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

ssin.S (19058B)


      1|
      2|	ssin.sa 3.3 7/29/91
      3|
      4|	The entry point sSIN computes the sine of an input argument
      5|	sCOS computes the cosine, and sSINCOS computes both. The
      6|	corresponding entry points with a "d" computes the same
      7|	corresponding function values for denormalized inputs.
      8|
      9|	Input: Double-extended number X in location pointed to
     10|		by address register a0.
     11|
     12|	Output: The function value sin(X) or cos(X) returned in Fp0 if SIN or
     13|		COS is requested. Otherwise, for SINCOS, sin(X) is returned
     14|		in Fp0, and cos(X) is returned in Fp1.
     15|
     16|	Modifies: Fp0 for SIN or COS; both Fp0 and Fp1 for SINCOS.
     17|
     18|	Accuracy and Monotonicity: The returned result is within 1 ulp in
     19|		64 significant bit, i.e. within 0.5001 ulp to 53 bits if the
     20|		result is subsequently rounded to double precision. The
     21|		result is provably monotonic in double precision.
     22|
     23|	Speed: The programs sSIN and sCOS take approximately 150 cycles for
     24|		input argument X such that |X| < 15Pi, which is the usual
     25|		situation. The speed for sSINCOS is approximately 190 cycles.
     26|
     27|	Algorithm:
     28|
     29|	SIN and COS:
     30|	1. If SIN is invoked, set AdjN := 0; otherwise, set AdjN := 1.
     31|
     32|	2. If |X| >= 15Pi or |X| < 2**(-40), go to 7.
     33|
     34|	3. Decompose X as X = N(Pi/2) + r where |r| <= Pi/4. Let
     35|		k = N mod 4, so in particular, k = 0,1,2,or 3. Overwrite
     36|		k by k := k + AdjN.
     37|
     38|	4. If k is even, go to 6.
     39|
     40|	5. (k is odd) Set j := (k-1)/2, sgn := (-1)**j. Return sgn*cos(r)
     41|		where cos(r) is approximated by an even polynomial in r,
     42|		1 + r*r*(B1+s*(B2+ ... + s*B8)),	s = r*r.
     43|		Exit.
     44|
     45|	6. (k is even) Set j := k/2, sgn := (-1)**j. Return sgn*sin(r)
     46|		where sin(r) is approximated by an odd polynomial in r
     47|		r + r*s*(A1+s*(A2+ ... + s*A7)),	s = r*r.
     48|		Exit.
     49|
     50|	7. If |X| > 1, go to 9.
     51|
     52|	8. (|X|<2**(-40)) If SIN is invoked, return X; otherwise return 1.
     53|
     54|	9. Overwrite X by X := X rem 2Pi. Now that |X| <= Pi, go back to 3.
     55|
     56|	SINCOS:
     57|	1. If |X| >= 15Pi or |X| < 2**(-40), go to 6.
     58|
     59|	2. Decompose X as X = N(Pi/2) + r where |r| <= Pi/4. Let
     60|		k = N mod 4, so in particular, k = 0,1,2,or 3.
     61|
     62|	3. If k is even, go to 5.
     63|
     64|	4. (k is odd) Set j1 := (k-1)/2, j2 := j1 (EOR) (k mod 2), i.e.
     65|		j1 exclusive or with the l.s.b. of k.
     66|		sgn1 := (-1)**j1, sgn2 := (-1)**j2.
     67|		SIN(X) = sgn1 * cos(r) and COS(X) = sgn2*sin(r) where
     68|		sin(r) and cos(r) are computed as odd and even polynomials
     69|		in r, respectively. Exit
     70|
     71|	5. (k is even) Set j1 := k/2, sgn1 := (-1)**j1.
     72|		SIN(X) = sgn1 * sin(r) and COS(X) = sgn1*cos(r) where
     73|		sin(r) and cos(r) are computed as odd and even polynomials
     74|		in r, respectively. Exit
     75|
     76|	6. If |X| > 1, go to 8.
     77|
     78|	7. (|X|<2**(-40)) SIN(X) = X and COS(X) = 1. Exit.
     79|
     80|	8. Overwrite X by X := X rem 2Pi. Now that |X| <= Pi, go back to 2.
     81|
     82
     83|		Copyright (C) Motorola, Inc. 1990
     84|			All Rights Reserved
     85|
     86|       For details on the license for this file, please see the
     87|       file, README, in this same directory.
