cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

poly_tan.c (6947B)


      1// SPDX-License-Identifier: GPL-2.0
      2/*---------------------------------------------------------------------------+
      3 |  poly_tan.c                                                               |
      4 |                                                                           |
      5 | Compute the tan of a FPU_REG, using a polynomial approximation.           |
      6 |                                                                           |
      7 | Copyright (C) 1992,1993,1994,1997,1999                                    |
      8 |                       W. Metzenthen, 22 Parker St, Ormond, Vic 3163,      |
      9 |                       Australia.  E-mail   billm@melbpc.org.au            |
     10 |                                                                           |
     11 |                                                                           |
     12 +---------------------------------------------------------------------------*/
     13
     14#include "exception.h"
     15#include "reg_constant.h"
     16#include "fpu_emu.h"
     17#include "fpu_system.h"
     18#include "control_w.h"
     19#include "poly.h"
     20
     21#define	HiPOWERop	3	/* odd poly, positive terms */
     22static const unsigned long long oddplterm[HiPOWERop] = {
     23	0x0000000000000000LL,
     24	0x0051a1cf08fca228LL,
     25	0x0000000071284ff7LL
     26};
     27
     28#define	HiPOWERon	2	/* odd poly, negative terms */
     29static const unsigned long long oddnegterm[HiPOWERon] = {
     30	0x1291a9a184244e80LL,
     31	0x0000583245819c21LL
     32};
     33
     34#define	HiPOWERep	2	/* even poly, positive terms */
     35static const unsigned long long evenplterm[HiPOWERep] = {
     36	0x0e848884b539e888LL,
     37	0x00003c7f18b887daLL
     38};
     39
     40#define	HiPOWERen	2	/* even poly, negative terms */
     41static const unsigned long long evennegterm[HiPOWERen] = {
     42	0xf1f0200fd51569ccLL,
     43	0x003afb46105c4432LL
     44};
     45
     46static const unsigned long long twothirds = 0xaaaaaaaaaaaaaaabLL;
     47
     48/*--- poly_tan() ------------------------------------------------------------+
     49 |                                                                           |
     50 +---------------------------------------------------------------------------*/
     51void poly_tan(FPU_REG *st0_ptr)
     52{
     53	long int exponent;
     54	int invert;
     55	Xsig argSq, argSqSq, accumulatoro, accumulatore, accum,
     56	    argSignif, fix_up;
     57	unsigned long adj;
     58
     59	exponent = exponent(st0_ptr);
     60
     61#ifdef PARANOID
     62	if (signnegative(st0_ptr)) {	/* Can't hack a number < 0.0 */
     63		arith_invalid(0);
     64		return;
     65	}			/* Need a positive number */
     66#endif /* PARANOID */
     67
     68	/* Split the problem into two domains, smaller and larger than pi/4 */
     69	if ((exponent == 0)
     70	    || ((exponent == -1) && (st0_ptr->sigh > 0xc90fdaa2))) {
     71		/* The argument is greater than (approx) pi/4 */
     72		invert = 1;
     73		accum.lsw = 0;
     74		XSIG_LL(accum) = significand(st0_ptr);
     75
     76		if (exponent == 0) {
     77			/* The argument is >= 1.0 */
     78			/* Put the binary point at the left. */
     79			XSIG_LL(accum) <<= 1;
     80		}
     81		/* pi/2 in hex is: 1.921fb54442d18469 898CC51701B839A2 52049C1 */
     82		XSIG_LL(accum) = 0x921fb54442d18469LL - XSIG_LL(accum);
     83		/* This is a special case which arises due to rounding. */
     84		if (XSIG_LL(accum) == 0xffffffffffffffffLL) {
     85			FPU_settag0(TAG_Valid);
     86			significand(st0_ptr) = 0x8a51e04daabda360LL;
     87			setexponent16(st0_ptr,
     88				      (0x41 + EXTENDED_Ebias) | SIGN_Negative);
     89			return;
     90		}
     91
     92		argSignif.lsw = accum.lsw;
     93		XSIG_LL(argSignif) = XSIG_LL(accum);
     94		exponent = -1 + norm_Xsig(&argSignif);
     95	} else {
     96		invert = 0;
     97		argSignif.lsw = 0;
     98		XSIG_LL(accum) = XSIG_LL(argSignif) = significand(st0_ptr);
     99
    100		if (exponent < -1) {
    101			/* shift the argument right by the required places */
