-
Notifications
You must be signed in to change notification settings - Fork 7
/
sign.c
1528 lines (1375 loc) · 40 KB
/
sign.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*
* Falcon signature generation.
*
* ==========================(LICENSE BEGIN)============================
*
* Copyright (c) 2017-2019 Falcon Project
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice shall be
* included in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
* ===========================(LICENSE END)=============================
*
* @author Thomas Pornin <[email protected]>
*/
#include "inner.h"
/* =================================================================== */
/*
* Compute degree N from logarithm 'logn'.
*/
#define MKN(logn) ((size_t)1 << (logn))
/* =================================================================== */
/*
* Binary case:
* N = 2^logn
* phi = X^N+1
*/
/*
* Get the size of the LDL tree for an input with polynomials of size
* 2^logn. The size is expressed in the number of elements.
*/
static inline unsigned
ffLDL_treesize(unsigned logn)
{
/*
* For logn = 0 (polynomials are constant), the "tree" is a
* single element. Otherwise, the tree node has size 2^logn, and
* has two child trees for size logn-1 each. Thus, treesize s()
* must fulfill these two relations:
*
* s(0) = 1
* s(logn) = (2^logn) + 2*s(logn-1)
*/
return (logn + 1) << logn;
}
/*
* Inner function for ffLDL_fft(). It expects the matrix to be both
* auto-adjoint and quasicyclic; also, it uses the source operands
* as modifiable temporaries.
*
* tmp[] must have room for at least one polynomial.
*/
static void
ffLDL_fft_inner(fpr *restrict tree,
fpr *restrict g0, fpr *restrict g1, unsigned logn, fpr *restrict tmp)
{
size_t n, hn;
n = MKN(logn);
if (n == 1) {
tree[0] = g0[0];
return;
}
hn = n >> 1;
/*
* The LDL decomposition yields L (which is written in the tree)
* and the diagonal of D. Since d00 = g0, we just write d11
* into tmp.
*/
Zf(poly_LDLmv_fft)(tmp, tree, g0, g1, g0, logn);
/*
* Split d00 (currently in g0) and d11 (currently in tmp). We
* reuse g0 and g1 as temporary storage spaces:
* d00 splits into g1, g1+hn
* d11 splits into g0, g0+hn
*/
Zf(poly_split_fft)(g1, g1 + hn, g0, logn);
Zf(poly_split_fft)(g0, g0 + hn, tmp, logn);
/*
* Each split result is the first row of a new auto-adjoint
* quasicyclic matrix for the next recursive step.
*/
ffLDL_fft_inner(tree + n,
g1, g1 + hn, logn - 1, tmp);
ffLDL_fft_inner(tree + n + ffLDL_treesize(logn - 1),
g0, g0 + hn, logn - 1, tmp);
}
/*
* Compute the ffLDL tree of an auto-adjoint matrix G. The matrix
* is provided as three polynomials (FFT representation).
*
* The "tree" array is filled with the computed tree, of size
* (logn+1)*(2^logn) elements (see ffLDL_treesize()).
*
* Input arrays MUST NOT overlap, except possibly the three unmodified
* arrays g00, g01 and g11. tmp[] should have room for at least three
* polynomials of 2^logn elements each.
*/
static void
ffLDL_fft(fpr *restrict tree, const fpr *restrict g00,
const fpr *restrict g01, const fpr *restrict g11,
unsigned logn, fpr *restrict tmp)
{
size_t n, hn;
fpr *d00, *d11;
n = MKN(logn);
if (n == 1) {
tree[0] = g00[0];
return;
}
hn = n >> 1;
d00 = tmp;
d11 = tmp + n;
tmp += n << 1;
memcpy(d00, g00, n * sizeof *g00);
Zf(poly_LDLmv_fft)(d11, tree, g00, g01, g11, logn);
Zf(poly_split_fft)(tmp, tmp + hn, d00, logn);
Zf(poly_split_fft)(d00, d00 + hn, d11, logn);
memcpy(d11, tmp, n * sizeof *tmp);
ffLDL_fft_inner(tree + n,
d11, d11 + hn, logn - 1, tmp);
ffLDL_fft_inner(tree + n + ffLDL_treesize(logn - 1),
d00, d00 + hn, logn - 1, tmp);
}
/*
* Normalize an ffLDL tree: each leaf of value x is replaced with
* sigma / sqrt(x).
*/
static void
ffLDL_binary_normalize(fpr *tree, unsigned orig_logn, unsigned logn)
{
/*
* TODO: make an iterative version.
