1 | //! Galois Field New Instructions (GFNI) |
2 | //! |
3 | //! The intrinsics here correspond to those in the `immintrin.h` C header. |
4 | //! |
5 | //! The reference is [Intel 64 and IA-32 Architectures Software Developer's |
6 | //! Manual Volume 2: Instruction Set Reference, A-Z][intel64_ref]. |
7 | //! |
8 | //! [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 |
9 | |
10 | use crate::core_arch::simd::i8x16; |
11 | use crate::core_arch::simd::i8x32; |
12 | use crate::core_arch::simd::i8x64; |
13 | use crate::core_arch::x86::__m128i; |
14 | use crate::core_arch::x86::__m256i; |
15 | use crate::core_arch::x86::__m512i; |
16 | use crate::core_arch::x86::__mmask16; |
17 | use crate::core_arch::x86::__mmask32; |
18 | use crate::core_arch::x86::__mmask64; |
19 | use crate::core_arch::x86::_mm256_setzero_si256; |
20 | use crate::core_arch::x86::_mm512_setzero_si512; |
21 | use crate::core_arch::x86::_mm_setzero_si128; |
22 | use crate::core_arch::x86::m128iExt; |
23 | use crate::core_arch::x86::m256iExt; |
24 | use crate::core_arch::x86::m512iExt; |
25 | use crate::intrinsics::simd::simd_select_bitmask; |
26 | use crate::mem::transmute; |
27 | |
28 | #[cfg (test)] |
29 | use stdarch_test::assert_instr; |
30 | |
31 | #[allow (improper_ctypes)] |
32 | extern "C" { |
33 | #[link_name = "llvm.x86.vgf2p8affineinvqb.512" ] |
34 | fn vgf2p8affineinvqb_512(x: i8x64, a: i8x64, imm8: u8) -> i8x64; |
35 | #[link_name = "llvm.x86.vgf2p8affineinvqb.256" ] |
36 | fn vgf2p8affineinvqb_256(x: i8x32, a: i8x32, imm8: u8) -> i8x32; |
37 | #[link_name = "llvm.x86.vgf2p8affineinvqb.128" ] |
38 | fn vgf2p8affineinvqb_128(x: i8x16, a: i8x16, imm8: u8) -> i8x16; |
39 | #[link_name = "llvm.x86.vgf2p8affineqb.512" ] |
40 | fn vgf2p8affineqb_512(x: i8x64, a: i8x64, imm8: u8) -> i8x64; |
41 | #[link_name = "llvm.x86.vgf2p8affineqb.256" ] |
42 | fn vgf2p8affineqb_256(x: i8x32, a: i8x32, imm8: u8) -> i8x32; |
43 | #[link_name = "llvm.x86.vgf2p8affineqb.128" ] |
44 | fn vgf2p8affineqb_128(x: i8x16, a: i8x16, imm8: u8) -> i8x16; |
45 | #[link_name = "llvm.x86.vgf2p8mulb.512" ] |
46 | fn vgf2p8mulb_512(a: i8x64, b: i8x64) -> i8x64; |
47 | #[link_name = "llvm.x86.vgf2p8mulb.256" ] |
48 | fn vgf2p8mulb_256(a: i8x32, b: i8x32) -> i8x32; |
49 | #[link_name = "llvm.x86.vgf2p8mulb.128" ] |
50 | fn vgf2p8mulb_128(a: i8x16, b: i8x16) -> i8x16; |
51 | } |
52 | |
53 | // LLVM requires AVX512BW for a lot of these instructions, see |
54 | // https://github.com/llvm/llvm-project/blob/release/9.x/clang/include/clang/Basic/BuiltinsX86.def#L457 |
55 | // however our tests also require the target feature list to match Intel's |
56 | // which *doesn't* require AVX512BW but only AVX512F, so we added the redundant AVX512F |
57 | // requirement (for now) |
58 | // also see |
59 | // https://github.com/llvm/llvm-project/blob/release/9.x/clang/lib/Headers/gfniintrin.h |
60 | // for forcing GFNI, BW and optionally VL extension |
61 | |
62 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
63 | /// The field is in polynomial representation with the reduction polynomial |
64 | /// x^8 + x^4 + x^3 + x + 1. |
65 | /// |
66 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_gf2p8mul_epi8) |
67 | #[inline ] |
68 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
69 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
70 | #[cfg_attr (test, assert_instr(vgf2p8mulb))] |
71 | pub unsafe fn _mm512_gf2p8mul_epi8(a: __m512i, b: __m512i) -> __m512i { |
72 | transmute(src:vgf2p8mulb_512(a:a.as_i8x64(), b:b.as_i8x64())) |
73 | } |
74 | |
75 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
76 | /// The field is in polynomial representation with the reduction polynomial |
77 | /// x^8 + x^4 + x^3 + x + 1. |
78 | /// |
79 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
80 | /// Otherwise the computation result is written into the result. |
81 | /// |
82 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_mask_gf2p8mul_epi8) |
83 | #[inline ] |
84 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
85 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
86 | #[cfg_attr (test, assert_instr(vgf2p8mulb))] |
87 | pub unsafe fn _mm512_mask_gf2p8mul_epi8( |
88 | src: __m512i, |
89 | k: __mmask64, |
90 | a: __m512i, |
91 | b: __m512i, |
92 | ) -> __m512i { |
93 | transmute(src:simd_select_bitmask( |
94 | m:k, |
95 | yes:vgf2p8mulb_512(a.as_i8x64(), b.as_i8x64()), |
96 | no:src.as_i8x64(), |
97 | )) |
98 | } |
99 | |
100 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
101 | /// The field is in polynomial representation with the reduction polynomial |
102 | /// x^8 + x^4 + x^3 + x + 1. |
103 | /// |
104 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
105 | /// Otherwise the computation result is written into the result. |
106 | /// |
107 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_maskz_gf2p8mul_epi8) |
108 | #[inline ] |
109 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
110 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
111 | #[cfg_attr (test, assert_instr(vgf2p8mulb))] |
112 | pub unsafe fn _mm512_maskz_gf2p8mul_epi8(k: __mmask64, a: __m512i, b: __m512i) -> __m512i { |
113 | let zero: i8x64 = _mm512_setzero_si512().as_i8x64(); |
114 | transmute(src:simd_select_bitmask( |
115 | m:k, |
116 | yes:vgf2p8mulb_512(a.as_i8x64(), b.as_i8x64()), |
117 | no:zero, |
118 | )) |
119 | } |
120 | |
121 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
122 | /// The field is in polynomial representation with the reduction polynomial |
123 | /// x^8 + x^4 + x^3 + x + 1. |
124 | /// |
125 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_gf2p8mul_epi8) |
126 | #[inline ] |
127 | #[target_feature (enable = "gfni,avx" )] |
128 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
129 | #[cfg_attr (test, assert_instr(vgf2p8mulb))] |
130 | pub unsafe fn _mm256_gf2p8mul_epi8(a: __m256i, b: __m256i) -> __m256i { |
131 | transmute(src:vgf2p8mulb_256(a:a.as_i8x32(), b:b.as_i8x32())) |
132 | } |
133 | |
134 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
135 | /// The field is in polynomial representation with the reduction polynomial |
136 | /// x^8 + x^4 + x^3 + x + 1. |
137 | /// |
138 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
139 | /// Otherwise the computation result is written into the result. |
140 | /// |
141 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_mask_gf2p8mul_epi8) |
142 | #[inline ] |
143 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
144 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
145 | #[cfg_attr (test, assert_instr(vgf2p8mulb))] |
146 | pub unsafe fn _mm256_mask_gf2p8mul_epi8( |
147 | src: __m256i, |
148 | k: __mmask32, |
149 | a: __m256i, |
150 | b: __m256i, |
151 | ) -> __m256i { |
152 | transmute(src:simd_select_bitmask( |
153 | m:k, |
154 | yes:vgf2p8mulb_256(a.as_i8x32(), b.as_i8x32()), |
155 | no:src.as_i8x32(), |
156 | )) |
157 | } |
158 | |
159 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
160 | /// The field is in polynomial representation with the reduction polynomial |
161 | /// x^8 + x^4 + x^3 + x + 1. |
162 | /// |
163 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
164 | /// Otherwise the computation result is written into the result. |
165 | /// |
166 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_maskz_gf2p8mul_epi8) |
167 | #[inline ] |
168 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
169 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
170 | #[cfg_attr (test, assert_instr(vgf2p8mulb))] |
171 | pub unsafe fn _mm256_maskz_gf2p8mul_epi8(k: __mmask32, a: __m256i, b: __m256i) -> __m256i { |
172 | let zero: i8x32 = _mm256_setzero_si256().as_i8x32(); |
173 | transmute(src:simd_select_bitmask( |
174 | m:k, |
175 | yes:vgf2p8mulb_256(a.as_i8x32(), b.as_i8x32()), |
176 | no:zero, |
177 | )) |
178 | } |
179 | |
180 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
181 | /// The field is in polynomial representation with the reduction polynomial |
182 | /// x^8 + x^4 + x^3 + x + 1. |
183 | /// |
184 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_gf2p8mul_epi8) |
185 | #[inline ] |
186 | #[target_feature (enable = "gfni" )] |
187 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
188 | #[cfg_attr (test, assert_instr(gf2p8mulb))] |
189 | pub unsafe fn _mm_gf2p8mul_epi8(a: __m128i, b: __m128i) -> __m128i { |
190 | transmute(src:vgf2p8mulb_128(a:a.as_i8x16(), b:b.as_i8x16())) |
191 | } |
192 | |
193 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
194 | /// The field is in polynomial representation with the reduction polynomial |
195 | /// x^8 + x^4 + x^3 + x + 1. |
196 | /// |
197 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
