| 1 | //! Supplemental Streaming SIMD Extensions 3 (SSSE3) |
| 2 | |
| 3 | use crate::{ |
| 4 | core_arch::{simd::*, x86::*}, |
| 5 | intrinsics::simd::*, |
| 6 | }; |
| 7 | |
| 8 | #[cfg (test)] |
| 9 | use stdarch_test::assert_instr; |
| 10 | |
| 11 | /// Computes the absolute value of packed 8-bit signed integers in `a` and |
| 12 | /// return the unsigned results. |
| 13 | /// |
| 14 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_abs_epi8) |
| 15 | #[inline ] |
| 16 | #[target_feature (enable = "ssse3" )] |
| 17 | #[cfg_attr (test, assert_instr(pabsb))] |
| 18 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 19 | pub fn _mm_abs_epi8(a: __m128i) -> __m128i { |
| 20 | unsafe { |
| 21 | let a: i8x16 = a.as_i8x16(); |
| 22 | let zero: i8x16 = i8x16::ZERO; |
| 23 | let r: i8x16 = simd_select::<m8x16, _>(mask:simd_lt(a, zero), if_true:simd_neg(a), if_false:a); |
| 24 | transmute(src:r) |
| 25 | } |
| 26 | } |
| 27 | |
| 28 | /// Computes the absolute value of each of the packed 16-bit signed integers in |
| 29 | /// `a` and |
| 30 | /// return the 16-bit unsigned integer |
| 31 | /// |
| 32 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_abs_epi16) |
| 33 | #[inline ] |
| 34 | #[target_feature (enable = "ssse3" )] |
| 35 | #[cfg_attr (test, assert_instr(pabsw))] |
| 36 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 37 | pub fn _mm_abs_epi16(a: __m128i) -> __m128i { |
| 38 | unsafe { |
| 39 | let a: i16x8 = a.as_i16x8(); |
| 40 | let zero: i16x8 = i16x8::ZERO; |
| 41 | let r: i16x8 = simd_select::<m16x8, _>(mask:simd_lt(a, zero), if_true:simd_neg(a), if_false:a); |
| 42 | transmute(src:r) |
| 43 | } |
| 44 | } |
| 45 | |
| 46 | /// Computes the absolute value of each of the packed 32-bit signed integers in |
| 47 | /// `a` and |
| 48 | /// return the 32-bit unsigned integer |
| 49 | /// |
| 50 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_abs_epi32) |
| 51 | #[inline ] |
| 52 | #[target_feature (enable = "ssse3" )] |
| 53 | #[cfg_attr (test, assert_instr(pabsd))] |
| 54 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 55 | pub fn _mm_abs_epi32(a: __m128i) -> __m128i { |
| 56 | unsafe { |
| 57 | let a: i32x4 = a.as_i32x4(); |
| 58 | let zero: i32x4 = i32x4::ZERO; |
| 59 | let r: i32x4 = simd_select::<m32x4, _>(mask:simd_lt(a, zero), if_true:simd_neg(a), if_false:a); |
| 60 | transmute(src:r) |
| 61 | } |
| 62 | } |
| 63 | |
| 64 | /// Shuffles bytes from `a` according to the content of `b`. |
| 65 | /// |
| 66 | /// The last 4 bits of each byte of `b` are used as addresses |
| 67 | /// into the 16 bytes of `a`. |
| 68 | /// |
| 69 | /// In addition, if the highest significant bit of a byte of `b` |
| 70 | /// is set, the respective destination byte is set to 0. |
| 71 | /// |
| 72 | /// Picturing `a` and `b` as `[u8; 16]`, `_mm_shuffle_epi8` is |
| 73 | /// logically equivalent to: |
| 74 | /// |
| 75 | /// ``` |
| 76 | /// fn mm_shuffle_epi8(a: [u8; 16], b: [u8; 16]) -> [u8; 16] { |
| 77 | /// let mut r = [0u8; 16]; |
| 78 | /// for i in 0..16 { |
| 79 | /// // if the most significant bit of b is set, |
| 80 | /// // then the destination byte is set to 0. |
| 81 | /// if b[i] & 0x80 == 0u8 { |
| 82 | /// r[i] = a[(b[i] % 16) as usize]; |
| 83 | /// } |
| 84 | /// } |
| 85 | /// r |
