2 * Copyright 2015 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
23 #include <asm/div64.h>
25 #define SHIFT_AMOUNT 16 /* We multiply all original integers with 2^SHIFT_AMOUNT to get the fInt representation */
27 #define PRECISION 5 /* Change this value to change the number of decimal places in the final output - 5 is a good default */
29 #define SHIFTED_2 (2 << SHIFT_AMOUNT)
30 #define MAX (1 << (SHIFT_AMOUNT - 1)) - 1 /* 32767 - Might change in the future */
32 /* -------------------------------------------------------------------------------
34 * -------------------------------------------------------------------------------
35 * A variable of type fInt can be accessed in 3 ways using the dot (.) operator
37 * A.full => The full number as it is. Generally not easy to read
38 * A.partial.real => Only the integer portion
39 * A.partial.decimal => Only the fractional portion
44 unsigned int decimal: SHIFT_AMOUNT; /*Needs to always be unsigned*/
45 int real: 32 - SHIFT_AMOUNT;
49 /* -------------------------------------------------------------------------------
50 * Function Declarations
51 * -------------------------------------------------------------------------------
53 fInt ConvertToFraction(int); /* Use this to convert an INT to a FINT */
54 fInt Convert_ULONG_ToFraction(uint32_t); /* Use this to convert an uint32_t to a FINT */
55 fInt GetScaledFraction(int, int); /* Use this to convert an INT to a FINT after scaling it by a factor */
56 int ConvertBackToInteger(fInt); /* Convert a FINT back to an INT that is scaled by 1000 (i.e. last 3 digits are the decimal digits) */
58 fInt fNegate(fInt); /* Returns -1 * input fInt value */
59 fInt fAdd (fInt, fInt); /* Returns the sum of two fInt numbers */
60 fInt fSubtract (fInt A, fInt B); /* Returns A-B - Sometimes easier than Adding negative numbers */
61 fInt fMultiply (fInt, fInt); /* Returns the product of two fInt numbers */
62 fInt fDivide (fInt A, fInt B); /* Returns A/B */
63 fInt fGetSquare(fInt); /* Returns the square of a fInt number */
64 fInt fSqrt(fInt); /* Returns the Square Root of a fInt number */
66 int uAbs(int); /* Returns the Absolute value of the Int */
67 fInt fAbs(fInt); /* Returns the Absolute value of the fInt */
68 int uPow(int base, int exponent); /* Returns base^exponent an INT */
70 void SolveQuadracticEqn(fInt, fInt, fInt, fInt[]); /* Returns the 2 roots via the array */
71 bool Equal(fInt, fInt); /* Returns true if two fInts are equal to each other */
72 bool GreaterThan(fInt A, fInt B); /* Returns true if A > B */
74 fInt fExponential(fInt exponent); /* Can be used to calculate e^exponent */
75 fInt fNaturalLog(fInt value); /* Can be used to calculate ln(value) */
77 /* Fuse decoding functions
78 * -------------------------------------------------------------------------------------
80 fInt fDecodeLinearFuse(uint32_t fuse_value, fInt f_min, fInt f_range, uint32_t bitlength);
81 fInt fDecodeLogisticFuse(uint32_t fuse_value, fInt f_average, fInt f_range, uint32_t bitlength);
82 fInt fDecodeLeakageID (uint32_t leakageID_fuse, fInt ln_max_div_min, fInt f_min, uint32_t bitlength);
84 /* Internal Support Functions - Use these ONLY for testing or adding to internal functions
85 * -------------------------------------------------------------------------------------
86 * Some of the following functions take two INTs as their input - This is unsafe for a variety of reasons.
