git.oblomov.eu Git - linux-2.6/blob - arch/m68k/fpsp040/stan.S

   1 |
   2 |       stan.sa 3.3 7/29/91
   3 |
   4 |       The entry point stan computes the tangent of
   5 |       an input argument;
   6 |       stand does the same except for denormalized input.
   7 |
   8 |       Input: Double-extended number X in location pointed to
   9 |               by address register a0.
  10 |
  11 |       Output: The value tan(X) returned in floating-point register Fp0.
  12 |
  13 |       Accuracy and Monotonicity: The returned result is within 3 ulp in
  14 |               64 significant bit, i.e. within 0.5001 ulp to 53 bits if the
  15 |               result is subsequently rounded to double precision. The
  16 |               result is provably monotonic in double precision.
  17 |
  18 |       Speed: The program sTAN takes approximately 170 cycles for
  19 |               input argument X such that |X| < 15Pi, which is the usual
  20 |               situation.
  21 |
  22 |       Algorithm:
  23 |
  24 |       1. If |X| >= 15Pi or |X| < 2**(-40), go to 6.
  25 |
  26 |       2. Decompose X as X = N(Pi/2) + r where |r| <= Pi/4. Let
  27 |               k = N mod 2, so in particular, k = 0 or 1.
  28 |
  29 |       3. If k is odd, go to 5.
  30 |
  31 |       4. (k is even) Tan(X) = tan(r) and tan(r) is approximated by a
  32 |               rational function U/V where
  33 |               U = r + r*s*(P1 + s*(P2 + s*P3)), and
  34 |               V = 1 + s*(Q1 + s*(Q2 + s*(Q3 + s*Q4))),  s = r*r.
  35 |               Exit.
  36 |
  37 |       4. (k is odd) Tan(X) = -cot(r). Since tan(r) is approximated by a
  38 |               rational function U/V where
  39 |               U = r + r*s*(P1 + s*(P2 + s*P3)), and
  40 |               V = 1 + s*(Q1 + s*(Q2 + s*(Q3 + s*Q4))), s = r*r,
  41 |               -Cot(r) = -V/U. Exit.
  42 |
  43 |       6. If |X| > 1, go to 8.
  44 |
  45 |       7. (|X|<2**(-40)) Tan(X) = X. Exit.
  46 |
  47 |       8. Overwrite X by X := X rem 2Pi. Now that |X| <= Pi, go back to 2.
  48 |
  49
  50 |               Copyright (C) Motorola, Inc. 1990
  51 |                       All Rights Reserved
  52 |
  53 |       For details on the license for this file, please see the
  54 |       file, README, in this same directory.
  55
  56 |STAN   idnt    2,1 | Motorola 040 Floating Point Software Package
  57
  58         |section        8
  59
  60 #include "fpsp.h"
  61
  62 BOUNDS1:        .long 0x3FD78000,0x4004BC7E
  63 TWOBYPI:        .long 0x3FE45F30,0x6DC9C883
  64
  65 TANQ4:  .long 0x3EA0B759,0xF50F8688
  66 TANP3:  .long 0xBEF2BAA5,0xA8924F04
  67
  68 TANQ3:  .long 0xBF346F59,0xB39BA65F,0x00000000,0x00000000
  69
  70 TANP2:  .long 0x3FF60000,0xE073D3FC,0x199C4A00,0x00000000
  71
  72 TANQ2:  .long 0x3FF90000,0xD23CD684,0x15D95FA1,0x00000000
  73
  74 TANP1:  .long 0xBFFC0000,0x8895A6C5,0xFB423BCA,0x00000000
  75
  76 TANQ1:  .long 0xBFFD0000,0xEEF57E0D,0xA84BC8CE,0x00000000
  77
  78 INVTWOPI: .long 0x3FFC0000,0xA2F9836E,0x4E44152A,0x00000000
  79
  80 TWOPI1: .long 0x40010000,0xC90FDAA2,0x00000000,0x00000000
  81 TWOPI2: .long 0x3FDF0000,0x85A308D4,0x00000000,0x00000000
  82
  83 |--N*PI/2, -32 <= N <= 32, IN A LEADING TERM IN EXT. AND TRAILING
  84 |--TERM IN SGL. NOTE THAT PI IS 64-BIT LONG, THUS N*PI/2 IS AT
  85 |--MOST 69 BITS LONG.
