• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *  Licensed to the Apache Software Foundation (ASF) under one or more
3  *  contributor license agreements.  See the NOTICE file distributed with
4  *  this work for additional information regarding copyright ownership.
5  *  The ASF licenses this file to You under the Apache License, Version 2.0
6  *  (the "License"); you may not use this file except in compliance with
7  *  the License.  You may obtain a copy of the License at
8  *
9  *     http://www.apache.org/licenses/LICENSE-2.0
10  *
11  *  Unless required by applicable law or agreed to in writing, software
12  *  distributed under the License is distributed on an "AS IS" BASIS,
13  *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14  *  See the License for the specific language governing permissions and
15  *  limitations under the License.
16  */
17 
18 #include <string.h>
19 #include "cbigint.h"
20 
21 #define HIGH_IN_U64(u64) ((u64) >> 32)
22 #define LOW_IN_U64(u64) ((u64) & 0x00000000FFFFFFFFLL)
23 
24 #define TEN_E1 (0xALL)
25 #define TEN_E2 (0x64LL)
26 #define TEN_E3 (0x3E8LL)
27 #define TEN_E4 (0x2710LL)
28 #define TEN_E5 (0x186A0LL)
29 #define TEN_E6 (0xF4240LL)
30 #define TEN_E7 (0x989680LL)
31 #define TEN_E8 (0x5F5E100LL)
32 #define TEN_E9 (0x3B9ACA00LL)
33 #define TEN_E19 (0x8AC7230489E80000LL)
34 
35 #define TIMES_TEN(x) (((x) << 3) + ((x) << 1))
36 #define bitSection(x, mask, shift) (((x) & (mask)) >> (shift))
37 #define CREATE_DOUBLE_BITS(normalizedM, e) (((normalizedM) & MANTISSA_MASK) | ((static_cast<uint64_t>((e) + E_OFFSET)) << 52))
38 
39 #define MANTISSA_MASK (0x000FFFFFFFFFFFFFLL)
40 #define EXPONENT_MASK (0x7FF0000000000000LL)
41 #define NORMAL_MASK (0x0010000000000000LL)
42 #define SIGN_MASK (0x8000000000000000LL)
43 
44 #define E_OFFSET (1075)
45 
46 #define FLOAT_MANTISSA_MASK (0x007FFFFF)
47 #define FLOAT_EXPONENT_MASK (0x7F800000)
48 #define FLOAT_NORMAL_MASK   (0x00800000)
49 #define FLOAT_E_OFFSET (150)
50 
51 int32_t
simpleAddHighPrecision(uint64_t * arg1,int32_t length,uint64_t arg2)52 simpleAddHighPrecision (uint64_t * arg1, int32_t length, uint64_t arg2)
53 {
54   /* assumes length > 0 */
55   int32_t index = 1;
56 
57   *arg1 += arg2;
58   if (arg2 <= *arg1)
59     return 0;
60   else if (length == 1)
61     return 1;
62 
63   while (++arg1[index] == 0 && ++index < length) {
64   }
65 
66   return index == length;
67 }
68 
69 int32_t
addHighPrecision(uint64_t * arg1,int32_t length1,uint64_t * arg2,int32_t length2)70 addHighPrecision (uint64_t * arg1, int32_t length1, uint64_t * arg2, int32_t length2)
71 {
72   /* addition is limited by length of arg1 as it this function is
73    * storing the result in arg1 */
74   /* fix for cc (GCC) 3.2 20020903 (Red Hat Linux 8.0 3.2-7): code generated does not
75    * do the temp1 + temp2 + carry addition correct.  carry is 64 bit because gcc has
76    * subtle issues when you mix 64 / 32 bit maths. */
77   uint64_t temp1, temp2, temp3;     /* temporary variables to help the SH-4, and gcc */
78   uint64_t carry;
79   int32_t index;
80 
81   if (length1 == 0 || length2 == 0)
82     {
83       return 0;
84     }
85   else if (length1 < length2)
86     {
87       length2 = length1;
88     }
89 
90   carry = 0;
91   index = 0;
92   do
93     {
94       temp1 = arg1[index];
95       temp2 = arg2[index];
96       temp3 = temp1 + temp2;
97       arg1[index] = temp3 + carry;
98       if (arg2[index] < arg1[index])
99         carry = 0;
100       else if (arg2[index] != arg1[index])
101         carry = 1;
102     }
103   while (++index < length2);
104   if (!carry)
105     return 0;
106   else if (index == length1)
107     return 1;
108 
109   while (++arg1[index] == 0 && ++index < length1) {
