Home | History | Annotate | Line # | Download | only in fpsp
setox.sa revision 1.1
      1 *	MOTOROLA MICROPROCESSOR & MEMORY TECHNOLOGY GROUP
      2 *	M68000 Hi-Performance Microprocessor Division
      3 *	M68040 Software Package 
      4 *
      5 *	M68040 Software Package Copyright (c) 1993, 1994 Motorola Inc.
      6 *	All rights reserved.
      7 *
      8 *	THE SOFTWARE is provided on an "AS IS" basis and without warranty.
      9 *	To the maximum extent permitted by applicable law,
     10 *	MOTOROLA DISCLAIMS ALL WARRANTIES WHETHER EXPRESS OR IMPLIED,
     11 *	INCLUDING IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A
     12 *	PARTICULAR PURPOSE and any warranty against infringement with
     13 *	regard to the SOFTWARE (INCLUDING ANY MODIFIED VERSIONS THEREOF)
     14 *	and any accompanying written materials. 
     15 *
     16 *	To the maximum extent permitted by applicable law,
     17 *	IN NO EVENT SHALL MOTOROLA BE LIABLE FOR ANY DAMAGES WHATSOEVER
     18 *	(INCLUDING WITHOUT LIMITATION, DAMAGES FOR LOSS OF BUSINESS
     19 *	PROFITS, BUSINESS INTERRUPTION, LOSS OF BUSINESS INFORMATION, OR
     20 *	OTHER PECUNIARY LOSS) ARISING OF THE USE OR INABILITY TO USE THE
     21 *	SOFTWARE.  Motorola assumes no responsibility for the maintenance
     22 *	and support of the SOFTWARE.  
     23 *
     24 *	You are hereby granted a copyright license to use, modify, and
     25 *	distribute the SOFTWARE so long as this entire notice is retained
     26 *	without alteration in any modified and/or redistributed versions,
     27 *	and that such modified versions are clearly identified as such.
     28 *	No licenses are granted by implication, estoppel or otherwise
     29 *	under any patents or trademarks of Motorola, Inc.
     30 
     31 *
     32 *	setox.sa 3.1 12/10/90
     33 *
     34 *	The entry point setox computes the exponential of a value.
     35 *	setoxd does the same except the input value is a denormalized
     36 *	number.	setoxm1 computes exp(X)-1, and setoxm1d computes
     37 *	exp(X)-1 for denormalized X.
     38 *
     39 *	INPUT
     40 *	-----
     41 *	Double-extended value in memory location pointed to by address
     42 *	register a0.
     43 *
     44 *	OUTPUT
     45 *	------
     46 *	exp(X) or exp(X)-1 returned in floating-point register fp0.
     47 *
     48 *	ACCURACY and MONOTONICITY
     49 *	-------------------------
     50 *	The returned result is within 0.85 ulps in 64 significant bit, i.e.
     51 *	within 0.5001 ulp to 53 bits if the result is subsequently rounded
     52 *	to double precision. The result is provably monotonic in double
     53 *	precision.
     54 *
     55 *	SPEED
     56 *	-----
     57 *	Two timings are measured, both in the copy-back mode. The
     58 *	first one is measured when the function is invoked the first time
     59 *	(so the instructions and data are not in cache), and the
     60 *	second one is measured when the function is reinvoked at the same
     61 *	input argument.
     62 *
     63 *	The program setox takes approximately 210/190 cycles for input
     64 *	argument X whose magnitude is less than 16380 log2, which
     65 *	is the usual situation.	For the less common arguments,
     66 *	depending on their values, the program may run faster or slower --
     67 *	but no worse than 10% slower even in the extreme cases.
     68 *
     69 *	The program setoxm1 takes approximately ???/??? cycles for input
     70 *	argument X, 0.25 <= |X| < 70log2. For |X| < 0.25, it takes
     71 *	approximately ???/??? cycles. For the less common arguments,
     72 *	depending on their values, the program may run faster or slower --
     73 *	but no worse than 10% slower even in the extreme cases.
     74 *
     75 *	ALGORITHM and IMPLEMENTATION NOTES
     76 *	----------------------------------
     77 *
     78 *	setoxd
     79 *	------
     80 *	Step 1.	Set ans := 1.0
     81 *
     82 *	Step 2.	Return	ans := ans + sign(X)*2^(-126). Exit.
     83 *	Notes:	This will always generate one exception -- inexact.
     84 *
     85 *
     86 *	setox
     87 *	-----
     88 *
     89 *	Step 1.	Filter out extreme cases of input argument.
     90 *		1.1	If |X| >= 2^(-65), go to Step 1.3.
     91 *		1.2	Go to Step 7.
     92 *		1.3	If |X| < 16380 log(2), go to Step 2.
     93 *		1.4	Go to Step 8.
     94 *	Notes:	The usual case should take the branches 1.1 -> 1.3 -> 2.
     95 *		 To avoid the use of floating-point comparisons, a
     96 *		 compact representation of |X| is used. This format is a
     97 *		 32-bit integer, the upper (more significant) 16 bits are
     98 *		 the sign and biased exponent field of |X|; the lower 16
     99 *		 bits are the 16 most significant fraction (including the
    100 *		 explicit bit) bits of |X|. Consequently, the comparisons
    101 *		 in Steps 1.1 and 1.3 can be performed by integer comparison.
    102 *		 Note also that the constant 16380 log(2) used in Step 1.3
    103 *		 is also in the compact form. Thus taking the branch
    104 *		 to Step 2 guarantees |X| < 16380 log(2). There is no harm
    105 *		 to have a small number of cases where |X| is less than,
    106 *		 but close to, 16380 log(2) and the branch to Step 9 is
    107 *		 taken.
    108 *
    109 *	Step 2.	Calculate N = round-to-nearest-int( X * 64/log2 ).
    110 *		2.1	Set AdjFlag := 0 (indicates the branch 1.3 -> 2 was taken)
    111 *		2.2	N := round-to-nearest-integer( X * 64/log2 ).
    112 *		2.3	Calculate	J = N mod 64; so J = 0,1,2,..., or 63.
    113 *		2.4	Calculate	M = (N - J)/64; so N = 64M + J.
    114 *		2.5	Calculate the address of the stored value of 2^(J/64).
    115 *		2.6	Create the value Scale = 2^M.
    116 *	Notes:	The calculation in 2.2 is really performed by
    117 *
    118 *			Z := X * constant
    119 *			N := round-to-nearest-integer(Z)
    120 *
    121 *		 where
    122 *
    123 *			constant := single-precision( 64/log 2 ).
