| 1 | //! `x86_64`'s Streaming SIMD Extensions 2 (SSE2) |
| 2 | |
| 3 | use crate::core_arch::x86::*; |
| 4 | |
| 5 | #[cfg (test)] |
| 6 | use stdarch_test::assert_instr; |
| 7 | |
| 8 | #[allow (improper_ctypes)] |
| 9 | unsafe extern "C" { |
| 10 | #[link_name = "llvm.x86.sse2.cvtsd2si64" ] |
| 11 | unsafefn cvtsd2si64(a: __m128d) -> i64; |
| 12 | #[link_name = "llvm.x86.sse2.cvttsd2si64" ] |
| 13 | unsafefn cvttsd2si64(a: __m128d) -> i64; |
| 14 | } |
| 15 | |
| 16 | /// Converts the lower double-precision (64-bit) floating-point element in a to |
| 17 | /// a 64-bit integer. |
| 18 | /// |
| 19 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsd_si64) |
| 20 | #[inline ] |
| 21 | #[target_feature (enable = "sse2" )] |
| 22 | #[cfg_attr (test, assert_instr(cvtsd2si))] |
| 23 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 24 | pub fn _mm_cvtsd_si64(a: __m128d) -> i64 { |
| 25 | unsafe { cvtsd2si64(a) } |
| 26 | } |
| 27 | |
| 28 | /// Alias for `_mm_cvtsd_si64` |
| 29 | /// |
| 30 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsd_si64x) |
| 31 | #[inline ] |
| 32 | #[target_feature (enable = "sse2" )] |
| 33 | #[cfg_attr (test, assert_instr(cvtsd2si))] |
| 34 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 35 | pub fn _mm_cvtsd_si64x(a: __m128d) -> i64 { |
| 36 | _mm_cvtsd_si64(a) |
| 37 | } |
| 38 | |
| 39 | /// Converts the lower double-precision (64-bit) floating-point element in `a` |
| 40 | /// to a 64-bit integer with truncation. |
| 41 | /// |
| 42 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvttsd_si64) |
| 43 | #[inline ] |
| 44 | #[target_feature (enable = "sse2" )] |
| 45 | #[cfg_attr (test, assert_instr(cvttsd2si))] |
| 46 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 47 | pub fn _mm_cvttsd_si64(a: __m128d) -> i64 { |
| 48 | unsafe { cvttsd2si64(a) } |
| 49 | } |
| 50 | |
| 51 | /// Alias for `_mm_cvttsd_si64` |
| 52 | /// |
| 53 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvttsd_si64x) |
| 54 | #[inline ] |
| 55 | #[target_feature (enable = "sse2" )] |
| 56 | #[cfg_attr (test, assert_instr(cvttsd2si))] |
| 57 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 58 | pub fn _mm_cvttsd_si64x(a: __m128d) -> i64 { |
| 59 | _mm_cvttsd_si64(a) |
| 60 | } |
| 61 | |
| 62 | /// Stores a 64-bit integer value in the specified memory location. |
| 63 | /// To minimize caching, the data is flagged as non-temporal (unlikely to be |
| 64 | /// used again soon). |
| 65 | /// |
| 66 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_stream_si64) |
| 67 | /// |
| 68 | /// # Safety of non-temporal stores |
| 69 | /// |
| 70 | /// After using this intrinsic, but before any other access to the memory that this intrinsic |
| 71 | /// mutates, a call to [`_mm_sfence`] must be performed by the thread that used the intrinsic. In |
| 72 | /// particular, functions that call this intrinsic should generally call `_mm_sfence` before they |
| 73 | /// return. |
| 74 | /// |
| 75 | /// See [`_mm_sfence`] for details. |
| 76 | #[inline ] |
| 77 | #[target_feature (enable = "sse2" )] |
| 78 | #[cfg_attr (test, assert_instr(movnti))] |
| 79 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 80 | pub unsafe fn _mm_stream_si64(mem_addr: *mut i64, a: i64) { |
| 81 | crate::arch::asm!( |
| 82 | vps!("movnti" , ",{a}" ), |
| 83 | p = in(reg) mem_addr, |
| 84 | a = in(reg) a, |
| 85 | options(nostack, preserves_flags), |
| 86 | ); |
| 87 | } |
| 88 | |
| 89 | /// Returns a vector whose lowest element is `a` and all higher elements are |
| 90 | /// `0`. |
| 91 | /// |
| 92 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsi64_si128) |
| 93 | #[inline ] |
| 94 | #[target_feature (enable = "sse2" )] |
| 95 | #[cfg_attr (test, assert_instr(movq))] |
| 96 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 97 | pub fn _mm_cvtsi64_si128(a: i64) -> __m128i { |
| 98 | _mm_set_epi64x(e1:0, e0:a) |
| 99 | } |
| 100 | |
| 101 | /// Returns a vector whose lowest element is `a` and all higher elements are |
| 102 | /// `0`. |
| 103 | /// |
| 104 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsi64x_si128) |
| 105 | #[inline ] |
| 106 | #[target_feature (enable = "sse2" )] |
| 107 | #[cfg_attr (test, assert_instr(movq))] |
