THE SOFTWARE is provided on an "AS IS" basis and without warranty.
To the maximum extent permitted by applicable law,
MOTOROLA DISCLAIMS ALL WARRANTIES WHETHER EXPRESS OR IMPLIED,
INCLUDING IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE
and any warranty against infringement with regard to the SOFTWARE
(INCLUDING ANY MODIFIED VERSIONS THEREOF) and any accompanying written materials.
To the maximum extent permitted by applicable law,
IN NO EVENT SHALL MOTOROLA BE LIABLE FOR ANY DAMAGES WHATSOEVER
(INCLUDING WITHOUT LIMITATION, DAMAGES FOR LOSS OF BUSINESS PROFITS,
BUSINESS INTERRUPTION, LOSS OF BUSINESS INFORMATION, OR OTHER PECUNIARY LOSS)
ARISING OF THE USE OR INABILITY TO USE THE SOFTWARE.
Motorola assumes no responsibility for the maintenance and support of the SOFTWARE.
You are hereby granted a copyright license to use, modify, and distribute the SOFTWARE
so long as this entire notice is retained without alteration in any modified and/or
redistributed versions, and that such modified versions are clearly identified as such.
No licenses are granted by implication, estoppel or otherwise under any patents
or trademarks of Motorola, Inc.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# freal.s:
# This file is appended to the top of the 060FPSP package
# and contains the entry points into the package. The user, in
# effect, branches to one of the branch table entries located
# after _060FPSP_TABLE.
# Also, subroutine stubs exist in this file (_fpsp_done for
# example) that are referenced by the FPSP package itself in order
# to call a given routine. The stub routine actually performs the
# callout. The FPSP code does a "bsr" to the stub routine. This
# extra layer of hierarchy adds a slight performance penalty but
# it makes the FPSP code easier to read and more mainatinable.
#
set _off_bsun, 0x00 set _off_snan, 0x04 set _off_operr, 0x08 set _off_ovfl, 0x0c set _off_unfl, 0x10 set _off_dz, 0x14 set _off_inex, 0x18 set _off_fline, 0x1c set _off_fpu_dis, 0x20 set _off_trap, 0x24 set _off_trace, 0x28 set _off_access, 0x2c set _off_done, 0x30
set _off_imr, 0x40 set _off_dmr, 0x44 set _off_dmw, 0x48 set _off_irw, 0x4c set _off_irl, 0x50 set _off_drb, 0x54 set _off_drw, 0x58 set _off_drl, 0x5c set _off_dwb, 0x60 set _off_dww, 0x64 set _off_dwl, 0x68
# Here's the table of ENTRY POINTS for those linking the package.
bra.l _fpsp_snan
short 0x0000
bra.l _fpsp_operr
short 0x0000
bra.l _fpsp_ovfl
short 0x0000
bra.l _fpsp_unfl
short 0x0000
bra.l _fpsp_dz
short 0x0000
bra.l _fpsp_inex
short 0x0000
bra.l _fpsp_fline
short 0x0000
bra.l _fpsp_unsupp
short 0x0000
bra.l _fpsp_effadd
short 0x0000
#
# This file contains a set of define statements for constants
# in order to promote readability within the corecode itself.
#
set LOCAL_SIZE, 192 # stack frame size(bytes) set LV, -LOCAL_SIZE # stack offset
set EXC_SR, 0x4 # stack status register set EXC_PC, 0x6 # stack pc set EXC_VOFF, 0xa # stacked vector offset set EXC_EA, 0xc # stacked <ea>
set EXC_FP, 0x0 # frame pointer
set EXC_AREGS, -68 # offset of all address regs set EXC_DREGS, -100 # offset of all data regs set EXC_FPREGS, -36 # offset of all fp regs
set EXC_A7, EXC_AREGS+(7*4) # offset of saved a7 set OLD_A7, EXC_AREGS+(6*4) # extra copy of saved a7 set EXC_A6, EXC_AREGS+(6*4) # offset of saved a6 set EXC_A5, EXC_AREGS+(5*4) set EXC_A4, EXC_AREGS+(4*4) set EXC_A3, EXC_AREGS+(3*4) set EXC_A2, EXC_AREGS+(2*4) set EXC_A1, EXC_AREGS+(1*4) set EXC_A0, EXC_AREGS+(0*4) set EXC_D7, EXC_DREGS+(7*4) set EXC_D6, EXC_DREGS+(6*4) set EXC_D5, EXC_DREGS+(5*4) set EXC_D4, EXC_DREGS+(4*4) set EXC_D3, EXC_DREGS+(3*4) set EXC_D2, EXC_DREGS+(2*4) set EXC_D1, EXC_DREGS+(1*4) set EXC_D0, EXC_DREGS+(0*4)
set EXC_FP0, EXC_FPREGS+(0*12) # offset of saved fp0 set EXC_FP1, EXC_FPREGS+(1*12) # offset of saved fp1 set EXC_FP2, EXC_FPREGS+(2*12) # offset of saved fp2 (not used)
set FP_SCR1, LV+80 # fp scratch 1 set FP_SCR1_EX, FP_SCR1+0 set FP_SCR1_SGN, FP_SCR1+2 set FP_SCR1_HI, FP_SCR1+4 set FP_SCR1_LO, FP_SCR1+8
set FP_SCR0, LV+68 # fp scratch 0 set FP_SCR0_EX, FP_SCR0+0 set FP_SCR0_SGN, FP_SCR0+2 set FP_SCR0_HI, FP_SCR0+4 set FP_SCR0_LO, FP_SCR0+8
set FP_DST, LV+56 # fp destination operand set FP_DST_EX, FP_DST+0 set FP_DST_SGN, FP_DST+2 set FP_DST_HI, FP_DST+4 set FP_DST_LO, FP_DST+8
set FP_SRC, LV+44 # fp source operand set FP_SRC_EX, FP_SRC+0 set FP_SRC_SGN, FP_SRC+2 set FP_SRC_HI, FP_SRC+4 set FP_SRC_LO, FP_SRC+8
set USER_FPIAR, LV+40 # FP instr address register
set USER_FPSR, LV+36 # FP status register set FPSR_CC, USER_FPSR+0 # FPSR condition codes set FPSR_QBYTE, USER_FPSR+1 # FPSR qoutient byte set FPSR_EXCEPT, USER_FPSR+2 # FPSR exception status byte set FPSR_AEXCEPT, USER_FPSR+3 # FPSR accrued exception byte
set USER_FPCR, LV+32 # FP control register set FPCR_ENABLE, USER_FPCR+2 # FPCR exception enable set FPCR_MODE, USER_FPCR+3 # FPCR rounding mode control
set L_SCR3, LV+28 # integer scratch 3 set L_SCR2, LV+24 # integer scratch 2 set L_SCR1, LV+20 # integer scratch 1
set STORE_FLG, LV+19 # flag: operand store (ie. not fcmp/ftst)
set EXC_TEMP2, LV+24 # temporary space set EXC_TEMP, LV+16 # temporary space
set DTAG, LV+15 # destination operand type set STAG, LV+14 # source operand type
set SPCOND_FLG, LV+10 # flag: special case (see below)
set EXC_CC, LV+8 # saved condition codes set EXC_EXTWPTR, LV+4 # saved current PC (active) set EXC_EXTWORD, LV+2 # saved extension word set EXC_CMDREG, LV+2 # saved extension word set EXC_OPWORD, LV+0 # saved operation word
################################
# Helpful macros
set FTEMP, 0 # offsets within an set FTEMP_EX, 0 # extended precision set FTEMP_SGN, 2 # value saved in memory. set FTEMP_HI, 4 set FTEMP_LO, 8 set FTEMP_GRS, 12
set LOCAL, 0 # offsets within an set LOCAL_EX, 0 # extended precision set LOCAL_SGN, 2 # value saved in memory. set LOCAL_HI, 4 set LOCAL_LO, 8 set LOCAL_GRS, 12
set DST, 0 # offsets within an set DST_EX, 0 # extended precision set DST_HI, 4 # value saved in memory. set DST_LO, 8
set SRC, 0 # offsets within an set SRC_EX, 0 # extended precision set SRC_HI, 4 # value saved in memory. set SRC_LO, 8
set SGL_LO, 0x3f81 # min sgl prec exponent set SGL_HI, 0x407e # max sgl prec exponent set DBL_LO, 0x3c01 # min dbl prec exponent set DBL_HI, 0x43fe # max dbl prec exponent set EXT_LO, 0x0 # min ext prec exponent set EXT_HI, 0x7ffe # max ext prec exponent
set EXT_BIAS, 0x3fff # extended precision bias set SGL_BIAS, 0x007f # single precision bias set DBL_BIAS, 0x03ff # double precision bias
set NORM, 0x00 # operand type for STAG/DTAG set ZERO, 0x01 # operand type for STAG/DTAG set INF, 0x02 # operand type for STAG/DTAG set QNAN, 0x03 # operand type for STAG/DTAG set DENORM, 0x04 # operand type for STAG/DTAG set SNAN, 0x05 # operand type for STAG/DTAG set UNNORM, 0x06 # operand type for STAG/DTAG
##################
# FPSR/FPCR bits #
################## set neg_bit, 0x3 # negative result set z_bit, 0x2 # zero result set inf_bit, 0x1 # infinite result set nan_bit, 0x0 # NAN result
set q_sn_bit, 0x7 # sign bit of quotient byte
set bsun_bit, 7 # branch on unordered set snan_bit, 6 # signalling NAN set operr_bit, 5 # operand error set ovfl_bit, 4 # overflow set unfl_bit, 3 # underflow set dz_bit, 2 # divide by zero set inex2_bit, 1 # inexact result 2 set inex1_bit, 0 # inexact result 1
set aiop_bit, 7 # accrued inexact operation bit set aovfl_bit, 6 # accrued overflow bit set aunfl_bit, 5 # accrued underflow bit set adz_bit, 4 # accrued dz bit set ainex_bit, 3 # accrued inexact bit
#############################
# FPSR individual bit masks #
############################# set neg_mask, 0x08000000 # negative bit mask (lw) set inf_mask, 0x02000000 # infinity bit mask (lw) set z_mask, 0x04000000 # zero bit mask (lw) set nan_mask, 0x01000000 # nan bit mask (lw)
set neg_bmask, 0x08 # negative bit mask (byte) set inf_bmask, 0x02 # infinity bit mask (byte) set z_bmask, 0x04 # zero bit mask (byte) set nan_bmask, 0x01 # nan bit mask (byte)
set bsun_mask, 0x00008000 # bsun exception mask set snan_mask, 0x00004000 # snan exception mask set operr_mask, 0x00002000 # operr exception mask set ovfl_mask, 0x00001000 # overflow exception mask set unfl_mask, 0x00000800 # underflow exception mask set dz_mask, 0x00000400 # dz exception mask set inex2_mask, 0x00000200 # inex2 exception mask set inex1_mask, 0x00000100 # inex1 exception mask
set aiop_mask, 0x00000080 # accrued illegal operation set aovfl_mask, 0x00000040 # accrued overflow set aunfl_mask, 0x00000020 # accrued underflow set adz_mask, 0x00000010 # accrued divide by zero set ainex_mask, 0x00000008 # accrued inexact
######################################
# FPSR combinations used in the FPSP #
###################################### set dzinf_mask, inf_mask+dz_mask+adz_mask set opnan_mask, nan_mask+operr_mask+aiop_mask set nzi_mask, 0x01ffffff #clears N, Z, and I set unfinx_mask, unfl_mask+inex2_mask+aunfl_mask+ainex_mask set unf2inx_mask, unfl_mask+inex2_mask+ainex_mask set ovfinx_mask, ovfl_mask+inex2_mask+aovfl_mask+ainex_mask set inx1a_mask, inex1_mask+ainex_mask set inx2a_mask, inex2_mask+ainex_mask set snaniop_mask, nan_mask+snan_mask+aiop_mask set snaniop2_mask, snan_mask+aiop_mask set naniop_mask, nan_mask+aiop_mask set neginf_mask, neg_mask+inf_mask set infaiop_mask, inf_mask+aiop_mask set negz_mask, neg_mask+z_mask set opaop_mask, operr_mask+aiop_mask set unfl_inx_mask, unfl_mask+aunfl_mask+ainex_mask set ovfl_inx_mask, ovfl_mask+aovfl_mask+ainex_mask
#########
# misc. #
######### set rnd_stky_bit, 29 # stky bit pos in longword
set sign_bit, 0x7 # sign bit set signan_bit, 0x6 # signalling nan bit
set sgl_thresh, 0x3f81 # minimum sgl exponent set dbl_thresh, 0x3c01 # minimum dbl exponent
set x_mode, 0x0 # extended precision set s_mode, 0x4 # single precision set d_mode, 0x8 # double precision
set rn_mode, 0x0 # round-to-nearest set rz_mode, 0x1 # round-to-zero set rm_mode, 0x2 # round-tp-minus-infinity set rp_mode, 0x3 # round-to-plus-infinity
set BSUN_VEC, 0xc0 # bsun vector offset set INEX_VEC, 0xc4 # inexact vector offset set DZ_VEC, 0xc8 # dz vector offset set UNFL_VEC, 0xcc # unfl vector offset set OPERR_VEC, 0xd0 # operr vector offset set OVFL_VEC, 0xd4 # ovfl vector offset set SNAN_VEC, 0xd8 # snan vector offset
###########################
# SPecial CONDition FLaGs #
########################### set ftrapcc_flg, 0x01 # flag bit: ftrapcc exception set fbsun_flg, 0x02 # flag bit: bsun exception set mia7_flg, 0x04 # flag bit: (a7)+ <ea> set mda7_flg, 0x08 # flag bit: -(a7) <ea> set fmovm_flg, 0x40 # flag bit: fmovm instruction set immed_flg, 0x80 # flag bit: &<data> <ea>
set ftrapcc_bit, 0x0 set fbsun_bit, 0x1 set mia7_bit, 0x2 set mda7_bit, 0x3 set immed_bit, 0x7
##################################
# TRANSCENDENTAL "LAST-OP" FLAGS #
################################## set FMUL_OP, 0x0 # fmul instr performed last set FDIV_OP, 0x1 # fdiv performed last set FADD_OP, 0x2 # fadd performed last set FMOV_OP, 0x3 # fmov performed last
#############
# CONSTANTS #
#############
T1: long 0x40C62D38,0xD3D64634 # 16381 LOG2 LEAD
T2: long 0x3D6F90AE,0xB1E75CC7 # 16381 LOG2 TRAIL
PI: long 0x40000000,0xC90FDAA2,0x2168C235,0x00000000
PIBY2: long 0x3FFF0000,0xC90FDAA2,0x2168C235,0x00000000
TWOBYPI:
long 0x3FE45F30,0x6DC9C883
#########################################################################
# XDEF **************************************************************** #
# _fpsp_ovfl(): 060FPSP entry point for FP Overflow exception. #
# #
# This handler should be the first code executed upon taking the #
# FP Overflow exception in an operating system. #
# #
# XREF **************************************************************** #
# _imem_read_long() - read instruction longword #
# fix_skewed_ops() - adjust src operand in fsave frame #
# set_tag_x() - determine optype of src/dst operands #
# store_fpreg() - store opclass 0 or 2 result to FP regfile #
# unnorm_fix() - change UNNORM operands to NORM or ZERO #
# load_fpn2() - load dst operand from FP regfile #
# fout() - emulate an opclass 3 instruction #
# tbl_unsupp - add of table of emulation routines for opclass 0,2 #
# _fpsp_done() - "callout" for 060FPSP exit (all work done!) #
# _real_ovfl() - "callout" for Overflow exception enabled code #
# _real_inex() - "callout" for Inexact exception enabled code #
# _real_trace() - "callout" for Trace exception code #
# #
# INPUT *************************************************************** #
# - The system stack contains the FP Ovfl exception stack frame #
# - The fsave frame contains the source operand #
# #
# OUTPUT ************************************************************** #
# Overflow Exception enabled: #
# - The system stack is unchanged #
# - The fsave frame contains the adjusted src op for opclass 0,2 #
# Overflow Exception disabled: #
# - The system stack is unchanged #
# - The "exception present" flag in the fsave frame is cleared #
# #
# ALGORITHM *********************************************************** #
# On the 060, if an FP overflow is present as the result of any #
# instruction, the 060 will take an overflow exception whether the #
# exception is enabled or disabled in the FPCR. For the disabled case, #
# This handler emulates the instruction to determine what the correct #
# default result should be for the operation. This default result is #
# then stored in either the FP regfile, data regfile, or memory. #
# Finally, the handler exits through the "callout" _fpsp_done() #
# denoting that no exceptional conditions exist within the machine. #
# If the exception is enabled, then this handler must create the #
# exceptional operand and plave it in the fsave state frame, and store #
# the default result (only if the instruction is opclass 3). For #
# exceptions enabled, this handler must exit through the "callout" #
# _real_ovfl() so that the operating system enabled overflow handler #
# can handle this case. #
# Two other conditions exist. First, if overflow was disabled #
# but the inexact exception was enabled, this handler must exit #
# through the "callout" _real_inex() regardless of whether the result #
# was inexact. #
# Also, in the case of an opclass three instruction where #
# overflow was disabled and the trace exception was enabled, this #
# handler must exit through the "callout" _real_trace(). #
# #
#########################################################################
global _fpsp_ovfl
_fpsp_ovfl:
#$# sub.l &24,%sp # make room for src/dst
link.w %a6,&-LOCAL_SIZE # init stack frame
fsave FP_SRC(%a6) # grab the "busy" frame
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs
fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack
# the FPIAR holds the "current PC" of the faulting instruction
mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6)
mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr
addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr
bsr.l _imem_read_long # fetch the instruction words
mov.l %d0,EXC_OPWORD(%a6)
btst &0x5,EXC_CMDREG(%a6) # is instr an fmove out?
bne.w fovfl_out
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l fix_skewed_ops # fix src op
# since, I believe, only NORMs and DENORMs can come through here,
# maybe we can avoid the subroutine call.
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l set_tag_x # tag the operand type
mov.b %d0,STAG(%a6) # maybe NORM,DENORM
# bit five of the fp extension word separates the monadic and dyadic operations
# that can pass through fpsp_ovfl(). remember that fcmp, ftst, and fsincos
# will never take this exception.
btst &0x5,1+EXC_CMDREG(%a6) # is operation monadic or dyadic?
beq.b fovfl_extract # monadic
lea FP_DST(%a6),%a0 # pass: ptr to dst op
bsr.l set_tag_x # tag the operand type
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b fovfl_op2_done # no
bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO
fovfl_op2_done:
mov.b %d0,DTAG(%a6) # save dst optype tag
andi.l &0x00ff01ff,USER_FPSR(%a6) # zero all but accured field
fmov.l &0x0,%fpcr # zero current control regs
fmov.l &0x0,%fpsr
lea FP_SRC(%a6),%a0
lea FP_DST(%a6),%a1
# maybe we can make these entry points ONLY the OVFL entry points of each routine.
mov.l (tbl_unsupp.l,%pc,%d1.w*4),%d1 # fetch routine addr
jsr (tbl_unsupp.l,%pc,%d1.l*1)
# the operation has been emulated. the result is in fp0.
# the EXOP, if an exception occurred, is in fp1.
# we must save the default result regardless of whether
# traps are enabled or disabled.
bfextu EXC_CMDREG(%a6){&6:&3},%d0
bsr.l store_fpreg
# the exceptional possibilities we have left ourselves with are ONLY overflow
# and inexact. and, the inexact is such that overflow occurred and was disabled
# but inexact was enabled.
btst &ovfl_bit,FPCR_ENABLE(%a6)
bne.b fovfl_ovfl_on
# overflow is enabled AND overflow, of course, occurred. so, we have the EXOP
# in fp1. now, simply jump to _real_ovfl()!
fovfl_ovfl_on:
fmovm.x &0x40,FP_SRC(%a6) # save EXOP (fp1) to stack
btst &0x7,(%sp) # is trace on?
beq.l _fpsp_done # no
fmov.l %fpiar,0x8(%sp) # "Current PC" is in FPIAR
mov.w &0x2024,0x6(%sp) # stk fmt = 0x2; voff = 0x024
bra.l _real_trace
#########################################################################
# XDEF **************************************************************** #
# _fpsp_unfl(): 060FPSP entry point for FP Underflow exception. #
# #
# This handler should be the first code executed upon taking the #
# FP Underflow exception in an operating system. #
# #
# XREF **************************************************************** #
# _imem_read_long() - read instruction longword #
# fix_skewed_ops() - adjust src operand in fsave frame #
# set_tag_x() - determine optype of src/dst operands #
# store_fpreg() - store opclass 0 or 2 result to FP regfile #
# unnorm_fix() - change UNNORM operands to NORM or ZERO #
# load_fpn2() - load dst operand from FP regfile #
# fout() - emulate an opclass 3 instruction #
# tbl_unsupp - add of table of emulation routines for opclass 0,2 #
# _fpsp_done() - "callout" for 060FPSP exit (all work done!) #
# _real_ovfl() - "callout" for Overflow exception enabled code #
# _real_inex() - "callout" for Inexact exception enabled code #
# _real_trace() - "callout" for Trace exception code #
# #
# INPUT *************************************************************** #
# - The system stack contains the FP Unfl exception stack frame #
# - The fsave frame contains the source operand #
# #
# OUTPUT ************************************************************** #
# Underflow Exception enabled: #
# - The system stack is unchanged #
# - The fsave frame contains the adjusted src op for opclass 0,2 #
# Underflow Exception disabled: #
# - The system stack is unchanged #
# - The "exception present" flag in the fsave frame is cleared #
# #
# ALGORITHM *********************************************************** #
# On the 060, if an FP underflow is present as the result of any #
# instruction, the 060 will take an underflow exception whether the #
# exception is enabled or disabled in the FPCR. For the disabled case, #
# This handler emulates the instruction to determine what the correct #
# default result should be for the operation. This default result is #
# then stored in either the FP regfile, data regfile, or memory. #
# Finally, the handler exits through the "callout" _fpsp_done() #
# denoting that no exceptional conditions exist within the machine. #
# If the exception is enabled, then this handler must create the #
# exceptional operand and plave it in the fsave state frame, and store #
# the default result (only if the instruction is opclass 3). For #
# exceptions enabled, this handler must exit through the "callout" #
# _real_unfl() so that the operating system enabled overflow handler #
# can handle this case. #
# Two other conditions exist. First, if underflow was disabled #
# but the inexact exception was enabled and the result was inexact, #
# this handler must exit through the "callout" _real_inex(). #
# was inexact. #
# Also, in the case of an opclass three instruction where #
# underflow was disabled and the trace exception was enabled, this #
# handler must exit through the "callout" _real_trace(). #
# #
#########################################################################
global _fpsp_unfl
_fpsp_unfl:
#$# sub.l &24,%sp # make room for src/dst
link.w %a6,&-LOCAL_SIZE # init stack frame
fsave FP_SRC(%a6) # grab the "busy" frame
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs
fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack
# the FPIAR holds the "current PC" of the faulting instruction
mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6)
mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr
addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr
bsr.l _imem_read_long # fetch the instruction words
mov.l %d0,EXC_OPWORD(%a6)
btst &0x5,EXC_CMDREG(%a6) # is instr an fmove out?
bne.w funfl_out
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l fix_skewed_ops # fix src op
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l set_tag_x # tag the operand type
mov.b %d0,STAG(%a6) # maybe NORM,DENORM
# bit five of the fp ext word separates the monadic and dyadic operations
# that can pass through fpsp_unfl(). remember that fcmp, and ftst
# will never take this exception.
btst &0x5,1+EXC_CMDREG(%a6) # is op monadic or dyadic?
beq.b funfl_extract # monadic
# now, what's left that's not dyadic is fsincos. we can distinguish it
# from all dyadics by the '0110xxx pattern
btst &0x4,1+EXC_CMDREG(%a6) # is op an fsincos?
bne.b funfl_extract # yes
lea FP_DST(%a6),%a0 # pass: ptr to dst op
bsr.l set_tag_x # tag the operand type
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b funfl_op2_done # no
bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO
funfl_op2_done:
mov.b %d0,DTAG(%a6) # save dst optype tag
fmov.l &0x0,%fpcr # zero current control regs
fmov.l &0x0,%fpsr
lea FP_SRC(%a6),%a0
lea FP_DST(%a6),%a1
# maybe we can make these entry points ONLY the OVFL entry points of each routine.
mov.l (tbl_unsupp.l,%pc,%d1.w*4),%d1 # fetch routine addr
jsr (tbl_unsupp.l,%pc,%d1.l*1)
# The `060 FPU multiplier hardware is such that if the result of a
# multiply operation is the smallest possible normalized number
# (0x00000000_80000000_00000000), then the machine will take an
# underflow exception. Since this is incorrect, we need to check
# if our emulation, after re-doing the operation, decided that
# no underflow was called for. We do these checks only in
# funfl_{unfl,inex}_on() because w/ both exceptions disabled, this
# special case will simply exit gracefully with the correct result.
# the exceptional possibilities we have left ourselves with are ONLY overflow
# and inexact. and, the inexact is such that overflow occurred and was disabled
# but inexact was enabled.
btst &unfl_bit,FPCR_ENABLE(%a6)
bne.b funfl_unfl_on
# overflow is enabled AND overflow, of course, occurred. so, we have the EXOP
# in fp1 (don't forget to save fp0). what to do now?
# well, we simply have to get to go to _real_unfl()!
funfl_unfl_on:
# The `060 FPU multiplier hardware is such that if the result of a
# multiply operation is the smallest possible normalized number
# (0x00000000_80000000_00000000), then the machine will take an
# underflow exception. Since this is incorrect, we check here to see
# if our emulation, after re-doing the operation, decided that
# no underflow was called for.
btst &unfl_bit,FPSR_EXCEPT(%a6)
beq.w funfl_chkinex
funfl_unfl_on2:
fmovm.x &0x40,FP_SRC(%a6) # save EXOP (fp1) to stack
frestore FP_SRC(%a6) # do this after fmovm,other f<op>s!
unlk %a6
bra.l _real_unfl
# underflow occurred but is disabled. meanwhile, inexact is enabled. Therefore,
# we must jump to real_inex().
funfl_inex_on:
# The `060 FPU multiplier hardware is such that if the result of a
# multiply operation is the smallest possible normalized number
# (0x00000000_80000000_00000000), then the machine will take an
# underflow exception.
# But, whether bogus or not, if inexact is enabled AND it occurred,
# then we have to branch to real_inex.
btst &inex2_bit,FPSR_EXCEPT(%a6)
beq.w funfl_exit
funfl_inex_on2:
fmovm.x &0x40,FP_SRC(%a6) # save EXOP to stack
mov.b &0xc4,1+EXC_VOFF(%a6) # vector offset = 0xc4
mov.w &0xe001,2+FP_SRC(%a6) # save exc status
btst &0x7,(%sp) # is trace on?
beq.l _fpsp_done # no
fmov.l %fpiar,0x8(%sp) # "Current PC" is in FPIAR
mov.w &0x2024,0x6(%sp) # stk fmt = 0x2; voff = 0x024
bra.l _real_trace
#########################################################################
# XDEF **************************************************************** #
# _fpsp_unsupp(): 060FPSP entry point for FP "Unimplemented #
# Data Type" exception. #
# #
# This handler should be the first code executed upon taking the #
# FP Unimplemented Data Type exception in an operating system. #
# #
# XREF **************************************************************** #
# _imem_read_{word,long}() - read instruction word/longword #
# fix_skewed_ops() - adjust src operand in fsave frame #
# set_tag_x() - determine optype of src/dst operands #
# store_fpreg() - store opclass 0 or 2 result to FP regfile #
# unnorm_fix() - change UNNORM operands to NORM or ZERO #
# load_fpn2() - load dst operand from FP regfile #
# load_fpn1() - load src operand from FP regfile #
# fout() - emulate an opclass 3 instruction #
# tbl_unsupp - add of table of emulation routines for opclass 0,2 #
# _real_inex() - "callout" to operating system inexact handler #
# _fpsp_done() - "callout" for exit; work all done #
# _real_trace() - "callout" for Trace enabled exception #
# funimp_skew() - adjust fsave src ops to "incorrect" value #
# _real_snan() - "callout" for SNAN exception #
# _real_operr() - "callout" for OPERR exception #
# _real_ovfl() - "callout" for OVFL exception #
# _real_unfl() - "callout" for UNFL exception #
# get_packed() - fetch packed operand from memory #
# #
# INPUT *************************************************************** #
# - The system stack contains the "Unimp Data Type" stk frame #
# - The fsave frame contains the ssrc op (for UNNORM/DENORM) #
# #
# OUTPUT ************************************************************** #
# If Inexact exception (opclass 3): #
# - The system stack is changed to an Inexact exception stk frame #
# If SNAN exception (opclass 3): #
# - The system stack is changed to an SNAN exception stk frame #
# If OPERR exception (opclass 3): #
# - The system stack is changed to an OPERR exception stk frame #
# If OVFL exception (opclass 3): #
# - The system stack is changed to an OVFL exception stk frame #
# If UNFL exception (opclass 3): #
# - The system stack is changed to an UNFL exception stack frame #
# If Trace exception enabled: #
# - The system stack is changed to a Trace exception stack frame #
# Else: (normal case) #
# - Correct result has been stored as appropriate #
# #
# ALGORITHM *********************************************************** #
# Two main instruction types can enter here: (1) DENORM or UNNORM #
# unimplemented data types. These can be either opclass 0,2 or 3 #
# instructions, and (2) PACKED unimplemented data format instructions #
# also of opclasses 0,2, or 3. #
# For UNNORM/DENORM opclass 0 and 2, the handler fetches the src #
# operand from the fsave state frame and the dst operand (if dyadic) #
# from the FP register file. The instruction is then emulated by #
# choosing an emulation routine from a table of routines indexed by #
# instruction type. Once the instruction has been emulated and result #
# saved, then we check to see if any enabled exceptions resulted from #
# instruction emulation. If none, then we exit through the "callout" #
# _fpsp_done(). If there is an enabled FP exception, then we insert #
# this exception into the FPU in the fsave state frame and then exit #
# through _fpsp_done(). #
# PACKED opclass 0 and 2 is similar in how the instruction is #
# emulated and exceptions handled. The differences occur in how the #
# handler loads the packed op (by calling get_packed() routine) and #
# by the fact that a Trace exception could be pending for PACKED ops. #
# If a Trace exception is pending, then the current exception stack #
# frame is changed to a Trace exception stack frame and an exit is #
# made through _real_trace(). #
# For UNNORM/DENORM opclass 3, the actual move out to memory is #
# performed by calling the routine fout(). If no exception should occur #
# as the result of emulation, then an exit either occurs through #
# _fpsp_done() or through _real_trace() if a Trace exception is pending #
# (a Trace stack frame must be created here, too). If an FP exception #
# should occur, then we must create an exception stack frame of that #
# type and jump to either _real_snan(), _real_operr(), _real_inex(), #
# _real_unfl(), or _real_ovfl() as appropriate. PACKED opclass 3 #
# emulation is performed in a similar manner. #
# #
#########################################################################
#
# (1) DENORM and UNNORM (unimplemented) data types:
#
# post-instruction
# *****************
# * EA *
# pre-instruction * *
# ***************** *****************
# * 0x0 * 0x0dc * * 0x3 * 0x0dc *
# ***************** *****************
# * Next * * Next *
# * PC * * PC *
# ***************** *****************
# * SR * * SR *
# ***************** *****************
#
# (2) PACKED format (unsupported) opclasses two and three:
# *****************
# * EA *
# * *
# *****************
# * 0x2 * 0x0dc *
# *****************
# * Next *
# * PC *
# *****************
# * SR *
# *****************
# global _fpsp_unsupp
_fpsp_unsupp:
link.w %a6,&-LOCAL_SIZE # init stack frame
fsave FP_SRC(%a6) # save fp state
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs
fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack
btst &0x5,EXC_SR(%a6) # user or supervisor mode?
bne.b fu_s
fu_u:
mov.l %usp,%a0 # fetch user stack pointer
mov.l %a0,EXC_A7(%a6) # save on stack
bra.b fu_cont
# if the exception is an opclass zero or two unimplemented data type
# exception, then the a7' calculated here is wrong since it doesn't
# stack an ea. however, we don't need an a7' for this case anyways.
fu_s:
lea 0x4+EXC_EA(%a6),%a0 # load old a7'
mov.l %a0,EXC_A7(%a6) # save on stack
fu_cont:
# the FPIAR holds the "current PC" of the faulting instruction
# the FPIAR should be set correctly for ALL exceptions passing through
# this point.
mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6)
mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr
addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr
bsr.l _imem_read_long # fetch the instruction words
mov.l %d0,EXC_OPWORD(%a6) # store OPWORD and EXTWORD
############################
clr.b SPCOND_FLG(%a6) # clear special condition flag
# Separate opclass three (fpn-to-mem) ops since they have a different
# stack frame and protocol.
btst &0x5,EXC_CMDREG(%a6) # is it an fmove out?
bne.w fu_out # yes
# Separate packed opclass two instructions.
bfextu EXC_CMDREG(%a6){&0:&6},%d0
cmpi.b %d0,&0x13
beq.w fu_in_pack
# I'm not sure at this point what FPSR bits are valid for this instruction.
# so, since the emulation routines re-create them anyways, zero exception field
andi.l &0x00ff00ff,USER_FPSR(%a6) # zero exception field
fmov.l &0x0,%fpcr # zero current control regs
fmov.l &0x0,%fpsr
# Opclass two w/ memory-to-fpn operation will have an incorrect extended
# precision format if the src format was single or double and the
# source data type was an INF, NAN, DENORM, or UNNORM
lea FP_SRC(%a6),%a0 # pass ptr to input
bsr.l fix_skewed_ops
# we don't know whether the src operand or the dst operand (or both) is the
# UNNORM or DENORM. call the function that tags the operand type. if the
# input is an UNNORM, then convert it to a NORM, DENORM, or ZERO.
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l set_tag_x # tag the operand type
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b fu_op2 # no
bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO
# bit five of the fp extension word separates the monadic and dyadic operations
# at this point
btst &0x5,1+EXC_CMDREG(%a6) # is operation monadic or dyadic?
beq.b fu_extract # monadic
cmpi.b 1+EXC_CMDREG(%a6),&0x3a # is operation an ftst?
beq.b fu_extract # yes, so it's monadic, too
bsr.l load_fpn2 # load dst into FP_DST
lea FP_DST(%a6),%a0 # pass: ptr to dst op
bsr.l set_tag_x # tag the operand type
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b fu_op2_done # no
bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO
fu_op2_done:
mov.b %d0,DTAG(%a6) # save dst optype tag
#
# Exceptions in order of precedence:
# BSUN : none
# SNAN : all dyadic ops
# OPERR : fsqrt(-NORM)
# OVFL : all except ftst,fcmp
# UNFL : all except ftst,fcmp
# DZ : fdiv
# INEX2 : all except ftst,fcmp
# INEX1 : none (packed doesn't go through here)
#
# we determine the highest priority exception(if any) set by the
# emulation routine that has also been enabled by the user.
mov.b FPCR_ENABLE(%a6),%d0 # fetch exceptions set
bne.b fu_in_ena # some are enabled
fu_in_cont:
# fcmp and ftst do not store any result.
mov.b 1+EXC_CMDREG(%a6),%d0 # fetch extension
andi.b &0x38,%d0 # extract bits 3-5
cmpi.b %d0,&0x38 # is instr fcmp or ftst?
beq.b fu_in_exit # yes
bfextu EXC_CMDREG(%a6){&6:&3},%d0 # dyadic; load dst reg
bsr.l store_fpreg # store the result
fu_in_ena:
and.b FPSR_EXCEPT(%a6),%d0 # keep only ones enabled
bfffo %d0{&24:&8},%d0 # find highest priority exception
bne.b fu_in_exc # there is at least one set
#
# No exceptions occurred that were also enabled. Now:
#
# if (OVFL && ovfl_disabled && inexact_enabled) {
# branch to _real_inex() (even if the result was exact!);
# } else {
# save the result in the proper fp reg (unless the op is fcmp or ftst);
# return;
# }
#
btst &ovfl_bit,FPSR_EXCEPT(%a6) # was overflow set?
beq.b fu_in_cont # no
fu_in_ovflchk:
btst &inex2_bit,FPCR_ENABLE(%a6) # was inexact enabled?
beq.b fu_in_cont # no
bra.w fu_in_exc_ovfl # go insert overflow frame
#
# An exception occurred and that exception was enabled:
#
# shift enabled exception field into lo byte of d0;
# if (((INEX2 || INEX1) && inex_enabled && OVFL && ovfl_disabled) ||
# ((INEX2 || INEX1) && inex_enabled && UNFL && unfl_disabled)) {
# /* # * this is the case where we must call _real_inex() now or else # * there will be no other way to pass it the exceptional operand
# */
# call _real_inex();
# } else {
# restore exc state (SNAN||OPERR||OVFL||UNFL||DZ||INEX) into the FPU;
# }
#
fu_in_exc:
subi.l &24,%d0 # fix offset to be 0-8
cmpi.b %d0,&0x6 # is exception INEX? (6)
bne.b fu_in_exc_exit # no
# the enabled exception was inexact
btst &unfl_bit,FPSR_EXCEPT(%a6) # did disabled underflow occur?
bne.w fu_in_exc_unfl # yes
btst &ovfl_bit,FPSR_EXCEPT(%a6) # did disabled overflow occur?
bne.w fu_in_exc_ovfl # yes
# here, we insert the correct fsave status value into the fsave frame for the
# corresponding exception. the operand in the fsave frame should be the original
# src operand.
fu_in_exc_exit:
mov.l %d0,-(%sp) # save d0
bsr.l funimp_skew # skew sgl or dbl inputs
mov.l (%sp)+,%d0 # restore d0
mov.w (tbl_except.b,%pc,%d0.w*2),2+FP_SRC(%a6) # create exc status
# If the input operand to this operation was opclass two and a single
# or double precision denorm, inf, or nan, the operand needs to be
# "corrected" in order to have the proper equivalent extended precision
# number. global fix_skewed_ops
fix_skewed_ops:
bfextu EXC_CMDREG(%a6){&0:&6},%d0 # extract opclass,src fmt
cmpi.b %d0,&0x11 # is class = 2 & fmt = sgl?
