File: mpi-pow.c

package info (click to toggle)
gnupg 1.4.18-7+deb8u5
  • links: PTS, VCS
  • area: main
  • in suites: jessie
  • size: 34,476 kB
  • sloc: ansic: 127,074; sh: 7,535; asm: 4,610; makefile: 1,186; yacc: 291; perl: 196; pascal: 72; sed: 16
file content (719 lines) | stat: -rw-r--r-- 20,359 bytes parent folder | download | duplicates (2)
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
/* mpi-pow.c  -  MPI functions
 * Copyright (C) 1994, 1996, 1998, 2000 Free Software Foundation, Inc.
 * Copyright (C) 2013 Werner Koch
 *
 * This file is part of GnuPG.
 *
 * GnuPG is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 3 of the License, or
 * (at your option) any later version.
 *
 * GnuPG is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, see <http://www.gnu.org/licenses/>.
 *
 * Note: This code is heavily based on the GNU MP Library.
 *	 Actually it's the same code with only minor changes in the
 *	 way the data is stored; this is to support the abstraction
 *	 of an optional secure memory allocation which may be used
 *	 to avoid revealing of sensitive data due to paging etc.
 *	 The GNU MP Library itself is published under the LGPL;
 *	 however I decided to publish this code under the plain GPL.
 */

#include <config.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "mpi-internal.h"
#include "longlong.h"
#include <assert.h>

/*
 * When you need old implementation, please add compilation option
 * -DUSE_ALGORITHM_SIMPLE_EXPONENTIATION
 * or expose this line:
#define USE_ALGORITHM_SIMPLE_EXPONENTIATION 1
 */

#if defined(USE_ALGORITHM_SIMPLE_EXPONENTIATION)
/****************
 * RES = BASE ^ EXP mod MOD
 */
void
mpi_powm( MPI res, MPI base, MPI exponent, MPI mod)
{
    mpi_ptr_t  rp, ep, mp, bp;
    mpi_size_t esize, msize, bsize, rsize;
    int               msign, bsign, rsign;
    int        esec,  msec,  bsec,  rsec;
    mpi_size_t size;
    int mod_shift_cnt;
    int negative_result;
    mpi_ptr_t mp_marker=NULL, bp_marker=NULL, ep_marker=NULL;
    mpi_ptr_t xp_marker=NULL;
    int assign_rp=0;
    mpi_ptr_t tspace = NULL;
    mpi_size_t tsize=0;   /* to avoid compiler warning */
			  /* fixme: we should check that the warning is void*/

    esize = exponent->nlimbs;
    msize = mod->nlimbs;
    size = 2 * msize;
    msign = mod->sign;

    esec = mpi_is_secure(exponent);
    msec = mpi_is_secure(mod);
    bsec = mpi_is_secure(base);
    rsec = mpi_is_secure(res);

    rp = res->d;
    ep = exponent->d;

    if( !msize )
	msize = 1 / msize;	    /* provoke a signal */

    if( !esize ) {
	/* Exponent is zero, result is 1 mod MOD, i.e., 1 or 0
	 * depending on if MOD equals 1.  */
	rp[0] = 1;
	res->nlimbs = (msize == 1 && mod->d[0] == 1) ? 0 : 1;
	res->sign = 0;
	goto leave;
    }

    /* Normalize MOD (i.e. make its most significant bit set) as required by
     * mpn_divrem.  This will make the intermediate values in the calculation
     * slightly larger, but the correct result is obtained after a final
     * reduction using the original MOD value.	*/
    mp = mp_marker = mpi_alloc_limb_space(msize, msec);
    count_leading_zeros( mod_shift_cnt, mod->d[msize-1] );
    if( mod_shift_cnt )
	mpihelp_lshift( mp, mod->d, msize, mod_shift_cnt );
    else
	MPN_COPY( mp, mod->d, msize );

