drm/amd/powerplay: Mark functions of ppevvmath.h static
[deliverable/linux.git] / drivers / gpu / drm / amd / powerplay / hwmgr / ppevvmath.h
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 */
41 typedef 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 */
53 static fInt ConvertToFraction(int); /* Use this to convert an INT to a FINT */
54 static fInt Convert_ULONG_ToFraction(uint32_t); /* Use this to convert an uint32_t to a FINT */
55 static fInt GetScaledFraction(int, int); /* Use this to convert an INT to a FINT after scaling it by a factor */
56 static 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
58 static fInt fNegate(fInt); /* Returns -1 * input fInt value */
59 static fInt fAdd (fInt, fInt); /* Returns the sum of two fInt numbers */
60 static fInt fSubtract (fInt A, fInt B); /* Returns A-B - Sometimes easier than Adding negative numbers */
61 static fInt fMultiply (fInt, fInt); /* Returns the product of two fInt numbers */
62 static fInt fDivide (fInt A, fInt B); /* Returns A/B */
63 static fInt fGetSquare(fInt); /* Returns the square of a fInt number */
64 static fInt fSqrt(fInt); /* Returns the Square Root of a fInt number */
65
66 static int uAbs(int); /* Returns the Absolute value of the Int */
67 static fInt fAbs(fInt); /* Returns the Absolute value of the fInt */
68 static int uPow(int base, int exponent); /* Returns base^exponent an INT */
69
70 static void SolveQuadracticEqn(fInt, fInt, fInt, fInt[]); /* Returns the 2 roots via the array */
71 static bool Equal(fInt, fInt); /* Returns true if two fInts are equal to each other */
72 static bool GreaterThan(fInt A, fInt B); /* Returns true if A > B */
73
74 static fInt fExponential(fInt exponent); /* Can be used to calculate e^exponent */
75 static fInt fNaturalLog(fInt value); /* Can be used to calculate ln(value) */
76
77 /* Fuse decoding functions
78 * -------------------------------------------------------------------------------------
79 */
80 static fInt fDecodeLinearFuse(uint32_t fuse_value, fInt f_min, fInt f_range, uint32_t bitlength);
81 static fInt fDecodeLogisticFuse(uint32_t fuse_value, fInt f_average, fInt f_range, uint32_t bitlength);
82 static fInt fDecodeLeakageID (uint32_t leakageID_fuse, fInt ln_max_div_min, fInt f_min, uint32_t bitlength);
83
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.
87 */
88 static fInt Add (int, int); /* Add two INTs and return Sum as FINT */
89 static fInt Multiply (int, int); /* Multiply two INTs and return Product as FINT */
90 static fInt Divide (int, int); /* You get the idea... */
91 static fInt fNegate(fInt);
92
93 static int uGetScaledDecimal (fInt); /* Internal function */
94 static int GetReal (fInt A); /* Internal function */
95
96 /* Future Additions and Incomplete Functions
97 * -------------------------------------------------------------------------------------
98 */
99 static 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 */
102
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
112 */
113
114 /* -------------------------------------------------------------------------------------
115 * START OF CODE
116 * -------------------------------------------------------------------------------------
117 */
118 static fInt fExponential(fInt exponent) /*Can be used to calculate e^exponent*/
119 {
120 uint32_t i;
121 bool bNegated = false;
122
123 fInt fPositiveOne = ConvertToFraction(1);
124 fInt fZERO = ConvertToFraction(0);
125
126 fInt lower_bound = Divide(78, 10000);
127 fInt solution = fPositiveOne; /*Starting off with baseline of 1 */
128 fInt error_term;
129
130 static const uint32_t k_array[11] = {55452, 27726, 13863, 6931, 4055, 2231, 1178, 606, 308, 155, 78};
131 static const uint32_t expk_array[11] = {2560000, 160000, 40000, 20000, 15000, 12500, 11250, 10625, 10313, 10156, 10078};
132
133 if (GreaterThan(fZERO, exponent)) {
134 exponent = fNegate(exponent);
135 bNegated = true;
136 }
137