     88
     89|SSIN	idnt	2,1 | Motorola 040 Floating Point Software Package
     90
     91	|section	8
     92
     93#include "fpsp.h"
     94
     95BOUNDS1:	.long 0x3FD78000,0x4004BC7E
     96TWOBYPI:	.long 0x3FE45F30,0x6DC9C883
     97
     98SINA7:	.long 0xBD6AAA77,0xCCC994F5
     99SINA6:	.long 0x3DE61209,0x7AAE8DA1
    100
    101SINA5:	.long 0xBE5AE645,0x2A118AE4
    102SINA4:	.long 0x3EC71DE3,0xA5341531
    103
    104SINA3:	.long 0xBF2A01A0,0x1A018B59,0x00000000,0x00000000
    105
    106SINA2:	.long 0x3FF80000,0x88888888,0x888859AF,0x00000000
    107
    108SINA1:	.long 0xBFFC0000,0xAAAAAAAA,0xAAAAAA99,0x00000000
    109
    110COSB8:	.long 0x3D2AC4D0,0xD6011EE3
    111COSB7:	.long 0xBDA9396F,0x9F45AC19
    112
    113COSB6:	.long 0x3E21EED9,0x0612C972
    114COSB5:	.long 0xBE927E4F,0xB79D9FCF
    115
    116COSB4:	.long 0x3EFA01A0,0x1A01D423,0x00000000,0x00000000
    117
    118COSB3:	.long 0xBFF50000,0xB60B60B6,0x0B61D438,0x00000000
    119
    120COSB2:	.long 0x3FFA0000,0xAAAAAAAA,0xAAAAAB5E
    121COSB1:	.long 0xBF000000
    122
    123INVTWOPI: .long 0x3FFC0000,0xA2F9836E,0x4E44152A
    124
    125TWOPI1:	.long 0x40010000,0xC90FDAA2,0x00000000,0x00000000
    126TWOPI2:	.long 0x3FDF0000,0x85A308D4,0x00000000,0x00000000
    127
    128	|xref	PITBL
    129
    130	.set	INARG,FP_SCR4
    131
    132	.set	X,FP_SCR5
    133	.set	XDCARE,X+2
    134	.set	XFRAC,X+4
    135
    136	.set	RPRIME,FP_SCR1
    137	.set	SPRIME,FP_SCR2
    138
    139	.set	POSNEG1,L_SCR1
    140	.set	TWOTO63,L_SCR1
    141
    142	.set	ENDFLAG,L_SCR2
    143	.set	N,L_SCR2
    144
    145	.set	ADJN,L_SCR3
    146
    147	| xref	t_frcinx
    148	|xref	t_extdnrm
    149	|xref	sto_cos
    150
    151	.global	ssind
    152ssind:
    153|--SIN(X) = X FOR DENORMALIZED X
    154	bra		t_extdnrm
    155
    156	.global	scosd
    157scosd:
    158|--COS(X) = 1 FOR DENORMALIZED X
    159
    160	fmoves		#0x3F800000,%fp0
    161|
    162|	9D25B Fix: Sometimes the previous fmove.s sets fpsr bits
    163|
    164	fmovel		#0,%fpsr
    165|
    166	bra		t_frcinx
    167
    168	.global	ssin
    169ssin:
    170|--SET ADJN TO 0
    171	movel		#0,ADJN(%a6)
    172	bras		SINBGN
    173
    174	.global	scos
    175scos:
    176|--SET ADJN TO 1
    177	movel		#1,ADJN(%a6)
    178
    179SINBGN:
    180|--SAVE FPCR, FP1. CHECK IF |X| IS TOO SMALL OR LARGE
    181
    182	fmovex		(%a0),%fp0	| ...LOAD INPUT
    183
    184	movel		(%a0),%d0
    185	movew		4(%a0),%d0
    186	fmovex		%fp0,X(%a6)
    187	andil		#0x7FFFFFFF,%d0		| ...COMPACTIFY X
    188
    189	cmpil		#0x3FD78000,%d0		| ...|X| >= 2**(-40)?
    190	bges		SOK1
    191	bra		SINSM
    192
    193SOK1:
    194	cmpil		#0x4004BC7E,%d0		| ...|X| < 15 PI?
    195	blts		SINMAIN
    196	bra		REDUCEX
    197
    198SINMAIN:
    199|--THIS IS THE USUAL CASE, |X| <= 15 PI.
    200|--THE ARGUMENT REDUCTION IS DONE BY TABLE LOOK UP.