    124 *
    125 *		 Using a single-precision constant avoids memory access.
    126 *		 Another effect of using a single-precision "constant" is
    127 *		 that the calculated value Z is
    128 *
    129 *			Z = X*(64/log2)*(1+eps), |eps| <= 2^(-24).
    130 *
    131 *		 This error has to be considered later in Steps 3 and 4.
    132 *
    133 *	Step 3.	Calculate X - N*log2/64.
    134 *		3.1	R := X + N*L1, where L1 := single-precision(-log2/64).
    135 *		3.2	R := R + N*L2, L2 := extended-precision(-log2/64 - L1).
    136 *	Notes:	a) The way L1 and L2 are chosen ensures L1+L2 approximate
    137 *		 the value	-log2/64	to 88 bits of accuracy.
    138 *		 b) N*L1 is exact because N is no longer than 22 bits and
    139 *		 L1 is no longer than 24 bits.
    140 *		 c) The calculation X+N*L1 is also exact due to cancellation.
    141 *		 Thus, R is practically X+N(L1+L2) to full 64 bits.
    142 *		 d) It is important to estimate how large can |R| be after
    143 *		 Step 3.2.
    144 *
    145 *			N = rnd-to-int( X*64/log2 (1+eps) ), |eps|<=2^(-24)
    146 *			X*64/log2 (1+eps)	=	N + f,	|f| <= 0.5
    147 *			X*64/log2 - N	=	f - eps*X 64/log2
    148 *			X - N*log2/64	=	f*log2/64 - eps*X
    149 *
    150 *
    151 *		 Now |X| <= 16446 log2, thus
    152 *
    153 *			|X - N*log2/64| <= (0.5 + 16446/2^(18))*log2/64
    154 *					<= 0.57 log2/64.
    155 *		 This bound will be used in Step 4.
    156 *
    157 *	Step 4.	Approximate exp(R)-1 by a polynomial
    158 *			p = R + R*R*(A1 + R*(A2 + R*(A3 + R*(A4 + R*A5))))
    159 *	Notes:	a) In order to reduce memory access, the coefficients are
    160 *		 made as "short" as possible: A1 (which is 1/2), A4 and A5
    161 *		 are single precision; A2 and A3 are double precision.
    162 *		 b) Even with the restrictions above,
    163 *			|p - (exp(R)-1)| < 2^(-68.8) for all |R| <= 0.0062.
    164 *		 Note that 0.0062 is slightly bigger than 0.57 log2/64.
    165 *		 c) To fully utilize the pipeline, p is separated into
    166 *		 two independent pieces of roughly equal complexities
    167 *			p = [ R + R*S*(A2 + S*A4) ]	+
    168 *				[ S*(A1 + S*(A3 + S*A5)) ]
    169 *		 where S = R*R.
    170 *
    171 *	Step 5.	Compute 2^(J/64)*exp(R) = 2^(J/64)*(1+p) by
    172 *				ans := T + ( T*p + t)
    173 *		 where T and t are the stored values for 2^(J/64).
    174 *	Notes:	2^(J/64) is stored as T and t where T+t approximates
    175 *		 2^(J/64) to roughly 85 bits; T is in extended precision
    176 *		 and t is in single precision. Note also that T is rounded
    177 *		 to 62 bits so that the last two bits of T are zero. The
    178 *		 reason for such a special form is that T-1, T-2, and T-8
    179 *		 will all be exact --- a property that will give much
    180 *		 more accurate computation of the function EXPM1.
    181 *
    182 *	Step 6.	Reconstruction of exp(X)
    183 *			exp(X) = 2^M * 2^(J/64) * exp(R).
    184 *		6.1	If AdjFlag = 0, go to 6.3
    185 *		6.2	ans := ans * AdjScale
    186 *		6.3	Restore the user FPCR
    187 *		6.4	Return ans := ans * Scale. Exit.
    188 *	Notes:	If AdjFlag = 0, we have X = Mlog2 + Jlog2/64 + R,
    189 *		 |M| <= 16380, and Scale = 2^M. Moreover, exp(X) will
    190 *		 neither overflow nor underflow. If AdjFlag = 1, that
    191 *		 means that
    192 *			X = (M1+M)log2 + Jlog2/64 + R, |M1+M| >= 16380.
    193 *		 Hence, exp(X) may overflow or underflow or neither.
    194 *		 When that is the case, AdjScale = 2^(M1) where M1 is
    195 *		 approximately M. Thus 6.2 will never cause over/underflow.
    196 *		 Possible exception in 6.4 is overflow or underflow.
    197 *		 The inexact exception is not generated in 6.4. Although
    198 *		 one can argue that the inexact flag should always be
    199 *		 raised, to simulate that exception cost to much than the
    200 *		 flag is worth in practical uses.
    201 *
    202 *	Step 7.	Return 1 + X.
    203 *		7.1	ans := X
    204 *		7.2	Restore user FPCR.
    205 *		7.3	Return ans := 1 + ans. Exit
    206 *	Notes:	For non-zero X, the inexact exception will always be
    207 *		 raised by 7.3. That is the only exception raised by 7.3.
    208 *		 Note also that we use the FMOVEM instruction to move X
    209 *		 in Step 7.1 to avoid unnecessary trapping. (Although
    210 *		 the FMOVEM may not seem relevant since X is normalized,
    211 *		 the precaution will be useful in the library version of
    212 *		 this code where the separate entry for denormalized inputs
    213 *		 will be done away with.)
    214 *
    215 *	Step 8.	Handle exp(X) where |X| >= 16380log2.
    216 *		8.1	If |X| > 16480 log2, go to Step 9.
    217 *		(mimic 2.2 - 2.6)
    218 *		8.2	N := round-to-integer( X * 64/log2 )
    219 *		8.3	Calculate J = N mod 64, J = 0,1,...,63
    220 *		8.4	K := (N-J)/64, M1 := truncate(K/2), M = K-M1, AdjFlag := 1.
    221 *		8.5	Calculate the address of the stored value 2^(J/64).
    222 *		8.6	Create the values Scale = 2^M, AdjScale = 2^M1.
    223 *		8.7	Go to Step 3.
    224 *	Notes:	Refer to notes for 2.2 - 2.6.
    225 *
    226 *	Step 9.	Handle exp(X), |X| > 16480 log2.
    227 *		9.1	If X < 0, go to 9.3
    228 *		9.2	ans := Huge, go to 9.4
    229 *		9.3	ans := Tiny.