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));
143 }
144 }
145 }
146
147 error_term = fAdd(fPositiveOne, exponent);
148
149 solution = fMultiply(solution, error_term);
150
151 if (bNegated)
152 solution = fDivide(fPositiveOne, solution);
153
154 return solution;
155 }
156
157 static fInt fNaturalLog(fInt value)
158 {
159 uint32_t i;
160 fInt upper_bound = Divide(8, 1000);
161 fInt fNegativeOne = ConvertToFraction(-1);
162 fInt solution = ConvertToFraction(0); /*Starting off with baseline of 0 */
163 fInt error_term;
164
165 static const uint32_t k_array[10] = {160000, 40000, 20000, 15000, 12500, 11250, 10625, 10313, 10156, 10078};
166 static const uint32_t logk_array[10] = {27726, 13863, 6931, 4055, 2231, 1178, 606, 308, 155, 78};
167
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));
173 }
174 }
175 }
176
177 error_term = fAdd(fNegativeOne, value);
178
179 return (fAdd(solution, error_term));
180 }
181
182 static fInt fDecodeLinearFuse(uint32_t fuse_value, fInt f_min, fInt f_range, uint32_t bitlength)
183 {
184 fInt f_fuse_value = Convert_ULONG_ToFraction(fuse_value);
185 fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
186
187 fInt f_decoded_value;
188
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);
192
193 return f_decoded_value;
194 }
195
196
197 static fInt fDecodeLogisticFuse(uint32_t fuse_value, fInt f_average, fInt f_range, uint32_t bitlength)
198 {
199 fInt f_fuse_value = Convert_ULONG_ToFraction(fuse_value);
200 fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
201
202 fInt f_CONSTANT_NEG13 = ConvertToFraction(-13);
203 fInt f_CONSTANT1 = ConvertToFraction(1);
204
205 fInt f_decoded_value;
206
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);
211
212 return f_decoded_value;
213 }
214
215 static fInt fDecodeLeakageID (uint32_t leakageID_fuse, fInt ln_max_div_min, fInt f_min, uint32_t bitlength)
216 {
217 fInt fLeakage;
218 fInt f_bit_max_value = Convert_ULONG_ToFraction((uPow(2, bitlength)) - 1);
219
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);
224
225 return fLeakage;
226 }
227
228 static fInt ConvertToFraction(int X) /*Add all range checking here. Is it possible to make fInt a private declaration? */
229 {
230 fInt temp;
231
232 if (X <= MAX)
233 temp.full = (X << SHIFT_AMOUNT);
234 else
235 temp.full = 0;
236
237 return temp;
238 }
239
240 static fInt fNegate(fInt X)
241 {
242 fInt CONSTANT_NEGONE = ConvertToFraction(-1);
243 return (fMultiply(X, CONSTANT_NEGONE));
244 }
245
246 static fInt Convert_ULONG_ToFraction(uint32_t X)
247 {
248 fInt temp;
249
250 if (X <= MAX)
251 temp.full = (X << SHIFT_AMOUNT);
252 else
253 temp.full = 0;
254
255 return temp;
256 }
257
258 static fInt GetScaledFraction(int X, int factor)
259 {
260 int times_shifted, factor_shifted;
261 bool bNEGATED;
262 fInt fValue;
263
264 times_shifted = 0;
265 factor_shifted = 0;
266 bNEGATED = false;
267
268 if (X < 0) {
269 X = -1*X;
270 bNEGATED = true;
271 }
272
273 if (factor < 0) {
274 factor = -1*factor;
275 bNEGATED = !bNEGATED; /*If bNEGATED = true due to X < 0, this will cover the case of negative cancelling negative */
276 }
277
278 if ((X > MAX) || factor > MAX) {
279 if ((X/factor) <= MAX) {
280 while (X > MAX) {
281 X = X >> 1;
282 times_shifted++;
283 }
284
285 while (factor > MAX) {
286 factor = factor >> 1;
287 factor_shifted++;
288 }
289 } else {
290 fValue.full = 0;
291 return fValue;
292 }
293 }
294
295 if (factor == 1)
296 return ConvertToFraction(X);
297
298 fValue = fDivide(ConvertToFraction(X * uPow(-1, bNEGATED)), ConvertToFraction(factor));
299
300 fValue.full = fValue.full << times_shifted;
301 fValue.full = fValue.full >> factor_shifted;
302
303 return fValue;
304 }
305
306 /* Addition using two fInts */
307 static fInt fAdd (fInt X, fInt Y)
308 {
309 fInt Sum;
310
311 Sum.full = X.full + Y.full;
312
313 return Sum;
314 }
315
316 /* Addition using two fInts */
317 static fInt fSubtract (fInt X, fInt Y)
318 {
319 fInt Difference;
