a);
1737        let e = _mm_setr_epi16(13, 5, 0, 0, 0, 0, 0, 0);
1738        assert_eq_m128i(r, e);
1739    }
1740
1741    #[simd_test(enable = "sse4.1")]
1742    unsafe fn test_mm_minpos_epu16_2() {
1743        let a = _mm_setr_epi16(0, 18, 44, 97, 50, 13, 67, 66);
1744        let r = _mm_minpos_epu16(a);
1745        let e = _mm_setr_epi16(0, 0, 0, 0, 0, 0, 0, 0);
1746        assert_eq_m128i(r, e);
1747    }
1748
1749    #[simd_test(enable = "sse4.1")]
1750    unsafe fn test_mm_minpos_epu16_3() {
1751        // Case where the minimum value is repeated
1752        let a = _mm_setr_epi16(23, 18, 44, 97, 50, 13, 67, 13);
1753        let r = _mm_minpos_epu16(a);
1754        let e = _mm_setr_epi16(13, 5, 0, 0, 0, 0, 0, 0);
1755        assert_eq_m128i(r, e);
1756    }
1757
1758    #[simd_test(enable = "sse4.1")]
1759    unsafe fn test_mm_mul_epi32() {
1760        {
1761            let a = _mm_setr_epi32(1, 1, 1, 1);
1762            let b = _mm_setr_epi32(1, 2, 3, 4);
1763            let r = _mm_mul_epi32(a, b);
1764            let e = _mm_setr_epi64x(1, 3);
1765            assert_eq_m128i(r, e);
1766        }
1767        {
1768            let a = _mm_setr_epi32(15, 2 /* ignored */, 1234567, 4 /* ignored */);
1769            let b = _mm_setr_epi32(
1770                -20, -256, /* ignored */
1771                666666, 666666, /* ignored */
1772            );
1773            let r = _mm_mul_epi32(a, b);
1774            let e = _mm_setr_epi64x(-300, 823043843622);
1775            assert_eq_m128i(r, e);
1776        }
1777    }
1778
1779    #[simd_test(enable = "sse4.1")]
1780    unsafe fn test_mm_mullo_epi32() {
1781        {
1782            let a = _mm_setr_epi32(1, 1, 1, 1);
1783            let b = _mm_setr_epi32(1, 2, 3, 4);
1784            let r = _mm_mullo_epi32(a, b);
1785            let e = _mm_setr_epi32(1, 2, 3, 4);
1786            assert_eq_m128i(r, e);
1787        }
1788        {
1789            let a = _mm_setr_epi32(15, -2, 1234567, 99999);
1790            let b = _mm_setr_epi32(-20, -256, 666666, -99999);
1791            let r = _mm_mullo_epi32(a, b);
1792            // Attention, most significant bit in r[2] is treated
1793            // as a sign bit:
1794            // 1234567 * 666666 = -1589877210
1795            let e = _mm_setr_epi32(-300, 512, -1589877210, -1409865409);
1796            assert_eq_m128i(r, e);
1797        }
1798    }
1799
1800    #[simd_test(enable = "sse4.1")]
1801    unsafe fn test_mm_minpos_epu16() {
1802        let a = _mm_setr_epi16(8, 7, 6, 5, 4, 1, 2, 3);
1803        let r = _mm_minpos_epu16(a);
1804        let e = _mm_setr_epi16(1, 5, 0, 0, 0, 0, 0, 0);
1805        assert_eq_m128i(r, e);
1806    }
1807
1808    #[simd_test(enable = "sse4.1")]
1809    unsafe fn test_mm_mpsadbw_epu8() {
1810        #[rustfmt::skip]
1811        let a = _mm_setr_epi8(
1812            0, 1, 2, 3, 4, 5, 6, 7,
1813            8, 9, 10, 11, 12, 13, 14, 15,
1814        );
1815
1816        let r = _mm_mpsadbw_epu8::<0b000>(a, a);
1817        let e = _mm_setr_epi16(0, 4, 8, 12, 16, 20, 24, 28);
1818        assert_eq_m128i(r, e);
1819
1820        let r = _mm_mpsadbw_epu8::<0b001>(a, a);
1821        let e = _mm_setr_epi16(16, 12, 8, 4, 0, 4, 8, 12);
1822        assert_eq_m128i(r, e);
1823
1824        let r = _mm_mpsadbw_epu8::<0b100>(a, a);
1825        let e = _mm_setr_epi16(16, 20, 24, 28, 32, 36, 40, 44);
1826        assert_eq_m128i(r, e);
1827
1828        let r = _mm_mpsadbw_epu8::<0b101>(a, a);
1829        let e = _mm_setr_epi16(0, 4, 8, 12, 16, 20, 24, 28);
1830        assert_eq_m128i(r, e);
1831
1832        let r = _mm_mpsadbw_epu8::<0b111>(a, a);
1833        let e = _mm_setr_epi16(32, 28, 24, 20, 16, 12, 8, 4);
1834        assert_eq_m128i(r, e);
1835    }
1836
1837    #[simd_test(enable = "sse4.1")]
