Crypto++  8.2
Free C&
speck128_simd.cpp
1 // speck128_simd.cpp - written and placed in the public domain by Jeffrey Walton
2 //
3 // This source file uses intrinsics and built-ins to gain access to
4 // SSSE3, ARM NEON and ARMv8a, and Power7 Altivec instructions. A separate
5 // source file is needed because additional CXXFLAGS are required to enable
6 // the appropriate instructions sets in some build configurations.
7 
8 #include "pch.h"
9 #include "config.h"
10 
11 #include "speck.h"
12 #include "misc.h"
13 
14 // Uncomment for benchmarking C++ against SSE or NEON.
15 // Do so in both speck.cpp and speck-simd.cpp.
16 // #undef CRYPTOPP_SSSE3_AVAILABLE
17 // #undef CRYPTOPP_ARM_NEON_AVAILABLE
18 
19 #if (CRYPTOPP_SSSE3_AVAILABLE)
20 # include "adv_simd.h"
21 # include <pmmintrin.h>
22 # include <tmmintrin.h>
23 #endif
24 
25 #if defined(__XOP__)
26 # include <ammintrin.h>
27 #endif
28 
29 #if defined(__AVX512F__)
30 # define CRYPTOPP_AVX512_ROTATE 1
31 # include <immintrin.h>
32 #endif
33 
34 // C1189: error: This header is specific to ARM targets
35 #if (CRYPTOPP_ARM_NEON_AVAILABLE)
36 # include "adv_simd.h"
37 # ifndef _M_ARM64
38 # include <arm_neon.h>
39 # endif
40 #endif
41 
42 #if (CRYPTOPP_ARM_ACLE_AVAILABLE)
43 # include <stdint.h>
44 # include <arm_acle.h>
45 #endif
46 
47 #if defined(CRYPTOPP_POWER8_AVAILABLE)
48 # include "adv_simd.h"
49 # include "ppc_simd.h"
50 #endif
51 
52 // Squash MS LNK4221 and libtool warnings
53 extern const char SPECK128_SIMD_FNAME[] = __FILE__;
54 
55 ANONYMOUS_NAMESPACE_BEGIN
56 
57 using CryptoPP::byte;
58 using CryptoPP::word32;
59 using CryptoPP::word64;
60 
61 // *************************** ARM NEON ************************** //
62 
63 #if (CRYPTOPP_ARM_NEON_AVAILABLE)
64 
65 // Missing from Microsoft's ARM A-32 implementation
66 #if defined(_MSC_VER) && !defined(_M_ARM64)
67 inline uint64x2_t vld1q_dup_u64(const uint64_t* ptr)
68 {
69  return vmovq_n_u64(*ptr);
70 }
71 #endif
72 
73 template <class T>
74 inline T UnpackHigh64(const T& a, const T& b)
75 {
76  const uint64x1_t x(vget_high_u64((uint64x2_t)a));
77  const uint64x1_t y(vget_high_u64((uint64x2_t)b));
78  return (T)vcombine_u64(x, y);
79 }
80 
81 template <class T>
82 inline T UnpackLow64(const T& a, const T& b)
83 {
84  const uint64x1_t x(vget_low_u64((uint64x2_t)a));
85  const uint64x1_t y(vget_low_u64((uint64x2_t)b));
86  return (T)vcombine_u64(x, y);
87 }
88 
89 template <unsigned int R>
90 inline uint64x2_t RotateLeft64(const uint64x2_t& val)
91 {
92  const uint64x2_t a(vshlq_n_u64(val, R));
93  const uint64x2_t b(vshrq_n_u64(val, 64 - R));
94  return vorrq_u64(a, b);
95 }
96 
97 template <unsigned int R>
98 inline uint64x2_t RotateRight64(const uint64x2_t& val)
99 {
100  const uint64x2_t a(vshlq_n_u64(val, 64 - R));
101  const uint64x2_t b(vshrq_n_u64(val, R));
102  return vorrq_u64(a, b);
103 }
104 
105 #if defined(__aarch32__) || defined(__aarch64__)
106 // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
107 template <>
108 inline uint64x2_t RotateLeft64<8>(const uint64x2_t& val)
109 {
110  const uint8_t maskb[16] = { 7,0,1,2, 3,4,5,6, 15,8,9,10, 11,12,13,14 };
111  const uint8x16_t mask = vld1q_u8(maskb);
112 
113  return vreinterpretq_u64_u8(