    201	fmovex		%fp0,%fp1
    202	fmuld		TWOBYPI,%fp1	| ...X*2/PI
    203
    204|--HIDE THE NEXT THREE INSTRUCTIONS
    205	lea		PITBL+0x200,%a1 | ...TABLE OF N*PI/2, N = -32,...,32
    206
    207
    208|--FP1 IS NOW READY
    209	fmovel		%fp1,N(%a6)		| ...CONVERT TO INTEGER
    210
    211	movel		N(%a6),%d0
    212	asll		#4,%d0
    213	addal		%d0,%a1	| ...A1 IS THE ADDRESS OF N*PIBY2
    214|				...WHICH IS IN TWO PIECES Y1 & Y2
    215
    216	fsubx		(%a1)+,%fp0	| ...X-Y1
    217|--HIDE THE NEXT ONE
    218	fsubs		(%a1),%fp0	| ...FP0 IS R = (X-Y1)-Y2
    219
    220SINCONT:
    221|--continuation from REDUCEX
    222
    223|--GET N+ADJN AND SEE IF SIN(R) OR COS(R) IS NEEDED
    224	movel		N(%a6),%d0
    225	addl		ADJN(%a6),%d0	| ...SEE IF D0 IS ODD OR EVEN
    226	rorl		#1,%d0	| ...D0 WAS ODD IFF D0 IS NEGATIVE
    227	cmpil		#0,%d0
    228	blt		COSPOLY
    229
    230SINPOLY:
    231|--LET J BE THE LEAST SIG. BIT OF D0, LET SGN := (-1)**J.
    232|--THEN WE RETURN	SGN*SIN(R). SGN*SIN(R) IS COMPUTED BY
    233|--R' + R'*S*(A1 + S(A2 + S(A3 + S(A4 + ... + SA7)))), WHERE
    234|--R' = SGN*R, S=R*R. THIS CAN BE REWRITTEN AS
    235|--R' + R'*S*( [A1+T(A3+T(A5+TA7))] + [S(A2+T(A4+TA6))])
    236|--WHERE T=S*S.
    237|--NOTE THAT A3 THROUGH A7 ARE STORED IN DOUBLE PRECISION
    238|--WHILE A1 AND A2 ARE IN DOUBLE-EXTENDED FORMAT.
    239	fmovex		%fp0,X(%a6)	| ...X IS R
    240	fmulx		%fp0,%fp0	| ...FP0 IS S
    241|---HIDE THE NEXT TWO WHILE WAITING FOR FP0
    242	fmoved		SINA7,%fp3
    243	fmoved		SINA6,%fp2
    244|--FP0 IS NOW READY
    245	fmovex		%fp0,%fp1
    246	fmulx		%fp1,%fp1	| ...FP1 IS T
    247|--HIDE THE NEXT TWO WHILE WAITING FOR FP1
    248
    249	rorl		#1,%d0
    250	andil		#0x80000000,%d0
    251|				...LEAST SIG. BIT OF D0 IN SIGN POSITION
    252	eorl		%d0,X(%a6)	| ...X IS NOW R'= SGN*R
    253
    254	fmulx		%fp1,%fp3	| ...TA7
    255	fmulx		%fp1,%fp2	| ...TA6
    256
    257	faddd		SINA5,%fp3 | ...A5+TA7
    258	faddd		SINA4,%fp2 | ...A4+TA6
    259
    260	fmulx		%fp1,%fp3	| ...T(A5+TA7)
    261	fmulx		%fp1,%fp2	| ...T(A4+TA6)
    262
    263	faddd		SINA3,%fp3 | ...A3+T(A5+TA7)
    264	faddx		SINA2,%fp2 | ...A2+T(A4+TA6)
    265
    266	fmulx		%fp3,%fp1	| ...T(A3+T(A5+TA7))
    267
    268	fmulx		%fp0,%fp2	| ...S(A2+T(A4+TA6))
    269	faddx		SINA1,%fp1 | ...A1+T(A3+T(A5+TA7))
    270	fmulx		X(%a6),%fp0	| ...R'*S
    271
    272	faddx		%fp2,%fp1	| ...[A1+T(A3+T(A5+TA7))]+[S(A2+T(A4+TA6))]
    273|--FP3 RELEASED, RESTORE NOW AND TAKE SOME ADVANTAGE OF HIDING
    274|--FP2 RELEASED, RESTORE NOW AND TAKE FULL ADVANTAGE OF HIDING
    275
    276
    277	fmulx		%fp1,%fp0		| ...SIN(R')-R'
    278|--FP1 RELEASED.
    279
    280	fmovel		%d1,%FPCR		|restore users exceptions
    281	faddx		X(%a6),%fp0		|last inst - possible exception set
    282	bra		t_frcinx
    283
    284
    285COSPOLY:
    286|--LET J BE THE LEAST SIG. BIT OF D0, LET SGN := (-1)**J.
    287|--THEN WE RETURN	SGN*COS(R). SGN*COS(R) IS COMPUTED BY
    288|--SGN + S'*(B1 + S(B2 + S(B3 + S(B4 + ... + SB8)))), WHERE
    289|--S=R*R AND S'=SGN*S. THIS CAN BE REWRITTEN AS
    290|--SGN + S'*([B1+T(B3+T(B5+TB7))] + [S(B2+T(B4+T(B6+TB8)))])
    291|--WHERE T=S*S.