| 86 | /// } |
| 87 | /// ``` |
| 88 | /// |
| 89 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_shuffle_epi8) |
| 90 | #[inline ] |
| 91 | #[target_feature (enable = "ssse3" )] |
| 92 | #[cfg_attr (test, assert_instr(pshufb))] |
| 93 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 94 | pub fn _mm_shuffle_epi8(a: __m128i, b: __m128i) -> __m128i { |
| 95 | unsafe { transmute(src:pshufb128(a.as_u8x16(), b.as_u8x16())) } |
| 96 | } |
| 97 | |
| 98 | /// Concatenate 16-byte blocks in `a` and `b` into a 32-byte temporary result, |
| 99 | /// shift the result right by `n` bytes, and returns the low 16 bytes. |
| 100 | /// |
| 101 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_alignr_epi8) |
| 102 | #[inline ] |
| 103 | #[target_feature (enable = "ssse3" )] |
| 104 | #[cfg_attr (test, assert_instr(palignr, IMM8 = 15))] |
| 105 | #[rustc_legacy_const_generics (2)] |
| 106 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 107 | pub fn _mm_alignr_epi8<const IMM8: i32>(a: __m128i, b: __m128i) -> __m128i { |
| 108 | static_assert_uimm_bits!(IMM8, 8); |
| 109 | // If palignr is shifting the pair of vectors more than the size of two |
| 110 | // lanes, emit zero. |
| 111 | if IMM8 > 32 { |
| 112 | return _mm_setzero_si128(); |
| 113 | } |
| 114 | // If palignr is shifting the pair of input vectors more than one lane, |
| 115 | // but less than two lanes, convert to shifting in zeroes. |
| 116 | let (a, b) = if IMM8 > 16 { |
| 117 | (_mm_setzero_si128(), a) |
| 118 | } else { |
| 119 | (a, b) |
| 120 | }; |
| 121 | const fn mask(shift: u32, i: u32) -> u32 { |
| 122 | if shift > 32 { |
| 123 | // Unused, but needs to be a valid index. |
| 124 | i |
| 125 | } else if shift > 16 { |
| 126 | shift - 16 + i |
| 127 | } else { |
| 128 | shift + i |
| 129 | } |
| 130 | } |
| 131 | unsafe { |
| 132 | let r: i8x16 = simd_shuffle!( |
| 133 | b.as_i8x16(), |
| 134 | a.as_i8x16(), |
| 135 | [ |
| 136 | mask(IMM8 as u32, 0), |
| 137 | mask(IMM8 as u32, 1), |
| 138 | mask(IMM8 as u32, 2), |
| 139 | mask(IMM8 as u32, 3), |
| 140 | mask(IMM8 as u32, 4), |
| 141 | mask(IMM8 as u32, 5), |
| 142 | mask(IMM8 as u32, 6), |
| 143 | mask(IMM8 as u32, 7), |
| 144 | mask(IMM8 as u32, 8), |
| 145 | mask(IMM8 as u32, 9), |
| 146 | mask(IMM8 as u32, 10), |
| 147 | mask(IMM8 as u32, 11), |
| 148 | mask(IMM8 as u32, 12), |
| 149 | mask(IMM8 as u32, 13), |
| 150 | mask(IMM8 as u32, 14), |
| 151 | mask(IMM8 as u32, 15), |
| 152 | ], |
| 153 | ); |
| 154 | transmute(r) |
| 155 | } |
| 156 | } |
| 157 | |
| 158 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 159 | /// 128-bit vectors of `[8 x i16]`. |
| 160 | /// |
| 161 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_hadd_epi16) |
| 162 | #[inline ] |
| 163 | #[target_feature (enable = "ssse3" )] |
| 164 | #[cfg_attr (test, assert_instr(phaddw))] |
| 165 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 166 | pub fn _mm_hadd_epi16(a: __m128i, b: __m128i) -> __m128i { |
| 167 | let a: i16x8 = a.as_i16x8(); |
| 168 | let b: i16x8 = b.as_i16x8(); |
| 169 | unsafe { |
| 170 | let even: i16x8 = simd_shuffle!(a, b, [0, 2, 4, 6, 8, 10, 12, 14]); |
| 171 | let odd: i16x8 = simd_shuffle!(a, b, [1, 3, 5, 7, 9, 11, 13, 15]); |
| 172 | simd_add(x:even, y:odd).as_m128i() |
| 173 | } |