  86         .global PITBL
  87 PITBL:
  88   .long  0xC0040000,0xC90FDAA2,0x2168C235,0x21800000
  89   .long  0xC0040000,0xC2C75BCD,0x105D7C23,0xA0D00000
  90   .long  0xC0040000,0xBC7EDCF7,0xFF523611,0xA1E80000
  91   .long  0xC0040000,0xB6365E22,0xEE46F000,0x21480000
  92   .long  0xC0040000,0xAFEDDF4D,0xDD3BA9EE,0xA1200000
  93   .long  0xC0040000,0xA9A56078,0xCC3063DD,0x21FC0000
  94   .long  0xC0040000,0xA35CE1A3,0xBB251DCB,0x21100000
  95   .long  0xC0040000,0x9D1462CE,0xAA19D7B9,0xA1580000
  96   .long  0xC0040000,0x96CBE3F9,0x990E91A8,0x21E00000
  97   .long  0xC0040000,0x90836524,0x88034B96,0x20B00000
  98   .long  0xC0040000,0x8A3AE64F,0x76F80584,0xA1880000
  99   .long  0xC0040000,0x83F2677A,0x65ECBF73,0x21C40000
 100   .long  0xC0030000,0xFB53D14A,0xA9C2F2C2,0x20000000
 101   .long  0xC0030000,0xEEC2D3A0,0x87AC669F,0x21380000
 102   .long  0xC0030000,0xE231D5F6,0x6595DA7B,0xA1300000
 103   .long  0xC0030000,0xD5A0D84C,0x437F4E58,0x9FC00000
 104   .long  0xC0030000,0xC90FDAA2,0x2168C235,0x21000000
 105   .long  0xC0030000,0xBC7EDCF7,0xFF523611,0xA1680000
 106   .long  0xC0030000,0xAFEDDF4D,0xDD3BA9EE,0xA0A00000
 107   .long  0xC0030000,0xA35CE1A3,0xBB251DCB,0x20900000
 108   .long  0xC0030000,0x96CBE3F9,0x990E91A8,0x21600000
 109   .long  0xC0030000,0x8A3AE64F,0x76F80584,0xA1080000
 110   .long  0xC0020000,0xFB53D14A,0xA9C2F2C2,0x1F800000
 111   .long  0xC0020000,0xE231D5F6,0x6595DA7B,0xA0B00000
 112   .long  0xC0020000,0xC90FDAA2,0x2168C235,0x20800000
 113   .long  0xC0020000,0xAFEDDF4D,0xDD3BA9EE,0xA0200000
 114   .long  0xC0020000,0x96CBE3F9,0x990E91A8,0x20E00000
 115   .long  0xC0010000,0xFB53D14A,0xA9C2F2C2,0x1F000000
 116   .long  0xC0010000,0xC90FDAA2,0x2168C235,0x20000000
 117   .long  0xC0010000,0x96CBE3F9,0x990E91A8,0x20600000
 118   .long  0xC0000000,0xC90FDAA2,0x2168C235,0x1F800000
 119   .long  0xBFFF0000,0xC90FDAA2,0x2168C235,0x1F000000
 120   .long  0x00000000,0x00000000,0x00000000,0x00000000
 121   .long  0x3FFF0000,0xC90FDAA2,0x2168C235,0x9F000000
 122   .long  0x40000000,0xC90FDAA2,0x2168C235,0x9F800000
 123   .long  0x40010000,0x96CBE3F9,0x990E91A8,0xA0600000
 124   .long  0x40010000,0xC90FDAA2,0x2168C235,0xA0000000
 125   .long  0x40010000,0xFB53D14A,0xA9C2F2C2,0x9F000000
 126   .long  0x40020000,0x96CBE3F9,0x990E91A8,0xA0E00000
 127   .long  0x40020000,0xAFEDDF4D,0xDD3BA9EE,0x20200000
 128   .long  0x40020000,0xC90FDAA2,0x2168C235,0xA0800000
