| 1 | //! Bit Manipulation Instruction (BMI) Set 1.0. |
| 2 | //! |
| 3 | //! The reference is [Intel 64 and IA-32 Architectures Software Developer's |
| 4 | //! Manual Volume 2: Instruction Set Reference, A-Z][intel64_ref]. |
| 5 | //! |
| 6 | //! [Wikipedia][wikipedia_bmi] provides a quick overview of the instructions |
| 7 | //! available. |
| 8 | //! |
| 9 | //! [intel64_ref]: http://www.intel.de/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-instruction-set-reference-manual-325383.pdf |
| 10 | //! [wikipedia_bmi]: https://en.wikipedia.org/wiki/Bit_Manipulation_Instruction_Sets#ABM_.28Advanced_Bit_Manipulation.29 |
| 11 | |
| 12 | #[cfg (test)] |
| 13 | use stdarch_test::assert_instr; |
| 14 | |
| 15 | /// Extracts bits in range [`start`, `start` + `length`) from `a` into |
| 16 | /// the least significant bits of the result. |
| 17 | /// |
| 18 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_bextr_u32) |
| 19 | #[inline ] |
| 20 | #[target_feature (enable = "bmi1" )] |
| 21 | #[cfg_attr (test, assert_instr(bextr))] |
| 22 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 23 | pub fn _bextr_u32(a: u32, start: u32, len: u32) -> u32 { |
| 24 | _bextr2_u32(a, (start & 0xff_u32) | ((len & 0xff_u32) << 8_u32)) |
| 25 | } |
| 26 | |
| 27 | /// Extracts bits of `a` specified by `control` into |
| 28 | /// the least significant bits of the result. |
| 29 | /// |
| 30 | /// Bits `[7,0]` of `control` specify the index to the first bit in the range |
| 31 | /// to be extracted, and bits `[15,8]` specify the length of the range. |
| 32 | /// |
| 33 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_bextr2_u32) |
| 34 | #[inline ] |
| 35 | #[target_feature (enable = "bmi1" )] |
| 36 | #[cfg_attr (test, assert_instr(bextr))] |
| 37 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 38 | pub fn _bextr2_u32(a: u32, control: u32) -> u32 { |
| 39 | unsafe { x86_bmi_bextr_32(x:a, y:control) } |
| 40 | } |
| 41 | |
| 42 | /// Bitwise logical `AND` of inverted `a` with `b`. |
| 43 | /// |
| 44 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_andn_u32) |
| 45 | #[inline ] |
| 46 | #[target_feature (enable = "bmi1" )] |
| 47 | #[cfg_attr (test, assert_instr(andn))] |
| 48 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 49 | pub fn _andn_u32(a: u32, b: u32) -> u32 { |
| 50 | !a & b |
| 51 | } |
| 52 | |
| 53 | /// Extracts lowest set isolated bit. |
| 54 | /// |
| 55 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_blsi_u32) |
| 56 | #[inline ] |
| 57 | #[target_feature (enable = "bmi1" )] |
| 58 | #[cfg_attr (test, assert_instr(blsi))] |
| 59 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 60 | pub fn _blsi_u32(x: u32) -> u32 { |
| 61 | x & x.wrapping_neg() |
| 62 | } |
| 63 | |
| 64 | /// Gets mask up to lowest set bit. |
| 65 | /// |
| 66 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_blsmsk_u32) |
| 67 | #[inline ] |
| 68 | #[target_feature (enable = "bmi1" )] |
| 69 | #[cfg_attr (test, assert_instr(blsmsk))] |
| 70 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 71 | pub fn _blsmsk_u32(x: u32) -> u32 { |
| 72 | x ^ (x.wrapping_sub(1_u32)) |
| 73 | } |
| 74 | |
| 75 | /// Resets the lowest set bit of `x`. |
| 76 | /// |
| 77 | /// If `x` is sets CF. |
| 78 | /// |
| 79 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_blsr_u32) |
| 80 | #[inline ] |
| 81 | #[target_feature (enable = "bmi1" )] |
| 82 | #[cfg_attr (test, assert_instr(blsr))] |
| 83 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 84 | pub fn _blsr_u32(x: u32) -> u32 { |
| 85 | x & (x.wrapping_sub(1)) |
| 86 | } |
| 87 | |
| 88 | /// Counts the number of trailing least significant zero bits. |
| 89 | /// |
| 90 | /// When the source operand is `0`, it returns its size in bits. |
| 91 | /// |
| 92 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_tzcnt_u16) |