| 174 | } |
| 175 | |
| 176 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 177 | /// 128-bit vectors of `[8 x i16]`. Positive sums greater than 7FFFh are |
| 178 | /// saturated to 7FFFh. Negative sums less than 8000h are saturated to 8000h. |
| 179 | /// |
| 180 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_hadds_epi16) |
| 181 | #[inline ] |
| 182 | #[target_feature (enable = "ssse3" )] |
| 183 | #[cfg_attr (test, assert_instr(phaddsw))] |
| 184 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 185 | pub fn _mm_hadds_epi16(a: __m128i, b: __m128i) -> __m128i { |
| 186 | unsafe { transmute(src:phaddsw128(a.as_i16x8(), b.as_i16x8())) } |
| 187 | } |
| 188 | |
| 189 | /// Horizontally adds the adjacent pairs of values contained in 2 packed |
| 190 | /// 128-bit vectors of `[4 x i32]`. |
| 191 | /// |
| 192 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_hadd_epi32) |
| 193 | #[inline ] |
| 194 | #[target_feature (enable = "ssse3" )] |
| 195 | #[cfg_attr (test, assert_instr(phaddd))] |
| 196 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 197 | pub fn _mm_hadd_epi32(a: __m128i, b: __m128i) -> __m128i { |
| 198 | let a: i32x4 = a.as_i32x4(); |
| 199 | let b: i32x4 = b.as_i32x4(); |
| 200 | unsafe { |
| 201 | let even: i32x4 = simd_shuffle!(a, b, [0, 2, 4, 6]); |
| 202 | let odd: i32x4 = simd_shuffle!(a, b, [1, 3, 5, 7]); |
| 203 | simd_add(x:even, y:odd).as_m128i() |
| 204 | } |
| 205 | } |
| 206 | |
| 207 | /// Horizontally subtract the adjacent pairs of values contained in 2 |
| 208 | /// packed 128-bit vectors of `[8 x i16]`. |
| 209 | /// |
| 210 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_hsub_epi16) |
| 211 | #[inline ] |
| 212 | #[target_feature (enable = "ssse3" )] |
| 213 | #[cfg_attr (test, assert_instr(phsubw))] |
| 214 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 215 | pub fn _mm_hsub_epi16(a: __m128i, b: __m128i) -> __m128i { |
| 216 | let a: i16x8 = a.as_i16x8(); |
| 217 | let b: i16x8 = b.as_i16x8(); |
| 218 | unsafe { |
| 219 | let even: i16x8 = simd_shuffle!(a, b, [0, 2, 4, 6, 8, 10, 12, 14]); |
| 220 | let odd: i16x8 = simd_shuffle!(a, b, [1, 3, 5, 7, 9, 11, 13, 15]); |
| 221 | simd_sub(lhs:even, rhs:odd).as_m128i() |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | /// Horizontally subtract the adjacent pairs of values contained in 2 |
| 226 | /// packed 128-bit vectors of `[8 x i16]`. Positive differences greater than |
| 227 | /// 7FFFh are saturated to 7FFFh. Negative differences less than 8000h are |
| 228 | /// saturated to 8000h. |
| 229 | /// |
| 230 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_hsubs_epi16) |
| 231 | #[inline ] |
| 232 | #[target_feature (enable = "ssse3" )] |
| 233 | #[cfg_attr (test, assert_instr(phsubsw))] |
| 234 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 235 | pub fn _mm_hsubs_epi16(a: __m128i, b: __m128i) -> __m128i { |
| 236 | unsafe { transmute(src:phsubsw128(a.as_i16x8(), b.as_i16x8())) } |
| 237 | } |
| 238 | |
| 239 | /// Horizontally subtract the adjacent pairs of values contained in 2 |
| 240 | /// packed 128-bit vectors of `[4 x i32]`. |
| 241 | /// |
| 242 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_hsub_epi32) |
| 243 | #[inline ] |
| 244 | #[target_feature (enable = "ssse3" )] |