    bsize = base->nlimbs;
    bsign = base->sign;
    if( bsize > msize ) { /* The base is larger than the module. Reduce it. */
	/* Allocate (BSIZE + 1) with space for remainder and quotient.
	 * (The quotient is (bsize - msize + 1) limbs.)  */
	bp = bp_marker = mpi_alloc_limb_space( bsize + 1, bsec );
	MPN_COPY( bp, base->d, bsize );
	/* We don't care about the quotient, store it above the remainder,
	 * at BP + MSIZE.  */
	mpihelp_divrem( bp + msize, 0, bp, bsize, mp, msize );
	bsize = msize;
	/* Canonicalize the base, since we are going to multiply with it
	 * quite a few times.  */
	MPN_NORMALIZE( bp, bsize );
    }
    else
	bp = base->d;

    if( !bsize ) {
	res->nlimbs = 0;
	res->sign = 0;
	goto leave;
    }

    if( res->alloced < size ) {
	/* We have to allocate more space for RES.  If any of the input
	 * parameters are identical to RES, defer deallocation of the old
	 * space.  */
	if( rp == ep || rp == mp || rp == bp ) {
	    rp = mpi_alloc_limb_space( size, rsec );
	    assign_rp = 1;
	}
	else {
	    mpi_resize( res, size );
	    rp = res->d;
	}
    }
    else { /* Make BASE, EXPONENT and MOD not overlap with RES.  */
	if( rp == bp ) {
	    /* RES and BASE are identical.  Allocate temp. space for BASE.  */
	    assert( !bp_marker );
	    bp = bp_marker = mpi_alloc_limb_space( bsize, bsec );
	    MPN_COPY(bp, rp, bsize);
	}
	if( rp == ep ) {
	    /* RES and EXPONENT are identical.
               Allocate temp. space for EXPONENT.  */
	    ep = ep_marker = mpi_alloc_limb_space( esize, esec );
	    MPN_COPY(ep, rp, esize);
	}
	if( rp == mp ) {
	    /* RES and MOD are identical.  Allocate temporary space for MOD.*/
	    assert( !mp_marker );
	    mp = mp_marker = mpi_alloc_limb_space( msize, msec );
	    MPN_COPY(mp, rp, msize);
	}
    }

    MPN_COPY( rp, bp, bsize );
    rsize = bsize;
    rsign = bsign;

    {
	mpi_size_t i;
	mpi_ptr_t xp = xp_marker = mpi_alloc_limb_space( size, msec );
	int c;
	mpi_limb_t e;
	mpi_limb_t carry_limb;
	struct karatsuba_ctx karactx;

	memset( &karactx, 0, sizeof karactx );
	negative_result = (ep[0] & 1) && base->sign;

	i = esize - 1;
	e = ep[i];
	count_leading_zeros (c, e);
	e = (e << c) << 1;     /* shift the exp bits to the left, lose msb */
	c = BITS_PER_MPI_LIMB - 1 - c;

	/* Main loop.
	 *
	 * Make the result be pointed to alternately by XP and RP.  This
	 * helps us avoid block copying, which would otherwise be necessary
	 * with the overlap restrictions of mpihelp_divmod. With 50% probability
	 * the result after this loop will be in the area originally pointed
	 * by RP (==RES->d), and with 50% probability in the area originally
	 * pointed to by XP.
	 */

	for(;;) {
	    while( c ) {
		mpi_ptr_t tp;
		mpi_size_t xsize;

		/*mpihelp_mul_n(xp, rp, rp, rsize);*/
		if( rsize < KARATSUBA_THRESHOLD )
		    mpih_sqr_n_basecase( xp, rp, rsize );
		else {
		    if( !tspace ) {
			tsize = 2 * rsize;
			tspace = mpi_alloc_limb_space( tsize, 0 );
		    }
		    else if( tsize < (2*rsize) ) {
			mpi_free_limb_space( tspace );
			tsize = 2 * rsize;
			tspace = mpi_alloc_limb_space( tsize, 0 );
		    }
		    mpih_sqr_n( xp, rp, rsize, tspace );
		}

		xsize = 2 * rsize;
		if( xsize > msize ) {
		    mpihelp_divrem(xp + msize, 0, xp, xsize, mp, msize);
		    xsize = msize;
		}

		tp = rp; rp = xp; xp = tp;
		rsize = xsize;