198 | /// Otherwise the computation result is written into the result. |
199 | /// |
200 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mask_gf2p8mul_epi8) |
201 | #[inline ] |
202 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
203 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
204 | #[cfg_attr (test, assert_instr(vgf2p8mulb))] |
205 | pub unsafe fn _mm_mask_gf2p8mul_epi8( |
206 | src: __m128i, |
207 | k: __mmask16, |
208 | a: __m128i, |
209 | b: __m128i, |
210 | ) -> __m128i { |
211 | transmute(src:simd_select_bitmask( |
212 | m:k, |
213 | yes:vgf2p8mulb_128(a.as_i8x16(), b.as_i8x16()), |
214 | no:src.as_i8x16(), |
215 | )) |
216 | } |
217 | |
218 | /// Performs a multiplication in GF(2^8) on the packed bytes. |
219 | /// The field is in polynomial representation with the reduction polynomial |
220 | /// x^8 + x^4 + x^3 + x + 1. |
221 | /// |
222 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
223 | /// Otherwise the computation result is written into the result. |
224 | /// |
225 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_maskz_gf2p8mul_epi8) |
226 | #[inline ] |
227 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
228 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
229 | #[cfg_attr (test, assert_instr(vgf2p8mulb))] |
230 | pub unsafe fn _mm_maskz_gf2p8mul_epi8(k: __mmask16, a: __m128i, b: __m128i) -> __m128i { |
231 | let zero: i8x16 = _mm_setzero_si128().as_i8x16(); |
232 | transmute(src:simd_select_bitmask( |
233 | m:k, |
234 | yes:vgf2p8mulb_128(a.as_i8x16(), b.as_i8x16()), |
235 | no:zero, |
236 | )) |
237 | } |
238 | |
239 | /// Performs an affine transformation on the packed bytes in x. |
240 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
241 | /// and b being a constant 8-bit immediate value. |
242 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
243 | /// |
244 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_gf2p8affine_epi8) |
245 | #[inline ] |
246 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
247 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
248 | #[cfg_attr (test, assert_instr(vgf2p8affineqb, B = 0))] |
249 | #[rustc_legacy_const_generics (2)] |
250 | pub unsafe fn _mm512_gf2p8affine_epi64_epi8<const B: i32>(x: __m512i, a: __m512i) -> __m512i { |
251 | static_assert_uimm_bits!(B, 8); |
252 | let b: u8 = B as u8; |
253 | let x: i8x64 = x.as_i8x64(); |
254 | let a: i8x64 = a.as_i8x64(); |
255 | let r: i8x64 = vgf2p8affineqb_512(x, a, imm8:b); |
256 | transmute(src:r) |
257 | } |
258 | |
259 | /// Performs an affine transformation on the packed bytes in x. |
260 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
261 | /// and b being a constant 8-bit immediate value. |
262 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
263 | /// |
264 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
265 | /// Otherwise the computation result is written into the result. |
266 | /// |
267 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_maskz_gf2p8affine_epi8) |
268 | #[inline ] |
269 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
270 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
271 | #[cfg_attr (test, assert_instr(vgf2p8affineqb, B = 0))] |
272 | #[rustc_legacy_const_generics (3)] |
273 | pub unsafe fn _mm512_maskz_gf2p8affine_epi64_epi8<const B: i32>( |
274 | k: __mmask64, |
275 | x: __m512i, |
276 | a: __m512i, |
277 | ) -> __m512i { |
278 | static_assert_uimm_bits!(B, 8); |
279 | let b: u8 = B as u8; |
280 | let zero: i8x64 = _mm512_setzero_si512().as_i8x64(); |
281 | let x: i8x64 = x.as_i8x64(); |
282 | let a: i8x64 = a.as_i8x64(); |
283 | let r: i8x64 = vgf2p8affineqb_512(x, a, imm8:b); |
284 | transmute(src:simd_select_bitmask(m:k, yes:r, no:zero)) |
285 | } |
286 | |
287 | /// Performs an affine transformation on the packed bytes in x. |
288 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
289 | /// and b being a constant 8-bit immediate value. |
290 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
291 | /// |
292 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
293 | /// Otherwise the computation result is written into the result. |
294 | /// |
295 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_mask_gf2p8affine_epi8) |
296 | #[inline ] |
297 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
298 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
299 | #[cfg_attr (test, assert_instr(vgf2p8affineqb, B = 0))] |
300 | #[rustc_legacy_const_generics (4)] |
301 | pub unsafe fn _mm512_mask_gf2p8affine_epi64_epi8<const B: i32>( |
302 | src: __m512i, |
303 | k: __mmask64, |
304 | x: __m512i, |
305 | a: __m512i, |
306 | ) -> __m512i { |
307 | static_assert_uimm_bits!(B, 8); |
308 | let b: u8 = B as u8; |
309 | let x: i8x64 = x.as_i8x64(); |
310 | let a: i8x64 = a.as_i8x64(); |
311 | let r: i8x64 = vgf2p8affineqb_512(x, a, imm8:b); |
312 | transmute(src:simd_select_bitmask(m:k, yes:r, no:src.as_i8x64())) |
313 | } |
314 | |
315 | /// Performs an affine transformation on the packed bytes in x. |
316 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
317 | /// and b being a constant 8-bit immediate value. |
318 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
319 | /// |
320 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_gf2p8affine_epi8) |
321 | #[inline ] |
322 | #[target_feature (enable = "gfni,avx" )] |
323 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
324 | #[cfg_attr (test, assert_instr(vgf2p8affineqb, B = 0))] |
325 | #[rustc_legacy_const_generics (2)] |
326 | pub unsafe fn _mm256_gf2p8affine_epi64_epi8<const B: i32>(x: __m256i, a: __m256i) -> __m256i { |
327 | static_assert_uimm_bits!(B, 8); |
328 | let b: u8 = B as u8; |
329 | let x: i8x32 = x.as_i8x32(); |
330 | let a: i8x32 = a.as_i8x32(); |
331 | let r: i8x32 = vgf2p8affineqb_256(x, a, imm8:b); |
332 | transmute(src:r) |
333 | } |
334 | |
335 | /// Performs an affine transformation on the packed bytes in x. |
336 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
337 | /// and b being a constant 8-bit immediate value. |
338 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
339 | /// |
340 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
341 | /// Otherwise the computation result is written into the result. |
342 | /// |
343 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_maskz_gf2p8affine_epi8) |
344 | #[inline ] |
345 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
346 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
347 | #[cfg_attr (test, assert_instr(vgf2p8affineqb, B = 0))] |
348 | #[rustc_legacy_const_generics (3)] |
349 | pub unsafe fn _mm256_maskz_gf2p8affine_epi64_epi8<const B: i32>( |
350 | k: __mmask32, |
351 | x: __m256i, |
352 | a: __m256i, |
353 | ) -> __m256i { |
354 | static_assert_uimm_bits!(B, 8); |
355 | let b: u8 = B as u8; |
356 | let zero: i8x32 = _mm256_setzero_si256().as_i8x32(); |
357 | let x: i8x32 = x.as_i8x32(); |
358 | let a: i8x32 = a.as_i8x32(); |
359 | let r: i8x32 = vgf2p8affineqb_256(x, a, imm8:b); |
360 | transmute(src:simd_select_bitmask(m:k, yes:r, no:zero)) |
361 | } |
362 | |
363 | /// Performs an affine transformation on the packed bytes in x. |
364 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
365 | /// and b being a constant 8-bit immediate value. |
366 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
367 | /// |
368 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
369 | /// Otherwise the computation result is written into the result. |
370 | /// |
371 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_mask_gf2p8affine_epi8) |
372 | #[inline ] |
373 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
374 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
375 | #[cfg_attr (test, assert_instr(vgf2p8affineqb, B = 0))] |
376 | #[rustc_legacy_const_generics (4)] |
377 | pub unsafe fn _mm256_mask_gf2p8affine_epi64_epi8<const B: i32>( |
378 | src: __m256i, |
379 | k: __mmask32, |
380 | x: __m256i, |
381 | a: __m256i, |
382 | ) -> __m256i { |
383 | static_assert_uimm_bits!(B, 8); |
384 | let b: u8 = B as u8; |
385 | let x: i8x32 = x.as_i8x32(); |
386 | let a: i8x32 = a.as_i8x32(); |
387 | let r: i8x32 = vgf2p8affineqb_256(x, a, imm8:b); |
388 | transmute(src:simd_select_bitmask(m:k, yes:r, no:src.as_i8x32())) |
389 | } |
390 | |
391 | /// Performs an affine transformation on the packed bytes in x. |
392 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
393 | /// and b being a constant 8-bit immediate value. |
394 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
395 | /// |
396 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_gf2p8affine_epi8) |
397 | #[inline ] |
398 | #[target_feature (enable = "gfni" )] |
399 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