 129   .long  0x40020000,0xE231D5F6,0x6595DA7B,0x20B00000
 130   .long  0x40020000,0xFB53D14A,0xA9C2F2C2,0x9F800000
 131   .long  0x40030000,0x8A3AE64F,0x76F80584,0x21080000
 132   .long  0x40030000,0x96CBE3F9,0x990E91A8,0xA1600000
 133   .long  0x40030000,0xA35CE1A3,0xBB251DCB,0xA0900000
 134   .long  0x40030000,0xAFEDDF4D,0xDD3BA9EE,0x20A00000
 135   .long  0x40030000,0xBC7EDCF7,0xFF523611,0x21680000
 136   .long  0x40030000,0xC90FDAA2,0x2168C235,0xA1000000
 137   .long  0x40030000,0xD5A0D84C,0x437F4E58,0x1FC00000
 138   .long  0x40030000,0xE231D5F6,0x6595DA7B,0x21300000
 139   .long  0x40030000,0xEEC2D3A0,0x87AC669F,0xA1380000
 140   .long  0x40030000,0xFB53D14A,0xA9C2F2C2,0xA0000000
 141   .long  0x40040000,0x83F2677A,0x65ECBF73,0xA1C40000
 142   .long  0x40040000,0x8A3AE64F,0x76F80584,0x21880000
 143   .long  0x40040000,0x90836524,0x88034B96,0xA0B00000
 144   .long  0x40040000,0x96CBE3F9,0x990E91A8,0xA1E00000
 145   .long  0x40040000,0x9D1462CE,0xAA19D7B9,0x21580000
 146   .long  0x40040000,0xA35CE1A3,0xBB251DCB,0xA1100000
 147   .long  0x40040000,0xA9A56078,0xCC3063DD,0xA1FC0000
 148   .long  0x40040000,0xAFEDDF4D,0xDD3BA9EE,0x21200000
 149   .long  0x40040000,0xB6365E22,0xEE46F000,0xA1480000
 150   .long  0x40040000,0xBC7EDCF7,0xFF523611,0x21E80000
 151   .long  0x40040000,0xC2C75BCD,0x105D7C23,0x20D00000
 152   .long  0x40040000,0xC90FDAA2,0x2168C235,0xA1800000
 153
 154         .set    INARG,FP_SCR4
 155
 156         .set    TWOTO63,L_SCR1
 157         .set    ENDFLAG,L_SCR2
 158         .set    N,L_SCR3
 159
 160         | xref  t_frcinx
 161         |xref   t_extdnrm
 162
 163         .global stand
 164 stand:
 165 |--TAN(X) = X FOR DENORMALIZED X
 166
 167         bra             t_extdnrm
 168
 169         .global stan
 170 stan:
 171         fmovex          (%a0),%fp0      | ...LOAD INPUT
 172
 173         movel           (%a0),%d0
 174         movew           4(%a0),%d0
 175         andil           #0x7FFFFFFF,%d0
 176
 177         cmpil           #0x3FD78000,%d0         | ...|X| >= 2**(-40)?
 178         bges            TANOK1
 179         bra             TANSM
 180 TANOK1:
 181         cmpil           #0x4004BC7E,%d0         | ...|X| < 15 PI?
 182         blts            TANMAIN
 183         bra             REDUCEX
 184
 185
 186 TANMAIN:
 187 |--THIS IS THE USUAL CASE, |X| <= 15 PI.
 188 |--THE ARGUMENT REDUCTION IS DONE BY TABLE LOOK UP.