    292|--NOTE THAT B4 THROUGH B8 ARE STORED IN DOUBLE PRECISION
    293|--WHILE B2 AND B3 ARE IN DOUBLE-EXTENDED FORMAT, B1 IS -1/2
    294|--AND IS THEREFORE STORED AS SINGLE PRECISION.
    295
    296	fmulx		%fp0,%fp0	| ...FP0 IS S
    297|---HIDE THE NEXT TWO WHILE WAITING FOR FP0
    298	fmoved		COSB8,%fp2
    299	fmoved		COSB7,%fp3
    300|--FP0 IS NOW READY
    301	fmovex		%fp0,%fp1
    302	fmulx		%fp1,%fp1	| ...FP1 IS T
    303|--HIDE THE NEXT TWO WHILE WAITING FOR FP1
    304	fmovex		%fp0,X(%a6)	| ...X IS S
    305	rorl		#1,%d0
    306	andil		#0x80000000,%d0
    307|			...LEAST SIG. BIT OF D0 IN SIGN POSITION
    308
    309	fmulx		%fp1,%fp2	| ...TB8
    310|--HIDE THE NEXT TWO WHILE WAITING FOR THE XU
    311	eorl		%d0,X(%a6)	| ...X IS NOW S'= SGN*S
    312	andil		#0x80000000,%d0
    313
    314	fmulx		%fp1,%fp3	| ...TB7
    315|--HIDE THE NEXT TWO WHILE WAITING FOR THE XU
    316	oril		#0x3F800000,%d0	| ...D0 IS SGN IN SINGLE
    317	movel		%d0,POSNEG1(%a6)
    318
    319	faddd		COSB6,%fp2 | ...B6+TB8
    320	faddd		COSB5,%fp3 | ...B5+TB7
    321
    322	fmulx		%fp1,%fp2	| ...T(B6+TB8)
    323	fmulx		%fp1,%fp3	| ...T(B5+TB7)
    324
    325	faddd		COSB4,%fp2 | ...B4+T(B6+TB8)
    326	faddx		COSB3,%fp3 | ...B3+T(B5+TB7)
    327
    328	fmulx		%fp1,%fp2	| ...T(B4+T(B6+TB8))
    329	fmulx		%fp3,%fp1	| ...T(B3+T(B5+TB7))
    330
    331	faddx		COSB2,%fp2 | ...B2+T(B4+T(B6+TB8))
    332	fadds		COSB1,%fp1 | ...B1+T(B3+T(B5+TB7))
    333
    334	fmulx		%fp2,%fp0	| ...S(B2+T(B4+T(B6+TB8)))
    335|--FP3 RELEASED, RESTORE NOW AND TAKE SOME ADVANTAGE OF HIDING
    336|--FP2 RELEASED.
    337
    338
    339	faddx		%fp1,%fp0
    340|--FP1 RELEASED
    341
    342	fmulx		X(%a6),%fp0
    343
    344	fmovel		%d1,%FPCR		|restore users exceptions
    345	fadds		POSNEG1(%a6),%fp0	|last inst - possible exception set
    346	bra		t_frcinx
    347
    348
    349SINBORS:
    350|--IF |X| > 15PI, WE USE THE GENERAL ARGUMENT REDUCTION.
    351|--IF |X| < 2**(-40), RETURN X OR 1.
    352	cmpil		#0x3FFF8000,%d0
    353	bgts		REDUCEX
    354
    355
    356SINSM:
    357	movel		ADJN(%a6),%d0
    358	cmpil		#0,%d0
    359	bgts		COSTINY
    360
    361SINTINY:
    362	movew		#0x0000,XDCARE(%a6)	| ...JUST IN CASE
    363	fmovel		%d1,%FPCR		|restore users exceptions
    364	fmovex		X(%a6),%fp0		|last inst - possible exception set
    365	bra		t_frcinx
    366
    367
    368COSTINY:
    369	fmoves		#0x3F800000,%fp0
    370
    371	fmovel		%d1,%FPCR		|restore users exceptions
    372	fsubs		#0x00800000,%fp0	|last inst - possible exception set
    373	bra		t_frcinx
    374
    375
    376REDUCEX:
    377|--WHEN REDUCEX IS USED, THE CODE WILL INEVITABLY BE SLOW.
    378|--THIS REDUCTION METHOD, HOWEVER, IS MUCH FASTER THAN USING
    379|--THE REMAINDER INSTRUCTION WHICH IS NOW IN SOFTWARE.