320
321 Difference.full = X.full - Y.full;
322
323 return Difference;
324 }
325
326 static bool Equal(fInt A, fInt B)
327 {
328 if (A.full == B.full)
329 return true;
330 else
331 return false;
332 }
333
334 static bool GreaterThan(fInt A, fInt B)
335 {
336 if (A.full > B.full)
337 return true;
338 else
339 return false;
340 }
341
342 static fInt fMultiply (fInt X, fInt Y) /* Uses 64-bit integers (int64_t) */
343 {
344 fInt Product;
345 int64_t tempProduct;
346 bool X_LessThanOne, Y_LessThanOne;
347
348 X_LessThanOne = (X.partial.real == 0 && X.partial.decimal != 0 && X.full >= 0);
349 Y_LessThanOne = (Y.partial.real == 0 && Y.partial.decimal != 0 && Y.full >= 0);
350
351 /*The following is for a very specific common case: Non-zero number with ONLY fractional portion*/
352 /* TEMPORARILY DISABLED - CAN BE USED TO IMPROVE PRECISION
353
354 if (X_LessThanOne && Y_LessThanOne) {
355 Product.full = X.full * Y.full;
356 return Product
357 }*/
358
359 tempProduct = ((int64_t)X.full) * ((int64_t)Y.full); /*Q(16,16)*Q(16,16) = Q(32, 32) - Might become a negative number! */
360 tempProduct = tempProduct >> 16; /*Remove lagging 16 bits - Will lose some precision from decimal; */
361 Product.full = (int)tempProduct; /*The int64_t will lose the leading 16 bits that were part of the integer portion */
362
363 return Product;
364 }
365
366 static fInt fDivide (fInt X, fInt Y)
367 {
368 fInt fZERO, fQuotient;
369 int64_t longlongX, longlongY;
370
371 fZERO = ConvertToFraction(0);
372
373 if (Equal(Y, fZERO))
374 return fZERO;
375
376 longlongX = (int64_t)X.full;
377 longlongY = (int64_t)Y.full;
378
379 longlongX = longlongX << 16; /*Q(16,16) -> Q(32,32) */
380
381 div64_s64(longlongX, longlongY); /*Q(32,32) divided by Q(16,16) = Q(16,16) Back to original format */
382
383 fQuotient.full = (int)longlongX;
384 return fQuotient;
385 }
386
387 static int ConvertBackToInteger (fInt A) /*THIS is the function that will be used to check with the Golden settings table*/
388 {
389 fInt fullNumber, scaledDecimal, scaledReal;
390
391 scaledReal.full = GetReal(A) * uPow(10, PRECISION-1); /* DOUBLE CHECK THISSSS!!! */
392
393 scaledDecimal.full = uGetScaledDecimal(A);
394
395 fullNumber = fAdd(scaledDecimal,scaledReal);
396
397 return fullNumber.full;
398 }
399
400 static fInt fGetSquare(fInt A)
401 {
402 return fMultiply(A,A);
403 }
404
405 /* x_new = x_old - (x_old^2 - C) / (2 * x_old) */
406 static fInt fSqrt(fInt num)
407 {
408 fInt F_divide_Fprime, Fprime;
409 fInt test;
410 fInt twoShifted;
411 int seed, counter, error;
412 fInt x_new, x_old, C, y;
413
414 fInt fZERO = ConvertToFraction(0);
415
416 /* (0 > num) is the same as (num < 0), i.e., num is negative */
417
418 if (GreaterThan(fZERO, num) || Equal(fZERO, num))
419 return fZERO;
420
421 C = num;
422
423 if (num.partial.real > 3000)
424 seed = 60;
425 else if (num.partial.real > 1000)
426 seed = 30;
427 else if (num.partial.real > 100)
428 seed = 10;
429 else
430 seed = 2;
431
432 counter = 0;
433
434 if (Equal(num, fZERO)) /*Square Root of Zero is zero */
435 return fZERO;
436
437 twoShifted = ConvertToFraction(2);
438 x_new = ConvertToFraction(seed);
439
440 do {
441 counter++;
442
443 x_old.full = x_new.full;
444
445 test = fGetSquare(x_old); /*1.75*1.75 is reverting back to 1 when shifted down */
446 y = fSubtract(test, C); /*y = f(x) = x^2 - C; */
447
448 Fprime = fMultiply(twoShifted, x_old);
449 F_divide_Fprime = fDivide(y, Fprime);
450
451 x_new = fSubtract(x_old, F_divide_Fprime);
452
453 error = ConvertBackToInteger(x_new) - ConvertBackToInteger(x_old);
454
455 if (counter > 20) /*20 is already way too many iterations. If we dont have an answer by then, we never will*/
456 return x_new;
457
458 } while (uAbs(error) > 0);
459
460 return (x_new);
461 }
462
463 static void SolveQuadracticEqn(fInt A, fInt B, fInt C, fInt Roots[])
464 {
465 fInt *pRoots = &Roots[0];
466 fInt temp, root_first, root_second;