114  vqtbl1q_u8(vreinterpretq_u8_u64(val), mask));
115 }
116 
117 // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
118 template <>
119 inline uint64x2_t RotateRight64<8>(const uint64x2_t& val)
120 {
121  const uint8_t maskb[16] = { 1,2,3,4, 5,6,7,0, 9,10,11,12, 13,14,15,8 };
122  const uint8x16_t mask = vld1q_u8(maskb);
123 
124  return vreinterpretq_u64_u8(
125  vqtbl1q_u8(vreinterpretq_u8_u64(val), mask));
126 }
127 #endif
128 
129 inline void SPECK128_Enc_Block(uint64x2_t &block0, uint64x2_t &block1,
130  const word64 *subkeys, unsigned int rounds)
131 {
132  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
133  uint64x2_t x1 = UnpackHigh64(block0, block1);
134  uint64x2_t y1 = UnpackLow64(block0, block1);
135 
136  for (int i=0; i < static_cast<int>(rounds); ++i)
137  {
138  const uint64x2_t rk = vld1q_dup_u64(subkeys+i);
139 
140  x1 = RotateRight64<8>(x1);
141  x1 = vaddq_u64(x1, y1);
142  x1 = veorq_u64(x1, rk);
143  y1 = RotateLeft64<3>(y1);
144  y1 = veorq_u64(y1, x1);
145  }
146 
147  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
148  block0 = UnpackLow64(y1, x1);
149  block1 = UnpackHigh64(y1, x1);
150 }
151 
152 inline void SPECK128_Enc_6_Blocks(uint64x2_t &block0, uint64x2_t &block1,
153  uint64x2_t &block2, uint64x2_t &block3, uint64x2_t &block4, uint64x2_t &block5,
154  const word64 *subkeys, unsigned int rounds)
155 {
156  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
157  uint64x2_t x1 = UnpackHigh64(block0, block1);
158  uint64x2_t y1 = UnpackLow64(block0, block1);
159  uint64x2_t x2 = UnpackHigh64(block2, block3);
160  uint64x2_t y2 = UnpackLow64(block2, block3);
161  uint64x2_t x3 = UnpackHigh64(block4, block5);
162  uint64x2_t y3 = UnpackLow64(block4, block5);
163 
164  for (int i=0; i < static_cast<int>(rounds); ++i)
165  {
166  const uint64x2_t rk = vld1q_dup_u64(subkeys+i);
167 
168  x1 = RotateRight64<8>(x1);
169  x2 = RotateRight64<8>(x2);
170  x3 = RotateRight64<8>(x3);
171  x1 = vaddq_u64(x1, y1);
172  x2 = vaddq_u64(x2, y2);
173  x3 = vaddq_u64(x3, y3);
174  x1 = veorq_u64(x1, rk);
175  x2 = veorq_u64(x2, rk);
176  x3 = veorq_u64(x3, rk);
177  y1 = RotateLeft64<3>(y1);
178  y2 = RotateLeft64<3>(y2);
179  y3 = RotateLeft64<3>(y3);
180  y1 = veorq_u64(y1, x1);
181  y2 = veorq_u64(y2, x2);
182  y3 = veorq_u64(y3, x3);
183  }
184 
185  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
186  block0 = UnpackLow64(y1, x1);
187  block1 = UnpackHigh64(y1, x1);
188  block2 = UnpackLow64(y2, x2);
189  block3 = UnpackHigh64(y2, x2);
190  block4 = UnpackLow64(y3, x3);
191  block5 = UnpackHigh64(y3, x3);
192 }
193 
194 inline void SPECK128_Dec_Block(uint64x2_t &block0, uint64x2_t &block1,
195  const word64 *subkeys, unsigned int rounds)
196 {
197  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
198  uint64x2_t x1 = UnpackHigh64(block0, block1);
199  uint64x2_t y1 = UnpackLow64(block0, block1);
200 
201  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
202  {
203  const uint64x2_t rk = vld1q_dup_u64(subkeys+i);
204 
205  y1 = veorq_u64(y1, x1);
206  y1 = RotateRight64<3>(y1);
207  x1 = veorq_u64(x1, rk);
208  x1 = vsubq_u64(x1, y1);
209  x1 = RotateLeft64<8>(x1);
210  }
211 
212  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