400 | #[cfg_attr (test, assert_instr(gf2p8affineqb, B = 0))] |
401 | #[rustc_legacy_const_generics (2)] |
402 | pub unsafe fn _mm_gf2p8affine_epi64_epi8<const B: i32>(x: __m128i, a: __m128i) -> __m128i { |
403 | static_assert_uimm_bits!(B, 8); |
404 | let b: u8 = B as u8; |
405 | let x: i8x16 = x.as_i8x16(); |
406 | let a: i8x16 = a.as_i8x16(); |
407 | let r: i8x16 = vgf2p8affineqb_128(x, a, imm8:b); |
408 | transmute(src:r) |
409 | } |
410 | |
411 | /// Performs an affine transformation on the packed bytes in x. |
412 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
413 | /// and b being a constant 8-bit immediate value. |
414 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
415 | /// |
416 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
417 | /// Otherwise the computation result is written into the result. |
418 | /// |
419 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_maskz_gf2p8affine_epi8) |
420 | #[inline ] |
421 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
422 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
423 | #[cfg_attr (test, assert_instr(vgf2p8affineqb, B = 0))] |
424 | #[rustc_legacy_const_generics (3)] |
425 | pub unsafe fn _mm_maskz_gf2p8affine_epi64_epi8<const B: i32>( |
426 | k: __mmask16, |
427 | x: __m128i, |
428 | a: __m128i, |
429 | ) -> __m128i { |
430 | static_assert_uimm_bits!(B, 8); |
431 | let b: u8 = B as u8; |
432 | let zero: i8x16 = _mm_setzero_si128().as_i8x16(); |
433 | let x: i8x16 = x.as_i8x16(); |
434 | let a: i8x16 = a.as_i8x16(); |
435 | let r: i8x16 = vgf2p8affineqb_128(x, a, imm8:b); |
436 | transmute(src:simd_select_bitmask(m:k, yes:r, no:zero)) |
437 | } |
438 | |
439 | /// Performs an affine transformation on the packed bytes in x. |
440 | /// That is computes a*x+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
441 | /// and b being a constant 8-bit immediate value. |
442 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
443 | /// |
444 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
445 | /// Otherwise the computation result is written into the result. |
446 | /// |
447 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mask_gf2p8affine_epi8) |
448 | #[inline ] |
449 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
450 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
451 | #[cfg_attr (test, assert_instr(vgf2p8affineqb, B = 0))] |
452 | #[rustc_legacy_const_generics (4)] |
453 | pub unsafe fn _mm_mask_gf2p8affine_epi64_epi8<const B: i32>( |
454 | src: __m128i, |
455 | k: __mmask16, |
456 | x: __m128i, |
457 | a: __m128i, |
458 | ) -> __m128i { |
459 | static_assert_uimm_bits!(B, 8); |
460 | let b: u8 = B as u8; |
461 | let x: i8x16 = x.as_i8x16(); |
462 | let a: i8x16 = a.as_i8x16(); |
463 | let r: i8x16 = vgf2p8affineqb_128(x, a, imm8:b); |
464 | transmute(src:simd_select_bitmask(m:k, yes:r, no:src.as_i8x16())) |
465 | } |
466 | |
467 | /// Performs an affine transformation on the inverted packed bytes in x. |
468 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
469 | /// and b being a constant 8-bit immediate value. |
470 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
471 | /// The inverse of 0 is 0. |
472 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
473 | /// |
474 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_gf2p8affineinv_epi64_epi8) |
475 | #[inline ] |
476 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
477 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
478 | #[cfg_attr (test, assert_instr(vgf2p8affineinvqb, B = 0))] |
479 | #[rustc_legacy_const_generics (2)] |
480 | pub unsafe fn _mm512_gf2p8affineinv_epi64_epi8<const B: i32>(x: __m512i, a: __m512i) -> __m512i { |
481 | static_assert_uimm_bits!(B, 8); |
482 | let b: u8 = B as u8; |
483 | let x: i8x64 = x.as_i8x64(); |
484 | let a: i8x64 = a.as_i8x64(); |
485 | let r: i8x64 = vgf2p8affineinvqb_512(x, a, imm8:b); |
486 | transmute(src:r) |
487 | } |
488 | |
489 | /// Performs an affine transformation on the inverted packed bytes in x. |
490 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
491 | /// and b being a constant 8-bit immediate value. |
492 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
493 | /// The inverse of 0 is 0. |
494 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
495 | /// |
496 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
497 | /// Otherwise the computation result is written into the result. |
498 | /// |
499 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_maskz_gf2p8affineinv_epi64_epi8) |
500 | #[inline ] |
501 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
502 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
503 | #[cfg_attr (test, assert_instr(vgf2p8affineinvqb, B = 0))] |
504 | #[rustc_legacy_const_generics (3)] |
505 | pub unsafe fn _mm512_maskz_gf2p8affineinv_epi64_epi8<const B: i32>( |
506 | k: __mmask64, |
507 | x: __m512i, |
508 | a: __m512i, |
509 | ) -> __m512i { |
510 | static_assert_uimm_bits!(B, 8); |
511 | let b: u8 = B as u8; |
512 | let zero: i8x64 = _mm512_setzero_si512().as_i8x64(); |
513 | let x: i8x64 = x.as_i8x64(); |
514 | let a: i8x64 = a.as_i8x64(); |
515 | let r: i8x64 = vgf2p8affineinvqb_512(x, a, imm8:b); |
516 | transmute(src:simd_select_bitmask(m:k, yes:r, no:zero)) |
517 | } |
518 | |
519 | /// Performs an affine transformation on the inverted packed bytes in x. |
520 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
521 | /// and b being a constant 8-bit immediate value. |
522 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
523 | /// The inverse of 0 is 0. |
524 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
525 | /// |
526 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
527 | /// Otherwise the computation result is written into the result. |
528 | /// |
529 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_mask_gf2p8affineinv_epi64_epi8) |
530 | #[inline ] |
531 | #[target_feature (enable = "gfni,avx512bw,avx512f" )] |
532 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
533 | #[cfg_attr (test, assert_instr(vgf2p8affineinvqb, B = 0))] |
534 | #[rustc_legacy_const_generics (4)] |
535 | pub unsafe fn _mm512_mask_gf2p8affineinv_epi64_epi8<const B: i32>( |
536 | src: __m512i, |
537 | k: __mmask64, |
538 | x: __m512i, |
539 | a: __m512i, |
540 | ) -> __m512i { |
541 | static_assert_uimm_bits!(B, 8); |
542 | let b: u8 = B as u8; |
543 | let x: i8x64 = x.as_i8x64(); |
544 | let a: i8x64 = a.as_i8x64(); |
545 | let r: i8x64 = vgf2p8affineinvqb_512(x, a, imm8:b); |
546 | transmute(src:simd_select_bitmask(m:k, yes:r, no:src.as_i8x64())) |
547 | } |
548 | |
549 | /// Performs an affine transformation on the inverted packed bytes in x. |
550 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
551 | /// and b being a constant 8-bit immediate value. |
552 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
553 | /// The inverse of 0 is 0. |
554 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
555 | /// |
556 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_gf2p8affineinv_epi64_epi8) |
557 | #[inline ] |
558 | #[target_feature (enable = "gfni,avx" )] |
559 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
560 | #[cfg_attr (test, assert_instr(vgf2p8affineinvqb, B = 0))] |
561 | #[rustc_legacy_const_generics (2)] |
562 | pub unsafe fn _mm256_gf2p8affineinv_epi64_epi8<const B: i32>(x: __m256i, a: __m256i) -> __m256i { |
563 | static_assert_uimm_bits!(B, 8); |
564 | let b: u8 = B as u8; |
565 | let x: i8x32 = x.as_i8x32(); |
566 | let a: i8x32 = a.as_i8x32(); |
567 | let r: i8x32 = vgf2p8affineinvqb_256(x, a, imm8:b); |
568 | transmute(src:r) |
569 | } |
570 | |
571 | /// Performs an affine transformation on the inverted packed bytes in x. |
572 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
573 | /// and b being a constant 8-bit immediate value. |
574 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
575 | /// The inverse of 0 is 0. |
576 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
577 | /// |
578 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
579 | /// Otherwise the computation result is written into the result. |
580 | /// |
581 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_maskz_gf2p8affineinv_epi64_epi8) |
582 | #[inline ] |
583 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
584 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
585 | #[cfg_attr (test, assert_instr(vgf2p8affineinvqb, B = 0))] |
586 | #[rustc_legacy_const_generics (3)] |
587 | pub unsafe fn _mm256_maskz_gf2p8affineinv_epi64_epi8<const B: i32>( |
588 | k: __mmask32, |
589 | x: __m256i, |
590 | a: __m256i, |
591 | ) -> __m256i { |
592 | static_assert_uimm_bits!(B, 8); |
593 | let b: u8 = B as u8; |