beq.b fso_sgl # yes
cmpi.b %d0,&0x15 # is class = 2 & fmt = dbl?
beq.b fso_dbl # yes
rts # no
fso_sgl_dnrm_zero:
andi.l &0x7fffffff,LOCAL_HI(%a0) # clear j-bit
beq.b fso_zero # it's a skewed zero
fso_sgl_dnrm:
# here, we count on norm not to alter a0...
bsr.l norm # normalize mantissa
neg.w %d0 # -shft amt
addi.w &0x3f81,%d0 # adjust new exponent
andi.w &0x8000,LOCAL_EX(%a0) # clear old exponent
or.w %d0,LOCAL_EX(%a0) # insert new exponent
rts
fso_dbl_dnrm_zero:
andi.l &0x7fffffff,LOCAL_HI(%a0) # clear j-bit
bne.b fso_dbl_dnrm # it's a skewed denorm
tst.l LOCAL_LO(%a0) # is it a zero?
beq.b fso_zero # yes
fso_dbl_dnrm:
# here, we count on norm not to alter a0...
bsr.l norm # normalize mantissa
neg.w %d0 # -shft amt
addi.w &0x3c01,%d0 # adjust new exponent
andi.w &0x8000,LOCAL_EX(%a0) # clear old exponent
or.w %d0,LOCAL_EX(%a0) # insert new exponent
rts
# fmove out took an unimplemented data type exception.
# the src operand is in FP_SRC. Call _fout() to write out the result and
# to determine which exceptions, if any, to take.
fu_out:
# Separate packed move outs from the UNNORM and DENORM move outs.
bfextu EXC_CMDREG(%a6){&3:&3},%d0
cmpi.b %d0,&0x3
beq.w fu_out_pack
cmpi.b %d0,&0x7
beq.w fu_out_pack
# I'm not sure at this point what FPSR bits are valid for this instruction.
# so, since the emulation routines re-create them anyways, zero exception field.
# fmove out doesn't affect ccodes.
and.l &0xffff00ff,USER_FPSR(%a6) # zero exception field
fmov.l &0x0,%fpcr # zero current control regs
fmov.l &0x0,%fpsr
# the src can ONLY be a DENORM or an UNNORM! so, don't make any big subroutine
# call here. just figure out what it is...
mov.w FP_SRC_EX(%a6),%d0 # get exponent
andi.w &0x7fff,%d0 # strip sign
beq.b fu_out_denorm # it's a DENORM
lea FP_SRC(%a6),%a0
bsr.l unnorm_fix # yes; fix it
mov.l (%a6),EXC_A6(%a6) # in case a6 changes
bsr.l fout # call fmove out routine
# Exceptions in order of precedence:
# BSUN : none
# SNAN : none
# OPERR : fmove.{b,w,l} out of large UNNORM
# OVFL : fmove.{s,d}
# UNFL : fmove.{s,d,x}
# DZ : none
# INEX2 : all
# INEX1 : none (packed doesn't travel through here)
# determine the highest priority exception(if any) set by the
# emulation routine that has also been enabled by the user.
mov.b FPCR_ENABLE(%a6),%d0 # fetch exceptions enabled
bne.w fu_out_ena # some are enabled
fu_out_done:
mov.l EXC_A6(%a6),(%a6) # in case a6 changed
# on extended precision opclass three instructions using pre-decrement or
# post-increment addressing mode, the address register is not updated. is the
# address register was the stack pointer used from user mode, then let's update
# it here. if it was used from supervisor mode, then we have to handle this
# as a special case.
btst &0x5,EXC_SR(%a6)
bne.b fu_out_done_s
btst &0x7,(%sp) # is trace on?
bne.b fu_out_trace # yes
bra.l _fpsp_done
# is the ea mode pre-decrement of the stack pointer from supervisor mode?
# ("fmov.x fpm,-(a7)") if so,
fu_out_done_s:
cmpi.b SPCOND_FLG(%a6),&mda7_flg
bne.b fu_out_done_cont
# the extended precision result is still in fp0. but, we need to save it
# somewhere on the stack until we can copy it to its final resting place.
# here, we're counting on the top of the stack to be the old place-holders
# for fp0/fp1 which have already been restored. that way, we can write
# over those destinations with the shifted stack frame.
fmovm.x &0x80,FP_SRC(%a6) # put answer on stack
# now, copy the result to the proper place on the stack
mov.l LOCAL_SIZE+FP_SRC_EX(%sp),LOCAL_SIZE+EXC_SR+0x0(%sp)
mov.l LOCAL_SIZE+FP_SRC_HI(%sp),LOCAL_SIZE+EXC_SR+0x4(%sp)
mov.l LOCAL_SIZE+FP_SRC_LO(%sp),LOCAL_SIZE+EXC_SR+0x8(%sp)
add.l &LOCAL_SIZE-0x8,%sp
btst &0x7,(%sp)
bne.b fu_out_trace
bra.l _fpsp_done
fu_out_ena:
and.b FPSR_EXCEPT(%a6),%d0 # keep only ones enabled
bfffo %d0{&24:&8},%d0 # find highest priority exception
bne.b fu_out_exc # there is at least one set
# no exceptions were set.
# if a disabled overflow occurred and inexact was enabled but the result
# was exact, then a branch to _real_inex() is made.
btst &ovfl_bit,FPSR_EXCEPT(%a6) # was overflow set?
beq.w fu_out_done # no
fu_out_ovflchk:
btst &inex2_bit,FPCR_ENABLE(%a6) # was inexact enabled?
beq.w fu_out_done # no
bra.w fu_inex # yes
#
# The fp move out that took the "Unimplemented Data Type" exception was
# being traced. Since the stack frames are similar, get the "current" PC
# from FPIAR and put it in the trace stack frame then jump to _real_trace().
#
# UNSUPP FRAME TRACE FRAME
# ***************** *****************
# * EA * * Current *
# * * * PC *
# ***************** *****************
# * 0x3 * 0x0dc * * 0x2 * 0x024 *
# ***************** *****************
# * Next * * Next *
# * PC * * PC *
# ***************** *****************
# * SR * * SR *
# ***************** *****************
#
fu_out_trace:
mov.w &0x2024,0x6(%sp)
fmov.l %fpiar,0x8(%sp)
bra.l _real_trace
# an exception occurred and that exception was enabled.
fu_out_exc:
subi.l &24,%d0 # fix offset to be 0-8
# we don't mess with the existing fsave frame. just re-insert it and
# jump to the "_real_{}()" handler...
mov.w (tbl_fu_out.b,%pc,%d0.w*2),%d0
jmp (tbl_fu_out.b,%pc,%d0.w*1)
swbeg &0x8
tbl_fu_out:
short tbl_fu_out - tbl_fu_out # BSUN can't happen
short tbl_fu_out - tbl_fu_out # SNAN can't happen
short fu_operr - tbl_fu_out # OPERR
short fu_ovfl - tbl_fu_out # OVFL
short fu_unfl - tbl_fu_out # UNFL
short tbl_fu_out - tbl_fu_out # DZ can't happen
short fu_inex - tbl_fu_out # INEX2
short tbl_fu_out - tbl_fu_out # INEX1 won't make it here
# for snan,operr,ovfl,unfl, src op is still in FP_SRC so just
# frestore it.
fu_snan:
fmovm.x EXC_FPREGS(%a6),&0xc0 # restore fp0/fp1
fmovm.l USER_FPCR(%a6),%fpcr,%fpsr,%fpiar # restore ctrl regs
movm.l EXC_DREGS(%a6),&0x0303 # restore d0-d1/a0-a1
# underflow can happen for extended precision. extended precision opclass
# three instruction exceptions don't update the stack pointer. so, if the
# exception occurred from user mode, then simply update a7 and exit normally.
# if the exception occurred from supervisor mode, check if
fu_unfl:
mov.l EXC_A6(%a6),(%a6) # restore a6
btst &0x5,EXC_SR(%a6)
bne.w fu_unfl_s
mov.l EXC_A7(%a6),%a0 # restore a7 whether we need
mov.l %a0,%usp # to or not...
fu_unfl_cont:
fmovm.x &0x40,FP_SRC(%a6) # save EXOP to the stack
fu_unfl_s:
cmpi.b SPCOND_FLG(%a6),&mda7_flg # was the <ea> mode -(sp)?
bne.b fu_unfl_cont
# the extended precision result is still in fp0. but, we need to save it
# somewhere on the stack until we can copy it to its final resting place
# (where the exc frame is currently). make sure it's not at the top of the
# frame or it will get overwritten when the exc stack frame is shifted "down".
fmovm.x &0x80,FP_SRC(%a6) # put answer on stack
fmovm.x &0x40,FP_DST(%a6) # put EXOP on stack
# now, copy the result to the proper place on the stack
mov.l LOCAL_SIZE+FP_SRC_EX(%sp),LOCAL_SIZE+EXC_SR+0x0(%sp)
mov.l LOCAL_SIZE+FP_SRC_HI(%sp),LOCAL_SIZE+EXC_SR+0x4(%sp)
mov.l LOCAL_SIZE+FP_SRC_LO(%sp),LOCAL_SIZE+EXC_SR+0x8(%sp)
add.l &LOCAL_SIZE-0x8,%sp
bra.l _real_unfl
# fmove in and out enter here.
fu_inex:
fmovm.x &0x40,FP_SRC(%a6) # save EXOP to the stack
# I'm not sure at this point what FPSR bits are valid for this instruction.
# so, since the emulation routines re-create them anyways, zero exception field
andi.l &0x0ff00ff,USER_FPSR(%a6) # zero exception field
fmov.l &0x0,%fpcr # zero current control regs
fmov.l &0x0,%fpsr
bsr.l get_packed # fetch packed src operand
lea FP_SRC(%a6),%a0 # pass ptr to src
bsr.l set_tag_x # set src optype tag
# bit five of the fp extension word separates the monadic and dyadic operations
# at this point
btst &0x5,1+EXC_CMDREG(%a6) # is operation monadic or dyadic?
beq.b fu_extract_p # monadic
cmpi.b 1+EXC_CMDREG(%a6),&0x3a # is operation an ftst?
beq.b fu_extract_p # yes, so it's monadic, too
bsr.l load_fpn2 # load dst into FP_DST
lea FP_DST(%a6),%a0 # pass: ptr to dst op
bsr.l set_tag_x # tag the operand type
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b fu_op2_done_p # no
bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO
fu_op2_done_p:
mov.b %d0,DTAG(%a6) # save dst optype tag
#
# Exceptions in order of precedence:
# BSUN : none
# SNAN : all dyadic ops
# OPERR : fsqrt(-NORM)
# OVFL : all except ftst,fcmp
# UNFL : all except ftst,fcmp
# DZ : fdiv
# INEX2 : all except ftst,fcmp
# INEX1 : all
#
# we determine the highest priority exception(if any) set by the
# emulation routine that has also been enabled by the user.
mov.b FPCR_ENABLE(%a6),%d0 # fetch exceptions enabled
bne.w fu_in_ena_p # some are enabled
fu_in_cont_p:
# fcmp and ftst do not store any result.
mov.b 1+EXC_CMDREG(%a6),%d0 # fetch extension
andi.b &0x38,%d0 # extract bits 3-5
cmpi.b %d0,&0x38 # is instr fcmp or ftst?
beq.b fu_in_exit_p # yes
bfextu EXC_CMDREG(%a6){&6:&3},%d0 # dyadic; load dst reg
bsr.l store_fpreg # store the result
fu_in_exit_p:
btst &0x5,EXC_SR(%a6) # user or supervisor?
bne.w fu_in_exit_s_p # supervisor
mov.l EXC_A7(%a6),%a0 # update user a7
mov.l %a0,%usp
btst &0x7,(%sp) # is trace on?
bne.w fu_trace_p # yes
bra.l _fpsp_done # exit to os
# the exception occurred in supervisor mode. check to see if the
# addressing mode was (a7)+. if so, we'll need to shift the
# stack frame "up".
fu_in_exit_s_p:
btst &mia7_bit,SPCOND_FLG(%a6) # was ea mode (a7)+
beq.b fu_in_exit_cont_p # no
# shift the stack frame "up". we don't really care about the <ea> field.
mov.l 0x4(%sp),0x10(%sp)
mov.l 0x0(%sp),0xc(%sp)
add.l &0xc,%sp
btst &0x7,(%sp) # is trace on?
bne.w fu_trace_p # yes
bra.l _fpsp_done # exit to os
fu_in_ena_p:
and.b FPSR_EXCEPT(%a6),%d0 # keep only ones enabled & set
bfffo %d0{&24:&8},%d0 # find highest priority exception
bne.b fu_in_exc_p # at least one was set
#
# No exceptions occurred that were also enabled. Now:
#
# if (OVFL && ovfl_disabled && inexact_enabled) {
# branch to _real_inex() (even if the result was exact!);
# } else {
# save the result in the proper fp reg (unless the op is fcmp or ftst);
# return;
# }
#
btst &ovfl_bit,FPSR_EXCEPT(%a6) # was overflow set?
beq.w fu_in_cont_p # no
fu_in_ovflchk_p:
btst &inex2_bit,FPCR_ENABLE(%a6) # was inexact enabled?
beq.w fu_in_cont_p # no
bra.w fu_in_exc_ovfl_p # do _real_inex() now
#
# An exception occurred and that exception was enabled:
#
# shift enabled exception field into lo byte of d0;
# if (((INEX2 || INEX1) && inex_enabled && OVFL && ovfl_disabled) ||
# ((INEX2 || INEX1) && inex_enabled && UNFL && unfl_disabled)) {
# /* # * this is the case where we must call _real_inex() now or else # * there will be no other way to pass it the exceptional operand
# */
# call _real_inex();
# } else {
# restore exc state (SNAN||OPERR||OVFL||UNFL||DZ||INEX) into the FPU;
# }
#
fu_in_exc_p:
subi.l &24,%d0 # fix offset to be 0-8
cmpi.b %d0,&0x6 # is exception INEX? (6 or 7)
blt.b fu_in_exc_exit_p # no
# the enabled exception was inexact
btst &unfl_bit,FPSR_EXCEPT(%a6) # did disabled underflow occur?
bne.w fu_in_exc_unfl_p # yes
btst &ovfl_bit,FPSR_EXCEPT(%a6) # did disabled overflow occur?
bne.w fu_in_exc_ovfl_p # yes
# here, we insert the correct fsave status value into the fsave frame for the
# corresponding exception. the operand in the fsave frame should be the original
# src operand.
# as a reminder for future predicted pain and agony, we are passing in fsave the
# "non-skewed" operand for cases of sgl and dbl src INFs,NANs, and DENORMs.
# this is INCORRECT for enabled SNAN which would give to the user the skewed SNAN!!!
fu_in_exc_exit_p:
btst &0x5,EXC_SR(%a6) # user or supervisor?
bne.w fu_in_exc_exit_s_p # supervisor
mov.l EXC_A7(%a6),%a0 # update user a7
mov.l %a0,%usp
# shift stack frame "up". who cares about <ea> field.
mov.l 0x4(%sp),0x10(%sp)
mov.l 0x0(%sp),0xc(%sp)
add.l &0xc,%sp
btst &0x7,(%sp) # is trace on?
bne.b fu_trace_p # yes
bra.l _fpsp_done # exit to os
#
# The opclass two PACKED instruction that took an "Unimplemented Data Type"
# exception was being traced. Make the "current" PC the FPIAR and put it in the
# trace stack frame then jump to _real_trace().
#
# UNSUPP FRAME TRACE FRAME
# ***************** *****************
# * EA * * Current *
# * * * PC *
# ***************** *****************
# * 0x2 * 0x0dc * * 0x2 * 0x024 *
# ***************** *****************
# * Next * * Next *
# * PC * * PC *
# ***************** *****************
# * SR * * SR *
# ***************** *****************
fu_trace_p:
mov.w &0x2024,0x6(%sp)
fmov.l %fpiar,0x8(%sp)
# I'm not sure at this point what FPSR bits are valid for this instruction.
# so, since the emulation routines re-create them anyways, zero exception field.
# fmove out doesn't affect ccodes.
and.l &0xffff00ff,USER_FPSR(%a6) # zero exception field
fmov.l &0x0,%fpcr # zero current control regs
fmov.l &0x0,%fpsr
bfextu EXC_CMDREG(%a6){&6:&3},%d0
bsr.l load_fpn1
# unlike other opclass 3, unimplemented data type exceptions, packed must be
# able to detect all operand types.
lea FP_SRC(%a6),%a0
bsr.l set_tag_x # tag the operand type
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b fu_op2_p # no
bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO
fu_op2_p:
mov.b %d0,STAG(%a6) # save src optype tag
mov.l (%a6),EXC_A6(%a6) # in case a6 changes
bsr.l fout # call fmove out routine
# Exceptions in order of precedence:
# BSUN : no
# SNAN : yes
# OPERR : if ((k_factor > +17) || (dec. exp exceeds 3 digits))
# OVFL : no
# UNFL : no
# DZ : no
# INEX2 : yes
# INEX1 : no
# determine the highest priority exception(if any) set by the
# emulation routine that has also been enabled by the user.
mov.b FPCR_ENABLE(%a6),%d0 # fetch exceptions enabled
bne.w fu_out_ena_p # some are enabled
btst &0x7,(%sp) # is trace on?
bne.w fu_trace_p # yes
bra.l _fpsp_done # exit to os
# the exception occurred in supervisor mode. check to see if the
# addressing mode was -(a7). if so, we'll need to shift the
# stack frame "down".
fu_out_exit_s_p:
btst &mda7_bit,SPCOND_FLG(%a6) # was ea mode -(a7)
beq.b fu_out_exit_cont_p # no
# now, copy the result to the proper place on the stack
mov.l LOCAL_SIZE+FP_DST_EX(%sp),LOCAL_SIZE+EXC_SR+0x0(%sp)
mov.l LOCAL_SIZE+FP_DST_HI(%sp),LOCAL_SIZE+EXC_SR+0x4(%sp)
mov.l LOCAL_SIZE+FP_DST_LO(%sp),LOCAL_SIZE+EXC_SR+0x8(%sp)
# an exception occurred and that exception was enabled.
# the only exception possible on packed move out are INEX, OPERR, and SNAN.
fu_out_exc_p:
cmpi.b %d0,&0x1a
bgt.w fu_inex_p2
beq.w fu_operr_p
# the instruction was "fmove.p fpn,-(a7)" from supervisor mode.
# the strategy is to move the exception frame "down" 12 bytes. then, we
# can store the default result where the exception frame was.
fmovm.x EXC_FPREGS(%a6),&0xc0 # restore fp0/fp1
fmovm.l USER_FPCR(%a6),%fpcr,%fpsr,%fpiar # restore ctrl regs
movm.l EXC_DREGS(%a6),&0x0303 # restore d0-d1/a0-a1
mov.w &0x30d8,EXC_VOFF(%a6) # vector offset = 0xd0
mov.w &0xe006,2+FP_SRC(%a6) # set fsave status
# now, we copy the default result to its proper location
mov.l LOCAL_SIZE+FP_DST_EX(%sp),LOCAL_SIZE+0x4(%sp)
mov.l LOCAL_SIZE+FP_DST_HI(%sp),LOCAL_SIZE+0x8(%sp)
mov.l LOCAL_SIZE+FP_DST_LO(%sp),LOCAL_SIZE+0xc(%sp)
# the instruction was "fmove.p fpn,-(a7)" from supervisor mode.
# the strategy is to move the exception frame "down" 12 bytes. then, we
# can store the default result where the exception frame was.
fmovm.x EXC_FPREGS(%a6),&0xc0 # restore fp0/fp1
fmovm.l USER_FPCR(%a6),%fpcr,%fpsr,%fpiar # restore ctrl regs
movm.l EXC_DREGS(%a6),&0x0303 # restore d0-d1/a0-a1
mov.w &0x30d0,EXC_VOFF(%a6) # vector offset = 0xd0
mov.w &0xe004,2+FP_SRC(%a6) # set fsave status
# now, we copy the default result to its proper location
mov.l LOCAL_SIZE+FP_DST_EX(%sp),LOCAL_SIZE+0x4(%sp)
mov.l LOCAL_SIZE+FP_DST_HI(%sp),LOCAL_SIZE+0x8(%sp)
mov.l LOCAL_SIZE+FP_DST_LO(%sp),LOCAL_SIZE+0xc(%sp)
# the instruction was "fmove.p fpn,-(a7)" from supervisor mode.
# the strategy is to move the exception frame "down" 12 bytes. then, we
# can store the default result where the exception frame was.
fmovm.x EXC_FPREGS(%a6),&0xc0 # restore fp0/fp1
fmovm.l USER_FPCR(%a6),%fpcr,%fpsr,%fpiar # restore ctrl regs
movm.l EXC_DREGS(%a6),&0x0303 # restore d0-d1/a0-a1
mov.w &0x30c4,EXC_VOFF(%a6) # vector offset = 0xc4
mov.w &0xe001,2+FP_SRC(%a6) # set fsave status
# now, we copy the default result to its proper location
mov.l LOCAL_SIZE+FP_DST_EX(%sp),LOCAL_SIZE+0x4(%sp)
mov.l LOCAL_SIZE+FP_DST_HI(%sp),LOCAL_SIZE+0x8(%sp)
mov.l LOCAL_SIZE+FP_DST_LO(%sp),LOCAL_SIZE+0xc(%sp)
#
# if we're stuffing a source operand back into an fsave frame then we
# have to make sure that for single or double source operands that the
# format stuffed is as weird as the hardware usually makes it.
# global funimp_skew
funimp_skew:
bfextu EXC_EXTWORD(%a6){&3:&3},%d0 # extract src specifier
cmpi.b %d0,&0x1 # was src sgl?
beq.b funimp_skew_sgl # yes
cmpi.b %d0,&0x5 # was src dbl?
beq.b funimp_skew_dbl # yes
rts
funimp_skew_sgl:
mov.w FP_SRC_EX(%a6),%d0 # fetch DENORM exponent
andi.w &0x7fff,%d0 # strip sign
beq.b funimp_skew_sgl_not
cmpi.w %d0,&0x3f80
bgt.b funimp_skew_sgl_not
neg.w %d0 # make exponent negative
addi.w &0x3f81,%d0 # find amt to shift
mov.l FP_SRC_HI(%a6),%d1 # fetch DENORM hi(man)
lsr.l %d0,%d1 # shift it
bset &31,%d1 # set j-bit
mov.l %d1,FP_SRC_HI(%a6) # insert new hi(man)
andi.w &0x8000,FP_SRC_EX(%a6) # clear old exponent
ori.w &0x3f80,FP_SRC_EX(%a6) # insert new "skewed" exponent
funimp_skew_sgl_not:
rts
#########################################################################
# XDEF **************************************************************** #
# _fpsp_effadd(): 060FPSP entry point for FP "Unimplemented #
# effective address" exception. #
# #
# This handler should be the first code executed upon taking the #
# FP Unimplemented Effective Address exception in an operating #
# system. #
# #
# XREF **************************************************************** #
# _imem_read_long() - read instruction longword #
# fix_skewed_ops() - adjust src operand in fsave frame #
# set_tag_x() - determine optype of src/dst operands #
# store_fpreg() - store opclass 0 or 2 result to FP regfile #
# unnorm_fix() - change UNNORM operands to NORM or ZERO #
# load_fpn2() - load dst operand from FP regfile #
# tbl_unsupp - add of table of emulation routines for opclass 0,2 #
# decbin() - convert packed data to FP binary data #
# _real_fpu_disabled() - "callout" for "FPU disabled" exception #
# _real_access() - "callout" for access error exception #
# _mem_read() - read extended immediate operand from memory #
# _fpsp_done() - "callout" for exit; work all done #
# _real_trace() - "callout" for Trace enabled exception #
# fmovm_dynamic() - emulate dynamic fmovm instruction #
# fmovm_ctrl() - emulate fmovm control instruction #
# #
# INPUT *************************************************************** #
# - The system stack contains the "Unimplemented <ea>" stk frame #
# #
# OUTPUT ************************************************************** #
# If access error: #
# - The system stack is changed to an access error stack frame #
# If FPU disabled: #
# - The system stack is changed to an FPU disabled stack frame #
# If Trace exception enabled: #
# - The system stack is changed to a Trace exception stack frame #
# Else: (normal case) #
# - None (correct result has been stored as appropriate) #
# #
# ALGORITHM *********************************************************** #
# This exception handles 3 types of operations: #
# (1) FP Instructions using extended precision or packed immediate #
# addressing mode. #
# (2) The "fmovm.x" instruction w/ dynamic register specification. #
# (3) The "fmovm.l" instruction w/ 2 or 3 control registers. #
# #
# For immediate data operations, the data is read in w/ a #
# _mem_read() "callout", converted to FP binary (if packed), and used #
# as the source operand to the instruction specified by the instruction #
# word. If no FP exception should be reported ads a result of the #
# emulation, then the result is stored to the destination register and #
# the handler exits through _fpsp_done(). If an enabled exc has been #
# signalled as a result of emulation, then an fsave state frame #
# corresponding to the FP exception type must be entered into the 060 #
# FPU before exiting. In either the enabled or disabled cases, we #
# must also check if a Trace exception is pending, in which case, we #
# must create a Trace exception stack frame from the current exception #
# stack frame. If no Trace is pending, we simply exit through #
# _fpsp_done(). #
# For "fmovm.x", call the routine fmovm_dynamic() which will #
# decode and emulate the instruction. No FP exceptions can be pending #
# as a result of this operation emulation. A Trace exception can be #
# pending, though, which means the current stack frame must be changed #
# to a Trace stack frame and an exit made through _real_trace(). #
# For the case of "fmovm.x Dn,-(a7)", where the offending instruction #
# was executed from supervisor mode, this handler must store the FP #
# register file values to the system stack by itself since #
# fmovm_dynamic() can't handle this. A normal exit is made through #
# fpsp_done(). #
# For "fmovm.l", fmovm_ctrl() is used to emulate the instruction. #
# Again, a Trace exception may be pending and an exit made through #
# _real_trace(). Else, a normal exit is made through _fpsp_done(). #
# #
# Before any of the above is attempted, it must be checked to #
# see if the FPU is disabled. Since the "Unimp <ea>" exception is taken #
# before the "FPU disabled" exception, but the "FPU disabled" exception #
# has higher priority, we check the disabled bit in the PCR. If set, #
# then we must create an 8 word"FPU disabled" exception stack frame #
# from the current 4 word exception stack frame. This includes #
# reproducing the effective address of the instruction to put on the #
# new stack frame. #
# #
# In the process of all emulation work, if a _mem_read() #
# "callout" returns a failing result indicating an access error, then #
# we must create an access error stack frame from the current stack #
# frame. This information includes a faulting address and a fault- #
# status-longword. These are created within this handler. #
# #
#########################################################################
global _fpsp_effadd
_fpsp_effadd:
# This exception type takes priority over the "Line F Emulator"
# exception. Therefore, the FPU could be disabled when entering here.
# So, we must check to see if it's disabled and handle that case separately.
mov.l %d0,-(%sp) # save d0
movc %pcr,%d0 # load proc cr
btst &0x1,%d0 # is FPU disabled?
bne.w iea_disabled # yes
mov.l (%sp)+,%d0 # restore d0
link %a6,&-LOCAL_SIZE # init stack frame
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs
fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack
# PC of instruction that took the exception is the PC in the frame
mov.l EXC_PC(%a6),EXC_EXTWPTR(%a6)
mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr
addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr
bsr.l _imem_read_long # fetch the instruction words
mov.l %d0,EXC_OPWORD(%a6) # store OPWORD and EXTWORD
tst.w %d0 # is operation fmovem?
bmi.w iea_fmovm # yes
#
# here, we will have:
# fabs fdabs fsabs facos fmod
# fadd fdadd fsadd fasin frem
# fcmp fatan fscale
# fdiv fddiv fsdiv fatanh fsin
# fint fcos fsincos
# fintrz fcosh fsinh
# fmove fdmove fsmove fetox ftan
# fmul fdmul fsmul fetoxm1 ftanh
# fneg fdneg fsneg fgetexp ftentox
# fsgldiv fgetman ftwotox
# fsglmul flog10
# fsqrt flog2
# fsub fdsub fssub flogn
# ftst flognp1
# which can all use f<op>.{x,p}
# so, now it's immediate data extended precision AND PACKED FORMAT!
#
iea_op:
andi.l &0x00ff00ff,USER_FPSR(%a6)
btst &0xa,%d0 # is src fmt x or p?
bne.b iea_op_pack # packed
mov.l EXC_EXTWPTR(%a6),%a0 # pass: ptr to #<data>
lea FP_SRC(%a6),%a1 # pass: ptr to super addr
mov.l &0xc,%d0 # pass: 12 bytes
bsr.l _imem_read # read extended immediate
tst.l %d1 # did ifetch fail?
bne.w iea_iacc # yes
bra.b iea_op_setsrc
iea_op_pack:
mov.l EXC_EXTWPTR(%a6),%a0 # pass: ptr to #<data>
lea FP_SRC(%a6),%a1 # pass: ptr to super dst
mov.l &0xc,%d0 # pass: 12 bytes
bsr.l _imem_read # read packed operand
tst.l %d1 # did ifetch fail?
bne.w iea_iacc # yes
# The packed operand is an INF or a NAN if the exponent field is all ones.
bfextu FP_SRC(%a6){&1:&15},%d0 # get exp
cmpi.w %d0,&0x7fff # INF or NAN?
beq.b iea_op_setsrc # operand is an INF or NAN
# The packed operand is a zero if the mantissa is all zero, else it's
# a normal packed op.
mov.b 3+FP_SRC(%a6),%d0 # get byte 4
andi.b &0x0f,%d0 # clear all but last nybble
bne.b iea_op_gp_not_spec # not a zero
tst.l FP_SRC_HI(%a6) # is lw 2 zero?
bne.b iea_op_gp_not_spec # not a zero
tst.l FP_SRC_LO(%a6) # is lw 3 zero?
beq.b iea_op_setsrc # operand is a ZERO
iea_op_gp_not_spec:
lea FP_SRC(%a6),%a0 # pass: ptr to packed op
bsr.l decbin # convert to extended
fmovm.x &0x80,FP_SRC(%a6) # make this the srcop
iea_op_setsrc:
addi.l &0xc,EXC_EXTWPTR(%a6) # update extension word pointer
# FP_SRC now holds the src operand.
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l set_tag_x # tag the operand type
mov.b %d0,STAG(%a6) # could be ANYTHING!!!
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b iea_op_getdst # no
bsr.l unnorm_fix # yes; convert to NORM/DENORM/ZERO
mov.b %d0,STAG(%a6) # set new optype tag
iea_op_getdst: clr.b STORE_FLG(%a6) # clear "store result" boolean
btst &0x5,1+EXC_CMDREG(%a6) # is operation monadic or dyadic?
beq.b iea_op_extract # monadic
btst &0x4,1+EXC_CMDREG(%a6) # is operation fsincos,ftst,fcmp?
bne.b iea_op_spec # yes
lea FP_DST(%a6),%a0 # pass: ptr to dst op
bsr.l set_tag_x # tag the operand type
mov.b %d0,DTAG(%a6) # could be ANYTHING!!!
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b iea_op_extract # no
bsr.l unnorm_fix # yes; convert to NORM/DENORM/ZERO
mov.b %d0,DTAG(%a6) # set new optype tag
bra.b iea_op_extract
# the operation is fsincos, ftst, or fcmp. only fcmp is dyadic
iea_op_spec:
btst &0x3,1+EXC_CMDREG(%a6) # is operation fsincos?
beq.b iea_op_extract # yes
# now, we're left with ftst and fcmp. so, first let's tag them so that they don't
# store a result. then, only fcmp will branch back and pick up a dst operand. st STORE_FLG(%a6) # don't store a final result
btst &0x1,1+EXC_CMDREG(%a6) # is operation fcmp?
beq.b iea_op_loaddst # yes
#
# Exceptions in order of precedence:
# BSUN : none
# SNAN : all operations
# OPERR : all reg-reg or mem-reg operations that can normally operr
# OVFL : same as OPERR
# UNFL : same as OPERR
# DZ : same as OPERR
# INEX2 : same as OPERR
# INEX1 : all packed immediate operations
#
# we determine the highest priority exception(if any) set by the
# emulation routine that has also been enabled by the user.
mov.b FPCR_ENABLE(%a6),%d0 # fetch exceptions enabled
bne.b iea_op_ena # some are enabled
# now, we save the result, unless, of course, the operation was ftst or fcmp.
# these don't save results.
iea_op_save:
tst.b STORE_FLG(%a6) # does this op store a result?
bne.b iea_op_exit1 # exit with no frestore
iea_op_store:
bfextu EXC_CMDREG(%a6){&6:&3},%d0 # fetch dst regno
bsr.l store_fpreg # store the result
iea_op_exit1:
mov.l EXC_PC(%a6),USER_FPIAR(%a6) # set FPIAR to "Current PC"
mov.l EXC_EXTWPTR(%a6),EXC_PC(%a6) # set"Next PC" in exc frame
btst &0x7,(%sp) # is trace on?
bne.w iea_op_trace # yes
bra.l _fpsp_done # exit to os
iea_op_ena:
and.b FPSR_EXCEPT(%a6),%d0 # keep only ones enable and set
bfffo %d0{&24:&8},%d0 # find highest priority exception
bne.b iea_op_exc # at least one was set
# no exception occurred. now, did a disabled, exact overflow occur with inexact
# enabled? if so, then we have to stuff an overflow frame into the FPU.
btst &ovfl_bit,FPSR_EXCEPT(%a6) # did overflow occur?
beq.b iea_op_save
iea_op_ovfl:
btst &inex2_bit,FPCR_ENABLE(%a6) # is inexact enabled?
beq.b iea_op_store # no
bra.b iea_op_exc_ovfl # yes
# an enabled exception occurred. we have to insert the exception type back into
# the machine.
iea_op_exc:
subi.l &24,%d0 # fix offset to be 0-8
cmpi.b %d0,&0x6 # is exception INEX?
bne.b iea_op_exc_force # no
# the enabled exception was inexact. so, if it occurs with an overflow
# or underflow that was disabled, then we have to force an overflow or
# underflow frame.
btst &ovfl_bit,FPSR_EXCEPT(%a6) # did overflow occur?
bne.b iea_op_exc_ovfl # yes
btst &unfl_bit,FPSR_EXCEPT(%a6) # did underflow occur?
bne.b iea_op_exc_unfl # yes
iea_op_exc_force:
mov.w (tbl_iea_except.b,%pc,%d0.w*2),2+FP_SRC(%a6)
bra.b iea_op_exit2 # exit with frestore
tbl_iea_except:
short 0xe002, 0xe006, 0xe004, 0xe005
short 0xe003, 0xe002, 0xe001, 0xe001
btst &0x7,(%sp) # is trace on?
bne.b iea_op_trace # yes
bra.l _fpsp_done # exit to os
#
# The opclass two instruction that took an "Unimplemented Effective Address"
# exception was being traced. Make the "current" PC the FPIAR and put it in
# the trace stack frame then jump to _real_trace().
#
# UNIMP EA FRAME TRACE FRAME
# ***************** *****************
# * 0x0 * 0x0f0 * * Current *
# ***************** * PC *
# * Current * *****************
# * PC * * 0x2 * 0x024 *
# ***************** *****************
# * SR * * Next *
# ***************** * PC *
# *****************
# * SR *
# *****************
iea_op_trace:
mov.l (%sp),-(%sp) # shift stack frame "down"
mov.w 0x8(%sp),0x4(%sp)
mov.w &0x2024,0x6(%sp) # stk fmt = 0x2; voff = 0x024
fmov.l %fpiar,0x8(%sp) # "Current PC" is in FPIAR
bra.l _real_trace
#########################################################################
iea_fmovm:
btst &14,%d0 # ctrl or data reg
beq.w iea_fmovm_ctrl
iea_fmovm_data:
btst &0x5,EXC_SR(%a6) # user or supervisor mode
bne.b iea_fmovm_data_s
iea_fmovm_data_u:
mov.l %usp,%a0
mov.l %a0,EXC_A7(%a6) # store current a7
bsr.l fmovm_dynamic # do dynamic fmovm
mov.l EXC_A7(%a6),%a0 # load possibly new a7
mov.l %a0,%usp # update usp
bra.w iea_fmovm_exit
# right now, d0 = the size.
# the data has been fetched from the supervisor stack, but we have not
# incremented the stack pointer by the appropriate number of bytes.
# do it here.
iea_fmovm_data_postinc:
btst &0x7,EXC_SR(%a6)
bne.b iea_fmovm_data_pi_trace
btst &0x7,EXC_SR(%a6) # is trace on?
bne.b iea_fmovm_trace # yes
mov.l EXC_EXTWPTR(%a6),EXC_PC(%a6) # set Next PC
unlk %a6 # unravel the frame
bra.l _fpsp_done # exit to os
#
# The control reg instruction that took an "Unimplemented Effective Address"
# exception was being traced. The "Current PC" for the trace frame is the
# PC stacked for Unimp EA. The "Next PC" is in EXC_EXTWPTR.
# After fixing the stack frame, jump to _real_trace().
#
# UNIMP EA FRAME TRACE FRAME
# ***************** *****************
# * 0x0 * 0x0f0 * * Current *
# ***************** * PC *
# * Current * *****************
# * PC * * 0x2 * 0x024 *
# ***************** *****************
# * SR * * Next *
# ***************** * PC *
# *****************
# * SR *
# *****************
# this ain't a pretty solution, but it works:
# -restore a6 (not with unlk)
# -shift stack frame down over where old a6 used to be
# -add LOCAL_SIZE to stack pointer
iea_fmovm_trace:
mov.l (%a6),%a6 # restore frame pointer
mov.w EXC_SR+LOCAL_SIZE(%sp),0x0+LOCAL_SIZE(%sp)
mov.l EXC_PC+LOCAL_SIZE(%sp),0x8+LOCAL_SIZE(%sp)
mov.l EXC_EXTWPTR+LOCAL_SIZE(%sp),0x2+LOCAL_SIZE(%sp)
mov.w &0x2024,0x6+LOCAL_SIZE(%sp) # stk fmt = 0x2; voff = 0x024
add.l &LOCAL_SIZE,%sp # clear stack frame
bra.l _real_trace
#########################################################################
# The FPU is disabled and so we should really have taken the "Line
# F Emulator" exception. So, here we create an 8-word stack frame
# from our 4-word stack frame. This means we must calculate the length
# the faulting instruction to get the "next PC". This is trivial for
# immediate operands but requires some extra work for fmovm dynamic
# which can use most addressing modes.
iea_disabled:
mov.l (%sp)+,%d0 # restore d0
link %a6,&-LOCAL_SIZE # init stack frame
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
# PC of instruction that took the exception is the PC in the frame
mov.l EXC_PC(%a6),EXC_EXTWPTR(%a6)
mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr
addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr
bsr.l _imem_read_long # fetch the instruction words
mov.l %d0,EXC_OPWORD(%a6) # store OPWORD and EXTWORD
tst.w %d0 # is instr fmovm?
