Crypto++  8.2
Free C++ class library of cryptographic schemes
speck64_simd.cpp
1 // speck64_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 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_SSE41_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 (CRYPTOPP_SSE41_AVAILABLE)
26 # include <smmintrin.h>
27 #endif
28 
29 #if defined(__XOP__)
30 # include <ammintrin.h>
31 #endif
32 
33 #if defined(__AVX512F__)
34 # define CRYPTOPP_AVX512_ROTATE 1
35 # include <immintrin.h>
36 #endif
37 
38 // C1189: error: This header is specific to ARM targets
39 #if (CRYPTOPP_ARM_NEON_AVAILABLE)
40 # include "adv_simd.h"
41 # ifndef _M_ARM64
42 # include <arm_neon.h>
43 # endif
44 #endif
45 
46 #if (CRYPTOPP_ARM_ACLE_AVAILABLE)
47 # include <stdint.h>
48 # include <arm_acle.h>
49 #endif
50 
51 #if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
52 # include "adv_simd.h"
53 # include "ppc_simd.h"
54 #endif
55 
56 // Squash MS LNK4221 and libtool warnings
57 extern const char SPECK64_SIMD_FNAME[] = __FILE__;
58 
59 ANONYMOUS_NAMESPACE_BEGIN
60 
61 using CryptoPP::byte;
62 using CryptoPP::word32;
63 using CryptoPP::word64;
64 
65 // *************************** ARM NEON ************************** //
66 
67 #if (CRYPTOPP_ARM_NEON_AVAILABLE)
68 
69 template <class T>
70 inline T UnpackHigh32(const T& a, const T& b)
71 {
72  const uint32x2_t x(vget_high_u32((uint32x4_t)a));
73  const uint32x2_t y(vget_high_u32((uint32x4_t)b));
74  const uint32x2x2_t r = vzip_u32(x, y);
75  return (T)vcombine_u32(r.val[0], r.val[1]);
76 }
77 
78 template <class T>
79 inline T UnpackLow32(const T& a, const T& b)
80 {
81  const uint32x2_t x(vget_low_u32((uint32x4_t)a));
82  const uint32x2_t y(vget_low_u32((uint32x4_t)b));
83  const uint32x2x2_t r = vzip_u32(x, y);
84  return (T)vcombine_u32(r.val[0], r.val[1]);
85 }
86 
87 template <unsigned int R>
88 inline uint32x4_t RotateLeft32(const uint32x4_t& val)
89 {
90  const uint32x4_t a(vshlq_n_u32(val, R));
91  const uint32x4_t b(vshrq_n_u32(val, 32 - R));
92  return vorrq_u32(a, b);
93 }
94 
95 template <unsigned int R>
96 inline uint32x4_t RotateRight32(const uint32x4_t& val)
97 {
98  const uint32x4_t a(vshlq_n_u32(val, 32 - R));
99  const uint32x4_t b(vshrq_n_u32(val, R));
100  return vorrq_u32(a, b);
101 }
102 
103 #if defined(__aarch32__) || defined(__aarch64__)
104 // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
105 template <>
106 inline uint32x4_t RotateLeft32<8>(const uint32x4_t& val)
107 {
108  const uint8_t maskb[16] = { 3,0,1,2, 7,4,5,6, 11,8,9,10, 15,12,13,14 };
109  const uint8x16_t mask = vld1q_u8(maskb);
110 
111  return vreinterpretq_u32_u8(
112  vqtbl1q_u8(vreinterpretq_u8_u32(val), mask));
113 }
114 
115 // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
116 template <>
117 inline uint32x4_t RotateRight32<8>(const uint32x4_t& val)
118 {
119  const uint8_t maskb[16] = { 1,2,3,0, 5,6,7,4, 9,10,11,8, 13,14,15,12 };
120  const uint8x16_t mask = vld1q_u8(maskb);
121 
122  return vreinterpretq_u32_u8(
123  vqtbl1q_u8(vreinterpretq_u8_u32(val), mask));
124 }
125 #endif // Aarch32 or Aarch64
126 
127 inline void SPECK64_Enc_Block(uint32x4_t &block0, uint32x4_t &block1,
128  const word32 *subkeys, unsigned int rounds)
129 {
130  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
131  uint32x4_t x1 = vuzpq_u32(block0, block1).val[1];
132  uint32x4_t y1 = vuzpq_u32(block0, block1).val[0];
133 
134  for (int i=0; i < static_cast<int>(rounds); ++i)
135  {