    230 *		9.4	Restore user FPCR.
    231 *		9.5	Return ans := ans * ans. Exit.
    232 *	Notes:	Exp(X) will surely overflow or underflow, depending on
    233 *		 X's sign. "Huge" and "Tiny" are respectively large/tiny
    234 *		 extended-precision numbers whose square over/underflow
    235 *		 with an inexact result. Thus, 9.5 always raises the
    236 *		 inexact together with either overflow or underflow.
    237 *
    238 *
    239 *	setoxm1d
    240 *	--------
    241 *
    242 *	Step 1.	Set ans := 0
    243 *
    244 *	Step 2.	Return	ans := X + ans. Exit.
    245 *	Notes:	This will return X with the appropriate rounding
    246 *		 precision prescribed by the user FPCR.
    247 *
    248 *	setoxm1
    249 *	-------
    250 *
    251 *	Step 1.	Check |X|
    252 *		1.1	If |X| >= 1/4, go to Step 1.3.
    253 *		1.2	Go to Step 7.
    254 *		1.3	If |X| < 70 log(2), go to Step 2.
    255 *		1.4	Go to Step 10.
    256 *	Notes:	The usual case should take the branches 1.1 -> 1.3 -> 2.
    257 *		 However, it is conceivable |X| can be small very often
    258 *		 because EXPM1 is intended to evaluate exp(X)-1 accurately
    259 *		 when |X| is small. For further details on the comparisons,
    260 *		 see the notes on Step 1 of setox.
    261 *
    262 *	Step 2.	Calculate N = round-to-nearest-int( X * 64/log2 ).
    263 *		2.1	N := round-to-nearest-integer( X * 64/log2 ).
    264 *		2.2	Calculate	J = N mod 64; so J = 0,1,2,..., or 63.
    265 *		2.3	Calculate	M = (N - J)/64; so N = 64M + J.
    266 *		2.4	Calculate the address of the stored value of 2^(J/64).
    267 *		2.5	Create the values Sc = 2^M and OnebySc := -2^(-M).
    268 *	Notes:	See the notes on Step 2 of setox.
    269 *
    270 *	Step 3.	Calculate X - N*log2/64.
    271 *		3.1	R := X + N*L1, where L1 := single-precision(-log2/64).
    272 *		3.2	R := R + N*L2, L2 := extended-precision(-log2/64 - L1).
    273 *	Notes:	Applying the analysis of Step 3 of setox in this case
    274 *		 shows that |R| <= 0.0055 (note that |X| <= 70 log2 in
    275 *		 this case).
    276 *
    277 *	Step 4.	Approximate exp(R)-1 by a polynomial
    278 *			p = R+R*R*(A1+R*(A2+R*(A3+R*(A4+R*(A5+R*A6)))))
    279 *	Notes:	a) In order to reduce memory access, the coefficients are
    280 *		 made as "short" as possible: A1 (which is 1/2), A5 and A6
    281 *		 are single precision; A2, A3 and A4 are double precision.
    282 *		 b) Even with the restriction above,
    283 *			|p - (exp(R)-1)| <	|R| * 2^(-72.7)
    284 *		 for all |R| <= 0.0055.
    285 *		 c) To fully utilize the pipeline, p is separated into
    286 *		 two independent pieces of roughly equal complexity
    287 *			p = [ R*S*(A2 + S*(A4 + S*A6)) ]	+
    288 *				[ R + S*(A1 + S*(A3 + S*A5)) ]
    289 *		 where S = R*R.
    290 *
    291 *	Step 5.	Compute 2^(J/64)*p by
    292 *				p := T*p
    293 *		 where T and t are the stored values for 2^(J/64).
    294 *	Notes:	2^(J/64) is stored as T and t where T+t approximates
    295 *		 2^(J/64) to roughly 85 bits; T is in extended precision
    296 *		 and t is in single precision. Note also that T is rounded
    297 *		 to 62 bits so that the last two bits of T are zero. The
    298 *		 reason for such a special form is that T-1, T-2, and T-8
    299 *		 will all be exact --- a property that will be exploited
    300 *		 in Step 6 below. The total relative error in p is no
    301 *		 bigger than 2^(-67.7) compared to the final result.
    302 *
    303 *	Step 6.	Reconstruction of exp(X)-1
    304 *			exp(X)-1 = 2^M * ( 2^(J/64) + p - 2^(-M) ).
    305 *		6.1	If M <= 63, go to Step 6.3.
    306 *		6.2	ans := T + (p + (t + OnebySc)). Go to 6.6
    307 *		6.3	If M >= -3, go to 6.5.
    308 *		6.4	ans := (T + (p + t)) + OnebySc. Go to 6.6
    309 *		6.5	ans := (T + OnebySc) + (p + t).
    310 *		6.6	Restore user FPCR.
    311 *		6.7	Return ans := Sc * ans. Exit.
    312 *	Notes:	The various arrangements of the expressions give accurate
    313 *		 evaluations.
    314 *
    315 *	Step 7.	exp(X)-1 for |X| < 1/4.
    316 *		7.1	If |X| >= 2^(-65), go to Step 9.
    317 *		7.2	Go to Step 8.
    318 *
    319 *	Step 8.	Calculate exp(X)-1, |X| < 2^(-65).
    320 *		8.1	If |X| < 2^(-16312), goto 8.3
    321 *		8.2	Restore FPCR; return ans := X - 2^(-16382). Exit.
    322 *		8.3	X := X * 2^(140).
    323 *		8.4	Restore FPCR; ans := ans - 2^(-16382).
    324 *		 Return ans := ans*2^(140). Exit
    325 *	Notes:	The idea is to return "X - tiny" under the user
    326 *		 precision and rounding modes. To avoid unnecessary
    327 *		 inefficiency, we stay away from denormalized numbers the
    328 *		 best we can. For |X| >= 2^(-16312), the straightforward
    329 *		 8.2 generates the inexact exception as the case warrants.
    330 *
    331 *	Step 9.	Calculate exp(X)-1, |X| < 1/4, by a polynomial
    332 *			p = X + X*X*(B1 + X*(B2 + ... + X*B12))
    333 *	Notes:	a) In order to reduce memory access, the coefficients are
    334 *		 made as "short" as possible: B1 (which is 1/2), B9 to B12
    335 *		 are single precision; B3 to B8 are double precision; and
    336 *		 B2 is double extended.
    337 *		 b) Even with the restriction above,
    338 *			|p - (exp(X)-1)| < |X| 2^(-70.6)
    339 *		 for all |X| <= 0.251.