*/
size_t n;
n = MKN(logn);
if (n == 1) {
/*
* We actually store in the tree leaf the inverse of
* the value mandated by the specification: this
* saves a division both here and in the sampler.
*/
tree[0] = fpr_mul(fpr_sqrt(tree[0]), fpr_inv_sigma[orig_logn]);
} else {
ffLDL_binary_normalize(tree + n, orig_logn, logn - 1);
ffLDL_binary_normalize(tree + n + ffLDL_treesize(logn - 1),
orig_logn, logn - 1);
}
}
/* =================================================================== */
/*
* Convert an integer polynomial (with small values) into the
* representation with complex numbers.
*/
static void
smallints_to_fpr(fpr *r, const int8_t *t, unsigned logn)
{
size_t n, u;
n = MKN(logn);
for (u = 0; u < n; u ++) {
r[u] = fpr_of(t[u]);
}
}
/*
* The expanded private key contains:
* - The B0 matrix (four elements)
* - The ffLDL tree
*/
static inline size_t
skoff_b00(unsigned logn)
{
(void)logn;
return 0;
}
static inline size_t
skoff_b01(unsigned logn)
{
return MKN(logn);
}
static inline size_t
skoff_b10(unsigned logn)
{
return 2 * MKN(logn);
}
static inline size_t
skoff_b11(unsigned logn)
{
return 3 * MKN(logn);
}
static inline size_t
skoff_tree(unsigned logn)
{
return 4 * MKN(logn);
}
/* see inner.h */
void
Zf(expand_privkey)(fpr *restrict expanded_key,
const int8_t *f, const int8_t *g,
const int8_t *F, const int8_t *G,
unsigned logn, uint8_t *restrict tmp)
{
size_t n;
fpr *rf, *rg, *rF, *rG;
fpr *b00, *b01, *b10, *b11;
fpr *g00, *g01, *g11, *gxx;
fpr *tree;
n = MKN(logn);
b00 = expanded_key + skoff_b00(logn);
b01 = expanded_key + skoff_b01(logn);
b10 = expanded_key + skoff_b10(logn);
b11 = expanded_key + skoff_b11(logn);
tree = expanded_key + skoff_tree(logn);
/*
* We load the private key elements directly into the B0 matrix,
* since B0 = [[g, -f], [G, -F]].
*/
rf = b01;
rg = b00;
rF = b11;
rG = b10;
smallints_to_fpr(rf, f, logn);
smallints_to_fpr(rg, g, logn);
smallints_to_fpr(rF, F, logn);
smallints_to_fpr(rG, G, logn);
/*
* Compute the FFT for the key elements, and negate f and F.
*/
Zf(FFT)(rf, logn);
Zf(FFT)(rg, logn);
Zf(FFT)(rF, logn);
Zf(FFT)(rG, logn);
Zf(poly_neg)(rf, logn);
Zf(poly_neg)(rF, logn);
/*
* The Gram matrix is G = B·B*. Formulas are:
* g00 = b00*adj(b00) + b01*adj(b01)
* g01 = b00*adj(b10) + b01*adj(b11)
* g10 = b10*adj(b00) + b11*adj(b01)
* g11 = b10*adj(b10) + b11*adj(b11)
*
* For historical reasons, this implementation uses
* g00, g01 and g11 (upper triangle).
*/
g00 = (fpr *)tmp;
g01 = g00 + n;
g11 = g01 + n;
gxx = g11 + n;
memcpy(g00, b00, n * sizeof *b00);
Zf(poly_mulselfadj_fft)(g00, logn);
memcpy(gxx, b01, n * sizeof *b01);
Zf(poly_mulselfadj_fft)(gxx, logn);
Zf(poly_add)(g00, gxx, logn);
memcpy(g01, b00, n * sizeof *b00);
Zf(poly_muladj_fft)(g01, b10, logn);
memcpy(gxx, b01, n * sizeof *b01);
Zf(poly_muladj_fft)(gxx, b11, logn);
Zf(poly_add)(g01, gxx, logn);
memcpy(g11, b10, n * sizeof *b10);
Zf(poly_mulselfadj_fft)(g11, logn);
memcpy(gxx, b11, n * sizeof *b11);
Zf(poly_mulselfadj_fft)(gxx, logn);
Zf(poly_add)(g11, gxx, logn);
/*
* Compute the Falcon tree.
*/
ffLDL_fft(tree, g00, g01, g11, logn, gxx);
/*
* Normalize tree.