136  const uint32x4_t rk = vdupq_n_u32(subkeys[i]);
137 
138  x1 = RotateRight32<8>(x1);
139  x1 = vaddq_u32(x1, y1);
140  x1 = veorq_u32(x1, rk);
141  y1 = RotateLeft32<3>(y1);
142  y1 = veorq_u32(y1, x1);
143  }
144 
145  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
146  block0 = UnpackLow32(y1, x1);
147  block1 = UnpackHigh32(y1, x1);
148 }
149 
150 inline void SPECK64_Dec_Block(uint32x4_t &block0, uint32x4_t &block1,
151  const word32 *subkeys, unsigned int rounds)
152 {
153  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
154  uint32x4_t x1 = vuzpq_u32(block0, block1).val[1];
155  uint32x4_t y1 = vuzpq_u32(block0, block1).val[0];
156 
157  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
158  {
159  const uint32x4_t rk = vdupq_n_u32(subkeys[i]);
160 
161  y1 = veorq_u32(y1, x1);
162  y1 = RotateRight32<3>(y1);
163  x1 = veorq_u32(x1, rk);
164  x1 = vsubq_u32(x1, y1);
165  x1 = RotateLeft32<8>(x1);
166  }
167 
168  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
169  block0 = UnpackLow32(y1, x1);
170  block1 = UnpackHigh32(y1, x1);
171 }
172 
173 inline void SPECK64_Enc_6_Blocks(uint32x4_t &block0, uint32x4_t &block1,
174  uint32x4_t &block2, uint32x4_t &block3, uint32x4_t &block4, uint32x4_t &block5,
175  const word32 *subkeys, unsigned int rounds)
176 {
177  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
178  uint32x4_t x1 = vuzpq_u32(block0, block1).val[1];
179  uint32x4_t y1 = vuzpq_u32(block0, block1).val[0];
180  uint32x4_t x2 = vuzpq_u32(block2, block3).val[1];
181  uint32x4_t y2 = vuzpq_u32(block2, block3).val[0];
182  uint32x4_t x3 = vuzpq_u32(block4, block5).val[1];
183  uint32x4_t y3 = vuzpq_u32(block4, block5).val[0];
184 
185  for (int i=0; i < static_cast<int>(rounds); ++i)
186  {
187  const uint32x4_t rk = vdupq_n_u32(subkeys[i]);
188 
189  x1 = RotateRight32<8>(x1);
190  x2 = RotateRight32<8>(x2);
191  x3 = RotateRight32<8>(x3);
192  x1 = vaddq_u32(x1, y1);
193  x2 = vaddq_u32(x2, y2);
194  x3 = vaddq_u32(x3, y3);
195  x1 = veorq_u32(x1, rk);
196  x2 = veorq_u32(x2, rk);
197  x3 = veorq_u32(x3, rk);
198  y1 = RotateLeft32<3>(y1);
199  y2 = RotateLeft32<3>(y2);
200  y3 = RotateLeft32<3>(y3);
201  y1 = veorq_u32(y1, x1);
202  y2 = veorq_u32(y2, x2);
203  y3 = veorq_u32(y3, x3);
204  }
205 
206  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
207  block0 = UnpackLow32(y1, x1);
208  block1 = UnpackHigh32(y1, x1);
209  block2 = UnpackLow32(y2, x2);
210  block3 = UnpackHigh32(y2, x2);
211  block4 = UnpackLow32(y3, x3);
212  block5 = UnpackHigh32(y3, x3);
213 }
214 
215 inline void SPECK64_Dec_6_Blocks(uint32x4_t &block0, uint32x4_t &block1,
216  uint32x4_t &block2, uint32x4_t &block3, uint32x4_t &block4, uint32x4_t &block5,
217  const word32 *subkeys, unsigned int rounds)
218 {
219  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
220  uint32x4_t x1 = vuzpq_u32(block0, block1).val[1];
221  uint32x4_t y1 = vuzpq_u32(block0, block1).val[0];
222  uint32x4_t x2 = vuzpq_u32(block2, block3).val[1];
223  uint32x4_t y2 = vuzpq_u32(block2, block3).val[0];
224  uint32x4_t x3 = vuzpq_u32(block4, block5).val[1];
225  uint32x4_t y3 = vuzpq_u32(block4, block5).val[0];
226 
227  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
228  {
229  const uint32x4_t rk = vdupq_n_u32(subkeys[i]);
230 
231  y1 = veorq_u32(y1, x1);
232  y2 = veorq_u32(y2, x2);
233  y3 = veorq_u32(y3, x3);
234  y1 = RotateRight32<3>(y1);
235  y2 = RotateRight32<3>(y2);
236  y3 = RotateRight32<3>(y3);