213  block0 = UnpackLow64(y1, x1);
214  block1 = UnpackHigh64(y1, x1);
215 }
216 
217 inline void SPECK128_Dec_6_Blocks(uint64x2_t &block0, uint64x2_t &block1,
218  uint64x2_t &block2, uint64x2_t &block3, uint64x2_t &block4, uint64x2_t &block5,
219  const word64 *subkeys, unsigned int rounds)
220 {
221  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
222  uint64x2_t x1 = UnpackHigh64(block0, block1);
223  uint64x2_t y1 = UnpackLow64(block0, block1);
224  uint64x2_t x2 = UnpackHigh64(block2, block3);
225  uint64x2_t y2 = UnpackLow64(block2, block3);
226  uint64x2_t x3 = UnpackHigh64(block4, block5);
227  uint64x2_t y3 = UnpackLow64(block4, block5);
228 
229  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
230  {
231  const uint64x2_t rk = vld1q_dup_u64(subkeys+i);
232 
233  y1 = veorq_u64(y1, x1);
234  y2 = veorq_u64(y2, x2);
235  y3 = veorq_u64(y3, x3);
236  y1 = RotateRight64<3>(y1);
237  y2 = RotateRight64<3>(y2);
238  y3 = RotateRight64<3>(y3);
239  x1 = veorq_u64(x1, rk);
240  x2 = veorq_u64(x2, rk);
241  x3 = veorq_u64(x3, rk);
242  x1 = vsubq_u64(x1, y1);
243  x2 = vsubq_u64(x2, y2);
244  x3 = vsubq_u64(x3, y3);
245  x1 = RotateLeft64<8>(x1);
246  x2 = RotateLeft64<8>(x2);
247  x3 = RotateLeft64<8>(x3);
248  }
249 
250  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
251  block0 = UnpackLow64(y1, x1);
252  block1 = UnpackHigh64(y1, x1);
253  block2 = UnpackLow64(y2, x2);
254  block3 = UnpackHigh64(y2, x2);
255  block4 = UnpackLow64(y3, x3);
256  block5 = UnpackHigh64(y3, x3);
257 }
258 
259 #endif // CRYPTOPP_ARM_NEON_AVAILABLE
260 
261 // ***************************** IA-32 ***************************** //
262 
263 #if defined(CRYPTOPP_SSSE3_AVAILABLE)
264 
265 // Clang __m128i casts, http://bugs.llvm.org/show_bug.cgi?id=20670
266 #ifndef M128_CAST
267 # define M128_CAST(x) ((__m128i *)(void *)(x))
268 #endif
269 #ifndef CONST_M128_CAST
270 # define CONST_M128_CAST(x) ((const __m128i *)(const void *)(x))
271 #endif
272 
273 // GCC double casts, https://www.spinics.net/lists/gcchelp/msg47735.html
274 #ifndef DOUBLE_CAST
275 # define DOUBLE_CAST(x) ((double *)(void *)(x))
276 #endif
277 #ifndef CONST_DOUBLE_CAST
278 # define CONST_DOUBLE_CAST(x) ((const double *)(const void *)(x))
279 #endif
280 
281 template <unsigned int R>
282 inline __m128i RotateLeft64(const __m128i& val)
283 {
284 #if defined(CRYPTOPP_AVX512_ROTATE)
285  return _mm_rol_epi64(val, R);
286 #elif defined(__XOP__)
287  return _mm_roti_epi64(val, R);
288 #else
289  return _mm_or_si128(
290  _mm_slli_epi64(val, R), _mm_srli_epi64(val, 64-R));
291 #endif
292 }
293 
294 template <unsigned int R>
295 inline __m128i RotateRight64(const __m128i& val)
296 {
297 #if defined(CRYPTOPP_AVX512_ROTATE)
298  return _mm_ror_epi64(val, R);
299 #elif defined(__XOP__)
300  return _mm_roti_epi64(val, 64-R);
301 #else
302  return _mm_or_si128(
303  _mm_slli_epi64(val, 64-R), _mm_srli_epi64(val, R));
304 #endif
305 }
306 
307 // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
308 template <>
309 __m128i RotateLeft64<8>(const __m128i& val)
310 {
311 #if defined(__XOP__)
312  return _mm_roti_epi64(val, 8);
313 #else
314  const __m128i mask = _mm_set_epi8(14,13,12,11, 10,9,8,15, 6,5,4,3, 2,1,0,7);