*/
ffLDL_binary_normalize(tree, logn, logn);
}
typedef int (*samplerZ)(void *ctx, fpr mu, fpr sigma);
/*
* Perform Fast Fourier Sampling for target vector t. The Gram matrix
* is provided (G = [[g00, g01], [adj(g01), g11]]). The sampled vector
* is written over (t0,t1). The Gram matrix is modified as well. The
* tmp[] buffer must have room for four polynomials.
*/
TARGET_AVX2
static void
ffSampling_fft_dyntree(samplerZ samp, void *samp_ctx,
fpr *restrict t0, fpr *restrict t1,
fpr *restrict g00, fpr *restrict g01, fpr *restrict g11,
unsigned orig_logn, unsigned logn, fpr *restrict tmp)
{
size_t n, hn;
fpr *z0, *z1;
/*
* Deepest level: the LDL tree leaf value is just g00 (the
* array has length only 1 at this point); we normalize it
* with regards to sigma, then use it for sampling.
*/
if (logn == 0) {
fpr leaf;
leaf = g00[0];
leaf = fpr_mul(fpr_sqrt(leaf), fpr_inv_sigma[orig_logn]);
t0[0] = fpr_of(samp(samp_ctx, t0[0], leaf));
t1[0] = fpr_of(samp(samp_ctx, t1[0], leaf));
return;
}
n = (size_t)1 << logn;
hn = n >> 1;
/*
* Decompose G into LDL. We only need d00 (identical to g00),
* d11, and l10; we do that in place.
*/
Zf(poly_LDL_fft)(g00, g01, g11, logn);
/*
* Split d00 and d11 and expand them into half-size quasi-cyclic
* Gram matrices. We also save l10 in tmp[].
*/
Zf(poly_split_fft)(tmp, tmp + hn, g00, logn);
memcpy(g00, tmp, n * sizeof *tmp);
Zf(poly_split_fft)(tmp, tmp + hn, g11, logn);
memcpy(g11, tmp, n * sizeof *tmp);
memcpy(tmp, g01, n * sizeof *g01);
memcpy(g01, g00, hn * sizeof *g00);
memcpy(g01 + hn, g11, hn * sizeof *g00);
/*
* The half-size Gram matrices for the recursive LDL tree
* building are now:
* - left sub-tree: g00, g00+hn, g01
* - right sub-tree: g11, g11+hn, g01+hn
* l10 is in tmp[].
*/
/*
* We split t1 and use the first recursive call on the two
* halves, using the right sub-tree. The result is merged
* back into tmp + 2*n.
*/
z1 = tmp + n;
Zf(poly_split_fft)(z1, z1 + hn, t1, logn);
ffSampling_fft_dyntree(samp, samp_ctx, z1, z1 + hn,
g11, g11 + hn, g01 + hn, orig_logn, logn - 1, z1 + n);
Zf(poly_merge_fft)(tmp + (n << 1), z1, z1 + hn, logn);
/*
* Compute tb0 = t0 + (t1 - z1) * l10.
* At that point, l10 is in tmp, t1 is unmodified, and z1 is
* in tmp + (n << 1). The buffer in z1 is free.
*
* In the end, z1 is written over t1, and tb0 is in t0.
*/
memcpy(z1, t1, n * sizeof *t1);
Zf(poly_sub)(z1, tmp + (n << 1), logn);
memcpy(t1, tmp + (n << 1), n * sizeof *tmp);
Zf(poly_mul_fft)(tmp, z1, logn);
Zf(poly_add)(t0, tmp, logn);
/*
* Second recursive invocation, on the split tb0 (currently in t0)
* and the left sub-tree.
*/
z0 = tmp;
Zf(poly_split_fft)(z0, z0 + hn, t0, logn);
ffSampling_fft_dyntree(samp, samp_ctx, z0, z0 + hn,
g00, g00 + hn, g01, orig_logn, logn - 1, z0 + n);
Zf(poly_merge_fft)(t0, z0, z0 + hn, logn);
}
/*
* Perform Fast Fourier Sampling for target vector t and LDL tree T.
* tmp[] must have size for at least two polynomials of size 2^logn.
*/
TARGET_AVX2
static void
ffSampling_fft(samplerZ samp, void *samp_ctx,
fpr *restrict z0, fpr *restrict z1,
const fpr *restrict tree,
const fpr *restrict t0, const fpr *restrict t1, unsigned logn,
fpr *restrict tmp)
{
size_t n, hn;
const fpr *tree0, *tree1;
/*
* When logn == 2, we inline the last two recursion levels.