237  x1 = veorq_u32(x1, rk);
238  x2 = veorq_u32(x2, rk);
239  x3 = veorq_u32(x3, rk);
240  x1 = vsubq_u32(x1, y1);
241  x2 = vsubq_u32(x2, y2);
242  x3 = vsubq_u32(x3, y3);
243  x1 = RotateLeft32<8>(x1);
244  x2 = RotateLeft32<8>(x2);
245  x3 = RotateLeft32<8>(x3);
246  }
247 
248  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
249  block0 = UnpackLow32(y1, x1);
250  block1 = UnpackHigh32(y1, x1);
251  block2 = UnpackLow32(y2, x2);
252  block3 = UnpackHigh32(y2, x2);
253  block4 = UnpackLow32(y3, x3);
254  block5 = UnpackHigh32(y3, x3);
255 }
256 
257 #endif // CRYPTOPP_ARM_NEON_AVAILABLE
258 
259 // ***************************** IA-32 ***************************** //
260 
261 #if defined(CRYPTOPP_SSE41_AVAILABLE)
262 
263 template <unsigned int R>
264 inline __m128i RotateLeft32(const __m128i& val)
265 {
266 #if defined(__XOP__)
267  return _mm_roti_epi32(val, R);
268 #else
269  return _mm_or_si128(
270  _mm_slli_epi32(val, R), _mm_srli_epi32(val, 32-R));
271 #endif
272 }
273 
274 template <unsigned int R>
275 inline __m128i RotateRight32(const __m128i& val)
276 {
277 #if defined(__XOP__)
278  return _mm_roti_epi32(val, 32-R);
279 #else
280  return _mm_or_si128(
281  _mm_slli_epi32(val, 32-R), _mm_srli_epi32(val, R));
282 #endif
283 }
284 
285 // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
286 template <>
287 __m128i RotateLeft32<8>(const __m128i& val)
288 {
289 #if defined(__XOP__)
290  return _mm_roti_epi32(val, 8);
291 #else
292  const __m128i mask = _mm_set_epi8(14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3);
293  return _mm_shuffle_epi8(val, mask);
294 #endif
295 }
296 
297 // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
298 template <>
299 __m128i RotateRight32<8>(const __m128i& val)
300 {
301 #if defined(__XOP__)
302  return _mm_roti_epi32(val, 32-8);
303 #else
304  const __m128i mask = _mm_set_epi8(12,15,14,13, 8,11,10,9, 4,7,6,5, 0,3,2,1);
305  return _mm_shuffle_epi8(val, mask);
306 #endif
307 }
308 
309 inline void SPECK64_Enc_Block(__m128i &block0, __m128i &block1,
310  const word32 *subkeys, unsigned int rounds)
311 {
312  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
313  const __m128 t0 = _mm_castsi128_ps(block0);
314  const __m128 t1 = _mm_castsi128_ps(block1);
315  __m128i x1 = _mm_castps_si128(_mm_shuffle_ps(t0, t1, _MM_SHUFFLE(3,1,3,1)));
316  __m128i y1 = _mm_castps_si128(_mm_shuffle_ps(t0, t1, _MM_SHUFFLE(2,0,2,0)));
317 
318  for (int i=0; i < static_cast<int>(rounds); ++i)
319  {
320  const __m128i rk = _mm_set1_epi32(subkeys[i]);
321 
322  x1 = RotateRight32<8>(x1);
323  x1 = _mm_add_epi32(x1, y1);
324  x1 = _mm_xor_si128(x1, rk);
325  y1 = RotateLeft32<3>(y1);
326  y1 = _mm_xor_si128(y1, x1);
327  }
328 
329  // The is roughly the SSE equivalent to ARM vzp32
330  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
331  block0 = _mm_unpacklo_epi32(y1, x1);
332  block1 = _mm_unpackhi_epi32(y1, x1);
333 }
334 
335 inline void SPECK64_Dec_Block(__m128i &block0, __m128i &block1,
336  const word32 *subkeys, unsigned int rounds)
337 {
338  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
339  const __m128 t0 = _mm_castsi128_ps(block0);
340  const __m128 t1 = _mm_castsi128_ps(block1);
341  __m128i x1 = _mm_castps_si128(_mm_shuffle_ps(t0, t1, _MM_SHUFFLE(3,1,3,1)));
342  __m128i y1 = _mm_castps_si128(_mm_shuffle_ps(t0, t1, _MM_SHUFFLE(2,0,2,0)));
343 
344  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
345  {
346  const __m128i rk = _mm_set1_epi32(subkeys[i]);