                /* To mitigate the Yarom/Falkner flush+reload cache
                 * side-channel attack on the RSA secret exponent, we
                 * do the multiplication regardless of the value of
                 * the high-bit of E.  But to avoid this performance
                 * penalty we do it only if the exponent has been
                 * stored in secure memory and we can thus assume it
                 * is a secret exponent.  */
                if (esec || (mpi_limb_signed_t)e < 0) {
		    /*mpihelp_mul( xp, rp, rsize, bp, bsize );*/
		    if( bsize < KARATSUBA_THRESHOLD ) {
			mpihelp_mul( xp, rp, rsize, bp, bsize );
		    }
		    else {
			mpihelp_mul_karatsuba_case(
				     xp, rp, rsize, bp, bsize, &karactx );
		    }

		    xsize = rsize + bsize;
		    if( xsize > msize ) {
			mpihelp_divrem(xp + msize, 0, xp, xsize, mp, msize);
			xsize = msize;
		    }
                }
		if ((mpi_limb_signed_t)e < 0) {
		    tp = rp; rp = xp; xp = tp;
		    rsize = xsize;
		}
		e <<= 1;
		c--;
	    }

	    i--;
	    if( i < 0 )
		break;
	    e = ep[i];
	    c = BITS_PER_MPI_LIMB;
	}

	/* We shifted MOD, the modulo reduction argument, left MOD_SHIFT_CNT
	 * steps.  Adjust the result by reducing it with the original MOD.
	 *
	 * Also make sure the result is put in RES->d (where it already
	 * might be, see above).
	 */
	if( mod_shift_cnt ) {
	    carry_limb = mpihelp_lshift( res->d, rp, rsize, mod_shift_cnt);
	    rp = res->d;
	    if( carry_limb ) {
		rp[rsize] = carry_limb;
		rsize++;
	    }
	}
	else {
	    MPN_COPY( res->d, rp, rsize);
	    rp = res->d;
	}

	if( rsize >= msize ) {
	    mpihelp_divrem(rp + msize, 0, rp, rsize, mp, msize);
	    rsize = msize;
	}

	/* Remove any leading zero words from the result.  */
	if( mod_shift_cnt )
	    mpihelp_rshift( rp, rp, rsize, mod_shift_cnt);
	MPN_NORMALIZE (rp, rsize);

	mpihelp_release_karatsuba_ctx( &karactx );
    }

    if( negative_result && rsize ) {
	if( mod_shift_cnt )
	    mpihelp_rshift( mp, mp, msize, mod_shift_cnt);
	mpihelp_sub( rp, mp, msize, rp, rsize);
	rsize = msize;
	rsign = msign;
	MPN_NORMALIZE(rp, rsize);
    }
    res->nlimbs = rsize;
    res->sign = rsign;

  leave:
    if( assign_rp ) mpi_assign_limb_space( res, rp, size );
    if( mp_marker ) mpi_free_limb_space( mp_marker );
    if( bp_marker ) mpi_free_limb_space( bp_marker );
    if( ep_marker ) mpi_free_limb_space( ep_marker );
    if( xp_marker ) mpi_free_limb_space( xp_marker );
    if( tspace )    mpi_free_limb_space( tspace );
}
#else /*!USE_ALGORITHM_SIMPLE_EXPONENTIATION */

/**
 * Internal function to compute
 *
 *    X = R * S mod M
 *
 * and set the size of X at the pointer XSIZE_P.
 * Use karatsuba structure at KARACTX_P.
 *
 * Condition:
 *   RSIZE >= SSIZE
 *   Enough space for X is allocated beforehand.
 *
 * For generic cases, we can/should use mpi_mulm.
 * This function is use for specific internal case.
 */
static void
mul_mod (mpi_ptr_t xp, mpi_size_t *xsize_p,
         mpi_ptr_t rp, mpi_size_t rsize,
         mpi_ptr_t sp, mpi_size_t ssize,
         mpi_ptr_t mp, mpi_size_t msize,
         struct karatsuba_ctx *karactx_p)
{
  if( ssize < KARATSUBA_THRESHOLD )
    mpihelp_mul ( xp, rp, rsize, sp, ssize );
  else
    mpihelp_mul_karatsuba_case (xp, rp, rsize, sp, ssize, karactx_p);

   if (rsize + ssize > msize)
    {
      mpihelp_divrem (xp + msize, 0, xp, rsize + ssize, mp, msize);
      *xsize_p = msize;
    }
   else
     *xsize_p = rsize + ssize;
}