| 245 | #[cfg_attr (test, assert_instr(phsubd))] |
| 246 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 247 | pub fn _mm_hsub_epi32(a: __m128i, b: __m128i) -> __m128i { |
| 248 | let a: i32x4 = a.as_i32x4(); |
| 249 | let b: i32x4 = b.as_i32x4(); |
| 250 | unsafe { |
| 251 | let even: i32x4 = simd_shuffle!(a, b, [0, 2, 4, 6]); |
| 252 | let odd: i32x4 = simd_shuffle!(a, b, [1, 3, 5, 7]); |
| 253 | simd_sub(lhs:even, rhs:odd).as_m128i() |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | /// Multiplies corresponding pairs of packed 8-bit unsigned integer |
| 258 | /// values contained in the first source operand and packed 8-bit signed |
| 259 | /// integer values contained in the second source operand, add pairs of |
| 260 | /// contiguous products with signed saturation, and writes the 16-bit sums to |
| 261 | /// the corresponding bits in the destination. |
| 262 | /// |
| 263 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_maddubs_epi16) |
| 264 | #[inline ] |
| 265 | #[target_feature (enable = "ssse3" )] |
| 266 | #[cfg_attr (test, assert_instr(pmaddubsw))] |
| 267 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 268 | pub fn _mm_maddubs_epi16(a: __m128i, b: __m128i) -> __m128i { |
| 269 | unsafe { transmute(src:pmaddubsw128(a.as_u8x16(), b.as_i8x16())) } |
| 270 | } |
| 271 | |
| 272 | /// Multiplies packed 16-bit signed integer values, truncate the 32-bit |
| 273 | /// product to the 18 most significant bits by right-shifting, round the |
| 274 | /// truncated value by adding 1, and write bits `[16:1]` to the destination. |
| 275 | /// |
| 276 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mulhrs_epi16) |
| 277 | #[inline ] |
| 278 | #[target_feature (enable = "ssse3" )] |
| 279 | #[cfg_attr (test, assert_instr(pmulhrsw))] |
| 280 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 281 | pub fn _mm_mulhrs_epi16(a: __m128i, b: __m128i) -> __m128i { |
| 282 | unsafe { transmute(src:pmulhrsw128(a.as_i16x8(), b.as_i16x8())) } |
| 283 | } |
| 284 | |
| 285 | /// Negates packed 8-bit integers in `a` when the corresponding signed 8-bit |
| 286 | /// integer in `b` is negative, and returns the result. |
| 287 | /// Elements in result are zeroed out when the corresponding element in `b` |
| 288 | /// is zero. |
| 289 | /// |
| 290 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_sign_epi8) |
| 291 | #[inline ] |
| 292 | #[target_feature (enable = "ssse3" )] |
| 293 | #[cfg_attr (test, assert_instr(psignb))] |
| 294 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 295 | pub fn _mm_sign_epi8(a: __m128i, b: __m128i) -> __m128i { |
| 296 | unsafe { transmute(src:psignb128(a.as_i8x16(), b.as_i8x16())) } |
| 297 | } |
| 298 | |
| 299 | /// Negates packed 16-bit integers in `a` when the corresponding signed 16-bit |
| 300 | /// integer in `b` is negative, and returns the results. |
| 301 | /// Elements in result are zeroed out when the corresponding element in `b` |
| 302 | /// is zero. |
| 303 | /// |
| 304 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_sign_epi16) |
| 305 | #[inline ] |
| 306 | #[target_feature (enable = "ssse3" )] |
| 307 | #[cfg_attr (test, assert_instr(psignw))] |
| 308 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 309 | pub fn _mm_sign_epi16(a: __m128i, b: __m128i) -> __m128i { |