110   }
111 
112   return index == length1;
113 }
114 
115 void
subtractHighPrecision(uint64_t * arg1,int32_t length1,uint64_t * arg2,int32_t length2)116 subtractHighPrecision (uint64_t * arg1, int32_t length1, uint64_t * arg2, int32_t length2)
117 {
118   /* assumes arg1 > arg2 */
119   int32_t index;
120   for (index = 0; index < length1; ++index)
121     arg1[index] = ~arg1[index];
122   simpleAddHighPrecision (arg1, length1, 1);
123 
124   while (length2 > 0 && arg2[length2 - 1] == 0)
125     --length2;
126 
127   addHighPrecision (arg1, length1, arg2, length2);
128 
129   for (index = 0; index < length1; ++index)
130     arg1[index] = ~arg1[index];
131   simpleAddHighPrecision (arg1, length1, 1);
132 }
133 
simpleMultiplyHighPrecision(uint64_t * arg1,int32_t length,uint64_t arg2)134 static uint32_t simpleMultiplyHighPrecision(uint64_t* arg1, int32_t length, uint64_t arg2) {
135   /* assumes arg2 only holds 32 bits of information */
136   uint64_t product;
137   int32_t index;
138 
139   index = 0;
140   product = 0;
141 
142   do
143     {
144       product =
145         HIGH_IN_U64 (product) + arg2 * LOW_U32_FROM_PTR (arg1 + index);
146       LOW_U32_FROM_PTR (arg1 + index) = LOW_U32_FROM_VAR (product);
147       product =
148         HIGH_IN_U64 (product) + arg2 * HIGH_U32_FROM_PTR (arg1 + index);
149       HIGH_U32_FROM_PTR (arg1 + index) = LOW_U32_FROM_VAR (product);
150     }
151   while (++index < length);
152 
153   return HIGH_U32_FROM_VAR (product);
154 }
155 
156 static void
simpleMultiplyAddHighPrecision(uint64_t * arg1,int32_t length,uint64_t arg2,uint32_t * result)157 simpleMultiplyAddHighPrecision (uint64_t * arg1, int32_t length, uint64_t arg2,
158                                 uint32_t * result)
159 {
160   /* Assumes result can hold the product and arg2 only holds 32 bits
161      of information */
162   uint64_t product;
163   int32_t index, resultIndex;
164 
165   index = resultIndex = 0;
166   product = 0;
167 
168   do
169     {
170       product =
171         HIGH_IN_U64 (product) + result[resultIndex] +
172         arg2 * LOW_U32_FROM_PTR (arg1 + index);
173       result[resultIndex] = LOW_U32_FROM_VAR (product);
174       ++resultIndex;
175       product =
176         HIGH_IN_U64 (product) + result[resultIndex] +
177         arg2 * HIGH_U32_FROM_PTR (arg1 + index);
178       result[resultIndex] = LOW_U32_FROM_VAR (product);
179       ++resultIndex;
180     }
181   while (++index < length);
182 
183   result[resultIndex] += HIGH_U32_FROM_VAR (product);
184   if (result[resultIndex] < HIGH_U32_FROM_VAR (product))
185     {
186       /* must be careful with ++ operator and macro expansion */
187       ++resultIndex;
188       while (++result[resultIndex] == 0)
189         ++resultIndex;
190     }
191 }
192 
193 void
multiplyHighPrecision(uint64_t * arg1,int32_t length1,uint64_t * arg2,int32_t length2,uint64_t * result,int32_t length)194 multiplyHighPrecision (uint64_t * arg1, int32_t length1, uint64_t * arg2, int32_t length2,
195                        uint64_t * result, int32_t length)
196 {
197   /* assumes result is large enough to hold product */
198   uint64_t* temp;
199   uint32_t* resultIn32;
200   int32_t count, index;
201 
202   if (length1 < length2)
203     {
204       temp = arg1;
205       arg1 = arg2;
206       arg2 = temp;
207       count = length1;
208       length1 = length2;
209       length2 = count;
210     }
211 
212   memset (result, 0, sizeof (uint64_t) * length);
213 
214   /* length1 > length2 */
215   resultIn32 = reinterpret_cast<uint32_t*>(result);
216   index = -1;
217   for (count = 0; count < length2; ++count)
218     {
219       simpleMultiplyAddHighPrecision (arg1, length1, LOW_IN_U64 (arg2[count]),
