linux/drivers/gpu/drm/amd/pm/powerplay/hwmgr/ppevvmath.h
<<
>>
Prefs
   1/*
   2 * Copyright 2015 Advanced Micro Devices, Inc.
   3 *
   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:
  10 *
  11 * The above copyright notice and this permission notice shall be included in
  12 * all copies or substantial portions of the Software.
  13 *
  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.
  21 *
  22 */
  23#include <asm/div64.h>
  24
  25#define SHIFT_AMOUNT 16 /* We multiply all original integers with 2^SHIFT_AMOUNT to get the fInt representation */
  26
  27#define PRECISION 5 /* Change this value to change the number of decimal places in the final output - 5 is a good default */
  28
  29#define SHIFTED_2 (2 << SHIFT_AMOUNT)
  30#define MAX (1 << (SHIFT_AMOUNT - 1)) - 1 /* 32767 - Might change in the future */
  31
  32/* -------------------------------------------------------------------------------
  33 * NEW TYPE - fINT
  34 * -------------------------------------------------------------------------------
  35 * A variable of type fInt can be accessed in 3 ways using the dot (.) operator
  36 * fInt A;
  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
  40 */
  41typedef union _fInt {
  42    int full;
  43    struct _partial {
  44        unsigned int decimal: SHIFT_AMOUNT; /*Needs to always be unsigned*/
  45        int real: 32 - SHIFT_AMOUNT;
  46    } partial;
  47} fInt;
  48
  49/* -------------------------------------------------------------------------------
  50 * Function Declarations
  51 *  -------------------------------------------------------------------------------
  52 */
  53static fInt ConvertToFraction(int);                       /* Use this to convert an INT to a FINT */
  54static fInt Convert_ULONG_ToFraction(uint32_t);           /* Use this to convert an uint32_t to a FINT */
  55static fInt GetScaledFraction(int, int);                  /* Use this to convert an INT to a FINT after scaling it by a factor */
  56static int ConvertBackToInteger(fInt);                    /* Convert a FINT back to an INT that is scaled by 1000 (i.e. last 3 digits are the decimal digits) */
  57
  58static fInt fNegate(fInt);                                /* Returns -1 * input fInt value */
  59static fInt fAdd (fInt, fInt);                            /* Returns the sum of two fInt numbers */
  60static fInt fSubtract (fInt A, fInt B);                   /* Returns A-B - Sometimes easier than Adding negative numbers */
  61static fInt fMultiply (fInt, fInt);                       /* Returns the product of two fInt numbers */
  62static fInt fDivide (fInt A, fInt B);                     /* Returns A/B */
  63static fInt fGetSquare(fInt);                             /* Returns the square of a fInt number */
  64static fInt fSqrt(fInt);                                  /* Returns the Square Root of a fInt number */
  65
  66static int uAbs(int);                                     /* Returns the Absolute value of the Int */
  67static int uPow(int base, int exponent);                  /* Returns base^exponent an INT */
  68
  69static void SolveQuadracticEqn(fInt, fInt, fInt, fInt[]); /* Returns the 2 roots via the array */
  70static bool Equal(fInt, fInt);                            /* Returns true if two fInts are equal to each other */
  71static bool GreaterThan(fInt A, fInt B);                  /* Returns true if A > B */
  72
  73static fInt fExponential(fInt exponent);                  /* Can be used to calculate e^exponent */
  74static fInt fNaturalLog(fInt value);                      /* Can be used to calculate ln(value) */
  75
  76/* Fuse decoding functions
  77 * -------------------------------------------------------------------------------------
  78 */
  79static fInt fDecodeLinearFuse(uint32_t fuse_value, fInt f_min, fInt f_range, uint32_t bitlength);
  80static fInt fDecodeLogisticFuse(uint32_t fuse_value, fInt f_average, fInt f_range, uint32_t bitlength);
  81static fInt fDecodeLeakageID (uint32_t leakageID_fuse, fInt ln_max_div_min, fInt f_min, uint32_t bitlength);
  82
  83/* Internal Support Functions - Use these ONLY for testing or adding to internal functions
  84 * -------------------------------------------------------------------------------------
  85 * Some of the following functions take two INTs as their input - This is unsafe for a variety of reasons.