594 | let zero: i8x32 = _mm256_setzero_si256().as_i8x32(); |
595 | let x: i8x32 = x.as_i8x32(); |
596 | let a: i8x32 = a.as_i8x32(); |
597 | let r: i8x32 = vgf2p8affineinvqb_256(x, a, imm8:b); |
598 | transmute(src:simd_select_bitmask(m:k, yes:r, no:zero)) |
599 | } |
600 | |
601 | /// Performs an affine transformation on the inverted packed bytes in x. |
602 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
603 | /// and b being a constant 8-bit immediate value. |
604 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
605 | /// The inverse of 0 is 0. |
606 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
607 | /// |
608 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
609 | /// Otherwise the computation result is written into the result. |
610 | /// |
611 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm256_mask_gf2p8affineinv_epi64_epi8) |
612 | #[inline ] |
613 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
614 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
615 | #[cfg_attr (test, assert_instr(vgf2p8affineinvqb, B = 0))] |
616 | #[rustc_legacy_const_generics (4)] |
617 | pub unsafe fn _mm256_mask_gf2p8affineinv_epi64_epi8<const B: i32>( |
618 | src: __m256i, |
619 | k: __mmask32, |
620 | x: __m256i, |
621 | a: __m256i, |
622 | ) -> __m256i { |
623 | static_assert_uimm_bits!(B, 8); |
624 | let b: u8 = B as u8; |
625 | let x: i8x32 = x.as_i8x32(); |
626 | let a: i8x32 = a.as_i8x32(); |
627 | let r: i8x32 = vgf2p8affineinvqb_256(x, a, imm8:b); |
628 | transmute(src:simd_select_bitmask(m:k, yes:r, no:src.as_i8x32())) |
629 | } |
630 | |
631 | /// Performs an affine transformation on the inverted packed bytes in x. |
632 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
633 | /// and b being a constant 8-bit immediate value. |
634 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
635 | /// The inverse of 0 is 0. |
636 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
637 | /// |
638 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_gf2p8affineinv_epi64_epi8) |
639 | #[inline ] |
640 | #[target_feature (enable = "gfni" )] |
641 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
642 | #[cfg_attr (test, assert_instr(gf2p8affineinvqb, B = 0))] |
643 | #[rustc_legacy_const_generics (2)] |
644 | pub unsafe fn _mm_gf2p8affineinv_epi64_epi8<const B: i32>(x: __m128i, a: __m128i) -> __m128i { |
645 | static_assert_uimm_bits!(B, 8); |
646 | let b: u8 = B as u8; |
647 | let x: i8x16 = x.as_i8x16(); |
648 | let a: i8x16 = a.as_i8x16(); |
649 | let r: i8x16 = vgf2p8affineinvqb_128(x, a, imm8:b); |
650 | transmute(src:r) |
651 | } |
652 | |
653 | /// Performs an affine transformation on the inverted packed bytes in x. |
654 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
655 | /// and b being a constant 8-bit immediate value. |
656 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
657 | /// The inverse of 0 is 0. |
658 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
659 | /// |
660 | /// Uses the writemask in k - elements are zeroed in the result if the corresponding mask bit is not set. |
661 | /// Otherwise the computation result is written into the result. |
662 | /// |
663 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_maskz_gf2p8affineinv_epi64_epi8) |
664 | #[inline ] |
665 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
666 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
667 | #[cfg_attr (test, assert_instr(vgf2p8affineinvqb, B = 0))] |
668 | #[rustc_legacy_const_generics (3)] |
669 | pub unsafe fn _mm_maskz_gf2p8affineinv_epi64_epi8<const B: i32>( |
670 | k: __mmask16, |
671 | x: __m128i, |
672 | a: __m128i, |
673 | ) -> __m128i { |
674 | static_assert_uimm_bits!(B, 8); |
675 | let b: u8 = B as u8; |
676 | let zero: i8x16 = _mm_setzero_si128().as_i8x16(); |
677 | let x: i8x16 = x.as_i8x16(); |
678 | let a: i8x16 = a.as_i8x16(); |
679 | let r: i8x16 = vgf2p8affineinvqb_128(x, a, imm8:b); |
680 | transmute(src:simd_select_bitmask(m:k, yes:r, no:zero)) |
681 | } |
682 | |
683 | /// Performs an affine transformation on the inverted packed bytes in x. |
684 | /// That is computes a*inv(x)+b over the Galois Field 2^8 for each packed byte with a being a 8x8 bit matrix |
685 | /// and b being a constant 8-bit immediate value. |
686 | /// The inverse of a byte is defined with respect to the reduction polynomial x^8+x^4+x^3+x+1. |
687 | /// The inverse of 0 is 0. |
688 | /// Each pack of 8 bytes in x is paired with the 64-bit word at the same position in a. |
689 | /// |
690 | /// Uses the writemask in k - elements are copied from src if the corresponding mask bit is not set. |
691 | /// Otherwise the computation result is written into the result. |
692 | /// |
693 | /// [Intel's documentation](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_mask_gf2p8affineinv_epi64_epi8) |
694 | #[inline ] |
695 | #[target_feature (enable = "gfni,avx512bw,avx512vl" )] |
696 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
697 | #[cfg_attr (test, assert_instr(vgf2p8affineinvqb, B = 0))] |
698 | #[rustc_legacy_const_generics (4)] |
699 | pub unsafe fn _mm_mask_gf2p8affineinv_epi64_epi8<const B: i32>( |
700 | src: __m128i, |
701 | k: __mmask16, |
702 | x: __m128i, |
703 | a: __m128i, |
704 | ) -> __m128i { |
705 | static_assert_uimm_bits!(B, 8); |
706 | let b: u8 = B as u8; |
707 | let x: i8x16 = x.as_i8x16(); |
708 | let a: i8x16 = a.as_i8x16(); |
709 | let r: i8x16 = vgf2p8affineinvqb_128(x, a, imm8:b); |
710 | transmute(src:simd_select_bitmask(m:k, yes:r, no:src.as_i8x16())) |
711 | } |
712 | |
713 | #[cfg (test)] |
714 | mod tests { |
715 | // The constants in the tests below are just bit patterns. They should not |
716 | // be interpreted as integers; signedness does not make sense for them, but |
717 | // __mXXXi happens to be defined in terms of signed integers. |
718 | #![allow (overflowing_literals)] |
719 | |
720 | use core::hint::black_box; |
721 | use core::intrinsics::size_of; |
722 | use stdarch_test::simd_test; |
723 | |
724 | use crate::core_arch::x86::*; |
725 | |
726 | fn mulbyte(left: u8, right: u8) -> u8 { |
727 | // this implementation follows the description in |
728 | // https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_gf2p8mul_epi8 |
729 | const REDUCTION_POLYNOMIAL: u16 = 0x11b; |
730 | let left: u16 = left.into(); |
731 | let right: u16 = right.into(); |
732 | let mut carryless_product: u16 = 0; |
733 | |
734 | // Carryless multiplication |
735 | for i in 0..8 { |
736 | if ((left >> i) & 0x01) != 0 { |
737 | carryless_product ^= right << i; |
738 | } |
739 | } |
740 | |
741 | // reduction, adding in "0" where appropriate to clear out high bits |
742 | // note that REDUCTION_POLYNOMIAL is zero in this context |
743 | for i in (8..=14).rev() { |
744 | if ((carryless_product >> i) & 0x01) != 0 { |
745 | carryless_product ^= REDUCTION_POLYNOMIAL << (i - 8); |
746 | } |
747 | } |
748 | |
749 | carryless_product as u8 |
750 | } |
751 | |
752 | const NUM_TEST_WORDS_512: usize = 4; |
753 | const NUM_TEST_WORDS_256: usize = NUM_TEST_WORDS_512 * 2; |
754 | const NUM_TEST_WORDS_128: usize = NUM_TEST_WORDS_256 * 2; |
755 | const NUM_TEST_ENTRIES: usize = NUM_TEST_WORDS_512 * 64; |
756 | const NUM_TEST_WORDS_64: usize = NUM_TEST_WORDS_128 * 2; |
757 | const NUM_BYTES: usize = 256; |
758 | const NUM_BYTES_WORDS_128: usize = NUM_BYTES / 16; |
759 | const NUM_BYTES_WORDS_256: usize = NUM_BYTES_WORDS_128 / 2; |
760 | const NUM_BYTES_WORDS_512: usize = NUM_BYTES_WORDS_256 / 2; |
761 | |
762 | fn parity(input: u8) -> u8 { |
763 | let mut accumulator = 0; |
764 | for i in 0..8 { |
765 | accumulator ^= (input >> i) & 0x01; |
766 | } |
767 | accumulator |
768 | } |
769 | |
770 | fn mat_vec_multiply_affine(matrix: u64, x: u8, b: u8) -> u8 { |
771 | // this implementation follows the description in |
772 | // https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm_gf2p8affine_epi64_epi8 |
773 | let mut accumulator = 0; |
774 | |
775 | for bit in 0..8 { |
776 | accumulator |= parity(x & matrix.to_le_bytes()[bit]) << (7 - bit); |
777 | } |
778 | |
779 | accumulator ^ b |
780 | } |
781 | |
782 | fn generate_affine_mul_test_data( |
783 | immediate: u8, |
784 | ) -> ( |
785 | [u64; NUM_TEST_WORDS_64], |
786 | [u8; NUM_TEST_ENTRIES], |
787 | [u8; NUM_TEST_ENTRIES], |
788 | ) { |
789 | let mut left: [u64; NUM_TEST_WORDS_64] = [0; NUM_TEST_WORDS_64]; |
790 | let mut right: [u8; NUM_TEST_ENTRIES] = [0; NUM_TEST_ENTRIES]; |
791 | let mut result: [u8; NUM_TEST_ENTRIES] = [0; NUM_TEST_ENTRIES]; |
792 | |
793 | for i in 0..NUM_TEST_WORDS_64 { |
794 | left[i] = (i as u64) * 103 * 101; |
795 | for j in 0..8 { |
796 | let j64 = j as u64; |
797 | right[i * 8 + j] = ((left[i] + j64) % 256) as u8; |