220                                       resultIn32 + (++index));
221       simpleMultiplyAddHighPrecision(arg1, length1, HIGH_IN_U64(arg2[count]), resultIn32 + (++index));
222     }
223 }
224 
225 uint32_t
simpleAppendDecimalDigitHighPrecision(uint64_t * arg1,int32_t length,uint64_t digit)226 simpleAppendDecimalDigitHighPrecision (uint64_t * arg1, int32_t length, uint64_t digit)
227 {
228   /* assumes digit is less than 32 bits */
229   uint64_t arg;
230   int32_t index = 0;
231 
232   digit <<= 32;
233   do
234     {
235       arg = LOW_IN_U64 (arg1[index]);
236       digit = HIGH_IN_U64 (digit) + TIMES_TEN (arg);
237       LOW_U32_FROM_PTR (arg1 + index) = LOW_U32_FROM_VAR (digit);
238 
239       arg = HIGH_IN_U64 (arg1[index]);
240       digit = HIGH_IN_U64 (digit) + TIMES_TEN (arg);
241       HIGH_U32_FROM_PTR (arg1 + index) = LOW_U32_FROM_VAR (digit);
242     }
243   while (++index < length);
244 
245   return HIGH_U32_FROM_VAR (digit);
246 }
247 
248 void
simpleShiftLeftHighPrecision(uint64_t * arg1,int32_t length,int32_t arg2)249 simpleShiftLeftHighPrecision (uint64_t * arg1, int32_t length, int32_t arg2)
250 {
251   /* assumes length > 0 */
252   int32_t index, offset;
253   if (arg2 >= 64)
254     {
255       offset = arg2 >> 6;
256       index = length;
257 
258       while (--index - offset >= 0)
259         arg1[index] = arg1[index - offset];
260       do
261         {
262           arg1[index] = 0;
263         }
264       while (--index >= 0);
265 
266       arg2 &= 0x3F;
267     }
268 
269   if (arg2 == 0)
270     return;
271   while (--length > 0)
272     {
273       arg1[length] = arg1[length] << arg2 | arg1[length - 1] >> (64 - arg2);
274     }
275   *arg1 <<= arg2;
276 }
277 
278 int32_t
highestSetBit(uint64_t * y)279 highestSetBit (uint64_t * y)
280 {
281   uint32_t x;
282   int32_t result;
283 
284   if (*y == 0)
285     return 0;
286 
287   if (*y & 0xFFFFFFFF00000000LL)
288     {
289       x = HIGH_U32_FROM_PTR (y);
290       result = 32;
291     }
292   else
293     {
294       x = LOW_U32_FROM_PTR (y);
295       result = 0;
296     }
297 
298   if (x & 0xFFFF0000)
299     {
300       x = bitSection (x, 0xFFFF0000, 16);
301       result += 16;
302     }
303   if (x & 0xFF00)
304     {
305       x = bitSection (x, 0xFF00, 8);
306       result += 8;
307     }
308   if (x & 0xF0)
309     {
310       x = bitSection (x, 0xF0, 4);
311       result += 4;
312     }
313   if (x > 0x7)
314     return result + 4;
315   else if (x > 0x3)
316     return result + 3;
317   else if (x > 0x1)
318     return result + 2;
319   else
320     return result + 1;
321 }
322 
323 int32_t
lowestSetBit(uint64_t * y)324 lowestSetBit (uint64_t * y)
325 {
326   uint32_t x;
327   int32_t result;
328 
329   if (*y == 0)
330     return 0;
331 
332   if (*y & 0x00000000FFFFFFFFLL)
333     {
334       x = LOW_U32_FROM_PTR (y);
335       result = 0;
336     }
337   else
338     {
339       x = HIGH_U32_FROM_PTR (y);
340       result = 32;
341     }
342 
343   if (!(x & 0xFFFF))
344     {
345       x = bitSection (x, 0xFFFF0000, 16);
346       result += 16;
347     }
348   if (!(x & 0xFF))
349     {
350       x = bitSection (x, 0xFF00, 8);
351       result += 8;
352     }
353   if (!(x & 0xF))
354     {
355       x = bitSection (x, 0xF0, 4);
356       result += 4;
357     }
358 
359   if (x & 0x1)
360     return result + 1;
361   else if (x & 0x2)
362     return result + 2;
363   else if (x & 0x4)
364     return result + 3;
365   else
366     return result + 4;
367 }
368 
369 int32_t
highestSetBitHighPrecision(uint64_t * arg,int32_t length)370 highestSetBitHighPrecision (uint64_t * arg, int32_t length)
371 {
372   int32_t highBit;
373 
374   while (--length >= 0)
375     {
