| 1 |
30 |
unneback |
//
|
| 2 |
|
|
// $Id: scale.S,v 1.2 2001-09-27 12:01:22 chris Exp $
|
| 3 |
|
|
//
|
| 4 |
|
|
// scale.sa 3.3 7/30/91
|
| 5 |
|
|
//
|
| 6 |
|
|
// The entry point sSCALE computes the destination operand
|
| 7 |
|
|
// scaled by the source operand. If the absolute value of
|
| 8 |
|
|
// the source operand is (>= 2^14) an overflow or underflow
|
| 9 |
|
|
// is returned.
|
| 10 |
|
|
//
|
| 11 |
|
|
// The entry point sscale is called from do_func to emulate
|
| 12 |
|
|
// the fscale unimplemented instruction.
|
| 13 |
|
|
//
|
| 14 |
|
|
// Input: Double-extended destination operand in FPTEMP,
|
| 15 |
|
|
// double-extended source operand in ETEMP.
|
| 16 |
|
|
//
|
| 17 |
|
|
// Output: The function returns scale(X,Y) to fp0.
|
| 18 |
|
|
//
|
| 19 |
|
|
// Modifies: fp0.
|
| 20 |
|
|
//
|
| 21 |
|
|
// Algorithm:
|
| 22 |
|
|
//
|
| 23 |
|
|
// Copyright (C) Motorola, Inc. 1990
|
| 24 |
|
|
// All Rights Reserved
|
| 25 |
|
|
//
|
| 26 |
|
|
// THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA
|
| 27 |
|
|
// The copyright notice above does not evidence any
|
| 28 |
|
|
// actual or intended publication of such source code.
|
| 29 |
|
|
|
| 30 |
|
|
//SCALE idnt 2,1 | Motorola 040 Floating Point Software Package
|
| 31 |
|
|
|
| 32 |
|
|
|section 8
|
| 33 |
|
|
|
| 34 |
|
|
#include "fpsp.defs"
|
| 35 |
|
|
|
| 36 |
|
|
|xref t_ovfl2
|
| 37 |
|
|
|xref t_unfl
|
| 38 |
|
|
|xref round
|
| 39 |
|
|
|xref t_resdnrm
|
| 40 |
|
|
|
| 41 |
|
|
SRC_BNDS: .short 0x3fff,0x400c
|
| 42 |
|
|
|
| 43 |
|
|
//
|
| 44 |
|
|
// This entry point is used by the unimplemented instruction exception
|
| 45 |
|
|
// handler.
|
| 46 |
|
|
//
|
| 47 |
|
|
//
|
| 48 |
|
|
//
|
| 49 |
|
|
// FSCALE
|
| 50 |
|
|
//
|
| 51 |
|
|
.global sscale
|
| 52 |
|
|
sscale:
|
| 53 |
|
|
fmovel #0,%fpcr //clr user enabled exc
|
| 54 |
|
|
clrl %d1
|
| 55 |
|
|
movew FPTEMP(%a6),%d1 //get dest exponent
|
| 56 |
|
|
smi L_SCR1(%a6) //use L_SCR1 to hold sign
|
| 57 |
|
|
andil #0x7fff,%d1 //strip sign
|
| 58 |
|
|
movew ETEMP(%a6),%d0 //check src bounds
|
| 59 |
|
|
andiw #0x7fff,%d0 //clr sign bit
|
| 60 |
|
|
cmp2w SRC_BNDS,%d0
|
| 61 |
|
|
bccs src_in
|
| 62 |
|
|
cmpiw #0x400c,%d0 //test for too large
|
| 63 |
|
|
bge src_out
|
| 64 |
|
|
//
|
| 65 |
|
|
// The source input is below 1, so we check for denormalized numbers
|
| 66 |
|
|
// and set unfl.