    102			if (FPU_shrx(&XSIG_LL(accum), -1 - exponent) >=
    103			    0x80000000U)
    104				XSIG_LL(accum)++;	/* round up */
    105		}
    106	}
    107
    108	XSIG_LL(argSq) = XSIG_LL(accum);
    109	argSq.lsw = accum.lsw;
    110	mul_Xsig_Xsig(&argSq, &argSq);
    111	XSIG_LL(argSqSq) = XSIG_LL(argSq);
    112	argSqSq.lsw = argSq.lsw;
    113	mul_Xsig_Xsig(&argSqSq, &argSqSq);
    114
    115	/* Compute the negative terms for the numerator polynomial */
    116	accumulatoro.msw = accumulatoro.midw = accumulatoro.lsw = 0;
    117	polynomial_Xsig(&accumulatoro, &XSIG_LL(argSqSq), oddnegterm,
    118			HiPOWERon - 1);
    119	mul_Xsig_Xsig(&accumulatoro, &argSq);
    120	negate_Xsig(&accumulatoro);
    121	/* Add the positive terms */
    122	polynomial_Xsig(&accumulatoro, &XSIG_LL(argSqSq), oddplterm,
    123			HiPOWERop - 1);
    124
    125	/* Compute the positive terms for the denominator polynomial */
    126	accumulatore.msw = accumulatore.midw = accumulatore.lsw = 0;
    127	polynomial_Xsig(&accumulatore, &XSIG_LL(argSqSq), evenplterm,
    128			HiPOWERep - 1);
    129	mul_Xsig_Xsig(&accumulatore, &argSq);
    130	negate_Xsig(&accumulatore);
    131	/* Add the negative terms */
    132	polynomial_Xsig(&accumulatore, &XSIG_LL(argSqSq), evennegterm,
    133			HiPOWERen - 1);
    134	/* Multiply by arg^2 */
    135	mul64_Xsig(&accumulatore, &XSIG_LL(argSignif));
    136	mul64_Xsig(&accumulatore, &XSIG_LL(argSignif));
    137	/* de-normalize and divide by 2 */
    138	shr_Xsig(&accumulatore, -2 * (1 + exponent) + 1);
    139	negate_Xsig(&accumulatore);	/* This does 1 - accumulator */
    140
    141	/* Now find the ratio. */
    142	if (accumulatore.msw == 0) {
    143		/* accumulatoro must contain 1.0 here, (actually, 0) but it
    144		   really doesn't matter what value we use because it will
    145		   have negligible effect in later calculations
    146		 */
    147		XSIG_LL(accum) = 0x8000000000000000LL;
    148		accum.lsw = 0;
    149	} else {
    150		div_Xsig(&accumulatoro, &accumulatore, &accum);
    151	}
    152
    153	/* Multiply by 1/3 * arg^3 */
    154	mul64_Xsig(&accum, &XSIG_LL(argSignif));
    155	mul64_Xsig(&accum, &XSIG_LL(argSignif));
    156	mul64_Xsig(&accum, &XSIG_LL(argSignif));
    157	mul64_Xsig(&accum, &twothirds);
    158	shr_Xsig(&accum, -2 * (exponent + 1));
    159
    160	/* tan(arg) = arg + accum */
    161	add_two_Xsig(&accum, &argSignif, &exponent);
    162
    163	if (invert) {
    164		/* We now have the value of tan(pi_2 - arg) where pi_2 is an
    165		   approximation for pi/2
    166		 */
    167		/* The next step is to fix the answer to compensate for the
    168		   error due to the approximation used for pi/2
    169		 */
    170
    171		/* This is (approx) delta, the error in our approx for pi/2
    172		   (see above). It has an exponent of -65
    173		 */
    174		XSIG_LL(fix_up) = 0x898cc51701b839a2LL;
    175		fix_up.lsw = 0;
    176
    177		if (exponent == 0)
    178			adj = 0xffffffff;	/* We want approx 1.0 here, but
    179						   this is close enough. */
    180		else if (exponent > -30) {
    181			adj = accum.msw >> -(exponent + 1);	/* tan */
    182			adj = mul_32_32(adj, adj);	/* tan^2 */
    183		} else
    184			adj = 0;
    185		adj = mul_32_32(0x898cc517, adj);	/* delta * tan^2 */
    186
    187		fix_up.msw += adj;
    188		if (!(fix_up.msw & 0x80000000)) {	/* did fix_up overflow ? */
    189			/* Yes, we need to add an msb */
    190			shr_Xsig(&fix_up, 1);
    191			fix_up.msw |= 0x80000000;
    192			shr_Xsig(&fix_up, 64 + exponent);
    193		} else
    194			shr_Xsig(&fix_up, 65 + exponent);
    195
    196		add_two_Xsig(&accum, &fix_up, &exponent);
    197
    198		/* accum now contains tan(pi/2 - arg).
    199		   Use tan(arg) = 1.0 / tan(pi/2 - arg)
    200		 */
    201		accumulatoro.lsw = accumulatoro.midw = 0;
    202		accumulatoro.msw = 0x80000000;
    203		div_Xsig(&accumulatoro, &accum, &accum);
    204		exponent = -exponent - 1;
    205	}
    206
    207	/* Transfer the result */
    208	round_Xsig(&accum);
    209	FPU_settag0(TAG_Valid);
    210	significand(st0_ptr) = XSIG_LL(accum);
    211	setexponent16(st0_ptr, exponent + EXTENDED_Ebias);	/* Result is positive. */
    212
    213}