376       highBit = highestSetBit (arg + length);
377       if (highBit)
378         return highBit + 64 * length;
379     }
380 
381   return 0;
382 }
383 
384 int32_t
lowestSetBitHighPrecision(uint64_t * arg,int32_t length)385 lowestSetBitHighPrecision (uint64_t * arg, int32_t length)
386 {
387   int32_t lowBit, index = -1;
388 
389   while (++index < length)
390     {
391       lowBit = lowestSetBit (arg + index);
392       if (lowBit)
393         return lowBit + 64 * index;
394     }
395 
396   return 0;
397 }
398 
399 int32_t
compareHighPrecision(uint64_t * arg1,int32_t length1,uint64_t * arg2,int32_t length2)400 compareHighPrecision (uint64_t * arg1, int32_t length1, uint64_t * arg2, int32_t length2)
401 {
402   while (--length1 >= 0 && arg1[length1] == 0) {
403   }
404   while (--length2 >= 0 && arg2[length2] == 0) {
405   }
406 
407   if (length1 > length2)
408     return 1;
409   else if (length1 < length2)
410     return -1;
411   else if (length1 > -1)
412     {
413       do
414         {
415           if (arg1[length1] > arg2[length1])
416             return 1;
417           else if (arg1[length1] < arg2[length1])
418             return -1;
419         }
420       while (--length1 >= 0);
421     }
422 
423   return 0;
424 }
425 
426 jdouble
toDoubleHighPrecision(uint64_t * arg,int32_t length)427 toDoubleHighPrecision (uint64_t * arg, int32_t length)
428 {
429   int32_t highBit;
430   uint64_t mantissa, test64;
431   uint32_t test;
432   jdouble result;
433 
434   while (length > 0 && arg[length - 1] == 0)
435     --length;
436 
437   if (length == 0)
438     result = 0.0;
439   else if (length > 16)
440     {
441       DOUBLE_TO_LONGBITS (result) = EXPONENT_MASK;
442     }
443   else if (length == 1)
444     {
445       highBit = highestSetBit (arg);
446       if (highBit <= 53)
447         {
448           highBit = 53 - highBit;
449           mantissa = *arg << highBit;
450           DOUBLE_TO_LONGBITS (result) =
451             CREATE_DOUBLE_BITS (mantissa, -highBit);
452         }
453       else
454         {
455           highBit -= 53;
456           mantissa = *arg >> highBit;
457           DOUBLE_TO_LONGBITS (result) =
458             CREATE_DOUBLE_BITS (mantissa, highBit);
459 
460           /* perform rounding, round to even in case of tie */
461           test = (LOW_U32_FROM_PTR (arg) << (11 - highBit)) & 0x7FF;
462           if (test > 0x400 || ((test == 0x400) && (mantissa & 1)))
463             DOUBLE_TO_LONGBITS (result) = DOUBLE_TO_LONGBITS (result) + 1;
464         }
465     }
466   else
467     {
468       highBit = highestSetBit (arg + (--length));
469       if (highBit <= 53)
470         {
471           highBit = 53 - highBit;
472           if (highBit > 0)
473             {
474               mantissa =
475                 (arg[length] << highBit) | (arg[length - 1] >>
476                                             (64 - highBit));
477             }
478           else
479             {
480               mantissa = arg[length];
481             }
482           DOUBLE_TO_LONGBITS (result) =
483             CREATE_DOUBLE_BITS (mantissa, length * 64 - highBit);
484 
485           /* perform rounding, round to even in case of tie */
486           test64 = arg[--length] << highBit;
487           if (test64 > SIGN_MASK || ((test64 == SIGN_MASK) && (mantissa & 1)))
488             DOUBLE_TO_LONGBITS (result) = DOUBLE_TO_LONGBITS (result) + 1;
489           else if (test64 == SIGN_MASK)
490             {
491               while (--length >= 0)
492                 {
493                   if (arg[length] != 0)
494                     {
495                       DOUBLE_TO_LONGBITS (result) =
496                         DOUBLE_TO_LONGBITS (result) + 1;
497                       break;