798 | result[i * 8 + j] = mat_vec_multiply_affine(left[i], right[i * 8 + j], immediate); |
799 | } |
800 | } |
801 | |
802 | (left, right, result) |
803 | } |
804 | |
805 | fn generate_inv_tests_data() -> ([u8; NUM_BYTES], [u8; NUM_BYTES]) { |
806 | let mut input: [u8; NUM_BYTES] = [0; NUM_BYTES]; |
807 | let mut result: [u8; NUM_BYTES] = [0; NUM_BYTES]; |
808 | |
809 | for i in 0..NUM_BYTES { |
810 | input[i] = (i % 256) as u8; |
811 | result[i] = if i == 0 { 0 } else { 1 }; |
812 | } |
813 | |
814 | (input, result) |
815 | } |
816 | |
817 | const AES_S_BOX: [u8; NUM_BYTES] = [ |
818 | 0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, |
819 | 0x76, 0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, |
820 | 0x72, 0xc0, 0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, |
821 | 0xd8, 0x31, 0x15, 0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 0x07, 0x12, 0x80, 0xe2, |
822 | 0xeb, 0x27, 0xb2, 0x75, 0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 0x52, 0x3b, 0xd6, |
823 | 0xb3, 0x29, 0xe3, 0x2f, 0x84, 0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 0x6a, 0xcb, |
824 | 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf, 0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 0x45, |
825 | 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8, 0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, |
826 | 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2, 0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, |
827 | 0x17, 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73, 0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, |
828 | 0x90, 0x88, 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb, 0xe0, 0x32, 0x3a, 0x0a, 0x49, |
829 | 0x06, 0x24, 0x5c, 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79, 0xe7, 0xc8, 0x37, 0x6d, |
830 | 0x8d, 0xd5, 0x4e, 0xa9, 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08, 0xba, 0x78, 0x25, |
831 | 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a, 0x70, 0x3e, |
832 | 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e, 0xe1, |
833 | 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf, |
834 | 0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, |
835 | 0x16, |
836 | ]; |
837 | |
838 | fn generate_byte_mul_test_data() -> ( |
839 | [u8; NUM_TEST_ENTRIES], |
840 | [u8; NUM_TEST_ENTRIES], |
841 | [u8; NUM_TEST_ENTRIES], |
842 | ) { |
843 | let mut left: [u8; NUM_TEST_ENTRIES] = [0; NUM_TEST_ENTRIES]; |
844 | let mut right: [u8; NUM_TEST_ENTRIES] = [0; NUM_TEST_ENTRIES]; |
845 | let mut result: [u8; NUM_TEST_ENTRIES] = [0; NUM_TEST_ENTRIES]; |
846 | |
847 | for i in 0..NUM_TEST_ENTRIES { |
848 | left[i] = (i % 256) as u8; |
849 | right[i] = left[i].wrapping_mul(101); |
850 | result[i] = mulbyte(left[i], right[i]); |
851 | } |
852 | |
853 | (left, right, result) |
854 | } |
855 | |
856 | #[target_feature (enable = "sse2" )] |
857 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
858 | unsafe fn load_m128i_word<T>(data: &[T], word_index: usize) -> __m128i { |
859 | let byte_offset = word_index * 16 / size_of::<T>(); |
860 | let pointer = data.as_ptr().add(byte_offset) as *const __m128i; |
861 | _mm_loadu_si128(black_box(pointer)) |
862 | } |
863 | |
864 | #[target_feature (enable = "avx" )] |
865 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
866 | unsafe fn load_m256i_word<T>(data: &[T], word_index: usize) -> __m256i { |
867 | let byte_offset = word_index * 32 / size_of::<T>(); |
868 | let pointer = data.as_ptr().add(byte_offset) as *const __m256i; |
869 | _mm256_loadu_si256(black_box(pointer)) |
870 | } |
871 | |
872 | #[target_feature (enable = "avx512f" )] |
873 | #[unstable (feature = "stdarch_x86_avx512" , issue = "111137" )] |
874 | unsafe fn load_m512i_word<T>(data: &[T], word_index: usize) -> __m512i { |
875 | let byte_offset = word_index * 64 / size_of::<T>(); |
876 | let pointer = data.as_ptr().add(byte_offset) as *const i32; |
877 | _mm512_loadu_si512(black_box(pointer)) |
878 | } |
879 | |
880 | #[simd_test(enable = "gfni,avx512bw" )] |
881 | unsafe fn test_mm512_gf2p8mul_epi8() { |
882 | let (left, right, expected) = generate_byte_mul_test_data(); |
883 | |
884 | for i in 0..NUM_TEST_WORDS_512 { |
885 | let left = load_m512i_word(&left, i); |
886 | let right = load_m512i_word(&right, i); |
887 | let expected = load_m512i_word(&expected, i); |
888 | let result = _mm512_gf2p8mul_epi8(left, right); |
889 | assert_eq_m512i(result, expected); |
890 | } |
891 | } |
892 | |
893 | #[simd_test(enable = "gfni,avx512bw" )] |
894 | unsafe fn test_mm512_maskz_gf2p8mul_epi8() { |
895 | let (left, right, _expected) = generate_byte_mul_test_data(); |
896 | |
897 | for i in 0..NUM_TEST_WORDS_512 { |
898 | let left = load_m512i_word(&left, i); |
899 | let right = load_m512i_word(&right, i); |
900 | let result_zero = _mm512_maskz_gf2p8mul_epi8(0, left, right); |
901 | assert_eq_m512i(result_zero, _mm512_setzero_si512()); |
902 | let mask_bytes: __mmask64 = 0x0F_0F_0F_0F_FF_FF_00_00; |
903 | let mask_words: __mmask16 = 0b01_01_01_01_11_11_00_00; |
904 | let expected_result = _mm512_gf2p8mul_epi8(left, right); |
905 | let result_masked = _mm512_maskz_gf2p8mul_epi8(mask_bytes, left, right); |
906 | let expected_masked = |
907 | _mm512_mask_blend_epi32(mask_words, _mm512_setzero_si512(), expected_result); |
908 | assert_eq_m512i(result_masked, expected_masked); |
909 | } |
910 | } |
911 | |
912 | #[simd_test(enable = "gfni,avx512bw" )] |
913 | unsafe fn test_mm512_mask_gf2p8mul_epi8() { |
914 | let (left, right, _expected) = generate_byte_mul_test_data(); |
915 | |
916 | for i in 0..NUM_TEST_WORDS_512 { |
917 | let left = load_m512i_word(&left, i); |
918 | let right = load_m512i_word(&right, i); |
919 | let result_left = _mm512_mask_gf2p8mul_epi8(left, 0, left, right); |
920 | assert_eq_m512i(result_left, left); |
921 | let mask_bytes: __mmask64 = 0x0F_0F_0F_0F_FF_FF_00_00; |
922 | let mask_words: __mmask16 = 0b01_01_01_01_11_11_00_00; |
923 | let expected_result = _mm512_gf2p8mul_epi8(left, right); |
924 | let result_masked = _mm512_mask_gf2p8mul_epi8(left, mask_bytes, left, right); |
925 | let expected_masked = _mm512_mask_blend_epi32(mask_words, left, expected_result); |
926 | assert_eq_m512i(result_masked, expected_masked); |
927 | } |
928 | } |
929 | |
930 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
931 | unsafe fn test_mm256_gf2p8mul_epi8() { |
932 | let (left, right, expected) = generate_byte_mul_test_data(); |
933 | |
934 | for i in 0..NUM_TEST_WORDS_256 { |
935 | let left = load_m256i_word(&left, i); |
936 | let right = load_m256i_word(&right, i); |
937 | let expected = load_m256i_word(&expected, i); |
938 | let result = _mm256_gf2p8mul_epi8(left, right); |
939 | assert_eq_m256i(result, expected); |
940 | } |
941 | } |
942 | |
943 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
944 | unsafe fn test_mm256_maskz_gf2p8mul_epi8() { |
945 | let (left, right, _expected) = generate_byte_mul_test_data(); |
946 | |
947 | for i in 0..NUM_TEST_WORDS_256 { |
948 | let left = load_m256i_word(&left, i); |
949 | let right = load_m256i_word(&right, i); |
950 | let result_zero = _mm256_maskz_gf2p8mul_epi8(0, left, right); |
951 | assert_eq_m256i(result_zero, _mm256_setzero_si256()); |
952 | let mask_bytes: __mmask32 = 0x0F_F0_FF_00; |
953 | const MASK_WORDS: i32 = 0b01_10_11_00; |
954 | let expected_result = _mm256_gf2p8mul_epi8(left, right); |
955 | let result_masked = _mm256_maskz_gf2p8mul_epi8(mask_bytes, left, right); |
956 | let expected_masked = |
957 | _mm256_blend_epi32::<MASK_WORDS>(_mm256_setzero_si256(), expected_result); |
958 | assert_eq_m256i(result_masked, expected_masked); |
959 | } |
960 | } |
961 | |
962 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
963 | unsafe fn test_mm256_mask_gf2p8mul_epi8() { |
964 | let (left, right, _expected) = generate_byte_mul_test_data(); |
965 | |
966 | for i in 0..NUM_TEST_WORDS_256 { |
967 | let left = load_m256i_word(&left, i); |
968 | let right = load_m256i_word(&right, i); |
969 | let result_left = _mm256_mask_gf2p8mul_epi8(left, 0, left, right); |
970 | assert_eq_m256i(result_left, left); |
971 | let mask_bytes: __mmask32 = 0x0F_F0_FF_00; |
972 | const MASK_WORDS: i32 = 0b01_10_11_00; |
973 | let expected_result = _mm256_gf2p8mul_epi8(left, right); |
974 | let result_masked = _mm256_mask_gf2p8mul_epi8(left, mask_bytes, left, right); |
975 | let expected_masked = _mm256_blend_epi32::<MASK_WORDS>(left, expected_result); |
976 | assert_eq_m256i(result_masked, expected_masked); |
977 | } |
978 | } |
979 | |
980 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
981 | unsafe fn test_mm_gf2p8mul_epi8() { |
982 | let (left, right, expected) = generate_byte_mul_test_data(); |