bmi.b iea_dis_fmovm # yes
# instruction is using an extended precision immediate operand. Therefore,
# the total instruction length is 16 bytes.
iea_dis_immed:
mov.l &0x10,%d0 # 16 bytes of instruction
bra.b iea_dis_cont
iea_dis_fmovm:
btst &0xe,%d0 # is instr fmovm ctrl
bne.b iea_dis_fmovm_data # no
# the instruction is a fmovm.l with 2 or 3 registers.
bfextu %d0{&19:&3},%d1
mov.l &0xc,%d0
cmpi.b %d1,&0x7 # move all regs?
bne.b iea_dis_cont
addq.l &0x4,%d0
bra.b iea_dis_cont
# the instruction is an fmovm.x dynamic which can use many addressing
# modes and thus can have several different total instruction lengths.
# call fmovm_calc_ea which will go through the ea calc process and,
# as a by-product, will tell us how long the instruction is.
iea_dis_fmovm_data: clr.l %d0
bsr.l fmovm_calc_ea
mov.l EXC_EXTWPTR(%a6),%d0 sub.l EXC_PC(%a6),%d0
iea_dis_cont:
mov.w %d0,EXC_VOFF(%a6) # store stack shift value
# here, we actually create the 8-word frame from the 4-word frame,
# with the "next PC" as additional info.
# the <ea> field is let as undefined.
subq.l &0x8,%sp # make room for new stack
mov.l %d0,-(%sp) # save d0
mov.w 0xc(%sp),0x4(%sp) # move SR
mov.l 0xe(%sp),0x6(%sp) # move Current PC clr.l %d0
mov.w 0x12(%sp),%d0
mov.l 0x6(%sp),0x10(%sp) # move Current PC
add.l %d0,0x6(%sp) # make Next PC
mov.w &0x402c,0xa(%sp) # insert offset,frame format
mov.l (%sp)+,%d0 # restore d0
subq.w &0x8,%sp # make stack frame bigger
mov.l 0x8(%sp),(%sp) # store SR,hi(PC)
mov.w 0xc(%sp),0x4(%sp) # store lo(PC)
mov.w &0x4008,0x6(%sp) # store voff
mov.l 0x2(%sp),0x8(%sp) # store ea
mov.l &0x09428001,0xc(%sp) # store fslw
iea_acc_done:
btst &0x5,(%sp) # user or supervisor mode?
beq.b iea_acc_done2 # user
bset &0x2,0xd(%sp) # set supervisor TM bit
#########################################################################
# XDEF **************************************************************** #
# _fpsp_operr(): 060FPSP entry point for FP Operr exception. #
# #
# This handler should be the first code executed upon taking the #
# FP Operand Error exception in an operating system. #
# #
# XREF **************************************************************** #
# _imem_read_long() - read instruction longword #
# fix_skewed_ops() - adjust src operand in fsave frame #
# _real_operr() - "callout" to operating system operr handler #
# _dmem_write_{byte,word,long}() - store data to mem (opclass 3) #
# store_dreg_{b,w,l}() - store data to data regfile (opclass 3) #
# facc_out_{b,w,l}() - store to memory took access error (opcl 3) #
# #
# INPUT *************************************************************** #
# - The system stack contains the FP Operr exception frame #
# - The fsave frame contains the source operand #
# #
# OUTPUT ************************************************************** #
# No access error: #
# - The system stack is unchanged #
# - The fsave frame contains the adjusted src op for opclass 0,2 #
# #
# ALGORITHM *********************************************************** #
# In a system where the FP Operr exception is enabled, the goal #
# is to get to the handler specified at _real_operr(). But, on the 060, #
# for opclass zero and two instruction taking this exception, the #
# input operand in the fsave frame may be incorrect for some cases #
# and needs to be corrected. This handler calls fix_skewed_ops() to #
# do just this and then exits through _real_operr(). #
# For opclass 3 instructions, the 060 doesn't store the default #
# operr result out to memory or data register file as it should. #
# This code must emulate the move out before finally exiting through #
# _real_inex(). The move out, if to memory, is performed using #
# _mem_write() "callout" routines that may return a failing result. #
# In this special case, the handler must exit through facc_out() #
# which creates an access error stack frame from the current operr #
# stack frame. #
# #
#########################################################################
global _fpsp_operr
_fpsp_operr:
link.w %a6,&-LOCAL_SIZE # init stack frame
fsave FP_SRC(%a6) # grab the "busy" frame
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs
fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack
# the FPIAR holds the "current PC" of the faulting instruction
mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6)
btst &13,%d0 # is instr an fmove out?
bne.b foperr_out # fmove out
# here, we simply see if the operand in the fsave frame needs to be "unskewed".
# this would be the case for opclass two operations with a source infinity or
# denorm operand in the sgl or dbl format. NANs also become skewed, but can't
# cause an operr so we don't need to check for them here.
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l fix_skewed_ops # fix src op
#
# the hardware does not save the default result to memory on enabled
# operand error exceptions. we do this here before passing control to
# the user operand error handler.
#
# byte, word, and long destination format operations can pass
# through here. we simply need to test the sign of the src
# operand and save the appropriate minimum or maximum integer value
# to the effective address as pointed to by the stacked effective address.
#
# although packed opclass three operations can take operand error
# exceptions, they won't pass through here since they are caught
# first by the unsupported data format exception handler. that handler
# sends them directly to _real_operr() if necessary.
#
foperr_out:
mov.w FP_SRC_EX(%a6),%d1 # fetch exponent
andi.w &0x7fff,%d1
cmpi.w %d1,&0x7fff
bne.b foperr_out_not_qnan
# the operand is either an infinity or a QNAN.
tst.l FP_SRC_LO(%a6)
bne.b foperr_out_qnan
mov.l FP_SRC_HI(%a6),%d1
andi.l &0x7fffffff,%d1
beq.b foperr_out_not_qnan
foperr_out_qnan:
mov.l FP_SRC_HI(%a6),L_SCR1(%a6)
bra.b foperr_out_jmp
foperr_out_jmp:
bfextu %d0{&19:&3},%d0 # extract dst format field
mov.b 1+EXC_OPWORD(%a6),%d1 # extract <ea> mode,reg
mov.w (tbl_operr.b,%pc,%d0.w*2),%a0
jmp (tbl_operr.b,%pc,%a0)
tbl_operr:
short foperr_out_l - tbl_operr # long word integer
short tbl_operr - tbl_operr # sgl prec shouldn't happen
short tbl_operr - tbl_operr # ext prec shouldn't happen
short foperr_exit - tbl_operr # packed won't enter here
short foperr_out_w - tbl_operr # word integer
short tbl_operr - tbl_operr # dbl prec shouldn't happen
short foperr_out_b - tbl_operr # byte integer
short tbl_operr - tbl_operr # packed won't enter here
foperr_out_b:
mov.b L_SCR1(%a6),%d0 # load positive default result
cmpi.b %d1,&0x7 # is <ea> mode a data reg?
ble.b foperr_out_b_save_dn # yes
mov.l EXC_EA(%a6),%a0 # pass: <ea> of default result
bsr.l _dmem_write_byte # write the default result
tst.l %d1 # did dstore fail?
bne.l facc_out_b # yes
bra.w foperr_exit
foperr_out_b_save_dn:
andi.w &0x0007,%d1
bsr.l store_dreg_b # store result to regfile
bra.w foperr_exit
foperr_out_w:
mov.w L_SCR1(%a6),%d0 # load positive default result
cmpi.b %d1,&0x7 # is <ea> mode a data reg?
ble.b foperr_out_w_save_dn # yes
mov.l EXC_EA(%a6),%a0 # pass: <ea> of default result
bsr.l _dmem_write_word # write the default result
tst.l %d1 # did dstore fail?
bne.l facc_out_w # yes
bra.w foperr_exit
foperr_out_w_save_dn:
andi.w &0x0007,%d1
bsr.l store_dreg_w # store result to regfile
bra.w foperr_exit
foperr_out_l:
mov.l L_SCR1(%a6),%d0 # load positive default result
cmpi.b %d1,&0x7 # is <ea> mode a data reg?
ble.b foperr_out_l_save_dn # yes
mov.l EXC_EA(%a6),%a0 # pass: <ea> of default result
bsr.l _dmem_write_long # write the default result
tst.l %d1 # did dstore fail?
bne.l facc_out_l # yes
bra.w foperr_exit
foperr_out_l_save_dn:
andi.w &0x0007,%d1
bsr.l store_dreg_l # store result to regfile
bra.w foperr_exit
#########################################################################
# XDEF **************************************************************** #
# _fpsp_snan(): 060FPSP entry point for FP SNAN exception. #
# #
# This handler should be the first code executed upon taking the #
# FP Signalling NAN exception in an operating system. #
# #
# XREF **************************************************************** #
# _imem_read_long() - read instruction longword #
# fix_skewed_ops() - adjust src operand in fsave frame #
# _real_snan() - "callout" to operating system SNAN handler #
# _dmem_write_{byte,word,long}() - store data to mem (opclass 3) #
# store_dreg_{b,w,l}() - store data to data regfile (opclass 3) #
# facc_out_{b,w,l,d,x}() - store to mem took acc error (opcl 3) #
# _calc_ea_fout() - fix An if <ea> is -() or ()+; also get <ea> #
# #
# INPUT *************************************************************** #
# - The system stack contains the FP SNAN exception frame #
# - The fsave frame contains the source operand #
# #
# OUTPUT ************************************************************** #
# No access error: #
# - The system stack is unchanged #
# - The fsave frame contains the adjusted src op for opclass 0,2 #
# #
# ALGORITHM *********************************************************** #
# In a system where the FP SNAN exception is enabled, the goal #
# is to get to the handler specified at _real_snan(). But, on the 060, #
# for opclass zero and two instructions taking this exception, the #
# input operand in the fsave frame may be incorrect for some cases #
# and needs to be corrected. This handler calls fix_skewed_ops() to #
# do just this and then exits through _real_snan(). #
# For opclass 3 instructions, the 060 doesn't store the default #
# SNAN result out to memory or data register file as it should. #
# This code must emulate the move out before finally exiting through #
# _real_snan(). The move out, if to memory, is performed using #
# _mem_write() "callout" routines that may return a failing result. #
# In this special case, the handler must exit through facc_out() #
# which creates an access error stack frame from the current SNAN #
# stack frame. #
# For the case of an extended precision opclass 3 instruction, #
# if the effective addressing mode was -() or ()+, then the address #
# register must get updated by calling _calc_ea_fout(). If the <ea> #
# was -(a7) from supervisor mode, then the exception frame currently #
# on the system stack must be carefully moved "down" to make room #
# for the operand being moved. #
# #
#########################################################################
global _fpsp_snan
_fpsp_snan:
link.w %a6,&-LOCAL_SIZE # init stack frame
fsave FP_SRC(%a6) # grab the "busy" frame
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs
fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack
# the FPIAR holds the "current PC" of the faulting instruction
mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6)
btst &13,%d0 # is instr an fmove out?
bne.w fsnan_out # fmove out
# here, we simply see if the operand in the fsave frame needs to be "unskewed".
# this would be the case for opclass two operations with a source infinity or
# denorm operand in the sgl or dbl format. NANs also become skewed and must be
# fixed here.
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l fix_skewed_ops # fix src op
#
# the hardware does not save the default result to memory on enabled
# snan exceptions. we do this here before passing control to
# the user snan handler.
#
# byte, word, long, and packed destination format operations can pass
# through here. since packed format operations already were handled by
# fpsp_unsupp(), then we need to do nothing else for them here.
# for byte, word, and long, we simply need to test the sign of the src
# operand and save the appropriate minimum or maximum integer value
# to the effective address as pointed to by the stacked effective address.
#
fsnan_out:
bfextu %d0{&19:&3},%d0 # extract dst format field
mov.b 1+EXC_OPWORD(%a6),%d1 # extract <ea> mode,reg
mov.w (tbl_snan.b,%pc,%d0.w*2),%a0
jmp (tbl_snan.b,%pc,%a0)
tbl_snan:
short fsnan_out_l - tbl_snan # long word integer
short fsnan_out_s - tbl_snan # sgl prec shouldn't happen
short fsnan_out_x - tbl_snan # ext prec shouldn't happen
short tbl_snan - tbl_snan # packed needs no help
short fsnan_out_w - tbl_snan # word integer
short fsnan_out_d - tbl_snan # dbl prec shouldn't happen
short fsnan_out_b - tbl_snan # byte integer
short tbl_snan - tbl_snan # packed needs no help
fsnan_out_b:
mov.b FP_SRC_HI(%a6),%d0 # load upper byte of SNAN
bset &6,%d0 # set SNAN bit
cmpi.b %d1,&0x7 # is <ea> mode a data reg?
ble.b fsnan_out_b_dn # yes
mov.l EXC_EA(%a6),%a0 # pass: <ea> of default result
bsr.l _dmem_write_byte # write the default result
tst.l %d1 # did dstore fail?
bne.l facc_out_b # yes
bra.w fsnan_exit
fsnan_out_b_dn:
andi.w &0x0007,%d1
bsr.l store_dreg_b # store result to regfile
bra.w fsnan_exit
fsnan_out_w:
mov.w FP_SRC_HI(%a6),%d0 # load upper word of SNAN
bset &14,%d0 # set SNAN bit
cmpi.b %d1,&0x7 # is <ea> mode a data reg?
ble.b fsnan_out_w_dn # yes
mov.l EXC_EA(%a6),%a0 # pass: <ea> of default result
bsr.l _dmem_write_word # write the default result
tst.l %d1 # did dstore fail?
bne.l facc_out_w # yes
bra.w fsnan_exit
fsnan_out_w_dn:
andi.w &0x0007,%d1
bsr.l store_dreg_w # store result to regfile
bra.w fsnan_exit
fsnan_out_l:
mov.l FP_SRC_HI(%a6),%d0 # load upper longword of SNAN
bset &30,%d0 # set SNAN bit
cmpi.b %d1,&0x7 # is <ea> mode a data reg?
ble.b fsnan_out_l_dn # yes
mov.l EXC_EA(%a6),%a0 # pass: <ea> of default result
bsr.l _dmem_write_long # write the default result
tst.l %d1 # did dstore fail?
bne.l facc_out_l # yes
bra.w fsnan_exit
fsnan_out_l_dn:
andi.w &0x0007,%d1
bsr.l store_dreg_l # store result to regfile
bra.w fsnan_exit
fsnan_out_s:
cmpi.b %d1,&0x7 # is <ea> mode a data reg?
ble.b fsnan_out_d_dn # yes
mov.l FP_SRC_EX(%a6),%d0 # fetch SNAN sign
andi.l &0x80000000,%d0 # keep sign
ori.l &0x7fc00000,%d0 # insert new exponent,SNAN bit
mov.l FP_SRC_HI(%a6),%d1 # load mantissa
lsr.l &0x8,%d1 # shift mantissa for sgl
or.l %d1,%d0 # create sgl SNAN
mov.l EXC_EA(%a6),%a0 # pass: <ea> of default result
bsr.l _dmem_write_long # write the default result
tst.l %d1 # did dstore fail?
bne.l facc_out_l # yes
bra.w fsnan_exit
fsnan_out_d_dn:
mov.l FP_SRC_EX(%a6),%d0 # fetch SNAN sign
andi.l &0x80000000,%d0 # keep sign
ori.l &0x7fc00000,%d0 # insert new exponent,SNAN bit
mov.l %d1,-(%sp)
mov.l FP_SRC_HI(%a6),%d1 # load mantissa
lsr.l &0x8,%d1 # shift mantissa for sgl
or.l %d1,%d0 # create sgl SNAN
mov.l (%sp)+,%d1
andi.w &0x0007,%d1
bsr.l store_dreg_l # store result to regfile
bra.w fsnan_exit
fsnan_out_d:
mov.l FP_SRC_EX(%a6),%d0 # fetch SNAN sign
andi.l &0x80000000,%d0 # keep sign
ori.l &0x7ff80000,%d0 # insert new exponent,SNAN bit
mov.l FP_SRC_HI(%a6),%d1 # load hi mantissa
mov.l %d0,FP_SCR0_EX(%a6) # store to temp space
mov.l &11,%d0 # load shift amt
lsr.l %d0,%d1
or.l %d1,FP_SCR0_EX(%a6) # create dbl hi
mov.l FP_SRC_HI(%a6),%d1 # load hi mantissa
andi.l &0x000007ff,%d1
ror.l %d0,%d1
mov.l %d1,FP_SCR0_HI(%a6) # store to temp space
mov.l FP_SRC_LO(%a6),%d1 # load lo mantissa
lsr.l %d0,%d1
or.l %d1,FP_SCR0_HI(%a6) # create dbl lo
lea FP_SCR0(%a6),%a0 # pass: ptr to operand
mov.l EXC_EA(%a6),%a1 # pass: dst addr
movq.l &0x8,%d0 # pass: size of 8 bytes
bsr.l _dmem_write # write the default result
tst.l %d1 # did dstore fail?
bne.l facc_out_d # yes
bra.w fsnan_exit
# for extended precision, if the addressing mode is pre-decrement or
# post-increment, then the address register did not get updated.
# in addition, for pre-decrement, the stacked <ea> is incorrect.
fsnan_out_x: clr.b SPCOND_FLG(%a6) # clear special case flag
#########################################################################
# XDEF **************************************************************** #
# _fpsp_inex(): 060FPSP entry point for FP Inexact exception. #
# #
# This handler should be the first code executed upon taking the #
# FP Inexact exception in an operating system. #
# #
# XREF **************************************************************** #
# _imem_read_long() - read instruction longword #
# fix_skewed_ops() - adjust src operand in fsave frame #
# set_tag_x() - determine optype of src/dst operands #
# store_fpreg() - store opclass 0 or 2 result to FP regfile #
# unnorm_fix() - change UNNORM operands to NORM or ZERO #
# load_fpn2() - load dst operand from FP regfile #
# smovcr() - emulate an "fmovcr" instruction #
# fout() - emulate an opclass 3 instruction #
# tbl_unsupp - add of table of emulation routines for opclass 0,2 #
# _real_inex() - "callout" to operating system inexact handler #
# #
# INPUT *************************************************************** #
# - The system stack contains the FP Inexact exception frame #
# - The fsave frame contains the source operand #
# #
# OUTPUT ************************************************************** #
# - The system stack is unchanged #
# - The fsave frame contains the adjusted src op for opclass 0,2 #
# #
# ALGORITHM *********************************************************** #
# In a system where the FP Inexact exception is enabled, the goal #
# is to get to the handler specified at _real_inex(). But, on the 060, #
# for opclass zero and two instruction taking this exception, the #
# hardware doesn't store the correct result to the destination FP #
# register as did the '040 and '881/2. This handler must emulate the #
# instruction in order to get this value and then store it to the #
# correct register before calling _real_inex(). #
# For opclass 3 instructions, the 060 doesn't store the default #
# inexact result out to memory or data register file as it should. #
# This code must emulate the move out by calling fout() before finally #
# exiting through _real_inex(). #
# #
#########################################################################
global _fpsp_inex
_fpsp_inex:
link.w %a6,&-LOCAL_SIZE # init stack frame
fsave FP_SRC(%a6) # grab the "busy" frame
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs
fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack
# the FPIAR holds the "current PC" of the faulting instruction
mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6)
btst &13,%d0 # is instr an fmove out?
bne.w finex_out # fmove out
# the hardware, for "fabs" and "fneg" w/ a long source format, puts the
# longword integer directly into the upper longword of the mantissa along
# w/ an exponent value of 0x401e. we convert this to extended precision here.
bfextu %d0{&19:&3},%d0 # fetch instr size
bne.b finex_cont # instr size is not long
cmpi.w FP_SRC_EX(%a6),&0x401e # is exponent 0x401e?
bne.b finex_cont # no
fmov.l &0x0,%fpcr
fmov.l FP_SRC_HI(%a6),%fp0 # load integer src
fmov.x %fp0,FP_SRC(%a6) # store integer as extended precision
mov.w &0xe001,0x2+FP_SRC(%a6)
finex_cont:
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l fix_skewed_ops # fix src op
# Here, we zero the ccode and exception byte field since we're going to
# emulate the whole instruction. Notice, though, that we don't kill the
# INEX1 bit. This is because a packed op has long since been converted
# to extended before arriving here. Therefore, we need to retain the
# INEX1 bit from when the operand was first converted.
andi.l &0x00ff01ff,USER_FPSR(%a6) # zero all but accured field
fmov.l &0x0,%fpcr # zero current control regs
fmov.l &0x0,%fpsr
bfextu EXC_EXTWORD(%a6){&0:&6},%d1 # extract upper 6 of cmdreg
cmpi.b %d1,&0x17 # is op an fmovecr?
beq.w finex_fmovcr # yes
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l set_tag_x # tag the operand type
mov.b %d0,STAG(%a6) # maybe NORM,DENORM
# bits four and five of the fp extension word separate the monadic and dyadic
# operations that can pass through fpsp_inex(). remember that fcmp and ftst
# will never take this exception, but fsincos will.
btst &0x5,1+EXC_CMDREG(%a6) # is operation monadic or dyadic?
beq.b finex_extract # monadic
btst &0x4,1+EXC_CMDREG(%a6) # is operation an fsincos?
bne.b finex_extract # yes
lea FP_DST(%a6),%a0 # pass: ptr to dst op
bsr.l set_tag_x # tag the operand type
cmpi.b %d0,&UNNORM # is operand an UNNORM?
bne.b finex_op2_done # no
bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO
finex_op2_done:
mov.b %d0,DTAG(%a6) # save dst optype tag
#
# the hardware does not save the default result to memory on enabled
# inexact exceptions. we do this here before passing control to
# the user inexact handler.
#
# byte, word, and long destination format operations can pass
# through here. so can double and single precision.
# although packed opclass three operations can take inexact
# exceptions, they won't pass through here since they are caught
# first by the unsupported data format exception handler. that handler
# sends them directly to _real_inex() if necessary.
#
finex_out:
andi.l &0xffff00ff,USER_FPSR(%a6) # zero exception field
lea FP_SRC(%a6),%a0 # pass ptr to src operand
bsr.l fout # store the default result
bra.b finex_exit
#########################################################################
# XDEF **************************************************************** #
# _fpsp_dz(): 060FPSP entry point for FP DZ exception. #
# #
# This handler should be the first code executed upon taking #
# the FP DZ exception in an operating system. #
# #
# XREF **************************************************************** #
# _imem_read_long() - read instruction longword from memory #
# fix_skewed_ops() - adjust fsave operand #
# _real_dz() - "callout" exit point from FP DZ handler #
# #
# INPUT *************************************************************** #
# - The system stack contains the FP DZ exception stack. #
# - The fsave frame contains the source operand. #
# #
# OUTPUT ************************************************************** #
# - The system stack contains the FP DZ exception stack. #
# - The fsave frame contains the adjusted source operand. #
# #
# ALGORITHM *********************************************************** #
# In a system where the DZ exception is enabled, the goal is to #
# get to the handler specified at _real_dz(). But, on the 060, when the #
# exception is taken, the input operand in the fsave state frame may #
# be incorrect for some cases and need to be adjusted. So, this package #
# adjusts the operand using fix_skewed_ops() and then branches to #
# _real_dz(). #
# #
#########################################################################
global _fpsp_dz
_fpsp_dz:
link.w %a6,&-LOCAL_SIZE # init stack frame
fsave FP_SRC(%a6) # grab the "busy" frame
movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1
fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs
fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack
# the FPIAR holds the "current PC" of the faulting instruction
mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6)
# here, we simply see if the operand in the fsave frame needs to be "unskewed".
# this would be the case for opclass two operations with a source zero
# in the sgl or dbl format.
lea FP_SRC(%a6),%a0 # pass: ptr to src op
bsr.l fix_skewed_ops # fix src op
#########################################################################
# XDEF **************************************************************** #
# _fpsp_fline(): 060FPSP entry point for "Line F emulator" #
# exception when the "reduced" version of the #
# FPSP is implemented that does not emulate #
# FP unimplemented instructions. #
# #
# This handler should be the first code executed upon taking a #
# "Line F Emulator" exception in an operating system integrating #
# the reduced version of 060FPSP. #
# #
# XREF **************************************************************** #
# _real_fpu_disabled() - Handle "FPU disabled" exceptions #
# _real_fline() - Handle all other cases (treated equally) #
# #
# INPUT *************************************************************** #
# - The system stack contains a "Line F Emulator" exception #
# stack frame. #
# #
# OUTPUT ************************************************************** #
# - The system stack is unchanged. #
# #
# ALGORITHM *********************************************************** #
# When a "Line F Emulator" exception occurs in a system where #
# "FPU Unimplemented" instructions will not be emulated, the exception #
# can occur because then FPU is disabled or the instruction is to be #
# classifed as "Line F". This module determines which case exists and #
# calls the appropriate "callout". #
# #
#########################################################################
global _fpsp_fline
_fpsp_fline:
# check to see if the FPU is disabled. if so, jump to the OS entry
# point for that condition.
cmpi.w 0x6(%sp),&0x402c
beq.l _real_fpu_disabled
bra.l _real_fline
#########################################################################
# XDEF **************************************************************** #
# _dcalc_ea(): calc correct <ea> from <ea> stacked on exception #
# #
# XREF **************************************************************** #
# inc_areg() - increment an address register #
# dec_areg() - decrement an address register #
# #
# INPUT *************************************************************** #
# d0 = number of bytes to adjust <ea> by #
# #
# OUTPUT ************************************************************** #
# None #
# #
# ALGORITHM *********************************************************** #
# "Dummy" CALCulate Effective Address: #
# The stacked <ea> for FP unimplemented instructions and opclass #
# two packed instructions is correct with the exception of... #
# #
# 1) -(An) : The register is not updated regardless of size. #
# Also, for extended precision and packed, the #
# stacked <ea> value is 8 bytes too big #
# 2) (An)+ : The register is not updated. #
# 3) #<data> : The upper longword of the immediate operand is #
# stacked b,w,l and s sizes are completely stacked. #
# d,x, and p are not. #
# #
#########################################################################
global _dcalc_ea
_dcalc_ea:
mov.l %d0, %a0 # move # bytes to %a0
mov.b 1+EXC_OPWORD(%a6), %d0 # fetch opcode word
mov.l %d0, %d1 # make a copy
andi.w &0x38, %d0 # extract mode field
andi.l &0x7, %d1 # extract reg field
# need to set immediate data flag here since we'll need to do
# an imem_read to fetch this later.
dcea_imm:
mov.b &immed_flg,SPCOND_FLG(%a6)
lea ([USER_FPIAR,%a6],0x4),%a0 # no; return <ea>
rts
# here, the <ea> is stacked correctly. however, we must update the
# address register...
dcea_pi:
mov.l %a0,%d0 # pass amt to inc by
bsr.l inc_areg # inc addr register
mov.l EXC_EA(%a6),%a0 # stacked <ea> is correct
rts
# the <ea> is stacked correctly for all but extended and packed which
# the <ea>s are 8 bytes too large.
# it would make no sense to have a pre-decrement to a7 in supervisor
# mode so we don't even worry about this tricky case here : )
dcea_pd:
mov.l %a0,%d0 # pass amt to dec by
bsr.l dec_areg # dec addr register
mov.l EXC_EA(%a6),%a0 # stacked <ea> is correct
cmpi.b %d0,&0xc # is opsize ext or packed?
beq.b dcea_pd2 # yes
rts
dcea_pd2: sub.l &0x8,%a0 # correct <ea>
mov.l %a0,EXC_EA(%a6) # put correct <ea> on stack
rts
#########################################################################
# XDEF **************************************************************** #
# _calc_ea_fout(): calculate correct stacked <ea> for extended #
# and packed data opclass 3 operations. #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# None #
# #
# OUTPUT ************************************************************** #
# a0 = return correct effective address #
# #
# ALGORITHM *********************************************************** #
# For opclass 3 extended and packed data operations, the <ea> #
# stacked for the exception is incorrect for -(an) and (an)+ addressing #
# modes. Also, while we're at it, the index register itself must get #
# updated. #
# So, for -(an), we must subtract 8 off of the stacked <ea> value #
# and return that value as the correct <ea> and store that value in An. #
# For (an)+, the stacked <ea> is correct but we must adjust An by +12. #
# #
#########################################################################
# This calc_ea is currently used to retrieve the correct <ea>
# for fmove outs of type extended and packed. global _calc_ea_fout
_calc_ea_fout:
mov.b 1+EXC_OPWORD(%a6),%d0 # fetch opcode word
mov.l %d0,%d1 # make a copy
andi.w &0x38,%d0 # extract mode field
andi.l &0x7,%d1 # extract reg field
mov.l EXC_EA(%a6),%a0 # stacked <ea> is correct
rts
# (An)+ : extended and packed fmove out
# : stacked <ea> is correct
# : "An" not updated
ceaf_pi:
mov.w (tbl_ceaf_pi.b,%pc,%d1.w*2),%d1
mov.l EXC_EA(%a6),%a0
jmp (tbl_ceaf_pi.b,%pc,%d1.w*1)
swbeg &0x8
tbl_ceaf_pi:
short ceaf_pi0 - tbl_ceaf_pi
short ceaf_pi1 - tbl_ceaf_pi
short ceaf_pi2 - tbl_ceaf_pi
short ceaf_pi3 - tbl_ceaf_pi
short ceaf_pi4 - tbl_ceaf_pi
short ceaf_pi5 - tbl_ceaf_pi
short ceaf_pi6 - tbl_ceaf_pi
short ceaf_pi7 - tbl_ceaf_pi
# -(An) : extended and packed fmove out
# : stacked <ea> = actual <ea> + 8
# : "An" not updated
ceaf_pd:
mov.w (tbl_ceaf_pd.b,%pc,%d1.w*2),%d1
mov.l EXC_EA(%a6),%a0 sub.l &0x8,%a0 sub.l &0x8,EXC_EA(%a6)
jmp (tbl_ceaf_pd.b,%pc,%d1.w*1)
swbeg &0x8
tbl_ceaf_pd:
short ceaf_pd0 - tbl_ceaf_pd
short ceaf_pd1 - tbl_ceaf_pd
short ceaf_pd2 - tbl_ceaf_pd
short ceaf_pd3 - tbl_ceaf_pd
short ceaf_pd4 - tbl_ceaf_pd
short ceaf_pd5 - tbl_ceaf_pd
short ceaf_pd6 - tbl_ceaf_pd
short ceaf_pd7 - tbl_ceaf_pd
#
# This table holds the offsets of the emulation routines for each individual
# math operation relative to the address of this table. Included are
# routines like fadd/fmul/fabs. The transcendentals ARE NOT. This is because
# this table is for the version if the 060FPSP without transcendentals.
# The location within the table is determined by the extension bits of the
# operation longword.
#
swbeg &109
tbl_unsupp:
long fin - tbl_unsupp # 00: fmove
long fint - tbl_unsupp # 01: fint
long tbl_unsupp - tbl_unsupp # 02: fsinh
long fintrz - tbl_unsupp # 03: fintrz
long fsqrt - tbl_unsupp # 04: fsqrt
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp # 06: flognp1
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp # 08: fetoxm1
long tbl_unsupp - tbl_unsupp # 09: ftanh
long tbl_unsupp - tbl_unsupp # 0a: fatan
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp # 0c: fasin
long tbl_unsupp - tbl_unsupp # 0d: fatanh
long tbl_unsupp - tbl_unsupp # 0e: fsin
long tbl_unsupp - tbl_unsupp # 0f: ftan
long tbl_unsupp - tbl_unsupp # 10: fetox
long tbl_unsupp - tbl_unsupp # 11: ftwotox
long tbl_unsupp - tbl_unsupp # 12: ftentox
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp # 14: flogn
long tbl_unsupp - tbl_unsupp # 15: flog10
long tbl_unsupp - tbl_unsupp # 16: flog2
long tbl_unsupp - tbl_unsupp
long fabs - tbl_unsupp # 18: fabs
long tbl_unsupp - tbl_unsupp # 19: fcosh
long fneg - tbl_unsupp # 1a: fneg
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp # 1c: facos
long tbl_unsupp - tbl_unsupp # 1d: fcos
long tbl_unsupp - tbl_unsupp # 1e: fgetexp
long tbl_unsupp - tbl_unsupp # 1f: fgetman
long fdiv - tbl_unsupp # 20: fdiv
long tbl_unsupp - tbl_unsupp # 21: fmod
long fadd - tbl_unsupp # 22: fadd
long fmul - tbl_unsupp # 23: fmul
long fsgldiv - tbl_unsupp # 24: fsgldiv
long tbl_unsupp - tbl_unsupp # 25: frem
long tbl_unsupp - tbl_unsupp # 26: fscale
long fsglmul - tbl_unsupp # 27: fsglmul
long fsub - tbl_unsupp # 28: fsub
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp # 30: fsincos
long tbl_unsupp - tbl_unsupp # 31: fsincos
long tbl_unsupp - tbl_unsupp # 32: fsincos
long tbl_unsupp - tbl_unsupp # 33: fsincos
long tbl_unsupp - tbl_unsupp # 34: fsincos
long tbl_unsupp - tbl_unsupp # 35: fsincos
long tbl_unsupp - tbl_unsupp # 36: fsincos
long tbl_unsupp - tbl_unsupp # 37: fsincos
long fcmp - tbl_unsupp # 38: fcmp
long tbl_unsupp - tbl_unsupp
long ftst - tbl_unsupp # 3a: ftst
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long fsin - tbl_unsupp # 40: fsmove
long fssqrt - tbl_unsupp # 41: fssqrt
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long fdin - tbl_unsupp # 44: fdmove
long fdsqrt - tbl_unsupp # 45: fdsqrt
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long fsabs - tbl_unsupp # 58: fsabs
long tbl_unsupp - tbl_unsupp
long fsneg - tbl_unsupp # 5a: fsneg
long tbl_unsupp - tbl_unsupp
long fdabs - tbl_unsupp # 5c: fdabs
long tbl_unsupp - tbl_unsupp
long fdneg - tbl_unsupp # 5e: fdneg
long tbl_unsupp - tbl_unsupp
long fsdiv - tbl_unsupp # 60: fsdiv
long tbl_unsupp - tbl_unsupp
long fsadd - tbl_unsupp # 62: fsadd
long fsmul - tbl_unsupp # 63: fsmul
long fddiv - tbl_unsupp # 64: fddiv
long tbl_unsupp - tbl_unsupp
long fdadd - tbl_unsupp # 66: fdadd
long fdmul - tbl_unsupp # 67: fdmul
long fssub - tbl_unsupp # 68: fssub
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long tbl_unsupp - tbl_unsupp
long fdsub - tbl_unsupp # 6c: fdsub
#################################################
# Add this here so non-fp modules can compile.
# (smovcr is called from fpsp_inex.) global smovcr
smovcr:
bra.b smovcr
#########################################################################
# XDEF **************************************************************** #
# fmovm_dynamic(): emulate "fmovm" dynamic instruction #
# #
# XREF **************************************************************** #
# fetch_dreg() - fetch data register #
# {i,d,}mem_read() - fetch data from memory #
# _mem_write() - write data to memory #
# iea_iacc() - instruction memory access error occurred #
# iea_dacc() - data memory access error occurred #
# restore() - restore An index regs if access error occurred #
# #
# INPUT *************************************************************** #
# None #
# #
# OUTPUT ************************************************************** #
# If instr is "fmovm Dn,-(A7)" from supervisor mode, #
# d0 = size of dump #
# d1 = Dn #
# Else if instruction access error, #
# d0 = FSLW #
# Else if data access error, #
# d0 = FSLW #
# a0 = address of fault #
# Else #
# none. #
# #
# ALGORITHM *********************************************************** #
# The effective address must be calculated since this is entered #
# from an "Unimplemented Effective Address" exception handler. So, we #
# have our own fcalc_ea() routine here. If an access error is flagged #
# by a _{i,d,}mem_read() call, we must exit through the special #
# handler. #
# The data register is determined and its value loaded to get the #
# string of FP registers affected. This value is used as an index into #
# a lookup table such that we can determine the number of bytes #
# involved. #
# If the instruction is "fmovm.x <ea>,Dn", a _mem_read() is used #
# to read in all FP values. Again, _mem_read() may fail and require a #
# special exit. #
# If the instruction is "fmovm.x DN,<ea>", a _mem_write() is used #
# to write all FP values. _mem_write() may also fail. #
# If the instruction is "fmovm.x DN,-(a7)" from supervisor mode, #
# then we return the size of the dump and the string to the caller #
# so that the move can occur outside of this routine. This special #
# case is required so that moves to the system stack are handled #
# correctly. #
# #
# DYNAMIC: #
# fmovm.x dn, <ea> #
# fmovm.x <ea>, dn #
# #
# <WORD 1> <WORD2> #
# 1111 0010 00 |<ea>| 11@& 1000 0$$$ 0000 #
# #
# & = (0): predecrement addressing mode #
# (1): postincrement or control addressing mode #
# @ = (0): move listed regs from memory to the FPU #
# (1): move listed regs from the FPU to memory #
# $$$ : index of data register holding reg select mask #
# #
# NOTES: #
# If the data register holds a zero, then the #
# instruction is a nop. #
# #
#########################################################################
global fmovm_dynamic
fmovm_dynamic:
# extract the data register in which the bit string resides...
mov.b 1+EXC_EXTWORD(%a6),%d1 # fetch extword
andi.w &0x70,%d1 # extract reg bits
lsr.b &0x4,%d1 # shift into lo bits
# fetch the bit string into d0...
bsr.l fetch_dreg # fetch reg string
# if the bit string is a zero, then the operation is a no-op
# but, make sure that we've calculated ea and advanced the opword pointer
beq.w fmovm_data_done
# separate move ins from move outs...
btst &0x5,EXC_EXTWORD(%a6) # is it a move in or out?