| 310 | unsafe { transmute(src:psignw128(a.as_i16x8(), b.as_i16x8())) } |
| 311 | } |
| 312 | |
| 313 | /// Negates packed 32-bit integers in `a` when the corresponding signed 32-bit |
| 314 | /// integer in `b` is negative, and returns the results. |
| 315 | /// Element in result are zeroed out when the corresponding element in `b` |
| 316 | /// is zero. |
| 317 | /// |
| 318 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_sign_epi32) |
| 319 | #[inline ] |
| 320 | #[target_feature (enable = "ssse3" )] |
| 321 | #[cfg_attr (test, assert_instr(psignd))] |
| 322 | #[stable (feature = "simd_x86" , since = "1.27.0" )] |
| 323 | pub fn _mm_sign_epi32(a: __m128i, b: __m128i) -> __m128i { |
| 324 | unsafe { transmute(src:psignd128(a.as_i32x4(), b.as_i32x4())) } |
| 325 | } |
| 326 | |
| 327 | #[allow (improper_ctypes)] |
| 328 | unsafe extern "C" { |
| 329 | #[link_name = "llvm.x86.ssse3.pshuf.b.128" ] |
| 330 | unsafefn pshufb128(a: u8x16, b: u8x16) -> u8x16; |
| 331 | |
| 332 | #[link_name = "llvm.x86.ssse3.phadd.sw.128" ] |
| 333 | unsafefn phaddsw128(a: i16x8, b: i16x8) -> i16x8; |
| 334 | |
| 335 | #[link_name = "llvm.x86.ssse3.phsub.sw.128" ] |
| 336 | unsafefn phsubsw128(a: i16x8, b: i16x8) -> i16x8; |
| 337 | |
| 338 | #[link_name = "llvm.x86.ssse3.pmadd.ub.sw.128" ] |
| 339 | unsafefn pmaddubsw128(a: u8x16, b: i8x16) -> i16x8; |
| 340 | |
| 341 | #[link_name = "llvm.x86.ssse3.pmul.hr.sw.128" ] |
| 342 | unsafefn pmulhrsw128(a: i16x8, b: i16x8) -> i16x8; |
| 343 | |
| 344 | #[link_name = "llvm.x86.ssse3.psign.b.128" ] |
| 345 | unsafefn psignb128(a: i8x16, b: i8x16) -> i8x16; |
| 346 | |
| 347 | #[link_name = "llvm.x86.ssse3.psign.w.128" ] |
| 348 | unsafefn psignw128(a: i16x8, b: i16x8) -> i16x8; |
| 349 | |
| 350 | #[link_name = "llvm.x86.ssse3.psign.d.128" ] |
| 351 | unsafefn psignd128(a: i32x4, b: i32x4) -> i32x4; |
| 352 | } |
| 353 | |
| 354 | #[cfg (test)] |
| 355 | mod tests { |
| 356 | use stdarch_test::simd_test; |
| 357 | |
| 358 | use crate::core_arch::x86::*; |
| 359 | |
| 360 | #[simd_test(enable = "ssse3" )] |
| 361 | unsafe fn test_mm_abs_epi8() { |
| 362 | let r = _mm_abs_epi8(_mm_set1_epi8(-5)); |
| 363 | assert_eq_m128i(r, _mm_set1_epi8(5)); |
| 364 | } |
| 365 | |
| 366 | #[simd_test(enable = "ssse3" )] |
| 367 | unsafe fn test_mm_abs_epi16() { |
| 368 | let r = _mm_abs_epi16(_mm_set1_epi16(-5)); |
| 369 | assert_eq_m128i(r, _mm_set1_epi16(5)); |
| 370 | } |
| 371 | |
| 372 | #[simd_test(enable = "ssse3" )] |
| 373 | unsafe fn test_mm_abs_epi32() { |
| 374 | let r = _mm_abs_epi32(_mm_set1_epi32(-5)); |
| 375 | assert_eq_m128i(r, _mm_set1_epi32(5)); |
| 376 | } |
| 377 | |
| 378 | #[simd_test(enable = "ssse3" )] |
| 379 | unsafe fn test_mm_shuffle_epi8() { |
| 380 | #[rustfmt::skip] |
| 381 | let a = _mm_setr_epi8( |
| 382 | 1, 2, 3, 4, 5, 6, 7, 8, |
| 383 | 9, 10, 11, 12, 13, 14, 15, 16, |
| 384 | ); |
| 385 | #[rustfmt::skip] |
| 386 | let b = _mm_setr_epi8( |
| 387 | 4, 128_u8 as i8, 4, 3, |
| 388 | 24, 12, 6, 19, |
| 389 | 12, 5, 5, 10, |
| 390 | 4, 1, 8, 0, |
| 391 | ); |
| 392 | let expected = _mm_setr_epi8(5, 0, 5, 4, 9, 13, 7, 4, 13, 6, 6, 11, 5, 2, 9, 1); |
| 393 | let r = _mm_shuffle_epi8(a, b); |
| 394 | assert_eq_m128i(r, expected); |
| 395 | |