88 fInt Add (int, int); /* Add two INTs and return Sum as FINT */
89 fInt Multiply (int, int); /* Multiply two INTs and return Product as FINT */
90 fInt Divide (int, int); /* You get the idea... */
93 int uGetScaledDecimal (fInt); /* Internal function */
94 int GetReal (fInt A); /* Internal function */
96 /* Future Additions and Incomplete Functions
97 * -------------------------------------------------------------------------------------
99 int GetRoundedValue(fInt); /* Incomplete function - Useful only when Precision is lacking */
100 /* Let us say we have 2.126 but can only handle 2 decimal points. We could */
101 /* either chop of 6 and keep 2.12 or use this function to get 2.13, which is more accurate */
103 /* -------------------------------------------------------------------------------------
104 * TROUBLESHOOTING INFORMATION
105 * -------------------------------------------------------------------------------------
106 * 1) ConvertToFraction - InputOutOfRangeException: Only accepts numbers smaller than MAX (default: 32767)
107 * 2) fAdd - OutputOutOfRangeException: Output bigger than MAX (default: 32767)
108 * 3) fMultiply - OutputOutOfRangeException:
109 * 4) fGetSquare - OutputOutOfRangeException:
110 * 5) fDivide - DivideByZeroException
111 * 6) fSqrt - NegativeSquareRootException: Input cannot be a negative number
114 /* -------------------------------------------------------------------------------------
116 * -------------------------------------------------------------------------------------
118 fInt fExponential(fInt exponent) /*Can be used to calculate e^exponent*/
121 bool bNegated = false;
123 fInt fPositiveOne = ConvertToFraction(1);
124 fInt fZERO = ConvertToFraction(0);
126 fInt lower_bound = Divide(78, 10000);
127 fInt solution = fPositiveOne; /*Starting off with baseline of 1 */
130 uint32_t k_array[11] = {55452, 27726, 13863, 6931, 4055, 2231, 1178, 606, 308, 155, 78};
131 uint32_t expk_array[11] = {2560000, 160000, 40000, 20000, 15000, 12500, 11250, 10625, 10313, 10156, 10078};
133 if (GreaterThan(fZERO, exponent)) {
134 exponent = fNegate(exponent);
138 while (GreaterThan(exponent, lower_bound)) {
139 for (i = 0; i < 11; i++) {
140 if (GreaterThan(exponent, GetScaledFraction(k_array[i], 10000))) {
141 exponent = fSubtract(exponent, GetScaledFraction(k_array[i], 10000));
142 solution = fMultiply(solution, GetScaledFraction(expk_array[i], 10000));
147 error_term = fAdd(fPositiveOne, exponent);
149 solution = fMultiply(solution, error_term);
152 solution = fDivide(fPositiveOne, solution);
157 fInt fNaturalLog(fInt value)
160 fInt upper_bound = Divide(8, 1000);
161 fInt fNegativeOne = ConvertToFraction(-1);
162 fInt solution = ConvertToFraction(0); /*Starting off with baseline of 0 */
165 uint32_t k_array[10] = {160000, 40000, 20000, 15000, 12500, 11250, 10625, 10313, 10156, 10078};
166 uint32_t logk_array[10] = {27726, 13863, 6931, 4055, 2231, 1178, 606, 308, 155, 78};
168 while (GreaterThan(fAdd(value, fNegativeOne), upper_bound)) {
169 for (i = 0; i < 10; i++) {
170 if (GreaterThan(value, GetScaledFraction(k_array[i], 10000))) {
171 value = fDivide(value, GetScaledFraction(k_array[i], 10000));
172 solution = fAdd(solution, GetScaledFraction(logk_array[i], 10000));
177 error_term = fAdd(fNegativeOne, value);
179 return (fAdd(solution, error_term));
182 fInt fDecodeLinearFuse(uint32_t fuse_value, fInt f_min, fInt f_range, uint32_t bitlength)
184 fInt f_fuse_value = Convert_ULONG_ToFraction(fuse_value);
185 fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
187 fInt f_decoded_value;
189 f_decoded_value = fDivide(f_fuse_value, f_bit_max_value);
190 f_decoded_value = fMultiply(f_decoded_value, f_range);
191 f_decoded_value = fAdd(f_decoded_value, f_min);
193 return f_decoded_value;
197 fInt fDecodeLogisticFuse(uint32_t fuse_value, fInt f_average, fInt f_range, uint32_t bitlength)
199 fInt f_fuse_value = Convert_ULONG_ToFraction(fuse_value);
200 fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