498                     }
499                 }
500             }
501         }
502       else
503         {
504           highBit -= 53;
505           mantissa = arg[length] >> highBit;
506           DOUBLE_TO_LONGBITS (result) =
507             CREATE_DOUBLE_BITS (mantissa, length * 64 + highBit);
508 
509           /* perform rounding, round to even in case of tie */
510           test = (LOW_U32_FROM_PTR (arg + length) << (11 - highBit)) & 0x7FF;
511           if (test > 0x400 || ((test == 0x400) && (mantissa & 1)))
512             DOUBLE_TO_LONGBITS (result) = DOUBLE_TO_LONGBITS (result) + 1;
513           else if (test == 0x400)
514             {
515               do
516                 {
517                   if (arg[--length] != 0)
518                     {
519                       DOUBLE_TO_LONGBITS (result) =
520                         DOUBLE_TO_LONGBITS (result) + 1;
521                       break;
522                     }
523                 }
524               while (length > 0);
525             }
526         }
527     }
528 
529   return result;
530 }
531 
532 static uint64_t simpleMultiplyHighPrecision64(uint64_t* arg1, int32_t length, uint64_t arg2);
533 
534 int32_t
timesTenToTheEHighPrecision(uint64_t * result,int32_t length,jint e)535 timesTenToTheEHighPrecision (uint64_t * result, int32_t length, jint e)
536 {
537   /* assumes result can hold value */
538   uint64_t overflow;
539   int exp10 = e;
540 
541   if (e == 0)
542     return length;
543 
544   /* bad O(n) way of doing it, but simple */
545   /*
546      do {
547      overflow = simpleAppendDecimalDigitHighPrecision(result, length, 0);
548      if (overflow)
549      result[length++] = overflow;
550      } while (--e);
551    */
552   /* Replace the current implementaion which performs a
553    * "multiplication" by 10 e number of times with an actual
554    * multiplication. 10e19 is the largest exponent to the power of ten
555    * that will fit in a 64-bit integer, and 10e9 is the largest exponent to
556    * the power of ten that will fit in a 64-bit integer. Not sure where the
557    * break-even point is between an actual multiplication and a
558    * simpleAappendDecimalDigit() so just pick 10e3 as that point for
559    * now.
560    */
561   while (exp10 >= 19)
562     {
563       overflow = simpleMultiplyHighPrecision64 (result, length, TEN_E19);
564       if (overflow)
565         result[length++] = overflow;
566       exp10 -= 19;
567     }
568   while (exp10 >= 9)
569     {
570       overflow = simpleMultiplyHighPrecision (result, length, TEN_E9);
571       if (overflow)
572         result[length++] = overflow;
573       exp10 -= 9;
574     }
575   if (exp10 == 0)
576     return length;
577   else if (exp10 == 1)
578     {
579       overflow = simpleAppendDecimalDigitHighPrecision (result, length, 0);
580       if (overflow)
581         result[length++] = overflow;
582     }
583   else if (exp10 == 2)
584     {
585       overflow = simpleAppendDecimalDigitHighPrecision (result, length, 0);
586       if (overflow)
587         result[length++] = overflow;
588       overflow = simpleAppendDecimalDigitHighPrecision (result, length, 0);
589       if (overflow)
590         result[length++] = overflow;
591     }
592   else if (exp10 == 3)
593     {
594       overflow = simpleMultiplyHighPrecision (result, length, TEN_E3);
595       if (overflow)
596         result[length++] = overflow;
597     }
598   else if (exp10 == 4)
599     {
600       overflow = simpleMultiplyHighPrecision (result, length, TEN_E4);
601       if (overflow)
602         result[length++] = overflow;
603     }
604   else if (exp10 == 5)
605     {
606       overflow = simpleMultiplyHighPrecision (result, length, TEN_E5);