    340 *		 Note that 0.251 is slightly bigger than 1/4.
    341 *		 c) To fully preserve accuracy, the polynomial is computed
    342 *		 as	X + ( S*B1 +	Q ) where S = X*X and
    343 *			Q	=	X*S*(B2 + X*(B3 + ... + X*B12))
    344 *		 d) To fully utilize the pipeline, Q is separated into
    345 *		 two independent pieces of roughly equal complexity
    346 *			Q = [ X*S*(B2 + S*(B4 + ... + S*B12)) ] +
    347 *				[ S*S*(B3 + S*(B5 + ... + S*B11)) ]
    348 *
    349 *	Step 10.	Calculate exp(X)-1 for |X| >= 70 log 2.
    350 *		10.1 If X >= 70log2 , exp(X) - 1 = exp(X) for all practical
    351 *		 purposes. Therefore, go to Step 1 of setox.
    352 *		10.2 If X <= -70log2, exp(X) - 1 = -1 for all practical purposes.
    353 *		 ans := -1
    354 *		 Restore user FPCR
    355 *		 Return ans := ans + 2^(-126). Exit.
    356 *	Notes:	10.2 will always create an inexact and return -1 + tiny
    357 *		 in the user rounding precision and mode.
    358 *
    359 
    360 setox	IDNT	2,1 Motorola 040 Floating Point Software Package
    361 
    362 	section	8
    363 
    364 	include	fpsp.h
    365 
    366 L2	DC.L	$3FDC0000,$82E30865,$4361C4C6,$00000000
    367 
    368 EXPA3	DC.L	$3FA55555,$55554431
    369 EXPA2	DC.L	$3FC55555,$55554018
    370 
    371 HUGE	DC.L	$7FFE0000,$FFFFFFFF,$FFFFFFFF,$00000000
    372 TINY	DC.L	$00010000,$FFFFFFFF,$FFFFFFFF,$00000000
    373 
    374 EM1A4	DC.L	$3F811111,$11174385
    375 EM1A3	DC.L	$3FA55555,$55554F5A
    376 
    377 EM1A2	DC.L	$3FC55555,$55555555,$00000000,$00000000
    378 
    379 EM1B8	DC.L	$3EC71DE3,$A5774682
    380 EM1B7	DC.L	$3EFA01A0,$19D7CB68
    381 
    382 EM1B6	DC.L	$3F2A01A0,$1A019DF3
    383 EM1B5	DC.L	$3F56C16C,$16C170E2
    384 
    385 EM1B4	DC.L	$3F811111,$11111111
    386 EM1B3	DC.L	$3FA55555,$55555555
    387 
    388 EM1B2	DC.L	$3FFC0000,$AAAAAAAA,$AAAAAAAB
    389 	DC.L	$00000000
    390 
    391 TWO140	DC.L	$48B00000,$00000000
    392 TWON140	DC.L	$37300000,$00000000
    393 
    394 EXPTBL
    395 	DC.L	$3FFF0000,$80000000,$00000000,$00000000
    396 	DC.L	$3FFF0000,$8164D1F3,$BC030774,$9F841A9B
    397 	DC.L	$3FFF0000,$82CD8698,$AC2BA1D8,$9FC1D5B9
    398 	DC.L	$3FFF0000,$843A28C3,$ACDE4048,$A0728369
    399 	DC.L	$3FFF0000,$85AAC367,$CC487B14,$1FC5C95C
    400 	DC.L	$3FFF0000,$871F6196,$9E8D1010,$1EE85C9F
    401 	DC.L	$3FFF0000,$88980E80,$92DA8528,$9FA20729
    402 	DC.L	$3FFF0000,$8A14D575,$496EFD9C,$A07BF9AF
    403 	DC.L	$3FFF0000,$8B95C1E3,$EA8BD6E8,$A0020DCF
    404 	DC.L	$3FFF0000,$8D1ADF5B,$7E5BA9E4,$205A63DA
    405 	DC.L	$3FFF0000,$8EA4398B,$45CD53C0,$1EB70051
    406 	DC.L	$3FFF0000,$9031DC43,$1466B1DC,$1F6EB029
    407 	DC.L	$3FFF0000,$91C3D373,$AB11C338,$A0781494
    408 	DC.L	$3FFF0000,$935A2B2F,$13E6E92C,$9EB319B0
    409 	DC.L	$3FFF0000,$94F4EFA8,$FEF70960,$2017457D
    410 	DC.L	$3FFF0000,$96942D37,$20185A00,$1F11D537
    411 	DC.L	$3FFF0000,$9837F051,$8DB8A970,$9FB952DD
    412 	DC.L	$3FFF0000,$99E04593,$20B7FA64,$1FE43087
    413 	DC.L	$3FFF0000,$9B8D39B9,$D54E5538,$1FA2A818
    414 	DC.L	$3FFF0000,$9D3ED9A7,$2CFFB750,$1FDE494D
    415 	DC.L	$3FFF0000,$9EF53260,$91A111AC,$20504890
    416 	DC.L	$3FFF0000,$A0B0510F,$B9714FC4,$A073691C
    417 	DC.L	$3FFF0000,$A2704303,$0C496818,$1F9B7A05
    418 	DC.L	$3FFF0000,$A43515AE,$09E680A0,$A0797126
    419 	DC.L	$3FFF0000,$A5FED6A9,$B15138EC,$A071A140
    420 	DC.L	$3FFF0000,$A7CD93B4,$E9653568,$204F62DA
    421 	DC.L	$3FFF0000,$A9A15AB4,$EA7C0EF8,$1F283C4A
    422 	DC.L	$3FFF0000,$AB7A39B5,$A93ED338,$9F9A7FDC
    423 	DC.L	$3FFF0000,$AD583EEA,$42A14AC8,$A05B3FAC
    424 	DC.L	$3FFF0000,$AF3B78AD,$690A4374,$1FDF2610
    425 	DC.L	$3FFF0000,$B123F581,$D2AC2590,$9F705F90
    426 	DC.L	$3FFF0000,$B311C412,$A9112488,$201F678A
    427 	DC.L	$3FFF0000,$B504F333,$F9DE6484,$1F32FB13
    428 	DC.L	$3FFF0000,$B6FD91E3,$28D17790,$20038B30
    429 	DC.L	$3FFF0000,$B8FBAF47,$62FB9EE8,$200DC3CC
    430 	DC.L	$3FFF0000,$BAFF5AB2,$133E45FC,$9F8B2AE6