467 fInt f_CONSTANT10, f_CONSTANT100;
468
469 f_CONSTANT100 = ConvertToFraction(100);
470 f_CONSTANT10 = ConvertToFraction(10);
471
472 while(GreaterThan(A, f_CONSTANT100) || GreaterThan(B, f_CONSTANT100) || GreaterThan(C, f_CONSTANT100)) {
473 A = fDivide(A, f_CONSTANT10);
474 B = fDivide(B, f_CONSTANT10);
475 C = fDivide(C, f_CONSTANT10);
476 }
477
478 temp = fMultiply(ConvertToFraction(4), A); /* root = 4*A */
479 temp = fMultiply(temp, C); /* root = 4*A*C */
480 temp = fSubtract(fGetSquare(B), temp); /* root = b^2 - 4AC */
481 temp = fSqrt(temp); /*root = Sqrt (b^2 - 4AC); */
482
483 root_first = fSubtract(fNegate(B), temp); /* b - Sqrt(b^2 - 4AC) */
484 root_second = fAdd(fNegate(B), temp); /* b + Sqrt(b^2 - 4AC) */
485
486 root_first = fDivide(root_first, ConvertToFraction(2)); /* [b +- Sqrt(b^2 - 4AC)]/[2] */
487 root_first = fDivide(root_first, A); /*[b +- Sqrt(b^2 - 4AC)]/[2*A] */
488
489 root_second = fDivide(root_second, ConvertToFraction(2)); /* [b +- Sqrt(b^2 - 4AC)]/[2] */
490 root_second = fDivide(root_second, A); /*[b +- Sqrt(b^2 - 4AC)]/[2*A] */
491
492 *(pRoots + 0) = root_first;
493 *(pRoots + 1) = root_second;
494 }
495
496 /* -----------------------------------------------------------------------------
497 * SUPPORT FUNCTIONS
498 * -----------------------------------------------------------------------------
499 */
500
501 /* Addition using two normal ints - Temporary - Use only for testing purposes?. */
502 static fInt Add (int X, int Y)
503 {
504 fInt A, B, Sum;
505
506 A.full = (X << SHIFT_AMOUNT);
507 B.full = (Y << SHIFT_AMOUNT);
508
509 Sum.full = A.full + B.full;
510
511 return Sum;
512 }
513
514 /* Conversion Functions */
515 static int GetReal (fInt A)
516 {
517 return (A.full >> SHIFT_AMOUNT);
518 }
519
520 /* Temporarily Disabled */
521 static int GetRoundedValue(fInt A) /*For now, round the 3rd decimal place */
522 {
523 /* ROUNDING TEMPORARLY DISABLED
524 int temp = A.full;
525 int decimal_cutoff, decimal_mask = 0x000001FF;
526 decimal_cutoff = temp & decimal_mask;
527 if (decimal_cutoff > 0x147) {
528 temp += 673;
529 }*/
530
531 return ConvertBackToInteger(A)/10000; /*Temporary - in case this was used somewhere else */
532 }
533
534 static fInt Multiply (int X, int Y)
535 {
536 fInt A, B, Product;
537
538 A.full = X << SHIFT_AMOUNT;
539 B.full = Y << SHIFT_AMOUNT;
540
541 Product = fMultiply(A, B);
542
543 return Product;
544 }
545
546 static fInt Divide (int X, int Y)
547 {
548 fInt A, B, Quotient;
549
550 A.full = X << SHIFT_AMOUNT;
551 B.full = Y << SHIFT_AMOUNT;
552
553 Quotient = fDivide(A, B);
554
555 return Quotient;
556 }
557
558 static int uGetScaledDecimal (fInt A) /*Converts the fractional portion to whole integers - Costly function */
559 {
560 int dec[PRECISION];
561 int i, scaledDecimal = 0, tmp = A.partial.decimal;
562
563 for (i = 0; i < PRECISION; i++) {
564 dec[i] = tmp / (1 << SHIFT_AMOUNT);
565 tmp = tmp - ((1 << SHIFT_AMOUNT)*dec[i]);
566 tmp *= 10;
567 scaledDecimal = scaledDecimal + dec[i]*uPow(10, PRECISION - 1 -i);
568 }
569
570 return scaledDecimal;
571 }
572
573 static int uPow(int base, int power)
574 {
575 if (power == 0)
576 return 1;
577 else
578 return (base)*uPow(base, power - 1);
579 }
580
581 static fInt fAbs(fInt A)
582 {
583 if (A.partial.real < 0)
584 return (fMultiply(A, ConvertToFraction(-1)));
585 else
586 return A;
587 }
588
589 static int uAbs(int X)
590 {
591 if (X < 0)
592 return (X * -1);
593 else
594 return X;
595 }
596
597 static fInt fRoundUpByStepSize(fInt A, fInt fStepSize, bool error_term)
598 {
599 fInt solution;
600
601 solution = fDivide(A, fStepSize);
602 solution.partial.decimal = 0; /*All fractional digits changes to 0 */
603
604 if (error_term)
605 solution.partial.real += 1; /*Error term of 1 added */
606
607 solution = fMultiply(solution, fStepSize);
608 solution = fAdd(solution, fStepSize);
609
610 return solution;
611 }
612
This page took 0.044608 seconds and 5 git commands to generate.