315  return _mm_shuffle_epi8(val, mask);
316 #endif
317 }
318 
319 // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
320 template <>
321 __m128i RotateRight64<8>(const __m128i& val)
322 {
323 #if defined(__XOP__)
324  return _mm_roti_epi64(val, 64-8);
325 #else
326  const __m128i mask = _mm_set_epi8(8,15,14,13, 12,11,10,9, 0,7,6,5, 4,3,2,1);
327  return _mm_shuffle_epi8(val, mask);
328 #endif
329 }
330 
331 inline void SPECK128_Enc_Block(__m128i &block0, __m128i &block1,
332  const word64 *subkeys, unsigned int rounds)
333 {
334  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
335  __m128i x1 = _mm_unpackhi_epi64(block0, block1);
336  __m128i y1 = _mm_unpacklo_epi64(block0, block1);
337 
338  for (int i=0; i < static_cast<int>(rounds); ++i)
339  {
340  const __m128i rk = _mm_castpd_si128(
341  _mm_loaddup_pd(CONST_DOUBLE_CAST(subkeys+i)));
342 
343  x1 = RotateRight64<8>(x1);
344  x1 = _mm_add_epi64(x1, y1);
345  x1 = _mm_xor_si128(x1, rk);
346  y1 = RotateLeft64<3>(y1);
347  y1 = _mm_xor_si128(y1, x1);
348  }
349 
350  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
351  block0 = _mm_unpacklo_epi64(y1, x1);
352  block1 = _mm_unpackhi_epi64(y1, x1);
353 }
354 
355 inline void SPECK128_Enc_6_Blocks(__m128i &block0, __m128i &block1,
356  __m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5,
357  const word64 *subkeys, unsigned int rounds)
358 {
359  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
360  __m128i x1 = _mm_unpackhi_epi64(block0, block1);
361  __m128i y1 = _mm_unpacklo_epi64(block0, block1);
362  __m128i x2 = _mm_unpackhi_epi64(block2, block3);
363  __m128i y2 = _mm_unpacklo_epi64(block2, block3);
364  __m128i x3 = _mm_unpackhi_epi64(block4, block5);
365  __m128i y3 = _mm_unpacklo_epi64(block4, block5);
366 
367  for (int i=0; i < static_cast<int>(rounds); ++i)
368  {
369  const __m128i rk = _mm_castpd_si128(
370  _mm_loaddup_pd(CONST_DOUBLE_CAST(subkeys+i)));
371 
372  x1 = RotateRight64<8>(x1);
373  x2 = RotateRight64<8>(x2);
374  x3 = RotateRight64<8>(x3);
375  x1 = _mm_add_epi64(x1, y1);
376  x2 = _mm_add_epi64(x2, y2);
377  x3 = _mm_add_epi64(x3, y3);
378  x1 = _mm_xor_si128(x1, rk);
379  x2 = _mm_xor_si128(x2, rk);
380  x3 = _mm_xor_si128(x3, rk);
381  y1 = RotateLeft64<3>(y1);
382  y2 = RotateLeft64<3>(y2);
383  y3 = RotateLeft64<3>(y3);
384  y1 = _mm_xor_si128(y1, x1);
385  y2 = _mm_xor_si128(y2, x2);
386  y3 = _mm_xor_si128(y3, x3);
387  }
388 
389  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
390  block0 = _mm_unpacklo_epi64(y1, x1);
391  block1 = _mm_unpackhi_epi64(y1, x1);
392  block2 = _mm_unpacklo_epi64(y2, x2);
393  block3 = _mm_unpackhi_epi64(y2, x2);
394  block4 = _mm_unpacklo_epi64(y3, x3);
395  block5 = _mm_unpackhi_epi64(y3, x3);
396 }
397 
398 inline void SPECK128_Dec_Block(__m128i &block0, __m128i &block1,
399  const word64 *subkeys, unsigned int rounds)
400 {
401  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
402  __m128i x1 = _mm_unpackhi_epi64(block0, block1);
403  __m128i y1 = _mm_unpacklo_epi64(block0, block1);
404 
405  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
406  {
407  const __m128i rk = _mm_castpd_si128(
408  _mm_loaddup_pd(CONST_DOUBLE_CAST(subkeys+i)));
409 