983 | |
984 | for i in 0..NUM_TEST_WORDS_128 { |
985 | let left = load_m128i_word(&left, i); |
986 | let right = load_m128i_word(&right, i); |
987 | let expected = load_m128i_word(&expected, i); |
988 | let result = _mm_gf2p8mul_epi8(left, right); |
989 | assert_eq_m128i(result, expected); |
990 | } |
991 | } |
992 | |
993 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
994 | unsafe fn test_mm_maskz_gf2p8mul_epi8() { |
995 | let (left, right, _expected) = generate_byte_mul_test_data(); |
996 | |
997 | for i in 0..NUM_TEST_WORDS_128 { |
998 | let left = load_m128i_word(&left, i); |
999 | let right = load_m128i_word(&right, i); |
1000 | let result_zero = _mm_maskz_gf2p8mul_epi8(0, left, right); |
1001 | assert_eq_m128i(result_zero, _mm_setzero_si128()); |
1002 | let mask_bytes: __mmask16 = 0x0F_F0; |
1003 | const MASK_WORDS: i32 = 0b01_10; |
1004 | let expected_result = _mm_gf2p8mul_epi8(left, right); |
1005 | let result_masked = _mm_maskz_gf2p8mul_epi8(mask_bytes, left, right); |
1006 | let expected_masked = |
1007 | _mm_blend_epi32::<MASK_WORDS>(_mm_setzero_si128(), expected_result); |
1008 | assert_eq_m128i(result_masked, expected_masked); |
1009 | } |
1010 | } |
1011 | |
1012 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1013 | unsafe fn test_mm_mask_gf2p8mul_epi8() { |
1014 | let (left, right, _expected) = generate_byte_mul_test_data(); |
1015 | |
1016 | for i in 0..NUM_TEST_WORDS_128 { |
1017 | let left = load_m128i_word(&left, i); |
1018 | let right = load_m128i_word(&right, i); |
1019 | let result_left = _mm_mask_gf2p8mul_epi8(left, 0, left, right); |
1020 | assert_eq_m128i(result_left, left); |
1021 | let mask_bytes: __mmask16 = 0x0F_F0; |
1022 | const MASK_WORDS: i32 = 0b01_10; |
1023 | let expected_result = _mm_gf2p8mul_epi8(left, right); |
1024 | let result_masked = _mm_mask_gf2p8mul_epi8(left, mask_bytes, left, right); |
1025 | let expected_masked = _mm_blend_epi32::<MASK_WORDS>(left, expected_result); |
1026 | assert_eq_m128i(result_masked, expected_masked); |
1027 | } |
1028 | } |
1029 | |
1030 | #[simd_test(enable = "gfni,avx512bw" )] |
1031 | unsafe fn test_mm512_gf2p8affine_epi64_epi8() { |
1032 | let identity: i64 = 0x01_02_04_08_10_20_40_80; |
1033 | const IDENTITY_BYTE: i32 = 0; |
1034 | let constant: i64 = 0; |
1035 | const CONSTANT_BYTE: i32 = 0x63; |
1036 | let identity = _mm512_set1_epi64(identity); |
1037 | let constant = _mm512_set1_epi64(constant); |
1038 | let constant_reference = _mm512_set1_epi8(CONSTANT_BYTE as i8); |
1039 | |
1040 | let (bytes, more_bytes, _) = generate_byte_mul_test_data(); |
1041 | let (matrices, vectors, references) = generate_affine_mul_test_data(IDENTITY_BYTE as u8); |
1042 | |
1043 | for i in 0..NUM_TEST_WORDS_512 { |
1044 | let data = load_m512i_word(&bytes, i); |
1045 | let result = _mm512_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(data, identity); |
1046 | assert_eq_m512i(result, data); |
1047 | let result = _mm512_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(data, constant); |
1048 | assert_eq_m512i(result, constant_reference); |
1049 | let data = load_m512i_word(&more_bytes, i); |
1050 | let result = _mm512_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(data, identity); |
1051 | assert_eq_m512i(result, data); |
1052 | let result = _mm512_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(data, constant); |
1053 | assert_eq_m512i(result, constant_reference); |
1054 | |
1055 | let matrix = load_m512i_word(&matrices, i); |
1056 | let vector = load_m512i_word(&vectors, i); |
1057 | let reference = load_m512i_word(&references, i); |
1058 | |
1059 | let result = _mm512_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(vector, matrix); |
1060 | assert_eq_m512i(result, reference); |
1061 | } |
1062 | } |
1063 | |
1064 | #[simd_test(enable = "gfni,avx512bw" )] |
1065 | unsafe fn test_mm512_maskz_gf2p8affine_epi64_epi8() { |
1066 | const CONSTANT_BYTE: i32 = 0x63; |
1067 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1068 | |
1069 | for i in 0..NUM_TEST_WORDS_512 { |
1070 | let matrix = load_m512i_word(&matrices, i); |
1071 | let vector = load_m512i_word(&vectors, i); |
1072 | let result_zero = |
1073 | _mm512_maskz_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(0, vector, matrix); |
1074 | assert_eq_m512i(result_zero, _mm512_setzero_si512()); |
1075 | let mask_bytes: __mmask64 = 0x0F_0F_0F_0F_FF_FF_00_00; |
1076 | let mask_words: __mmask16 = 0b01_01_01_01_11_11_00_00; |
1077 | let expected_result = _mm512_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1078 | let result_masked = |
1079 | _mm512_maskz_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(mask_bytes, vector, matrix); |
1080 | let expected_masked = |
1081 | _mm512_mask_blend_epi32(mask_words, _mm512_setzero_si512(), expected_result); |
1082 | assert_eq_m512i(result_masked, expected_masked); |
1083 | } |
1084 | } |
1085 | |
1086 | #[simd_test(enable = "gfni,avx512bw" )] |
1087 | unsafe fn test_mm512_mask_gf2p8affine_epi64_epi8() { |
1088 | const CONSTANT_BYTE: i32 = 0x63; |
1089 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1090 | |
1091 | for i in 0..NUM_TEST_WORDS_512 { |
1092 | let left = load_m512i_word(&vectors, i); |
1093 | let right = load_m512i_word(&matrices, i); |
1094 | let result_left = |
1095 | _mm512_mask_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, 0, left, right); |
1096 | assert_eq_m512i(result_left, left); |
1097 | let mask_bytes: __mmask64 = 0x0F_0F_0F_0F_FF_FF_00_00; |
1098 | let mask_words: __mmask16 = 0b01_01_01_01_11_11_00_00; |
1099 | let expected_result = _mm512_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, right); |
1100 | let result_masked = |
1101 | _mm512_mask_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, mask_bytes, left, right); |
1102 | let expected_masked = _mm512_mask_blend_epi32(mask_words, left, expected_result); |
1103 | assert_eq_m512i(result_masked, expected_masked); |
1104 | } |
1105 | } |
1106 | |
1107 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1108 | unsafe fn test_mm256_gf2p8affine_epi64_epi8() { |
1109 | let identity: i64 = 0x01_02_04_08_10_20_40_80; |
1110 | const IDENTITY_BYTE: i32 = 0; |
1111 | let constant: i64 = 0; |
1112 | const CONSTANT_BYTE: i32 = 0x63; |
1113 | let identity = _mm256_set1_epi64x(identity); |
1114 | let constant = _mm256_set1_epi64x(constant); |
1115 | let constant_reference = _mm256_set1_epi8(CONSTANT_BYTE as i8); |
1116 | |
1117 | let (bytes, more_bytes, _) = generate_byte_mul_test_data(); |
1118 | let (matrices, vectors, references) = generate_affine_mul_test_data(IDENTITY_BYTE as u8); |
1119 | |
1120 | for i in 0..NUM_TEST_WORDS_256 { |
1121 | let data = load_m256i_word(&bytes, i); |
1122 | let result = _mm256_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(data, identity); |
1123 | assert_eq_m256i(result, data); |
1124 | let result = _mm256_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(data, constant); |
1125 | assert_eq_m256i(result, constant_reference); |
1126 | let data = load_m256i_word(&more_bytes, i); |
1127 | let result = _mm256_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(data, identity); |
1128 | assert_eq_m256i(result, data); |
1129 | let result = _mm256_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(data, constant); |
1130 | assert_eq_m256i(result, constant_reference); |
1131 | |
1132 | let matrix = load_m256i_word(&matrices, i); |
1133 | let vector = load_m256i_word(&vectors, i); |
1134 | let reference = load_m256i_word(&references, i); |
1135 | |
1136 | let result = _mm256_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(vector, matrix); |
1137 | assert_eq_m256i(result, reference); |
1138 | } |
1139 | } |
1140 | |
1141 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1142 | unsafe fn test_mm256_maskz_gf2p8affine_epi64_epi8() { |
1143 | const CONSTANT_BYTE: i32 = 0x63; |
1144 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1145 | |
1146 | for i in 0..NUM_TEST_WORDS_256 { |
1147 | let matrix = load_m256i_word(&matrices, i); |
1148 | let vector = load_m256i_word(&vectors, i); |
1149 | let result_zero = |
1150 | _mm256_maskz_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(0, vector, matrix); |
1151 | assert_eq_m256i(result_zero, _mm256_setzero_si256()); |
1152 | let mask_bytes: __mmask32 = 0xFF_0F_F0_00; |
1153 | const MASK_WORDS: i32 = 0b11_01_10_00; |
1154 | let expected_result = _mm256_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1155 | let result_masked = |
1156 | _mm256_maskz_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(mask_bytes, vector, matrix); |
1157 | let expected_masked = |
1158 | _mm256_blend_epi32::<MASK_WORDS>(_mm256_setzero_si256(), expected_result); |