    380
    381	fmovemx	%fp2-%fp5,-(%a7)	| ...save FP2 through FP5
    382	movel		%d2,-(%a7)
    383        fmoves         #0x00000000,%fp1
    384|--If compact form of abs(arg) in d0=$7ffeffff, argument is so large that
    385|--there is a danger of unwanted overflow in first LOOP iteration.  In this
    386|--case, reduce argument by one remainder step to make subsequent reduction
    387|--safe.
    388	cmpil	#0x7ffeffff,%d0		|is argument dangerously large?
    389	bnes	LOOP
    390	movel	#0x7ffe0000,FP_SCR2(%a6)	|yes
    391|					;create 2**16383*PI/2
    392	movel	#0xc90fdaa2,FP_SCR2+4(%a6)
    393	clrl	FP_SCR2+8(%a6)
    394	ftstx	%fp0			|test sign of argument
    395	movel	#0x7fdc0000,FP_SCR3(%a6)	|create low half of 2**16383*
    396|					;PI/2 at FP_SCR3
    397	movel	#0x85a308d3,FP_SCR3+4(%a6)
    398	clrl   FP_SCR3+8(%a6)
    399	fblt	red_neg
    400	orw	#0x8000,FP_SCR2(%a6)	|positive arg
    401	orw	#0x8000,FP_SCR3(%a6)
    402red_neg:
    403	faddx  FP_SCR2(%a6),%fp0		|high part of reduction is exact
    404	fmovex  %fp0,%fp1		|save high result in fp1
    405	faddx  FP_SCR3(%a6),%fp0		|low part of reduction
    406	fsubx  %fp0,%fp1			|determine low component of result
    407	faddx  FP_SCR3(%a6),%fp1		|fp0/fp1 are reduced argument.
    408
    409|--ON ENTRY, FP0 IS X, ON RETURN, FP0 IS X REM PI/2, |X| <= PI/4.
    410|--integer quotient will be stored in N
    411|--Intermediate remainder is 66-bit long; (R,r) in (FP0,FP1)
    412
    413LOOP:
    414	fmovex		%fp0,INARG(%a6)	| ...+-2**K * F, 1 <= F < 2
    415	movew		INARG(%a6),%d0
    416        movel          %d0,%a1		| ...save a copy of D0
    417	andil		#0x00007FFF,%d0
    418	subil		#0x00003FFF,%d0	| ...D0 IS K
    419	cmpil		#28,%d0
    420	bles		LASTLOOP
    421CONTLOOP:
    422	subil		#27,%d0	 | ...D0 IS L := K-27
    423	movel		#0,ENDFLAG(%a6)
    424	bras		WORK
    425LASTLOOP:
    426	clrl		%d0		| ...D0 IS L := 0
    427	movel		#1,ENDFLAG(%a6)
    428
    429WORK:
    430|--FIND THE REMAINDER OF (R,r) W.R.T.	2**L * (PI/2). L IS SO CHOSEN
    431|--THAT	INT( X * (2/PI) / 2**(L) ) < 2**29.
    432
    433|--CREATE 2**(-L) * (2/PI), SIGN(INARG)*2**(63),
    434|--2**L * (PIby2_1), 2**L * (PIby2_2)
    435
    436	movel		#0x00003FFE,%d2	| ...BIASED EXPO OF 2/PI
    437	subl		%d0,%d2		| ...BIASED EXPO OF 2**(-L)*(2/PI)
    438
    439	movel		#0xA2F9836E,FP_SCR1+4(%a6)
    440	movel		#0x4E44152A,FP_SCR1+8(%a6)
    441	movew		%d2,FP_SCR1(%a6)	| ...FP_SCR1 is 2**(-L)*(2/PI)
    442
    443	fmovex		%fp0,%fp2
    444	fmulx		FP_SCR1(%a6),%fp2
    445|--WE MUST NOW FIND INT(FP2). SINCE WE NEED THIS VALUE IN
    446|--FLOATING POINT FORMAT, THE TWO FMOVE'S	FMOVE.L FP <--> N
    447|--WILL BE TOO INEFFICIENT. THE WAY AROUND IT IS THAT
    448|--(SIGN(INARG)*2**63	+	FP2) - SIGN(INARG)*2**63 WILL GIVE
    449|--US THE DESIRED VALUE IN FLOATING POINT.