410  y1 = _mm_xor_si128(y1, x1);
411  y1 = RotateRight64<3>(y1);
412  x1 = _mm_xor_si128(x1, rk);
413  x1 = _mm_sub_epi64(x1, y1);
414  x1 = RotateLeft64<8>(x1);
415  }
416 
417  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
418  block0 = _mm_unpacklo_epi64(y1, x1);
419  block1 = _mm_unpackhi_epi64(y1, x1);
420 }
421 
422 inline void SPECK128_Dec_6_Blocks(__m128i &block0, __m128i &block1,
423  __m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5,
424  const word64 *subkeys, unsigned int rounds)
425 {
426  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
427  __m128i x1 = _mm_unpackhi_epi64(block0, block1);
428  __m128i y1 = _mm_unpacklo_epi64(block0, block1);
429  __m128i x2 = _mm_unpackhi_epi64(block2, block3);
430  __m128i y2 = _mm_unpacklo_epi64(block2, block3);
431  __m128i x3 = _mm_unpackhi_epi64(block4, block5);
432  __m128i y3 = _mm_unpacklo_epi64(block4, block5);
433 
434  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
435  {
436  const __m128i rk = _mm_castpd_si128(
437  _mm_loaddup_pd(CONST_DOUBLE_CAST(subkeys+i)));
438 
439  y1 = _mm_xor_si128(y1, x1);
440  y2 = _mm_xor_si128(y2, x2);
441  y3 = _mm_xor_si128(y3, x3);
442  y1 = RotateRight64<3>(y1);
443  y2 = RotateRight64<3>(y2);
444  y3 = RotateRight64<3>(y3);
445  x1 = _mm_xor_si128(x1, rk);
446  x2 = _mm_xor_si128(x2, rk);
447  x3 = _mm_xor_si128(x3, rk);
448  x1 = _mm_sub_epi64(x1, y1);
449  x2 = _mm_sub_epi64(x2, y2);
450  x3 = _mm_sub_epi64(x3, y3);
451  x1 = RotateLeft64<8>(x1);
452  x2 = RotateLeft64<8>(x2);
453  x3 = RotateLeft64<8>(x3);
454  }
455 
456  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
457  block0 = _mm_unpacklo_epi64(y1, x1);
458  block1 = _mm_unpackhi_epi64(y1, x1);
459  block2 = _mm_unpacklo_epi64(y2, x2);
460  block3 = _mm_unpackhi_epi64(y2, x2);
461  block4 = _mm_unpacklo_epi64(y3, x3);
462  block5 = _mm_unpackhi_epi64(y3, x3);
463 }
464 
465 #endif // CRYPTOPP_SSSE3_AVAILABLE
466 
467 // ***************************** Power8 ***************************** //
468 
469 #if defined(CRYPTOPP_POWER8_AVAILABLE)
470 
474 
475 using CryptoPP::VecAdd;
476 using CryptoPP::VecSub;
477 using CryptoPP::VecXor;
479 
480 // Rotate left by bit count
481 template<unsigned int C>
482 inline uint64x2_p RotateLeft64(const uint64x2_p val)
483 {
484  const uint64x2_p m = {C, C};
485  return vec_rl(val, m);
486 }
487 
488 // Rotate right by bit count
489 template<unsigned int C>
490 inline uint64x2_p RotateRight64(const uint64x2_p val)
491 {
492  const uint64x2_p m = {64-C, 64-C};
493  return vec_rl(val, m);
494 }
495 
496 void SPECK128_Enc_Block(uint32x4_p &block, const word64 *subkeys, unsigned int rounds)
497 {
498 #if (CRYPTOPP_BIG_ENDIAN)
499  const uint8x16_p m1 = {31,30,29,28,27,26,25,24, 15,14,13,12,11,10,9,8};
500  const uint8x16_p m2 = {23,22,21,20,19,18,17,16, 7,6,5,4,3,2,1,0};
501 #else
502  const uint8x16_p m1 = {7,6,5,4,3,2,1,0, 23,22,21,20,19,18,17,16};
503  const uint8x16_p m2 = {15,14,13,12,11,10,9,8, 31,30,29,28,27,26,25,24};
504 #endif
505 
506  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
507  uint64x2_p x1 = (uint64x2_p)VecPermute(block, block, m1);
508  uint64x2_p y1 = (uint64x2_p)VecPermute(block, block, m2);
509 
510  for (int i=0; i < static_cast<int>(rounds); ++i)