*/
if (logn == 2) {
#if FALCON_AVX2 // yyyAVX2+1
fpr w0, w1, w2, w3, sigma;
__m128d ww0, ww1, wa, wb, wc, wd;
__m128d wy0, wy1, wz0, wz1;
__m128d half, invsqrt8, invsqrt2, neghi, neglo;
int si0, si1, si2, si3;
tree0 = tree + 4;
tree1 = tree + 8;
half = _mm_set1_pd(0.5);
invsqrt8 = _mm_set1_pd(0.353553390593273762200422181052);
invsqrt2 = _mm_set1_pd(0.707106781186547524400844362105);
neghi = _mm_set_pd(-0.0, 0.0);
neglo = _mm_set_pd(0.0, -0.0);
/*
* We split t1 into w*, then do the recursive invocation,
* with output in w*. We finally merge back into z1.
*/
ww0 = _mm_loadu_pd(&t1[0].v);
ww1 = _mm_loadu_pd(&t1[2].v);
wa = _mm_unpacklo_pd(ww0, ww1);
wb = _mm_unpackhi_pd(ww0, ww1);
wc = _mm_add_pd(wa, wb);
ww0 = _mm_mul_pd(wc, half);
wc = _mm_sub_pd(wa, wb);
wd = _mm_xor_pd(_mm_permute_pd(wc, 1), neghi);
ww1 = _mm_mul_pd(_mm_add_pd(wc, wd), invsqrt8);
w2.v = _mm_cvtsd_f64(ww1);
w3.v = _mm_cvtsd_f64(_mm_permute_pd(ww1, 1));
wa = ww1;
sigma = tree1[3];
si2 = samp(samp_ctx, w2, sigma);
si3 = samp(samp_ctx, w3, sigma);
ww1 = _mm_set_pd((double)si3, (double)si2);
wa = _mm_sub_pd(wa, ww1);
wb = _mm_loadu_pd(&tree1[0].v);
wc = _mm_mul_pd(wa, wb);
wd = _mm_mul_pd(wa, _mm_permute_pd(wb, 1));
wa = _mm_unpacklo_pd(wc, wd);
wb = _mm_unpackhi_pd(wc, wd);
ww0 = _mm_add_pd(ww0, _mm_add_pd(wa, _mm_xor_pd(wb, neglo)));
w0.v = _mm_cvtsd_f64(ww0);
w1.v = _mm_cvtsd_f64(_mm_permute_pd(ww0, 1));
sigma = tree1[2];
si0 = samp(samp_ctx, w0, sigma);
si1 = samp(samp_ctx, w1, sigma);
ww0 = _mm_set_pd((double)si1, (double)si0);
wc = _mm_mul_pd(
_mm_set_pd((double)(si2 + si3), (double)(si2 - si3)),
invsqrt2);
wa = _mm_add_pd(ww0, wc);
wb = _mm_sub_pd(ww0, wc);
ww0 = _mm_unpacklo_pd(wa, wb);
ww1 = _mm_unpackhi_pd(wa, wb);
_mm_storeu_pd(&z1[0].v, ww0);
_mm_storeu_pd(&z1[2].v, ww1);
/*
* Compute tb0 = t0 + (t1 - z1) * L. Value tb0 ends up in w*.
*/
wy0 = _mm_sub_pd(_mm_loadu_pd(&t1[0].v), ww0);
wy1 = _mm_sub_pd(_mm_loadu_pd(&t1[2].v), ww1);
wz0 = _mm_loadu_pd(&tree[0].v);
wz1 = _mm_loadu_pd(&tree[2].v);
ww0 = _mm_sub_pd(_mm_mul_pd(wy0, wz0), _mm_mul_pd(wy1, wz1));
ww1 = _mm_add_pd(_mm_mul_pd(wy0, wz1), _mm_mul_pd(wy1, wz0));
ww0 = _mm_add_pd(ww0, _mm_loadu_pd(&t0[0].v));
ww1 = _mm_add_pd(ww1, _mm_loadu_pd(&t0[2].v));
/*
* Second recursive invocation.