| 396 | // Test indices greater than 15 wrapping around |
| 397 | let b = _mm_add_epi8(b, _mm_set1_epi8(32)); |
| 398 | let r = _mm_shuffle_epi8(a, b); |
| 399 | assert_eq_m128i(r, expected); |
| 400 | } |
| 401 | |
| 402 | #[simd_test(enable = "ssse3" )] |
| 403 | unsafe fn test_mm_alignr_epi8() { |
| 404 | #[rustfmt::skip] |
| 405 | let a = _mm_setr_epi8( |
| 406 | 1, 2, 3, 4, 5, 6, 7, 8, |
| 407 | 9, 10, 11, 12, 13, 14, 15, 16, |
| 408 | ); |
| 409 | #[rustfmt::skip] |
| 410 | let b = _mm_setr_epi8( |
| 411 | 4, 63, 4, 3, |
| 412 | 24, 12, 6, 19, |
| 413 | 12, 5, 5, 10, |
| 414 | 4, 1, 8, 0, |
| 415 | ); |
| 416 | let r = _mm_alignr_epi8::<33>(a, b); |
| 417 | assert_eq_m128i(r, _mm_set1_epi8(0)); |
| 418 | |
| 419 | let r = _mm_alignr_epi8::<17>(a, b); |
| 420 | #[rustfmt::skip] |
| 421 | let expected = _mm_setr_epi8( |
| 422 | 2, 3, 4, 5, 6, 7, 8, 9, |
| 423 | 10, 11, 12, 13, 14, 15, 16, 0, |
| 424 | ); |
| 425 | assert_eq_m128i(r, expected); |
| 426 | |
| 427 | let r = _mm_alignr_epi8::<16>(a, b); |
| 428 | assert_eq_m128i(r, a); |
| 429 | |
| 430 | let r = _mm_alignr_epi8::<15>(a, b); |
| 431 | #[rustfmt::skip] |
| 432 | let expected = _mm_setr_epi8( |
| 433 | 0, 1, 2, 3, 4, 5, 6, 7, |
| 434 | 8, 9, 10, 11, 12, 13, 14, 15, |
| 435 | ); |
| 436 | assert_eq_m128i(r, expected); |
| 437 | |
| 438 | let r = _mm_alignr_epi8::<0>(a, b); |
| 439 | assert_eq_m128i(r, b); |
| 440 | } |
| 441 | |
| 442 | #[simd_test(enable = "ssse3" )] |
| 443 | unsafe fn test_mm_hadd_epi16() { |
| 444 | let a = _mm_setr_epi16(1, 2, 3, 4, 5, 6, 7, 8); |
| 445 | let b = _mm_setr_epi16(4, 128, 4, 3, 24, 12, 6, 19); |
| 446 | let expected = _mm_setr_epi16(3, 7, 11, 15, 132, 7, 36, 25); |
| 447 | let r = _mm_hadd_epi16(a, b); |
| 448 | assert_eq_m128i(r, expected); |
| 449 | |
| 450 | // Test wrapping on overflow |
| 451 | let a = _mm_setr_epi16(i16::MAX, 1, i16::MAX, 2, i16::MAX, 3, i16::MAX, 4); |
| 452 | let b = _mm_setr_epi16(i16::MIN, -1, i16::MIN, -2, i16::MIN, -3, i16::MIN, -4); |
| 453 | let expected = _mm_setr_epi16( |
| 454 | i16::MIN, |
| 455 | i16::MIN + 1, |
| 456 | i16::MIN + 2, |
| 457 | i16::MIN + 3, |
| 458 | i16::MAX, |
| 459 | i16::MAX - 1, |
| 460 | i16::MAX - 2, |
| 461 | i16::MAX - 3, |
| 462 | ); |
| 463 | let r = _mm_hadd_epi16(a, b); |
| 464 | assert_eq_m128i(r, expected); |
| 465 | } |
| 466 | |
| 467 | #[simd_test(enable = "ssse3" )] |
| 468 | unsafe fn test_mm_hadds_epi16() { |
| 469 | let a = _mm_setr_epi16(1, 2, 3, 4, 5, 6, 7, 8); |
| 470 | let b = _mm_setr_epi16(4, 128, 4, 3, 32767, 1, -32768, -1); |
| 471 | let expected = _mm_setr_epi16(3, 7, 11, 15, 132, 7, 32767, -32768); |
| 472 | let r = _mm_hadds_epi16(a, b); |
| 473 | assert_eq_m128i(r, expected); |
| 474 | |
| 475 | // Test saturating on overflow |
| 476 | let a = _mm_setr_epi16(i16::MAX, 1, i16::MAX, 2, i16::MAX, 3, i16::MAX, 4); |
| 477 | let b = _mm_setr_epi16(i16::MIN, -1, i16::MIN, -2, i16::MIN, -3, i16::MIN, -4); |
| 478 | let expected = _mm_setr_epi16( |
| 479 | i16::MAX, |
| 480 | i16::MAX, |
| 481 | i16::MAX, |
| 482 | i16::MAX, |
| 483 | i16::MIN, |
| 484 | i16::MIN, |
| 485 | i16::MIN, |
| 486 | i16::MIN, |
| 487 | ); |
| 488 | let r = _mm_hadds_epi16(a, b); |
| 489 | assert_eq_m128i(r, expected); |
| 490 | } |
| 491 | |
| 492 | #[simd_test(enable = "ssse3" )] |