beq.w fmovm_data_in # it's a move out
#############
# MOVE OUT: #
#############
fmovm_data_out:
btst &0x4,EXC_EXTWORD(%a6) # control or predecrement?
bne.w fmovm_out_ctrl # control
############################
fmovm_out_predec:
# for predecrement mode, the bit string is the opposite of both control
# operations and postincrement mode. (bit7 = FP7 ... bit0 = FP0)
# here, we convert it to be just like the others...
mov.b (tbl_fmovm_convert.w,%pc,%d1.w*1),%d1
btst &0x5,EXC_SR(%a6) # user or supervisor mode?
beq.b fmovm_out_ctrl # user
fmovm_out_predec_s:
cmpi.b SPCOND_FLG(%a6),&mda7_flg # is <ea> mode -(a7)?
bne.b fmovm_out_ctrl
# the operation was unfortunately an: fmovm.x dn,-(sp)
# called from supervisor mode.
# we're also passing "size" and "strg" back to the calling routine
rts
############################
fmovm_out_ctrl:
mov.l %a0,%a1 # move <ea> to a1
sub.l %d0,%sp # subtract size of dump
lea (%sp),%a0
tst.b %d1 # should FP0 be moved?
bpl.b fmovm_out_ctrl_fp1 # no
global fmovm_calc_ea
###############################################
# _fmovm_calc_ea: calculate effective address #
###############################################
fmovm_calc_ea:
mov.l %d0,%a0 # move # bytes to a0
# currently, MODE and REG are taken from the EXC_OPWORD. this could be
# easily changed if they were inputs passed in registers.
mov.w EXC_OPWORD(%a6),%d0 # fetch opcode word
mov.w %d0,%d1 # make a copy
andi.w &0x3f,%d0 # extract mode field
andi.l &0x7,%d1 # extract reg field
# jump to the corresponding function for each {MODE,REG} pair.
mov.w (tbl_fea_mode.b,%pc,%d0.w*2),%d0 # fetch jmp distance
jmp (tbl_fea_mode.b,%pc,%d0.w*1) # jmp to correct ea mode
swbeg &64
tbl_fea_mode:
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short faddr_ind_a0 - tbl_fea_mode
short faddr_ind_a1 - tbl_fea_mode
short faddr_ind_a2 - tbl_fea_mode
short faddr_ind_a3 - tbl_fea_mode
short faddr_ind_a4 - tbl_fea_mode
short faddr_ind_a5 - tbl_fea_mode
short faddr_ind_a6 - tbl_fea_mode
short faddr_ind_a7 - tbl_fea_mode
short faddr_ind_p_a0 - tbl_fea_mode
short faddr_ind_p_a1 - tbl_fea_mode
short faddr_ind_p_a2 - tbl_fea_mode
short faddr_ind_p_a3 - tbl_fea_mode
short faddr_ind_p_a4 - tbl_fea_mode
short faddr_ind_p_a5 - tbl_fea_mode
short faddr_ind_p_a6 - tbl_fea_mode
short faddr_ind_p_a7 - tbl_fea_mode
short faddr_ind_m_a0 - tbl_fea_mode
short faddr_ind_m_a1 - tbl_fea_mode
short faddr_ind_m_a2 - tbl_fea_mode
short faddr_ind_m_a3 - tbl_fea_mode
short faddr_ind_m_a4 - tbl_fea_mode
short faddr_ind_m_a5 - tbl_fea_mode
short faddr_ind_m_a6 - tbl_fea_mode
short faddr_ind_m_a7 - tbl_fea_mode
short faddr_ind_disp_a0 - tbl_fea_mode
short faddr_ind_disp_a1 - tbl_fea_mode
short faddr_ind_disp_a2 - tbl_fea_mode
short faddr_ind_disp_a3 - tbl_fea_mode
short faddr_ind_disp_a4 - tbl_fea_mode
short faddr_ind_disp_a5 - tbl_fea_mode
short faddr_ind_disp_a6 - tbl_fea_mode
short faddr_ind_disp_a7 - tbl_fea_mode
short faddr_ind_ext - tbl_fea_mode
short faddr_ind_ext - tbl_fea_mode
short faddr_ind_ext - tbl_fea_mode
short faddr_ind_ext - tbl_fea_mode
short faddr_ind_ext - tbl_fea_mode
short faddr_ind_ext - tbl_fea_mode
short faddr_ind_ext - tbl_fea_mode
short faddr_ind_ext - tbl_fea_mode
short fabs_short - tbl_fea_mode
short fabs_long - tbl_fea_mode
short fpc_ind - tbl_fea_mode
short fpc_ind_ext - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
short tbl_fea_mode - tbl_fea_mode
###################################
# Address register indirect: (An) #
###################################
faddr_ind_a0:
mov.l EXC_DREGS+0x8(%a6),%a0 # Get current a0
rts
faddr_ind_a1:
mov.l EXC_DREGS+0xc(%a6),%a0 # Get current a1
rts
faddr_ind_a2:
mov.l %a2,%a0 # Get current a2
rts
faddr_ind_a3:
mov.l %a3,%a0 # Get current a3
rts
faddr_ind_a4:
mov.l %a4,%a0 # Get current a4
rts
faddr_ind_a5:
mov.l %a5,%a0 # Get current a5
rts
faddr_ind_a6:
mov.l (%a6),%a0 # Get current a6
rts
faddr_ind_a7:
mov.l EXC_A7(%a6),%a0 # Get current a7
rts
#####################################################
# Address register indirect w/ postincrement: (An)+ #
#####################################################
faddr_ind_p_a0:
mov.l EXC_DREGS+0x8(%a6),%d0 # Get current a0
mov.l %d0,%d1
add.l %a0,%d1 # Increment
mov.l %d1,EXC_DREGS+0x8(%a6) # Save incr value
mov.l %d0,%a0
rts
faddr_ind_p_a1:
mov.l EXC_DREGS+0xc(%a6),%d0 # Get current a1
mov.l %d0,%d1
add.l %a0,%d1 # Increment
mov.l %d1,EXC_DREGS+0xc(%a6) # Save incr value
mov.l %d0,%a0
rts
faddr_ind_p_a2:
mov.l %a2,%d0 # Get current a2
mov.l %d0,%d1
add.l %a0,%d1 # Increment
mov.l %d1,%a2 # Save incr value
mov.l %d0,%a0
rts
faddr_ind_p_a3:
mov.l %a3,%d0 # Get current a3
mov.l %d0,%d1
add.l %a0,%d1 # Increment
mov.l %d1,%a3 # Save incr value
mov.l %d0,%a0
rts
faddr_ind_p_a4:
mov.l %a4,%d0 # Get current a4
mov.l %d0,%d1
add.l %a0,%d1 # Increment
mov.l %d1,%a4 # Save incr value
mov.l %d0,%a0
rts
faddr_ind_p_a5:
mov.l %a5,%d0 # Get current a5
mov.l %d0,%d1
add.l %a0,%d1 # Increment
mov.l %d1,%a5 # Save incr value
mov.l %d0,%a0
rts
faddr_ind_p_a6:
mov.l (%a6),%d0 # Get current a6
mov.l %d0,%d1
add.l %a0,%d1 # Increment
mov.l %d1,(%a6) # Save incr value
mov.l %d0,%a0
rts
faddr_ind_p_a7:
mov.b &mia7_flg,SPCOND_FLG(%a6) # set"special case" flag
mov.l EXC_A7(%a6),%d0 # Get current a7
mov.l %d0,%d1
add.l %a0,%d1 # Increment
mov.l %d1,EXC_A7(%a6) # Save incr value
mov.l %d0,%a0
rts
####################################################
# Address register indirect w/ predecrement: -(An) #
####################################################
faddr_ind_m_a0:
mov.l EXC_DREGS+0x8(%a6),%d0 # Get current a0 sub.l %a0,%d0 # Decrement
mov.l %d0,EXC_DREGS+0x8(%a6) # Save decr value
mov.l %d0,%a0
rts
faddr_ind_m_a1:
mov.l EXC_DREGS+0xc(%a6),%d0 # Get current a1 sub.l %a0,%d0 # Decrement
mov.l %d0,EXC_DREGS+0xc(%a6) # Save decr value
mov.l %d0,%a0
rts
faddr_ind_m_a2:
mov.l %a2,%d0 # Get current a2 sub.l %a0,%d0 # Decrement
mov.l %d0,%a2 # Save decr value
mov.l %d0,%a0
rts
faddr_ind_m_a3:
mov.l %a3,%d0 # Get current a3 sub.l %a0,%d0 # Decrement
mov.l %d0,%a3 # Save decr value
mov.l %d0,%a0
rts
faddr_ind_m_a4:
mov.l %a4,%d0 # Get current a4 sub.l %a0,%d0 # Decrement
mov.l %d0,%a4 # Save decr value
mov.l %d0,%a0
rts
faddr_ind_m_a5:
mov.l %a5,%d0 # Get current a5 sub.l %a0,%d0 # Decrement
mov.l %d0,%a5 # Save decr value
mov.l %d0,%a0
rts
faddr_ind_m_a6:
mov.l (%a6),%d0 # Get current a6 sub.l %a0,%d0 # Decrement
mov.l %d0,(%a6) # Save decr value
mov.l %d0,%a0
rts
faddr_ind_m_a7:
mov.b &mda7_flg,SPCOND_FLG(%a6) # set"special case" flag
mov.l EXC_A7(%a6),%d0 # Get current a7 sub.l %a0,%d0 # Decrement
mov.l %d0,EXC_A7(%a6) # Save decr value
mov.l %d0,%a0
rts
mov.l %d0,%d1
rol.w &0x4,%d1
andi.w &0xf,%d1 # extract index regno
# count on fetch_dreg() not to alter a0...
bsr.l fetch_dreg # fetch index
mov.l %d2,-(%sp) # save d2
mov.l L_SCR1(%a6),%d2 # fetch opword
btst &0xb,%d2 # is it word or long?
bne.b faii8_long
ext.l %d0 # sign extend word index
faii8_long:
mov.l %d2,%d1
rol.w &0x7,%d1
andi.l &0x3,%d1 # extract scale value
lsl.l %d1,%d0 # shift index by scale
extb.l %d2 # sign extend displacement
add.l %d2,%d0 # index + disp
add.l %d0,%a0 # An + (index + disp)
mov.l EXC_EXTWPTR(%a6),%a0 # put base in a0
subq.l &0x2,%a0 # adjust base
btst &0x8,%d0 # is disp only 8 bits?
bne.w fcalc_mem_ind # calc memory indirect
mov.l %d0,L_SCR1(%a6) # store opword
mov.l %d0,%d1 # make extword copy
rol.w &0x4,%d1 # rotate reg num into place
andi.w &0xf,%d1 # extract register number
# count on fetch_dreg() not to alter a0...
bsr.l fetch_dreg # fetch index
mov.l %d2,-(%sp) # save d2
mov.l L_SCR1(%a6),%d2 # fetch opword
btst &0xb,%d2 # is index word or long?
bne.b fpii8_long # long
ext.l %d0 # sign extend word index
fpii8_long:
mov.l %d2,%d1
rol.w &0x7,%d1 # rotate scale value into place
andi.l &0x3,%d1 # extract scale value
lsl.l %d1,%d0 # shift index by scale
extb.l %d2 # sign extend displacement
add.l %d2,%d0 # disp + index
add.l %d0,%a0 # An + (index + disp)
mov.l (%sp)+,%d2 # restore temp register
rts
# d2 = index
# d3 = base
# d4 = od
# d5 = extword
fcalc_mem_ind:
btst &0x6,%d0 # is the index suppressed?
beq.b fcalc_index
movm.l &0x3c00,-(%sp) # save d2-d5
mov.l %d0,%d5 # put extword in d5
mov.l %a0,%d3 # put base in d3
clr.l %d2 # yes, so index = 0
bra.b fbase_supp_ck
# index:
fcalc_index:
mov.l %d0,L_SCR1(%a6) # save d0 (opword)
bfextu %d0{&16:&4},%d1 # fetch dreg index
bsr.l fetch_dreg
movm.l &0x3c00,-(%sp) # save d2-d5
mov.l %d0,%d2 # put index in d2
mov.l L_SCR1(%a6),%d5
mov.l %a0,%d3
btst &0xb,%d5 # is index word or long?
bne.b fno_ext
ext.l %d2
fno_ext:
bfextu %d5{&21:&2},%d0
lsl.l %d0,%d2
# base address (passed as parameter in d3):
# we clear the value here if it should actually be suppressed.
fbase_supp_ck:
btst &0x7,%d5 # is the bd suppressed?
beq.b fno_base_sup clr.l %d3
# base displacement:
fno_base_sup:
bfextu %d5{&26:&2},%d0 # get bd size
# beq.l fmovm_error # if (size == 0) it's reserved
mov.l %d0,USER_FPCR(%a6) # store new FPCR to mem
mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr
addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr
bsr.l _imem_read_long # fetch FPSR from mem
tst.l %d1 # did ifetch fail?
bne.l iea_iacc # yes
mov.l %d0,USER_FPSR(%a6) # store new FPSR to mem
rts
mov.l %d0,USER_FPCR(%a6) # store new FPCR to mem
mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr
addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr
bsr.l _imem_read_long # fetch FPSR from mem
tst.l %d1 # did ifetch fail?
bne.l iea_iacc # yes
mov.l %d0,USER_FPSR(%a6) # store new FPSR to mem
mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr
addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr
bsr.l _imem_read_long # fetch FPIAR from mem
tst.l %d1 # did ifetch fail?
bne.l iea_iacc # yes
mov.l %d0,USER_FPIAR(%a6) # store new FPIAR to mem
rts
#########################################################################
# XDEF **************************************************************** #
# addsub_scaler2(): scale inputs to fadd/fsub such that no #
# OVFL/UNFL exceptions will result #
# #
# XREF **************************************************************** #
# norm() - normalize mantissa after adjusting exponent #
# #
# INPUT *************************************************************** #
# FP_SRC(a6) = fp op1(src) #
# FP_DST(a6) = fp op2(dst) #
# #
# OUTPUT ************************************************************** #
# FP_SRC(a6) = fp op1 scaled(src) #
# FP_DST(a6) = fp op2 scaled(dst) #
# d0 = scale amount #
# #
# ALGORITHM *********************************************************** #
# If the DST exponent is > the SRC exponent, set the DST exponent #
# equal to 0x3fff and scale the SRC exponent by the value that the #
# DST exponent was scaled by. If the SRC exponent is greater or equal, #
# do the opposite. Return this scale factor in d0. #
# If the two exponents differ by > the number of mantissa bits #
# plus two, then set the smallest exponent to a very small value as a #
# quick shortcut. #
# #
#########################################################################
quick_scale12:
andi.w &0x8000,FP_SCR0_EX(%a6) # zero src exponent
bset &0x0,1+FP_SCR0_EX(%a6) # set exp = 1
mov.l (%sp)+,%d0 # return SCALE factor
rts
# src exp is >= dst exp; scale src to exp = 0x3fff
src_exp_ge2:
bsr.l scale_to_zero_src
mov.l %d0,-(%sp) # save scale factor
cmpi.b DTAG(%a6),&DENORM # is dst denormalized?
bne.b cmpexp22
lea FP_SCR1(%a6),%a0
bsr.l norm # normalize the denorm; result is new exp
neg.w %d0 # new exp = -(shft val)
mov.w %d0,2+L_SCR1(%a6) # inset new exp
#########################################################################
# XDEF **************************************************************** #
# scale_to_zero_src(): scale the exponent of extended precision #
# value at FP_SCR0(a6). #
# #
# XREF **************************************************************** #
# norm() - normalize the mantissa if the operand was a DENORM #
# #
# INPUT *************************************************************** #
# FP_SCR0(a6) = extended precision operand to be scaled #
# #
# OUTPUT ************************************************************** #
# FP_SCR0(a6) = scaled extended precision operand #
# d0 = scale value #
# #
# ALGORITHM *********************************************************** #
# Set the exponent of the input operand to 0x3fff. Save the value #
# of the difference between the original and new exponent. Then, #
# normalize the operand if it was a DENORM. Add this normalization #
# value to the previous value. Return the result. #
# #
#########################################################################
global scale_to_zero_src
scale_to_zero_src:
mov.w FP_SCR0_EX(%a6),%d1 # extract operand's {sgn,exp}
mov.w %d1,%d0 # make a copy
stzs_denorm:
lea FP_SCR0(%a6),%a0 # pass ptr to src op
bsr.l norm # normalize denorm
neg.l %d0 # new exponent = -(shft val)
mov.l %d0,%d1 # prepare for op_norm call
bra.b stzs_norm # finish scaling
###
#########################################################################
# XDEF **************************************************************** #
# scale_sqrt(): scale the input operand exponent so a subsequent #
# fsqrt operation won't take an exception. #
# #
# XREF **************************************************************** #
# norm() - normalize the mantissa if the operand was a DENORM #
# #
# INPUT *************************************************************** #
# FP_SCR0(a6) = extended precision operand to be scaled #
# #
# OUTPUT ************************************************************** #
# FP_SCR0(a6) = scaled extended precision operand #
# d0 = scale value #
# #
# ALGORITHM *********************************************************** #
# If the input operand is a DENORM, normalize it. #
# If the exponent of the input operand is even, set the exponent #
# to 0x3ffe and return a scale factor of "(exp-0x3ffe)/2". If the #
# exponent of the input operand is off, set the exponent to ox3fff and #
# return a scale factor of "(exp-0x3fff)/2". #
# #
#########################################################################
global scale_sqrt
scale_sqrt:
cmpi.b STAG(%a6),&DENORM # is operand normalized?
beq.b ss_denorm # normalize the DENORM
#########################################################################
# XDEF **************************************************************** #
# scale_to_zero_dst(): scale the exponent of extended precision #
# value at FP_SCR1(a6). #
# #
# XREF **************************************************************** #
# norm() - normalize the mantissa if the operand was a DENORM #
# #
# INPUT *************************************************************** #
# FP_SCR1(a6) = extended precision operand to be scaled #
# #
# OUTPUT ************************************************************** #
# FP_SCR1(a6) = scaled extended precision operand #
# d0 = scale value #
# #
# ALGORITHM *********************************************************** #
# Set the exponent of the input operand to 0x3fff. Save the value #
# of the difference between the original and new exponent. Then, #
# normalize the operand if it was a DENORM. Add this normalization #
# value to the previous value. Return the result. #
# #
#########################################################################
global scale_to_zero_dst
scale_to_zero_dst:
mov.w FP_SCR1_EX(%a6),%d1 # extract operand's {sgn,exp}
mov.w %d1,%d0 # make a copy
#########################################################################
# XDEF **************************************************************** #
# res_qnan(): return default result w/ QNAN operand for dyadic #
# res_snan(): return default result w/ SNAN operand for dyadic #
# res_qnan_1op(): return dflt result w/ QNAN operand for monadic #
# res_snan_1op(): return dflt result w/ SNAN operand for monadic #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# FP_SRC(a6) = pointer to extended precision src operand #
# FP_DST(a6) = pointer to extended precision dst operand #
# #
# OUTPUT ************************************************************** #
# fp0 = default result #
# #
# ALGORITHM *********************************************************** #
# If either operand (but not both operands) of an operation is a #
# nonsignalling NAN, then that NAN is returned as the result. If both #
# operands are nonsignalling NANs, then the destination operand #
# nonsignalling NAN is returned as the result. #
# If either operand to an operation is a signalling NAN (SNAN), #
# then, the SNAN bit is set in the FPSR EXC byte. If the SNAN trap #
# enable bit is set in the FPCR, then the trap is taken and the #
# destination is not modified. If the SNAN trap enable bit is not set, #
# then the SNAN is converted to a nonsignalling NAN (by setting the #
# SNAN bit in the operand to one), and the operation continues as #
# described in the preceding paragraph, for nonsignalling NANs. #
# Make sure the appropriate FPSR bits are set before exiting. #
# #
#########################################################################
global res_qnan global res_snan
res_qnan:
res_snan:
cmp.b DTAG(%a6), &SNAN # is the dst an SNAN?
beq.b dst_snan2
cmp.b DTAG(%a6), &QNAN # is the dst a QNAN?
beq.b dst_qnan2
src_nan:
cmp.b STAG(%a6), &QNAN
beq.b src_qnan2 global res_snan_1op
res_snan_1op:
src_snan2:
bset &0x6, FP_SRC_HI(%a6) # set SNAN bit
or.l &nan_mask+aiop_mask+snan_mask, USER_FPSR(%a6)
lea FP_SRC(%a6), %a0
bra.b nan_comp global res_qnan_1op
res_qnan_1op:
src_qnan2:
or.l &nan_mask, USER_FPSR(%a6)
lea FP_SRC(%a6), %a0
bra.b nan_comp
dst_snan2:
or.l &nan_mask+aiop_mask+snan_mask, USER_FPSR(%a6)
bset &0x6, FP_DST_HI(%a6) # set SNAN bit
lea FP_DST(%a6), %a0
bra.b nan_comp
dst_qnan2:
lea FP_DST(%a6), %a0
cmp.b STAG(%a6), &SNAN
bne nan_done
or.l &aiop_mask+snan_mask, USER_FPSR(%a6)
nan_done:
or.l &nan_mask, USER_FPSR(%a6)
nan_comp:
btst &0x7, FTEMP_EX(%a0) # is NAN neg?
beq.b nan_not_neg
or.l &neg_mask, USER_FPSR(%a6)
nan_not_neg:
fmovm.x (%a0), &0x80
rts
#########################################################################
# XDEF **************************************************************** #
# res_operr(): return default result during operand error #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# None #
# #
# OUTPUT ************************************************************** #
# fp0 = default operand error result #
# #
# ALGORITHM *********************************************************** #
# An nonsignalling NAN is returned as the default result when #
# an operand error occurs for the following cases: #
# #
# Multiply: (Infinity x Zero) #
# Divide : (Zero / Zero) || (Infinity / Infinity) #
# #
#########################################################################
global res_operr
res_operr:
or.l &nan_mask+operr_mask+aiop_mask, USER_FPSR(%a6)
fmovm.x nan_return(%pc), &0x80
rts
nan_return:
long 0x7fff0000, 0xffffffff, 0xffffffff
#########################################################################
# XDEF **************************************************************** #
# _denorm(): denormalize an intermediate result #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# a0 = points to the operand to be denormalized #
# (in the internal extended format) #
# #
# d0 = rounding precision #
# #
# OUTPUT ************************************************************** #
# a0 = pointer to the denormalized result #
# (in the internal extended format) #
# #
# d0 = guard,round,sticky #
# #
# ALGORITHM *********************************************************** #
# According to the exponent underflow threshold for the given #
# precision, shift the mantissa bits to the right in order raise the #
# exponent of the operand to the threshold value. While shifting the #
# mantissa bits right, maintain the value of the guard, round, and #
# sticky bits. #
# other notes: #
# (1) _denorm() is called by the underflow routines #
# (2) _denorm() does NOT affect the status register #
# #
#########################################################################
#
# table of exponent threshold values for each precision
#
tbl_thresh:
short 0x0
short sgl_thresh
short dbl_thresh
global _denorm
_denorm:
#
# Load the exponent threshold for the precision selected and check
# to see if (threshold - exponent) is > 65 in which case we can
# simply calculate the sticky bit and zero the mantissa. otherwise
# we have to call the denormalization routine.
#
lsr.b &0x2, %d0 # shift prec to lo bits
mov.w (tbl_thresh.b,%pc,%d0.w*2), %d1 # load prec threshold
mov.w %d1, %d0 # copy d1 into d0 sub.w FTEMP_EX(%a0), %d0 # diff = threshold - exp
cmpi.w %d0, &66 # is diff > 65? (mant + g,r bits)
bpl.b denorm_set_stky # yes; just calc sticky
clr.l %d0 # clear g,r,s
btst &inex2_bit, FPSR_EXCEPT(%a6) # yes; was INEX2 set?
beq.b denorm_call # no; don't change anything
bset &29, %d0 # yes; set sticky bit
denorm_call:
bsr.l dnrm_lp # denormalize the number
rts
#
# all bit would have been shifted off during the denorm so simply
# calculate if the sticky should be set and clear the entire mantissa.
#
denorm_set_stky:
mov.l &0x20000000, %d0 # set sticky bit in return value
mov.w %d1, FTEMP_EX(%a0) # load exp with threshold clr.l FTEMP_HI(%a0) # set d1 = 0 (ms mantissa) clr.l FTEMP_LO(%a0) # set d2 = 0 (ms mantissa)
rts
# #
# dnrm_lp(): normalize exponent/mantissa to specified threshold #
# #
# INPUT: #
# %a0 : points to the operand to be denormalized #
# %d0{31:29} : initial guard,round,sticky #
# %d1{15:0} : denormalization threshold #
# OUTPUT: #
# %a0 : points to the denormalized operand #
# %d0{31:29} : final guard,round,sticky #
# #
# *** Local Equates *** # set GRS, L_SCR2 # g,r,s temp storage set FTEMP_LO2, L_SCR1 # FTEMP_LO copy
global dnrm_lp
dnrm_lp:
#
# make a copy of FTEMP_LO and place the g,r,s bits directly after it
# in memory so as to make the bitfield extraction for denormalization easier.
#
mov.l FTEMP_LO(%a0), FTEMP_LO2(%a6) # make FTEMP_LO copy
mov.l %d0, GRS(%a6) # place g,r,s after it
#
# check to see how much less than the underflow threshold the operand
# exponent is.
#
mov.l %d1, %d0 # copy the denorm threshold sub.w FTEMP_EX(%a0), %d1 # d1 = threshold - uns exponent
ble.b dnrm_no_lp # d1 <= 0
cmpi.w %d1, &0x20 # is ( 0 <= d1 < 32) ?
blt.b case_1 # yes
cmpi.w %d1, &0x40 # is (32 <= d1 < 64) ?
blt.b case_2 # yes
bra.w case_3 # (d1 >= 64)
#
# No normalization necessary
#
dnrm_no_lp:
mov.l GRS(%a6), %d0 # restore original g,r,s
rts
mov.w %d0, FTEMP_EX(%a0) # exponent = denorm threshold
subi.w &0x20, %d1 # %d1 now between 0 and 32
mov.l &0x20, %d0 sub.w %d1, %d0 # %d0 = 32 - %d1
# subtle step here; or in the g,r,s at the bottom of FTEMP_LO to minimize
# the number of bits to check for the sticky detect.
# it only plays a role in shift amounts of 61-63.
mov.b GRS(%a6), %d2
or.b %d2, 3+FTEMP_LO2(%a6)
bfextu FTEMP_HI(%a0){&0:%d0}, %d2 # %d2 = new FTEMP_LO
bfextu FTEMP_HI(%a0){%d0:&32}, %d1 # %d1 = new G,R,S
bftst %d1{&2:&30} # were any bits shifted off?
bne.b case2_set_sticky # yes; set sticky bit
bftst FTEMP_LO2(%a6){%d0:&31} # were any bits shifted off?
bne.b case2_set_sticky # yes; set sticky bit
mov.l %d1, %d0 # move new G,R,S to %d0
bra.b case2_end
case2_set_sticky:
mov.l %d1, %d0 # move new G,R,S to %d0
bset &rnd_stky_bit, %d0 # set sticky bit
case2_end: clr.l FTEMP_HI(%a0) # store FTEMP_HI = 0
mov.l %d2, FTEMP_LO(%a0) # store FTEMP_LO
and.l &0xe0000000, %d0 # clear all but G,R,S
mov.l (%sp)+,%d2 # restore temp register
rts
#
# case (d1>=64)
#
# %d0 = denorm threshold
# %d1 = amt to shift
#
case_3:
mov.w %d0, FTEMP_EX(%a0) # insert denorm threshold
#
# case (d1>65)
#
# Shift value is > 65 and out of range. All bits are shifted off.
# Return a zero mantissa with the sticky bit set
# clr.l FTEMP_HI(%a0) # clear hi(mantissa) clr.l FTEMP_LO(%a0) # clear lo(mantissa)
mov.l &0x20000000, %d0 # set sticky bit
rts
#
# case (d1 == 65)
#
# ---------------------------------------------------------
# | FTEMP_HI | FTEMP_LO |grs000.........000|
# ---------------------------------------------------------
# <-------(32)------>
# \ \
# \ \
# \ \
# \ ------------------------------
# -------------------------------- \
# \ \
# \ \
# \ \
# <-------(31)----->
# ---------------------------------------------------------
# |0...............0|0................0|0rs |
# ---------------------------------------------------------
#
case3_65:
mov.l FTEMP_HI(%a0), %d0 # fetch hi(mantissa)
and.l &0x80000000, %d0 # extract R bit
lsr.l &0x1, %d0 # shift high bit into R bit
and.l &0x7fffffff, %d1 # extract other bits
case3_complete:
# last operation done was an "and" of the bits shifted off so the condition
# codes are already set so branch accordingly.
bne.b case3_set_sticky # yes; go set new sticky
tst.l FTEMP_LO(%a0) # were any bits shifted off?
bne.b case3_set_sticky # yes; go set new sticky
tst.b GRS(%a6) # were any bits shifted off?
bne.b case3_set_sticky # yes; go set new sticky
#
# no bits were shifted off so don't set the sticky bit.
# the guard and
# the entire mantissa is zero.
# clr.l FTEMP_HI(%a0) # clear hi(mantissa) clr.l FTEMP_LO(%a0) # clear lo(mantissa)
rts
#
# some bits were shifted off so set the sticky bit.
# the entire mantissa is zero.
#
case3_set_sticky:
bset &rnd_stky_bit,%d0 # set new sticky bit clr.l FTEMP_HI(%a0) # clear hi(mantissa) clr.l FTEMP_LO(%a0) # clear lo(mantissa)
rts
#########################################################################
# XDEF **************************************************************** #
# _round(): round result according to precision/mode #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# a0 = ptr to input operand in internal extended format #
# d1(hi) = contains rounding precision: #
# ext = $0000xxxx #
# sgl = $0004xxxx #
# dbl = $0008xxxx #
# d1(lo) = contains rounding mode: #
# RN = $xxxx0000 #
# RZ = $xxxx0001 #
# RM = $xxxx0002 #
# RP = $xxxx0003 #
# d0{31:29} = contains the g,r,s bits (extended) #
# #
# OUTPUT ************************************************************** #
# a0 = pointer to rounded result #
# #
# ALGORITHM *********************************************************** #
# On return the value pointed to by a0 is correctly rounded, #
# a0 is preserved and the g-r-s bits in d0 are cleared. #
# The result is not typed - the tag field is invalid. The #
# result is still in the internal extended format. #
# #
# The INEX bit of USER_FPSR will be set if the rounded result was #
# inexact (i.e. if any of the g-r-s bits were set). #
# #
#########################################################################
global _round
_round:
#
# ext_grs() looks at the rounding precision and sets the appropriate
# G,R,S bits.
# If (G,R,S == 0) then result is exact and round is done, else set
# the inex flag in status reg and continue.
#
bsr.l ext_grs # extract G,R,S
tst.l %d0 # are G,R,S zero?
beq.w truncate # yes; round is complete
or.w &inx2a_mask, 2+USER_FPSR(%a6) # set inex2/ainex
#
# Use rounding mode as an index into a jump table for these modes.
# All of the following assumes grs != 0.
#
mov.w (tbl_mode.b,%pc,%d1.w*2), %a1 # load jump offset
jmp (tbl_mode.b,%pc,%a1) # jmp to rnd mode handler
tbl_mode:
short rnd_near - tbl_mode
short truncate - tbl_mode # RZ always truncates
short rnd_mnus - tbl_mode
short rnd_plus - tbl_mode
#################################################################
# ROUND PLUS INFINITY #
# #
# If sign of fp number = 0 (positive), then add 1 to l. #
#################################################################
rnd_plus:
tst.b FTEMP_SGN(%a0) # check for sign
bmi.w truncate # if positive then truncate
mov.l &0xffffffff, %d0 # force g,r,s to be all f's
swap %d1 # set up d1 for round prec.
#################################################################
# ROUND MINUS INFINITY #
# #
# If sign of fp number = 1 (negative), then add 1 to l. #
#################################################################
rnd_mnus:
tst.b FTEMP_SGN(%a0) # check for sign
bpl.w truncate # if negative then truncate
mov.l &0xffffffff, %d0 # force g,r,s to be all f's
swap %d1 # set up d1 for round prec.