    431 	DC.L	$3FFF0000,$BD08A39F,$580C36C0,$A02BBF70
    432 	DC.L	$3FFF0000,$BF1799B6,$7A731084,$A00BF518
    433 	DC.L	$3FFF0000,$C12C4CCA,$66709458,$A041DD41
    434 	DC.L	$3FFF0000,$C346CCDA,$24976408,$9FDF137B
    435 	DC.L	$3FFF0000,$C5672A11,$5506DADC,$201F1568
    436 	DC.L	$3FFF0000,$C78D74C8,$ABB9B15C,$1FC13A2E
    437 	DC.L	$3FFF0000,$C9B9BD86,$6E2F27A4,$A03F8F03
    438 	DC.L	$3FFF0000,$CBEC14FE,$F2727C5C,$1FF4907D
    439 	DC.L	$3FFF0000,$CE248C15,$1F8480E4,$9E6E53E4
    440 	DC.L	$3FFF0000,$D06333DA,$EF2B2594,$1FD6D45C
    441 	DC.L	$3FFF0000,$D2A81D91,$F12AE45C,$A076EDB9
    442 	DC.L	$3FFF0000,$D4F35AAB,$CFEDFA20,$9FA6DE21
    443 	DC.L	$3FFF0000,$D744FCCA,$D69D6AF4,$1EE69A2F
    444 	DC.L	$3FFF0000,$D99D15C2,$78AFD7B4,$207F439F
    445 	DC.L	$3FFF0000,$DBFBB797,$DAF23754,$201EC207
    446 	DC.L	$3FFF0000,$DE60F482,$5E0E9124,$9E8BE175
    447 	DC.L	$3FFF0000,$E0CCDEEC,$2A94E110,$20032C4B
    448 	DC.L	$3FFF0000,$E33F8972,$BE8A5A50,$2004DFF5
    449 	DC.L	$3FFF0000,$E5B906E7,$7C8348A8,$1E72F47A
    450 	DC.L	$3FFF0000,$E8396A50,$3C4BDC68,$1F722F22
    451 	DC.L	$3FFF0000,$EAC0C6E7,$DD243930,$A017E945
    452 	DC.L	$3FFF0000,$ED4F301E,$D9942B84,$1F401A5B
    453 	DC.L	$3FFF0000,$EFE4B99B,$DCDAF5CC,$9FB9A9E3
    454 	DC.L	$3FFF0000,$F281773C,$59FFB138,$20744C05
    455 	DC.L	$3FFF0000,$F5257D15,$2486CC2C,$1F773A19
    456 	DC.L	$3FFF0000,$F7D0DF73,$0AD13BB8,$1FFE90D5
    457 	DC.L	$3FFF0000,$FA83B2DB,$722A033C,$A041ED22
    458 	DC.L	$3FFF0000,$FD3E0C0C,$F486C174,$1F853F3A
    459 
    460 ADJFLAG	equ L_SCR2
    461 SCALE	equ FP_SCR1
    462 ADJSCALE equ FP_SCR2
    463 SC	equ FP_SCR3
    464 ONEBYSC	equ FP_SCR4
    465 
    466 	xref	t_frcinx
    467 	xref	t_extdnrm
    468 	xref	t_unfl
    469 	xref	t_ovfl
    470 
    471 	xdef	setoxd
    472 setoxd:
    473 *--entry point for EXP(X), X is denormalized
    474 	MOVE.L		(a0),d0
    475 	ANDI.L		#$80000000,d0
    476 	ORI.L		#$00800000,d0		...sign(X)*2^(-126)
    477 	MOVE.L		d0,-(sp)
    478 	FMOVE.S		#:3F800000,fp0
    479 	fmove.l		d1,fpcr
    480 	FADD.S		(sp)+,fp0
    481 	bra		t_frcinx
    482 
    483 	xdef	setox
    484 setox:
    485 *--entry point for EXP(X), here X is finite, non-zero, and not NaN's
    486 
    487 *--Step 1.
    488 	MOVE.L		(a0),d0	 ...load part of input X
    489 	ANDI.L		#$7FFF0000,d0	...biased expo. of X
    490 	CMPI.L		#$3FBE0000,d0	...2^(-65)
    491 	BGE.B		EXPC1		...normal case
    492 	BRA.W		EXPSM
    493 
    494 EXPC1:
    495 *--The case |X| >= 2^(-65)
    496 	MOVE.W		4(a0),d0	...expo. and partial sig. of |X|
    497 	CMPI.L		#$400CB167,d0	...16380 log2 trunc. 16 bits
    498 	BLT.B		EXPMAIN	 ...normal case
    499 	BRA.W		EXPBIG
    500 
    501 EXPMAIN:
    502 *--Step 2.
    503 *--This is the normal branch:	2^(-65) <= |X| < 16380 log2.
    504 	FMOVE.X		(a0),fp0	...load input from (a0)
    505 
    506 	FMOVE.X		fp0,fp1
    507 	FMUL.S		#:42B8AA3B,fp0	...64/log2 * X
    508 	fmovem.x	fp2/fp3,-(a7)		...save fp2
    509 	MOVE.L		#0,ADJFLAG(a6)
    510 	FMOVE.L		fp0,d0		...N = int( X * 64/log2 )
    511 	LEA		EXPTBL,a1
    512 	FMOVE.L		d0,fp0		...convert to floating-format
    513 
    514 	MOVE.L		d0,L_SCR1(a6)	...save N temporarily
    515 	ANDI.L		#$3F,d0		...D0 is J = N mod 64
    516 	LSL.L		#4,d0
    517 	ADDA.L		d0,a1		...address of 2^(J/64)
    518 	MOVE.L		L_SCR1(a6),d0
    519 	ASR.L		#6,d0		...D0 is M
    520 	ADDI.W		#$3FFF,d0	...biased expo. of 2^(M)
    521 	MOVE.W		L2,L_SCR1(a6)	...prefetch L2, no need in CB
    522 
    523 EXPCONT1:
    524 *--Step 3.
    525 *--fp1,fp2 saved on the stack. fp0 is N, fp1 is X,
    526 *--a0 points to 2^(J/64), D0 is biased expo. of 2^(M)
    527 	FMOVE.X		fp0,fp2
    528 	FMUL.S		#:BC317218,fp0	...N * L1, L1 = lead(-log2/64)
    529 	FMUL.X		L2,fp2		...N * L2, L1+L2 = -log2/64
    530 	FADD.X		fp1,fp0	 	...X + N*L1
    531 	FADD.X		fp2,fp0		...fp0 is R, reduced arg.
    532 *	MOVE.W		#$3FA5,EXPA3	...load EXPA3 in cache
    533 
    534 *--Step 4.