607       if (overflow)
608         result[length++] = overflow;
609     }
610   else if (exp10 == 6)
611     {
612       overflow = simpleMultiplyHighPrecision (result, length, TEN_E6);
613       if (overflow)
614         result[length++] = overflow;
615     }
616   else if (exp10 == 7)
617     {
618       overflow = simpleMultiplyHighPrecision (result, length, TEN_E7);
619       if (overflow)
620         result[length++] = overflow;
621     }
622   else if (exp10 == 8)
623     {
624       overflow = simpleMultiplyHighPrecision (result, length, TEN_E8);
625       if (overflow)
626         result[length++] = overflow;
627     }
628   return length;
629 }
630 
631 uint64_t
doubleMantissa(jdouble z)632 doubleMantissa (jdouble z)
633 {
634   uint64_t m = DOUBLE_TO_LONGBITS (z);
635 
636   if ((m & EXPONENT_MASK) != 0)
637     m = (m & MANTISSA_MASK) | NORMAL_MASK;
638   else
639     m = (m & MANTISSA_MASK);
640 
641   return m;
642 }
643 
644 int32_t
doubleExponent(jdouble z)645 doubleExponent (jdouble z)
646 {
647   /* assumes positive double */
648   int32_t k = HIGH_U32_FROM_VAR (z) >> 20;
649 
650   if (k)
651     k -= E_OFFSET;
652   else
653     k = 1 - E_OFFSET;
654 
655   return k;
656 }
657 
floatMantissa(jfloat z)658 uint32_t floatMantissa(jfloat z) {
659   uint32_t m = FLOAT_TO_INTBITS (z);
660 
661   if ((m & FLOAT_EXPONENT_MASK) != 0)
662     m = (m & FLOAT_MANTISSA_MASK) | FLOAT_NORMAL_MASK;
663   else
664     m = (m & FLOAT_MANTISSA_MASK);
665 
666   return m;
667 }
668 
669 int32_t
floatExponent(jfloat z)670 floatExponent (jfloat z)
671 {
672   /* assumes positive float */
673   int32_t k = FLOAT_TO_INTBITS (z) >> 23;
674   if (k)
675     k -= FLOAT_E_OFFSET;
676   else
677     k = 1 - FLOAT_E_OFFSET;
678 
679   return k;
680 }
681 
682 /* Allow a 64-bit value in arg2 */
683 uint64_t
simpleMultiplyHighPrecision64(uint64_t * arg1,int32_t length,uint64_t arg2)684 simpleMultiplyHighPrecision64 (uint64_t * arg1, int32_t length, uint64_t arg2)
685 {
686   uint64_t intermediate, carry1, carry2, prod1, prod2, sum;
687   uint64_t* pArg1;
688   int32_t index;
689   uint32_t buf32;
690 
691   index = 0;
692   intermediate = 0;
693   pArg1 = arg1 + index;
694   carry1 = carry2 = 0;
695 
696   do
697     {
698       if ((*pArg1 != 0) || (intermediate != 0))
699         {
700           prod1 =
701             static_cast<uint64_t>(LOW_U32_FROM_VAR (arg2)) * static_cast<uint64_t>(LOW_U32_FROM_PTR (pArg1));
702           sum = intermediate + prod1;
703           if ((sum < prod1) || (sum < intermediate))
704             {
705               carry1 = 1;
706             }
707           else
708             {
709               carry1 = 0;
710             }
711           prod1 =
712             static_cast<uint64_t>(LOW_U32_FROM_VAR (arg2)) * static_cast<uint64_t>(HIGH_U32_FROM_PTR (pArg1));
713           prod2 =
714             static_cast<uint64_t>(HIGH_U32_FROM_VAR (arg2)) * static_cast<uint64_t>(LOW_U32_FROM_PTR (pArg1));
715           intermediate = carry2 + HIGH_IN_U64 (sum) + prod1 + prod2;
716           if ((intermediate < prod1) || (intermediate < prod2))
717             {
718               carry2 = 1;
719             }
720           else
721             {
722               carry2 = 0;
723             }
724           LOW_U32_FROM_PTR (pArg1) = LOW_U32_FROM_VAR (sum);
725           buf32 = HIGH_U32_FROM_PTR (pArg1);
726           HIGH_U32_FROM_PTR (pArg1) = LOW_U32_FROM_VAR (intermediate);
727           intermediate = carry1 + HIGH_IN_U64 (intermediate)
728             + static_cast<uint64_t>(HIGH_U32_FROM_VAR (arg2)) * static_cast<uint64_t>(buf32);
729         }
730       pArg1++;
731     }
732   while (++index < length);
733   return intermediate;
734 }
735