| 493 | unsafe fn test_mm_hadd_epi32() { |
| 494 | let a = _mm_setr_epi32(1, 2, 3, 4); |
| 495 | let b = _mm_setr_epi32(4, 128, 4, 3); |
| 496 | let expected = _mm_setr_epi32(3, 7, 132, 7); |
| 497 | let r = _mm_hadd_epi32(a, b); |
| 498 | assert_eq_m128i(r, expected); |
| 499 | |
| 500 | // Test wrapping on overflow |
| 501 | let a = _mm_setr_epi32(i32::MAX, 1, i32::MAX, 2); |
| 502 | let b = _mm_setr_epi32(i32::MIN, -1, i32::MIN, -2); |
| 503 | let expected = _mm_setr_epi32(i32::MIN, i32::MIN + 1, i32::MAX, i32::MAX - 1); |
| 504 | let r = _mm_hadd_epi32(a, b); |
| 505 | assert_eq_m128i(r, expected); |
| 506 | } |
| 507 | |
| 508 | #[simd_test(enable = "ssse3" )] |
| 509 | unsafe fn test_mm_hsub_epi16() { |
| 510 | let a = _mm_setr_epi16(1, 2, 3, 4, 5, 6, 7, 8); |
| 511 | let b = _mm_setr_epi16(4, 128, 4, 3, 24, 12, 6, 19); |
| 512 | let expected = _mm_setr_epi16(-1, -1, -1, -1, -124, 1, 12, -13); |
| 513 | let r = _mm_hsub_epi16(a, b); |
| 514 | assert_eq_m128i(r, expected); |
| 515 | |
| 516 | // Test wrapping on overflow |
| 517 | let a = _mm_setr_epi16(i16::MAX, -1, i16::MAX, -2, i16::MAX, -3, i16::MAX, -4); |
| 518 | let b = _mm_setr_epi16(i16::MIN, 1, i16::MIN, 2, i16::MIN, 3, i16::MIN, 4); |
| 519 | let expected = _mm_setr_epi16( |
| 520 | i16::MIN, |
| 521 | i16::MIN + 1, |
| 522 | i16::MIN + 2, |
| 523 | i16::MIN + 3, |
| 524 | i16::MAX, |
| 525 | i16::MAX - 1, |
| 526 | i16::MAX - 2, |
| 527 | i16::MAX - 3, |
| 528 | ); |
| 529 | let r = _mm_hsub_epi16(a, b); |
| 530 | assert_eq_m128i(r, expected); |
| 531 | } |
| 532 | |
| 533 | #[simd_test(enable = "ssse3" )] |
| 534 | unsafe fn test_mm_hsubs_epi16() { |
| 535 | let a = _mm_setr_epi16(1, 2, 3, 4, 5, 6, 7, 8); |
| 536 | let b = _mm_setr_epi16(4, 128, 4, 3, 32767, -1, -32768, 1); |
| 537 | let expected = _mm_setr_epi16(-1, -1, -1, -1, -124, 1, 32767, -32768); |
| 538 | let r = _mm_hsubs_epi16(a, b); |
| 539 | assert_eq_m128i(r, expected); |
| 540 | |
| 541 | // Test saturating on overflow |
| 542 | let a = _mm_setr_epi16(i16::MAX, -1, i16::MAX, -2, i16::MAX, -3, i16::MAX, -4); |
| 543 | let b = _mm_setr_epi16(i16::MIN, 1, i16::MIN, 2, i16::MIN, 3, i16::MIN, 4); |
| 544 | let expected = _mm_setr_epi16( |
| 545 | i16::MAX, |
| 546 | i16::MAX, |
| 547 | i16::MAX, |
| 548 | i16::MAX, |
| 549 | i16::MIN, |
| 550 | i16::MIN, |
| 551 | i16::MIN, |
| 552 | i16::MIN, |
| 553 | ); |
| 554 | let r = _mm_hsubs_epi16(a, b); |
| 555 | assert_eq_m128i(r, expected); |
| 556 | } |
| 557 | |
| 558 | #[simd_test(enable = "ssse3" )] |
| 559 | unsafe fn test_mm_hsub_epi32() { |
| 560 | let a = _mm_setr_epi32(1, 2, 3, 4); |
| 561 | let b = _mm_setr_epi32(4, 128, 4, 3); |
| 562 | let expected = _mm_setr_epi32(-1, -1, -124, 1); |
| 563 | let r = _mm_hsub_epi32(a, b); |
| 564 | assert_eq_m128i(r, expected); |
| 565 | |
| 566 | // Test wrapping on overflow |
| 567 | let a = _mm_setr_epi32(i32::MAX, -1, i32::MAX, -2); |
| 568 | let b = _mm_setr_epi32(i32::MIN, 1, i32::MIN, 2); |
| 569 | let expected = _mm_setr_epi32(i32::MIN, i32::MIN + 1, i32::MAX, i32::MAX - 1); |
| 570 | let r = _mm_hsub_epi32(a, b); |
| 571 | assert_eq_m128i(r, expected); |
| 572 | } |
| 573 | |
| 574 | #[simd_test(enable = "ssse3" )] |
| 575 | unsafe fn test_mm_maddubs_epi16() { |
| 576 | #[rustfmt::skip] |