#################################################################
# ROUND NEAREST #
# #
# If (g=1), then add 1 to l and if (r=s=0), then clear l #
# Note that this will round to even in case of a tie. #
#################################################################
rnd_near:
asl.l &0x1, %d0 # shift g-bit to c-bit
bcc.w truncate # if (g=1) then
# *** LOCAL EQUATES *** set ad_1_sgl, 0x00000100 # constant to add 1 to l-bit in sgl prec set ad_1_dbl, 0x00000800 # constant to add 1 to l-bit in dbl prec
#########################
# ADD SINGLE #
#########################
add_sgl:
add.l &ad_1_sgl, FTEMP_HI(%a0)
bcc.b scc_clr # no mantissa overflow
roxr.w FTEMP_HI(%a0) # shift v-bit back in
roxr.w FTEMP_HI+2(%a0) # shift v-bit back in
add.w &0x1, FTEMP_EX(%a0) # and incr exponent
scc_clr:
tst.l %d0 # test for rs = 0
bne.b sgl_done
and.w &0xfe00, FTEMP_HI+2(%a0) # clear the l-bit
sgl_done:
and.l &0xffffff00, FTEMP_HI(%a0) # truncate bits beyond sgl limit clr.l FTEMP_LO(%a0) # clear d2
rts
#########################
# ADD EXTENDED #
#########################
add_ext:
addq.l &1,FTEMP_LO(%a0) # add 1 to l-bit
bcc.b xcc_clr # test for carry out
addq.l &1,FTEMP_HI(%a0) # propagate carry
bcc.b xcc_clr
roxr.w FTEMP_HI(%a0) # mant is 0 so restore v-bit
roxr.w FTEMP_HI+2(%a0) # mant is 0 so restore v-bit
roxr.w FTEMP_LO(%a0)
roxr.w FTEMP_LO+2(%a0)
add.w &0x1,FTEMP_EX(%a0) # and inc exp
xcc_clr:
tst.l %d0 # test rs = 0
bne.b add_ext_done
and.b &0xfe,FTEMP_LO+3(%a0) # clear the l bit
add_ext_done:
rts
#
# ext_grs(): extract guard, round and sticky bits according to
# rounding precision.
#
# INPUT
# d0 = extended precision g,r,s (in d0{31:29})
# d1 = {PREC,ROUND}
# OUTPUT
# d0{31:29} = guard, round, sticky
#
# The ext_grs extract the guard/round/sticky bits according to the
# selected rounding precision. It is called by the round subroutine
# only. All registers except d0 are kept intact. d0 becomes an
# updated guard,round,sticky in d0{31:29}
#
# Notes: the ext_grs uses the round PREC, and therefore has to swap d1
# prior to usage, and needs to restore d1 to original. this
# routine is tightly tied to the round routine and not meant to
# uphold standard subroutine calling practices.
#
ext_grs:
swap %d1 # have d1.w point to round precision
tst.b %d1 # is rnd prec = extended?
bne.b ext_grs_not_ext # no; go handle sgl or dbl
#
# %d0 actually already hold g,r,s since _round() had it before calling
# this function. so, as long as we don't disturb it, we are "returning" it.
#
ext_grs_ext:
swap %d1 # yes; return to correct positions
rts
ext_grs_not_ext:
movm.l &0x3000, -(%sp) # make some temp registers {d2/d3}
cmpi.b %d1, &s_mode # is rnd prec = sgl?
bne.b ext_grs_dbl # no; go handle dbl
#
# sgl:
# 96 64 40 32 0
# -----------------------------------------------------
# | EXP |XXXXXXX| |xx | |grs|
# -----------------------------------------------------
# <--(24)--->nn\ /
# ee ---------------------
# ww |
# v
# gr new sticky
#
ext_grs_sgl:
bfextu FTEMP_HI(%a0){&24:&2}, %d3 # sgl prec. g-r are 2 bits right
mov.l &30, %d2 # of the sgl prec. limits
lsl.l %d2, %d3 # shift g-r bits to MSB of d3
mov.l FTEMP_HI(%a0), %d2 # get word 2 for s-bit test
and.l &0x0000003f, %d2 # s bit is the or of all other
bne.b ext_grs_st_stky # bits to the right of g-r
tst.l FTEMP_LO(%a0) # test lower mantissa
bne.b ext_grs_st_stky # if any are set, set sticky
tst.l %d0 # test original g,r,s
bne.b ext_grs_st_stky # if any are set, set sticky
bra.b ext_grs_end_sd # if words 3 and 4 are clr, exit
#
# dbl:
# 96 64 32 11 0
# -----------------------------------------------------
# | EXP |XXXXXXX| | |xx |grs|
# -----------------------------------------------------
# nn\ /
# ee -------
# ww |
# v
# gr new sticky
#
ext_grs_dbl:
bfextu FTEMP_LO(%a0){&21:&2}, %d3 # dbl-prec. g-r are 2 bits right
mov.l &30, %d2 # of the dbl prec. limits
lsl.l %d2, %d3 # shift g-r bits to the MSB of d3
mov.l FTEMP_LO(%a0), %d2 # get lower mantissa for s-bit test
and.l &0x000001ff, %d2 # s bit is the or-ing of all
bne.b ext_grs_st_stky # other bits to the right of g-r
tst.l %d0 # test word original g,r,s
bne.b ext_grs_st_stky # if any are set, set sticky
bra.b ext_grs_end_sd # if clear, exit
ext_grs_st_stky:
bset &rnd_stky_bit, %d3 # set sticky bit
ext_grs_end_sd:
mov.l %d3, %d0 # return grs to d0
#########################################################################
# unnorm_fix(): - changes an UNNORM to one of NORM, DENORM, or ZERO #
# - returns corresponding optype tag #
# #
# XDEF **************************************************************** #
# unnorm_fix() #
# #
# XREF **************************************************************** #
# norm() - normalize the mantissa #
# #
# INPUT *************************************************************** #
# a0 = pointer to unnormalized extended precision number #
# #
# OUTPUT ************************************************************** #
# d0 = optype tag - is corrected to one of NORM, DENORM, or ZERO #
# a0 = input operand has been converted to a norm, denorm, or #
# zero; both the exponent and mantissa are changed. #
# #
#########################################################################
global unnorm_fix
unnorm_fix:
bfffo FTEMP_HI(%a0){&0:&32}, %d0 # how many shifts are needed?
bne.b unnorm_shift # hi(man) is not all zeroes
#
# hi(man) is all zeroes so see if any bits in lo(man) are set
#
unnorm_chk_lo:
bfffo FTEMP_LO(%a0){&0:&32}, %d0 # is operand really a zero?
beq.w unnorm_zero # yes
add.w &32, %d0 # no; fix shift distance
#
# d0 = # shifts needed for complete normalization
#
unnorm_shift: clr.l %d1 # clear top word
mov.w FTEMP_EX(%a0), %d1 # extract exponent
and.w &0x7fff, %d1 # strip off sgn
cmp.w %d0, %d1 # will denorm push exp < 0?
bgt.b unnorm_nrm_zero # yes; denorm only until exp = 0
#
# exponent would not go < 0. Therefore, number stays normalized
# sub.w %d0, %d1 # shift exponent value
mov.w FTEMP_EX(%a0), %d0 # load old exponent
and.w &0x8000, %d0 # save old sign
or.w %d0, %d1 # {sgn,new exp}
mov.w %d1, FTEMP_EX(%a0) # insert new exponent
bsr.l norm # normalize UNNORM
mov.b &NORM, %d0 # return new optype tag
rts
#
# exponent would go < 0, so only denormalize until exp = 0
#
unnorm_nrm_zero:
cmp.b %d1, &32 # is exp <= 32?
bgt.b unnorm_nrm_zero_lrg # no; go handle large exponent
bfextu FTEMP_HI(%a0){%d1:&32}, %d0 # extract new hi(man)
mov.l %d0, FTEMP_HI(%a0) # save new hi(man)
mov.l FTEMP_LO(%a0), %d0 # fetch old lo(man)
lsl.l %d1, %d0 # extract new lo(man)
mov.l %d0, FTEMP_LO(%a0) # save new lo(man)
and.w &0x8000, FTEMP_EX(%a0) # set exp = 0
mov.b &DENORM, %d0 # return new optype tag
rts
#
# only mantissa bits set are in lo(man)
#
unnorm_nrm_zero_lrg: sub.w &32, %d1 # adjust shft amt by 32
mov.l FTEMP_LO(%a0), %d0 # fetch old lo(man)
lsl.l %d1, %d0 # left shift lo(man)
mov.l %d0, FTEMP_HI(%a0) # store new hi(man) clr.l FTEMP_LO(%a0) # lo(man) = 0
and.w &0x8000, FTEMP_EX(%a0) # set exp = 0
mov.b &DENORM, %d0 # return new optype tag
rts
#
# whole mantissa is zero so this UNNORM is actually a zero
#
unnorm_zero:
and.w &0x8000, FTEMP_EX(%a0) # force exponent to zero
mov.b &ZERO, %d0 # fix optype tag
rts
#########################################################################
# XDEF **************************************************************** #
# set_tag_x(): return the optype of the input ext fp number #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision operand #
# #
# OUTPUT ************************************************************** #
# d0 = value of type tag #
# one of: NORM, INF, QNAN, SNAN, DENORM, UNNORM, ZERO #
# #
# ALGORITHM *********************************************************** #
# Simply test the exponent, j-bit, and mantissa values to #
# determine the type of operand. #
# If it's an unnormalized zero, alter the operand and force it #
# to be a normal zero. #
# #
#########################################################################
global set_tag_x
set_tag_x:
mov.w FTEMP_EX(%a0), %d0 # extract exponent
andi.w &0x7fff, %d0 # strip off sign
cmpi.w %d0, &0x7fff # is (EXP == MAX)?
beq.b inf_or_nan_x
not_inf_or_nan_x:
btst &0x7,FTEMP_HI(%a0)
beq.b not_norm_x
is_norm_x:
mov.b &NORM, %d0
rts
not_norm_x:
tst.w %d0 # is exponent = 0?
bne.b is_unnorm_x
not_unnorm_x:
tst.l FTEMP_HI(%a0)
bne.b is_denorm_x
tst.l FTEMP_LO(%a0)
bne.b is_denorm_x
is_zero_x:
mov.b &ZERO, %d0
rts
is_denorm_x:
mov.b &DENORM, %d0
rts
# must distinguish now "Unnormalized zeroes" which we
# must convert to zero.
is_unnorm_x:
tst.l FTEMP_HI(%a0)
bne.b is_unnorm_reg_x
tst.l FTEMP_LO(%a0)
bne.b is_unnorm_reg_x
# it's an "unnormalized zero". let's convert it to an actual zero...
andi.w &0x8000,FTEMP_EX(%a0) # clear exponent
mov.b &ZERO, %d0
rts
is_unnorm_reg_x:
mov.b &UNNORM, %d0
rts
inf_or_nan_x:
tst.l FTEMP_LO(%a0)
bne.b is_nan_x
mov.l FTEMP_HI(%a0), %d0
and.l &0x7fffffff, %d0 # msb is a don't care!
bne.b is_nan_x
is_inf_x:
mov.b &INF, %d0
rts
is_nan_x:
btst &0x6, FTEMP_HI(%a0)
beq.b is_snan_x
mov.b &QNAN, %d0
rts
is_snan_x:
mov.b &SNAN, %d0
rts
#########################################################################
# XDEF **************************************************************** #
# set_tag_d(): return the optype of the input dbl fp number #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# a0 = points to double precision operand #
# #
# OUTPUT ************************************************************** #
# d0 = value of type tag #
# one of: NORM, INF, QNAN, SNAN, DENORM, ZERO #
# #
# ALGORITHM *********************************************************** #
# Simply test the exponent, j-bit, and mantissa values to #
# determine the type of operand. #
# #
#########################################################################
global set_tag_d
set_tag_d:
mov.l FTEMP(%a0), %d0
mov.l %d0, %d1
#########################################################################
# XDEF **************************************************************** #
# set_tag_s(): return the optype of the input sgl fp number #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# a0 = pointer to single precision operand #
# #
# OUTPUT ************************************************************** #
# d0 = value of type tag #
# one of: NORM, INF, QNAN, SNAN, DENORM, ZERO #
# #
# ALGORITHM *********************************************************** #
# Simply test the exponent, j-bit, and mantissa values to #
# determine the type of operand. #
# #
#########################################################################
global set_tag_s
set_tag_s:
mov.l FTEMP(%a0), %d0
mov.l %d0, %d1
#########################################################################
# XDEF **************************************************************** #
# unf_res(): routine to produce default underflow result of a #
# scaled extended precision number; this is used by #
# fadd/fdiv/fmul/etc. emulation routines. #
# unf_res4(): same as above but for fsglmul/fsgldiv which use #
# single round prec and extended prec mode. #
# #
# XREF **************************************************************** #
# _denorm() - denormalize according to scale factor #
# _round() - round denormalized number according to rnd prec #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precison operand #
# d0 = scale factor #
# d1 = rounding precision/mode #
# #
# OUTPUT ************************************************************** #
# a0 = pointer to default underflow result in extended precision #
# d0.b = result FPSR_cc which caller may or may not want to save #
# #
# ALGORITHM *********************************************************** #
# Convert the input operand to "internal format" which means the #
# exponent is extended to 16 bits and the sign is stored in the unused #
# portion of the extended precison operand. Denormalize the number #
# according to the scale factor passed in d0. Then, round the #
# denormalized result. #
# Set the FPSR_exc bits as appropriate but return the cc bits in #
# d0 in case the caller doesn't want to save them (as is the case for #
# fmove out). #
# unf_res4() for fsglmul/fsgldiv forces the denorm to extended #
# precision and the rounding mode to single. #
# #
######################################################################### global unf_res
unf_res:
mov.l %d1, -(%sp) # save rnd prec,mode on stack
btst &0x7, FTEMP_EX(%a0) # make "internal" format
sne FTEMP_SGN(%a0)
# result is now rounded properly. convert back to normal format
bclr &0x7, FTEMP_EX(%a0) # clear sgn first; may have residue
tst.b FTEMP_SGN(%a0) # is "internal result" sign set?
beq.b unf_res_chkifzero # no; result is positive
bset &0x7, FTEMP_EX(%a0) # set result sgn clr.b FTEMP_SGN(%a0) # clear temp sign
# the number may have become zero after rounding. set ccodes accordingly.
unf_res_chkifzero: clr.l %d0
tst.l FTEMP_HI(%a0) # is value now a zero?
bne.b unf_res_cont # no
tst.l FTEMP_LO(%a0)
bne.b unf_res_cont # no
# bset &z_bit, FPSR_CC(%a6) # yes; set zero ccode bit
bset &z_bit, %d0 # yes; set zero ccode bit
unf_res_cont:
#
# can inex1 also be set along with unfl and inex2???
#
# we know that underflow has occurred. aunfl should be set if INEX2 is also set.
#
btst &inex2_bit, FPSR_EXCEPT(%a6) # is INEX2 set?
beq.b unf_res_end # no
bset &aunfl_bit, FPSR_AEXCEPT(%a6) # yes; set aunfl
unf_res_end:
add.l &0x4, %sp # clear stack
rts
# unf_res() for fsglmul() and fsgldiv(). global unf_res4
unf_res4:
mov.l %d1,-(%sp) # save rnd prec,mode on stack
btst &0x7,FTEMP_EX(%a0) # make "internal" format
sne FTEMP_SGN(%a0)
# result is now rounded properly. convert back to normal format
bclr &0x7,FTEMP_EX(%a0) # clear sgn first; may have residue
tst.b FTEMP_SGN(%a0) # is "internal result" sign set?
beq.b unf_res4_chkifzero # no; result is positive
bset &0x7,FTEMP_EX(%a0) # set result sgn clr.b FTEMP_SGN(%a0) # clear temp sign
# the number may have become zero after rounding. set ccodes accordingly.
unf_res4_chkifzero: clr.l %d0
tst.l FTEMP_HI(%a0) # is value now a zero?
bne.b unf_res4_cont # no
tst.l FTEMP_LO(%a0)
bne.b unf_res4_cont # no
# bset &z_bit,FPSR_CC(%a6) # yes; set zero ccode bit
bset &z_bit,%d0 # yes; set zero ccode bit
unf_res4_cont:
#
# can inex1 also be set along with unfl and inex2???
#
# we know that underflow has occurred. aunfl should be set if INEX2 is also set.
#
btst &inex2_bit,FPSR_EXCEPT(%a6) # is INEX2 set?
beq.b unf_res4_end # no
bset &aunfl_bit,FPSR_AEXCEPT(%a6) # yes; set aunfl
unf_res4_end:
add.l &0x4,%sp # clear stack
rts
#########################################################################
# XDEF **************************************************************** #
# ovf_res(): routine to produce the default overflow result of #
# an overflowing number. #
# ovf_res2(): same as above but the rnd mode/prec are passed #
# differently. #
# #
# XREF **************************************************************** #
# none #
# #
# INPUT *************************************************************** #
# d1.b = '-1' => (-); '0' => (+) #
# ovf_res(): #
# d0 = rnd mode/prec #
# ovf_res2(): #
# hi(d0) = rnd prec #
# lo(d0) = rnd mode #
# #
# OUTPUT ************************************************************** #
# a0 = points to extended precision result #
# d0.b = condition code bits #
# #
# ALGORITHM *********************************************************** #
# The default overflow result can be determined by the sign of #
# the result and the rounding mode/prec in effect. These bits are #
# concatenated together to create an index into the default result #
# table. A pointer to the correct result is returned in a0. The #
# resulting condition codes are returned in d0 in case the caller #
# doesn't want FPSR_cc altered (as is the case for fmove out). #
# #
#########################################################################
global ovf_res
ovf_res:
andi.w &0x10,%d1 # keep result sign
lsr.b &0x4,%d0 # shift prec/mode
or.b %d0,%d1 # concat the two
mov.w %d1,%d0 # make a copy
lsl.b &0x1,%d1 # multiply d1 by 2
bra.b ovf_res_load
global ovf_res2
ovf_res2:
and.w &0x10, %d1 # keep result sign
or.b %d0, %d1 # insert rnd mode
swap %d0
or.b %d0, %d1 # insert rnd prec
mov.w %d1, %d0 # make a copy
lsl.b &0x1, %d1 # shift left by 1
#
# use the rounding mode, precision, and result sign as in index into the
# two tables below to fetch the default result and the result ccodes.
#
ovf_res_load:
mov.b (tbl_ovfl_cc.b,%pc,%d0.w*1), %d0 # fetch result ccodes
lea (tbl_ovfl_result.b,%pc,%d1.w*8), %a0 # return result ptr
tbl_ovfl_result:
long 0x7fff0000,0x00000000,0x00000000,0x00000000 # +INF; RN
long 0x7ffe0000,0xffffffff,0xffffffff,0x00000000 # +EXT; RZ
long 0x7ffe0000,0xffffffff,0xffffffff,0x00000000 # +EXT; RM
long 0x7fff0000,0x00000000,0x00000000,0x00000000 # +INF; RP
long 0x7fff0000,0x00000000,0x00000000,0x00000000 # +INF; RN
long 0x407e0000,0xffffff00,0x00000000,0x00000000 # +SGL; RZ
long 0x407e0000,0xffffff00,0x00000000,0x00000000 # +SGL; RM
long 0x7fff0000,0x00000000,0x00000000,0x00000000 # +INF; RP
long 0x7fff0000,0x00000000,0x00000000,0x00000000 # +INF; RN
long 0x43fe0000,0xffffffff,0xfffff800,0x00000000 # +DBL; RZ
long 0x43fe0000,0xffffffff,0xfffff800,0x00000000 # +DBL; RM
long 0x7fff0000,0x00000000,0x00000000,0x00000000 # +INF; RP
long 0x00000000,0x00000000,0x00000000,0x00000000
long 0x00000000,0x00000000,0x00000000,0x00000000
long 0x00000000,0x00000000,0x00000000,0x00000000
long 0x00000000,0x00000000,0x00000000,0x00000000
long 0xffff0000,0x00000000,0x00000000,0x00000000 # -INF; RN
long 0xfffe0000,0xffffffff,0xffffffff,0x00000000 # -EXT; RZ
long 0xffff0000,0x00000000,0x00000000,0x00000000 # -INF; RM
long 0xfffe0000,0xffffffff,0xffffffff,0x00000000 # -EXT; RP
long 0xffff0000,0x00000000,0x00000000,0x00000000 # -INF; RN
long 0xc07e0000,0xffffff00,0x00000000,0x00000000 # -SGL; RZ
long 0xffff0000,0x00000000,0x00000000,0x00000000 # -INF; RM
long 0xc07e0000,0xffffff00,0x00000000,0x00000000 # -SGL; RP
long 0xffff0000,0x00000000,0x00000000,0x00000000 # -INF; RN
long 0xc3fe0000,0xffffffff,0xfffff800,0x00000000 # -DBL; RZ
long 0xffff0000,0x00000000,0x00000000,0x00000000 # -INF; RM
long 0xc3fe0000,0xffffffff,0xfffff800,0x00000000 # -DBL; RP
#########################################################################
# XDEF **************************************************************** #
# fout(): move from fp register to memory or data register #
# #
# XREF **************************************************************** #
# _round() - needed to create EXOP for sgl/dbl precision #
# norm() - needed to create EXOP for extended precision #
# ovf_res() - create default overflow result for sgl/dbl precision#
# unf_res() - create default underflow result for sgl/dbl prec. #
# dst_dbl() - create rounded dbl precision result. #
# dst_sgl() - create rounded sgl precision result. #
# fetch_dreg() - fetch dynamic k-factor reg for packed. #
# bindec() - convert FP binary number to packed number. #
# _mem_write() - write data to memory. #
# _mem_write2() - write data to memory unless supv mode -(a7) exc.#
# _dmem_write_{byte,word,long}() - write data to memory. #
# store_dreg_{b,w,l}() - store data to data register file. #
# facc_out_{b,w,l,d,x}() - data access error occurred. #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# d0 = round prec,mode #
# #
# OUTPUT ************************************************************** #
# fp0 : intermediate underflow or overflow result if #
# OVFL/UNFL occurred for a sgl or dbl operand #
# #
# ALGORITHM *********************************************************** #
# This routine is accessed by many handlers that need to do an #
# opclass three move of an operand out to memory. #
# Decode an fmove out (opclass 3) instruction to determine if #
# it's b,w,l,s,d,x, or p in size. b,w,l can be stored to either a data #
# register or memory. The algorithm uses a standard "fmove" to create #
# the rounded result. Also, since exceptions are disabled, this also #
# create the correct OPERR default result if appropriate. #
# For sgl or dbl precision, overflow or underflow can occur. If #
# either occurs and is enabled, the EXOP. #
# For extended precision, the stacked <ea> must be fixed along #
# w/ the address index register as appropriate w/ _calc_ea_fout(). If #
# the source is a denorm and if underflow is enabled, an EXOP must be #
# created. #
# For packed, the k-factor must be fetched from the instruction #
# word or a data register. The <ea> must be fixed as w/ extended #
# precision. Then, bindec() is called to create the appropriate #
# packed result. #
# If at any time an access error is flagged by one of the move- #
# to-memory routines, then a special exit must be made so that the #
# access error can be handled properly. #
# #
#########################################################################
global fout
fout:
bfextu EXC_CMDREG(%a6){&3:&3},%d1 # extract dst fmt
mov.w (tbl_fout.b,%pc,%d1.w*2),%a1 # use as index
jmp (tbl_fout.b,%pc,%a1) # jump to routine
swbeg &0x8
tbl_fout:
short fout_long - tbl_fout
short fout_sgl - tbl_fout
short fout_ext - tbl_fout
short fout_pack - tbl_fout
short fout_word - tbl_fout
short fout_dbl - tbl_fout
short fout_byte - tbl_fout
short fout_pack - tbl_fout
#################################################################
# fmove.b out ###################################################
#################################################################
# Only "Unimplemented Data Type" exceptions enter here. The operand
# is either a DENORM or a NORM.
fout_byte:
tst.b STAG(%a6) # is operand normalized?
bne.b fout_byte_denorm # no
#################################################################
# fmove.w out ###################################################
#################################################################
# Only "Unimplemented Data Type" exceptions enter here. The operand
# is either a DENORM or a NORM.
fout_word:
tst.b STAG(%a6) # is operand normalized?
bne.b fout_word_denorm # no
#################################################################
# fmove.l out ###################################################
#################################################################
# Only "Unimplemented Data Type" exceptions enter here. The operand
# is either a DENORM or a NORM.
fout_long:
tst.b STAG(%a6) # is operand normalized?
bne.b fout_long_denorm # no
#################################################################
# fmove.x out ###################################################
#################################################################
# Only "Unimplemented Data Type" exceptions enter here. The operand
# is either a DENORM or a NORM.
# The DENORM causes an Underflow exception.
fout_ext:
# we copy the extended precision result to FP_SCR0 so that the reserved
# 16-bit field gets zeroed. we do this since we promise not to disturb
# what's at SRC(a0).
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6) clr.w 2+FP_SCR0_EX(%a6) # clear reserved field
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
# we must not yet write the extended precision data to the stack
# in the pre-decrement case from supervisor mode or else we'll corrupt
# the stack frame. so, leave it in FP_SRC for now and deal with it later...
cmpi.b SPCOND_FLG(%a6),&mda7_flg
beq.b fout_ext_a7
bsr.l _dmem_write # write ext prec number to memory
tst.l %d1 # did dstore fail?
bne.w fout_ext_err # yes
tst.b STAG(%a6) # is operand normalized?
bne.b fout_ext_denorm # no
rts
# the number is a DENORM. must set the underflow exception bit
fout_ext_denorm:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set underflow exc bit
mov.b FPCR_ENABLE(%a6),%d0
andi.b &0x0a,%d0 # is UNFL or INEX enabled?
bne.b fout_ext_exc # yes
rts
# we don't want to do the write if the exception occurred in supervisor mode
# so _mem_write2() handles this for us.
fout_ext_a7:
bsr.l _mem_write2 # write ext prec number to memory
tst.l %d1 # did dstore fail?
bne.w fout_ext_err # yes
tst.b STAG(%a6) # is operand normalized?
bne.b fout_ext_denorm # no
rts
fout_ext_exc:
lea FP_SCR0(%a6),%a0
bsr.l norm # normalize the mantissa
neg.w %d0 # new exp = -(shft amt)
andi.w &0x7fff,%d0
andi.w &0x8000,FP_SCR0_EX(%a6) # keep only old sign
or.w %d0,FP_SCR0_EX(%a6) # insert new exponent
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
rts
#########################################################################
# fmove.s out ###########################################################
#########################################################################
fout_sgl:
andi.b &0x30,%d0 # clear rnd prec
ori.b &s_mode*0x10,%d0 # insert sgl prec
mov.l %d0,L_SCR3(%a6) # save rnd prec,mode on stack
#
# operand is a normalized number. first, we check to see if the move out
# would cause either an underflow or overflow. these cases are handled
# separately. otherwise, set the FPCR to the proper rounding mode and
# execute the move.
#
mov.w SRC_EX(%a0),%d0 # extract exponent
andi.w &0x7fff,%d0 # strip sign
cmpi.w %d0,&SGL_HI # will operand overflow?
bgt.w fout_sgl_ovfl # yes; go handle OVFL
beq.w fout_sgl_may_ovfl # maybe; go handle possible OVFL
cmpi.w %d0,&SGL_LO # will operand underflow?
blt.w fout_sgl_unfl # yes; go handle underflow
#
# NORMs(in range) can be stored out by a simple "fmov.s"
# Unnormalized inputs can come through this point.
#
fout_sgl_exg:
fmovm.x SRC(%a0),&0x80 # fetch fop from stack
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fmov.s %fp0,%d0 # store does convert and round
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d1 # save FPSR
or.w %d1,2+USER_FPSR(%a6) # set possible inex2/ainex
fout_sgl_exg_write:
mov.b 1+EXC_OPWORD(%a6),%d1 # extract dst mode
andi.b &0x38,%d1 # is mode == 0? (Dreg dst)
beq.b fout_sgl_exg_write_dn # must save to integer regfile
mov.l EXC_EA(%a6),%a0 # stacked <ea> is correct
bsr.l _dmem_write_long # write long
tst.l %d1 # did dstore fail?
bne.l facc_out_l # yes
#
# here, we know that the operand would UNFL if moved out to single prec,
# so, denorm and round and then use generic store single routine to
# write the value to memory.
#
fout_sgl_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set UNFL
fout_sgl_unfl_chkexc:
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0a,%d1 # is UNFL or INEX enabled?
bne.w fout_sd_exc_unfl # yes
addq.l &0x4,%sp
rts
#
# it's definitely an overflow so call ovf_res to get the correct answer
#
fout_sgl_ovfl:
tst.b 3+SRC_HI(%a0) # is result inexact?
bne.b fout_sgl_ovfl_inex2
tst.l SRC_LO(%a0) # is result inexact?
bne.b fout_sgl_ovfl_inex2
ori.w &ovfl_inx_mask,2+USER_FPSR(%a6) # set ovfl/aovfl/ainex
bra.b fout_sgl_ovfl_cont
fout_sgl_ovfl_inex2:
ori.w &ovfinx_mask,2+USER_FPSR(%a6) # set ovfl/aovfl/ainex/inex2
fout_sgl_ovfl_cont:
mov.l %a0,-(%sp)
# call ovf_res() w/ sgl prec and the correct rnd mode to create the default
# overflow result. DON'T save the returned ccodes from ovf_res() since
# fmove out doesn't alter them.
tst.b SRC_EX(%a0) # is operand negative?
smi %d1 # set if so
mov.l L_SCR3(%a6),%d0 # pass: sgl prec,rnd mode
bsr.l ovf_res # calc OVFL result
fmovm.x (%a0),&0x80 # load default overflow result
fmov.s %fp0,%d0 # store to single
mov.b 1+EXC_OPWORD(%a6),%d1 # extract dst mode
andi.b &0x38,%d1 # is mode == 0? (Dreg dst)
beq.b fout_sgl_ovfl_dn # must save to integer regfile
mov.l EXC_EA(%a6),%a0 # stacked <ea> is correct
bsr.l _dmem_write_long # write long
tst.l %d1 # did dstore fail?
bne.l facc_out_l # yes
fout_sgl_ovfl_chkexc:
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0a,%d1 # is UNFL or INEX enabled?
bne.w fout_sd_exc_ovfl # yes
addq.l &0x4,%sp
rts
#
# move out MAY overflow:
# (1) force the exp to 0x3fff
# (2) do a move w/ appropriate rnd mode
# (3) if exp still equals zero, then insert original exponent
# for the correct result.
# if exp now equals one, then it overflowed so call ovf_res.
#
fout_sgl_may_ovfl:
mov.w SRC_EX(%a0),%d1 # fetch current sign
andi.w &0x8000,%d1 # keep it,clear exp
ori.w &0x3fff,%d1 # insert exp = 0
mov.w %d1,FP_SCR0_EX(%a6) # insert scaled exp
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6) # copy hi(man)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6) # copy lo(man)
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.x FP_SCR0(%a6),%fp0 # force fop to be rounded
fmov.l &0x0,%fpcr # clear FPCR
fabs.x %fp0 # need absolute value
fcmp.b %fp0,&0x2 # did exponent increase?
fblt.w fout_sgl_exg # no; go finish NORM
bra.w fout_sgl_ovfl # yes; go handle overflow
tst.b 2+FP_SCR0_EX(%a6) # is EXOP negative?
beq.b fout_sd_exc_done # no
bset &0x7,FP_SCR0_EX(%a6) # yes
fout_sd_exc_done:
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
rts
#################################################################
# fmove.d out ###################################################
#################################################################
fout_dbl:
andi.b &0x30,%d0 # clear rnd prec
ori.b &d_mode*0x10,%d0 # insert dbl prec
mov.l %d0,L_SCR3(%a6) # save rnd prec,mode on stack
#
# operand is a normalized number. first, we check to see if the move out
# would cause either an underflow or overflow. these cases are handled
# separately. otherwise, set the FPCR to the proper rounding mode and
# execute the move.
#
mov.w SRC_EX(%a0),%d0 # extract exponent
andi.w &0x7fff,%d0 # strip sign
cmpi.w %d0,&DBL_HI # will operand overflow?
bgt.w fout_dbl_ovfl # yes; go handle OVFL
beq.w fout_dbl_may_ovfl # maybe; go handle possible OVFL
cmpi.w %d0,&DBL_LO # will operand underflow?
blt.w fout_dbl_unfl # yes; go handle underflow
#
# NORMs(in range) can be stored out by a simple "fmov.d"
# Unnormalized inputs can come through this point.
#
fout_dbl_exg:
fmovm.x SRC(%a0),&0x80 # fetch fop from stack
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fmov.d %fp0,L_SCR1(%a6) # store does convert and round
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d0 # save FPSR
or.w %d0,2+USER_FPSR(%a6) # set possible inex2/ainex
mov.l EXC_EA(%a6),%a1 # pass: dst addr
lea L_SCR1(%a6),%a0 # pass: src addr
movq.l &0x8,%d0 # pass: opsize is 8 bytes
bsr.l _dmem_write # store dbl fop to memory
tst.l %d1 # did dstore fail?
bne.l facc_out_d # yes
rts # no; so we're finished
#
# here, we know that the operand would UNFL if moved out to double prec,
# so, denorm and round and then use generic store double routine to
# write the value to memory.
#
fout_dbl_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set UNFL
lea FP_SCR0(%a6),%a0 # pass: ptr to fop
bsr.l dst_dbl # convert to single prec
mov.l %d0,L_SCR1(%a6)
mov.l %d1,L_SCR2(%a6)
mov.l EXC_EA(%a6),%a1 # pass: dst addr
lea L_SCR1(%a6),%a0 # pass: src addr
movq.l &0x8,%d0 # pass: opsize is 8 bytes
bsr.l _dmem_write # store dbl fop to memory
tst.l %d1 # did dstore fail?
bne.l facc_out_d # yes
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0a,%d1 # is UNFL or INEX enabled?
bne.w fout_sd_exc_unfl # yes
addq.l &0x4,%sp
rts
#
# it's definitely an overflow so call ovf_res to get the correct answer
#
fout_dbl_ovfl:
mov.w 2+SRC_LO(%a0),%d0
andi.w &0x7ff,%d0
bne.b fout_dbl_ovfl_inex2
ori.w &ovfl_inx_mask,2+USER_FPSR(%a6) # set ovfl/aovfl/ainex
bra.b fout_dbl_ovfl_cont
fout_dbl_ovfl_inex2:
ori.w &ovfinx_mask,2+USER_FPSR(%a6) # set ovfl/aovfl/ainex/inex2
fout_dbl_ovfl_cont:
mov.l %a0,-(%sp)
# call ovf_res() w/ dbl prec and the correct rnd mode to create the default
# overflow result. DON'T save the returned ccodes from ovf_res() since
# fmove out doesn't alter them.
tst.b SRC_EX(%a0) # is operand negative?
smi %d1 # set if so
mov.l L_SCR3(%a6),%d0 # pass: dbl prec,rnd mode
bsr.l ovf_res # calc OVFL result
fmovm.x (%a0),&0x80 # load default overflow result
fmov.d %fp0,L_SCR1(%a6) # store to double
mov.l EXC_EA(%a6),%a1 # pass: dst addr
lea L_SCR1(%a6),%a0 # pass: src addr
movq.l &0x8,%d0 # pass: opsize is 8 bytes
bsr.l _dmem_write # store dbl fop to memory
tst.l %d1 # did dstore fail?
bne.l facc_out_d # yes
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0a,%d1 # is UNFL or INEX enabled?
bne.w fout_sd_exc_ovfl # yes
addq.l &0x4,%sp
rts
#
# move out MAY overflow:
# (1) force the exp to 0x3fff
# (2) do a move w/ appropriate rnd mode
# (3) if exp still equals zero, then insert original exponent
# for the correct result.
# if exp now equals one, then it overflowed so call ovf_res.
#
fout_dbl_may_ovfl:
mov.w SRC_EX(%a0),%d1 # fetch current sign
andi.w &0x8000,%d1 # keep it,clear exp
ori.w &0x3fff,%d1 # insert exp = 0
mov.w %d1,FP_SCR0_EX(%a6) # insert scaled exp
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6) # copy hi(man)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6) # copy lo(man)
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.x FP_SCR0(%a6),%fp0 # force fop to be rounded
fmov.l &0x0,%fpcr # clear FPCR
fabs.x %fp0 # need absolute value
fcmp.b %fp0,&0x2 # did exponent increase?
fblt.w fout_dbl_exg # no; go finish NORM
bra.w fout_dbl_ovfl # yes; go handle overflow
dst_dbl: clr.l %d0 # clear d0
mov.w FTEMP_EX(%a0),%d0 # get exponent
subi.w &EXT_BIAS,%d0 # subtract extended precision bias
addi.w &DBL_BIAS,%d0 # add double precision bias
tst.b FTEMP_HI(%a0) # is number a denorm?
bmi.b dst_get_dupper # no
subq.w &0x1,%d0 # yes; denorm bias = DBL_BIAS - 1
dst_get_dupper:
swap %d0 # d0 now in upper word
lsl.l &0x4,%d0 # d0 in proper place for dbl prec exp
tst.b FTEMP_EX(%a0) # test sign
bpl.b dst_get_dman # if positive, go process mantissa
bset &0x1f,%d0 # if negative, set sign
dst_get_dman:
mov.l FTEMP_HI(%a0),%d1 # get ms mantissa
bfextu %d1{&1:&20},%d1 # get upper 20 bits of ms
or.l %d1,%d0 # put these bits in ms word of double
mov.l %d0,L_SCR1(%a6) # put the new exp back on the stack
mov.l FTEMP_HI(%a0),%d1 # get ms mantissa
mov.l &21,%d0 # load shift count
lsl.l %d0,%d1 # put lower 11 bits in upper bits
mov.l %d1,L_SCR2(%a6) # build lower lword in memory
mov.l FTEMP_LO(%a0),%d1 # get ls mantissa
bfextu %d1{&0:&21},%d0 # get ls 21 bits of double
mov.l L_SCR2(%a6),%d1
or.l %d0,%d1 # put them in double result
mov.l L_SCR1(%a6),%d0
rts
dst_sgl: clr.l %d0
mov.w FTEMP_EX(%a0),%d0 # get exponent
subi.w &EXT_BIAS,%d0 # subtract extended precision bias
addi.w &SGL_BIAS,%d0 # add single precision bias
tst.b FTEMP_HI(%a0) # is number a denorm?
bmi.b dst_get_supper # no
subq.w &0x1,%d0 # yes; denorm bias = SGL_BIAS - 1
dst_get_supper:
swap %d0 # put exp in upper word of d0
lsl.l &0x7,%d0 # shift it into single exp bits
tst.b FTEMP_EX(%a0) # test sign
bpl.b dst_get_sman # if positive, continue
bset &0x1f,%d0 # if negative, put in sign first
dst_get_sman:
mov.l FTEMP_HI(%a0),%d1 # get ms mantissa
andi.l &0x7fffff00,%d1 # get upper 23 bits of ms
lsr.l &0x8,%d1 # and put them flush right
or.l %d1,%d0 # put these bits in ms word of single
rts
##############################################################################
fout_pack:
bsr.l _calc_ea_fout # fetch the <ea>
mov.l %a0,-(%sp)
mov.b STAG(%a6),%d0 # fetch input type
bne.w fout_pack_not_norm # input is not NORM
# bindec is currently scrambling FP_SRC for denorm inputs.
# we'll have to change this, but for now, tough luck!!!
bsr.l bindec # convert xprec to packed
# add the extra condition that only if the k-factor was zero, too, should
# we zero the exponent
tst.l %d0
bne.b fout_pack_set
# "mantissa" is all zero which means that the answer is zero. but, the '040
# algorithm allows the exponent to be non-zero. the 881/2 do not. Therefore,
# if the mantissa is zero, I will zero the exponent, too.
# the question now is whether the exponents sign bit is allowed to be non-zero
# for a zero, also...
andi.w &0xf000,FP_SCR0(%a6)
bsr.l _dmem_write # write ext prec number to memory
tst.l %d1 # did dstore fail?
bne.w fout_ext_err # yes
rts
# we don't want to do the write if the exception occurred in supervisor mode
# so _mem_write2() handles this for us.
fout_pack_a7:
bsr.l _mem_write2 # write ext prec number to memory
tst.l %d1 # did dstore fail?
bne.w fout_ext_err # yes
rts
fout_pack_not_norm:
cmpi.b %d0,&DENORM # is it a DENORM?
beq.w fout_pack_norm # yes
lea FP_SRC(%a6),%a0 clr.w 2+FP_SRC_EX(%a6)
cmpi.b %d0,&SNAN # is it an SNAN?
beq.b fout_pack_snan # yes
bra.b fout_pack_write # no
fout_pack_snan:
ori.w &snaniop2_mask,FPSR_EXCEPT(%a6) # set SNAN/AIOP
bset &0x6,FP_SRC_HI(%a6) # set snan bit
bra.b fout_pack_write
#########################################################################
# XDEF **************************************************************** #
# fmul(): emulates the fmul instruction #
# fsmul(): emulates the fsmul instruction #
# fdmul(): emulates the fdmul instruction #
# #
# XREF **************************************************************** #
# scale_to_zero_src() - scale src exponent to zero #
# scale_to_zero_dst() - scale dst exponent to zero #
# unf_res() - return default underflow result #
# ovf_res() - return default overflow result #
# res_qnan() - return QNAN result #
# res_snan() - return SNAN result #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# a1 = pointer to extended precision destination operand #
# d0 rnd prec,mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms/denorms into ext/sgl/dbl precision. #
# For norms/denorms, scale the exponents such that a multiply #
# instruction won't cause an exception. Use the regular fmul to #
# compute a result. Check if the regular operands would have taken #
# an exception. If so, return the default overflow/underflow result #
# and return the EXOP if exceptions are enabled. Else, scale the #
# result operand to the proper exponent. #
# #
#########################################################################
align 0x10
tbl_fmul_ovfl:
long 0x3fff - 0x7ffe # ext_max
long 0x3fff - 0x407e # sgl_max
long 0x3fff - 0x43fe # dbl_max
tbl_fmul_unfl:
long 0x3fff + 0x0001 # ext_unfl
long 0x3fff - 0x3f80 # sgl_unfl
long 0x3fff - 0x3c00 # dbl_unfl
mov.w 2+L_SCR3(%a6),%d1 # fetch precision
lsr.b &0x6,%d1 # shift to lo bits
mov.l (%sp)+,%d0 # load S.F.
cmp.l %d0,(tbl_fmul_ovfl.w,%pc,%d1.w*4) # would result ovfl?
beq.w fmul_may_ovfl # result may rnd to overflow
blt.w fmul_ovfl # result will overflow
cmp.l %d0,(tbl_fmul_unfl.w,%pc,%d1.w*4) # would result unfl?
beq.w fmul_may_unfl # result may rnd to no unfl
bgt.w fmul_unfl # result will underflow
#
# NORMAL:
# - the result of the multiply operation will neither overflow nor underflow.
# - do the multiply to the proper precision and rounding mode.
# - scale the result exponent using the scale factor. if both operands were
# normalized then we really don't need to go through this scaling. but for now,
# this will do.
#
fmul_normal:
fmovm.x FP_SCR1(%a6),&0x80 # load dst operand
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fmul.x FP_SCR0(%a6),%fp0 # execute multiply
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fmul_normal_exit:
fmovm.x &0x80,FP_SCR0(%a6) # store out result
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # load {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
or.w %d2,%d1 # concat old sign,new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# OVERFLOW:
# - the result of the multiply operation is an overflow.
# - do the multiply to the proper precision and rounding mode in order to
# set the inexact bits.
# - calculate the default result and return it in fp0.
# - if overflow or inexact is enabled, we need a multiply result rounded to
# extended precision. if the original operation was extended, then we have this
# result. if the original operation was single or double, we have to do another
# multiply using extended precision and the correct rounding mode. the result
# of this operation then has its exponent scaled by -0x6000 to create the
# exceptional operand.
#
fmul_ovfl:
fmovm.x FP_SCR1(%a6),&0x80 # load dst operand
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fmul.x FP_SCR0(%a6),%fp0 # execute multiply
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
# save setting this until now because this is where fmul_may_ovfl may jump in
fmul_ovfl_tst:
or.l &ovfl_inx_mask,USER_FPSR(%a6) # set ovfl/aovfl/ainex
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x13,%d1 # is OVFL or INEX enabled?
bne.b fmul_ovfl_ena # yes
# calculate the default result
fmul_ovfl_dis:
btst &neg_bit,FPSR_CC(%a6) # is result negative?
sne %d1 # set sign param accordingly
mov.l L_SCR3(%a6),%d0 # pass rnd prec,mode
bsr.l ovf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # set INF,N if applicable
fmovm.x (%a0),&0x80 # return default result in fp0
rts
#
# OVFL is enabled; Create EXOP:
# - if precision is extended, then we have the EXOP. simply bias the exponent
# with an extra -0x6000. if the precision is single or double, we need to
# calculate a result rounded to extended precision.
#
fmul_ovfl_ena:
mov.l L_SCR3(%a6),%d1
andi.b &0xc0,%d1 # test the rnd prec
bne.b fmul_ovfl_ena_sd # it's sgl or dbl
fmul_ovfl_ena_cont:
fmovm.x &0x80,FP_SCR0(%a6) # move result to stack
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.w %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign sub.l %d0,%d1 # add scale factor
subi.l &0x6000,%d1 # subtract bias
andi.w &0x7fff,%d1 # clear sign bit
andi.w &0x8000,%d2 # keep old sign
or.w %d2,%d1 # concat old sign,new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
bra.b fmul_ovfl_dis
#
# may OVERFLOW:
# - the result of the multiply operation MAY overflow.
# - do the multiply to the proper precision and rounding mode in order to
# set the inexact bits.
# - calculate the default result and return it in fp0.
#
fmul_may_ovfl:
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fmul.x FP_SCR0(%a6),%fp0 # execute multiply
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x2 # is |result| >= 2.b?
fbge.w fmul_ovfl_tst # yes; overflow has occurred
# no, it didn't overflow; we have correct result
bra.w fmul_normal_exit
#
# UNDERFLOW:
# - the result of the multiply operation is an underflow.
# - do the multiply to the proper precision and rounding mode in order to
# set the inexact bits.
# - calculate the default result and return it in fp0.
# - if overflow or inexact is enabled, we need a multiply result rounded to
# extended precision. if the original operation was extended, then we have this
# result. if the original operation was single or double, we have to do another
# multiply using extended precision and the correct rounding mode. the result
# of this operation then has its exponent scaled by -0x6000 to create the
# exceptional operand.
#
fmul_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
# for fun, let's use only extended precision, round to zero. then, let
# the unf_res() routine figure out all the rest.
# will we get the correct answer.
fmovm.x FP_SCR1(%a6),&0x80 # load dst operand
fmov.l &rz_mode*0x10,%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fmul.x FP_SCR0(%a6),%fp0 # execute multiply
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fmul_unfl_ena # yes
fmul_unfl_dis:
fmovm.x &0x80,FP_SCR0(%a6) # store out result
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # unf_res2 may have set'Z'
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# UNFL is enabled.