347 
348  y1 = _mm_xor_si128(y1, x1);
349  y1 = RotateRight32<3>(y1);
350  x1 = _mm_xor_si128(x1, rk);
351  x1 = _mm_sub_epi32(x1, y1);
352  x1 = RotateLeft32<8>(x1);
353  }
354 
355  // The is roughly the SSE equivalent to ARM vzp32
356  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
357  block0 = _mm_unpacklo_epi32(y1, x1);
358  block1 = _mm_unpackhi_epi32(y1, x1);
359 }
360 
361 inline void SPECK64_Enc_6_Blocks(__m128i &block0, __m128i &block1,
362  __m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5,
363  const word32 *subkeys, unsigned int rounds)
364 {
365  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
366  const __m128 t0 = _mm_castsi128_ps(block0);
367  const __m128 t1 = _mm_castsi128_ps(block1);
368  __m128i x1 = _mm_castps_si128(_mm_shuffle_ps(t0, t1, _MM_SHUFFLE(3,1,3,1)));
369  __m128i y1 = _mm_castps_si128(_mm_shuffle_ps(t0, t1, _MM_SHUFFLE(2,0,2,0)));
370 
371  const __m128 t2 = _mm_castsi128_ps(block2);
372  const __m128 t3 = _mm_castsi128_ps(block3);
373  __m128i x2 = _mm_castps_si128(_mm_shuffle_ps(t2, t3, _MM_SHUFFLE(3,1,3,1)));
374  __m128i y2 = _mm_castps_si128(_mm_shuffle_ps(t2, t3, _MM_SHUFFLE(2,0,2,0)));
375 
376  const __m128 t4 = _mm_castsi128_ps(block4);
377  const __m128 t5 = _mm_castsi128_ps(block5);
378  __m128i x3 = _mm_castps_si128(_mm_shuffle_ps(t4, t5, _MM_SHUFFLE(3,1,3,1)));
379  __m128i y3 = _mm_castps_si128(_mm_shuffle_ps(t4, t5, _MM_SHUFFLE(2,0,2,0)));
380 
381  for (int i=0; i < static_cast<int>(rounds); ++i)
382  {
383  const __m128i rk = _mm_set1_epi32(subkeys[i]);
384 
385  x1 = RotateRight32<8>(x1);
386  x2 = RotateRight32<8>(x2);
387  x3 = RotateRight32<8>(x3);
388  x1 = _mm_add_epi32(x1, y1);
389  x2 = _mm_add_epi32(x2, y2);
390  x3 = _mm_add_epi32(x3, y3);
391  x1 = _mm_xor_si128(x1, rk);
392  x2 = _mm_xor_si128(x2, rk);
393  x3 = _mm_xor_si128(x3, rk);
394  y1 = RotateLeft32<3>(y1);
395  y2 = RotateLeft32<3>(y2);
396  y3 = RotateLeft32<3>(y3);
397  y1 = _mm_xor_si128(y1, x1);
398  y2 = _mm_xor_si128(y2, x2);
399  y3 = _mm_xor_si128(y3, x3);
400  }
401 
402  // The is roughly the SSE equivalent to ARM vzp32
403  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
404  block0 = _mm_unpacklo_epi32(y1, x1);
405  block1 = _mm_unpackhi_epi32(y1, x1);
406  block2 = _mm_unpacklo_epi32(y2, x2);
407  block3 = _mm_unpackhi_epi32(y2, x2);
408  block4 = _mm_unpacklo_epi32(y3, x3);
409  block5 = _mm_unpackhi_epi32(y3, x3);
410 }
411 
412 inline void SPECK64_Dec_6_Blocks(__m128i &block0, __m128i &block1,
413  __m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5,
414  const word32 *subkeys, unsigned int rounds)
415 {
416  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
417  const __m128 t0 = _mm_castsi128_ps(block0);
418  const __m128 t1 = _mm_castsi128_ps(block1);
419  __m128i x1 = _mm_castps_si128(_mm_shuffle_ps(t0, t1, _MM_SHUFFLE(3,1,3,1)));
420  __m128i y1 = _mm_castps_si128(_mm_shuffle_ps(t0, t1, _MM_SHUFFLE(2,0,2,0)));
421 
422  const __m128 t2 = _mm_castsi128_ps(block2);
423  const __m128 t3 = _mm_castsi128_ps(block3);
424  __m128i x2 = _mm_castps_si128(_mm_shuffle_ps(t2, t3, _MM_SHUFFLE(3,1,3,1)));
425  __m128i y2 = _mm_castps_si128(_mm_shuffle_ps(t2, t3, _MM_SHUFFLE(2,0,2,0)));
426 
427  const __m128 t4 = _mm_castsi128_ps(block4);
428  const __m128 t5 = _mm_castsi128_ps(block5);
429  __m128i x3 = _mm_castps_si128(_mm_shuffle_ps(t4, t5, _MM_SHUFFLE(3,1,3,1)));
430  __m128i y3 = _mm_castps_si128(_mm_shuffle_ps(t4, t5, _MM_SHUFFLE(2,0,2,0)));