|
| 67 |
|
|
//
|
| 68 |
|
|
src_small:
|
| 69 |
|
|
moveb DTAG(%a6),%d0
|
| 70 |
|
|
andib #0xe0,%d0
|
| 71 |
|
|
tstb %d0
|
| 72 |
|
|
beqs no_denorm
|
| 73 |
|
|
st STORE_FLG(%a6) //dest already contains result
|
| 74 |
|
|
orl #unfl_mask,USER_FPSR(%a6) //set UNFL
|
| 75 |
|
|
den_done:
|
| 76 |
|
|
leal FPTEMP(%a6),%a0
|
| 77 |
|
|
bra t_resdnrm
|
| 78 |
|
|
no_denorm:
|
| 79 |
|
|
fmovel USER_FPCR(%a6),%FPCR
|
| 80 |
|
|
fmovex FPTEMP(%a6),%fp0 //simply return dest
|
| 81 |
|
|
rts
|
| 82 |
|
|
|
| 83 |
|
|
|
| 84 |
|
|
//
|
| 85 |
|
|
// Source is within 2^14 range. To perform the int operation,
|
| 86 |
|
|
// move it to d0.
|
| 87 |
|
|
//
|
| 88 |
|
|
src_in:
|
| 89 |
|
|
fmovex ETEMP(%a6),%fp0 //move in src for int
|
| 90 |
|
|
fmovel #rz_mode,%fpcr //force rz for src conversion
|
| 91 |
|
|
fmovel %fp0,%d0 //int src to d0
|
| 92 |
|
|
fmovel #0,%FPSR //clr status from above
|
| 93 |
|
|
tstw ETEMP(%a6) //check src sign
|
| 94 |
|
|
blt src_neg
|
| 95 |
|
|
//
|
| 96 |
|
|
// Source is positive. Add the src to the dest exponent.
|
| 97 |
|
|
// The result can be denormalized, if src = 0, or overflow,
|
| 98 |
|
|
// if the result of the add sets a bit in the upper word.
|
| 99 |
|
|
//
|
| 100 |
|
|
src_pos:
|
| 101 |
|
|
tstw %d1 //check for denorm
|
| 102 |
|
|
beq dst_dnrm
|
| 103 |
|
|
addl %d0,%d1 //add src to dest exp
|
| 104 |
|
|
beqs denorm //if zero, result is denorm
|
| 105 |
|
|
cmpil #0x7fff,%d1 //test for overflow
|
| 106 |
|
|
bges ovfl
|
| 107 |
|
|
tstb L_SCR1(%a6)
|
| 108 |
|
|
beqs spos_pos
|
| 109 |
|
|
orw #0x8000,%d1
|
| 110 |
|
|
spos_pos:
|
| 111 |
|
|
movew %d1,FPTEMP(%a6) //result in FPTEMP
|
| 112 |
|
|
fmovel USER_FPCR(%a6),%FPCR
|
| 113 |
|
|
fmovex FPTEMP(%a6),%fp0 //write result to fp0
|
| 114 |
|
|
rts
|
| 115 |
|
|
ovfl:
|
| 116 |
|
|
tstb L_SCR1(%a6)
|
| 117 |
|
|
beqs sovl_pos
|
| 118 |
|
|
orw #0x8000,%d1
|
| 119 |
|
|
sovl_pos:
|
| 120 |
|
|
movew FPTEMP(%a6),ETEMP(%a6) //result in ETEMP
|
| 121 |
|
|
movel FPTEMP_HI(%a6),ETEMP_HI(%a6)
|
| 122 |
|
|
movel FPTEMP_LO(%a6),ETEMP_LO(%a6)
|
| 123 |
|
|
bra t_ovfl2
|
| 124 |
|
|
|
| 125 |
|
|
denorm:
|
| 126 |
|
|
tstb L_SCR1(%a6)
|
| 127 |
|
|
beqs den_pos
|
| 128 |
|
|
orw #0x8000,%d1
|
| 129 |
|
|
den_pos:
|
| 130 |
|
|
tstl FPTEMP_HI(%a6) //check j bit
|
| 131 |
|
|
blts nden_exit //if set, not denorm
|
| 132 |
|
|
movew %d1,ETEMP(%a6) //input expected in ETEMP
|
| 133 |
|
|
movel FPTEMP_HI(%a6),ETEMP_HI(%a6)
|
| 134 |
|
|
movel FPTEMP_LO(%a6),ETEMP_LO(%a6)
|
| 135 |
|
|
orl #unfl_bit,USER_FPSR(%a6) //set unfl
|
| 136 |
|
|
leal ETEMP(%a6),%a0
|
| 137 |
|
|
bra t_resdnrm
|
| 138 |
|
|
nden_exit:
|
| 139 |
|
|
movew %d1,FPTEMP(%a6) //result in FPTEMP
|
| 140 |
|
|
fmovel USER_FPCR(%a6),%FPCR
|
| 141 |
|
|
fmovex FPTEMP(%a6),%fp0 //write result to fp0
|
| 142 |
|
|
rts
|
| 143 |
|
|
|
| 144 |
|
|
//
|
| 145 |
|
|
// Source is negative. Add the src to the dest exponent.