#define SIZE_PRECOMP ((1 << (5 - 1)))

/****************
 * RES = BASE ^ EXPO mod MOD
 *
 * To mitigate the Yarom/Falkner flush+reload cache side-channel
 * attack on the RSA secret exponent, we don't use the square
 * routine but multiplication.
 *
 * Reference:
 *   Handbook of Applied Cryptography
 *       Algorithm 14.83: Modified left-to-right k-ary exponentiation
 */
void
mpi_powm (MPI res, MPI base, MPI expo, MPI mod)
{
  /* Pointer to the limbs of the arguments, their size and signs. */
  mpi_ptr_t  rp, ep, mp, bp;
  mpi_size_t esize, msize, bsize, rsize;
  int               msign, bsign, rsign;
  /* Flags telling the secure allocation status of the arguments.  */
  int        esec,  msec,  bsec;
  /* Size of the result including space for temporary values.  */
  mpi_size_t size;
  /* Helper.  */
  int mod_shift_cnt;
  int negative_result;
  mpi_ptr_t mp_marker = NULL;
  mpi_ptr_t bp_marker = NULL;
  mpi_ptr_t ep_marker = NULL;
  mpi_ptr_t xp_marker = NULL;
  mpi_ptr_t precomp[SIZE_PRECOMP]; /* Pre-computed array: BASE^1, ^3, ^5, ... */
  mpi_size_t precomp_size[SIZE_PRECOMP];
  mpi_size_t W;
  mpi_ptr_t base_u;
  mpi_size_t base_u_size;
  mpi_size_t max_u_size;

  esize = expo->nlimbs;
  msize = mod->nlimbs;
  size = 2 * msize;
  msign = mod->sign;

  ep = expo->d;
  MPN_NORMALIZE(ep, esize);

  if (esize * BITS_PER_MPI_LIMB > 512)
    W = 5;
  else if (esize * BITS_PER_MPI_LIMB > 256)
    W = 4;
  else if (esize * BITS_PER_MPI_LIMB > 128)
    W = 3;
  else if (esize * BITS_PER_MPI_LIMB > 64)
    W = 2;
  else
    W = 1;

  esec = mpi_is_secure(expo);
  msec = mpi_is_secure(mod);
  bsec = mpi_is_secure(base);

  rp = res->d;

  if (!msize)
    msize = 1 / msize;	    /* provoke a signal */

  if (!esize)
    {
      /* Exponent is zero, result is 1 mod MOD, i.e., 1 or 0 depending
         on if MOD equals 1.  */
      res->nlimbs = (msize == 1 && mod->d[0] == 1) ? 0 : 1;
      if (res->nlimbs)
        {
          RESIZE_IF_NEEDED (res, 1);
          rp = res->d;
          rp[0] = 1;
        }
      res->sign = 0;
      goto leave;
    }

  /* Normalize MOD (i.e. make its most significant bit set) as
     required by mpn_divrem.  This will make the intermediate values
     in the calculation slightly larger, but the correct result is
     obtained after a final reduction using the original MOD value. */
  mp = mp_marker = mpi_alloc_limb_space(msize, msec);
  count_leading_zeros (mod_shift_cnt, mod->d[msize-1]);
  if (mod_shift_cnt)
    mpihelp_lshift (mp, mod->d, msize, mod_shift_cnt);
  else
    MPN_COPY( mp, mod->d, msize );

  bsize = base->nlimbs;
  bsign = base->sign;
  if (bsize > msize)
    {
      /* The base is larger than the module.  Reduce it.