#
fmul_unfl_ena:
fmovm.x FP_SCR1(%a6),&0x40 # load dst op
mov.l L_SCR3(%a6),%d1
andi.b &0xc0,%d1 # is precision extended?
bne.b fmul_unfl_ena_sd # no, sgl or dbl
# if the rnd mode is anything but RZ, then we have to re-do the above
# multiplication because we used RZ for all.
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmovm.x &0x40,FP_SCR0(%a6) # save result to stack
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
addi.l &0x6000,%d1 # add bias
andi.w &0x7fff,%d1
or.w %d2,%d1 # concat old sign,new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
bra.w fmul_unfl_dis
fmul_unfl_ena_sd:
mov.l L_SCR3(%a6),%d1
andi.b &0x30,%d1 # use only rnd mode
fmov.l %d1,%fpcr # set FPCR
bra.b fmul_unfl_ena_cont
# MAY UNDERFLOW:
# -use the correct rounding mode and precision. this code favors operations
# that do not underflow.
fmul_may_unfl:
fmovm.x FP_SCR1(%a6),&0x80 # load dst operand
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fmul.x FP_SCR0(%a6),%fp0 # execute multiply
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x2 # is |result| > 2.b?
fbgt.w fmul_normal_exit # no; no underflow occurred
fblt.w fmul_unfl # yes; underflow occurred
#
# we still don't know if underflow occurred. result is ~ equal to 2. but,
# we don't know if the result was an underflow that rounded up to a 2 or
# a normalized number that rounded down to a 2. so, redo the entire operation
# using RZ as the rounding mode to see what the pre-rounded result is.
# this case should be relatively rare.
#
fmovm.x FP_SCR1(%a6),&0x40 # load dst operand
#
# Multiply: inputs are not both normalized; what are they?
#
fmul_not_norm:
mov.w (tbl_fmul_op.b,%pc,%d1.w*2),%d1
jmp (tbl_fmul_op.b,%pc,%d1.w)
swbeg &48
tbl_fmul_op:
short fmul_norm - tbl_fmul_op # NORM x NORM
short fmul_zero - tbl_fmul_op # NORM x ZERO
short fmul_inf_src - tbl_fmul_op # NORM x INF
short fmul_res_qnan - tbl_fmul_op # NORM x QNAN
short fmul_norm - tbl_fmul_op # NORM x DENORM
short fmul_res_snan - tbl_fmul_op # NORM x SNAN
short tbl_fmul_op - tbl_fmul_op #
short tbl_fmul_op - tbl_fmul_op #
short fmul_zero - tbl_fmul_op # ZERO x NORM
short fmul_zero - tbl_fmul_op # ZERO x ZERO
short fmul_res_operr - tbl_fmul_op # ZERO x INF
short fmul_res_qnan - tbl_fmul_op # ZERO x QNAN
short fmul_zero - tbl_fmul_op # ZERO x DENORM
short fmul_res_snan - tbl_fmul_op # ZERO x SNAN
short tbl_fmul_op - tbl_fmul_op #
short tbl_fmul_op - tbl_fmul_op #
short fmul_inf_dst - tbl_fmul_op # INF x NORM
short fmul_res_operr - tbl_fmul_op # INF x ZERO
short fmul_inf_dst - tbl_fmul_op # INF x INF
short fmul_res_qnan - tbl_fmul_op # INF x QNAN
short fmul_inf_dst - tbl_fmul_op # INF x DENORM
short fmul_res_snan - tbl_fmul_op # INF x SNAN
short tbl_fmul_op - tbl_fmul_op #
short tbl_fmul_op - tbl_fmul_op #
short fmul_res_qnan - tbl_fmul_op # QNAN x NORM
short fmul_res_qnan - tbl_fmul_op # QNAN x ZERO
short fmul_res_qnan - tbl_fmul_op # QNAN x INF
short fmul_res_qnan - tbl_fmul_op # QNAN x QNAN
short fmul_res_qnan - tbl_fmul_op # QNAN x DENORM
short fmul_res_snan - tbl_fmul_op # QNAN x SNAN
short tbl_fmul_op - tbl_fmul_op #
short tbl_fmul_op - tbl_fmul_op #
short fmul_norm - tbl_fmul_op # NORM x NORM
short fmul_zero - tbl_fmul_op # NORM x ZERO
short fmul_inf_src - tbl_fmul_op # NORM x INF
short fmul_res_qnan - tbl_fmul_op # NORM x QNAN
short fmul_norm - tbl_fmul_op # NORM x DENORM
short fmul_res_snan - tbl_fmul_op # NORM x SNAN
short tbl_fmul_op - tbl_fmul_op #
short tbl_fmul_op - tbl_fmul_op #
short fmul_res_snan - tbl_fmul_op # SNAN x NORM
short fmul_res_snan - tbl_fmul_op # SNAN x ZERO
short fmul_res_snan - tbl_fmul_op # SNAN x INF
short fmul_res_snan - tbl_fmul_op # SNAN x QNAN
short fmul_res_snan - tbl_fmul_op # SNAN x DENORM
short fmul_res_snan - tbl_fmul_op # SNAN x SNAN
short tbl_fmul_op - tbl_fmul_op #
short tbl_fmul_op - tbl_fmul_op #
#
# Multiply: (Zero x Zero) || (Zero x norm) || (Zero x denorm)
# global fmul_zero # global for fsglmul
fmul_zero:
mov.b SRC_EX(%a0),%d0 # exclusive or the signs
mov.b DST_EX(%a1),%d1
eor.b %d0,%d1
bpl.b fmul_zero_p # result ZERO is pos.
fmul_zero_n:
fmov.s &0x80000000,%fp0 # load -ZERO
mov.b &z_bmask+neg_bmask,FPSR_CC(%a6) # set Z/N
rts
fmul_zero_p:
fmov.s &0x00000000,%fp0 # load +ZERO
mov.b &z_bmask,FPSR_CC(%a6) # set Z
rts
#
# Multiply: (inf x inf) || (inf x norm) || (inf x denorm)
#
# Note: The j-bit for an infinity is a don't-care. However, to be
# strictly compatible w/ the 68881/882, we make sure to return an
# INF w/ the j-bit set if the input INF j-bit was set. Destination
# INFs take priority.
# global fmul_inf_dst # global for fsglmul
fmul_inf_dst:
fmovm.x DST(%a1),&0x80 # return INF result in fp0
mov.b SRC_EX(%a0),%d0 # exclusive or the signs
mov.b DST_EX(%a1),%d1
eor.b %d0,%d1
bpl.b fmul_inf_dst_p # result INF is pos.
fmul_inf_dst_n:
fabs.x %fp0 # clear result sign
fneg.x %fp0 # set result sign
mov.b &inf_bmask+neg_bmask,FPSR_CC(%a6) # set INF/N
rts
fmul_inf_dst_p:
fabs.x %fp0 # clear result sign
mov.b &inf_bmask,FPSR_CC(%a6) # set INF
rts
global fmul_inf_src # global for fsglmul
fmul_inf_src:
fmovm.x SRC(%a0),&0x80 # return INF result in fp0
mov.b SRC_EX(%a0),%d0 # exclusive or the signs
mov.b DST_EX(%a1),%d1
eor.b %d0,%d1
bpl.b fmul_inf_dst_p # result INF is pos.
bra.b fmul_inf_dst_n
#########################################################################
# XDEF **************************************************************** #
# fin(): emulates the fmove instruction #
# fsin(): emulates the fsmove instruction #
# fdin(): emulates the fdmove instruction #
# #
# XREF **************************************************************** #
# norm() - normalize mantissa for EXOP on denorm #
# scale_to_zero_src() - scale src exponent to zero #
# ovf_res() - return default overflow result #
# unf_res() - return default underflow result #
# res_qnan_1op() - return QNAN result #
# res_snan_1op() - return SNAN result #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# d0 = round prec/mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms into extended, single, and double precision. #
# Norms can be emulated w/ a regular fmove instruction. For #
# sgl/dbl, must scale exponent and perform an "fmove". Check to see #
# if the result would have overflowed/underflowed. If so, use unf_res() #
# or ovf_res() to return the default result. Also return EXOP if #
# exception is enabled. If no exception, return the default result. #
# Unnorms don't pass through here. #
# #
#########################################################################
global fsin
fsin:
andi.b &0x30,%d0 # clear rnd prec
ori.b &s_mode*0x10,%d0 # insert sgl precision
bra.b fin
global fin
fin:
mov.l %d0,L_SCR3(%a6) # store rnd info
mov.b STAG(%a6),%d1 # fetch src optype tag
bne.w fin_not_norm # optimize on non-norm input
#
# FP MOVE IN: NORMs and DENORMs ONLY!
#
fin_norm:
andi.b &0xc0,%d0 # is precision extended?
bne.w fin_not_ext # no, so go handle dbl or sgl
#
# precision selected is extended. so...we cannot get an underflow
# or overflow because of rounding to the correct precision. so...
# skip the scaling and unscaling...
#
tst.b SRC_EX(%a0) # is the operand negative?
bpl.b fin_norm_done # no
bset &neg_bit,FPSR_CC(%a6) # yes, so set'N' ccode bit
fin_norm_done:
fmovm.x SRC(%a0),&0x80 # return result in fp0
rts
#
# for an extended precision DENORM, the UNFL exception bit is set
# the accrued bit is NOT set in this instance(no inexactness!)
#
fin_denorm:
andi.b &0xc0,%d0 # is precision extended?
bne.w fin_not_ext # no, so go handle dbl or sgl
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
tst.b SRC_EX(%a0) # is the operand negative?
bpl.b fin_denorm_done # no
bset &neg_bit,FPSR_CC(%a6) # yes, so set'N' ccode bit
fin_denorm_done:
fmovm.x SRC(%a0),&0x80 # return result in fp0
btst &unfl_bit,FPCR_ENABLE(%a6) # is UNFL enabled?
bne.b fin_denorm_unfl_ena # yes
rts
#
# the input is an extended DENORM and underflow is enabled in the FPCR.
# normalize the mantissa and add the bias of 0x6000 to the resulting negative
# exponent and insert back into the operand.
#
fin_denorm_unfl_ena:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
lea FP_SCR0(%a6),%a0 # pass: ptr to operand
bsr.l norm # normalize result
neg.w %d0 # new exponent = -(shft val)
addi.w &0x6000,%d0 # add new bias to exponent
mov.w FP_SCR0_EX(%a6),%d1 # fetch old sign,exp
andi.w &0x8000,%d1 # keep old sign
andi.w &0x7fff,%d0 # clear sign position
or.w %d1,%d0 # concat new exo,old sign
mov.w %d0,FP_SCR0_EX(%a6) # insert new exponent
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
rts
#
# operand is to be rounded to single or double precision
#
fin_not_ext:
cmpi.b %d0,&s_mode*0x10 # separate sgl/dbl prec
bne.b fin_dbl
#
# operand is to be rounded to single precision
#
fin_sgl:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_to_zero_src # calculate scale factor
cmpi.l %d0,&0x3fff-0x3f80 # will move in underflow?
bge.w fin_sd_unfl # yes; go handle underflow
cmpi.l %d0,&0x3fff-0x407e # will move in overflow?
beq.w fin_sd_may_ovfl # maybe; go check
blt.w fin_sd_ovfl # yes; go handle overflow
#
# operand will NOT overflow or underflow when moved into the fp reg file
#
fin_sd_normal:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.x FP_SCR0(%a6),%fp0 # perform move
fmov.l %fpsr,%d1 # save FPSR
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fin_sd_normal_exit:
mov.l %d2,-(%sp) # save d2
fmovm.x &0x80,FP_SCR0(%a6) # store out result
mov.w FP_SCR0_EX(%a6),%d1 # load {sgn,exp}
mov.w %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign sub.l %d0,%d1 # add scale factor
andi.w &0x8000,%d2 # keep old sign
or.w %d1,%d2 # concat old sign,new exponent
mov.w %d2,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x80 # return result in fp0
rts
#
# operand is to be rounded to double precision
#
fin_dbl:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_to_zero_src # calculate scale factor
cmpi.l %d0,&0x3fff-0x3c00 # will move in underflow?
bge.w fin_sd_unfl # yes; go handle underflow
cmpi.l %d0,&0x3fff-0x43fe # will move in overflow?
beq.w fin_sd_may_ovfl # maybe; go check
blt.w fin_sd_ovfl # yes; go handle overflow
bra.w fin_sd_normal # no; ho handle normalized op
#
# operand WILL underflow when moved in to the fp register file
#
fin_sd_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
tst.b FP_SCR0_EX(%a6) # is operand negative?
bpl.b fin_sd_unfl_tst
bset &neg_bit,FPSR_CC(%a6) # set'N' ccode bit
# if underflow or inexact is enabled, then go calculate the EXOP first.
fin_sd_unfl_tst:
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fin_sd_unfl_ena # yes
fin_sd_unfl_dis:
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # unf_res may have set'Z'
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# operand will underflow AND underflow or inexact is enabled.
# Therefore, we must return the result rounded to extended precision.
#
fin_sd_unfl_ena:
mov.l FP_SCR0_HI(%a6),FP_SCR1_HI(%a6)
mov.l FP_SCR0_LO(%a6),FP_SCR1_LO(%a6)
mov.w FP_SCR0_EX(%a6),%d1 # load current exponent
mov.l %d2,-(%sp) # save d2
mov.w %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign sub.l %d0,%d1 # subtract scale factor
andi.w &0x8000,%d2 # extract old sign
addi.l &0x6000,%d1 # add new bias
andi.w &0x7fff,%d1
or.w %d1,%d2 # concat old sign,new exp
mov.w %d2,FP_SCR1_EX(%a6) # insert new exponent
fmovm.x FP_SCR1(%a6),&0x40 # return EXOP in fp1
mov.l (%sp)+,%d2 # restore d2
bra.b fin_sd_unfl_dis
#
# operand WILL overflow.
#
fin_sd_ovfl:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.x FP_SCR0(%a6),%fp0 # perform move
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d1 # save FPSR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fin_sd_ovfl_tst:
or.l &ovfl_inx_mask,USER_FPSR(%a6) # set ovfl/aovfl/ainex
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x13,%d1 # is OVFL or INEX enabled?
bne.b fin_sd_ovfl_ena # yes
#
# OVFL is not enabled; therefore, we must create the default result by
# calling ovf_res().
#
fin_sd_ovfl_dis:
btst &neg_bit,FPSR_CC(%a6) # is result negative?
sne %d1 # set sign param accordingly
mov.l L_SCR3(%a6),%d0 # pass: prec,mode
bsr.l ovf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # set INF,N if applicable
fmovm.x (%a0),&0x80 # return default result in fp0
rts
#
# OVFL is enabled.
# the INEX2 bit has already been updated by the round to the correct precision.
# now, round to extended(and don't alter the FPSR).
#
fin_sd_ovfl_ena:
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor sub.l &0x6000,%d1 # subtract bias
andi.w &0x7fff,%d1
or.w %d2,%d1
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
bra.b fin_sd_ovfl_dis
#
# the move in MAY overflow. so...
#
fin_sd_may_ovfl:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.x FP_SCR0(%a6),%fp0 # perform the move
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x2 # is |result| >= 2.b?
fbge.w fin_sd_ovfl_tst # yes; overflow has occurred
# no, it didn't overflow; we have correct result
bra.w fin_sd_normal_exit
#
# operand is not a NORM: check its optype and branch accordingly
#
fin_not_norm:
cmpi.b %d1,&DENORM # weed out DENORM
beq.w fin_denorm
cmpi.b %d1,&SNAN # weed out SNANs
beq.l res_snan_1op
cmpi.b %d1,&QNAN # weed out QNANs
beq.l res_qnan_1op
#
# do the fmove in; at this point, only possible ops are ZERO and INF.
# use fmov to determine ccodes.
# prec:mode should be zero at this point but it won't affect answer anyways.
#
fmov.x SRC(%a0),%fp0 # do fmove in
fmov.l %fpsr,%d0 # no exceptions possible
rol.l &0x8,%d0 # put ccodes in lo byte
mov.b %d0,FPSR_CC(%a6) # insert correct ccodes
rts
#########################################################################
# XDEF **************************************************************** #
# fdiv(): emulates the fdiv instruction #
# fsdiv(): emulates the fsdiv instruction #
# fddiv(): emulates the fddiv instruction #
# #
# XREF **************************************************************** #
# scale_to_zero_src() - scale src exponent to zero #
# scale_to_zero_dst() - scale dst exponent to zero #
# unf_res() - return default underflow result #
# ovf_res() - return default overflow result #
# res_qnan() - return QNAN result #
# res_snan() - return SNAN result #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# a1 = pointer to extended precision destination operand #
# d0 rnd prec,mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms/denorms into ext/sgl/dbl precision. #
# For norms/denorms, scale the exponents such that a divide #
# instruction won't cause an exception. Use the regular fdiv to #
# compute a result. Check if the regular operands would have taken #
# an exception. If so, return the default overflow/underflow result #
# and return the EXOP if exceptions are enabled. Else, scale the #
# result operand to the proper exponent. #
# #
#########################################################################
align 0x10
tbl_fdiv_unfl:
long 0x3fff - 0x0000 # ext_unfl
long 0x3fff - 0x3f81 # sgl_unfl
long 0x3fff - 0x3c01 # dbl_unfl
tbl_fdiv_ovfl:
long 0x3fff - 0x7ffe # ext overflow exponent
long 0x3fff - 0x407e # sgl overflow exponent
long 0x3fff - 0x43fe # dbl overflow exponent
mov.w 2+L_SCR3(%a6),%d1 # fetch precision
lsr.b &0x6,%d1 # shift to lo bits
mov.l (%sp)+,%d0 # load S.F.
cmp.l %d0,(tbl_fdiv_ovfl.b,%pc,%d1.w*4) # will result overflow?
ble.w fdiv_may_ovfl # result will overflow
cmp.l %d0,(tbl_fdiv_unfl.w,%pc,%d1.w*4) # will result underflow?
beq.w fdiv_may_unfl # maybe
bgt.w fdiv_unfl # yes; go handle underflow
fdiv_normal:
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l L_SCR3(%a6),%fpcr # save FPCR
fmov.l &0x0,%fpsr # clear FPSR
fdiv.x FP_SCR0(%a6),%fp0 # perform divide
fmov.l %fpsr,%d1 # save FPSR
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fdiv_normal_exit:
fmovm.x &0x80,FP_SCR0(%a6) # store result on stack
mov.l %d2,-(%sp) # store d2
mov.w FP_SCR0_EX(%a6),%d1 # load {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
or.w %d2,%d1 # concat old sign,new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x80 # return result in fp0
rts
tbl_fdiv_ovfl2:
long 0x7fff
long 0x407f
long 0x43ff
fdiv_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l &rz_mode*0x10,%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fdiv.x FP_SCR0(%a6),%fp0 # execute divide
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fdiv_unfl_ena # yes
fdiv_unfl_dis:
fmovm.x &0x80,FP_SCR0(%a6) # store out result
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # 'Z' may have been set
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# UNFL is enabled.
#
fdiv_unfl_ena:
fmovm.x FP_SCR1(%a6),&0x40 # load dst op
mov.l L_SCR3(%a6),%d1
andi.b &0xc0,%d1 # is precision extended?
bne.b fdiv_unfl_ena_sd # no, sgl or dbl
fmovm.x &0x40,FP_SCR0(%a6) # save result to stack
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factoer
addi.l &0x6000,%d1 # add bias
andi.w &0x7fff,%d1
or.w %d2,%d1 # concat old sign,new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exp
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
bra.w fdiv_unfl_dis
fdiv_unfl_ena_sd:
mov.l L_SCR3(%a6),%d1
andi.b &0x30,%d1 # use only rnd mode
fmov.l %d1,%fpcr # set FPCR
bra.b fdiv_unfl_ena_cont
#
# the divide operation MAY underflow:
#
fdiv_may_unfl:
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fdiv.x FP_SCR0(%a6),%fp0 # execute divide
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x1 # is |result| > 1.b?
fbgt.w fdiv_normal_exit # no; no underflow occurred
fblt.w fdiv_unfl # yes; underflow occurred
#
# we still don't know if underflow occurred. result is ~ equal to 1. but,
# we don't know if the result was an underflow that rounded up to a 1
# or a normalized number that rounded down to a 1. so, redo the entire
# operation using RZ as the rounding mode to see what the pre-rounded
# result is. this case should be relatively rare.
#
fmovm.x FP_SCR1(%a6),&0x40 # load dst op into fp1
#
# Divide: inputs are not both normalized; what are they?
#
fdiv_not_norm:
mov.w (tbl_fdiv_op.b,%pc,%d1.w*2),%d1
jmp (tbl_fdiv_op.b,%pc,%d1.w*1)
swbeg &48
tbl_fdiv_op:
short fdiv_norm - tbl_fdiv_op # NORM / NORM
short fdiv_inf_load - tbl_fdiv_op # NORM / ZERO
short fdiv_zero_load - tbl_fdiv_op # NORM / INF
short fdiv_res_qnan - tbl_fdiv_op # NORM / QNAN
short fdiv_norm - tbl_fdiv_op # NORM / DENORM
short fdiv_res_snan - tbl_fdiv_op # NORM / SNAN
short tbl_fdiv_op - tbl_fdiv_op #
short tbl_fdiv_op - tbl_fdiv_op #
short fdiv_zero_load - tbl_fdiv_op # ZERO / NORM
short fdiv_res_operr - tbl_fdiv_op # ZERO / ZERO
short fdiv_zero_load - tbl_fdiv_op # ZERO / INF
short fdiv_res_qnan - tbl_fdiv_op # ZERO / QNAN
short fdiv_zero_load - tbl_fdiv_op # ZERO / DENORM
short fdiv_res_snan - tbl_fdiv_op # ZERO / SNAN
short tbl_fdiv_op - tbl_fdiv_op #
short tbl_fdiv_op - tbl_fdiv_op #
short fdiv_inf_dst - tbl_fdiv_op # INF / NORM
short fdiv_inf_dst - tbl_fdiv_op # INF / ZERO
short fdiv_res_operr - tbl_fdiv_op # INF / INF
short fdiv_res_qnan - tbl_fdiv_op # INF / QNAN
short fdiv_inf_dst - tbl_fdiv_op # INF / DENORM
short fdiv_res_snan - tbl_fdiv_op # INF / SNAN
short tbl_fdiv_op - tbl_fdiv_op #
short tbl_fdiv_op - tbl_fdiv_op #
short fdiv_res_qnan - tbl_fdiv_op # QNAN / NORM
short fdiv_res_qnan - tbl_fdiv_op # QNAN / ZERO
short fdiv_res_qnan - tbl_fdiv_op # QNAN / INF
short fdiv_res_qnan - tbl_fdiv_op # QNAN / QNAN
short fdiv_res_qnan - tbl_fdiv_op # QNAN / DENORM
short fdiv_res_snan - tbl_fdiv_op # QNAN / SNAN
short tbl_fdiv_op - tbl_fdiv_op #
short tbl_fdiv_op - tbl_fdiv_op #
short fdiv_norm - tbl_fdiv_op # DENORM / NORM
short fdiv_inf_load - tbl_fdiv_op # DENORM / ZERO
short fdiv_zero_load - tbl_fdiv_op # DENORM / INF
short fdiv_res_qnan - tbl_fdiv_op # DENORM / QNAN
short fdiv_norm - tbl_fdiv_op # DENORM / DENORM
short fdiv_res_snan - tbl_fdiv_op # DENORM / SNAN
short tbl_fdiv_op - tbl_fdiv_op #
short tbl_fdiv_op - tbl_fdiv_op #
short fdiv_res_snan - tbl_fdiv_op # SNAN / NORM
short fdiv_res_snan - tbl_fdiv_op # SNAN / ZERO
short fdiv_res_snan - tbl_fdiv_op # SNAN / INF
short fdiv_res_snan - tbl_fdiv_op # SNAN / QNAN
short fdiv_res_snan - tbl_fdiv_op # SNAN / DENORM
short fdiv_res_snan - tbl_fdiv_op # SNAN / SNAN
short tbl_fdiv_op - tbl_fdiv_op #
short tbl_fdiv_op - tbl_fdiv_op #
global fdiv_zero_load # global for fsgldiv
fdiv_zero_load:
mov.b SRC_EX(%a0),%d0 # result sign is exclusive
mov.b DST_EX(%a1),%d1 # or of input signs.
eor.b %d0,%d1
bpl.b fdiv_zero_load_p # result is positive
fmov.s &0x80000000,%fp0 # load a -ZERO
mov.b &z_bmask+neg_bmask,FPSR_CC(%a6) # set Z/N
rts
fdiv_zero_load_p:
fmov.s &0x00000000,%fp0 # load a +ZERO
mov.b &z_bmask,FPSR_CC(%a6) # set Z
rts
#
# The destination was In Range and the source was a ZERO. The result,
# Therefore, is an INF w/ the proper sign.
# So, determine the sign and return a new INF (w/ the j-bit cleared).
# global fdiv_inf_load # global for fsgldiv
fdiv_inf_load:
ori.w &dz_mask+adz_mask,2+USER_FPSR(%a6) # no; set DZ/ADZ
mov.b SRC_EX(%a0),%d0 # load both signs
mov.b DST_EX(%a1),%d1
eor.b %d0,%d1
bpl.b fdiv_inf_load_p # result is positive
fmov.s &0xff800000,%fp0 # make result -INF
mov.b &inf_bmask+neg_bmask,FPSR_CC(%a6) # set INF/N
rts
fdiv_inf_load_p:
fmov.s &0x7f800000,%fp0 # make result +INF
mov.b &inf_bmask,FPSR_CC(%a6) # set INF
rts
#
# The destination was an INF w/ an In Range or ZERO source, the result is
# an INF w/ the proper sign.
# The 68881/882 returns the destination INF w/ the new sign(if the j-bit of the
# dst INF is set, then then j-bit of the result INF is also set).
# global fdiv_inf_dst # global for fsgldiv
fdiv_inf_dst:
mov.b DST_EX(%a1),%d0 # load both signs
mov.b SRC_EX(%a0),%d1
eor.b %d0,%d1
bpl.b fdiv_inf_dst_p # result is positive
fmovm.x DST(%a1),&0x80 # return result in fp0
fabs.x %fp0 # clear sign bit
fneg.x %fp0 # set sign bit
mov.b &inf_bmask+neg_bmask,FPSR_CC(%a6) # set INF/NEG
rts
fdiv_inf_dst_p:
fmovm.x DST(%a1),&0x80 # return result in fp0
fabs.x %fp0 # return positive INF
mov.b &inf_bmask,FPSR_CC(%a6) # set INF
rts
#########################################################################
# XDEF **************************************************************** #
# fneg(): emulates the fneg instruction #
# fsneg(): emulates the fsneg instruction #
# fdneg(): emulates the fdneg instruction #
# #
# XREF **************************************************************** #
# norm() - normalize a denorm to provide EXOP #
# scale_to_zero_src() - scale sgl/dbl source exponent #
# ovf_res() - return default overflow result #
# unf_res() - return default underflow result #
# res_qnan_1op() - return QNAN result #
# res_snan_1op() - return SNAN result #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# d0 = rnd prec,mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, zeroes, and infinities as special cases. Separate #
# norms/denorms into ext/sgl/dbl precisions. Extended precision can be #
# emulated by simply setting sign bit. Sgl/dbl operands must be scaled #
# and an actual fneg performed to see if overflow/underflow would have #
# occurred. If so, return default underflow/overflow result. Else, #
# scale the result exponent and return result. FPSR gets set based on #
# the result value. #
# #
#########################################################################
global fneg
fneg:
mov.l %d0,L_SCR3(%a6) # store rnd info
mov.b STAG(%a6),%d1
bne.w fneg_not_norm # optimize on non-norm input
#
# NEGATE SIGN : norms and denorms ONLY!
#
fneg_norm:
andi.b &0xc0,%d0 # is precision extended?
bne.w fneg_not_ext # no; go handle sgl or dbl
#
# precision selected is extended. so...we can not get an underflow
# or overflow because of rounding to the correct precision. so...
# skip the scaling and unscaling...
#
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
mov.w SRC_EX(%a0),%d0
eori.w &0x8000,%d0 # negate sign
bpl.b fneg_norm_load # sign is positive
mov.b &neg_bmask,FPSR_CC(%a6) # set'N' ccode bit
fneg_norm_load:
mov.w %d0,FP_SCR0_EX(%a6)
fmovm.x FP_SCR0(%a6),&0x80 # return result in fp0
rts
#
# for an extended precision DENORM, the UNFL exception bit is set
# the accrued bit is NOT set in this instance(no inexactness!)
#
fneg_denorm:
andi.b &0xc0,%d0 # is precision extended?
bne.b fneg_not_ext # no; go handle sgl or dbl
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
mov.w SRC_EX(%a0),%d0
eori.w &0x8000,%d0 # negate sign
bpl.b fneg_denorm_done # no
mov.b &neg_bmask,FPSR_CC(%a6) # yes, set'N' ccode bit
fneg_denorm_done:
mov.w %d0,FP_SCR0_EX(%a6)
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
#
# the input is an extended DENORM and underflow is enabled in the FPCR.
# normalize the mantissa and add the bias of 0x6000 to the resulting negative
# exponent and insert back into the operand.
#
fneg_ext_unfl_ena:
lea FP_SCR0(%a6),%a0 # pass: ptr to operand
bsr.l norm # normalize result
neg.w %d0 # new exponent = -(shft val)
addi.w &0x6000,%d0 # add new bias to exponent
mov.w FP_SCR0_EX(%a6),%d1 # fetch old sign,exp
andi.w &0x8000,%d1 # keep old sign
andi.w &0x7fff,%d0 # clear sign position
or.w %d1,%d0 # concat old sign, new exponent
mov.w %d0,FP_SCR0_EX(%a6) # insert new exponent
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
rts
#
# operand is either single or double
#
fneg_not_ext:
cmpi.b %d0,&s_mode*0x10 # separate sgl/dbl prec
bne.b fneg_dbl
#
# operand is to be rounded to single precision
#
fneg_sgl:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_to_zero_src # calculate scale factor
cmpi.l %d0,&0x3fff-0x3f80 # will move in underflow?
bge.w fneg_sd_unfl # yes; go handle underflow
cmpi.l %d0,&0x3fff-0x407e # will move in overflow?
beq.w fneg_sd_may_ovfl # maybe; go check
blt.w fneg_sd_ovfl # yes; go handle overflow
#
# operand will NOT overflow or underflow when moved in to the fp reg file
#
fneg_sd_normal:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fneg.x FP_SCR0(%a6),%fp0 # perform negation
fmov.l %fpsr,%d1 # save FPSR
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fneg_sd_normal_exit:
mov.l %d2,-(%sp) # save d2
fmovm.x &0x80,FP_SCR0(%a6) # store out result
mov.w FP_SCR0_EX(%a6),%d1 # load sgn,exp
mov.w %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign sub.l %d0,%d1 # add scale factor
andi.w &0x8000,%d2 # keep old sign
or.w %d1,%d2 # concat old sign,new exp
mov.w %d2,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x80 # return result in fp0
rts
#
# operand is to be rounded to double precision
#
fneg_dbl:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_to_zero_src # calculate scale factor
cmpi.l %d0,&0x3fff-0x3c00 # will move in underflow?
bge.b fneg_sd_unfl # yes; go handle underflow
cmpi.l %d0,&0x3fff-0x43fe # will move in overflow?
beq.w fneg_sd_may_ovfl # maybe; go check
blt.w fneg_sd_ovfl # yes; go handle overflow
bra.w fneg_sd_normal # no; ho handle normalized op
#
# operand WILL underflow when moved in to the fp register file
#
fneg_sd_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
# if underflow or inexact is enabled, go calculate EXOP first.
fneg_sd_unfl_tst:
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fneg_sd_unfl_ena # yes
fneg_sd_unfl_dis:
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # unf_res may have set'Z'
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# operand will underflow AND underflow is enabled.