  86 */
  87static fInt Divide (int, int);                            /* Divide two INTs and return result as FINT */
  88static fInt fNegate(fInt);
  89
  90static int uGetScaledDecimal (fInt);                      /* Internal function */
  91static int GetReal (fInt A);                              /* Internal function */
  92
  93/* -------------------------------------------------------------------------------------
  94 * TROUBLESHOOTING INFORMATION
  95 * -------------------------------------------------------------------------------------
  96 * 1) ConvertToFraction - InputOutOfRangeException: Only accepts numbers smaller than MAX (default: 32767)
  97 * 2) fAdd - OutputOutOfRangeException: Output bigger than MAX (default: 32767)
  98 * 3) fMultiply - OutputOutOfRangeException:
  99 * 4) fGetSquare - OutputOutOfRangeException:
 100 * 5) fDivide - DivideByZeroException
 101 * 6) fSqrt - NegativeSquareRootException: Input cannot be a negative number
 102 */
 103
 104/* -------------------------------------------------------------------------------------
 105 * START OF CODE
 106 * -------------------------------------------------------------------------------------
 107 */
 108static fInt fExponential(fInt exponent)        /*Can be used to calculate e^exponent*/
 109{
 110        uint32_t i;
 111        bool bNegated = false;
 112
 113        fInt fPositiveOne = ConvertToFraction(1);
 114        fInt fZERO = ConvertToFraction(0);
 115
 116        fInt lower_bound = Divide(78, 10000);
 117        fInt solution = fPositiveOne; /*Starting off with baseline of 1 */
 118        fInt error_term;
 119
 120        static const uint32_t k_array[11] = {55452, 27726, 13863, 6931, 4055, 2231, 1178, 606, 308, 155, 78};
 121        static const uint32_t expk_array[11] = {2560000, 160000, 40000, 20000, 15000, 12500, 11250, 10625, 10313, 10156, 10078};
 122
 123        if (GreaterThan(fZERO, exponent)) {
 124                exponent = fNegate(exponent);
 125                bNegated = true;
 126        }
 127
 128        while (GreaterThan(exponent, lower_bound)) {
 129                for (i = 0; i < 11; i++) {
 130                        if (GreaterThan(exponent, GetScaledFraction(k_array[i], 10000))) {
 131                                exponent = fSubtract(exponent, GetScaledFraction(k_array[i], 10000));
 132                                solution = fMultiply(solution, GetScaledFraction(expk_array[i], 10000));
 133                        }
 134                }
 135        }
 136
 137        error_term = fAdd(fPositiveOne, exponent);
 138
 139        solution = fMultiply(solution, error_term);
 140
 141        if (bNegated)
 142                solution = fDivide(fPositiveOne, solution);
 143
 144        return solution;
 145}
 146
 147static fInt fNaturalLog(fInt value)
 148{
 149        uint32_t i;
 150        fInt upper_bound = Divide(8, 1000);
 151        fInt fNegativeOne = ConvertToFraction(-1);
 152        fInt solution = ConvertToFraction(0); /*Starting off with baseline of 0 */
 153        fInt error_term;
 154
 155        static const uint32_t k_array[10] = {160000, 40000, 20000, 15000, 12500, 11250, 10625, 10313, 10156, 10078};
 156        static const uint32_t logk_array[10] = {27726, 13863, 6931, 4055, 2231, 1178, 606, 308, 155, 78};
 157
 158        while (GreaterThan(fAdd(value, fNegativeOne), upper_bound)) {
 159                for (i = 0; i < 10; i++) {
 160                        if (GreaterThan(value, GetScaledFraction(k_array[i], 10000))) {
 161                                value = fDivide(value, GetScaledFraction(k_array[i], 10000));
 162                                solution = fAdd(solution, GetScaledFraction(logk_array[i], 10000));
 163                        }
 164                }
 165        }
 166
 167        error_term = fAdd(fNegativeOne, value);
 168
 169        return (fAdd(solution, error_term));
 170}
 171
 172static fInt fDecodeLinearFuse(uint32_t fuse_value, fInt f_min, fInt f_range, uint32_t bitlength)
 173{
 174        fInt f_fuse_value = Convert_ULONG_ToFraction(fuse_value);
 175        fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