 189         fmovex          %fp0,%fp1
 190         fmuld           TWOBYPI,%fp1    | ...X*2/PI
 191
 192 |--HIDE THE NEXT TWO INSTRUCTIONS
 193         leal            PITBL+0x200,%a1 | ...TABLE OF N*PI/2, N = -32,...,32
 194
 195 |--FP1 IS NOW READY
 196         fmovel          %fp1,%d0                | ...CONVERT TO INTEGER
 197
 198         asll            #4,%d0
 199         addal           %d0,%a1         | ...ADDRESS N*PIBY2 IN Y1, Y2
 200
 201         fsubx           (%a1)+,%fp0     | ...X-Y1
 202 |--HIDE THE NEXT ONE
 203
 204         fsubs           (%a1),%fp0      | ...FP0 IS R = (X-Y1)-Y2
 205
 206         rorl            #5,%d0
 207         andil           #0x80000000,%d0 | ...D0 WAS ODD IFF D0 < 0
 208
 209 TANCONT:
 210
 211         cmpil           #0,%d0
 212         blt             NODD
 213
 214         fmovex          %fp0,%fp1
 215         fmulx           %fp1,%fp1               | ...S = R*R
 216
 217         fmoved          TANQ4,%fp3
 218         fmoved          TANP3,%fp2
 219
 220         fmulx           %fp1,%fp3               | ...SQ4
 221         fmulx           %fp1,%fp2               | ...SP3
 222
 223         faddd           TANQ3,%fp3      | ...Q3+SQ4
 224         faddx           TANP2,%fp2      | ...P2+SP3
 225
 226         fmulx           %fp1,%fp3               | ...S(Q3+SQ4)
 227         fmulx           %fp1,%fp2               | ...S(P2+SP3)
 228
 229         faddx           TANQ2,%fp3      | ...Q2+S(Q3+SQ4)
 230         faddx           TANP1,%fp2      | ...P1+S(P2+SP3)
 231
 232         fmulx           %fp1,%fp3               | ...S(Q2+S(Q3+SQ4))
 233         fmulx           %fp1,%fp2               | ...S(P1+S(P2+SP3))
 234
 235         faddx           TANQ1,%fp3      | ...Q1+S(Q2+S(Q3+SQ4))
 236         fmulx           %fp0,%fp2               | ...RS(P1+S(P2+SP3))
 237
 238         fmulx           %fp3,%fp1               | ...S(Q1+S(Q2+S(Q3+SQ4)))
 239
 240
 241         faddx           %fp2,%fp0               | ...R+RS(P1+S(P2+SP3))
 242
 243
 244         fadds           #0x3F800000,%fp1        | ...1+S(Q1+...)
 245
 246         fmovel          %d1,%fpcr               |restore users exceptions
 247         fdivx           %fp1,%fp0               |last inst - possible exception set
 248
 249         bra             t_frcinx
 250
 251 NODD:
 252         fmovex          %fp0,%fp1
 253         fmulx           %fp0,%fp0               | ...S = R*R
 254
 255         fmoved          TANQ4,%fp3
 256         fmoved          TANP3,%fp2
 257
 258         fmulx           %fp0,%fp3               | ...SQ4
 259         fmulx           %fp0,%fp2               | ...SP3
 260
 261         faddd           TANQ3,%fp3      | ...Q3+SQ4
 262         faddx           TANP2,%fp2      | ...P2+SP3
 263
 264         fmulx           %fp0,%fp3               | ...S(Q3+SQ4)
 265         fmulx           %fp0,%fp2               | ...S(P2+SP3)
 266
 267         faddx           TANQ2,%fp3      | ...Q2+S(Q3+SQ4)
 268         faddx           TANP1,%fp2      | ...P1+S(P2+SP3)
 269
 270         fmulx           %fp0,%fp3               | ...S(Q2+S(Q3+SQ4))
 271         fmulx           %fp0,%fp2               | ...S(P1+S(P2+SP3))
 272
 273         faddx           TANQ1,%fp3      | ...Q1+S(Q2+S(Q3+SQ4))
 274         fmulx           %fp1,%fp2               | ...RS(P1+S(P2+SP3))
 275
 276         fmulx           %fp3,%fp0               | ...S(Q1+S(Q2+S(Q3+SQ4)))
 277
 278
 279         faddx           %fp2,%fp1               | ...R+RS(P1+S(P2+SP3))
 280         fadds           #0x3F800000,%fp0        | ...1+S(Q1+...)
 281
 282
 283         fmovex          %fp1,-(%sp)
 284         eoril           #0x80000000,(%sp)
 285
 286         fmovel          %d1,%fpcr               |restore users exceptions
 287         fdivx           (%sp)+,%fp0     |last inst - possible exception set
 288
 289         bra             t_frcinx
 290
 291 TANBORS:
 292 |--IF |X| > 15PI, WE USE THE GENERAL ARGUMENT REDUCTION.
 293 |--IF |X| < 2**(-40), RETURN X OR 1.
 294         cmpil           #0x3FFF8000,%d0
 295         bgts            REDUCEX
 296
 297 TANSM:
 298
 299         fmovex          %fp0,-(%sp)
 300         fmovel          %d1,%fpcr                |restore users exceptions
 301         fmovex          (%sp)+,%fp0     |last inst - possible exception set
 302
 303         bra             t_frcinx
 304
 305
 306 REDUCEX:
 307 |--WHEN REDUCEX IS USED, THE CODE WILL INEVITABLY BE SLOW.