# Therefore, we must return the result rounded to extended precision.
#
fneg_sd_unfl_ena:
mov.l FP_SCR0_HI(%a6),FP_SCR1_HI(%a6)
mov.l FP_SCR0_LO(%a6),FP_SCR1_LO(%a6)
mov.w FP_SCR0_EX(%a6),%d1 # load current exponent
mov.l %d2,-(%sp) # save d2
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # subtract scale factor
addi.l &0x6000,%d1 # add new bias
andi.w &0x7fff,%d1
or.w %d2,%d1 # concat new sign,new exp
mov.w %d1,FP_SCR1_EX(%a6) # insert new exp
fmovm.x FP_SCR1(%a6),&0x40 # return EXOP in fp1
mov.l (%sp)+,%d2 # restore d2
bra.b fneg_sd_unfl_dis
#
# operand WILL overflow.
#
fneg_sd_ovfl:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fneg.x FP_SCR0(%a6),%fp0 # perform negation
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d1 # save FPSR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fneg_sd_ovfl_tst:
or.l &ovfl_inx_mask,USER_FPSR(%a6) # set ovfl/aovfl/ainex
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x13,%d1 # is OVFL or INEX enabled?
bne.b fneg_sd_ovfl_ena # yes
#
# OVFL is not enabled; therefore, we must create the default result by
# calling ovf_res().
#
fneg_sd_ovfl_dis:
btst &neg_bit,FPSR_CC(%a6) # is result negative?
sne %d1 # set sign param accordingly
mov.l L_SCR3(%a6),%d0 # pass: prec,mode
bsr.l ovf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # set INF,N if applicable
fmovm.x (%a0),&0x80 # return default result in fp0
rts
#
# OVFL is enabled.
# the INEX2 bit has already been updated by the round to the correct precision.
# now, round to extended(and don't alter the FPSR).
#
fneg_sd_ovfl_ena:
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
subi.l &0x6000,%d1 # subtract bias
andi.w &0x7fff,%d1
or.w %d2,%d1 # concat sign,exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
mov.l (%sp)+,%d2 # restore d2
bra.b fneg_sd_ovfl_dis
#
# the move in MAY underflow. so...
#
fneg_sd_may_ovfl:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fneg.x FP_SCR0(%a6),%fp0 # perform negation
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x2 # is |result| >= 2.b?
fbge.w fneg_sd_ovfl_tst # yes; overflow has occurred
# no, it didn't overflow; we have correct result
bra.w fneg_sd_normal_exit
#
# input is not normalized; what is it?
#
fneg_not_norm:
cmpi.b %d1,&DENORM # weed out DENORM
beq.w fneg_denorm
cmpi.b %d1,&SNAN # weed out SNAN
beq.l res_snan_1op
cmpi.b %d1,&QNAN # weed out QNAN
beq.l res_qnan_1op
#
# do the fneg; at this point, only possible ops are ZERO and INF.
# use fneg to determine ccodes.
# prec:mode should be zero at this point but it won't affect answer anyways.
#
fneg.x SRC_EX(%a0),%fp0 # do fneg
fmov.l %fpsr,%d0
rol.l &0x8,%d0 # put ccodes in lo byte
mov.b %d0,FPSR_CC(%a6) # insert correct ccodes
rts
#########################################################################
# XDEF **************************************************************** #
# ftst(): emulates the ftest instruction #
# #
# XREF **************************************************************** #
# res{s,q}nan_1op() - set NAN result for monadic instruction #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# #
# OUTPUT ************************************************************** #
# none #
# #
# ALGORITHM *********************************************************** #
# Check the source operand tag (STAG) and set the FPCR according #
# to the operand type and sign. #
# #
#########################################################################
global ftst
ftst:
mov.b STAG(%a6),%d1
bne.b ftst_not_norm # optimize on non-norm input
#
# input is not normalized; what is it?
#
ftst_not_norm:
cmpi.b %d1,&ZERO # weed out ZERO
beq.b ftst_zero
cmpi.b %d1,&INF # weed out INF
beq.b ftst_inf
cmpi.b %d1,&SNAN # weed out SNAN
beq.l res_snan_1op
cmpi.b %d1,&QNAN # weed out QNAN
beq.l res_qnan_1op
#########################################################################
# XDEF **************************************************************** #
# fint(): emulates the fint instruction #
# #
# XREF **************************************************************** #
# res_{s,q}nan_1op() - set NAN result for monadic operation #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# d0 = round precision/mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# #
# ALGORITHM *********************************************************** #
# Separate according to operand type. Unnorms don't pass through #
# here. For norms, load the rounding mode/prec, execute a "fint", then #
# store the resulting FPSR bits. #
# For denorms, force the j-bit to a one and do the same as for #
# norms. Denorms are so low that the answer will either be a zero or a #
# one. #
# For zeroes/infs/NANs, return the same while setting the FPSR #
# as appropriate. #
# #
#########################################################################
global fint
fint:
mov.b STAG(%a6),%d1
bne.b fint_not_norm # optimize on non-norm input
fmov.l %d0,%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fint.x SRC(%a0),%fp0 # execute fint
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d0 # save FPSR
or.l %d0,USER_FPSR(%a6) # set exception bits
rts
#
# input is not normalized; what is it?
#
fint_not_norm:
cmpi.b %d1,&ZERO # weed out ZERO
beq.b fint_zero
cmpi.b %d1,&INF # weed out INF
beq.b fint_inf
cmpi.b %d1,&DENORM # weed out DENORM
beq.b fint_denorm
cmpi.b %d1,&SNAN # weed out SNAN
beq.l res_snan_1op
bra.l res_qnan_1op # weed out QNAN
#
# Denorm:
#
# for DENORMs, the result will be either (+/-)ZERO or (+/-)1.
# also, the INEX2 and AINEX exception bits will be set.
# so, we could either set these manually or force the DENORM
# to a very small NORM and ship it to the NORM routine.
# I do the latter.
#
fint_denorm:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6) # copy sign, zero exp
mov.b &0x80,FP_SCR0_HI(%a6) # force DENORM ==> small NORM
lea FP_SCR0(%a6),%a0
bra.b fint_norm
#
# Zero:
#
fint_zero:
tst.b SRC_EX(%a0) # is ZERO negative?
bmi.b fint_zero_m # yes
fint_zero_p:
fmov.s &0x00000000,%fp0 # return +ZERO in fp0
mov.b &z_bmask,FPSR_CC(%a6) # set'Z' ccode bit
rts
fint_zero_m:
fmov.s &0x80000000,%fp0 # return -ZERO in fp0
mov.b &z_bmask+neg_bmask,FPSR_CC(%a6) # set'Z','N' ccode bits
rts
#
# Infinity:
#
fint_inf:
fmovm.x SRC(%a0),&0x80 # return result in fp0
tst.b SRC_EX(%a0) # is INF negative?
bmi.b fint_inf_m # yes
fint_inf_p:
mov.b &inf_bmask,FPSR_CC(%a6) # set'I' ccode bit
rts
fint_inf_m:
mov.b &inf_bmask+neg_bmask,FPSR_CC(%a6) # set'N','I' ccode bits
rts
#########################################################################
# XDEF **************************************************************** #
# fintrz(): emulates the fintrz instruction #
# #
# XREF **************************************************************** #
# res_{s,q}nan_1op() - set NAN result for monadic operation #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# d0 = round precision/mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# #
# ALGORITHM *********************************************************** #
# Separate according to operand type. Unnorms don't pass through #
# here. For norms, load the rounding mode/prec, execute a "fintrz", #
# then store the resulting FPSR bits. #
# For denorms, force the j-bit to a one and do the same as for #
# norms. Denorms are so low that the answer will either be a zero or a #
# one. #
# For zeroes/infs/NANs, return the same while setting the FPSR #
# as appropriate. #
# #
#########################################################################
global fintrz
fintrz:
mov.b STAG(%a6),%d1
bne.b fintrz_not_norm # optimize on non-norm input
fmov.l %fpsr,%d0 # save FPSR
or.l %d0,USER_FPSR(%a6) # set exception bits
rts
#
# input is not normalized; what is it?
#
fintrz_not_norm:
cmpi.b %d1,&ZERO # weed out ZERO
beq.b fintrz_zero
cmpi.b %d1,&INF # weed out INF
beq.b fintrz_inf
cmpi.b %d1,&DENORM # weed out DENORM
beq.b fintrz_denorm
cmpi.b %d1,&SNAN # weed out SNAN
beq.l res_snan_1op
bra.l res_qnan_1op # weed out QNAN
#
# Denorm:
#
# for DENORMs, the result will be (+/-)ZERO.
# also, the INEX2 and AINEX exception bits will be set.
# so, we could either set these manually or force the DENORM
# to a very small NORM and ship it to the NORM routine.
# I do the latter.
#
fintrz_denorm:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6) # copy sign, zero exp
mov.b &0x80,FP_SCR0_HI(%a6) # force DENORM ==> small NORM
lea FP_SCR0(%a6),%a0
bra.b fintrz_norm
#
# Zero:
#
fintrz_zero:
tst.b SRC_EX(%a0) # is ZERO negative?
bmi.b fintrz_zero_m # yes
fintrz_zero_p:
fmov.s &0x00000000,%fp0 # return +ZERO in fp0
mov.b &z_bmask,FPSR_CC(%a6) # set'Z' ccode bit
rts
fintrz_zero_m:
fmov.s &0x80000000,%fp0 # return -ZERO in fp0
mov.b &z_bmask+neg_bmask,FPSR_CC(%a6) # set'Z','N' ccode bits
rts
#
# Infinity:
#
fintrz_inf:
fmovm.x SRC(%a0),&0x80 # return result in fp0
tst.b SRC_EX(%a0) # is INF negative?
bmi.b fintrz_inf_m # yes
fintrz_inf_p:
mov.b &inf_bmask,FPSR_CC(%a6) # set'I' ccode bit
rts
fintrz_inf_m:
mov.b &inf_bmask+neg_bmask,FPSR_CC(%a6) # set'N','I' ccode bits
rts
#########################################################################
# XDEF **************************************************************** #
# fabs(): emulates the fabs instruction #
# fsabs(): emulates the fsabs instruction #
# fdabs(): emulates the fdabs instruction #
# #
# XREF **************************************************************** #
# norm() - normalize denorm mantissa to provide EXOP #
# scale_to_zero_src() - make exponent. = 0; get scale factor #
# unf_res() - calculate underflow result #
# ovf_res() - calculate overflow result #
# res_{s,q}nan_1op() - set NAN result for monadic operation #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# d0 = rnd precision/mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms into extended, single, and double precision. #
# Simply clear sign for extended precision norm. Ext prec denorm #
# gets an EXOP created for it since it's an underflow. #
# Double and single precision can overflow and underflow. First, #
# scale the operand such that the exponent is zero. Perform an "fabs" #
# using the correct rnd mode/prec. Check to see if the original #
# exponent would take an exception. If so, use unf_res() or ovf_res() #
# to calculate the default result. Also, create the EXOP for the #
# exceptional case. If no exception should occur, insert the correct #
# result exponent and return. #
# Unnorms don't pass through here. #
# #
#########################################################################
global fabs
fabs:
mov.l %d0,L_SCR3(%a6) # store rnd info
mov.b STAG(%a6),%d1
bne.w fabs_not_norm # optimize on non-norm input
#
# ABSOLUTE VALUE: norms and denorms ONLY!
#
fabs_norm:
andi.b &0xc0,%d0 # is precision extended?
bne.b fabs_not_ext # no; go handle sgl or dbl
#
# precision selected is extended. so...we can not get an underflow
# or overflow because of rounding to the correct precision. so...
# skip the scaling and unscaling...
#
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
mov.w SRC_EX(%a0),%d1
bclr &15,%d1 # force absolute value
mov.w %d1,FP_SCR0_EX(%a6) # insert exponent
fmovm.x FP_SCR0(%a6),&0x80 # return result in fp0
rts
#
# for an extended precision DENORM, the UNFL exception bit is set
# the accrued bit is NOT set in this instance(no inexactness!)
#
fabs_denorm:
andi.b &0xc0,%d0 # is precision extended?
bne.b fabs_not_ext # no
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
btst &unfl_bit,FPCR_ENABLE(%a6) # is UNFL enabled?
bne.b fabs_ext_unfl_ena
rts
#
# the input is an extended DENORM and underflow is enabled in the FPCR.
# normalize the mantissa and add the bias of 0x6000 to the resulting negative
# exponent and insert back into the operand.
#
fabs_ext_unfl_ena:
lea FP_SCR0(%a6),%a0 # pass: ptr to operand
bsr.l norm # normalize result
neg.w %d0 # new exponent = -(shft val)
addi.w &0x6000,%d0 # add new bias to exponent
mov.w FP_SCR0_EX(%a6),%d1 # fetch old sign,exp
andi.w &0x8000,%d1 # keep old sign
andi.w &0x7fff,%d0 # clear sign position
or.w %d1,%d0 # concat old sign, new exponent
mov.w %d0,FP_SCR0_EX(%a6) # insert new exponent
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
rts
#
# operand is either single or double
#
fabs_not_ext:
cmpi.b %d0,&s_mode*0x10 # separate sgl/dbl prec
bne.b fabs_dbl
#
# operand is to be rounded to single precision
#
fabs_sgl:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_to_zero_src # calculate scale factor
cmpi.l %d0,&0x3fff-0x3f80 # will move in underflow?
bge.w fabs_sd_unfl # yes; go handle underflow
cmpi.l %d0,&0x3fff-0x407e # will move in overflow?
beq.w fabs_sd_may_ovfl # maybe; go check
blt.w fabs_sd_ovfl # yes; go handle overflow
#
# operand will NOT overflow or underflow when moved in to the fp reg file
#
fabs_sd_normal:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fabs.x FP_SCR0(%a6),%fp0 # perform absolute
fmov.l %fpsr,%d1 # save FPSR
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs_sd_normal_exit:
mov.l %d2,-(%sp) # save d2
fmovm.x &0x80,FP_SCR0(%a6) # store out result
mov.w FP_SCR0_EX(%a6),%d1 # load sgn,exp
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign sub.l %d0,%d1 # add scale factor
andi.w &0x8000,%d2 # keep old sign
or.w %d1,%d2 # concat old sign,new exp
mov.w %d2,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x80 # return result in fp0
rts
#
# operand is to be rounded to double precision
#
fabs_dbl:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_to_zero_src # calculate scale factor
cmpi.l %d0,&0x3fff-0x3c00 # will move in underflow?
bge.b fabs_sd_unfl # yes; go handle underflow
cmpi.l %d0,&0x3fff-0x43fe # will move in overflow?
beq.w fabs_sd_may_ovfl # maybe; go check
blt.w fabs_sd_ovfl # yes; go handle overflow
bra.w fabs_sd_normal # no; ho handle normalized op
#
# operand WILL underflow when moved in to the fp register file
#
fabs_sd_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
bclr &0x7,FP_SCR0_EX(%a6) # force absolute value
# if underflow or inexact is enabled, go calculate EXOP first.
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fabs_sd_unfl_ena # yes
fabs_sd_unfl_dis:
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # set possible 'Z' ccode
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# operand will underflow AND underflow is enabled.
# Therefore, we must return the result rounded to extended precision.
#
fabs_sd_unfl_ena:
mov.l FP_SCR0_HI(%a6),FP_SCR1_HI(%a6)
mov.l FP_SCR0_LO(%a6),FP_SCR1_LO(%a6)
mov.w FP_SCR0_EX(%a6),%d1 # load current exponent
mov.l %d2,-(%sp) # save d2
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # subtract scale factor
addi.l &0x6000,%d1 # add new bias
andi.w &0x7fff,%d1
or.w %d2,%d1 # concat new sign,new exp
mov.w %d1,FP_SCR1_EX(%a6) # insert new exp
fmovm.x FP_SCR1(%a6),&0x40 # return EXOP in fp1
mov.l (%sp)+,%d2 # restore d2
bra.b fabs_sd_unfl_dis
#
# operand WILL overflow.
#
fabs_sd_ovfl:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fabs.x FP_SCR0(%a6),%fp0 # perform absolute
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d1 # save FPSR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs_sd_ovfl_tst:
or.l &ovfl_inx_mask,USER_FPSR(%a6) # set ovfl/aovfl/ainex
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x13,%d1 # is OVFL or INEX enabled?
bne.b fabs_sd_ovfl_ena # yes
#
# OVFL is not enabled; therefore, we must create the default result by
# calling ovf_res().
#
fabs_sd_ovfl_dis:
btst &neg_bit,FPSR_CC(%a6) # is result negative?
sne %d1 # set sign param accordingly
mov.l L_SCR3(%a6),%d0 # pass: prec,mode
bsr.l ovf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # set INF,N if applicable
fmovm.x (%a0),&0x80 # return default result in fp0
rts
#
# OVFL is enabled.
# the INEX2 bit has already been updated by the round to the correct precision.
# now, round to extended(and don't alter the FPSR).
#
fabs_sd_ovfl_ena:
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
subi.l &0x6000,%d1 # subtract bias
andi.w &0x7fff,%d1
or.w %d2,%d1 # concat sign,exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
mov.l (%sp)+,%d2 # restore d2
bra.b fabs_sd_ovfl_dis
#
# the move in MAY underflow. so...
#
fabs_sd_may_ovfl:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fabs.x FP_SCR0(%a6),%fp0 # perform absolute
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x2 # is |result| >= 2.b?
fbge.w fabs_sd_ovfl_tst # yes; overflow has occurred
# no, it didn't overflow; we have correct result
bra.w fabs_sd_normal_exit
#
# input is not normalized; what is it?
#
fabs_not_norm:
cmpi.b %d1,&DENORM # weed out DENORM
beq.w fabs_denorm
cmpi.b %d1,&SNAN # weed out SNAN
beq.l res_snan_1op
cmpi.b %d1,&QNAN # weed out QNAN
beq.l res_qnan_1op
fabs.x SRC(%a0),%fp0 # force absolute value
cmpi.b %d1,&INF # weed out INF
beq.b fabs_inf
fabs_zero:
mov.b &z_bmask,FPSR_CC(%a6) # set'Z' ccode bit
rts
fabs_inf:
mov.b &inf_bmask,FPSR_CC(%a6) # set'I' ccode bit
rts
#########################################################################
# XDEF **************************************************************** #
# fcmp(): fp compare op routine #
# #
# XREF **************************************************************** #
# res_qnan() - return QNAN result #
# res_snan() - return SNAN result #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# a1 = pointer to extended precision destination operand #
# d0 = round prec/mode #
# #
# OUTPUT ************************************************************** #
# None #
# #
# ALGORITHM *********************************************************** #
# Handle NANs and denorms as special cases. For everything else, #
# just use the actual fcmp instruction to produce the correct condition #
# codes. #
# #
#########################################################################
global fcmp
fcmp: clr.w %d1
mov.b DTAG(%a6),%d1
lsl.b &0x3,%d1
or.b STAG(%a6),%d1
bne.b fcmp_not_norm # optimize on non-norm input
#
# COMPARE FP OPs : NORMs, ZEROs, INFs, and "corrected" DENORMs
#
fcmp_norm:
fmovm.x DST(%a1),&0x80 # load dst op
fcmp.x %fp0,SRC(%a0) # do compare
fmov.l %fpsr,%d0 # save FPSR
rol.l &0x8,%d0 # extract ccode bits
mov.b %d0,FPSR_CC(%a6) # set ccode bits(no exc bits are set)
rts
#
# fcmp: inputs are not both normalized; what are they?
#
fcmp_not_norm:
mov.w (tbl_fcmp_op.b,%pc,%d1.w*2),%d1
jmp (tbl_fcmp_op.b,%pc,%d1.w*1)
swbeg &48
tbl_fcmp_op:
short fcmp_norm - tbl_fcmp_op # NORM - NORM
short fcmp_norm - tbl_fcmp_op # NORM - ZERO
short fcmp_norm - tbl_fcmp_op # NORM - INF
short fcmp_res_qnan - tbl_fcmp_op # NORM - QNAN
short fcmp_nrm_dnrm - tbl_fcmp_op # NORM - DENORM
short fcmp_res_snan - tbl_fcmp_op # NORM - SNAN
short tbl_fcmp_op - tbl_fcmp_op #
short tbl_fcmp_op - tbl_fcmp_op #
short fcmp_norm - tbl_fcmp_op # ZERO - NORM
short fcmp_norm - tbl_fcmp_op # ZERO - ZERO
short fcmp_norm - tbl_fcmp_op # ZERO - INF
short fcmp_res_qnan - tbl_fcmp_op # ZERO - QNAN
short fcmp_dnrm_s - tbl_fcmp_op # ZERO - DENORM
short fcmp_res_snan - tbl_fcmp_op # ZERO - SNAN
short tbl_fcmp_op - tbl_fcmp_op #
short tbl_fcmp_op - tbl_fcmp_op #
short fcmp_norm - tbl_fcmp_op # INF - NORM
short fcmp_norm - tbl_fcmp_op # INF - ZERO
short fcmp_norm - tbl_fcmp_op # INF - INF
short fcmp_res_qnan - tbl_fcmp_op # INF - QNAN
short fcmp_dnrm_s - tbl_fcmp_op # INF - DENORM
short fcmp_res_snan - tbl_fcmp_op # INF - SNAN
short tbl_fcmp_op - tbl_fcmp_op #
short tbl_fcmp_op - tbl_fcmp_op #
short fcmp_res_qnan - tbl_fcmp_op # QNAN - NORM
short fcmp_res_qnan - tbl_fcmp_op # QNAN - ZERO
short fcmp_res_qnan - tbl_fcmp_op # QNAN - INF
short fcmp_res_qnan - tbl_fcmp_op # QNAN - QNAN
short fcmp_res_qnan - tbl_fcmp_op # QNAN - DENORM
short fcmp_res_snan - tbl_fcmp_op # QNAN - SNAN
short tbl_fcmp_op - tbl_fcmp_op #
short tbl_fcmp_op - tbl_fcmp_op #
short fcmp_dnrm_nrm - tbl_fcmp_op # DENORM - NORM
short fcmp_dnrm_d - tbl_fcmp_op # DENORM - ZERO
short fcmp_dnrm_d - tbl_fcmp_op # DENORM - INF
short fcmp_res_qnan - tbl_fcmp_op # DENORM - QNAN
short fcmp_dnrm_sd - tbl_fcmp_op # DENORM - DENORM
short fcmp_res_snan - tbl_fcmp_op # DENORM - SNAN
short tbl_fcmp_op - tbl_fcmp_op #
short tbl_fcmp_op - tbl_fcmp_op #
short fcmp_res_snan - tbl_fcmp_op # SNAN - NORM
short fcmp_res_snan - tbl_fcmp_op # SNAN - ZERO
short fcmp_res_snan - tbl_fcmp_op # SNAN - INF
short fcmp_res_snan - tbl_fcmp_op # SNAN - QNAN
short fcmp_res_snan - tbl_fcmp_op # SNAN - DENORM
short fcmp_res_snan - tbl_fcmp_op # SNAN - SNAN
short tbl_fcmp_op - tbl_fcmp_op #
short tbl_fcmp_op - tbl_fcmp_op #
# unlike all other functions for QNAN and SNAN, fcmp does NOT set the
# 'N' bit for a negative QNAN or SNAN input so we must squelch it here.
fcmp_res_qnan:
bsr.l res_qnan
andi.b &0xf7,FPSR_CC(%a6)
rts
fcmp_res_snan:
bsr.l res_snan
andi.b &0xf7,FPSR_CC(%a6)
rts
#
# DENORMs are a little more difficult.
# If you have a 2 DENORMs, then you can just force the j-bit to a one
# and use the fcmp_norm routine.
# If you have a DENORM and an INF or ZERO, just force the DENORM's j-bit to a one
# and use the fcmp_norm routine.
# If you have a DENORM and a NORM with opposite signs, then use fcmp_norm, also.
# But with a DENORM and a NORM of the same sign, the neg bit is set if the
# (1) signs are (+) and the DENORM is the dst or
# (2) signs are (-) and the DENORM is the src
#
fcmp_dnrm_s:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),%d0
bset &31,%d0 # DENORM src; make into small norm
mov.l %d0,FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
lea FP_SCR0(%a6),%a0
bra.w fcmp_norm
fcmp_dnrm_d:
mov.l DST_EX(%a1),FP_SCR0_EX(%a6)
mov.l DST_HI(%a1),%d0
bset &31,%d0 # DENORM src; make into small norm
mov.l %d0,FP_SCR0_HI(%a6)
mov.l DST_LO(%a1),FP_SCR0_LO(%a6)
lea FP_SCR0(%a6),%a1
bra.w fcmp_norm
fcmp_dnrm_sd:
mov.w DST_EX(%a1),FP_SCR1_EX(%a6)
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l DST_HI(%a1),%d0
bset &31,%d0 # DENORM dst; make into small norm
mov.l %d0,FP_SCR1_HI(%a6)
mov.l SRC_HI(%a0),%d0
bset &31,%d0 # DENORM dst; make into small norm
mov.l %d0,FP_SCR0_HI(%a6)
mov.l DST_LO(%a1),FP_SCR1_LO(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
lea FP_SCR1(%a6),%a1
lea FP_SCR0(%a6),%a0
bra.w fcmp_norm
fcmp_nrm_dnrm:
mov.b SRC_EX(%a0),%d0 # determine if like signs
mov.b DST_EX(%a1),%d1
eor.b %d0,%d1
bmi.w fcmp_dnrm_s
# signs are the same, so must determine the answer ourselves.
tst.b %d0 # is src op negative?
bmi.b fcmp_nrm_dnrm_m # yes
rts
fcmp_nrm_dnrm_m:
mov.b &neg_bmask,FPSR_CC(%a6) # set'Z' ccode bit
rts
fcmp_dnrm_nrm:
mov.b SRC_EX(%a0),%d0 # determine if like signs
mov.b DST_EX(%a1),%d1
eor.b %d0,%d1
bmi.w fcmp_dnrm_d
# signs are the same, so must determine the answer ourselves.
tst.b %d0 # is src op negative?
bpl.b fcmp_dnrm_nrm_m # no
rts
fcmp_dnrm_nrm_m:
mov.b &neg_bmask,FPSR_CC(%a6) # set'Z' ccode bit
rts
#########################################################################
# XDEF **************************************************************** #
# fsglmul(): emulates the fsglmul instruction #
# #
# XREF **************************************************************** #
# scale_to_zero_src() - scale src exponent to zero #
# scale_to_zero_dst() - scale dst exponent to zero #
# unf_res4() - return default underflow result for sglop #
# ovf_res() - return default overflow result #
# res_qnan() - return QNAN result #
# res_snan() - return SNAN result #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# a1 = pointer to extended precision destination operand #
# d0 rnd prec,mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms/denorms into ext/sgl/dbl precision. #
# For norms/denorms, scale the exponents such that a multiply #
# instruction won't cause an exception. Use the regular fsglmul to #
# compute a result. Check if the regular operands would have taken #
# an exception. If so, return the default overflow/underflow result #
# and return the EXOP if exceptions are enabled. Else, scale the #
# result operand to the proper exponent. #
# #
#########################################################################
global fsglmul
fsglmul:
mov.l %d0,L_SCR3(%a6) # store rnd info
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fsglmul_unfl_ena # yes
fsglmul_unfl_dis:
fmovm.x &0x80,FP_SCR0(%a6) # store out result
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res4 # calculate default result
or.b %d0,FPSR_CC(%a6) # 'Z' bit may have been set
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# UNFL is enabled.
#
fsglmul_unfl_ena:
fmovm.x FP_SCR1(%a6),&0x40 # load dst op
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x2 # is |result| > 2.b?
fbgt.w fsglmul_normal_exit # no; no underflow occurred
fblt.w fsglmul_unfl # yes; underflow occurred
#
# we still don't know if underflow occurred. result is ~ equal to 2. but,
# we don't know if the result was an underflow that rounded up to a 2 or
# a normalized number that rounded down to a 2. so, redo the entire operation
# using RZ as the rounding mode to see what the pre-rounded result is.
# this case should be relatively rare.
#
fmovm.x FP_SCR1(%a6),&0x40 # load dst op into fp1
#
# Single Precision Multiply: inputs are not both normalized; what are they?
#
fsglmul_not_norm:
mov.w (tbl_fsglmul_op.b,%pc,%d1.w*2),%d1
jmp (tbl_fsglmul_op.b,%pc,%d1.w*1)
swbeg &48
tbl_fsglmul_op:
short fsglmul_norm - tbl_fsglmul_op # NORM x NORM
short fsglmul_zero - tbl_fsglmul_op # NORM x ZERO
short fsglmul_inf_src - tbl_fsglmul_op # NORM x INF
short fsglmul_res_qnan - tbl_fsglmul_op # NORM x QNAN
short fsglmul_norm - tbl_fsglmul_op # NORM x DENORM
short fsglmul_res_snan - tbl_fsglmul_op # NORM x SNAN
short tbl_fsglmul_op - tbl_fsglmul_op #
short tbl_fsglmul_op - tbl_fsglmul_op #
short fsglmul_zero - tbl_fsglmul_op # ZERO x NORM
short fsglmul_zero - tbl_fsglmul_op # ZERO x ZERO
short fsglmul_res_operr - tbl_fsglmul_op # ZERO x INF
short fsglmul_res_qnan - tbl_fsglmul_op # ZERO x QNAN
short fsglmul_zero - tbl_fsglmul_op # ZERO x DENORM
short fsglmul_res_snan - tbl_fsglmul_op # ZERO x SNAN
short tbl_fsglmul_op - tbl_fsglmul_op #
short tbl_fsglmul_op - tbl_fsglmul_op #
short fsglmul_inf_dst - tbl_fsglmul_op # INF x NORM
short fsglmul_res_operr - tbl_fsglmul_op # INF x ZERO
short fsglmul_inf_dst - tbl_fsglmul_op # INF x INF
short fsglmul_res_qnan - tbl_fsglmul_op # INF x QNAN
short fsglmul_inf_dst - tbl_fsglmul_op # INF x DENORM
short fsglmul_res_snan - tbl_fsglmul_op # INF x SNAN
short tbl_fsglmul_op - tbl_fsglmul_op #
short tbl_fsglmul_op - tbl_fsglmul_op #
short fsglmul_res_qnan - tbl_fsglmul_op # QNAN x NORM
short fsglmul_res_qnan - tbl_fsglmul_op # QNAN x ZERO
short fsglmul_res_qnan - tbl_fsglmul_op # QNAN x INF
short fsglmul_res_qnan - tbl_fsglmul_op # QNAN x QNAN
short fsglmul_res_qnan - tbl_fsglmul_op # QNAN x DENORM
short fsglmul_res_snan - tbl_fsglmul_op # QNAN x SNAN
short tbl_fsglmul_op - tbl_fsglmul_op #
short tbl_fsglmul_op - tbl_fsglmul_op #
short fsglmul_norm - tbl_fsglmul_op # NORM x NORM
short fsglmul_zero - tbl_fsglmul_op # NORM x ZERO
short fsglmul_inf_src - tbl_fsglmul_op # NORM x INF
short fsglmul_res_qnan - tbl_fsglmul_op # NORM x QNAN
short fsglmul_norm - tbl_fsglmul_op # NORM x DENORM
short fsglmul_res_snan - tbl_fsglmul_op # NORM x SNAN
short tbl_fsglmul_op - tbl_fsglmul_op #
short tbl_fsglmul_op - tbl_fsglmul_op #
short fsglmul_res_snan - tbl_fsglmul_op # SNAN x NORM
short fsglmul_res_snan - tbl_fsglmul_op # SNAN x ZERO
short fsglmul_res_snan - tbl_fsglmul_op # SNAN x INF
short fsglmul_res_snan - tbl_fsglmul_op # SNAN x QNAN
short fsglmul_res_snan - tbl_fsglmul_op # SNAN x DENORM
short fsglmul_res_snan - tbl_fsglmul_op # SNAN x SNAN
short tbl_fsglmul_op - tbl_fsglmul_op #
short tbl_fsglmul_op - tbl_fsglmul_op #
#########################################################################
# XDEF **************************************************************** #
# fsgldiv(): emulates the fsgldiv instruction #
# #
# XREF **************************************************************** #
# scale_to_zero_src() - scale src exponent to zero #
# scale_to_zero_dst() - scale dst exponent to zero #
# unf_res4() - return default underflow result for sglop #
# ovf_res() - return default overflow result #
# res_qnan() - return QNAN result #
# res_snan() - return SNAN result #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# a1 = pointer to extended precision destination operand #
# d0 rnd prec,mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms/denorms into ext/sgl/dbl precision. #
# For norms/denorms, scale the exponents such that a divide #
# instruction won't cause an exception. Use the regular fsgldiv to #
# compute a result. Check if the regular operands would have taken #
# an exception. If so, return the default overflow/underflow result #
# and return the EXOP if exceptions are enabled. Else, scale the #
# result operand to the proper exponent. #
# #
#########################################################################
global fsgldiv
fsgldiv:
mov.l %d0,L_SCR3(%a6) # store rnd info
cmpi.l %d0,&0x3fff-0x0000 # will result underflow?
beq.w fsgldiv_may_unfl # maybe
bgt.w fsgldiv_unfl # yes; go handle underflow
fsgldiv_normal:
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l L_SCR3(%a6),%fpcr # save FPCR
fmov.l &0x0,%fpsr # clear FPSR
fsgldiv.x FP_SCR0(%a6),%fp0 # perform sgl divide
fmov.l %fpsr,%d1 # save FPSR
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fsgldiv_normal_exit:
fmovm.x &0x80,FP_SCR0(%a6) # store result on stack
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # load {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
or.w %d2,%d1 # concat old sign,new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x80 # return result in fp0
rts
fsgldiv_may_ovfl:
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # set FPSR
fsgldiv.x FP_SCR0(%a6),%fp0 # execute divide
fmov.l %fpsr,%d1
fmov.l &0x0,%fpcr
or.l %d1,USER_FPSR(%a6) # save INEX,N
fmovm.x &0x01,-(%sp) # save result to stack
mov.w (%sp),%d1 # fetch new exponent
add.l &0xc,%sp # clear result
andi.l &0x7fff,%d1 # strip sign sub.l %d0,%d1 # add scale factor
cmp.l %d1,&0x7fff # did divide overflow?
blt.b fsgldiv_normal_exit
fsgldiv_ovfl_tst:
or.w &ovfl_inx_mask,2+USER_FPSR(%a6) # set ovfl/aovfl/ainex
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x13,%d1 # is OVFL or INEX enabled?
bne.b fsgldiv_ovfl_ena # yes
fsgldiv_ovfl_dis:
btst &neg_bit,FPSR_CC(%a6) # is result negative
sne %d1 # set sign param accordingly
mov.l L_SCR3(%a6),%d0 # pass prec:rnd
andi.b &0x30,%d0 # kill precision
bsr.l ovf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # set INF if applicable
fmovm.x (%a0),&0x80 # return default result in fp0
rts
fsgldiv_ovfl_ena:
fmovm.x &0x80,FP_SCR0(%a6) # move result to stack
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
subi.l &0x6000,%d1 # subtract new bias
andi.w &0x7fff,%d1 # clear ms bit
or.w %d2,%d1 # concat old sign,new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
bra.b fsgldiv_ovfl_dis
fsgldiv_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l &rz_mode*0x10,%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fsgldiv.x FP_SCR0(%a6),%fp0 # execute sgl divide
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fsgldiv_unfl_ena # yes
fsgldiv_unfl_dis:
fmovm.x &0x80,FP_SCR0(%a6) # store out result
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res4 # calculate default result
or.b %d0,FPSR_CC(%a6) # 'Z' bit may have been set
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# UNFL is enabled.