    535 *--WE NOW COMPUTE EXP(R)-1 BY A POLYNOMIAL
    536 *-- R + R*R*(A1 + R*(A2 + R*(A3 + R*(A4 + R*A5))))
    537 *--TO FULLY UTILIZE THE PIPELINE, WE COMPUTE S = R*R
    538 *--[R+R*S*(A2+S*A4)] + [S*(A1+S*(A3+S*A5))]
    539 
    540 	FMOVE.X		fp0,fp1
    541 	FMUL.X		fp1,fp1	 	...fp1 IS S = R*R
    542 
    543 	FMOVE.S		#:3AB60B70,fp2	...fp2 IS A5
    544 *	MOVE.W		#0,2(a1)	...load 2^(J/64) in cache
    545 
    546 	FMUL.X		fp1,fp2	 	...fp2 IS S*A5
    547 	FMOVE.X		fp1,fp3
    548 	FMUL.S		#:3C088895,fp3	...fp3 IS S*A4
    549 
    550 	FADD.D		EXPA3,fp2	...fp2 IS A3+S*A5
    551 	FADD.D		EXPA2,fp3	...fp3 IS A2+S*A4
    552 
    553 	FMUL.X		fp1,fp2	 	...fp2 IS S*(A3+S*A5)
    554 	MOVE.W		d0,SCALE(a6)	...SCALE is 2^(M) in extended
    555 	clr.w		SCALE+2(a6)
    556 	move.l		#$80000000,SCALE+4(a6)
    557 	clr.l		SCALE+8(a6)
    558 
    559 	FMUL.X		fp1,fp3	 	...fp3 IS S*(A2+S*A4)
    560 
    561 	FADD.S		#:3F000000,fp2	...fp2 IS A1+S*(A3+S*A5)
    562 	FMUL.X		fp0,fp3	 	...fp3 IS R*S*(A2+S*A4)
    563 
    564 	FMUL.X		fp1,fp2	 	...fp2 IS S*(A1+S*(A3+S*A5))
    565 	FADD.X		fp3,fp0	 	...fp0 IS R+R*S*(A2+S*A4),
    566 *					...fp3 released
    567 
    568 	FMOVE.X		(a1)+,fp1	...fp1 is lead. pt. of 2^(J/64)
    569 	FADD.X		fp2,fp0	 	...fp0 is EXP(R) - 1
    570 *					...fp2 released
    571 
    572 *--Step 5
    573 *--final reconstruction process
    574 *--EXP(X) = 2^M * ( 2^(J/64) + 2^(J/64)*(EXP(R)-1) )
    575 
    576 	FMUL.X		fp1,fp0	 	...2^(J/64)*(Exp(R)-1)
    577 	fmovem.x	(a7)+,fp2/fp3	...fp2 restored
    578 	FADD.S		(a1),fp0	...accurate 2^(J/64)
    579 
    580 	FADD.X		fp1,fp0	 	...2^(J/64) + 2^(J/64)*...
    581 	MOVE.L		ADJFLAG(a6),d0
    582 
    583 *--Step 6
    584 	TST.L		D0
    585 	BEQ.B		NORMAL
    586 ADJUST:
    587 	FMUL.X		ADJSCALE(a6),fp0
    588 NORMAL:
    589 	FMOVE.L		d1,FPCR	 	...restore user FPCR
    590 	FMUL.X		SCALE(a6),fp0	...multiply 2^(M)
    591 	bra		t_frcinx
    592 
    593 EXPSM:
    594 *--Step 7
    595 	FMOVEM.X	(a0),fp0	...in case X is denormalized
    596 	FMOVE.L		d1,FPCR
    597 	FADD.S		#:3F800000,fp0	...1+X in user mode
    598 	bra		t_frcinx
    599 
    600 EXPBIG:
    601 *--Step 8
    602 	CMPI.L		#$400CB27C,d0	...16480 log2
    603 	BGT.B		EXP2BIG
    604 *--Steps 8.2 -- 8.6
    605 	FMOVE.X		(a0),fp0	...load input from (a0)
    606 
    607 	FMOVE.X		fp0,fp1
    608 	FMUL.S		#:42B8AA3B,fp0	...64/log2 * X
    609 	fmovem.x	 fp2/fp3,-(a7)		...save fp2
    610 	MOVE.L		#1,ADJFLAG(a6)
    611 	FMOVE.L		fp0,d0		...N = int( X * 64/log2 )
    612 	LEA		EXPTBL,a1
    613 	FMOVE.L		d0,fp0		...convert to floating-format
    614 	MOVE.L		d0,L_SCR1(a6)			...save N temporarily
    615 	ANDI.L		#$3F,d0		 ...D0 is J = N mod 64
    616 	LSL.L		#4,d0
    617 	ADDA.L		d0,a1			...address of 2^(J/64)
    618 	MOVE.L		L_SCR1(a6),d0
    619 	ASR.L		#6,d0			...D0 is K
    620 	MOVE.L		d0,L_SCR1(a6)			...save K temporarily
    621 	ASR.L		#1,d0			...D0 is M1
    622 	SUB.L		d0,L_SCR1(a6)			...a1 is M
    623 	ADDI.W		#$3FFF,d0		...biased expo. of 2^(M1)
    624 	MOVE.W		d0,ADJSCALE(a6)		...ADJSCALE := 2^(M1)
    625 	clr.w		ADJSCALE+2(a6)
    626 	move.l		#$80000000,ADJSCALE+4(a6)
    627 	clr.l		ADJSCALE+8(a6)
    628 	MOVE.L		L_SCR1(a6),d0			...D0 is M
    629 	ADDI.W		#$3FFF,d0		...biased expo. of 2^(M)
    630 	BRA.W		EXPCONT1		...go back to Step 3
    631 
    632 EXP2BIG:
    633 *--Step 9
    634 	FMOVE.L		d1,FPCR
    635 	MOVE.L		(a0),d0
    636 	bclr.b		#sign_bit,(a0)		...setox always returns positive
    637 	CMPI.L		#0,d0
    638 	BLT		t_unfl
    639 	BRA		t_ovfl
    640 
    641 	xdef	setoxm1d
    642 setoxm1d:
    643 *--entry point for EXPM1(X), here X is denormalized
    644 *--Step 0.
