| 1 | //! Trailing Bit Manipulation (TBM) instruction set. |
| 2 | //! |
| 3 | //! The reference is [AMD64 Architecture Programmer's Manual, Volume 3: |
| 4 | //! General-Purpose and System Instructions][amd64_ref]. |
| 5 | //! |
| 6 | //! [Wikipedia][wikipedia_bmi] provides a quick overview of the available |
| 7 | //! instructions. |
| 8 | //! |
| 9 | //! [amd64_ref]: https://docs.amd.com/v/u/en-US/24594_3.37 |
| 10 | //! [wikipedia_bmi]: |
| 11 | //! https://en.wikipedia.org/wiki/Bit_Manipulation_Instruction_Sets#ABM_.28Advanced_Bit_Manipulation.29 |
| 12 | |
| 13 | #[cfg (test)] |
| 14 | use stdarch_test::assert_instr; |
| 15 | |
| 16 | unsafe extern "C" { |
| 17 | #[link_name = "llvm.x86.tbm.bextri.u64" ] |
| 18 | unsafefn bextri_u64(a: u64, control: u64) -> u64; |
| 19 | } |
| 20 | |
| 21 | /// Extracts bits of `a` specified by `control` into |
| 22 | /// the least significant bits of the result. |
| 23 | /// |
| 24 | /// Bits `[7,0]` of `control` specify the index to the first bit in the range to |
| 25 | /// be extracted, and bits `[15,8]` specify the length of the range. For any bit |
| 26 | /// position in the specified range that lie beyond the MSB of the source operand, |
| 27 | /// zeroes will be written. If the range is empty, the result is zero. |
| 28 | #[inline ] |
| 29 | #[target_feature (enable = "tbm" )] |
| 30 | #[cfg_attr (test, assert_instr(bextr, CONTROL = 0x0404))] |
| 31 | #[rustc_legacy_const_generics (1)] |
| 32 | #[stable (feature = "simd_x86_updates" , since = "1.82.0" )] |
| 33 | pub fn _bextri_u64<const CONTROL: u64>(a: u64) -> u64 { |
| 34 | static_assert_uimm_bits!(CONTROL, 16); |
| 35 | unsafe { bextri_u64(a, CONTROL) } |
| 36 | } |
| 37 | |
| 38 | /// Clears all bits below the least significant zero bit of `x`. |
| 39 | /// |
| 40 | /// If there is no zero bit in `x`, it returns zero. |
| 41 | #[inline ] |
| 42 | #[target_feature (enable = "tbm" )] |
| 43 | #[cfg_attr (test, assert_instr(blcfill))] |
| 44 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 45 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 46 | pub const fn _blcfill_u64(x: u64) -> u64 { |
| 47 | x & x.wrapping_add(1) |
| 48 | } |
| 49 | |
| 50 | /// Sets all bits of `x` to 1 except for the least significant zero bit. |
| 51 | /// |
| 52 | /// If there is no zero bit in `x`, it sets all bits. |
| 53 | #[inline ] |
| 54 | #[target_feature (enable = "tbm" )] |
| 55 | #[cfg_attr (test, assert_instr(blci))] |
| 56 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 57 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 58 | pub const fn _blci_u64(x: u64) -> u64 { |
| 59 | x | !x.wrapping_add(1) |
| 60 | } |
| 61 | |
| 62 | /// Sets the least significant zero bit of `x` and clears all other bits. |
| 63 | /// |
| 64 | /// If there is no zero bit in `x`, it returns zero. |
| 65 | #[inline ] |
| 66 | #[target_feature (enable = "tbm" )] |
| 67 | #[cfg_attr (test, assert_instr(blcic))] |
| 68 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 69 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 70 | pub const fn _blcic_u64(x: u64) -> u64 { |
| 71 | !x & x.wrapping_add(1) |
| 72 | } |
| 73 | |
| 74 | /// Sets the least significant zero bit of `x` and clears all bits above |
| 75 | /// that bit. |
| 76 | /// |
| 77 | /// If there is no zero bit in `x`, it sets all the bits. |
| 78 | #[inline ] |
| 79 | #[target_feature (enable = "tbm" )] |
| 80 | #[cfg_attr (test, assert_instr(blcmsk))] |
| 81 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 82 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 83 | pub const fn _blcmsk_u64(x: u64) -> u64 { |
| 84 | x ^ x.wrapping_add(1) |
| 85 | } |
| 86 | |
| 87 | /// Sets the least significant zero bit of `x`. |
| 88 | /// |
| 89 | /// If there is no zero bit in `x`, it returns `x`. |
| 90 | #[inline ] |
| 91 | #[target_feature (enable = "tbm" )] |
| 92 | #[cfg_attr (test, assert_instr(blcs))] |
| 93 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 94 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 95 | pub const fn _blcs_u64(x: u64) -> u64 { |
| 96 | x | x.wrapping_add(1) |
| 97 | } |
| 98 | |
| 99 | /// Sets all bits of `x` below the least significant one. |
| 100 | /// |
| 101 | /// If there is no set bit in `x`, it sets all the bits. |
| 102 | #[inline ] |
| 103 | #[target_feature (enable = "tbm" )] |
| 104 | #[cfg_attr (test, assert_instr(blsfill))] |
| 105 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 106 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 107 | pub const fn _blsfill_u64(x: u64) -> u64 { |
| 108 | x | x.wrapping_sub(1) |
| 109 | } |
| 110 | |