|
| 146 |
|
|
// (The result exponent will be reduced). The result can be
|
| 147 |
|
|
// denormalized.
|
| 148 |
|
|
//
|
| 149 |
|
|
src_neg:
|
| 150 |
|
|
addl %d0,%d1 //add src to dest
|
| 151 |
|
|
beqs denorm //if zero, result is denorm
|
| 152 |
|
|
blts fix_dnrm //if negative, result is
|
| 153 |
|
|
// ;needing denormalization
|
| 154 |
|
|
tstb L_SCR1(%a6)
|
| 155 |
|
|
beqs sneg_pos
|
| 156 |
|
|
orw #0x8000,%d1
|
| 157 |
|
|
sneg_pos:
|
| 158 |
|
|
movew %d1,FPTEMP(%a6) //result in FPTEMP
|
| 159 |
|
|
fmovel USER_FPCR(%a6),%FPCR
|
| 160 |
|
|
fmovex FPTEMP(%a6),%fp0 //write result to fp0
|
| 161 |
|
|
rts
|
| 162 |
|
|
|
| 163 |
|
|
|
| 164 |
|
|
//
|
| 165 |
|
|
// The result exponent is below denorm value. Test for catastrophic
|
| 166 |
|
|
// underflow and force zero if true. If not, try to shift the
|
| 167 |
|
|
// mantissa right until a zero exponent exists.
|
| 168 |
|
|
//
|
| 169 |
|
|
fix_dnrm:
|
| 170 |
|
|
cmpiw #0xffc0,%d1 //lower bound for normalization
|
| 171 |
|
|
blt fix_unfl //if lower, catastrophic unfl
|
| 172 |
|
|
movew %d1,%d0 //use d0 for exp
|
| 173 |
|
|
movel %d2,-(%a7) //free d2 for norm
|
| 174 |
|
|
movel FPTEMP_HI(%a6),%d1
|
| 175 |
|
|
movel FPTEMP_LO(%a6),%d2
|
| 176 |
|
|
clrl L_SCR2(%a6)
|
| 177 |
|
|
fix_loop:
|
| 178 |
|
|
addw #1,%d0 //drive d0 to 0
|
| 179 |
|
|
lsrl #1,%d1 //while shifting the
|
| 180 |
|
|
roxrl #1,%d2 //mantissa to the right
|
| 181 |
|
|
bccs no_carry
|
| 182 |
|
|
st L_SCR2(%a6) //use L_SCR2 to capture inex
|
| 183 |
|
|
no_carry:
|
| 184 |
|
|
tstw %d0 //it is finished when
|
| 185 |
|
|
blts fix_loop //d0 is zero or the mantissa
|
| 186 |
|
|
tstb L_SCR2(%a6)
|
| 187 |
|
|
beqs tst_zero
|
| 188 |
|
|
orl #unfl_inx_mask,USER_FPSR(%a6)
|
| 189 |
|
|
// ;set unfl, aunfl, ainex
|
| 190 |
|
|
//
|
| 191 |
|
|
// Test for zero. If zero, simply use fmove to return +/- zero
|
| 192 |
|
|
// to the fpu.