202 fInt f_CONSTANT_NEG13 = ConvertToFraction(-13);
203 fInt f_CONSTANT1 = ConvertToFraction(1);
205 fInt f_decoded_value;
207 f_decoded_value = fSubtract(fDivide(f_bit_max_value, f_fuse_value), f_CONSTANT1);
208 f_decoded_value = fNaturalLog(f_decoded_value);
209 f_decoded_value = fMultiply(f_decoded_value, fDivide(f_range, f_CONSTANT_NEG13));
210 f_decoded_value = fAdd(f_decoded_value, f_average);
212 return f_decoded_value;
215 fInt fDecodeLeakageID (uint32_t leakageID_fuse, fInt ln_max_div_min, fInt f_min, uint32_t bitlength)
218 fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
220 fLeakage = fMultiply(ln_max_div_min, Convert_ULONG_ToFraction(leakageID_fuse));
221 fLeakage = fDivide(fLeakage, f_bit_max_value);
222 fLeakage = fExponential(fLeakage);
223 fLeakage = fMultiply(fLeakage, f_min);
228 fInt ConvertToFraction(int X) /*Add all range checking here. Is it possible to make fInt a private declaration? */
233 temp.full = (X << SHIFT_AMOUNT);
242 fInt CONSTANT_NEGONE = ConvertToFraction(-1);
243 return (fMultiply(X, CONSTANT_NEGONE));
246 fInt Convert_ULONG_ToFraction(uint32_t X)
251 temp.full = (X << SHIFT_AMOUNT);
258 fInt GetScaledFraction(int X, int factor)
260 int times_shifted, factor_shifted;
276 bNEGATED = !bNEGATED; /*If bNEGATED = true due to X < 0, this will cover the case of negative cancelling negative */
279 if ((X > MAX) || factor > MAX) {
280 if ((X/factor) <= MAX) {
286 while (factor > MAX) {
287 factor = factor >> 1;
297 return (ConvertToFraction(X));
299 fValue = fDivide(ConvertToFraction(X * uPow(-1, bNEGATED)), ConvertToFraction(factor));
301 fValue.full = fValue.full << times_shifted;
302 fValue.full = fValue.full >> factor_shifted;
307 /* Addition using two fInts */
308 fInt fAdd (fInt X, fInt Y)
312 Sum.full = X.full + Y.full;
317 /* Addition using two fInts */
318 fInt fSubtract (fInt X, fInt Y)
322 Difference.full = X.full - Y.full;
327 bool Equal(fInt A, fInt B)
329 if (A.full == B.full)
335 bool GreaterThan(fInt A, fInt B)
343 fInt fMultiply (fInt X, fInt Y) /* Uses 64-bit integers (int64_t) */
347 bool X_LessThanOne, Y_LessThanOne;
349 X_LessThanOne = (X.partial.real == 0 && X.partial.decimal != 0 && X.full >= 0);
350 Y_LessThanOne = (Y.partial.real == 0 && Y.partial.decimal != 0 && Y.full >= 0);
352 /*The following is for a very specific common case: Non-zero number with ONLY fractional portion*/
353 /* TEMPORARILY DISABLED - CAN BE USED TO IMPROVE PRECISION
355 if (X_LessThanOne && Y_LessThanOne) {
356 Product.full = X.full * Y.full;
360 tempProduct = ((int64_t)X.full) * ((int64_t)Y.full); /*Q(16,16)*Q(16,16) = Q(32, 32) - Might become a negative number! */
361 tempProduct = tempProduct >> 16; /*Remove lagging 16 bits - Will lose some precision from decimal; */
362 Product.full = (int)tempProduct; /*The int64_t will lose the leading 16 bits that were part of the integer portion */
367 fInt fDivide (fInt X, fInt Y)
369 fInt fZERO, fQuotient;
370 int64_t longlongX, longlongY;
372 fZERO = ConvertToFraction(0);
377 longlongX = (int64_t)X.full;
378 longlongY = (int64_t)Y.full;
380 longlongX = longlongX << 16; /*Q(16,16) -> Q(32,32) */
382 do_div(longlongX, longlongY); /*Q(32,32) divided by Q(16,16) = Q(16,16) Back to original format */
384 fQuotient.full = (int)longlongX;
388 int ConvertBackToInteger (fInt A) /*THIS is the function that will be used to check with the Golden settings table*/
390 fInt fullNumber, scaledDecimal, scaledReal;
392 scaledReal.full = GetReal(A) * uPow(10, PRECISION-1); /* DOUBLE CHECK THISSSS!!! */
394 scaledDecimal.full = uGetScaledDecimal(A);
396 fullNumber = fAdd(scaledDecimal,scaledReal);
398 return fullNumber.full;
401 fInt fGetSquare(fInt A)
403 return fMultiply(A,A);
406 /* x_new = x_old - (x_old^2 - C) / (2 * x_old) */
409 fInt F_divide_Fprime, Fprime;
412 int seed, counter, error;
413 fInt x_new, x_old, C, y;
415 fInt fZERO = ConvertToFraction(0);
416 /* (0 > num) is the same as (num < 0), i.e., num is negative */