 176
 177        fInt f_decoded_value;
 178
 179        f_decoded_value = fDivide(f_fuse_value, f_bit_max_value);
 180        f_decoded_value = fMultiply(f_decoded_value, f_range);
 181        f_decoded_value = fAdd(f_decoded_value, f_min);
 182
 183        return f_decoded_value;
 184}
 185
 186
 187static fInt fDecodeLogisticFuse(uint32_t fuse_value, fInt f_average, fInt f_range, uint32_t bitlength)
 188{
 189        fInt f_fuse_value = Convert_ULONG_ToFraction(fuse_value);
 190        fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
 191
 192        fInt f_CONSTANT_NEG13 = ConvertToFraction(-13);
 193        fInt f_CONSTANT1 = ConvertToFraction(1);
 194
 195        fInt f_decoded_value;
 196
 197        f_decoded_value = fSubtract(fDivide(f_bit_max_value, f_fuse_value), f_CONSTANT1);
 198        f_decoded_value = fNaturalLog(f_decoded_value);
 199        f_decoded_value = fMultiply(f_decoded_value, fDivide(f_range, f_CONSTANT_NEG13));
 200        f_decoded_value = fAdd(f_decoded_value, f_average);
 201
 202        return f_decoded_value;
 203}
 204
 205static fInt fDecodeLeakageID (uint32_t leakageID_fuse, fInt ln_max_div_min, fInt f_min, uint32_t bitlength)
 206{
 207        fInt fLeakage;
 208        fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
 209
 210        fLeakage = fMultiply(ln_max_div_min, Convert_ULONG_ToFraction(leakageID_fuse));
 211        fLeakage = fDivide(fLeakage, f_bit_max_value);
 212        fLeakage = fExponential(fLeakage);
 213        fLeakage = fMultiply(fLeakage, f_min);
 214
 215        return fLeakage;
 216}
 217
 218static fInt ConvertToFraction(int X) /*Add all range checking here. Is it possible to make fInt a private declaration? */
 219{
 220        fInt temp;
 221
 222        if (X <= MAX)
 223                temp.full = (X << SHIFT_AMOUNT);
 224        else
 225                temp.full = 0;
 226
 227        return temp;
 228}
 229
 230static fInt fNegate(fInt X)
 231{
 232        fInt CONSTANT_NEGONE = ConvertToFraction(-1);
 233        return (fMultiply(X, CONSTANT_NEGONE));
 234}
 235
 236static fInt Convert_ULONG_ToFraction(uint32_t X)
 237{
 238        fInt temp;
 239
 240        if (X <= MAX)
 241                temp.full = (X << SHIFT_AMOUNT);
 242        else
 243                temp.full = 0;
 244
 245        return temp;
 246}
 247
 248static fInt GetScaledFraction(int X, int factor)
 249{
 250        int times_shifted, factor_shifted;
 251        bool bNEGATED;
 252        fInt fValue;
 253
 254        times_shifted = 0;
 255        factor_shifted = 0;
 256        bNEGATED = false;
 257
 258        if (X < 0) {
 259                X = -1*X;
 260                bNEGATED = true;
 261        }
 262
 263        if (factor < 0) {
 264                factor = -1*factor;
 265                bNEGATED = !bNEGATED; /*If bNEGATED = true due to X < 0, this will cover the case of negative cancelling negative */
 266        }
 267
 268        if ((X > MAX) || factor > MAX) {
 269                if ((X/factor) <= MAX) {
 270                        while (X > MAX) {
 271                                X = X >> 1;
 272                                times_shifted++;
 273                        }
 274
 275                        while (factor > MAX) {
 276                                factor = factor >> 1;
 277                                factor_shifted++;
 278                        }
 279                } else {
 280                        fValue.full = 0;
 281                        return fValue;
 282                }
 283        }
 284
 285        if (factor == 1)
 286                return ConvertToFraction(X);
 287
 288        fValue = fDivide(ConvertToFraction(X * uPow(-1, bNEGATED)), ConvertToFraction(factor));
 289
 290        fValue.full = fValue.full << times_shifted;
 291        fValue.full = fValue.full >> factor_shifted;
 292
 293        return fValue;
 294}
 295
 296/* Addition using two fInts */
 297static fInt fAdd (fInt X, fInt Y)
 298{
 299        fInt Sum;
 300
 301        Sum.full = X.full + Y.full;
 302
 303        return Sum;
 304}
 305