| 93 | #[inline ] |
| 94 | #[target_feature (enable = "bmi1" )] |
| 95 | #[cfg_attr (test, assert_instr(tzcnt))] |
| 96 | #[stable (feature = "simd_x86_updates" , since = "1.82.0" )] |
| 97 | pub fn _tzcnt_u16(x: u16) -> u16 { |
| 98 | x.trailing_zeros() as u16 |
| 99 | } |
| 100 | |
| 101 | /// Counts the number of trailing least significant zero bits. |
| 102 | /// |
| 103 | /// When the source operand is `0`, it returns its size in bits. |
| 104 | /// |
| 105 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_tzcnt_u32) |
| 106 | #[inline ] |
| 107 | #[target_feature (enable = "bmi1" )] |
| 108 | #[cfg_attr (test, assert_instr(tzcnt))] |
| 109 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 110 | pub fn _tzcnt_u32(x: u32) -> u32 { |
| 111 | x.trailing_zeros() |
| 112 | } |
| 113 | |
| 114 | /// Counts the number of trailing least significant zero bits. |
| 115 | /// |
| 116 | /// When the source operand is `0`, it returns its size in bits. |
| 117 | /// |
| 118 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_tzcnt_32) |
| 119 | #[inline ] |
| 120 | #[target_feature (enable = "bmi1" )] |
| 121 | #[cfg_attr (test, assert_instr(tzcnt))] |
| 122 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 123 | pub fn _mm_tzcnt_32(x: u32) -> i32 { |
| 124 | x.trailing_zeros() as i32 |
| 125 | } |
| 126 | |
| 127 | unsafe extern "C" { |
| 128 | #[link_name = "llvm.x86.bmi.bextr.32" ] |
| 129 | unsafefn x86_bmi_bextr_32(x: u32, y: u32) -> u32; |
| 130 | } |
| 131 | |
| 132 | #[cfg (test)] |
| 133 | mod tests { |
| 134 | use stdarch_test::simd_test; |
| 135 | |
| 136 | use crate::core_arch::x86::*; |
| 137 | |
| 138 | #[simd_test(enable = "bmi1" )] |
| 139 | unsafe fn test_bextr_u32() { |
| 140 | let r = _bextr_u32(0b0101_0000u32, 4, 4); |
| 141 | assert_eq!(r, 0b0000_0101u32); |
| 142 | } |
| 143 | |
| 144 | #[simd_test(enable = "bmi1" )] |
| 145 | unsafe fn test_andn_u32() { |
| 146 | assert_eq!(_andn_u32(0, 0), 0); |
| 147 | assert_eq!(_andn_u32(0, 1), 1); |
| 148 | assert_eq!(_andn_u32(1, 0), 0); |
| 149 | assert_eq!(_andn_u32(1, 1), 0); |
| 150 | |
| 151 | let r = _andn_u32(0b0000_0000u32, 0b0000_0000u32); |
| 152 | assert_eq!(r, 0b0000_0000u32); |
| 153 | |
| 154 | let r = _andn_u32(0b0000_0000u32, 0b1111_1111u32); |
| 155 | assert_eq!(r, 0b1111_1111u32); |
| 156 | |
| 157 | let r = _andn_u32(0b1111_1111u32, 0b0000_0000u32); |
| 158 | assert_eq!(r, 0b0000_0000u32); |
| 159 | |
| 160 | let r = _andn_u32(0b1111_1111u32, 0b1111_1111u32); |
| 161 | assert_eq!(r, 0b0000_0000u32); |
| 162 | |
| 163 | let r = _andn_u32(0b0100_0000u32, 0b0101_1101u32); |
| 164 | assert_eq!(r, 0b0001_1101u32); |
| 165 | } |
| 166 | |
| 167 | #[simd_test(enable = "bmi1" )] |
| 168 | unsafe fn test_blsi_u32() { |
| 169 | assert_eq!(_blsi_u32(0b1101_0000u32), 0b0001_0000u32); |
| 170 | } |
| 171 | |
| 172 | #[simd_test(enable = "bmi1" )] |
| 173 | unsafe fn test_blsmsk_u32() { |
| 174 | let r = _blsmsk_u32(0b0011_0000u32); |
| 175 | assert_eq!(r, 0b0001_1111u32); |
| 176 | } |
| 177 | |
| 178 | #[simd_test(enable = "bmi1" )] |
| 179 | unsafe fn test_blsr_u32() { |
| 180 | // TODO: test the behavior when the input is `0`. |
| 181 | let r = _blsr_u32(0b0011_0000u32); |
| 182 | assert_eq!(r, 0b0010_0000u32); |
| 183 | } |
| 184 | |
| 185 | #[simd_test(enable = "bmi1" )] |
| 186 | unsafe fn test_tzcnt_u16() { |
| 187 | assert_eq!(_tzcnt_u16(0b0000_0001u16), 0u16); |
| 188 | assert_eq!(_tzcnt_u16(0b0000_0000u16), 16u16); |
| 189 | assert_eq!(_tzcnt_u16(0b1001_0000u16), 4u16); |
| 190 | } |
| 191 | |
| 192 | #[simd_test(enable = "bmi1" )] |
| 193 | unsafe fn test_tzcnt_u32() { |
| 194 | assert_eq!(_tzcnt_u32(0b0000_0001u32), 0u32); |
| 195 | assert_eq!(_tzcnt_u32(0b0000_0000u32), 32u32); |
| 196 | assert_eq!(_tzcnt_u32(0b1001_0000u32), 4u32); |
| 197 | } |
| 198 | } |
| 199 | |