417 if (GreaterThan(fZERO, num) || Equal(fZERO, num))
422 if (num.partial.real > 3000)
424 else if (num.partial.real > 1000)
426 else if (num.partial.real > 100)
433 if (Equal(num, fZERO)) /*Square Root of Zero is zero */
436 twoShifted = ConvertToFraction(2);
437 x_new = ConvertToFraction(seed);
442 x_old.full = x_new.full;
444 test = fGetSquare(x_old); /*1.75*1.75 is reverting back to 1 when shifted down */
445 y = fSubtract(test, C); /*y = f(x) = x^2 - C; */
447 Fprime = fMultiply(twoShifted, x_old);
448 F_divide_Fprime = fDivide(y, Fprime);
450 x_new = fSubtract(x_old, F_divide_Fprime);
452 error = ConvertBackToInteger(x_new) - ConvertBackToInteger(x_old);
454 if (counter > 20) /*20 is already way too many iterations. If we dont have an answer by then, we never will*/
457 } while (uAbs(error) > 0);
462 void SolveQuadracticEqn(fInt A, fInt B, fInt C, fInt Roots[])
464 fInt* pRoots = &Roots[0];
465 fInt temp, root_first, root_second;
466 fInt f_CONSTANT10, f_CONSTANT100;
468 f_CONSTANT100 = ConvertToFraction(100);
469 f_CONSTANT10 = ConvertToFraction(10);
471 while(GreaterThan(A, f_CONSTANT100) || GreaterThan(B, f_CONSTANT100) || GreaterThan(C, f_CONSTANT100)) {
472 A = fDivide(A, f_CONSTANT10);
473 B = fDivide(B, f_CONSTANT10);
474 C = fDivide(C, f_CONSTANT10);
477 temp = fMultiply(ConvertToFraction(4), A); /* root = 4*A */
478 temp = fMultiply(temp, C); /* root = 4*A*C */
479 temp = fSubtract(fGetSquare(B), temp); /* root = b^2 - 4AC */
480 temp = fSqrt(temp); /*root = Sqrt (b^2 - 4AC); */
482 root_first = fSubtract(fNegate(B), temp); /* b - Sqrt(b^2 - 4AC) */
483 root_second = fAdd(fNegate(B), temp); /* b + Sqrt(b^2 - 4AC) */
485 root_first = fDivide(root_first, ConvertToFraction(2)); /* [b +- Sqrt(b^2 - 4AC)]/[2] */
486 root_first = fDivide(root_first, A); /*[b +- Sqrt(b^2 - 4AC)]/[2*A] */
488 root_second = fDivide(root_second, ConvertToFraction(2)); /* [b +- Sqrt(b^2 - 4AC)]/[2] */
489 root_second = fDivide(root_second, A); /*[b +- Sqrt(b^2 - 4AC)]/[2*A] */
491 *(pRoots + 0) = root_first;
492 *(pRoots + 1) = root_second;
495 /* -----------------------------------------------------------------------------
497 * -----------------------------------------------------------------------------
500 /* Addition using two normal ints - Temporary - Use only for testing purposes?. */
501 fInt Add (int X, int Y)
505 A.full = (X << SHIFT_AMOUNT);
506 B.full = (Y << SHIFT_AMOUNT);
508 Sum.full = A.full + B.full;
513 /* Conversion Functions */
516 return (A.full >> SHIFT_AMOUNT);
519 /* Temporarily Disabled */
520 int GetRoundedValue(fInt A) /*For now, round the 3rd decimal place */
522 /* ROUNDING TEMPORARLY DISABLED
525 int decimal_cutoff, decimal_mask = 0x000001FF;
527 decimal_cutoff = temp & decimal_mask;
530 if (decimal_cutoff > 0x147) {
534 return ConvertBackToInteger(A)/10000; /*Temporary - in case this was used somewhere else */
537 fInt Multiply (int X, int Y)
541 A.full = X << SHIFT_AMOUNT;
542 B.full = Y << SHIFT_AMOUNT;
544 Product = fMultiply(A, B);
548 fInt Divide (int X, int Y)
552 A.full = X << SHIFT_AMOUNT;
553 B.full = Y << SHIFT_AMOUNT;
555 Quotient = fDivide(A, B);
560 int uGetScaledDecimal (fInt A) /*Converts the fractional portion to whole integers - Costly function */
563 int i, scaledDecimal = 0, tmp = A.partial.decimal;
565 for (i = 0; i < PRECISION; i++) {
566 dec[i] = tmp / (1 << SHIFT_AMOUNT);
568 tmp = tmp - ((1 << SHIFT_AMOUNT)*dec[i]);
572 scaledDecimal = scaledDecimal + dec[i]*uPow(10, PRECISION - 1 -i);
575 return scaledDecimal;
578 int uPow(int base, int power)
583 return (base)*uPow(base, power - 1);
588 if (A.partial.real < 0)
589 return (fMultiply(A, ConvertToFraction(-1)));
602 fInt fRoundUpByStepSize(fInt A, fInt fStepSize, bool error_term)
606 solution = fDivide(A, fStepSize);
607 solution.partial.decimal = 0; /*All fractional digits changes to 0 */
610 solution.partial.real += 1; /*Error term of 1 added */
612 solution = fMultiply(solution, fStepSize);
613 solution = fAdd(solution, fStepSize);