 306/* Addition using two fInts */
 307static fInt fSubtract (fInt X, fInt Y)
 308{
 309        fInt Difference;
 310
 311        Difference.full = X.full - Y.full;
 312
 313        return Difference;
 314}
 315
 316static bool Equal(fInt A, fInt B)
 317{
 318        if (A.full == B.full)
 319                return true;
 320        else
 321                return false;
 322}
 323
 324static bool GreaterThan(fInt A, fInt B)
 325{
 326        if (A.full > B.full)
 327                return true;
 328        else
 329                return false;
 330}
 331
 332static fInt fMultiply (fInt X, fInt Y) /* Uses 64-bit integers (int64_t) */
 333{
 334        fInt Product;
 335        int64_t tempProduct;
 336
 337        /*The following is for a very specific common case: Non-zero number with ONLY fractional portion*/
 338        /* TEMPORARILY DISABLED - CAN BE USED TO IMPROVE PRECISION
 339        bool X_LessThanOne, Y_LessThanOne;
 340
 341        X_LessThanOne = (X.partial.real == 0 && X.partial.decimal != 0 && X.full >= 0);
 342        Y_LessThanOne = (Y.partial.real == 0 && Y.partial.decimal != 0 && Y.full >= 0);
 343
 344        if (X_LessThanOne && Y_LessThanOne) {
 345                Product.full = X.full * Y.full;
 346                return Product
 347        }*/
 348
 349        tempProduct = ((int64_t)X.full) * ((int64_t)Y.full); /*Q(16,16)*Q(16,16) = Q(32, 32) - Might become a negative number! */
 350        tempProduct = tempProduct >> 16; /*Remove lagging 16 bits - Will lose some precision from decimal; */
 351        Product.full = (int)tempProduct; /*The int64_t will lose the leading 16 bits that were part of the integer portion */
 352
 353        return Product;
 354}
 355
 356static fInt fDivide (fInt X, fInt Y)
 357{
 358        fInt fZERO, fQuotient;
 359        int64_t longlongX, longlongY;
 360
 361        fZERO = ConvertToFraction(0);
 362
 363        if (Equal(Y, fZERO))
 364                return fZERO;
 365
 366        longlongX = (int64_t)X.full;
 367        longlongY = (int64_t)Y.full;
 368
 369        longlongX = longlongX << 16; /*Q(16,16) -> Q(32,32) */
 370
 371        div64_s64(longlongX, longlongY); /*Q(32,32) divided by Q(16,16) = Q(16,16) Back to original format */
 372
 373        fQuotient.full = (int)longlongX;
 374        return fQuotient;
 375}
 376
 377static int ConvertBackToInteger (fInt A) /*THIS is the function that will be used to check with the Golden settings table*/
 378{
 379        fInt fullNumber, scaledDecimal, scaledReal;
 380
 381        scaledReal.full = GetReal(A) * uPow(10, PRECISION-1); /* DOUBLE CHECK THISSSS!!! */
 382
 383        scaledDecimal.full = uGetScaledDecimal(A);
 384
 385        fullNumber = fAdd(scaledDecimal,scaledReal);
 386
 387        return fullNumber.full;
 388}
 389
 390static fInt fGetSquare(fInt A)
 391{
 392        return fMultiply(A,A);
 393}
 394
 395/* x_new = x_old - (x_old^2 - C) / (2 * x_old) */
 396static fInt fSqrt(fInt num)
 397{
 398        fInt F_divide_Fprime, Fprime;
 399        fInt test;
 400        fInt twoShifted;
 401        int seed, counter, error;
 402        fInt x_new, x_old, C, y;
 403
 404        fInt fZERO = ConvertToFraction(0);
 405
 406        /* (0 > num) is the same as (num < 0), i.e., num is negative */
 407
 408        if (GreaterThan(fZERO, num) || Equal(fZERO, num))
 409                return fZERO;
 410
 411        C = num;
 412
 413        if (num.partial.real > 3000)
 414                seed = 60;
 415        else if (num.partial.real > 1000)
 416                seed = 30;
 417        else if (num.partial.real > 100)
 418                seed = 10;
 419        else
 420                seed = 2;
 421
 422        counter = 0;
 423
 424        if (Equal(num, fZERO)) /*Square Root of Zero is zero */
 425                return fZERO;
 426
 427        twoShifted = ConvertToFraction(2);
 428        x_new = ConvertToFraction(seed);
 429
 430        do {
 431                counter++;
 432
 433                x_old.full = x_new.full;
 434
 435                test = fGetSquare(x_old); /*1.75*1.75 is reverting back to 1 when shifted down */