    450
    451|--HIDE SIX CYCLES OF INSTRUCTION
    452        movel		%a1,%d2
    453        swap		%d2
    454	andil		#0x80000000,%d2
    455	oril		#0x5F000000,%d2	| ...D2 IS SIGN(INARG)*2**63 IN SGL
    456	movel		%d2,TWOTO63(%a6)
    457
    458	movel		%d0,%d2
    459	addil		#0x00003FFF,%d2	| ...BIASED EXPO OF 2**L * (PI/2)
    460
    461|--FP2 IS READY
    462	fadds		TWOTO63(%a6),%fp2	| ...THE FRACTIONAL PART OF FP1 IS ROUNDED
    463
    464|--HIDE 4 CYCLES OF INSTRUCTION; creating 2**(L)*Piby2_1  and  2**(L)*Piby2_2
    465        movew		%d2,FP_SCR2(%a6)
    466	clrw           FP_SCR2+2(%a6)
    467	movel		#0xC90FDAA2,FP_SCR2+4(%a6)
    468	clrl		FP_SCR2+8(%a6)		| ...FP_SCR2 is  2**(L) * Piby2_1
    469
    470|--FP2 IS READY
    471	fsubs		TWOTO63(%a6),%fp2		| ...FP2 is N
    472
    473	addil		#0x00003FDD,%d0
    474        movew		%d0,FP_SCR3(%a6)
    475	clrw           FP_SCR3+2(%a6)
    476	movel		#0x85A308D3,FP_SCR3+4(%a6)
    477	clrl		FP_SCR3+8(%a6)		| ...FP_SCR3 is 2**(L) * Piby2_2
    478
    479	movel		ENDFLAG(%a6),%d0
    480
    481|--We are now ready to perform (R+r) - N*P1 - N*P2, P1 = 2**(L) * Piby2_1 and
    482|--P2 = 2**(L) * Piby2_2
    483	fmovex		%fp2,%fp4
    484	fmulx		FP_SCR2(%a6),%fp4		| ...W = N*P1
    485	fmovex		%fp2,%fp5
    486	fmulx		FP_SCR3(%a6),%fp5		| ...w = N*P2
    487	fmovex		%fp4,%fp3
    488|--we want P+p = W+w  but  |p| <= half ulp of P
    489|--Then, we need to compute  A := R-P   and  a := r-p
    490	faddx		%fp5,%fp3			| ...FP3 is P
    491	fsubx		%fp3,%fp4			| ...W-P
    492
    493	fsubx		%fp3,%fp0			| ...FP0 is A := R - P
    494        faddx		%fp5,%fp4			| ...FP4 is p = (W-P)+w
    495
    496	fmovex		%fp0,%fp3			| ...FP3 A
    497	fsubx		%fp4,%fp1			| ...FP1 is a := r - p
    498
    499|--Now we need to normalize (A,a) to  "new (R,r)" where R+r = A+a but
    500|--|r| <= half ulp of R.
    501	faddx		%fp1,%fp0			| ...FP0 is R := A+a
    502|--No need to calculate r if this is the last loop
    503	cmpil		#0,%d0
    504	bgt		RESTORE
    505
    506|--Need to calculate r
    507	fsubx		%fp0,%fp3			| ...A-R
    508	faddx		%fp3,%fp1			| ...FP1 is r := (A-R)+a
    509	bra		LOOP
    510
    511RESTORE:
    512        fmovel		%fp2,N(%a6)
    513	movel		(%a7)+,%d2
    514	fmovemx	(%a7)+,%fp2-%fp5
    515
    516
    517	movel		ADJN(%a6),%d0
    518	cmpil		#4,%d0
    519
    520	blt		SINCONT
    521	bras		SCCONT
    522
    523	.global	ssincosd
    524ssincosd:
    525|--SIN AND COS OF X FOR DENORMALIZED X
    526
    527	fmoves		#0x3F800000,%fp1
    528	bsr		sto_cos		|store cosine result
    529	bra		t_extdnrm
    530
    531	.global	ssincos
    532ssincos:
    533|--SET ADJN TO 4
    534	movel		#4,ADJN(%a6)
    535
    536	fmovex		(%a0),%fp0	| ...LOAD INPUT
    537
    538	movel		(%a0),%d0
    539	movew		4(%a0),%d0
    540	fmovex		%fp0,X(%a6)
    541	andil		#0x7FFFFFFF,%d0		| ...COMPACTIFY X
    542
    543	cmpil		#0x3FD78000,%d0		| ...|X| >= 2**(-40)?
    544	bges		SCOK1
    545	bra		SCSM
    546
    547SCOK1:
    548	cmpil		#0x4004BC7E,%d0		| ...|X| < 15 PI?
    549	blts		SCMAIN
    550	bra		REDUCEX
    551
    552
    553SCMAIN:
    554|--THIS IS THE USUAL CASE, |X| <= 15 PI.
    555|--THE ARGUMENT REDUCTION IS DONE BY TABLE LOOK UP.