511  {
512  const uint64x2_p rk = vec_splats((unsigned long long)subkeys[i]);
513 
514  x1 = RotateRight64<8>(x1);
515  x1 = VecAdd(x1, y1);
516  x1 = VecXor(x1, rk);
517 
518  y1 = RotateLeft64<3>(y1);
519  y1 = VecXor(y1, x1);
520  }
521 
522 #if (CRYPTOPP_BIG_ENDIAN)
523  const uint8x16_p m3 = {31,30,29,28,27,26,25,24, 15,14,13,12,11,10,9,8};
524  //const uint8x16_p m4 = {23,22,21,20,19,18,17,16, 7,6,5,4,3,2,1,0};
525 #else
526  const uint8x16_p m3 = {7,6,5,4,3,2,1,0, 23,22,21,20,19,18,17,16};
527  //const uint8x16_p m4 = {15,14,13,12,11,10,9,8, 31,30,29,28,27,26,25,24};
528 #endif
529 
530  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
531  block = (uint32x4_p)VecPermute(x1, y1, m3);
532 }
533 
534 void SPECK128_Dec_Block(uint32x4_p &block, const word64 *subkeys, unsigned int rounds)
535 {
536 #if (CRYPTOPP_BIG_ENDIAN)
537  const uint8x16_p m1 = {31,30,29,28,27,26,25,24, 15,14,13,12,11,10,9,8};
538  const uint8x16_p m2 = {23,22,21,20,19,18,17,16, 7,6,5,4,3,2,1,0};
539 #else
540  const uint8x16_p m1 = {7,6,5,4,3,2,1,0, 23,22,21,20,19,18,17,16};
541  const uint8x16_p m2 = {15,14,13,12,11,10,9,8, 31,30,29,28,27,26,25,24};
542 #endif
543 
544  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
545  uint64x2_p x1 = (uint64x2_p)VecPermute(block, block, m1);
546  uint64x2_p y1 = (uint64x2_p)VecPermute(block, block, m2);
547 
548  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
549  {
550  const uint64x2_p rk = vec_splats((unsigned long long)subkeys[i]);
551 
552  y1 = VecXor(y1, x1);
553  y1 = RotateRight64<3>(y1);
554  x1 = VecXor(x1, rk);
555  x1 = VecSub(x1, y1);
556  x1 = RotateLeft64<8>(x1);
557  }
558 
559 #if (CRYPTOPP_BIG_ENDIAN)
560  const uint8x16_p m3 = {31,30,29,28,27,26,25,24, 15,14,13,12,11,10,9,8};
561  //const uint8x16_p m4 = {23,22,21,20,19,18,17,16, 7,6,5,4,3,2,1,0};
562 #else
563  const uint8x16_p m3 = {7,6,5,4,3,2,1,0, 23,22,21,20,19,18,17,16};
564  //const uint8x16_p m4 = {15,14,13,12,11,10,9,8, 31,30,29,28,27,26,25,24};
565 #endif
566 
567  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
568  block = (uint32x4_p)VecPermute(x1, y1, m3);
569 }
570 
571 void SPECK128_Enc_6_Blocks(uint32x4_p &block0, uint32x4_p &block1,
572  uint32x4_p &block2, uint32x4_p &block3, uint32x4_p &block4,
573  uint32x4_p &block5, const word64 *subkeys, unsigned int rounds)
574 {
575 #if (CRYPTOPP_BIG_ENDIAN)
576  const uint8x16_p m1 = {31,30,29,28,27,26,25,24, 15,14,13,12,11,10,9,8};
577  const uint8x16_p m2 = {23,22,21,20,19,18,17,16, 7,6,5,4,3,2,1,0};
578 #else
579  const uint8x16_p m1 = {7,6,5,4,3,2,1,0, 23,22,21,20,19,18,17,16};
580  const uint8x16_p m2 = {15,14,13,12,11,10,9,8, 31,30,29,28,27,26,25,24};
581 #endif
582 
583  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
584  uint64x2_p x1 = (uint64x2_p)VecPermute(block0, block1, m1);
585  uint64x2_p y1 = (uint64x2_p)VecPermute(block0, block1, m2);
586  uint64x2_p x2 = (uint64x2_p)VecPermute(block2, block3, m1);
587  uint64x2_p y2 = (uint64x2_p)VecPermute(block2, block3, m2);
588  uint64x2_p x3 = (uint64x2_p)VecPermute(block4, block5, m1);
589  uint64x2_p y3 = (uint64x2_p)VecPermute(block4, block5, m2);
590 
591  for (int i=0; i < static_cast<int>(rounds); ++i)
592  {