 308 |--THIS REDUCTION METHOD, HOWEVER, IS MUCH FASTER THAN USING
 309 |--THE REMAINDER INSTRUCTION WHICH IS NOW IN SOFTWARE.
 310
 311         fmovemx %fp2-%fp5,-(%a7)        | ...save FP2 through FP5
 312         movel           %d2,-(%a7)
 313         fmoves         #0x00000000,%fp1
 314
 315 |--If compact form of abs(arg) in d0=$7ffeffff, argument is so large that
 316 |--there is a danger of unwanted overflow in first LOOP iteration.  In this
 317 |--case, reduce argument by one remainder step to make subsequent reduction
 318 |--safe.
 319         cmpil   #0x7ffeffff,%d0         |is argument dangerously large?
 320         bnes    LOOP
 321         movel   #0x7ffe0000,FP_SCR2(%a6)        |yes
 322 |                                       ;create 2**16383*PI/2
 323         movel   #0xc90fdaa2,FP_SCR2+4(%a6)
 324         clrl    FP_SCR2+8(%a6)
 325         ftstx   %fp0                    |test sign of argument
 326         movel   #0x7fdc0000,FP_SCR3(%a6)        |create low half of 2**16383*
 327 |                                       ;PI/2 at FP_SCR3
 328         movel   #0x85a308d3,FP_SCR3+4(%a6)
 329         clrl   FP_SCR3+8(%a6)
 330         fblt    red_neg
 331         orw     #0x8000,FP_SCR2(%a6)    |positive arg
 332         orw     #0x8000,FP_SCR3(%a6)
 333 red_neg:
 334         faddx  FP_SCR2(%a6),%fp0                |high part of reduction is exact
 335         fmovex  %fp0,%fp1               |save high result in fp1
 336         faddx  FP_SCR3(%a6),%fp0                |low part of reduction
 337         fsubx  %fp0,%fp1                        |determine low component of result
 338         faddx  FP_SCR3(%a6),%fp1                |fp0/fp1 are reduced argument.
 339
 340 |--ON ENTRY, FP0 IS X, ON RETURN, FP0 IS X REM PI/2, |X| <= PI/4.
 341 |--integer quotient will be stored in N
 342 |--Intermediate remainder is 66-bit long; (R,r) in (FP0,FP1)
 343
 344 LOOP:
 345         fmovex          %fp0,INARG(%a6) | ...+-2**K * F, 1 <= F < 2
 346         movew           INARG(%a6),%d0
 347         movel          %d0,%a1          | ...save a copy of D0
 348         andil           #0x00007FFF,%d0
 349         subil           #0x00003FFF,%d0 | ...D0 IS K
 350         cmpil           #28,%d0
 351         bles            LASTLOOP
 352 CONTLOOP:
 353         subil           #27,%d0  | ...D0 IS L := K-27
 354         movel           #0,ENDFLAG(%a6)
 355         bras            WORK
 356 LASTLOOP:
 357         clrl            %d0             | ...D0 IS L := 0
 358         movel           #1,ENDFLAG(%a6)
 359
 360 WORK:
 361 |--FIND THE REMAINDER OF (R,r) W.R.T.   2**L * (PI/2). L IS SO CHOSEN
 362 |--THAT INT( X * (2/PI) / 2**(L) ) < 2**29.
 363
 364 |--CREATE 2**(-L) * (2/PI), SIGN(INARG)*2**(63),
 365 |--2**L * (PIby2_1), 2**L * (PIby2_2)
 366
 367         movel           #0x00003FFE,%d2 | ...BIASED EXPO OF 2/PI
 368         subl            %d0,%d2         | ...BIASED EXPO OF 2**(-L)*(2/PI)
 369
 370         movel           #0xA2F9836E,FP_SCR1+4(%a6)
 371         movel           #0x4E44152A,FP_SCR1+8(%a6)
 372         movew           %d2,FP_SCR1(%a6)        | ...FP_SCR1 is 2**(-L)*(2/PI)
 373
 374         fmovex          %fp0,%fp2
 375         fmulx           FP_SCR1(%a6),%fp2
 376 |--WE MUST NOW FIND INT(FP2). SINCE WE NEED THIS VALUE IN
 377 |--FLOATING POINT FORMAT, THE TWO FMOVE'S       FMOVE.L FP <--> N
 378 |--WILL BE TOO INEFFICIENT. THE WAY AROUND IT IS THAT
 379 |--(SIGN(INARG)*2**63   +       FP2) - SIGN(INARG)*2**63 WILL GIVE
 380 |--US THE DESIRED VALUE IN FLOATING POINT.