431 
432  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
433  {
434  const __m128i rk = _mm_set1_epi32(subkeys[i]);
435 
436  y1 = _mm_xor_si128(y1, x1);
437  y2 = _mm_xor_si128(y2, x2);
438  y3 = _mm_xor_si128(y3, x3);
439  y1 = RotateRight32<3>(y1);
440  y2 = RotateRight32<3>(y2);
441  y3 = RotateRight32<3>(y3);
442  x1 = _mm_xor_si128(x1, rk);
443  x2 = _mm_xor_si128(x2, rk);
444  x3 = _mm_xor_si128(x3, rk);
445  x1 = _mm_sub_epi32(x1, y1);
446  x2 = _mm_sub_epi32(x2, y2);
447  x3 = _mm_sub_epi32(x3, y3);
448  x1 = RotateLeft32<8>(x1);
449  x2 = RotateLeft32<8>(x2);
450  x3 = RotateLeft32<8>(x3);
451  }
452 
453  // The is roughly the SSE equivalent to ARM vzp32
454  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
455  block0 = _mm_unpacklo_epi32(y1, x1);
456  block1 = _mm_unpackhi_epi32(y1, x1);
457  block2 = _mm_unpacklo_epi32(y2, x2);
458  block3 = _mm_unpackhi_epi32(y2, x2);
459  block4 = _mm_unpacklo_epi32(y3, x3);
460  block5 = _mm_unpackhi_epi32(y3, x3);
461 }
462 
463 #endif // CRYPTOPP_SSE41_AVAILABLE
464 
465 // ***************************** Altivec ***************************** //
466 
467 #if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
470 
471 using CryptoPP::VecAdd;
472 using CryptoPP::VecSub;
473 using CryptoPP::VecXor;
474 using CryptoPP::VecLoad;
476 
477 // Rotate left by bit count
478 template<unsigned int C>
479 inline uint32x4_p RotateLeft32(const uint32x4_p val)
480 {
481  const uint32x4_p m = {C, C, C, C};
482  return vec_rl(val, m);
483 }
484 
485 // Rotate right by bit count
486 template<unsigned int C>
487 inline uint32x4_p RotateRight32(const uint32x4_p val)
488 {
489  const uint32x4_p m = {32-C, 32-C, 32-C, 32-C};
490  return vec_rl(val, m);
491 }
492 
493 void SPECK64_Enc_Block(uint32x4_p &block0, uint32x4_p &block1,
494  const word32 *subkeys, unsigned int rounds)
495 {
496 #if (CRYPTOPP_BIG_ENDIAN)
497  const uint8x16_p m1 = {7,6,5,4, 15,14,13,12, 23,22,21,20, 31,30,29,28};
498  const uint8x16_p m2 = {3,2,1,0, 11,10,9,8, 19,18,17,16, 27,26,25,24};
499 #else
500  const uint8x16_p m1 = {3,2,1,0, 11,10,9,8, 19,18,17,16, 27,26,25,24};
501  const uint8x16_p m2 = {7,6,5,4, 15,14,13,12, 23,22,21,20, 31,30,29,28};
502 #endif
503 
504  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
505  uint32x4_p x1 = VecPermute(block0, block1, m1);
506  uint32x4_p y1 = VecPermute(block0, block1, m2);
507 
508  for (int i=0; i < static_cast<int>(rounds); ++i)
509  {
510 #if CRYPTOPP_POWER8_AVAILABLE
511  const uint32x4_p rk = vec_splats(subkeys[i]);
512 #else
513  // subkeys has extra elements so memory backs the last subkey
514  const uint8x16_p m = {0,1,2,3, 0,1,2,3, 0,1,2,3, 0,1,2,3};
515  uint32x4_p rk = VecLoad(subkeys+i);
516  rk = VecPermute(rk, rk, m);
517 #endif
518 
519  x1 = RotateRight32<8>(x1);
520  x1 = VecAdd(x1, y1);
521  x1 = VecXor(x1, rk);
522 
523  y1 = RotateLeft32<3>(y1);
524  y1 = VecXor(y1, x1);
525  }
526 
527 #if (CRYPTOPP_BIG_ENDIAN)
528  const uint8x16_p m3 = {19,18,17,16, 3,2,1,0, 23,22,21,20, 7,6,5,4};
529  const uint8x16_p m4 = {27,26,25,24, 11,10,9,8, 31,30,29,28, 15,14,13,12};
530 #else
531  const uint8x16_p m3 = {3,2,1,0, 19,18,17,16, 7,6,5,4, 23,22,21,20};
532  const uint8x16_p m4 = {11,10,9,8, 27,26,25,24, 15,14,13,12, 31,30,29,28};
533 #endif
534 
535  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
536  block0 = (uint32x4_p)VecPermute(x1, y1, m3);
537  block1 = (uint32x4_p)VecPermute(x1, y1, m4);
538 }
539 