| 577 | let a = _mm_setr_epi8( |
| 578 | 1, 2, 3, 4, 5, 6, 7, 8, |
| 579 | 9, 10, 11, 12, 13, 14, 15, 16, |
| 580 | ); |
| 581 | #[rustfmt::skip] |
| 582 | let b = _mm_setr_epi8( |
| 583 | 4, 63, 4, 3, |
| 584 | 24, 12, 6, 19, |
| 585 | 12, 5, 5, 10, |
| 586 | 4, 1, 8, 0, |
| 587 | ); |
| 588 | let expected = _mm_setr_epi16(130, 24, 192, 194, 158, 175, 66, 120); |
| 589 | let r = _mm_maddubs_epi16(a, b); |
| 590 | assert_eq_m128i(r, expected); |
| 591 | |
| 592 | // Test widening and saturation |
| 593 | #[rustfmt::skip] |
| 594 | let a = _mm_setr_epi8( |
| 595 | u8::MAX as i8, u8::MAX as i8, |
| 596 | u8::MAX as i8, u8::MAX as i8, |
| 597 | u8::MAX as i8, u8::MAX as i8, |
| 598 | 100, 100, 0, 0, |
| 599 | 0, 0, 0, 0, 0, 0, |
| 600 | ); |
| 601 | #[rustfmt::skip] |
| 602 | let b = _mm_setr_epi8( |
| 603 | i8::MAX, i8::MAX, |
| 604 | i8::MAX, i8::MIN, |
| 605 | i8::MIN, i8::MIN, |
| 606 | 50, 15, 0, 0, 0, |
| 607 | 0, 0, 0, 0, 0, |
| 608 | ); |
| 609 | let expected = _mm_setr_epi16(i16::MAX, -255, i16::MIN, 6500, 0, 0, 0, 0); |
| 610 | let r = _mm_maddubs_epi16(a, b); |
| 611 | assert_eq_m128i(r, expected); |
| 612 | } |
| 613 | |
| 614 | #[simd_test(enable = "ssse3" )] |
| 615 | unsafe fn test_mm_mulhrs_epi16() { |
| 616 | let a = _mm_setr_epi16(1, 2, 3, 4, 5, 6, 7, 8); |
| 617 | let b = _mm_setr_epi16(4, 128, 4, 3, 32767, -1, -32768, 1); |
| 618 | let expected = _mm_setr_epi16(0, 0, 0, 0, 5, 0, -7, 0); |
| 619 | let r = _mm_mulhrs_epi16(a, b); |
| 620 | assert_eq_m128i(r, expected); |
| 621 | |
| 622 | // Test extreme values |
| 623 | let a = _mm_setr_epi16(i16::MAX, i16::MIN, i16::MIN, 0, 0, 0, 0, 0); |
| 624 | let b = _mm_setr_epi16(i16::MAX, i16::MIN, i16::MAX, 0, 0, 0, 0, 0); |
| 625 | let expected = _mm_setr_epi16(i16::MAX - 1, i16::MIN, -i16::MAX, 0, 0, 0, 0, 0); |
| 626 | let r = _mm_mulhrs_epi16(a, b); |
| 627 | assert_eq_m128i(r, expected); |
| 628 | } |
| 629 | |
| 630 | #[simd_test(enable = "ssse3" )] |
| 631 | unsafe fn test_mm_sign_epi8() { |
| 632 | #[rustfmt::skip] |
| 633 | let a = _mm_setr_epi8( |
| 634 | 1, 2, 3, 4, 5, 6, 7, 8, |
| 635 | 9, 10, 11, 12, 13, -14, -15, 16, |
| 636 | ); |
| 637 | #[rustfmt::skip] |
| 638 | let b = _mm_setr_epi8( |
| 639 | 4, 63, -4, 3, 24, 12, -6, -19, |
| 640 | 12, 5, -5, 10, 4, 1, -8, 0, |
| 641 | ); |
| 642 | #[rustfmt::skip] |
| 643 | let expected = _mm_setr_epi8( |
| 644 | 1, 2, -3, 4, 5, 6, -7, -8, |
| 645 | 9, 10, -11, 12, 13, -14, 15, 0, |
| 646 | ); |
| 647 | let r = _mm_sign_epi8(a, b); |
| 648 | assert_eq_m128i(r, expected); |
| 649 | } |
| 650 | |
| 651 | #[simd_test(enable = "ssse3" )] |
| 652 | unsafe fn test_mm_sign_epi16() { |
| 653 | let a = _mm_setr_epi16(1, 2, 3, 4, -5, -6, 7, 8); |
| 654 | let b = _mm_setr_epi16(4, 128, 0, 3, 1, -1, -2, 1); |
| 655 | let expected = _mm_setr_epi16(1, 2, 0, 4, -5, 6, -7, 8); |
| 656 | let r = _mm_sign_epi16(a, b); |
| 657 | assert_eq_m128i(r, expected); |
| 658 | } |
| 659 | |
| 660 | #[simd_test(enable = "ssse3" )] |
| 661 | unsafe fn test_mm_sign_epi32() { |
| 662 | let a = _mm_setr_epi32(-1, 2, 3, 4); |
| 663 | let b = _mm_setr_epi32(1, -1, 1, 0); |
| 664 | let expected = _mm_setr_epi32(-1, -2, 3, 0); |
| 665 | let r = _mm_sign_epi32(a, b); |
| 666 | assert_eq_m128i(r, expected); |
| 667 | } |
| 668 | } |
| 669 | |