*/
wa = _mm_unpacklo_pd(ww0, ww1);
wb = _mm_unpackhi_pd(ww0, ww1);
wc = _mm_add_pd(wa, wb);
ww0 = _mm_mul_pd(wc, half);
wc = _mm_sub_pd(wa, wb);
wd = _mm_xor_pd(_mm_permute_pd(wc, 1), neghi);
ww1 = _mm_mul_pd(_mm_add_pd(wc, wd), invsqrt8);
w2.v = _mm_cvtsd_f64(ww1);
w3.v = _mm_cvtsd_f64(_mm_permute_pd(ww1, 1));
wa = ww1;
sigma = tree0[3];
si2 = samp(samp_ctx, w2, sigma);
si3 = samp(samp_ctx, w3, sigma);
ww1 = _mm_set_pd((double)si3, (double)si2);
wa = _mm_sub_pd(wa, ww1);
wb = _mm_loadu_pd(&tree0[0].v);
wc = _mm_mul_pd(wa, wb);
wd = _mm_mul_pd(wa, _mm_permute_pd(wb, 1));
wa = _mm_unpacklo_pd(wc, wd);
wb = _mm_unpackhi_pd(wc, wd);
ww0 = _mm_add_pd(ww0, _mm_add_pd(wa, _mm_xor_pd(wb, neglo)));
w0.v = _mm_cvtsd_f64(ww0);
w1.v = _mm_cvtsd_f64(_mm_permute_pd(ww0, 1));
sigma = tree0[2];
si0 = samp(samp_ctx, w0, sigma);
si1 = samp(samp_ctx, w1, sigma);
ww0 = _mm_set_pd((double)si1, (double)si0);
wc = _mm_mul_pd(
_mm_set_pd((double)(si2 + si3), (double)(si2 - si3)),
invsqrt2);
wa = _mm_add_pd(ww0, wc);
wb = _mm_sub_pd(ww0, wc);
ww0 = _mm_unpacklo_pd(wa, wb);
ww1 = _mm_unpackhi_pd(wa, wb);
_mm_storeu_pd(&z0[0].v, ww0);
_mm_storeu_pd(&z0[2].v, ww1);
return;
#else // yyyAVX2+0
fpr x0, x1, y0, y1, w0, w1, w2, w3, sigma;
fpr a_re, a_im, b_re, b_im, c_re, c_im;
tree0 = tree + 4;
tree1 = tree + 8;
/*
* We split t1 into w*, then do the recursive invocation,
* with output in w*. We finally merge back into z1.
*/
a_re = t1[0];
a_im = t1[2];
b_re = t1[1];
b_im = t1[3];
c_re = fpr_add(a_re, b_re);
c_im = fpr_add(a_im, b_im);
w0 = fpr_half(c_re);
w1 = fpr_half(c_im);
c_re = fpr_sub(a_re, b_re);
c_im = fpr_sub(a_im, b_im);
w2 = fpr_mul(fpr_add(c_re, c_im), fpr_invsqrt8);
w3 = fpr_mul(fpr_sub(c_im, c_re), fpr_invsqrt8);
x0 = w2;
x1 = w3;
sigma = tree1[3];
w2 = fpr_of(samp(samp_ctx, x0, sigma));
w3 = fpr_of(samp(samp_ctx, x1, sigma));
a_re = fpr_sub(x0, w2);
a_im = fpr_sub(x1, w3);
b_re = tree1[0];
b_im = tree1[1];
c_re = fpr_sub(fpr_mul(a_re, b_re), fpr_mul(a_im, b_im));
c_im = fpr_add(fpr_mul(a_re, b_im), fpr_mul(a_im, b_re));
x0 = fpr_add(c_re, w0);
x1 = fpr_add(c_im, w1);
sigma = tree1[2];
w0 = fpr_of(samp(samp_ctx, x0, sigma));
w1 = fpr_of(samp(samp_ctx, x1, sigma));
a_re = w0;
a_im = w1;
b_re = w2;
b_im = w3;
c_re = fpr_mul(fpr_sub(b_re, b_im), fpr_invsqrt2);
c_im = fpr_mul(fpr_add(b_re, b_im), fpr_invsqrt2);
z1[0] = w0 = fpr_add(a_re, c_re);
z1[2] = w2 = fpr_add(a_im, c_im);
z1[1] = w1 = fpr_sub(a_re, c_re);
z1[3] = w3 = fpr_sub(a_im, c_im);
/*
* Compute tb0 = t0 + (t1 - z1) * L. Value tb0 ends up in w*.
*/
w0 = fpr_sub(t1[0], w0);
w1 = fpr_sub(t1[1], w1);
w2 = fpr_sub(t1[2], w2);
w3 = fpr_sub(t1[3], w3);
a_re = w0;
a_im = w2;
b_re = tree[0];
b_im = tree[2];
w0 = fpr_sub(fpr_mul(a_re, b_re), fpr_mul(a_im, b_im));
w2 = fpr_add(fpr_mul(a_re, b_im), fpr_mul(a_im, b_re));
a_re = w1;
a_im = w3;
b_re = tree[1];
b_im = tree[3];
w1 = fpr_sub(fpr_mul(a_re, b_re), fpr_mul(a_im, b_im));
w3 = fpr_add(fpr_mul(a_re, b_im), fpr_mul(a_im, b_re));
w0 = fpr_add(w0, t0[0]);
w1 = fpr_add(w1, t0[1]);
w2 = fpr_add(w2, t0[2]);
w3 = fpr_add(w3, t0[3]);
/*
* Second recursive invocation.