    556	fmovex		%fp0,%fp1
    557	fmuld		TWOBYPI,%fp1	| ...X*2/PI
    558
    559|--HIDE THE NEXT THREE INSTRUCTIONS
    560	lea		PITBL+0x200,%a1 | ...TABLE OF N*PI/2, N = -32,...,32
    561
    562
    563|--FP1 IS NOW READY
    564	fmovel		%fp1,N(%a6)		| ...CONVERT TO INTEGER
    565
    566	movel		N(%a6),%d0
    567	asll		#4,%d0
    568	addal		%d0,%a1		| ...ADDRESS OF N*PIBY2, IN Y1, Y2
    569
    570	fsubx		(%a1)+,%fp0	| ...X-Y1
    571        fsubs		(%a1),%fp0	| ...FP0 IS R = (X-Y1)-Y2
    572
    573SCCONT:
    574|--continuation point from REDUCEX
    575
    576|--HIDE THE NEXT TWO
    577	movel		N(%a6),%d0
    578	rorl		#1,%d0
    579
    580	cmpil		#0,%d0		| ...D0 < 0 IFF N IS ODD
    581	bge		NEVEN
    582
    583NODD:
    584|--REGISTERS SAVED SO FAR: D0, A0, FP2.
    585
    586	fmovex		%fp0,RPRIME(%a6)
    587	fmulx		%fp0,%fp0	 | ...FP0 IS S = R*R
    588	fmoved		SINA7,%fp1	| ...A7
    589	fmoved		COSB8,%fp2	| ...B8
    590	fmulx		%fp0,%fp1	 | ...SA7
    591	movel		%d2,-(%a7)
    592	movel		%d0,%d2
    593	fmulx		%fp0,%fp2	 | ...SB8
    594	rorl		#1,%d2
    595	andil		#0x80000000,%d2
    596
    597	faddd		SINA6,%fp1	| ...A6+SA7
    598	eorl		%d0,%d2
    599	andil		#0x80000000,%d2
    600	faddd		COSB7,%fp2	| ...B7+SB8
    601
    602	fmulx		%fp0,%fp1	 | ...S(A6+SA7)
    603	eorl		%d2,RPRIME(%a6)
    604	movel		(%a7)+,%d2
    605	fmulx		%fp0,%fp2	 | ...S(B7+SB8)
    606	rorl		#1,%d0
    607	andil		#0x80000000,%d0
    608
    609	faddd		SINA5,%fp1	| ...A5+S(A6+SA7)
    610	movel		#0x3F800000,POSNEG1(%a6)
    611	eorl		%d0,POSNEG1(%a6)
    612	faddd		COSB6,%fp2	| ...B6+S(B7+SB8)
    613
    614	fmulx		%fp0,%fp1	 | ...S(A5+S(A6+SA7))
    615	fmulx		%fp0,%fp2	 | ...S(B6+S(B7+SB8))
    616	fmovex		%fp0,SPRIME(%a6)
    617
    618	faddd		SINA4,%fp1	| ...A4+S(A5+S(A6+SA7))
    619	eorl		%d0,SPRIME(%a6)
    620	faddd		COSB5,%fp2	| ...B5+S(B6+S(B7+SB8))
    621
    622	fmulx		%fp0,%fp1	 | ...S(A4+...)
    623	fmulx		%fp0,%fp2	 | ...S(B5+...)
    624
    625	faddd		SINA3,%fp1	| ...A3+S(A4+...)
    626	faddd		COSB4,%fp2	| ...B4+S(B5+...)
    627
    628	fmulx		%fp0,%fp1	 | ...S(A3+...)
    629	fmulx		%fp0,%fp2	 | ...S(B4+...)
    630
    631	faddx		SINA2,%fp1	| ...A2+S(A3+...)
    632	faddx		COSB3,%fp2	| ...B3+S(B4+...)
    633
    634	fmulx		%fp0,%fp1	 | ...S(A2+...)
    635	fmulx		%fp0,%fp2	 | ...S(B3+...)
    636
    637	faddx		SINA1,%fp1	| ...A1+S(A2+...)
    638	faddx		COSB2,%fp2	| ...B2+S(B3+...)
    639
    640	fmulx		%fp0,%fp1	 | ...S(A1+...)
    641	fmulx		%fp2,%fp0	 | ...S(B2+...)
    642
    643
    644
    645	fmulx		RPRIME(%a6),%fp1	| ...R'S(A1+...)
    646	fadds		COSB1,%fp0	| ...B1+S(B2...)
    647	fmulx		SPRIME(%a6),%fp0	| ...S'(B1+S(B2+...))