| 111 | /// Clears least significant bit and sets all other bits. |
| 112 | /// |
| 113 | /// If there is no set bit in `x`, it sets all the bits. |
| 114 | #[inline ] |
| 115 | #[target_feature (enable = "tbm" )] |
| 116 | #[cfg_attr (test, assert_instr(blsic))] |
| 117 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 118 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 119 | pub const fn _blsic_u64(x: u64) -> u64 { |
| 120 | !x | x.wrapping_sub(1) |
| 121 | } |
| 122 | |
| 123 | /// Clears all bits below the least significant zero of `x` and sets all other |
| 124 | /// bits. |
| 125 | /// |
| 126 | /// If the least significant bit of `x` is `0`, it sets all bits. |
| 127 | #[inline ] |
| 128 | #[target_feature (enable = "tbm" )] |
| 129 | #[cfg_attr (test, assert_instr(t1mskc))] |
| 130 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 131 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 132 | pub const fn _t1mskc_u64(x: u64) -> u64 { |
| 133 | !x | x.wrapping_add(1) |
| 134 | } |
| 135 | |
| 136 | /// Sets all bits below the least significant one of `x` and clears all other |
| 137 | /// bits. |
| 138 | /// |
| 139 | /// If the least significant bit of `x` is 1, it returns zero. |
| 140 | #[inline ] |
| 141 | #[target_feature (enable = "tbm" )] |
| 142 | #[cfg_attr (test, assert_instr(tzmsk))] |
| 143 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 144 | #[rustc_const_unstable (feature = "stdarch_const_x86" , issue = "149298" )] |
| 145 | pub const fn _tzmsk_u64(x: u64) -> u64 { |
| 146 | !x & x.wrapping_sub(1) |
| 147 | } |
| 148 | |
| 149 | #[cfg (test)] |
| 150 | mod tests { |
| 151 | use crate::core_arch::assert_eq_const as assert_eq; |
| 152 | use stdarch_test::simd_test; |
| 153 | |
| 154 | use crate::core_arch::x86_64::*; |
| 155 | |
| 156 | #[simd_test(enable = "tbm" )] |
| 157 | fn test_bextri_u64() { |
| 158 | assert_eq!(_bextri_u64::<0x0404>(0b0101_0000u64), 0b0000_0101u64); |
| 159 | } |
| 160 | |
| 161 | #[simd_test(enable = "tbm" )] |
| 162 | const fn test_blcfill_u64() { |
| 163 | assert_eq!(_blcfill_u64(0b0101_0111u64), 0b0101_0000u64); |
| 164 | assert_eq!(_blcfill_u64(0b1111_1111u64), 0u64); |
| 165 | } |
| 166 | |
| 167 | #[simd_test(enable = "tbm" )] |
| 168 | const fn test_blci_u64() { |
| 169 | assert_eq!( |
| 170 | _blci_u64(0b0101_0000u64), |
| 171 | 0b1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1110u64 |
| 172 | ); |
| 173 | assert_eq!( |
| 174 | _blci_u64(0b1111_1111u64), |
| 175 | 0b1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1110_1111_1111u64 |
| 176 | ); |
| 177 | } |
| 178 | |
| 179 | #[simd_test(enable = "tbm" )] |
| 180 | const fn test_blcic_u64() { |
| 181 | assert_eq!(_blcic_u64(0b0101_0001u64), 0b0000_0010u64); |
| 182 | assert_eq!(_blcic_u64(0b1111_1111u64), 0b1_0000_0000u64); |
| 183 | } |
| 184 | |
| 185 | #[simd_test(enable = "tbm" )] |
| 186 | const fn test_blcmsk_u64() { |
| 187 | assert_eq!(_blcmsk_u64(0b0101_0001u64), 0b0000_0011u64); |
| 188 | assert_eq!(_blcmsk_u64(0b1111_1111u64), 0b1_1111_1111u64); |
| 189 | } |
| 190 | |
| 191 | #[simd_test(enable = "tbm" )] |
| 192 | const fn test_blcs_u64() { |
| 193 | assert_eq!(_blcs_u64(0b0101_0001u64), 0b0101_0011u64); |
| 194 | assert_eq!(_blcs_u64(0b1111_1111u64), 0b1_1111_1111u64); |
| 195 | } |
| 196 | |
| 197 | #[simd_test(enable = "tbm" )] |
| 198 | const fn test_blsfill_u64() { |
| 199 | assert_eq!(_blsfill_u64(0b0101_0100u64), 0b0101_0111u64); |
| 200 | assert_eq!( |
| 201 | _blsfill_u64(0u64), |
| 202 | 0b1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111u64 |
| 203 | ); |
| 204 | } |
| 205 | |
| 206 | #[simd_test(enable = "tbm" )] |
| 207 | const fn test_blsic_u64() { |
| 208 | assert_eq!( |
| 209 | _blsic_u64(0b0101_0100u64), |
| 210 | 0b1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1011u64 |
| 211 | ); |
| 212 | assert_eq!( |
| 213 | _blsic_u64(0u64), |
| 214 | 0b1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111u64 |
| 215 | ); |
| 216 | } |
| 217 | |
| 218 | #[simd_test(enable = "tbm" )] |
| 219 | const fn test_t1mskc_u64() { |
| 220 | assert_eq!( |
| 221 | _t1mskc_u64(0b0101_0111u64), |
| 222 | 0b1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1000u64 |
| 223 | ); |
| 224 | assert_eq!( |
| 225 | _t1mskc_u64(0u64), |
| 226 | 0b1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111_1111u64 |
| 227 | ); |
| 228 | } |
| 229 | |
| 230 | #[simd_test(enable = "tbm" )] |
| 231 | const fn test_tzmsk_u64() { |
| 232 | assert_eq!(_tzmsk_u64(0b0101_1000u64), 0b0000_0111u64); |
| 233 | assert_eq!(_tzmsk_u64(0b0101_1001u64), 0b0000_0000u64); |
| 234 | } |
| 235 | } |
| 236 | |