 381
 382 |--HIDE SIX CYCLES OF INSTRUCTION
 383         movel           %a1,%d2
 384         swap            %d2
 385         andil           #0x80000000,%d2
 386         oril            #0x5F000000,%d2 | ...D2 IS SIGN(INARG)*2**63 IN SGL
 387         movel           %d2,TWOTO63(%a6)
 388
 389         movel           %d0,%d2
 390         addil           #0x00003FFF,%d2 | ...BIASED EXPO OF 2**L * (PI/2)
 391
 392 |--FP2 IS READY
 393         fadds           TWOTO63(%a6),%fp2       | ...THE FRACTIONAL PART OF FP1 IS ROUNDED
 394
 395 |--HIDE 4 CYCLES OF INSTRUCTION; creating 2**(L)*Piby2_1  and  2**(L)*Piby2_2
 396         movew           %d2,FP_SCR2(%a6)
 397         clrw           FP_SCR2+2(%a6)
 398         movel           #0xC90FDAA2,FP_SCR2+4(%a6)
 399         clrl            FP_SCR2+8(%a6)          | ...FP_SCR2 is  2**(L) * Piby2_1
 400
 401 |--FP2 IS READY
 402         fsubs           TWOTO63(%a6),%fp2               | ...FP2 is N
 403
 404         addil           #0x00003FDD,%d0
 405         movew           %d0,FP_SCR3(%a6)
 406         clrw           FP_SCR3+2(%a6)
 407         movel           #0x85A308D3,FP_SCR3+4(%a6)
 408         clrl            FP_SCR3+8(%a6)          | ...FP_SCR3 is 2**(L) * Piby2_2
 409
 410         movel           ENDFLAG(%a6),%d0
 411
 412 |--We are now ready to perform (R+r) - N*P1 - N*P2, P1 = 2**(L) * Piby2_1 and
 413 |--P2 = 2**(L) * Piby2_2
 414         fmovex          %fp2,%fp4
 415         fmulx           FP_SCR2(%a6),%fp4               | ...W = N*P1
 416         fmovex          %fp2,%fp5
 417         fmulx           FP_SCR3(%a6),%fp5               | ...w = N*P2
 418         fmovex          %fp4,%fp3
 419 |--we want P+p = W+w  but  |p| <= half ulp of P
 420 |--Then, we need to compute  A := R-P   and  a := r-p
 421         faddx           %fp5,%fp3                       | ...FP3 is P
 422         fsubx           %fp3,%fp4                       | ...W-P
 423
 424         fsubx           %fp3,%fp0                       | ...FP0 is A := R - P
 425         faddx           %fp5,%fp4                       | ...FP4 is p = (W-P)+w
 426
 427         fmovex          %fp0,%fp3                       | ...FP3 A
 428         fsubx           %fp4,%fp1                       | ...FP1 is a := r - p
 429
 430 |--Now we need to normalize (A,a) to  "new (R,r)" where R+r = A+a but
 431 |--|r| <= half ulp of R.
 432         faddx           %fp1,%fp0                       | ...FP0 is R := A+a
 433 |--No need to calculate r if this is the last loop
 434         cmpil           #0,%d0
 435         bgt             RESTORE
 436
 437 |--Need to calculate r
 438         fsubx           %fp0,%fp3                       | ...A-R
 439         faddx           %fp3,%fp1                       | ...FP1 is r := (A-R)+a
 440         bra             LOOP
 441
 442 RESTORE:
 443         fmovel          %fp2,N(%a6)
 444         movel           (%a7)+,%d2
 445         fmovemx (%a7)+,%fp2-%fp5
 446
 447
 448         movel           N(%a6),%d0
 449         rorl            #1,%d0
 450
 451
 452         bra             TANCONT
 453
 454         |end