    645 	bra		t_extdnrm
    646 
    647 
    648 	xdef	setoxm1
    649 setoxm1:
    650 *--entry point for EXPM1(X), here X is finite, non-zero, non-NaN
    651 
    652 *--Step 1.
    653 *--Step 1.1
    654 	MOVE.L		(a0),d0	 ...load part of input X
    655 	ANDI.L		#$7FFF0000,d0	...biased expo. of X
    656 	CMPI.L		#$3FFD0000,d0	...1/4
    657 	BGE.B		EM1CON1	 ...|X| >= 1/4
    658 	BRA.W		EM1SM
    659 
    660 EM1CON1:
    661 *--Step 1.3
    662 *--The case |X| >= 1/4
    663 	MOVE.W		4(a0),d0	...expo. and partial sig. of |X|
    664 	CMPI.L		#$4004C215,d0	...70log2 rounded up to 16 bits
    665 	BLE.B		EM1MAIN	 ...1/4 <= |X| <= 70log2
    666 	BRA.W		EM1BIG
    667 
    668 EM1MAIN:
    669 *--Step 2.
    670 *--This is the case:	1/4 <= |X| <= 70 log2.
    671 	FMOVE.X		(a0),fp0	...load input from (a0)
    672 
    673 	FMOVE.X		fp0,fp1
    674 	FMUL.S		#:42B8AA3B,fp0	...64/log2 * X
    675 	fmovem.x	fp2/fp3,-(a7)		...save fp2
    676 *	MOVE.W		#$3F81,EM1A4		...prefetch in CB mode
    677 	FMOVE.L		fp0,d0		...N = int( X * 64/log2 )
    678 	LEA		EXPTBL,a1
    679 	FMOVE.L		d0,fp0		...convert to floating-format
    680 
    681 	MOVE.L		d0,L_SCR1(a6)			...save N temporarily
    682 	ANDI.L		#$3F,d0		 ...D0 is J = N mod 64
    683 	LSL.L		#4,d0
    684 	ADDA.L		d0,a1			...address of 2^(J/64)
    685 	MOVE.L		L_SCR1(a6),d0
    686 	ASR.L		#6,d0			...D0 is M
    687 	MOVE.L		d0,L_SCR1(a6)			...save a copy of M
    688 *	MOVE.W		#$3FDC,L2		...prefetch L2 in CB mode
    689 
    690 *--Step 3.
    691 *--fp1,fp2 saved on the stack. fp0 is N, fp1 is X,
    692 *--a0 points to 2^(J/64), D0 and a1 both contain M
    693 	FMOVE.X		fp0,fp2
    694 	FMUL.S		#:BC317218,fp0	...N * L1, L1 = lead(-log2/64)
    695 	FMUL.X		L2,fp2		...N * L2, L1+L2 = -log2/64
    696 	FADD.X		fp1,fp0	 ...X + N*L1
    697 	FADD.X		fp2,fp0	 ...fp0 is R, reduced arg.
    698 *	MOVE.W		#$3FC5,EM1A2		...load EM1A2 in cache
    699 	ADDI.W		#$3FFF,d0		...D0 is biased expo. of 2^M
    700 
    701 *--Step 4.
    702 *--WE NOW COMPUTE EXP(R)-1 BY A POLYNOMIAL
    703 *-- R + R*R*(A1 + R*(A2 + R*(A3 + R*(A4 + R*(A5 + R*A6)))))
    704 *--TO FULLY UTILIZE THE PIPELINE, WE COMPUTE S = R*R
    705 *--[R*S*(A2+S*(A4+S*A6))] + [R+S*(A1+S*(A3+S*A5))]
    706 
    707 	FMOVE.X		fp0,fp1
    708 	FMUL.X		fp1,fp1		...fp1 IS S = R*R
    709 
    710 	FMOVE.S		#:3950097B,fp2	...fp2 IS a6
    711 *	MOVE.W		#0,2(a1)	...load 2^(J/64) in cache
    712 
    713 	FMUL.X		fp1,fp2		...fp2 IS S*A6
    714 	FMOVE.X		fp1,fp3
    715 	FMUL.S		#:3AB60B6A,fp3	...fp3 IS S*A5
    716 
    717 	FADD.D		EM1A4,fp2	...fp2 IS A4+S*A6
    718 	FADD.D		EM1A3,fp3	...fp3 IS A3+S*A5
    719 	MOVE.W		d0,SC(a6)		...SC is 2^(M) in extended
    720 	clr.w		SC+2(a6)
    721 	move.l		#$80000000,SC+4(a6)
    722 	clr.l		SC+8(a6)
    723 
    724 	FMUL.X		fp1,fp2		...fp2 IS S*(A4+S*A6)
    725 	MOVE.L		L_SCR1(a6),d0		...D0 is	M
    726 	NEG.W		D0		...D0 is -M
    727 	FMUL.X		fp1,fp3		...fp3 IS S*(A3+S*A5)
    728 	ADDI.W		#$3FFF,d0	...biased expo. of 2^(-M)
    729 	FADD.D		EM1A2,fp2	...fp2 IS A2+S*(A4+S*A6)
    730 	FADD.S		#:3F000000,fp3	...fp3 IS A1+S*(A3+S*A5)
    731 
    732 	FMUL.X		fp1,fp2		...fp2 IS S*(A2+S*(A4+S*A6))
    733 	ORI.W		#$8000,d0	...signed/expo. of -2^(-M)
    734 	MOVE.W		d0,ONEBYSC(a6)	...OnebySc is -2^(-M)
    735 	clr.w		ONEBYSC+2(a6)
    736 	move.l		#$80000000,ONEBYSC+4(a6)
    737 	clr.l		ONEBYSC+8(a6)
    738 	FMUL.X		fp3,fp1		...fp1 IS S*(A1+S*(A3+S*A5))
    739 *					...fp3 released
    740 
    741 	FMUL.X		fp0,fp2		...fp2 IS R*S*(A2+S*(A4+S*A6))
    742 	FADD.X		fp1,fp0		...fp0 IS R+S*(A1+S*(A3+S*A5))
    743 *					...fp1 released
    744 
    745 	FADD.X		fp2,fp0		...fp0 IS EXP(R)-1
    746 *					...fp2 released
    747 	fmovem.x	(a7)+,fp2/fp3	...fp2 restored
    748 
    749 *--Step 5
    750 *--Compute 2^(J/64)*p
    751 
    752 	FMUL.X		(a1),fp0	...2^(J/64)*(Exp(R)-1)
    753 
    754 *--Step 6
    755 *--Step 6.1
    756 	MOVE.L		L_SCR1(a6),d0		...retrieve M
    757 	CMPI.L		#63,d0
    758 	BLE.B		MLE63
    759 *--Step 6.2	M >= 64
    760 	FMOVE.S		12(a1),fp1	...fp1 is t
    761 	FADD.X		ONEBYSC(a6),fp1	...fp1 is t+OnebySc
    762 	FADD.X		fp1,fp0		...p+(t+OnebySc), fp1 released
    763 	FADD.X		(a1),fp0	...T+(p+(t+OnebySc))
    764 	BRA.B		EM1SCALE
    765 MLE63:
    766 *--Step 6.3	M <= 63
    767 	CMPI.L		#-3,d0
    768 	BGE.B		MGEN3
    769 MLTN3:
    770 *--Step 6.4	M <= -4
    771 	FADD.S		12(a1),fp0	...p+t
    772 	FADD.X		(a1),fp0	...T+(p+t)
    773 	FADD.X		ONEBYSC(a6),fp0	...OnebySc + (T+(p+t))
    774 	BRA.B		EM1SCALE
    775 MGEN3:
    776 *--Step 6.5	-3 <= M <= 63
    777 	FMOVE.X		(a1)+,fp1	...fp1 is T
    778 	FADD.S		(a1),fp0	...fp0 is p+t
    779 	FADD.X		ONEBYSC(a6),fp1	...fp1 is T+OnebySc
    780 	FADD.X		fp1,fp0		...(T+OnebySc)+(p+t)
    781 
    782 EM1SCALE:
    783 *--Step 6.6
    784 	FMOVE.L		d1,FPCR
    785 	FMUL.X		SC(a6),fp0
    786 
    787 	bra		t_frcinx
    788 
    789 EM1SM:
    790 *--Step 7	|X| < 1/4.