 436                y = fSubtract(test, C); /*y = f(x) = x^2 - C; */
 437
 438                Fprime = fMultiply(twoShifted, x_old);
 439                F_divide_Fprime = fDivide(y, Fprime);
 440
 441                x_new = fSubtract(x_old, F_divide_Fprime);
 442
 443                error = ConvertBackToInteger(x_new) - ConvertBackToInteger(x_old);
 444
 445                if (counter > 20) /*20 is already way too many iterations. If we dont have an answer by then, we never will*/
 446                        return x_new;
 447
 448        } while (uAbs(error) > 0);
 449
 450        return (x_new);
 451}
 452
 453static void SolveQuadracticEqn(fInt A, fInt B, fInt C, fInt Roots[])
 454{
 455        fInt *pRoots = &Roots[0];
 456        fInt temp, root_first, root_second;
 457        fInt f_CONSTANT10, f_CONSTANT100;
 458
 459        f_CONSTANT100 = ConvertToFraction(100);
 460        f_CONSTANT10 = ConvertToFraction(10);
 461
 462        while(GreaterThan(A, f_CONSTANT100) || GreaterThan(B, f_CONSTANT100) || GreaterThan(C, f_CONSTANT100)) {
 463                A = fDivide(A, f_CONSTANT10);
 464                B = fDivide(B, f_CONSTANT10);
 465                C = fDivide(C, f_CONSTANT10);
 466        }
 467
 468        temp = fMultiply(ConvertToFraction(4), A); /* root = 4*A */
 469        temp = fMultiply(temp, C); /* root = 4*A*C */
 470        temp = fSubtract(fGetSquare(B), temp); /* root = b^2 - 4AC */
 471        temp = fSqrt(temp); /*root = Sqrt (b^2 - 4AC); */
 472
 473        root_first = fSubtract(fNegate(B), temp); /* b - Sqrt(b^2 - 4AC) */
 474        root_second = fAdd(fNegate(B), temp); /* b + Sqrt(b^2 - 4AC) */
 475
 476        root_first = fDivide(root_first, ConvertToFraction(2)); /* [b +- Sqrt(b^2 - 4AC)]/[2] */
 477        root_first = fDivide(root_first, A); /*[b +- Sqrt(b^2 - 4AC)]/[2*A] */
 478
 479        root_second = fDivide(root_second, ConvertToFraction(2)); /* [b +- Sqrt(b^2 - 4AC)]/[2] */
 480        root_second = fDivide(root_second, A); /*[b +- Sqrt(b^2 - 4AC)]/[2*A] */
 481
 482        *(pRoots + 0) = root_first;
 483        *(pRoots + 1) = root_second;
 484}
 485
 486/* -----------------------------------------------------------------------------
 487 * SUPPORT FUNCTIONS
 488 * -----------------------------------------------------------------------------
 489 */
 490
 491/* Conversion Functions */
 492static int GetReal (fInt A)
 493{
 494        return (A.full >> SHIFT_AMOUNT);
 495}
 496
 497static fInt Divide (int X, int Y)
 498{
 499        fInt A, B, Quotient;
 500
 501        A.full = X << SHIFT_AMOUNT;
 502        B.full = Y << SHIFT_AMOUNT;
 503
 504        Quotient = fDivide(A, B);
 505
 506        return Quotient;
 507}
 508
 509static int uGetScaledDecimal (fInt A) /*Converts the fractional portion to whole integers - Costly function */
 510{
 511        int dec[PRECISION];
 512        int i, scaledDecimal = 0, tmp = A.partial.decimal;
 513
 514        for (i = 0; i < PRECISION; i++) {
 515                dec[i] = tmp / (1 << SHIFT_AMOUNT);
 516                tmp = tmp - ((1 << SHIFT_AMOUNT)*dec[i]);
 517                tmp *= 10;
 518                scaledDecimal = scaledDecimal + dec[i]*uPow(10, PRECISION - 1 -i);
 519        }
 520
 521        return scaledDecimal;
 522}
 523
 524static int uPow(int base, int power)
 525{
 526        if (power == 0)
 527                return 1;
 528        else
 529                return (base)*uPow(base, power - 1);
 530}
 531
 532static int uAbs(int X)
 533{
 534        if (X < 0)
 535                return (X * -1);
 536        else
 537                return X;
 538}
 539
 540static fInt fRoundUpByStepSize(fInt A, fInt fStepSize, bool error_term)
 541{
 542        fInt solution;
 543
 544        solution = fDivide(A, fStepSize);
 545        solution.partial.decimal = 0; /*All fractional digits changes to 0 */
 546
 547        if (error_term)
 548                solution.partial.real += 1; /*Error term of 1 added */
 549
 550        solution = fMultiply(solution, fStepSize);
 551        solution = fAdd(solution, fStepSize);
 552
 553        return solution;
 554}
 555
 556