540 void SPECK64_Dec_Block(uint32x4_p &block0, uint32x4_p &block1,
541  const word32 *subkeys, unsigned int rounds)
542 {
543 #if (CRYPTOPP_BIG_ENDIAN)
544  const uint8x16_p m1 = {7,6,5,4, 15,14,13,12, 23,22,21,20, 31,30,29,28};
545  const uint8x16_p m2 = {3,2,1,0, 11,10,9,8, 19,18,17,16, 27,26,25,24};
546 #else
547  const uint8x16_p m1 = {3,2,1,0, 11,10,9,8, 19,18,17,16, 27,26,25,24};
548  const uint8x16_p m2 = {7,6,5,4, 15,14,13,12, 23,22,21,20, 31,30,29,28};
549 #endif
550 
551  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
552  uint32x4_p x1 = VecPermute(block0, block1, m1);
553  uint32x4_p y1 = VecPermute(block0, block1, m2);
554 
555  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
556  {
557 #if CRYPTOPP_POWER8_AVAILABLE
558  const uint32x4_p rk = vec_splats(subkeys[i]);
559 #else
560  // subkeys has extra elements so memory backs the last subkey
561  const uint8x16_p m = {0,1,2,3, 0,1,2,3, 0,1,2,3, 0,1,2,3};
562  uint32x4_p rk = VecLoad(subkeys+i);
563  rk = VecPermute(rk, rk, m);
564 #endif
565 
566  y1 = VecXor(y1, x1);
567  y1 = RotateRight32<3>(y1);
568 
569  x1 = VecXor(x1, rk);
570  x1 = VecSub(x1, y1);
571  x1 = RotateLeft32<8>(x1);
572  }
573 
574 #if (CRYPTOPP_BIG_ENDIAN)
575  const uint8x16_p m3 = {19,18,17,16, 3,2,1,0, 23,22,21,20, 7,6,5,4};
576  const uint8x16_p m4 = {27,26,25,24, 11,10,9,8, 31,30,29,28, 15,14,13,12};
577 #else
578  const uint8x16_p m3 = {3,2,1,0, 19,18,17,16, 7,6,5,4, 23,22,21,20};
579  const uint8x16_p m4 = {11,10,9,8, 27,26,25,24, 15,14,13,12, 31,30,29,28};
580 #endif
581 
582  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
583  block0 = (uint32x4_p)VecPermute(x1, y1, m3);
584  block1 = (uint32x4_p)VecPermute(x1, y1, m4);
585 }
586 
587 void SPECK64_Enc_6_Blocks(uint32x4_p &block0, uint32x4_p &block1,
588  uint32x4_p &block2, uint32x4_p &block3, uint32x4_p &block4,
589  uint32x4_p &block5, const word32 *subkeys, unsigned int rounds)
590 {
591 #if (CRYPTOPP_BIG_ENDIAN)
592  const uint8x16_p m1 = {7,6,5,4, 15,14,13,12, 23,22,21,20, 31,30,29,28};
593  const uint8x16_p m2 = {3,2,1,0, 11,10,9,8, 19,18,17,16, 27,26,25,24};
594 #else
595  const uint8x16_p m1 = {3,2,1,0, 11,10,9,8, 19,18,17,16, 27,26,25,24};
596  const uint8x16_p m2 = {7,6,5,4, 15,14,13,12, 23,22,21,20, 31,30,29,28};
597 #endif
598 
599  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
600  uint32x4_p x1 = (uint32x4_p)VecPermute(block0, block1, m1);
601  uint32x4_p y1 = (uint32x4_p)VecPermute(block0, block1, m2);
602  uint32x4_p x2 = (uint32x4_p)VecPermute(block2, block3, m1);
603  uint32x4_p y2 = (uint32x4_p)VecPermute(block2, block3, m2);
604  uint32x4_p x3 = (uint32x4_p)VecPermute(block4, block5, m1);
605  uint32x4_p y3 = (uint32x4_p)VecPermute(block4, block5, m2);
606 
607  for (int i=0; i < static_cast<int>(rounds); ++i)
608  {
609 #if CRYPTOPP_POWER8_AVAILABLE
610  const uint32x4_p rk = vec_splats(subkeys[i]);
611 #else
612  // subkeys has extra elements so memory backs the last subkey
613  const uint8x16_p m = {0,1,2,3, 0,1,2,3, 0,1,2,3, 0,1,2,3};
614  uint32x4_p rk = VecLoad(subkeys+i);
615  rk = VecPermute(rk, rk, m);
616 #endif
617 
618  x1 = RotateRight32<8>(x1);
619  x2 = RotateRight32<8>(x2);
620  x3 = RotateRight32<8>(x3);
621 
622  x1 = VecAdd(x1, y1);
623  x2 = VecAdd(x2, y2);
624  x3 = VecAdd(x3, y3);
625 
626  x1 = VecXor(x1, rk);
627  x2 = VecXor(x2, rk);
628  x3 = VecXor(x3, rk);
629 
630  y1 = RotateLeft32<3>(y1);
631  y2 = RotateLeft32<3>(y2);
632  y3 = RotateLeft32<3>(y3);