593  const uint64x2_p rk = vec_splats((unsigned long long)subkeys[i]);
594 
595  x1 = RotateRight64<8>(x1);
596  x2 = RotateRight64<8>(x2);
597  x3 = RotateRight64<8>(x3);
598  x1 = VecAdd(x1, y1);
599  x2 = VecAdd(x2, y2);
600  x3 = VecAdd(x3, y3);
601  x1 = VecXor(x1, rk);
602  x2 = VecXor(x2, rk);
603  x3 = VecXor(x3, rk);
604 
605  y1 = RotateLeft64<3>(y1);
606  y2 = RotateLeft64<3>(y2);
607  y3 = RotateLeft64<3>(y3);
608  y1 = VecXor(y1, x1);
609  y2 = VecXor(y2, x2);
610  y3 = VecXor(y3, x3);
611  }
612 
613 #if (CRYPTOPP_BIG_ENDIAN)
614  const uint8x16_p m3 = {31,30,29,28,27,26,25,24, 15,14,13,12,11,10,9,8};
615  const uint8x16_p m4 = {23,22,21,20,19,18,17,16, 7,6,5,4,3,2,1,0};
616 #else
617  const uint8x16_p m3 = {7,6,5,4,3,2,1,0, 23,22,21,20,19,18,17,16};
618  const uint8x16_p m4 = {15,14,13,12,11,10,9,8, 31,30,29,28,27,26,25,24};
619 #endif
620 
621  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
622  block0 = (uint32x4_p)VecPermute(x1, y1, m3);
623  block1 = (uint32x4_p)VecPermute(x1, y1, m4);
624  block2 = (uint32x4_p)VecPermute(x2, y2, m3);
625  block3 = (uint32x4_p)VecPermute(x2, y2, m4);
626  block4 = (uint32x4_p)VecPermute(x3, y3, m3);
627  block5 = (uint32x4_p)VecPermute(x3, y3, m4);
628 }
629 
630 void SPECK128_Dec_6_Blocks(uint32x4_p &block0, uint32x4_p &block1,
631  uint32x4_p &block2, uint32x4_p &block3, uint32x4_p &block4,
632  uint32x4_p &block5, const word64 *subkeys, unsigned int rounds)
633 {
634 #if (CRYPTOPP_BIG_ENDIAN)
635  const uint8x16_p m1 = {31,30,29,28,27,26,25,24, 15,14,13,12,11,10,9,8};
636  const uint8x16_p m2 = {23,22,21,20,19,18,17,16, 7,6,5,4,3,2,1,0};
637 #else
638  const uint8x16_p m1 = {7,6,5,4,3,2,1,0, 23,22,21,20,19,18,17,16};
639  const uint8x16_p m2 = {15,14,13,12,11,10,9,8, 31,30,29,28,27,26,25,24};
640 #endif
641 
642  // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
643  uint64x2_p x1 = (uint64x2_p)VecPermute(block0, block1, m1);
644  uint64x2_p y1 = (uint64x2_p)VecPermute(block0, block1, m2);
645  uint64x2_p x2 = (uint64x2_p)VecPermute(block2, block3, m1);
646  uint64x2_p y2 = (uint64x2_p)VecPermute(block2, block3, m2);
647  uint64x2_p x3 = (uint64x2_p)VecPermute(block4, block5, m1);
648  uint64x2_p y3 = (uint64x2_p)VecPermute(block4, block5, m2);
649 
650  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
651  {
652  const uint64x2_p rk = vec_splats((unsigned long long)subkeys[i]);
653 
654  y1 = VecXor(y1, x1);
655  y2 = VecXor(y2, x2);
656  y3 = VecXor(y3, x3);
657  y1 = RotateRight64<3>(y1);
658  y2 = RotateRight64<3>(y2);
659  y3 = RotateRight64<3>(y3);
660 
661  x1 = VecXor(x1, rk);
662  x2 = VecXor(x2, rk);
663  x3 = VecXor(x3, rk);
664  x1 = VecSub(x1, y1);
665  x2 = VecSub(x2, y2);
666  x3 = VecSub(x3, y3);
667  x1 = RotateLeft64<8>(x1);
668  x2 = RotateLeft64<8>(x2);
669  x3 = RotateLeft64<8>(x3);
670  }
671 
672 #if (CRYPTOPP_BIG_ENDIAN)
673  const uint8x16_p m3 = {31,30,29,28,27,26,25,24, 15,14,13,12,11,10,9,8};
674  const uint8x16_p m4 = {23,22,21,20,19,18,17,16, 7,6,5,4,3,2,1,0};
675 #else
676  const uint8x16_p m3 = {7,6,5,4,3,2,1,0, 23,22,21,20,19,18,17,16};
677  const uint8x16_p m4 = {15,14,13,12,11,10,9,8, 31,30,29,28,27,26,25,24};
678 #endif
679 
680  // [A1 B1][A2 B2] ... => [A1 A2][B1 B2] ...