|
| 193 |
|
|
//
|
| 194 |
|
|
tst_zero:
|
| 195 |
|
|
clrw FPTEMP_EX(%a6)
|
| 196 |
|
|
tstb L_SCR1(%a6) //test for sign
|
| 197 |
|
|
beqs tst_con
|
| 198 |
|
|
orw #0x8000,FPTEMP_EX(%a6) //set sign bit
|
| 199 |
|
|
tst_con:
|
| 200 |
|
|
movel %d1,FPTEMP_HI(%a6)
|
| 201 |
|
|
movel %d2,FPTEMP_LO(%a6)
|
| 202 |
|
|
movel (%a7)+,%d2
|
| 203 |
|
|
tstl %d1
|
| 204 |
|
|
bnes not_zero
|
| 205 |
|
|
tstl FPTEMP_LO(%a6)
|
| 206 |
|
|
bnes not_zero
|
| 207 |
|
|
//
|
| 208 |
|
|
// Result is zero. Check for rounding mode to set lsb. If the
|
| 209 |
|
|
// mode is rp, and the zero is positive, return smallest denorm.
|
| 210 |
|
|
// If the mode is rm, and the zero is negative, return smallest
|
| 211 |
|
|
// negative denorm.
|
| 212 |
|
|
//
|
| 213 |
|
|
btstb #5,FPCR_MODE(%a6) //test if rm or rp
|
| 214 |
|
|
beqs no_dir
|
| 215 |
|
|
btstb #4,FPCR_MODE(%a6) //check which one
|
| 216 |
|
|
beqs zer_rm
|
| 217 |
|
|
zer_rp:
|
| 218 |
|
|
tstb L_SCR1(%a6) //check sign
|
| 219 |
|
|
bnes no_dir //if set, neg op, no inc
|
| 220 |
|
|
movel #1,FPTEMP_LO(%a6) //set lsb
|
| 221 |
|
|
bras sm_dnrm
|
| 222 |
|
|
zer_rm:
|
| 223 |
|
|
tstb L_SCR1(%a6) //check sign
|
| 224 |
|
|
beqs no_dir //if clr, neg op, no inc
|
| 225 |
|
|
movel #1,FPTEMP_LO(%a6) //set lsb
|
| 226 |
|
|
orl #neg_mask,USER_FPSR(%a6) //set N
|
| 227 |
|
|
bras sm_dnrm
|
| 228 |
|
|
no_dir:
|
| 229 |
|
|
fmovel USER_FPCR(%a6),%FPCR
|
| 230 |
|
|
fmovex FPTEMP(%a6),%fp0 //use fmove to set cc's
|
| 231 |
|
|
rts
|
| 232 |
|
|
|
| 233 |
|
|
//
|
| 234 |
|
|
// The rounding mode changed the zero to a smallest denorm. Call
|
| 235 |
|
|
// t_resdnrm with exceptional operand in ETEMP.
|
| 236 |
|
|
//
|
| 237 |
|
|
sm_dnrm:
|
| 238 |
|
|
movel FPTEMP_EX(%a6),ETEMP_EX(%a6)
|
| 239 |
|
|
movel FPTEMP_HI(%a6),ETEMP_HI(%a6)
|
| 240 |
|
|
movel FPTEMP_LO(%a6),ETEMP_LO(%a6)
|
| 241 |
|
|
leal ETEMP(%a6),%a0
|
| 242 |
|
|
bra t_resdnrm
|
| 243 |
|
|
|
| 244 |
|
|
//
|
| 245 |
|
|
// Result is still denormalized.
