home *** CD-ROM | disk | FTP | other *** search
- | mjr: not needed _normdf for -__M68881__ or the sfp004
- | however, _infinitydf is retained
-
- |#######################################################################
-
- | double floating point normalization routine
- |
- | written by Kai-Uwe Bloem (I5110401@dbstu1.bitnet).
- | Based on a 80x86 floating point packet from comp.os.minix, written by P.Housel
- |
- | Revision 1.6.3 michal 05-93 (ntomczak@vm.ucs.ualberta.ca)
- | + restored and ensured future synchronization with errno codes
- | + removed bogus error when normalizing legitimate zero
- | + small mods to shave off few cycles
- |
- | patched by Olaf Flebbe (flebbe@tat.physik.uni-tuebingen.de)
- |
- | Revision 1.6.2 olaf 12-92:
- | + added support for -0.
- |
- | Revision 1.6.1 olaf 10-92:
- | + corrected rounding in tie case: round up, not down.
- | (needed for enquire 4.3)
- |
- | Revision 1.6, kub 04-90 :
- | more robust handling exponent and sign handling for 32 bit integers. There
- | are now overflow tests for 32 bit exponents, and bit 31 of the sign flag
- | is or ed to bit 15 for later checks (i.e. both bits 31 and 15 are now sign
- | bits). Take care, the upper 16 bits of rounding info are ignored for 32 bit
- | integers !
- |
- | Revision 1.5, ++jrb 03-90:
- | change __normdf interface to expect ints instead of shorts. easier
- | to interface to 32 bit int code. this file is now pre-processed,
- | with __MSHORT__ defined when ints are 16 bits.
- |
- | Revision 1.4, kub 03-90 :
- | export ___normdf entry to C language. Rename the internal entry to a name
- | not accessible from C to prevent crashes
- |
- | Revision 1.3, kub 01-90 :
- | added support for denormalized numbers
- |
- | Revision 1.2, kub 01-90 :
- | replace far shifts by swaps to gain speed
- |
- | Revision 1.1, kub 12-89 :
- | Ported over to 68k assembler
- |
- | Revision 1.0:
- | original 8088 code from P.S.Housel
-
- .text
- .even
- .globl __infinitydf
-
- #if !defined (__M68881__) && !defined (sfp004)
- .globl ___normdf
- .globl norm_df
- #include "errbase.h"
- # ifdef ERROR_CHECK
- LC0:
- .ascii "normdf: OVERFLOW\12\15\0"
- .even
- # endif ERROR_CHECK
-
- | C entry, for procs dealing with the internal representation :
- | double __normdf(long long mant, int exp, int sign, int rbits);
- ___normdf:
- lea sp@(4),a0 | parameter pointer
- moveml d2-d7,sp@- | save working registers
- moveml a0@+,d4-d5 | get mantissa
-
- # ifdef __MSHORT__
- movew a0@+,d0 | get exponent
- movew a0@+,d2 | get sign
- movew a0@+,d1 | rounding information
- # else
- movel a0@+,d0 | get exponent
- movel a0@+,d2 | get sign
- bpl 0f | or bit 31 to bit 15 for later tests
- bset #15,d2
- 0: movel a0@+,d1 | rounding information
-
- movel #0x7fff,d3
- cmpl d3,d0 | test exponent
- bgt oflow
- notl d3 | #-0x8000 -> d3
- cmpl d3,d0
- blt retz
- # endif
-
- | internal entry for floating point package, saves time
- | d0=u.exp, d2=u.sign, d1=rounding bits, d4/d5=mantissa
- | registers d2-d7 must be saved on the stack !
- norm_df:
- movel d4,d3 | rounding and u.mant == 0 ?
- orl d5,d3
- bne 1f
- tstb d1
- beq retzok
- 1:
- movel d4,d3
- andl #0xfffff000,d3 | fast shift, 16 bits ?
- bne 2f
- cmpw #9,d0 | shift is going to far; do normal shift
- ble 2f | (minimize shifts here : 10l = 16l + 6r)
- swap d4 | yes, swap register halfs
- swap d5
- movew d5,d4
- moveb d1,d5 | some doubt about this one !
- lslw #8,d5
- clrw d1
- subw #16,d0 | account for swap
- bra 1b
- 2:
- clrb d2 | sticky byte
- movel #0xffe00000,d6
- 3: tstw d0 | divide (shift)
- ble 0f | denormalized number
- movel d4,d3
- andl d6,d3 | or until no bits above 53
- beq 4f
- 0: addw #1,d0 | increment exponent
- lsrl #1,d4
- roxrl #1,d5
- orb d1,d2 | set sticky
- roxrb #1,d1 | shift into rounding bits
- bra 3b
- 4:
- andb #1,d2
- orb d2,d1 | make least sig bit sticky
- asrl #1,d6 | #0xfff00000 -> d6
- 5: movel d4,d3 | multiply (shift) until
- andl d6,d3 | one in implied position
- bne 6f
- subw #1,d0 | decrement exponent
- beq 6f | too small. store as denormalized number
- addb d1,d1 | some doubt about this one *
- addxl d5,d5
- addxl d4,d4
- bra 5b
- 6:
- tstb d1 | check rounding bits
- bge 8f | round down - no action neccessary
- negb d1
- bvc 7f | round up
- movew d5,d1 | tie case - round to even
- | dont need rounding bits any more
- andw #1,d1 | check if even
- beq 8f | mantissa is even - no action necessary
- | fall through
- 7:
- clrl d1 | zero rounding bits
- addl #1,d5
- addxl d1,d4
- tstw d0
- bne 0f | renormalize if number was denormalized
- addw #1,d0 | correct exponent for denormalized numbers
- bra 2b
- 0: movel d4,d3 | check for rounding overflow
- asll #1,d6 | #0xffe00000 -> d3
- andl d6,d3
- bne 2b | go back and renormalize
- 8:
- movel d4,d3 | check if normalization caused an underflow
- orl d5,d3
- beq retz
- tstw d0 | check for exponent overflow or underflow
- blt retz
- cmpw #2047,d0
- bge oflow
-
- lslw #5,d0 | re-position exponent - one bit too high
- lslw #1,d2 | get X bit
- roxrw #1,d0 | shift it into sign position
- swap d0 | map to upper word
- clrw d0
- andl #0x0fffff,d4 | top mantissa bits
- orl d0,d4 | insert exponent and sign
- movel d4,d0
- movel d5,d1
- moveml sp@+,d2-d7
- rts
-
- retz:
- moveq #Erange,d0
- Emove d0,Errno
- retzok:
- moveq #0,d0 | return zero value
- movel d0,d1
- lslw #1,d2 | set value of extension
- roxrl #1,d0 | and move it to hight bit of d0
- 0: moveml sp@+,d2-d7
- rts
-
- oflow:
- #ifdef ERROR_CHECK
- pea pc@(LC0)
- pea Stderr
- jbsr _fprintf |
- addql #8,a7 |
- moveq #Erange,d0
- Emove d0,Errno
- #endif ERROR_CHECK
-
- | moveml pc@(__infinitydf),d0-d1 | return infinity value
- moveml __infinitydf,d0-d1 | return infinty value
- tstw d2
- bpl 1f
- bset #31,d0
- 1:
- moveml sp@+,d2-d7 | should really cause trap ?!? (mjr: why?)
- rts
-
- #endif __M68881__
-
- __infinitydf: | +infinity as proposed by IEEE
- .long 0x7ff00000,0x00000000
-