633 
634  y1 = VecXor(y1, x1);
635  y2 = VecXor(y2, x2);
636  y3 = VecXor(y3, x3);
637  }
638 
639 #if (CRYPTOPP_BIG_ENDIAN)
640  const uint8x16_p m3 = {19,18,17,16, 3,2,1,0, 23,22,21,20, 7,6,5,4};
641  const uint8x16_p m4 = {27,26,25,24, 11,10,9,8, 31,30,29,28, 15,14,13,12};
642 #else
643  const uint8x16_p m3 = {3,2,1,0, 19,18,17,16, 7,6,5,4, 23,22,21,20};
644  const uint8x16_p m4 = {11,10,9,8, 27,26,25,24, 15,14,13,12, 31,30,29,28};
645 #endif
646 
647  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
648  block0 = (uint32x4_p)VecPermute(x1, y1, m3);
649  block1 = (uint32x4_p)VecPermute(x1, y1, m4);
650  block2 = (uint32x4_p)VecPermute(x2, y2, m3);
651  block3 = (uint32x4_p)VecPermute(x2, y2, m4);
652  block4 = (uint32x4_p)VecPermute(x3, y3, m3);
653  block5 = (uint32x4_p)VecPermute(x3, y3, m4);
654 }
655 
656 void SPECK64_Dec_6_Blocks(uint32x4_p &block0, uint32x4_p &block1,
657  uint32x4_p &block2, uint32x4_p &block3, uint32x4_p &block4,
658  uint32x4_p &block5, const word32 *subkeys, unsigned int rounds)
659 {
660 #if (CRYPTOPP_BIG_ENDIAN)
661  const uint8x16_p m1 = {7,6,5,4, 15,14,13,12, 23,22,21,20, 31,30,29,28};
662  const uint8x16_p m2 = {3,2,1,0, 11,10,9,8, 19,18,17,16, 27,26,25,24};
663 #else
664  const uint8x16_p m1 = {3,2,1,0, 11,10,9,8, 19,18,17,16, 27,26,25,24};
665  const uint8x16_p m2 = {7,6,5,4, 15,14,13,12, 23,22,21,20, 31,30,29,28};
666 #endif
667 
668  // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
669  uint32x4_p x1 = (uint32x4_p)VecPermute(block0, block1, m1);
670  uint32x4_p y1 = (uint32x4_p)VecPermute(block0, block1, m2);
671  uint32x4_p x2 = (uint32x4_p)VecPermute(block2, block3, m1);
672  uint32x4_p y2 = (uint32x4_p)VecPermute(block2, block3, m2);
673  uint32x4_p x3 = (uint32x4_p)VecPermute(block4, block5, m1);
674  uint32x4_p y3 = (uint32x4_p)VecPermute(block4, block5, m2);
675 
676  for (int i = static_cast<int>(rounds-1); i >= 0; --i)
677  {
678 #if CRYPTOPP_POWER8_AVAILABLE
679  const uint32x4_p rk = vec_splats(subkeys[i]);
680 #else
681  // subkeys has extra elements so memory backs the last subkey
682  const uint8x16_p m = {0,1,2,3, 0,1,2,3, 0,1,2,3, 0,1,2,3};
683  uint32x4_p rk = VecLoad(subkeys+i);
684  rk = VecPermute(rk, rk, m);
685 #endif
686 
687  y1 = VecXor(y1, x1);
688  y2 = VecXor(y2, x2);
689  y3 = VecXor(y3, x3);
690 
691  y1 = RotateRight32<3>(y1);
692  y2 = RotateRight32<3>(y2);
693  y3 = RotateRight32<3>(y3);
694 
695  x1 = VecXor(x1, rk);
696  x2 = VecXor(x2, rk);
697  x3 = VecXor(x3, rk);
698 
699  x1 = VecSub(x1, y1);
700  x2 = VecSub(x2, y2);
701  x3 = VecSub(x3, y3);
702 
703  x1 = RotateLeft32<8>(x1);
704  x2 = RotateLeft32<8>(x2);
705  x3 = RotateLeft32<8>(x3);
706  }
707 
708 #if (CRYPTOPP_BIG_ENDIAN)
709  const uint8x16_p m3 = {19,18,17,16, 3,2,1,0, 23,22,21,20, 7,6,5,4};
710  const uint8x16_p m4 = {27,26,25,24, 11,10,9,8, 31,30,29,28, 15,14,13,12};
711 #else
712  const uint8x16_p m3 = {3,2,1,0, 19,18,17,16, 7,6,5,4, 23,22,21,20};
713  const uint8x16_p m4 = {11,10,9,8, 27,26,25,24, 15,14,13,12, 31,30,29,28};
714 #endif
715 
716  // [A1 A3 B1 B3][A2 A4 B2 B4] => [A1 A2 A3 A4][B1 B2 B3 B4]
717  block0 = (uint32x4_p)VecPermute(x1, y1, m3);
718  block1 = (uint32x4_p)VecPermute(x1, y1, m4);
719  block2 = (uint32x4_p)VecPermute(x2, y2, m3);
720  block3 = (uint32x4_p)VecPermute(x2, y2, m4);
721  block4 = (uint32x4_p)VecPermute(x3, y3, m3);
722  block5 = (uint32x4_p)VecPermute(x3, y3, m4);
723 }
724 
725 #endif // CRYPTOPP_ALTIVEC_AVAILABLE
726 