|
| 246 |
|
|
//
|
| 247 |
|
|
not_zero:
|
| 248 |
|
|
orl #unfl_mask,USER_FPSR(%a6) //set unfl
|
| 249 |
|
|
tstb L_SCR1(%a6) //check for sign
|
| 250 |
|
|
beqs fix_exit
|
| 251 |
|
|
orl #neg_mask,USER_FPSR(%a6) //set N
|
| 252 |
|
|
fix_exit:
|
| 253 |
|
|
bras sm_dnrm
|
| 254 |
|
|
|
| 255 |
|
|
|
| 256 |
|
|
//
|
| 257 |
|
|
// The result has underflowed to zero. Return zero and set
|
| 258 |
|
|
// unfl, aunfl, and ainex.
|
| 259 |
|
|
//
|
| 260 |
|
|
fix_unfl:
|
| 261 |
|
|
orl #unfl_inx_mask,USER_FPSR(%a6)
|
| 262 |
|
|
btstb #5,FPCR_MODE(%a6) //test if rm or rp
|
| 263 |
|
|
beqs no_dir2
|
| 264 |
|
|
btstb #4,FPCR_MODE(%a6) //check which one
|
| 265 |
|
|
beqs zer_rm2
|
| 266 |
|
|
zer_rp2:
|
| 267 |
|
|
tstb L_SCR1(%a6) //check sign
|
| 268 |
|
|
bnes no_dir2 //if set, neg op, no inc
|
| 269 |
|
|
clrl FPTEMP_EX(%a6)
|
| 270 |
|
|
clrl FPTEMP_HI(%a6)
|
| 271 |
|
|
movel #1,FPTEMP_LO(%a6) //set lsb
|
| 272 |
|
|
bras sm_dnrm //return smallest denorm
|
| 273 |
|
|
zer_rm2:
|
| 274 |
|
|
tstb L_SCR1(%a6) //check sign
|
| 275 |
|
|
beqs no_dir2 //if clr, neg op, no inc
|
| 276 |
|
|
movew #0x8000,FPTEMP_EX(%a6)
|
| 277 |
|
|
clrl FPTEMP_HI(%a6)
|
| 278 |
|
|
movel #1,FPTEMP_LO(%a6) //set lsb
|
| 279 |
|
|
orl #neg_mask,USER_FPSR(%a6) //set N
|
| 280 |
|
|
bra sm_dnrm //return smallest denorm
|
| 281 |
|
|
|
| 282 |
|
|
no_dir2:
|
| 283 |
|
|
tstb L_SCR1(%a6)
|
| 284 |
|
|
bges pos_zero
|
| 285 |
|
|
neg_zero:
|
| 286 |
|
|
clrl FP_SCR1(%a6) //clear the exceptional operand
|
| 287 |
|
|
clrl FP_SCR1+4(%a6) //for gen_except.
|
| 288 |
|
|
clrl FP_SCR1+8(%a6)
|
| 289 |
|
|
fmoves #0x80000000,%fp0
|
| 290 |
|
|
rts
|
| 291 |
|
|
pos_zero:
|
| 292 |
|
|
clrl FP_SCR1(%a6) //clear the exceptional operand
|
| 293 |
|
|
clrl FP_SCR1+4(%a6) //for gen_except.
|
| 294 |
|
|
clrl FP_SCR1+8(%a6)
|
| 295 |
|
|
fmoves #0x00000000,%fp0
|
| 296 |
|
|
rts
|
| 297 |
|
|
|
| 298 |
|
|
//
|
| 299 |
|
|
// The destination is a denormalized number. It must be handled
|
| 300 |
|
|
// by first shifting the bits in the mantissa until it is normalized,
|
| 301 |
|
|
// then adding the remainder of the source to the exponent.
|
| 302 |
|
|
//
|
| 303 |
|
|
dst_dnrm:
|
| 304 |
|
|
moveml %d2/%d3,-(%a7)
|
| 305 |
|
|
movew FPTEMP_EX(%a6),%d1
|
| 306 |
|
|
movel FPTEMP_HI(%a6),%d2
|
| 307 |
|
|
movel FPTEMP_LO(%a6),%d3
|
| 308 |
|
|
dst_loop:
|
| 309 |
|
|
tstl %d2 //test for normalized result
|
| 310 |
|
|
blts dst_norm //exit loop if so
|
| 311 |
|
|
tstl %d0 //otherwise, test shift count
|
| 312 |
|
|
beqs dst_fin //if zero, shifting is done
|
| 313 |
|
|
subil #1,%d0 //dec src
|
| 314 |
|
|
lsll #1,%d3
|
| 315 |
|
|
roxll #1,%d2
|
| 316 |
|
|
bras dst_loop
|
| 317 |
|
|
//
|
| 318 |
|
|
// Destination became normalized. Simply add the remaining
|
| 319 |
|
|
// portion of the src to the exponent.