*/
a_re = w0;
a_im = w2;
b_re = w1;
b_im = w3;
c_re = fpr_add(a_re, b_re);
c_im = fpr_add(a_im, b_im);
w0 = fpr_half(c_re);
w1 = fpr_half(c_im);
c_re = fpr_sub(a_re, b_re);
c_im = fpr_sub(a_im, b_im);
w2 = fpr_mul(fpr_add(c_re, c_im), fpr_invsqrt8);
w3 = fpr_mul(fpr_sub(c_im, c_re), fpr_invsqrt8);
x0 = w2;
x1 = w3;
sigma = tree0[3];
w2 = y0 = fpr_of(samp(samp_ctx, x0, sigma));
w3 = y1 = fpr_of(samp(samp_ctx, x1, sigma));
a_re = fpr_sub(x0, y0);
a_im = fpr_sub(x1, y1);
b_re = tree0[0];
b_im = tree0[1];
c_re = fpr_sub(fpr_mul(a_re, b_re), fpr_mul(a_im, b_im));
c_im = fpr_add(fpr_mul(a_re, b_im), fpr_mul(a_im, b_re));
x0 = fpr_add(c_re, w0);
x1 = fpr_add(c_im, w1);
sigma = tree0[2];
w0 = fpr_of(samp(samp_ctx, x0, sigma));
w1 = fpr_of(samp(samp_ctx, x1, sigma));
a_re = w0;
a_im = w1;
b_re = w2;
b_im = w3;
c_re = fpr_mul(fpr_sub(b_re, b_im), fpr_invsqrt2);
c_im = fpr_mul(fpr_add(b_re, b_im), fpr_invsqrt2);
z0[0] = fpr_add(a_re, c_re);
z0[2] = fpr_add(a_im, c_im);
z0[1] = fpr_sub(a_re, c_re);
z0[3] = fpr_sub(a_im, c_im);
return;
#endif // yyyAVX2-
}
/*
* Case logn == 1 is reachable only when using Falcon-2 (the
* smallest size for which Falcon is mathematically defined, but
* of course way too insecure to be of any use).
*/
if (logn == 1) {
fpr x0, x1, y0, y1, sigma;
fpr a_re, a_im, b_re, b_im, c_re, c_im;
x0 = t1[0];
x1 = t1[1];
sigma = tree[3];
z1[0] = y0 = fpr_of(samp(samp_ctx, x0, sigma));
z1[1] = y1 = fpr_of(samp(samp_ctx, x1, sigma));
a_re = fpr_sub(x0, y0);
a_im = fpr_sub(x1, y1);
b_re = tree[0];
b_im = tree[1];
c_re = fpr_sub(fpr_mul(a_re, b_re), fpr_mul(a_im, b_im));
c_im = fpr_add(fpr_mul(a_re, b_im), fpr_mul(a_im, b_re));
x0 = fpr_add(c_re, t0[0]);
x1 = fpr_add(c_im, t0[1]);
sigma = tree[2];
z0[0] = fpr_of(samp(samp_ctx, x0, sigma));
z0[1] = fpr_of(samp(samp_ctx, x1, sigma));
return;
}
/*
* Normal end of recursion is for logn == 0. Since the last
* steps of the recursions were inlined in the blocks above
* (when logn == 1 or 2), this case is not reachable, and is
* retained here only for documentation purposes.
if (logn == 0) {
fpr x0, x1, sigma;
x0 = t0[0];
x1 = t1[0];
sigma = tree[0];
z0[0] = fpr_of(samp(samp_ctx, x0, sigma));
z1[0] = fpr_of(samp(samp_ctx, x1, sigma));
return;
}
*/
/*
* General recursive case (logn >= 3).
*/
n = (size_t)1 << logn;
hn = n >> 1;
tree0 = tree + n;
tree1 = tree + n + ffLDL_treesize(logn - 1);
/*
* We split t1 into z1 (reused as temporary storage), then do
* the recursive invocation, with output in tmp. We finally
* merge back into z1.
*/
Zf(poly_split_fft)(z1, z1 + hn, t1, logn);
ffSampling_fft(samp, samp_ctx, tmp, tmp + hn,
tree1, z1, z1 + hn, logn - 1, tmp + n);
Zf(poly_merge_fft)(z1, tmp, tmp + hn, logn);
/*
* Compute tb0 = t0 + (t1 - z1) * L. Value tb0 ends up in tmp[].