1159 | assert_eq_m256i(result_masked, expected_masked); |
1160 | } |
1161 | } |
1162 | |
1163 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1164 | unsafe fn test_mm256_mask_gf2p8affine_epi64_epi8() { |
1165 | const CONSTANT_BYTE: i32 = 0x63; |
1166 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1167 | |
1168 | for i in 0..NUM_TEST_WORDS_256 { |
1169 | let left = load_m256i_word(&vectors, i); |
1170 | let right = load_m256i_word(&matrices, i); |
1171 | let result_left = |
1172 | _mm256_mask_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, 0, left, right); |
1173 | assert_eq_m256i(result_left, left); |
1174 | let mask_bytes: __mmask32 = 0xFF_0F_F0_00; |
1175 | const MASK_WORDS: i32 = 0b11_01_10_00; |
1176 | let expected_result = _mm256_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, right); |
1177 | let result_masked = |
1178 | _mm256_mask_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, mask_bytes, left, right); |
1179 | let expected_masked = _mm256_blend_epi32::<MASK_WORDS>(left, expected_result); |
1180 | assert_eq_m256i(result_masked, expected_masked); |
1181 | } |
1182 | } |
1183 | |
1184 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1185 | unsafe fn test_mm_gf2p8affine_epi64_epi8() { |
1186 | let identity: i64 = 0x01_02_04_08_10_20_40_80; |
1187 | const IDENTITY_BYTE: i32 = 0; |
1188 | let constant: i64 = 0; |
1189 | const CONSTANT_BYTE: i32 = 0x63; |
1190 | let identity = _mm_set1_epi64x(identity); |
1191 | let constant = _mm_set1_epi64x(constant); |
1192 | let constant_reference = _mm_set1_epi8(CONSTANT_BYTE as i8); |
1193 | |
1194 | let (bytes, more_bytes, _) = generate_byte_mul_test_data(); |
1195 | let (matrices, vectors, references) = generate_affine_mul_test_data(IDENTITY_BYTE as u8); |
1196 | |
1197 | for i in 0..NUM_TEST_WORDS_128 { |
1198 | let data = load_m128i_word(&bytes, i); |
1199 | let result = _mm_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(data, identity); |
1200 | assert_eq_m128i(result, data); |
1201 | let result = _mm_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(data, constant); |
1202 | assert_eq_m128i(result, constant_reference); |
1203 | let data = load_m128i_word(&more_bytes, i); |
1204 | let result = _mm_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(data, identity); |
1205 | assert_eq_m128i(result, data); |
1206 | let result = _mm_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(data, constant); |
1207 | assert_eq_m128i(result, constant_reference); |
1208 | |
1209 | let matrix = load_m128i_word(&matrices, i); |
1210 | let vector = load_m128i_word(&vectors, i); |
1211 | let reference = load_m128i_word(&references, i); |
1212 | |
1213 | let result = _mm_gf2p8affine_epi64_epi8::<IDENTITY_BYTE>(vector, matrix); |
1214 | assert_eq_m128i(result, reference); |
1215 | } |
1216 | } |
1217 | |
1218 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1219 | unsafe fn test_mm_maskz_gf2p8affine_epi64_epi8() { |
1220 | const CONSTANT_BYTE: i32 = 0x63; |
1221 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1222 | |
1223 | for i in 0..NUM_TEST_WORDS_128 { |
1224 | let matrix = load_m128i_word(&matrices, i); |
1225 | let vector = load_m128i_word(&vectors, i); |
1226 | let result_zero = _mm_maskz_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(0, vector, matrix); |
1227 | assert_eq_m128i(result_zero, _mm_setzero_si128()); |
1228 | let mask_bytes: __mmask16 = 0x0F_F0; |
1229 | const MASK_WORDS: i32 = 0b01_10; |
1230 | let expected_result = _mm_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1231 | let result_masked = |
1232 | _mm_maskz_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(mask_bytes, vector, matrix); |
1233 | let expected_masked = |
1234 | _mm_blend_epi32::<MASK_WORDS>(_mm_setzero_si128(), expected_result); |
1235 | assert_eq_m128i(result_masked, expected_masked); |
1236 | } |
1237 | } |
1238 | |
1239 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1240 | unsafe fn test_mm_mask_gf2p8affine_epi64_epi8() { |
1241 | const CONSTANT_BYTE: i32 = 0x63; |
1242 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1243 | |
1244 | for i in 0..NUM_TEST_WORDS_128 { |
1245 | let left = load_m128i_word(&vectors, i); |
1246 | let right = load_m128i_word(&matrices, i); |
1247 | let result_left = |
1248 | _mm_mask_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, 0, left, right); |
1249 | assert_eq_m128i(result_left, left); |
1250 | let mask_bytes: __mmask16 = 0x0F_F0; |
1251 | const MASK_WORDS: i32 = 0b01_10; |
1252 | let expected_result = _mm_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, right); |
1253 | let result_masked = |
1254 | _mm_mask_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(left, mask_bytes, left, right); |
1255 | let expected_masked = _mm_blend_epi32::<MASK_WORDS>(left, expected_result); |
1256 | assert_eq_m128i(result_masked, expected_masked); |
1257 | } |
1258 | } |
1259 | |
1260 | #[simd_test(enable = "gfni,avx512bw" )] |
1261 | unsafe fn test_mm512_gf2p8affineinv_epi64_epi8() { |
1262 | let identity: i64 = 0x01_02_04_08_10_20_40_80; |
1263 | const IDENTITY_BYTE: i32 = 0; |
1264 | const CONSTANT_BYTE: i32 = 0x63; |
1265 | let identity = _mm512_set1_epi64(identity); |
1266 | |
1267 | // validate inversion |
1268 | let (inputs, results) = generate_inv_tests_data(); |
1269 | |
1270 | for i in 0..NUM_BYTES_WORDS_512 { |
1271 | let input = load_m512i_word(&inputs, i); |
1272 | let reference = load_m512i_word(&results, i); |
1273 | let result = _mm512_gf2p8affineinv_epi64_epi8::<IDENTITY_BYTE>(input, identity); |
1274 | let remultiplied = _mm512_gf2p8mul_epi8(result, input); |
1275 | assert_eq_m512i(remultiplied, reference); |
1276 | } |
1277 | |
1278 | // validate subsequent affine operation |
1279 | let (matrices, vectors, _affine_expected) = |
1280 | generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1281 | |
1282 | for i in 0..NUM_TEST_WORDS_512 { |
1283 | let vector = load_m512i_word(&vectors, i); |
1284 | let matrix = load_m512i_word(&matrices, i); |
1285 | |
1286 | let inv_vec = _mm512_gf2p8affineinv_epi64_epi8::<IDENTITY_BYTE>(vector, identity); |
1287 | let reference = _mm512_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(inv_vec, matrix); |
1288 | let result = _mm512_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1289 | assert_eq_m512i(result, reference); |
1290 | } |
1291 | |
1292 | // validate everything by virtue of checking against the AES SBox |
1293 | const AES_S_BOX_MATRIX: i64 = 0xF1_E3_C7_8F_1F_3E_7C_F8; |
1294 | let sbox_matrix = _mm512_set1_epi64(AES_S_BOX_MATRIX); |
1295 | |
1296 | for i in 0..NUM_BYTES_WORDS_512 { |
1297 | let reference = load_m512i_word(&AES_S_BOX, i); |
1298 | let input = load_m512i_word(&inputs, i); |
1299 | let result = _mm512_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(input, sbox_matrix); |
1300 | assert_eq_m512i(result, reference); |
1301 | } |
1302 | } |
1303 | |
1304 | #[simd_test(enable = "gfni,avx512bw" )] |
1305 | unsafe fn test_mm512_maskz_gf2p8affineinv_epi64_epi8() { |
1306 | const CONSTANT_BYTE: i32 = 0x63; |
1307 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1308 | |
1309 | for i in 0..NUM_TEST_WORDS_512 { |
1310 | let matrix = load_m512i_word(&matrices, i); |
1311 | let vector = load_m512i_word(&vectors, i); |
1312 | let result_zero = |
1313 | _mm512_maskz_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(0, vector, matrix); |
1314 | assert_eq_m512i(result_zero, _mm512_setzero_si512()); |
1315 | let mask_bytes: __mmask64 = 0x0F_0F_0F_0F_FF_FF_00_00; |
1316 | let mask_words: __mmask16 = 0b01_01_01_01_11_11_00_00; |
1317 | let expected_result = _mm512_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1318 | let result_masked = |
1319 | _mm512_maskz_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(mask_bytes, vector, matrix); |
1320 | let expected_masked = |
1321 | _mm512_mask_blend_epi32(mask_words, _mm512_setzero_si512(), expected_result); |
1322 | assert_eq_m512i(result_masked, expected_masked); |
1323 | } |
1324 | } |
1325 | |
1326 | #[simd_test(enable = "gfni,avx512bw" )] |
1327 | unsafe fn test_mm512_mask_gf2p8affineinv_epi64_epi8() { |
1328 | const CONSTANT_BYTE: i32 = 0x63; |
1329 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1330 | |
1331 | for i in 0..NUM_TEST_WORDS_512 { |
1332 | let left = load_m512i_word(&vectors, i); |
1333 | let right = load_m512i_word(&matrices, i); |
1334 | let result_left = |
1335 | _mm512_mask_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(left, 0, left, right); |
1336 | assert_eq_m512i(result_left, left); |
1337 | let mask_bytes: __mmask64 = 0x0F_0F_0F_0F_FF_FF_00_00; |
1338 | let mask_words: __mmask16 = 0b01_01_01_01_11_11_00_00; |