| 108 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 109 | pub fn _mm_cvtsi64x_si128(a: i64) -> __m128i { |
| 110 | _mm_cvtsi64_si128(a) |
| 111 | } |
| 112 | |
| 113 | /// Returns the lowest element of `a`. |
| 114 | /// |
| 115 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsi128_si64) |
| 116 | #[inline ] |
| 117 | #[target_feature (enable = "sse2" )] |
| 118 | #[cfg_attr (test, assert_instr(movq))] |
| 119 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 120 | pub fn _mm_cvtsi128_si64(a: __m128i) -> i64 { |
| 121 | unsafe { simd_extract!(a.as_i64x2(), 0) } |
| 122 | } |
| 123 | |
| 124 | /// Returns the lowest element of `a`. |
| 125 | /// |
| 126 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsi128_si64x) |
| 127 | #[inline ] |
| 128 | #[target_feature (enable = "sse2" )] |
| 129 | #[cfg_attr (test, assert_instr(movq))] |
| 130 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 131 | pub fn _mm_cvtsi128_si64x(a: __m128i) -> i64 { |
| 132 | _mm_cvtsi128_si64(a) |
| 133 | } |
| 134 | |
| 135 | /// Returns `a` with its lower element replaced by `b` after converting it to |
| 136 | /// an `f64`. |
| 137 | /// |
| 138 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsi64_sd) |
| 139 | #[inline ] |
| 140 | #[target_feature (enable = "sse2" )] |
| 141 | #[cfg_attr (test, assert_instr(cvtsi2sd))] |
| 142 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 143 | pub fn _mm_cvtsi64_sd(a: __m128d, b: i64) -> __m128d { |
| 144 | unsafe { simd_insert!(a, 0, b as f64) } |
| 145 | } |
| 146 | |
| 147 | /// Returns `a` with its lower element replaced by `b` after converting it to |
| 148 | /// an `f64`. |
| 149 | /// |
| 150 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_cvtsi64x_sd) |
| 151 | #[inline ] |
| 152 | #[target_feature (enable = "sse2" )] |
| 153 | #[cfg_attr (test, assert_instr(cvtsi2sd))] |
| 154 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 155 | pub fn _mm_cvtsi64x_sd(a: __m128d, b: i64) -> __m128d { |
| 156 | _mm_cvtsi64_sd(a, b) |
| 157 | } |
| 158 | |
| 159 | #[cfg (test)] |
| 160 | mod tests { |
| 161 | use crate::core_arch::arch::x86_64::*; |
| 162 | use std::boxed; |
| 163 | use std::ptr; |
| 164 | use stdarch_test::simd_test; |
| 165 | |
| 166 | #[simd_test(enable = "sse2" )] |
| 167 | unsafe fn test_mm_cvtsd_si64() { |
| 168 | let r = _mm_cvtsd_si64(_mm_setr_pd(-2.0, 5.0)); |
| 169 | assert_eq!(r, -2_i64); |
| 170 | |
| 171 | let r = _mm_cvtsd_si64(_mm_setr_pd(f64::MAX, f64::MIN)); |
| 172 | assert_eq!(r, i64::MIN); |
| 173 | } |
| 174 | |
| 175 | #[simd_test(enable = "sse2" )] |
| 176 | unsafe fn test_mm_cvtsd_si64x() { |
| 177 | let r = _mm_cvtsd_si64x(_mm_setr_pd(f64::NAN, f64::NAN)); |
| 178 | assert_eq!(r, i64::MIN); |
| 179 | } |
| 180 | |
| 181 | #[simd_test(enable = "sse2" )] |
| 182 | unsafe fn test_mm_cvttsd_si64() { |
| 183 | let a = _mm_setr_pd(-1.1, 2.2); |
| 184 | let r = _mm_cvttsd_si64(a); |
| 185 | assert_eq!(r, -1_i64); |
| 186 | } |
| 187 | |
| 188 | #[simd_test(enable = "sse2" )] |
| 189 | unsafe fn test_mm_cvttsd_si64x() { |
| 190 | let a = _mm_setr_pd(f64::NEG_INFINITY, f64::NAN); |
| 191 | let r = _mm_cvttsd_si64x(a); |
| 192 | assert_eq!(r, i64::MIN); |
| 193 | } |
| 194 | |
| 195 | #[simd_test(enable = "sse2" )] |
| 196 | // Miri cannot support this until it is clear how it fits in the Rust memory model |
| 197 | // (non-temporal store) |
| 198 | #[cfg_attr (miri, ignore)] |
| 199 | unsafe fn test_mm_stream_si64() { |
| 200 | let a: i64 = 7; |
| 201 | let mut mem = boxed::Box::<i64>::new(-1); |
| 202 | _mm_stream_si64(ptr::addr_of_mut!(*mem), a); |
| 203 | assert_eq!(a, *mem); |
| 204 | } |
| 205 | |
| 206 | #[simd_test(enable = "sse2" )] |
| 207 | unsafe fn test_mm_cvtsi64_si128() { |
| 208 | let r = _mm_cvtsi64_si128(5); |
| 209 | assert_eq_m128i(r, _mm_setr_epi64x(5, 0)); |
| 210 | } |
| 211 | |
| 212 | #[simd_test(enable = "sse2" )] |
| 213 | unsafe fn test_mm_cvtsi128_si64() { |
| 214 | let r = _mm_cvtsi128_si64(_mm_setr_epi64x(5, 0)); |
| 215 | assert_eq!(r, 5); |
| 216 | } |
| 217 | |
| 218 | #[simd_test(enable = "sse2" )] |
| 219 | unsafe fn test_mm_cvtsi64_sd() { |
| 220 | let a = _mm_set1_pd(3.5); |
| 221 | let r = _mm_cvtsi64_sd(a, 5); |
| 222 | assert_eq_m128d(r, _mm_setr_pd(5.0, 3.5)); |
| 223 | } |
| 224 | } |
| 225 | |