*/
memcpy(tmp, t1, n * sizeof *t1);
Zf(poly_sub)(tmp, z1, logn);
Zf(poly_mul_fft)(tmp, tree, logn);
Zf(poly_add)(tmp, t0, logn);
/*
* Second recursive invocation.
*/
Zf(poly_split_fft)(z0, z0 + hn, tmp, logn);
ffSampling_fft(samp, samp_ctx, tmp, tmp + hn,
tree0, z0, z0 + hn, logn - 1, tmp + n);
Zf(poly_merge_fft)(z0, tmp, tmp + hn, logn);
}
/*
* Compute a signature: the signature contains two vectors, s1 and s2.
* The s1 vector is not returned. The squared norm of (s1,s2) is
* computed, and if it is short enough, then s2 is returned into the
* s2[] buffer, and 1 is returned; otherwise, s2[] is untouched and 0 is
* returned; the caller should then try again. This function uses an
* expanded key.
*
* tmp[] must have room for at least six polynomials.
*/
static int
do_sign_tree(samplerZ samp, void *samp_ctx, int16_t *s2,
const fpr *restrict expanded_key,
const uint16_t *hm,
unsigned logn, fpr *restrict tmp)
{
size_t n, u;
fpr *t0, *t1, *tx, *ty;
const fpr *b00, *b01, *b10, *b11, *tree;
fpr ni;
uint32_t sqn, ng;
int16_t *s1tmp, *s2tmp;
n = MKN(logn);
t0 = tmp;
t1 = t0 + n;
b00 = expanded_key + skoff_b00(logn);
b01 = expanded_key + skoff_b01(logn);
b10 = expanded_key + skoff_b10(logn);
b11 = expanded_key + skoff_b11(logn);
tree = expanded_key + skoff_tree(logn);
/*
* Set the target vector to [hm, 0] (hm is the hashed message).
*/
for (u = 0; u < n; u ++) {
t0[u] = fpr_of(hm[u]);
/* This is implicit.
t1[u] = fpr_zero;
*/
}
/*
* Apply the lattice basis to obtain the real target
* vector (after normalization with regards to modulus).
*/
Zf(FFT)(t0, logn);
ni = fpr_inverse_of_q;
memcpy(t1, t0, n * sizeof *t0);
Zf(poly_mul_fft)(t1, b01, logn);
Zf(poly_mulconst)(t1, fpr_neg(ni), logn);
Zf(poly_mul_fft)(t0, b11, logn);
Zf(poly_mulconst)(t0, ni, logn);
tx = t1 + n;
ty = tx + n;
/*
* Apply sampling. Output is written back in [tx, ty].
*/
ffSampling_fft(samp, samp_ctx, tx, ty, tree, t0, t1, logn, ty + n);
/*
* Get the lattice point corresponding to that tiny vector.
*/
memcpy(t0, tx, n * sizeof *tx);
memcpy(t1, ty, n * sizeof *ty);
Zf(poly_mul_fft)(tx, b00, logn);
Zf(poly_mul_fft)(ty, b10, logn);
Zf(poly_add)(tx, ty, logn);
memcpy(ty, t0, n * sizeof *t0);
Zf(poly_mul_fft)(ty, b01, logn);
memcpy(t0, tx, n * sizeof *tx);
Zf(poly_mul_fft)(t1, b11, logn);
Zf(poly_add)(t1, ty, logn);
Zf(iFFT)(t0, logn);
Zf(iFFT)(t1, logn);
/*
* Compute the signature.
*/
s1tmp = (int16_t *)tx;
sqn = 0;
ng = 0;
for (u = 0; u < n; u ++) {
int32_t z;
z = (int32_t)hm[u] - (int32_t)fpr_rint(t0[u]);
sqn += (uint32_t)(z * z);
ng |= sqn;
s1tmp[u] = (int16_t)z;
}
sqn |= -(ng >> 31);
/*
* With "normal" degrees (e.g. 512 or 1024), it is very
* improbable that the computed vector is not short enough;
* however, it may happen in practice for the very reduced
* versions (e.g. degree 16 or below). In that case, the caller
* will loop, and we must not write anything into s2[] because
* s2[] may overlap with the hashed message hm[] and we need
* hm[] for the next iteration.
*/
s2tmp = (int16_t *)tmp;
for (u = 0; u < n; u ++) {
s2tmp[u] = (int16_t)-fpr_rint(t1[u]);
}
if (Zf(is_short_half)(sqn, s2tmp, logn)) {
memcpy(s2, s2tmp, n * sizeof *s2);
memcpy(tmp, s1tmp, n * sizeof *s1tmp);
return 1;
}
return 0;
}
/*
* Compute a signature: the signature contains two vectors, s1 and s2.