         Allocate (BSIZE + 1) with space for remainder and quotient.
         (The quotient is (bsize - msize + 1) limbs.)  */
      bp = bp_marker = mpi_alloc_limb_space( bsize + 1, bsec );
      MPN_COPY ( bp, base->d, bsize );
      /* We don't care about the quotient, store it above the
       * remainder, at BP + MSIZE.  */
      mpihelp_divrem( bp + msize, 0, bp, bsize, mp, msize );
      bsize = msize;
      /* Canonicalize the base, since we are going to multiply with it
         quite a few times.  */
      MPN_NORMALIZE( bp, bsize );
    }
  else
    bp = base->d;

  if (!bsize)
    {
      res->nlimbs = 0;
      res->sign = 0;
      goto leave;
    }


  /* Make BASE, EXPO not overlap with RES.  We don't need to check MOD
     because that has already been copied to the MP var.  */
  if ( rp == bp )
    {
      /* RES and BASE are identical.  Allocate temp. space for BASE.  */
      assert (!bp_marker);
      bp = bp_marker = mpi_alloc_limb_space( bsize, bsec );
      MPN_COPY(bp, rp, bsize);
    }
  if ( rp == ep )
    {
      /* RES and EXPO are identical.  Allocate temp. space for EXPO.  */
      ep = ep_marker = mpi_alloc_limb_space( esize, esec );
      MPN_COPY(ep, rp, esize);
    }

  /* Copy base to the result.  */
  if (res->alloced < size)
    {
      mpi_resize (res, size);
      rp = res->d;
    }

  /* Main processing.  */
  {
    mpi_size_t i, j, k;
    mpi_ptr_t xp;
    mpi_size_t xsize;
    int c;
    mpi_limb_t e;
    mpi_limb_t carry_limb;
    struct karatsuba_ctx karactx;
    mpi_ptr_t tp;

    xp = xp_marker = mpi_alloc_limb_space( size, msec );

    memset( &karactx, 0, sizeof karactx );
    negative_result = (ep[0] & 1) && bsign;

    /* Precompute PRECOMP[], BASE^(2 * i + 1), BASE^1, ^3, ^5, ... */
    if (W > 1)                  /* X := BASE^2 */
      mul_mod (xp, &xsize, bp, bsize, bp, bsize, mp, msize, &karactx);
    base_u = precomp[0] = mpi_alloc_limb_space (bsize, esec);
    base_u_size = max_u_size = precomp_size[0] = bsize;
    MPN_COPY (precomp[0], bp, bsize);
    for (i = 1; i < (1 << (W - 1)); i++)
      {                         /* PRECOMP[i] = BASE^(2 * i + 1) */
        if (xsize >= base_u_size)
          mul_mod (rp, &rsize, xp, xsize, base_u, base_u_size,
                   mp, msize, &karactx);
        else
          mul_mod (rp, &rsize, base_u, base_u_size, xp, xsize,
                   mp, msize, &karactx);
        base_u = precomp[i] = mpi_alloc_limb_space (rsize, esec);
        base_u_size = precomp_size[i] = rsize;
        if (max_u_size < base_u_size)
          max_u_size = base_u_size;
        MPN_COPY (precomp[i], rp, rsize);
      }

    if (msize > max_u_size)
      max_u_size = msize;
    base_u = mpi_alloc_limb_space (max_u_size, esec);
    MPN_ZERO (base_u, max_u_size);

    i = esize - 1;

    /* Main loop.

       Make the result be pointed to alternately by XP and RP.  This
       helps us avoid block copying, which would otherwise be
       necessary with the overlap restrictions of mpihelp_divmod. With
       50% probability the result after this loop will be in the area
       originally pointed by RP (==RES->d), and with 50% probability
       in the area originally pointed to by XP. */
    rsign = 0;
    if (W == 1)
      {
        rsize = bsize;
      }
    else
      {
        rsize = msize;
        MPN_ZERO (rp, rsize);
      }
    MPN_COPY ( rp, bp, bsize );

    e = ep[i];
    count_leading_zeros (c, e);
    e = (e << c) << 1;
    c = BITS_PER_MPI_LIMB - 1 - c;

    j = 0;

    for (;;)
      if (e == 0)
        {
          j += c;
          if ( --i < 0 )
            break;

          e = ep[i];
          c = BITS_PER_MPI_LIMB;
        }
      else
        {
          int c0;
          mpi_limb_t e0;
          struct gcry_mpi w, u;
          w.sign = u.sign = 0;
          w.flags = u.flags = 0;
          w.d = base_u;