1838    unsafe fn test_mm_testz_si128() {
1839        let a = _mm_set1_epi8(1);
1840        let mask = _mm_set1_epi8(0);
1841        let r = _mm_testz_si128(a, mask);
1842        assert_eq!(r, 1);
1843        let a = _mm_set1_epi8(0b101);
1844        let mask = _mm_set1_epi8(0b110);
1845        let r = _mm_testz_si128(a, mask);
1846        assert_eq!(r, 0);
1847        let a = _mm_set1_epi8(0b011);
1848        let mask = _mm_set1_epi8(0b100);
1849        let r = _mm_testz_si128(a, mask);
1850        assert_eq!(r, 1);
1851    }
1852
1853    #[simd_test(enable = "sse4.1")]
1854    unsafe fn test_mm_testc_si128() {
1855        let a = _mm_set1_epi8(-1);
1856        let mask = _mm_set1_epi8(0);
1857        let r = _mm_testc_si128(a, mask);
1858        assert_eq!(r, 1);
1859        let a = _mm_set1_epi8(0b101);
1860        let mask = _mm_set1_epi8(0b110);
1861        let r = _mm_testc_si128(a, mask);
1862        assert_eq!(r, 0);
1863        let a = _mm_set1_epi8(0b101);
1864        let mask = _mm_set1_epi8(0b100);
1865        let r = _mm_testc_si128(a, mask);
1866        assert_eq!(r, 1);
1867    }
1868
1869    #[simd_test(enable = "sse4.1")]
1870    unsafe fn test_mm_testnzc_si128() {
1871        let a = _mm_set1_epi8(0);
1872        let mask = _mm_set1_epi8(1);
1873        let r = _mm_testnzc_si128(a, mask);
1874        assert_eq!(r, 0);
1875        let a = _mm_set1_epi8(-1);
1876        let mask = _mm_set1_epi8(0);
1877        let r = _mm_testnzc_si128(a, mask);
1878        assert_eq!(r, 0);
1879        let a = _mm_set1_epi8(0b101);
1880        let mask = _mm_set1_epi8(0b110);
1881        let r = _mm_testnzc_si128(a, mask);
1882        assert_eq!(r, 1);
1883        let a = _mm_set1_epi8(0b101);
1884        let mask = _mm_set1_epi8(0b101);
1885        let r = _mm_testnzc_si128(a, mask);
1886        assert_eq!(r, 0);
1887    }
1888
1889    #[simd_test(enable = "sse4.1")]
1890    unsafe fn test_mm_test_all_zeros() {
1891        let a = _mm_set1_epi8(1);
1892        let mask = _mm_set1_epi8(0);
1893        let r = _mm_test_all_zeros(a, mask);
1894        assert_eq!(r, 1);
1895        let a = _mm_set1_epi8(0b101);
1896        let mask = _mm_set1_epi8(0b110);
1897        let r = _mm_test_all_zeros(a, mask);
1898        assert_eq!(r, 0);
1899        let a = _mm_set1_epi8(0b011);
1900        let mask = _mm_set1_epi8(0b100);
1901        let r = _mm_test_all_zeros(a, mask);
1902        assert_eq!(r, 1);
1903    }
1904
1905    #[simd_test(enable = "sse4.1")]
1906    unsafe fn test_mm_test_all_ones() {
1907        let a = _mm_set1_epi8(-1);
1908        let r = _mm_test_all_ones(a);
1909        assert_eq!(r, 1);
1910        let a = _mm_set1_epi8(0b101);
1911        let r = _mm_test_all_ones(a);
1912        assert_eq!(r, 0);
1913    }
1914
1915    #[simd_test(enable = "sse4.1")]
1916    unsafe fn test_mm_test_mix_ones_zeros() {
1917        let a = _mm_set1_epi8(0);
1918        let mask = _mm_set1_epi8(1);
1919        let r = _mm_test_mix_ones_zeros(a, mask);
1920        assert_eq!(r, 0);
1921        let a = _mm_set1_epi8(-1);
1922        let mask = _mm_set1_epi8(0);
1923        let r = _mm_test_mix_ones_zeros(a, mask);
1924        assert_eq!(r, 0);
1925        let a = _mm_set1_epi8(0b101);
1926        let mask = _mm_set1_epi8(0b110);
1927        let r = _mm_test_mix_ones_zeros(a, mask);
1928        assert_eq!(r, 1);
1929        let a = _mm_set1_epi8(0b101);
1930        let mask = _mm_set1_epi8(0b101);
1931        let r = _mm_test_mix_ones_zeros(a, mask);
1932        assert_eq!(r, 0);
1933    }
1934
1935    #[simd_test(enable = "sse4.1")]
1936    unsafe fn test_mm_stream_load_si128() {
1937        let a = _mm_set_epi64x(5, 6);
1938        let r = _mm_stream_load_si128(core::ptr::addr_of!(a) as *const _);
1939        assert_eq_m128i(a, r);
1940    }
1941}