* The s1 vector is not returned. The squared norm of (s1,s2) is
* computed, and if it is short enough, then s2 is returned into the
* s2[] buffer, and 1 is returned; otherwise, s2[] is untouched and 0 is
* returned; the caller should then try again.
*
* tmp[] must have room for at least nine polynomials.
*/
static int
do_sign_dyn(samplerZ samp, void *samp_ctx, int16_t *s2,
const int8_t *restrict f, const int8_t *restrict g,
const int8_t *restrict F, const int8_t *restrict G,
const uint16_t *hm, unsigned logn, fpr *restrict tmp)
{
size_t n, u;
fpr *t0, *t1, *tx, *ty;
fpr *b00, *b01, *b10, *b11, *g00, *g01, *g11;
fpr ni;
uint32_t sqn, ng;
int16_t *s1tmp, *s2tmp;
n = MKN(logn);
/*
* Lattice basis is B = [[g, -f], [G, -F]]. We convert it to FFT.
*/
b00 = tmp;
b01 = b00 + n;
b10 = b01 + n;
b11 = b10 + n;
smallints_to_fpr(b01, f, logn);
smallints_to_fpr(b00, g, logn);
smallints_to_fpr(b11, F, logn);
smallints_to_fpr(b10, G, logn);
Zf(FFT)(b01, logn);
Zf(FFT)(b00, logn);
Zf(FFT)(b11, logn);
Zf(FFT)(b10, logn);
Zf(poly_neg)(b01, logn);
Zf(poly_neg)(b11, logn);
/*
* Compute the Gram matrix G = B·B*. Formulas are:
* g00 = b00*adj(b00) + b01*adj(b01)
* g01 = b00*adj(b10) + b01*adj(b11)
* g10 = b10*adj(b00) + b11*adj(b01)
* g11 = b10*adj(b10) + b11*adj(b11)
*
* For historical reasons, this implementation uses
* g00, g01 and g11 (upper triangle). g10 is not kept
* since it is equal to adj(g01).
*
* We _replace_ the matrix B with the Gram matrix, but we
* must keep b01 and b11 for computing the target vector.
*/
t0 = b11 + n;
t1 = t0 + n;
memcpy(t0, b01, n * sizeof *b01);
Zf(poly_mulselfadj_fft)(t0, logn); // t0 <- b01*adj(b01)
memcpy(t1, b00, n * sizeof *b00);
Zf(poly_muladj_fft)(t1, b10, logn); // t1 <- b00*adj(b10)
Zf(poly_mulselfadj_fft)(b00, logn); // b00 <- b00*adj(b00)
Zf(poly_add)(b00, t0, logn); // b00 <- g00
memcpy(t0, b01, n * sizeof *b01);
Zf(poly_muladj_fft)(b01, b11, logn); // b01 <- b01*adj(b11)
Zf(poly_add)(b01, t1, logn); // b01 <- g01
Zf(poly_mulselfadj_fft)(b10, logn); // b10 <- b10*adj(b10)
memcpy(t1, b11, n * sizeof *b11);
Zf(poly_mulselfadj_fft)(t1, logn); // t1 <- b11*adj(b11)
Zf(poly_add)(b10, t1, logn); // b10 <- g11
/*
* We rename variables to make things clearer. The three elements
* of the Gram matrix uses the first 3*n slots of tmp[], followed
* by b11 and b01 (in that order).
*/
g00 = b00;
g01 = b01;
g11 = b10;
b01 = t0;
t0 = b01 + n;
t1 = t0 + n;
/*
* Memory layout at that point:
* g00 g01 g11 b11 b01 t0 t1
*/
/*
* Set the target vector to [hm, 0] (hm is the hashed message).
*/
for (u = 0; u < n; u ++) {
t0[u] = fpr_of(hm[u]);
/* This is implicit.
t1[u] = fpr_zero;
*/
}
/*
* Apply the lattice basis to obtain the real target
* vector (after normalization with regards to modulus).
*/
Zf(FFT)(t0, logn);
ni = fpr_inverse_of_q;
memcpy(t1, t0, n * sizeof *t0);
Zf(poly_mul_fft)(t1, b01, logn);
Zf(poly_mulconst)(t1, fpr_neg(ni), logn);
Zf(poly_mul_fft)(t0, b11, logn);
Zf(poly_mulconst)(t0, ni, logn);
/*
* b01 and b11 can be discarded, so we move back (t0,t1).
* Memory layout is now:
* g00 g01 g11 t0 t1
*/