727 ANONYMOUS_NAMESPACE_END
728 
729 ///////////////////////////////////////////////////////////////////////
730 
731 NAMESPACE_BEGIN(CryptoPP)
732 
733 // *************************** ARM NEON **************************** //
734 
735 #if (CRYPTOPP_ARM_NEON_AVAILABLE)
736 size_t SPECK64_Enc_AdvancedProcessBlocks_NEON(const word32* subKeys, size_t rounds,
737  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
738 {
739  return AdvancedProcessBlocks64_6x2_NEON(SPECK64_Enc_Block, SPECK64_Enc_6_Blocks,
740  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
741 }
742 
743 size_t SPECK64_Dec_AdvancedProcessBlocks_NEON(const word32* subKeys, size_t rounds,
744  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
745 {
746  return AdvancedProcessBlocks64_6x2_NEON(SPECK64_Dec_Block, SPECK64_Dec_6_Blocks,
747  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
748 }
749 #endif
750 
751 // ***************************** IA-32 ***************************** //
752 
753 #if defined(CRYPTOPP_SSE41_AVAILABLE)
754 size_t SPECK64_Enc_AdvancedProcessBlocks_SSE41(const word32* subKeys, size_t rounds,
755  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
756 {
757  return AdvancedProcessBlocks64_6x2_SSE(SPECK64_Enc_Block, SPECK64_Enc_6_Blocks,
758  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
759 }
760 
761 size_t SPECK64_Dec_AdvancedProcessBlocks_SSE41(const word32* subKeys, size_t rounds,
762  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
763 {
764  return AdvancedProcessBlocks64_6x2_SSE(SPECK64_Dec_Block, SPECK64_Dec_6_Blocks,
765  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
766 }
767 #endif
768 
769 // ***************************** Altivec ***************************** //
770 
771 #if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
772 size_t SPECK64_Enc_AdvancedProcessBlocks_ALTIVEC(const word32* subKeys, size_t rounds,
773  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
774 {
775  return AdvancedProcessBlocks64_6x2_ALTIVEC(SPECK64_Enc_Block, SPECK64_Enc_6_Blocks,
776  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
777 }
778 
779 size_t SPECK64_Dec_AdvancedProcessBlocks_ALTIVEC(const word32* subKeys, size_t rounds,
780  const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
781 {
782  return AdvancedProcessBlocks64_6x2_ALTIVEC(SPECK64_Dec_Block, SPECK64_Dec_6_Blocks,
783  subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
784 }
785 #endif
786 
787 NAMESPACE_END
uint8x16_p
__vector unsigned char uint8x16_p
Vector of 8-bit elements.
Definition: ppc_simd.h:119
VecSub
T1 VecSub(const T1 vec1, const T2 vec2)
Subtract two vectors.
Definition: ppc_simd.h:956
uint32x4_p
__vector unsigned int uint32x4_p
Vector of 32-bit elements.
Definition: ppc_simd.h:129
pch.h
Precompiled header file.
misc.h
Utility functions for the Crypto++ library.
adv_simd.h
Template for AdvancedProcessBlocks and SIMD processing.
VecLoad
uint32x4_p VecLoad(const byte src[16])
Loads a vector from a byte array.
Definition: ppc_simd.h:253
speck.h
Classes for the Speck block cipher.
VecXor
T1 VecXor(const T1 vec1, const T2 vec2)
XOR two vectors.
Definition: ppc_simd.h:916
CryptoPP
Crypto++ library namespace.
config.h
Library configuration file.
VecPermute
T1 VecPermute(const T1 vec, const T2 mask)
Permutes a vector.
Definition: ppc_simd.h:1010
ppc_simd.h
Support functions for PowerPC and vector operations.
VecAdd
T1 VecAdd(const T1 vec1, const T2 vec2)
Add two vectors.
Definition: ppc_simd.h:939