    648
    649	movel		%d1,-(%sp)	|restore users mode & precision
    650	andil		#0xff,%d1		|mask off all exceptions
    651	fmovel		%d1,%FPCR
    652	faddx		RPRIME(%a6),%fp1	| ...COS(X)
    653	bsr		sto_cos		|store cosine result
    654	fmovel		(%sp)+,%FPCR	|restore users exceptions
    655	fadds		POSNEG1(%a6),%fp0	| ...SIN(X)
    656
    657	bra		t_frcinx
    658
    659
    660NEVEN:
    661|--REGISTERS SAVED SO FAR: FP2.
    662
    663	fmovex		%fp0,RPRIME(%a6)
    664	fmulx		%fp0,%fp0	 | ...FP0 IS S = R*R
    665	fmoved		COSB8,%fp1			| ...B8
    666	fmoved		SINA7,%fp2			| ...A7
    667	fmulx		%fp0,%fp1	 | ...SB8
    668	fmovex		%fp0,SPRIME(%a6)
    669	fmulx		%fp0,%fp2	 | ...SA7
    670	rorl		#1,%d0
    671	andil		#0x80000000,%d0
    672	faddd		COSB7,%fp1	| ...B7+SB8
    673	faddd		SINA6,%fp2	| ...A6+SA7
    674	eorl		%d0,RPRIME(%a6)
    675	eorl		%d0,SPRIME(%a6)
    676	fmulx		%fp0,%fp1	 | ...S(B7+SB8)
    677	oril		#0x3F800000,%d0
    678	movel		%d0,POSNEG1(%a6)
    679	fmulx		%fp0,%fp2	 | ...S(A6+SA7)
    680
    681	faddd		COSB6,%fp1	| ...B6+S(B7+SB8)
    682	faddd		SINA5,%fp2	| ...A5+S(A6+SA7)
    683
    684	fmulx		%fp0,%fp1	 | ...S(B6+S(B7+SB8))
    685	fmulx		%fp0,%fp2	 | ...S(A5+S(A6+SA7))
    686
    687	faddd		COSB5,%fp1	| ...B5+S(B6+S(B7+SB8))
    688	faddd		SINA4,%fp2	| ...A4+S(A5+S(A6+SA7))
    689
    690	fmulx		%fp0,%fp1	 | ...S(B5+...)
    691	fmulx		%fp0,%fp2	 | ...S(A4+...)
    692
    693	faddd		COSB4,%fp1	| ...B4+S(B5+...)
    694	faddd		SINA3,%fp2	| ...A3+S(A4+...)
    695
    696	fmulx		%fp0,%fp1	 | ...S(B4+...)
    697	fmulx		%fp0,%fp2	 | ...S(A3+...)
    698
    699	faddx		COSB3,%fp1	| ...B3+S(B4+...)
    700	faddx		SINA2,%fp2	| ...A2+S(A3+...)
    701
    702	fmulx		%fp0,%fp1	 | ...S(B3+...)
    703	fmulx		%fp0,%fp2	 | ...S(A2+...)
    704
    705	faddx		COSB2,%fp1	| ...B2+S(B3+...)
    706	faddx		SINA1,%fp2	| ...A1+S(A2+...)
    707
    708	fmulx		%fp0,%fp1	 | ...S(B2+...)
    709	fmulx		%fp2,%fp0	 | ...s(a1+...)
    710
    711
    712
    713	fadds		COSB1,%fp1	| ...B1+S(B2...)
    714	fmulx		RPRIME(%a6),%fp0	| ...R'S(A1+...)
    715	fmulx		SPRIME(%a6),%fp1	| ...S'(B1+S(B2+...))
    716
    717	movel		%d1,-(%sp)	|save users mode & precision
    718	andil		#0xff,%d1		|mask off all exceptions
    719	fmovel		%d1,%FPCR
    720	fadds		POSNEG1(%a6),%fp1	| ...COS(X)
    721	bsr		sto_cos		|store cosine result
    722	fmovel		(%sp)+,%FPCR	|restore users exceptions
    723	faddx		RPRIME(%a6),%fp0	| ...SIN(X)
    724
    725	bra		t_frcinx
    726
    727SCBORS:
    728	cmpil		#0x3FFF8000,%d0
    729	bgt		REDUCEX
    730
    731
    732SCSM:
    733	movew		#0x0000,XDCARE(%a6)
    734	fmoves		#0x3F800000,%fp1
    735
    736	movel		%d1,-(%sp)	|save users mode & precision
    737	andil		#0xff,%d1		|mask off all exceptions
    738	fmovel		%d1,%FPCR
    739	fsubs		#0x00800000,%fp1
    740	bsr		sto_cos		|store cosine result
    741	fmovel		(%sp)+,%FPCR	|restore users exceptions
    742	fmovex		X(%a6),%fp0
    743	bra		t_frcinx
    744
    745	|end