    791 	CMPI.L		#$3FBE0000,d0	...2^(-65)
    792 	BGE.B		EM1POLY
    793 
    794 EM1TINY:
    795 *--Step 8	|X| < 2^(-65)
    796 	CMPI.L		#$00330000,d0	...2^(-16312)
    797 	BLT.B		EM12TINY
    798 *--Step 8.2
    799 	MOVE.L		#$80010000,SC(a6)	...SC is -2^(-16382)
    800 	move.l		#$80000000,SC+4(a6)
    801 	clr.l		SC+8(a6)
    802 	FMOVE.X		(a0),fp0
    803 	FMOVE.L		d1,FPCR
    804 	FADD.X		SC(a6),fp0
    805 
    806 	bra		t_frcinx
    807 
    808 EM12TINY:
    809 *--Step 8.3
    810 	FMOVE.X		(a0),fp0
    811 	FMUL.D		TWO140,fp0
    812 	MOVE.L		#$80010000,SC(a6)
    813 	move.l		#$80000000,SC+4(a6)
    814 	clr.l		SC+8(a6)
    815 	FADD.X		SC(a6),fp0
    816 	FMOVE.L		d1,FPCR
    817 	FMUL.D		TWON140,fp0
    818 
    819 	bra		t_frcinx
    820 
    821 EM1POLY:
    822 *--Step 9	exp(X)-1 by a simple polynomial
    823 	FMOVE.X		(a0),fp0	...fp0 is X
    824 	FMUL.X		fp0,fp0		...fp0 is S := X*X
    825 	fmovem.x	fp2/fp3,-(a7)	...save fp2
    826 	FMOVE.S		#:2F30CAA8,fp1	...fp1 is B12
    827 	FMUL.X		fp0,fp1		...fp1 is S*B12
    828 	FMOVE.S		#:310F8290,fp2	...fp2 is B11
    829 	FADD.S		#:32D73220,fp1	...fp1 is B10+S*B12
    830 
    831 	FMUL.X		fp0,fp2		...fp2 is S*B11
    832 	FMUL.X		fp0,fp1		...fp1 is S*(B10 + ...
    833 
    834 	FADD.S		#:3493F281,fp2	...fp2 is B9+S*...
    835 	FADD.D		EM1B8,fp1	...fp1 is B8+S*...
    836 
    837 	FMUL.X		fp0,fp2		...fp2 is S*(B9+...
    838 	FMUL.X		fp0,fp1		...fp1 is S*(B8+...
    839 
    840 	FADD.D		EM1B7,fp2	...fp2 is B7+S*...
    841 	FADD.D		EM1B6,fp1	...fp1 is B6+S*...
    842 
    843 	FMUL.X		fp0,fp2		...fp2 is S*(B7+...
    844 	FMUL.X		fp0,fp1		...fp1 is S*(B6+...
    845 
    846 	FADD.D		EM1B5,fp2	...fp2 is B5+S*...
    847 	FADD.D		EM1B4,fp1	...fp1 is B4+S*...
    848 
    849 	FMUL.X		fp0,fp2		...fp2 is S*(B5+...
    850 	FMUL.X		fp0,fp1		...fp1 is S*(B4+...
    851 
    852 	FADD.D		EM1B3,fp2	...fp2 is B3+S*...
    853 	FADD.X		EM1B2,fp1	...fp1 is B2+S*...
    854 
    855 	FMUL.X		fp0,fp2		...fp2 is S*(B3+...
    856 	FMUL.X		fp0,fp1		...fp1 is S*(B2+...
    857 
    858 	FMUL.X		fp0,fp2		...fp2 is S*S*(B3+...)
    859 	FMUL.X		(a0),fp1	...fp1 is X*S*(B2...
    860 
    861 	FMUL.S		#:3F000000,fp0	...fp0 is S*B1
    862 	FADD.X		fp2,fp1		...fp1 is Q
    863 *					...fp2 released
    864 
    865 	fmovem.x	(a7)+,fp2/fp3	...fp2 restored
    866 
    867 	FADD.X		fp1,fp0		...fp0 is S*B1+Q
    868 *					...fp1 released
    869 
    870 	FMOVE.L		d1,FPCR
    871 	FADD.X		(a0),fp0
    872 
    873 	bra		t_frcinx
    874 
    875 EM1BIG:
    876 *--Step 10	|X| > 70 log2
    877 	MOVE.L		(a0),d0
    878 	CMPI.L		#0,d0
    879 	BGT.W		EXPC1
    880 *--Step 10.2
    881 	FMOVE.S		#:BF800000,fp0	...fp0 is -1
    882 	FMOVE.L		d1,FPCR
    883 	FADD.S		#:00800000,fp0	...-1 + 2^(-126)
    884 
    885 	bra		t_frcinx
    886 
    887 	end
    888