681  block0 = (uint32x4_p)VecPermute(x1, y1, m3);
682  block1 = (uint32x4_p)VecPermute(x1, y1, m4);
683  block2 = (uint32x4_p)VecPermute(x2, y2, m3);
684  block3 = (uint32x4_p)VecPermute(x2, y2, m4);
685  block4 = (uint32x4_p)VecPermute(x3, y3, m3);
686  block5 = (uint32x4_p)VecPermute(x3, y3, m4);
687 }
688 
689 #endif // CRYPTOPP_POWER8_AVAILABLE
690 
691 ANONYMOUS_NAMESPACE_END
692 
693 ///////////////////////////////////////////////////////////////////////
694 
695 NAMESPACE_BEGIN(CryptoPP)
696 
697 // *************************** ARM NEON **************************** //
698 
699 #if (CRYPTOPP_ARM_NEON_AVAILABLE)
700 size_t SPECK128_Enc_AdvancedProcessBlocks_NEON(const word64* subKeys, size_t rounds,
701  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
702 {
703  return AdvancedProcessBlocks128_6x2_NEON(SPECK128_Enc_Block, SPECK128_Enc_6_Blocks,
704  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
705 }
706 
707 size_t SPECK128_Dec_AdvancedProcessBlocks_NEON(const word64* subKeys, size_t rounds,
708  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
709 {
710  return AdvancedProcessBlocks128_6x2_NEON(SPECK128_Dec_Block, SPECK128_Dec_6_Blocks,
711  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
712 }
713 #endif // CRYPTOPP_ARM_NEON_AVAILABLE
714 
715 // ***************************** IA-32 ***************************** //
716 
717 #if defined(CRYPTOPP_SSSE3_AVAILABLE)
718 size_t SPECK128_Enc_AdvancedProcessBlocks_SSSE3(const word64* subKeys, size_t rounds,
719  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
720 {
721  return AdvancedProcessBlocks128_6x2_SSE(SPECK128_Enc_Block, SPECK128_Enc_6_Blocks,
722  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
723 }
724 
725 size_t SPECK128_Dec_AdvancedProcessBlocks_SSSE3(const word64* subKeys, size_t rounds,
726  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
727 {
728  return AdvancedProcessBlocks128_6x2_SSE(SPECK128_Dec_Block, SPECK128_Dec_6_Blocks,
729  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
730 }
731 #endif // CRYPTOPP_SSSE3_AVAILABLE
732 
733 // ***************************** Power8 ***************************** //
734 
735 #if defined(CRYPTOPP_POWER8_AVAILABLE)
736 size_t SPECK128_Enc_AdvancedProcessBlocks_POWER8(const word64* subKeys, size_t rounds,
737  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
738 {
739  return AdvancedProcessBlocks128_6x1_ALTIVEC(SPECK128_Enc_Block, SPECK128_Enc_6_Blocks,
740  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
741 }
742 
743 size_t SPECK128_Dec_AdvancedProcessBlocks_POWER8(const word64* subKeys, size_t rounds,
744  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
745 {
746  return AdvancedProcessBlocks128_6x1_ALTIVEC(SPECK128_Dec_Block, SPECK128_Dec_6_Blocks,
747  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
748 }
749 #endif // CRYPTOPP_POWER8_AVAILABLE
750 
751 NAMESPACE_END
Template for AdvancedProcessBlocks and SIMD processing.
Library configuration file.
Utility functions for the Crypto++ library.
Crypto++ library namespace.
Precompiled header file.
Support functions for PowerPC and vector operations.
__vector unsigned int uint32x4_p
Vector of 32-bit elements.
Definition: ppc_simd.h:129
T1 VecPermute(const T1 vec, const T2 mask)
Permutes a vector.
Definition: ppc_simd.h:1010
__vector unsigned char uint8x16_p
Vector of 8-bit elements.
Definition: ppc_simd.h:119
T1 VecXor(const T1 vec1, const T2 vec2)
XOR two vectors.
Definition: ppc_simd.h:916
__vector unsigned long long uint64x2_p
Vector of 64-bit elements.
Definition: ppc_simd.h:139
T1 VecSub(const T1 vec1, const T2 vec2)
Subtract two vectors.
Definition: ppc_simd.h:956
T1 VecAdd(const T1 vec1, const T2 vec2)
Add two vectors.
Definition: ppc_simd.h:939
Classes for the Speck block cipher.