          count_leading_zeros (c0, e);
          e = (e << c0);
          c -= c0;
          j += c0;

          e0 = (e >> (BITS_PER_MPI_LIMB - W));
          if (c >= W)
            c0 = 0;
          else
            {
              if ( --i < 0 )
                {
                  e0 = (e >> (BITS_PER_MPI_LIMB - c));
                  j += c - W;
                  goto last_step;
                }
              else
                {
                  c0 = c;
                  e = ep[i];
                  c = BITS_PER_MPI_LIMB;
                  e0 |= (e >> (BITS_PER_MPI_LIMB - (W - c0)));
		}
            }

          e = e << (W - c0);
          c -= (W - c0);

        last_step:
          count_trailing_zeros (c0, e0);
          e0 = (e0 >> c0) >> 1;

          for (j += W - c0; j >= 0; j--)
            {

              /*
               *  base_u <= precomp[e0]
               *  base_u_size <= precomp_size[e0]
               */
              base_u_size = 0;
              for (k = 0; k < (1<< (W - 1)); k++)
                {
                  w.alloced = w.nlimbs = precomp_size[k];
                  u.alloced = u.nlimbs = precomp_size[k];
                  u.d = precomp[k];

                  mpi_set_cond (&w, &u, k == e0);
                  base_u_size |= ( precomp_size[k] & (0UL - (k == e0)) );
                }

              w.alloced = w.nlimbs = rsize;
              u.alloced = u.nlimbs = rsize;
              u.d = rp;
              mpi_set_cond (&w, &u, j != 0);
              base_u_size ^= ((base_u_size ^ rsize)  & (0UL - (j != 0)));

              mul_mod (xp, &xsize, rp, rsize, base_u, base_u_size,
                       mp, msize, &karactx);
              tp = rp; rp = xp; xp = tp;
              rsize = xsize;
            }

          j = c0;
          if ( i < 0 )
            break;
        }

    while (j--)
      {
        mul_mod (xp, &xsize, rp, rsize, rp, rsize, mp, msize, &karactx);
        tp = rp; rp = xp; xp = tp;
        rsize = xsize;
      }

    /* We shifted MOD, the modulo reduction argument, left
       MOD_SHIFT_CNT steps.  Adjust the result by reducing it with the
       original MOD.

       Also make sure the result is put in RES->d (where it already
       might be, see above).  */
    if ( mod_shift_cnt )
      {
        carry_limb = mpihelp_lshift( res->d, rp, rsize, mod_shift_cnt);
        rp = res->d;
        if ( carry_limb )
          {
            rp[rsize] = carry_limb;
            rsize++;
          }
      }
    else if (res->d != rp)
      {
        MPN_COPY (res->d, rp, rsize);
        rp = res->d;
      }

    if ( rsize >= msize )
      {
        mpihelp_divrem(rp + msize, 0, rp, rsize, mp, msize);
        rsize = msize;
      }

    /* Remove any leading zero words from the result.  */
    if ( mod_shift_cnt )
      mpihelp_rshift (rp, rp, rsize, mod_shift_cnt);
    MPN_NORMALIZE (rp, rsize);

    mpihelp_release_karatsuba_ctx (&karactx );
    for (i = 0; i < (1 << (W - 1)); i++)
      mpi_free_limb_space (precomp[i]);
    mpi_free_limb_space (base_u);
  }

  /* Fixup for negative results.  */
  if ( negative_result && rsize )
    {
      if ( mod_shift_cnt )
        mpihelp_rshift (mp, mp, msize, mod_shift_cnt);
      mpihelp_sub (rp, mp, msize, rp, rsize);
      rsize = msize;
      rsign = msign;
      MPN_NORMALIZE(rp, rsize);
    }
  assert (res->d == rp);
  res->nlimbs = rsize;
  res->sign = rsign;

 leave:
  if (mp_marker)
    mpi_free_limb_space (mp_marker);
  if (bp_marker)
    mpi_free_limb_space (bp_marker);
  if (ep_marker)
    mpi_free_limb_space (ep_marker);
  if (xp_marker)
    mpi_free_limb_space (xp_marker);
}
#endif /*!USE_ALGORITHM_SIMPLE_EXPONENTIATION */