1339 | let expected_result = _mm512_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(left, right); |
1340 | let result_masked = _mm512_mask_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>( |
1341 | left, mask_bytes, left, right, |
1342 | ); |
1343 | let expected_masked = _mm512_mask_blend_epi32(mask_words, left, expected_result); |
1344 | assert_eq_m512i(result_masked, expected_masked); |
1345 | } |
1346 | } |
1347 | |
1348 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1349 | unsafe fn test_mm256_gf2p8affineinv_epi64_epi8() { |
1350 | let identity: i64 = 0x01_02_04_08_10_20_40_80; |
1351 | const IDENTITY_BYTE: i32 = 0; |
1352 | const CONSTANT_BYTE: i32 = 0x63; |
1353 | let identity = _mm256_set1_epi64x(identity); |
1354 | |
1355 | // validate inversion |
1356 | let (inputs, results) = generate_inv_tests_data(); |
1357 | |
1358 | for i in 0..NUM_BYTES_WORDS_256 { |
1359 | let input = load_m256i_word(&inputs, i); |
1360 | let reference = load_m256i_word(&results, i); |
1361 | let result = _mm256_gf2p8affineinv_epi64_epi8::<IDENTITY_BYTE>(input, identity); |
1362 | let remultiplied = _mm256_gf2p8mul_epi8(result, input); |
1363 | assert_eq_m256i(remultiplied, reference); |
1364 | } |
1365 | |
1366 | // validate subsequent affine operation |
1367 | let (matrices, vectors, _affine_expected) = |
1368 | generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1369 | |
1370 | for i in 0..NUM_TEST_WORDS_256 { |
1371 | let vector = load_m256i_word(&vectors, i); |
1372 | let matrix = load_m256i_word(&matrices, i); |
1373 | |
1374 | let inv_vec = _mm256_gf2p8affineinv_epi64_epi8::<IDENTITY_BYTE>(vector, identity); |
1375 | let reference = _mm256_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(inv_vec, matrix); |
1376 | let result = _mm256_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1377 | assert_eq_m256i(result, reference); |
1378 | } |
1379 | |
1380 | // validate everything by virtue of checking against the AES SBox |
1381 | const AES_S_BOX_MATRIX: i64 = 0xF1_E3_C7_8F_1F_3E_7C_F8; |
1382 | let sbox_matrix = _mm256_set1_epi64x(AES_S_BOX_MATRIX); |
1383 | |
1384 | for i in 0..NUM_BYTES_WORDS_256 { |
1385 | let reference = load_m256i_word(&AES_S_BOX, i); |
1386 | let input = load_m256i_word(&inputs, i); |
1387 | let result = _mm256_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(input, sbox_matrix); |
1388 | assert_eq_m256i(result, reference); |
1389 | } |
1390 | } |
1391 | |
1392 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1393 | unsafe fn test_mm256_maskz_gf2p8affineinv_epi64_epi8() { |
1394 | const CONSTANT_BYTE: i32 = 0x63; |
1395 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1396 | |
1397 | for i in 0..NUM_TEST_WORDS_256 { |
1398 | let matrix = load_m256i_word(&matrices, i); |
1399 | let vector = load_m256i_word(&vectors, i); |
1400 | let result_zero = |
1401 | _mm256_maskz_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(0, vector, matrix); |
1402 | assert_eq_m256i(result_zero, _mm256_setzero_si256()); |
1403 | let mask_bytes: __mmask32 = 0xFF_0F_F0_00; |
1404 | const MASK_WORDS: i32 = 0b11_01_10_00; |
1405 | let expected_result = _mm256_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1406 | let result_masked = |
1407 | _mm256_maskz_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(mask_bytes, vector, matrix); |
1408 | let expected_masked = |
1409 | _mm256_blend_epi32::<MASK_WORDS>(_mm256_setzero_si256(), expected_result); |
1410 | assert_eq_m256i(result_masked, expected_masked); |
1411 | } |
1412 | } |
1413 | |
1414 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1415 | unsafe fn test_mm256_mask_gf2p8affineinv_epi64_epi8() { |
1416 | const CONSTANT_BYTE: i32 = 0x63; |
1417 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1418 | |
1419 | for i in 0..NUM_TEST_WORDS_256 { |
1420 | let left = load_m256i_word(&vectors, i); |
1421 | let right = load_m256i_word(&matrices, i); |
1422 | let result_left = |
1423 | _mm256_mask_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(left, 0, left, right); |
1424 | assert_eq_m256i(result_left, left); |
1425 | let mask_bytes: __mmask32 = 0xFF_0F_F0_00; |
1426 | const MASK_WORDS: i32 = 0b11_01_10_00; |
1427 | let expected_result = _mm256_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(left, right); |
1428 | let result_masked = _mm256_mask_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>( |
1429 | left, mask_bytes, left, right, |
1430 | ); |
1431 | let expected_masked = _mm256_blend_epi32::<MASK_WORDS>(left, expected_result); |
1432 | assert_eq_m256i(result_masked, expected_masked); |
1433 | } |
1434 | } |
1435 | |
1436 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1437 | unsafe fn test_mm_gf2p8affineinv_epi64_epi8() { |
1438 | let identity: i64 = 0x01_02_04_08_10_20_40_80; |
1439 | const IDENTITY_BYTE: i32 = 0; |
1440 | const CONSTANT_BYTE: i32 = 0x63; |
1441 | let identity = _mm_set1_epi64x(identity); |
1442 | |
1443 | // validate inversion |
1444 | let (inputs, results) = generate_inv_tests_data(); |
1445 | |
1446 | for i in 0..NUM_BYTES_WORDS_128 { |
1447 | let input = load_m128i_word(&inputs, i); |
1448 | let reference = load_m128i_word(&results, i); |
1449 | let result = _mm_gf2p8affineinv_epi64_epi8::<IDENTITY_BYTE>(input, identity); |
1450 | let remultiplied = _mm_gf2p8mul_epi8(result, input); |
1451 | assert_eq_m128i(remultiplied, reference); |
1452 | } |
1453 | |
1454 | // validate subsequent affine operation |
1455 | let (matrices, vectors, _affine_expected) = |
1456 | generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1457 | |
1458 | for i in 0..NUM_TEST_WORDS_128 { |
1459 | let vector = load_m128i_word(&vectors, i); |
1460 | let matrix = load_m128i_word(&matrices, i); |
1461 | |
1462 | let inv_vec = _mm_gf2p8affineinv_epi64_epi8::<IDENTITY_BYTE>(vector, identity); |
1463 | let reference = _mm_gf2p8affine_epi64_epi8::<CONSTANT_BYTE>(inv_vec, matrix); |
1464 | let result = _mm_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1465 | assert_eq_m128i(result, reference); |
1466 | } |
1467 | |
1468 | // validate everything by virtue of checking against the AES SBox |
1469 | const AES_S_BOX_MATRIX: i64 = 0xF1_E3_C7_8F_1F_3E_7C_F8; |
1470 | let sbox_matrix = _mm_set1_epi64x(AES_S_BOX_MATRIX); |
1471 | |
1472 | for i in 0..NUM_BYTES_WORDS_128 { |
1473 | let reference = load_m128i_word(&AES_S_BOX, i); |
1474 | let input = load_m128i_word(&inputs, i); |
1475 | let result = _mm_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(input, sbox_matrix); |
1476 | assert_eq_m128i(result, reference); |
1477 | } |
1478 | } |
1479 | |
1480 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1481 | unsafe fn test_mm_maskz_gf2p8affineinv_epi64_epi8() { |
1482 | const CONSTANT_BYTE: i32 = 0x63; |
1483 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1484 | |
1485 | for i in 0..NUM_TEST_WORDS_128 { |
1486 | let matrix = load_m128i_word(&matrices, i); |
1487 | let vector = load_m128i_word(&vectors, i); |
1488 | let result_zero = |
1489 | _mm_maskz_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(0, vector, matrix); |
1490 | assert_eq_m128i(result_zero, _mm_setzero_si128()); |
1491 | let mask_bytes: __mmask16 = 0x0F_F0; |
1492 | const MASK_WORDS: i32 = 0b01_10; |
1493 | let expected_result = _mm_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(vector, matrix); |
1494 | let result_masked = |
1495 | _mm_maskz_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(mask_bytes, vector, matrix); |
1496 | let expected_masked = |
1497 | _mm_blend_epi32::<MASK_WORDS>(_mm_setzero_si128(), expected_result); |
1498 | assert_eq_m128i(result_masked, expected_masked); |
1499 | } |
1500 | } |
1501 | |
1502 | #[simd_test(enable = "gfni,avx512bw,avx512vl" )] |
1503 | unsafe fn test_mm_mask_gf2p8affineinv_epi64_epi8() { |
1504 | const CONSTANT_BYTE: i32 = 0x63; |
1505 | let (matrices, vectors, _expected) = generate_affine_mul_test_data(CONSTANT_BYTE as u8); |
1506 | |
1507 | for i in 0..NUM_TEST_WORDS_128 { |
1508 | let left = load_m128i_word(&vectors, i); |
1509 | let right = load_m128i_word(&matrices, i); |
1510 | let result_left = |
1511 | _mm_mask_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(left, 0, left, right); |
1512 | assert_eq_m128i(result_left, left); |
1513 | let mask_bytes: __mmask16 = 0x0F_F0; |
1514 | const MASK_WORDS: i32 = 0b01_10; |
1515 | let expected_result = _mm_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(left, right); |
1516 | let result_masked = |
1517 | _mm_mask_gf2p8affineinv_epi64_epi8::<CONSTANT_BYTE>(left, mask_bytes, left, right); |
1518 | let expected_masked = _mm_blend_epi32::<MASK_WORDS>(left, expected_result); |
1519 | assert_eq_m128i(result_masked, expected_masked); |
1520 | } |
1521 | } |
1522 | } |
1523 | |