|
| 320 |
|
|
//
|
| 321 |
|
|
dst_norm:
|
| 322 |
|
|
addw %d0,%d1 //dst is normalized; add src
|
| 323 |
|
|
tstb L_SCR1(%a6)
|
| 324 |
|
|
beqs dnrm_pos
|
| 325 |
|
|
orl #0x8000,%d1
|
| 326 |
|
|
dnrm_pos:
|
| 327 |
|
|
movemw %d1,FPTEMP_EX(%a6)
|
| 328 |
|
|
moveml %d2,FPTEMP_HI(%a6)
|
| 329 |
|
|
moveml %d3,FPTEMP_LO(%a6)
|
| 330 |
|
|
fmovel USER_FPCR(%a6),%FPCR
|
| 331 |
|
|
fmovex FPTEMP(%a6),%fp0
|
| 332 |
|
|
moveml (%a7)+,%d2/%d3
|
| 333 |
|
|
rts
|
| 334 |
|
|
|
| 335 |
|
|
//
|
| 336 |
|
|
// Destination remained denormalized. Call t_excdnrm with
|
| 337 |
|
|
// exceptional operand in ETEMP.
|
| 338 |
|
|
//
|
| 339 |
|
|
dst_fin:
|
| 340 |
|
|
tstb L_SCR1(%a6) //check for sign
|
| 341 |
|
|
beqs dst_exit
|
| 342 |
|
|
orl #neg_mask,USER_FPSR(%a6) //set N
|
| 343 |
|
|
orl #0x8000,%d1
|
| 344 |
|
|
dst_exit:
|
| 345 |
|
|
movemw %d1,ETEMP_EX(%a6)
|
| 346 |
|
|
moveml %d2,ETEMP_HI(%a6)
|
| 347 |
|
|
moveml %d3,ETEMP_LO(%a6)
|
| 348 |
|
|
orl #unfl_mask,USER_FPSR(%a6) //set unfl
|
| 349 |
|
|
moveml (%a7)+,%d2/%d3
|
| 350 |
|
|
leal ETEMP(%a6),%a0
|
| 351 |
|
|
bra t_resdnrm
|
| 352 |
|
|
|
| 353 |
|
|
//
|
| 354 |
|
|
// Source is outside of 2^14 range. Test the sign and branch
|
| 355 |
|
|
// to the appropriate exception handler.
|
| 356 |
|
|
//
|
| 357 |
|
|
src_out:
|
| 358 |
|
|
tstb L_SCR1(%a6)
|
| 359 |
|
|
beqs scro_pos
|
| 360 |
|
|
orl #0x8000,%d1
|
| 361 |
|
|
scro_pos:
|
| 362 |
|
|
movel FPTEMP_HI(%a6),ETEMP_HI(%a6)
|
| 363 |
|
|
movel FPTEMP_LO(%a6),ETEMP_LO(%a6)
|
| 364 |
|
|
tstw ETEMP(%a6)
|
| 365 |
|
|
blts res_neg
|
| 366 |
|
|
res_pos:
|
| 367 |
|
|
movew %d1,ETEMP(%a6) //result in ETEMP
|
| 368 |
|
|
bra t_ovfl2
|
| 369 |
|
|
res_neg:
|
| 370 |
|
|
movew %d1,ETEMP(%a6) //result in ETEMP
|
| 371 |
|
|
leal ETEMP(%a6),%a0
|
| 372 |
|
|
bra t_unfl
|
| 373 |
|
|
|end
|