#
fsgldiv_unfl_ena:
fmovm.x FP_SCR1(%a6),&0x40 # load dst op
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fsgldiv.x FP_SCR0(%a6),%fp1 # execute sgl divide
fmov.l &0x0,%fpcr # clear FPCR
fmovm.x &0x40,FP_SCR0(%a6) # save result to stack
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
addi.l &0x6000,%d1 # add bias
andi.w &0x7fff,%d1 # clear top bit
or.w %d2,%d1 # concat old sign, new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
bra.b fsgldiv_unfl_dis
#
# the divide operation MAY underflow:
#
fsgldiv_may_unfl:
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fsgldiv.x FP_SCR0(%a6),%fp0 # execute sgl divide
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fabs.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x1 # is |result| > 1.b?
fbgt.w fsgldiv_normal_exit # no; no underflow occurred
fblt.w fsgldiv_unfl # yes; underflow occurred
#
# we still don't know if underflow occurred. result is ~ equal to 1. but,
# we don't know if the result was an underflow that rounded up to a 1
# or a normalized number that rounded down to a 1. so, redo the entire
# operation using RZ as the rounding mode to see what the pre-rounded
# result is. this case should be relatively rare.
#
fmovm.x FP_SCR1(%a6),&0x40 # load dst op into %fp1
#
# Divide: inputs are not both normalized; what are they?
#
fsgldiv_not_norm:
mov.w (tbl_fsgldiv_op.b,%pc,%d1.w*2),%d1
jmp (tbl_fsgldiv_op.b,%pc,%d1.w*1)
swbeg &48
tbl_fsgldiv_op:
short fsgldiv_norm - tbl_fsgldiv_op # NORM / NORM
short fsgldiv_inf_load - tbl_fsgldiv_op # NORM / ZERO
short fsgldiv_zero_load - tbl_fsgldiv_op # NORM / INF
short fsgldiv_res_qnan - tbl_fsgldiv_op # NORM / QNAN
short fsgldiv_norm - tbl_fsgldiv_op # NORM / DENORM
short fsgldiv_res_snan - tbl_fsgldiv_op # NORM / SNAN
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short fsgldiv_zero_load - tbl_fsgldiv_op # ZERO / NORM
short fsgldiv_res_operr - tbl_fsgldiv_op # ZERO / ZERO
short fsgldiv_zero_load - tbl_fsgldiv_op # ZERO / INF
short fsgldiv_res_qnan - tbl_fsgldiv_op # ZERO / QNAN
short fsgldiv_zero_load - tbl_fsgldiv_op # ZERO / DENORM
short fsgldiv_res_snan - tbl_fsgldiv_op # ZERO / SNAN
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short fsgldiv_inf_dst - tbl_fsgldiv_op # INF / NORM
short fsgldiv_inf_dst - tbl_fsgldiv_op # INF / ZERO
short fsgldiv_res_operr - tbl_fsgldiv_op # INF / INF
short fsgldiv_res_qnan - tbl_fsgldiv_op # INF / QNAN
short fsgldiv_inf_dst - tbl_fsgldiv_op # INF / DENORM
short fsgldiv_res_snan - tbl_fsgldiv_op # INF / SNAN
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short fsgldiv_res_qnan - tbl_fsgldiv_op # QNAN / NORM
short fsgldiv_res_qnan - tbl_fsgldiv_op # QNAN / ZERO
short fsgldiv_res_qnan - tbl_fsgldiv_op # QNAN / INF
short fsgldiv_res_qnan - tbl_fsgldiv_op # QNAN / QNAN
short fsgldiv_res_qnan - tbl_fsgldiv_op # QNAN / DENORM
short fsgldiv_res_snan - tbl_fsgldiv_op # QNAN / SNAN
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short fsgldiv_norm - tbl_fsgldiv_op # DENORM / NORM
short fsgldiv_inf_load - tbl_fsgldiv_op # DENORM / ZERO
short fsgldiv_zero_load - tbl_fsgldiv_op # DENORM / INF
short fsgldiv_res_qnan - tbl_fsgldiv_op # DENORM / QNAN
short fsgldiv_norm - tbl_fsgldiv_op # DENORM / DENORM
short fsgldiv_res_snan - tbl_fsgldiv_op # DENORM / SNAN
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short fsgldiv_res_snan - tbl_fsgldiv_op # SNAN / NORM
short fsgldiv_res_snan - tbl_fsgldiv_op # SNAN / ZERO
short fsgldiv_res_snan - tbl_fsgldiv_op # SNAN / INF
short fsgldiv_res_snan - tbl_fsgldiv_op # SNAN / QNAN
short fsgldiv_res_snan - tbl_fsgldiv_op # SNAN / DENORM
short fsgldiv_res_snan - tbl_fsgldiv_op # SNAN / SNAN
short tbl_fsgldiv_op - tbl_fsgldiv_op #
short tbl_fsgldiv_op - tbl_fsgldiv_op #
#########################################################################
# XDEF **************************************************************** #
# fadd(): emulates the fadd instruction #
# fsadd(): emulates the fadd instruction #
# fdadd(): emulates the fdadd instruction #
# #
# XREF **************************************************************** #
# addsub_scaler2() - scale the operands so they won't take exc #
# ovf_res() - return default overflow result #
# unf_res() - return default underflow result #
# res_qnan() - set QNAN result #
# res_snan() - set SNAN result #
# res_operr() - set OPERR result #
# scale_to_zero_src() - set src operand exponent equal to zero #
# scale_to_zero_dst() - set dst operand exponent equal to zero #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# a1 = pointer to extended precision destination operand #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms into extended, single, and double precision. #
# Do addition after scaling exponents such that exception won't #
# occur. Then, check result exponent to see if exception would have #
# occurred. If so, return default result and maybe EXOP. Else, insert #
# the correct result exponent and return. Set FPSR bits as appropriate. #
# #
#########################################################################
fadd_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
add.l &0xc,%sp
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l &rz_mode*0x10,%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fadd.x FP_SCR0(%a6),%fp0 # execute add
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d1 # save status
or.l %d1,USER_FPSR(%a6) # save INEX,N
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fadd_unfl_ena # yes
fadd_unfl_dis:
fmovm.x &0x80,FP_SCR0(%a6) # store out result
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # 'Z' bit may have been set
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
mov.l (%sp)+,%d2 # restore d2
rts
fadd_unfl_ena:
fmovm.x FP_SCR1(%a6),&0x40 # load dst op
mov.l L_SCR3(%a6),%d1
andi.b &0xc0,%d1 # is precision extended?
bne.b fadd_unfl_ena_sd # no; sgl or dbl
fmovm.x &0x40,FP_SCR0(%a6) # save result to stack
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
addi.l &0x6000,%d1 # add new bias
andi.w &0x7fff,%d1 # clear top bit
or.w %d2,%d1 # concat sign,new exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
bra.w fadd_unfl_dis
fadd_unfl_ena_sd:
mov.l L_SCR3(%a6),%d1
andi.b &0x30,%d1 # use only rnd mode
fmov.l %d1,%fpcr # set FPCR
bra.b fadd_unfl_ena_cont
#
# result is equal to the smallest normalized number in the selected precision
# if the precision is extended, this result could not have come from an
# underflow that rounded up.
#
fadd_may_unfl:
mov.l L_SCR3(%a6),%d1
andi.b &0xc0,%d1
beq.w fadd_normal # yes; no underflow occurred
mov.l 0x4(%sp),%d1 # extract hi(man)
cmpi.l %d1,&0x80000000 # is hi(man) = 0x80000000?
bne.w fadd_normal # no; no underflow occurred
tst.l 0x8(%sp) # is lo(man) = 0x0?
bne.w fadd_normal # no; no underflow occurred
btst &inex2_bit,FPSR_EXCEPT(%a6) # is INEX2 set?
beq.w fadd_normal # no; no underflow occurred
#
# ok, so now the result has a exponent equal to the smallest normalized
# exponent for the selected precision. also, the mantissa is equal to
# 0x8000000000000000 and this mantissa is the result of rounding non-zero
# g,r,s.
# now, we must determine whether the pre-rounded result was an underflow
# rounded "up" or a normalized number rounded "down".
# so, we do this be re-executing the add using RZ as the rounding mode and
# seeing if the new result is smaller or equal to the current result.
#
fmovm.x FP_SCR1(%a6),&0x40 # load dst op into fp1
#
# Add: inputs are not both normalized; what are they?
#
fadd_not_norm:
mov.w (tbl_fadd_op.b,%pc,%d1.w*2),%d1
jmp (tbl_fadd_op.b,%pc,%d1.w*1)
swbeg &48
tbl_fadd_op:
short fadd_norm - tbl_fadd_op # NORM + NORM
short fadd_zero_src - tbl_fadd_op # NORM + ZERO
short fadd_inf_src - tbl_fadd_op # NORM + INF
short fadd_res_qnan - tbl_fadd_op # NORM + QNAN
short fadd_norm - tbl_fadd_op # NORM + DENORM
short fadd_res_snan - tbl_fadd_op # NORM + SNAN
short tbl_fadd_op - tbl_fadd_op #
short tbl_fadd_op - tbl_fadd_op #
short fadd_zero_dst - tbl_fadd_op # ZERO + NORM
short fadd_zero_2 - tbl_fadd_op # ZERO + ZERO
short fadd_inf_src - tbl_fadd_op # ZERO + INF
short fadd_res_qnan - tbl_fadd_op # NORM + QNAN
short fadd_zero_dst - tbl_fadd_op # ZERO + DENORM
short fadd_res_snan - tbl_fadd_op # NORM + SNAN
short tbl_fadd_op - tbl_fadd_op #
short tbl_fadd_op - tbl_fadd_op #
short fadd_inf_dst - tbl_fadd_op # INF + NORM
short fadd_inf_dst - tbl_fadd_op # INF + ZERO
short fadd_inf_2 - tbl_fadd_op # INF + INF
short fadd_res_qnan - tbl_fadd_op # NORM + QNAN
short fadd_inf_dst - tbl_fadd_op # INF + DENORM
short fadd_res_snan - tbl_fadd_op # NORM + SNAN
short tbl_fadd_op - tbl_fadd_op #
short tbl_fadd_op - tbl_fadd_op #
short fadd_res_qnan - tbl_fadd_op # QNAN + NORM
short fadd_res_qnan - tbl_fadd_op # QNAN + ZERO
short fadd_res_qnan - tbl_fadd_op # QNAN + INF
short fadd_res_qnan - tbl_fadd_op # QNAN + QNAN
short fadd_res_qnan - tbl_fadd_op # QNAN + DENORM
short fadd_res_snan - tbl_fadd_op # QNAN + SNAN
short tbl_fadd_op - tbl_fadd_op #
short tbl_fadd_op - tbl_fadd_op #
short fadd_norm - tbl_fadd_op # DENORM + NORM
short fadd_zero_src - tbl_fadd_op # DENORM + ZERO
short fadd_inf_src - tbl_fadd_op # DENORM + INF
short fadd_res_qnan - tbl_fadd_op # NORM + QNAN
short fadd_norm - tbl_fadd_op # DENORM + DENORM
short fadd_res_snan - tbl_fadd_op # NORM + SNAN
short tbl_fadd_op - tbl_fadd_op #
short tbl_fadd_op - tbl_fadd_op #
short fadd_res_snan - tbl_fadd_op # SNAN + NORM
short fadd_res_snan - tbl_fadd_op # SNAN + ZERO
short fadd_res_snan - tbl_fadd_op # SNAN + INF
short fadd_res_snan - tbl_fadd_op # SNAN + QNAN
short fadd_res_snan - tbl_fadd_op # SNAN + DENORM
short fadd_res_snan - tbl_fadd_op # SNAN + SNAN
short tbl_fadd_op - tbl_fadd_op #
short tbl_fadd_op - tbl_fadd_op #
#
# both operands are ZEROes
#
fadd_zero_2:
mov.b SRC_EX(%a0),%d0 # are the signs opposite
mov.b DST_EX(%a1),%d1
eor.b %d0,%d1
bmi.w fadd_zero_2_chk_rm # weed out (-ZERO)+(+ZERO)
# the signs are the same. so determine whether they are positive or negative
# and return the appropriately signed zero.
tst.b %d0 # are ZEROes positive or negative?
bmi.b fadd_zero_rm # negative
fmov.s &0x00000000,%fp0 # return +ZERO
mov.b &z_bmask,FPSR_CC(%a6) # set Z
rts
#
# the ZEROes have opposite signs:
# - Therefore, we return +ZERO if the rounding modes are RN,RZ, or RP.
# - -ZERO is returned in the case of RM.
#
fadd_zero_2_chk_rm:
mov.b 3+L_SCR3(%a6),%d1
andi.b &0x30,%d1 # extract rnd mode
cmpi.b %d1,&rm_mode*0x10 # is rnd mode == RM?
beq.b fadd_zero_rm # yes
fmov.s &0x00000000,%fp0 # return +ZERO
mov.b &z_bmask,FPSR_CC(%a6) # set Z
rts
#
# one operand is a ZERO and the other is a DENORM or NORM. scale
# the DENORM or NORM and jump to the regular fadd routine.
#
fadd_zero_dst:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_to_zero_src # scale the operand clr.w FP_SCR1_EX(%a6) clr.l FP_SCR1_HI(%a6) clr.l FP_SCR1_LO(%a6)
bra.w fadd_zero_entry # go execute fadd
#
# both operands are INFs. an OPERR will result if the INFs have
# different signs. else, an INF of the same sign is returned
#
fadd_inf_2:
mov.b SRC_EX(%a0),%d0 # exclusive or the signs
mov.b DST_EX(%a1),%d1
eor.b %d1,%d0
bmi.l res_operr # weed out (-INF)+(+INF)
# ok, so it's not an OPERR. but, we do have to remember to return the
# src INF since that's where the 881/882 gets the j-bit from...
#
# operands are INF and one of {ZERO, INF, DENORM, NORM}
#
fadd_inf_src:
fmovm.x SRC(%a0),&0x80 # return src INF
tst.b SRC_EX(%a0) # is INF positive?
bpl.b fadd_inf_done # yes; we're done
mov.b &neg_bmask+inf_bmask,FPSR_CC(%a6) # set INF/NEG
rts
#
# operands are INF and one of {ZERO, INF, DENORM, NORM}
#
fadd_inf_dst:
fmovm.x DST(%a1),&0x80 # return dst INF
tst.b DST_EX(%a1) # is INF positive?
bpl.b fadd_inf_done # yes; we're done
mov.b &neg_bmask+inf_bmask,FPSR_CC(%a6) # set INF/NEG
rts
fadd_inf_done:
mov.b &inf_bmask,FPSR_CC(%a6) # set INF
rts
#########################################################################
# XDEF **************************************************************** #
# fsub(): emulates the fsub instruction #
# fssub(): emulates the fssub instruction #
# fdsub(): emulates the fdsub instruction #
# #
# XREF **************************************************************** #
# addsub_scaler2() - scale the operands so they won't take exc #
# ovf_res() - return default overflow result #
# unf_res() - return default underflow result #
# res_qnan() - set QNAN result #
# res_snan() - set SNAN result #
# res_operr() - set OPERR result #
# scale_to_zero_src() - set src operand exponent equal to zero #
# scale_to_zero_dst() - set dst operand exponent equal to zero #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# a1 = pointer to extended precision destination operand #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms into extended, single, and double precision. #
# Do subtraction after scaling exponents such that exception won't#
# occur. Then, check result exponent to see if exception would have #
# occurred. If so, return default result and maybe EXOP. Else, insert #
# the correct result exponent and return. Set FPSR bits as appropriate. #
# #
#########################################################################
fsub_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
add.l &0xc,%sp
fmovm.x FP_SCR1(%a6),&0x80 # load dst op
fmov.l &rz_mode*0x10,%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fsub.x FP_SCR0(%a6),%fp0 # execute subtract
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d1 # save status
or.l %d1,USER_FPSR(%a6)
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fsub_unfl_ena # yes
fsub_unfl_dis:
fmovm.x &0x80,FP_SCR0(%a6) # store out result
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # 'Z' may have been set
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
mov.l (%sp)+,%d2 # restore d2
rts
fsub_unfl_ena:
fmovm.x FP_SCR1(%a6),&0x40
mov.l L_SCR3(%a6),%d1
andi.b &0xc0,%d1 # is precision extended?
bne.b fsub_unfl_ena_sd # no
#
# result is equal to the smallest normalized number in the selected precision
# if the precision is extended, this result could not have come from an
# underflow that rounded up.
#
fsub_may_unfl:
mov.l L_SCR3(%a6),%d1
andi.b &0xc0,%d1 # fetch rnd prec
beq.w fsub_normal # yes; no underflow occurred
mov.l 0x4(%sp),%d1
cmpi.l %d1,&0x80000000 # is hi(man) = 0x80000000?
bne.w fsub_normal # no; no underflow occurred
tst.l 0x8(%sp) # is lo(man) = 0x0?
bne.w fsub_normal # no; no underflow occurred
btst &inex2_bit,FPSR_EXCEPT(%a6) # is INEX2 set?
beq.w fsub_normal # no; no underflow occurred
#
# ok, so now the result has a exponent equal to the smallest normalized
# exponent for the selected precision. also, the mantissa is equal to
# 0x8000000000000000 and this mantissa is the result of rounding non-zero
# g,r,s.
# now, we must determine whether the pre-rounded result was an underflow
# rounded "up" or a normalized number rounded "down".
# so, we do this be re-executing the add using RZ as the rounding mode and
# seeing if the new result is smaller or equal to the current result.
#
fmovm.x FP_SCR1(%a6),&0x40 # load dst op into fp1
#
# Sub: inputs are not both normalized; what are they?
#
fsub_not_norm:
mov.w (tbl_fsub_op.b,%pc,%d1.w*2),%d1
jmp (tbl_fsub_op.b,%pc,%d1.w*1)
swbeg &48
tbl_fsub_op:
short fsub_norm - tbl_fsub_op # NORM - NORM
short fsub_zero_src - tbl_fsub_op # NORM - ZERO
short fsub_inf_src - tbl_fsub_op # NORM - INF
short fsub_res_qnan - tbl_fsub_op # NORM - QNAN
short fsub_norm - tbl_fsub_op # NORM - DENORM
short fsub_res_snan - tbl_fsub_op # NORM - SNAN
short tbl_fsub_op - tbl_fsub_op #
short tbl_fsub_op - tbl_fsub_op #
short fsub_zero_dst - tbl_fsub_op # ZERO - NORM
short fsub_zero_2 - tbl_fsub_op # ZERO - ZERO
short fsub_inf_src - tbl_fsub_op # ZERO - INF
short fsub_res_qnan - tbl_fsub_op # NORM - QNAN
short fsub_zero_dst - tbl_fsub_op # ZERO - DENORM
short fsub_res_snan - tbl_fsub_op # NORM - SNAN
short tbl_fsub_op - tbl_fsub_op #
short tbl_fsub_op - tbl_fsub_op #
short fsub_inf_dst - tbl_fsub_op # INF - NORM
short fsub_inf_dst - tbl_fsub_op # INF - ZERO
short fsub_inf_2 - tbl_fsub_op # INF - INF
short fsub_res_qnan - tbl_fsub_op # NORM - QNAN
short fsub_inf_dst - tbl_fsub_op # INF - DENORM
short fsub_res_snan - tbl_fsub_op # NORM - SNAN
short tbl_fsub_op - tbl_fsub_op #
short tbl_fsub_op - tbl_fsub_op #
short fsub_res_qnan - tbl_fsub_op # QNAN - NORM
short fsub_res_qnan - tbl_fsub_op # QNAN - ZERO
short fsub_res_qnan - tbl_fsub_op # QNAN - INF
short fsub_res_qnan - tbl_fsub_op # QNAN - QNAN
short fsub_res_qnan - tbl_fsub_op # QNAN - DENORM
short fsub_res_snan - tbl_fsub_op # QNAN - SNAN
short tbl_fsub_op - tbl_fsub_op #
short tbl_fsub_op - tbl_fsub_op #
short fsub_norm - tbl_fsub_op # DENORM - NORM
short fsub_zero_src - tbl_fsub_op # DENORM - ZERO
short fsub_inf_src - tbl_fsub_op # DENORM - INF
short fsub_res_qnan - tbl_fsub_op # NORM - QNAN
short fsub_norm - tbl_fsub_op # DENORM - DENORM
short fsub_res_snan - tbl_fsub_op # NORM - SNAN
short tbl_fsub_op - tbl_fsub_op #
short tbl_fsub_op - tbl_fsub_op #
short fsub_res_snan - tbl_fsub_op # SNAN - NORM
short fsub_res_snan - tbl_fsub_op # SNAN - ZERO
short fsub_res_snan - tbl_fsub_op # SNAN - INF
short fsub_res_snan - tbl_fsub_op # SNAN - QNAN
short fsub_res_snan - tbl_fsub_op # SNAN - DENORM
short fsub_res_snan - tbl_fsub_op # SNAN - SNAN
short tbl_fsub_op - tbl_fsub_op #
short tbl_fsub_op - tbl_fsub_op #
#
# both operands are ZEROes
#
fsub_zero_2:
mov.b SRC_EX(%a0),%d0
mov.b DST_EX(%a1),%d1
eor.b %d1,%d0
bpl.b fsub_zero_2_chk_rm
# the signs are opposite, so, return a ZERO w/ the sign of the dst ZERO
tst.b %d0 # is dst negative?
bmi.b fsub_zero_2_rm # yes
fmov.s &0x00000000,%fp0 # no; return +ZERO
mov.b &z_bmask,FPSR_CC(%a6) # set Z
rts
#
# the ZEROes have the same signs:
# - Therefore, we return +ZERO if the rounding mode is RN,RZ, or RP
# - -ZERO is returned in the case of RM.
#
fsub_zero_2_chk_rm:
mov.b 3+L_SCR3(%a6),%d1
andi.b &0x30,%d1 # extract rnd mode
cmpi.b %d1,&rm_mode*0x10 # is rnd mode = RM?
beq.b fsub_zero_2_rm # yes
fmov.s &0x00000000,%fp0 # no; return +ZERO
mov.b &z_bmask,FPSR_CC(%a6) # set Z
rts
#
# one operand is a ZERO and the other is a DENORM or a NORM.
# scale the DENORM or NORM and jump to the regular fsub routine.
#
fsub_zero_dst:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_to_zero_src # scale the operand clr.w FP_SCR1_EX(%a6) clr.l FP_SCR1_HI(%a6) clr.l FP_SCR1_LO(%a6)
bra.w fsub_zero_entry # go execute fsub
#
# both operands are INFs. an OPERR will result if the INFs have the
# same signs. else,
#
fsub_inf_2:
mov.b SRC_EX(%a0),%d0 # exclusive or the signs
mov.b DST_EX(%a1),%d1
eor.b %d1,%d0
bpl.l res_operr # weed out (-INF)+(+INF)
# ok, so it's not an OPERR. but we do have to remember to return
# the src INF since that's where the 881/882 gets the j-bit.
fsub_inf_src:
fmovm.x SRC(%a0),&0x80 # return src INF
fneg.x %fp0 # invert sign
fbge.w fsub_inf_done # sign is now positive
mov.b &neg_bmask+inf_bmask,FPSR_CC(%a6) # set INF/NEG
rts
fsub_inf_dst:
fmovm.x DST(%a1),&0x80 # return dst INF
tst.b DST_EX(%a1) # is INF negative?
bpl.b fsub_inf_done # no
mov.b &neg_bmask+inf_bmask,FPSR_CC(%a6) # set INF/NEG
rts
fsub_inf_done:
mov.b &inf_bmask,FPSR_CC(%a6) # set INF
rts
#########################################################################
# XDEF **************************************************************** #
# fsqrt(): emulates the fsqrt instruction #
# fssqrt(): emulates the fssqrt instruction #
# fdsqrt(): emulates the fdsqrt instruction #
# #
# XREF **************************************************************** #
# scale_sqrt() - scale the source operand #
# unf_res() - return default underflow result #
# ovf_res() - return default overflow result #
# res_qnan_1op() - return QNAN result #
# res_snan_1op() - return SNAN result #
# #
# INPUT *************************************************************** #
# a0 = pointer to extended precision source operand #
# d0 rnd prec,mode #
# #
# OUTPUT ************************************************************** #
# fp0 = result #
# fp1 = EXOP (if exception occurred) #
# #
# ALGORITHM *********************************************************** #
# Handle NANs, infinities, and zeroes as special cases. Divide #
# norms/denorms into ext/sgl/dbl precision. #
# For norms/denorms, scale the exponents such that a sqrt #
# instruction won't cause an exception. Use the regular fsqrt to #
# compute a result. Check if the regular operands would have taken #
# an exception. If so, return the default overflow/underflow result #
# and return the EXOP if exceptions are enabled. Else, scale the #
# result operand to the proper exponent. #
# #
#########################################################################
#
# operand is either single or double
#
fsqrt_not_ext:
cmpi.b %d0,&s_mode*0x10 # separate sgl/dbl prec
bne.w fsqrt_dbl
#
# operand is to be rounded to single precision
#
fsqrt_sgl:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_sqrt # calculate scale factor
cmpi.l %d0,&0x3fff-0x3f81 # will move in underflow?
beq.w fsqrt_sd_may_unfl
bgt.w fsqrt_sd_unfl # yes; go handle underflow
cmpi.l %d0,&0x3fff-0x407f # will move in overflow?
beq.w fsqrt_sd_may_ovfl # maybe; go check
blt.w fsqrt_sd_ovfl # yes; go handle overflow
#
# operand will NOT overflow or underflow when moved in to the fp reg file
#
fsqrt_sd_normal:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fsqrt.x FP_SCR0(%a6),%fp0 # perform absolute
fmov.l %fpsr,%d1 # save FPSR
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fsqrt_sd_normal_exit:
mov.l %d2,-(%sp) # save d2
fmovm.x &0x80,FP_SCR0(%a6) # store out result
mov.w FP_SCR0_EX(%a6),%d1 # load sgn,exp
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign sub.l %d0,%d1 # add scale factor
andi.w &0x8000,%d2 # keep old sign
or.w %d1,%d2 # concat old sign,new exp
mov.w %d2,FP_SCR0_EX(%a6) # insert new exponent
mov.l (%sp)+,%d2 # restore d2
fmovm.x FP_SCR0(%a6),&0x80 # return result in fp0
rts
#
# operand is to be rounded to double precision
#
fsqrt_dbl:
mov.w SRC_EX(%a0),FP_SCR0_EX(%a6)
mov.l SRC_HI(%a0),FP_SCR0_HI(%a6)
mov.l SRC_LO(%a0),FP_SCR0_LO(%a6)
bsr.l scale_sqrt # calculate scale factor
cmpi.l %d0,&0x3fff-0x3c01 # will move in underflow?
beq.w fsqrt_sd_may_unfl
bgt.b fsqrt_sd_unfl # yes; go handle underflow
cmpi.l %d0,&0x3fff-0x43ff # will move in overflow?
beq.w fsqrt_sd_may_ovfl # maybe; go check
blt.w fsqrt_sd_ovfl # yes; go handle overflow
bra.w fsqrt_sd_normal # no; ho handle normalized op
# we're on the line here and the distinguising characteristic is whether
# the exponent is 3fff or 3ffe. if it's 3ffe, then it's a safe number
# elsewise fall through to underflow.
fsqrt_sd_may_unfl:
btst &0x0,1+FP_SCR0_EX(%a6) # is exponent 0x3fff?
bne.w fsqrt_sd_normal # yes, so no underflow
#
# operand WILL underflow when moved in to the fp register file
#
fsqrt_sd_unfl:
bset &unfl_bit,FPSR_EXCEPT(%a6) # set unfl exc bit
fmov.l &rz_mode*0x10,%fpcr # set FPCR
fmov.l &0x0,%fpsr # clear FPSR
fsqrt.x FP_SCR0(%a6),%fp0 # execute square root
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
# if underflow or inexact is enabled, go calculate EXOP first.
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x0b,%d1 # is UNFL or INEX enabled?
bne.b fsqrt_sd_unfl_ena # yes
fsqrt_sd_unfl_dis:
fmovm.x &0x80,FP_SCR0(%a6) # store out result
lea FP_SCR0(%a6),%a0 # pass: result addr
mov.l L_SCR3(%a6),%d1 # pass: rnd prec,mode
bsr.l unf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # set possible 'Z' ccode
fmovm.x FP_SCR0(%a6),&0x80 # return default result in fp0
rts
#
# operand will underflow AND underflow is enabled.
# Therefore, we must return the result rounded to extended precision.
#
fsqrt_sd_unfl_ena:
mov.l FP_SCR0_HI(%a6),FP_SCR1_HI(%a6)
mov.l FP_SCR0_LO(%a6),FP_SCR1_LO(%a6)
mov.w FP_SCR0_EX(%a6),%d1 # load current exponent
mov.l %d2,-(%sp) # save d2
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # subtract scale factor
addi.l &0x6000,%d1 # add new bias
andi.w &0x7fff,%d1
or.w %d2,%d1 # concat new sign,new exp
mov.w %d1,FP_SCR1_EX(%a6) # insert new exp
fmovm.x FP_SCR1(%a6),&0x40 # return EXOP in fp1
mov.l (%sp)+,%d2 # restore d2
bra.b fsqrt_sd_unfl_dis
#
# operand WILL overflow.
#
fsqrt_sd_ovfl:
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fsqrt.x FP_SCR0(%a6),%fp0 # perform square root
fmov.l &0x0,%fpcr # clear FPCR
fmov.l %fpsr,%d1 # save FPSR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fsqrt_sd_ovfl_tst:
or.l &ovfl_inx_mask,USER_FPSR(%a6) # set ovfl/aovfl/ainex
mov.b FPCR_ENABLE(%a6),%d1
andi.b &0x13,%d1 # is OVFL or INEX enabled?
bne.b fsqrt_sd_ovfl_ena # yes
#
# OVFL is not enabled; therefore, we must create the default result by
# calling ovf_res().
#
fsqrt_sd_ovfl_dis:
btst &neg_bit,FPSR_CC(%a6) # is result negative?
sne %d1 # set sign param accordingly
mov.l L_SCR3(%a6),%d0 # pass: prec,mode
bsr.l ovf_res # calculate default result
or.b %d0,FPSR_CC(%a6) # set INF,N if applicable
fmovm.x (%a0),&0x80 # return default result in fp0
rts
#
# OVFL is enabled.
# the INEX2 bit has already been updated by the round to the correct precision.
# now, round to extended(and don't alter the FPSR).
#
fsqrt_sd_ovfl_ena:
mov.l %d2,-(%sp) # save d2
mov.w FP_SCR0_EX(%a6),%d1 # fetch {sgn,exp}
mov.l %d1,%d2 # make a copy
andi.l &0x7fff,%d1 # strip sign
andi.w &0x8000,%d2 # keep old sign sub.l %d0,%d1 # add scale factor
subi.l &0x6000,%d1 # subtract bias
andi.w &0x7fff,%d1
or.w %d2,%d1 # concat sign,exp
mov.w %d1,FP_SCR0_EX(%a6) # insert new exponent
fmovm.x FP_SCR0(%a6),&0x40 # return EXOP in fp1
mov.l (%sp)+,%d2 # restore d2
bra.b fsqrt_sd_ovfl_dis
#
# the move in MAY underflow. so...
#
fsqrt_sd_may_ovfl:
btst &0x0,1+FP_SCR0_EX(%a6) # is exponent 0x3fff?
bne.w fsqrt_sd_ovfl # yes, so overflow
fmov.l &0x0,%fpsr # clear FPSR
fmov.l L_SCR3(%a6),%fpcr # set FPCR
fsqrt.x FP_SCR0(%a6),%fp0 # perform absolute
fmov.l %fpsr,%d1 # save status
fmov.l &0x0,%fpcr # clear FPCR
or.l %d1,USER_FPSR(%a6) # save INEX2,N
fmov.x %fp0,%fp1 # make a copy of result
fcmp.b %fp1,&0x1 # is |result| >= 1.b?
fbge.w fsqrt_sd_ovfl_tst # yes; overflow has occurred
# no, it didn't overflow; we have correct result
bra.w fsqrt_sd_normal_exit
#
# input is not normalized; what is it?
#
fsqrt_not_norm:
cmpi.b %d1,&DENORM # weed out DENORM
beq.w fsqrt_denorm
cmpi.b %d1,&ZERO # weed out ZERO
beq.b fsqrt_zero
cmpi.b %d1,&INF # weed out INF
beq.b fsqrt_inf
cmpi.b %d1,&SNAN # weed out SNAN
beq.l res_snan_1op
bra.l res_qnan_1op
fsqrt_inf:
tst.b SRC_EX(%a0) # is INF positive or negative?
bmi.l res_operr # negative
fsqrt_inf_p:
fmovm.x SRC(%a0),&0x80 # return +INF in fp0
mov.b &inf_bmask,FPSR_CC(%a6) # set'I' ccode bit
rts
#########################################################################
# XDEF **************************************************************** #
# fetch_dreg(): fetch register according to index in d1 #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# d1 = index of register to fetch from #
# #
# OUTPUT ************************************************************** #
# d0 = value of register fetched #
# #
# ALGORITHM *********************************************************** #
# According to the index value in d1 which can range from zero #
# to fifteen, load the corresponding register file value (where #
# address register indexes start at 8). D0/D1/A0/A1/A6/A7 are on the #
# stack. The rest should still be in their original places. #
# #
#########################################################################
# this routine leaves d1 intact for subsequent store_dreg calls. global fetch_dreg
fetch_dreg:
mov.w (tbl_fdreg.b,%pc,%d1.w*2),%d0
jmp (tbl_fdreg.b,%pc,%d0.w*1)
tbl_fdreg:
short fdreg0 - tbl_fdreg
short fdreg1 - tbl_fdreg
short fdreg2 - tbl_fdreg
short fdreg3 - tbl_fdreg
short fdreg4 - tbl_fdreg
short fdreg5 - tbl_fdreg
short fdreg6 - tbl_fdreg
short fdreg7 - tbl_fdreg
short fdreg8 - tbl_fdreg
short fdreg9 - tbl_fdreg
short fdrega - tbl_fdreg
short fdregb - tbl_fdreg
short fdregc - tbl_fdreg
short fdregd - tbl_fdreg
short fdrege - tbl_fdreg
short fdregf - tbl_fdreg
#########################################################################
# XDEF **************************************************************** #
# store_dreg_l(): store longword to data register specified by d1 #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# d0 = longowrd value to store #
# d1 = index of register to fetch from #
# #
# OUTPUT ************************************************************** #
# (data register is updated) #
# #
# ALGORITHM *********************************************************** #
# According to the index value in d1, store the longword value #
# in d0 to the corresponding data register. D0/D1 are on the stack #
# while the rest are in their initial places. #
# #
#########################################################################
global store_dreg_l
store_dreg_l:
mov.w (tbl_sdregl.b,%pc,%d1.w*2),%d1
jmp (tbl_sdregl.b,%pc,%d1.w*1)
tbl_sdregl:
short sdregl0 - tbl_sdregl
short sdregl1 - tbl_sdregl
short sdregl2 - tbl_sdregl
short sdregl3 - tbl_sdregl
short sdregl4 - tbl_sdregl
short sdregl5 - tbl_sdregl
short sdregl6 - tbl_sdregl
short sdregl7 - tbl_sdregl
#########################################################################
# XDEF **************************************************************** #
# store_dreg_w(): store word to data register specified by d1 #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# d0 = word value to store #
# d1 = index of register to fetch from #
# #
# OUTPUT ************************************************************** #
# (data register is updated) #
# #
# ALGORITHM *********************************************************** #
# According to the index value in d1, store the word value #
# in d0 to the corresponding data register. D0/D1 are on the stack #
# while the rest are in their initial places. #
# #
#########################################################################
global store_dreg_w
store_dreg_w:
mov.w (tbl_sdregw.b,%pc,%d1.w*2),%d1
jmp (tbl_sdregw.b,%pc,%d1.w*1)
tbl_sdregw:
short sdregw0 - tbl_sdregw
short sdregw1 - tbl_sdregw
short sdregw2 - tbl_sdregw
short sdregw3 - tbl_sdregw
short sdregw4 - tbl_sdregw
short sdregw5 - tbl_sdregw
short sdregw6 - tbl_sdregw
short sdregw7 - tbl_sdregw
#########################################################################
# XDEF **************************************************************** #
# store_dreg_b(): store byte to data register specified by d1 #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# d0 = byte value to store #
# d1 = index of register to fetch from #
# #
# OUTPUT ************************************************************** #
# (data register is updated) #
# #
# ALGORITHM *********************************************************** #
# According to the index value in d1, store the byte value #
# in d0 to the corresponding data register. D0/D1 are on the stack #
# while the rest are in their initial places. #
# #
#########################################################################
global store_dreg_b
store_dreg_b:
mov.w (tbl_sdregb.b,%pc,%d1.w*2),%d1
jmp (tbl_sdregb.b,%pc,%d1.w*1)
tbl_sdregb:
short sdregb0 - tbl_sdregb
short sdregb1 - tbl_sdregb
short sdregb2 - tbl_sdregb
short sdregb3 - tbl_sdregb
short sdregb4 - tbl_sdregb
short sdregb5 - tbl_sdregb
short sdregb6 - tbl_sdregb
short sdregb7 - tbl_sdregb
#########################################################################
# XDEF **************************************************************** #
# inc_areg(): increment an address register by the value in d0 #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# d0 = amount to increment by #
# d1 = index of address register to increment #
# #
# OUTPUT ************************************************************** #
# (address register is updated) #
# #
# ALGORITHM *********************************************************** #
# Typically used for an instruction w/ a post-increment <ea>, #
# this routine adds the increment value in d0 to the address register #
# specified by d1. A0/A1/A6/A7 reside on the stack. The rest reside #
# in their original places. #
# For a7, if the increment amount is one, then we have to #
# increment by two. For any a7 update, set the mia7_flag so that if #
# an access error exception occurs later in emulation, this address #
# register update can be undone. #
# #
#########################################################################
global inc_areg
inc_areg:
mov.w (tbl_iareg.b,%pc,%d1.w*2),%d1
jmp (tbl_iareg.b,%pc,%d1.w*1)
tbl_iareg:
short iareg0 - tbl_iareg
short iareg1 - tbl_iareg
short iareg2 - tbl_iareg
short iareg3 - tbl_iareg
short iareg4 - tbl_iareg
short iareg5 - tbl_iareg
short iareg6 - tbl_iareg
short iareg7 - tbl_iareg
#########################################################################
# XDEF **************************************************************** #
# dec_areg(): decrement an address register by the value in d0 #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# d0 = amount to decrement by #
# d1 = index of address register to decrement #
# #
# OUTPUT ************************************************************** #
# (address register is updated) #
# #
# ALGORITHM *********************************************************** #
# Typically used for an instruction w/ a pre-decrement <ea>, #
# this routine adds the decrement value in d0 to the address register #
# specified by d1. A0/A1/A6/A7 reside on the stack. The rest reside #
# in their original places. #
# For a7, if the decrement amount is one, then we have to #
# decrement by two. For any a7 update, set the mda7_flag so that if #
# an access error exception occurs later in emulation, this address #
# register update can be undone. #
# #
#########################################################################
global dec_areg
dec_areg:
mov.w (tbl_dareg.b,%pc,%d1.w*2),%d1
jmp (tbl_dareg.b,%pc,%d1.w*1)
tbl_dareg:
short dareg0 - tbl_dareg
short dareg1 - tbl_dareg
short dareg2 - tbl_dareg
short dareg3 - tbl_dareg
short dareg4 - tbl_dareg
short dareg5 - tbl_dareg
short dareg6 - tbl_dareg
short dareg7 - tbl_dareg
#########################################################################
# XDEF **************************************************************** #
# load_fpn1(): load FP register value into FP_SRC(a6). #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# d0 = index of FP register to load #
# #
# OUTPUT ************************************************************** #
# FP_SRC(a6) = value loaded from FP register file #
# #
# ALGORITHM *********************************************************** #
# Using the index in d0, load FP_SRC(a6) with a number from the #
# FP register file. #
# #
#########################################################################
global load_fpn1
load_fpn1:
mov.w (tbl_load_fpn1.b,%pc,%d0.w*2), %d0
jmp (tbl_load_fpn1.b,%pc,%d0.w*1)
tbl_load_fpn1:
short load_fpn1_0 - tbl_load_fpn1
short load_fpn1_1 - tbl_load_fpn1
short load_fpn1_2 - tbl_load_fpn1
short load_fpn1_3 - tbl_load_fpn1
short load_fpn1_4 - tbl_load_fpn1
short load_fpn1_5 - tbl_load_fpn1
short load_fpn1_6 - tbl_load_fpn1
short load_fpn1_7 - tbl_load_fpn1
#########################################################################
# XDEF **************************************************************** #
# load_fpn2(): load FP register value into FP_DST(a6). #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# d0 = index of FP register to load #
# #
# OUTPUT ************************************************************** #
# FP_DST(a6) = value loaded from FP register file #
# #
# ALGORITHM *********************************************************** #
# Using the index in d0, load FP_DST(a6) with a number from the #
# FP register file. #
# #
#########################################################################
global load_fpn2
load_fpn2:
mov.w (tbl_load_fpn2.b,%pc,%d0.w*2), %d0
jmp (tbl_load_fpn2.b,%pc,%d0.w*1)
tbl_load_fpn2:
short load_fpn2_0 - tbl_load_fpn2
short load_fpn2_1 - tbl_load_fpn2
short load_fpn2_2 - tbl_load_fpn2
short load_fpn2_3 - tbl_load_fpn2
short load_fpn2_4 - tbl_load_fpn2
short load_fpn2_5 - tbl_load_fpn2
short load_fpn2_6 - tbl_load_fpn2
short load_fpn2_7 - tbl_load_fpn2
#########################################################################
# XDEF **************************************************************** #
# store_fpreg(): store an fp value to the fpreg designated d0. #
# #
# XREF **************************************************************** #
# None #
# #
# INPUT *************************************************************** #
# fp0 = extended precision value to store #
# d0 = index of floating-point register #
# #
# OUTPUT ************************************************************** #
# None #
# #
# ALGORITHM *********************************************************** #
# Store the value in fp0 to the FP register designated by the #
# value in d0. The FP number can be DENORM or SNAN so we have to be #
# careful that we don't take an exception here. #
# #
#########################################################################
global store_fpreg
store_fpreg:
mov.w (tbl_store_fpreg.b,%pc,%d0.w*2), %d0
jmp (tbl_store_fpreg.b,%pc,%d0.w*1)
tbl_store_fpreg:
short store_fpreg_0 - tbl_store_fpreg
short store_fpreg_1 - tbl_store_fpreg
short store_fpreg_2 - tbl_store_fpreg
short store_fpreg_3 - tbl_store_fpreg
short store_fpreg_4 - tbl_store_fpreg
short store_fpreg_5 - tbl_store_fpreg
short store_fpreg_6 - tbl_store_fpreg
short store_fpreg_7 - tbl_store_fpreg
#########################################################################
# XDEF **************************************************************** #
# get_packed(): fetch a packed operand from memory and then #
# convert it to a floating-point binary number. #
# #
# XREF **************************************************************** #
# _dcalc_ea() - calculate the correct <ea> #
# _mem_read() - fetch the packed operand from memory #
# facc_in_x() - the fetch failed so jump to special exit code #
# decbin() - convert packed to binary extended precision #
# #
# INPUT *************************************************************** #
# None #
# #
# OUTPUT ************************************************************** #
# If no failure on _mem_read(): #
# FP_SRC(a6) = packed operand now as a binary FP number #
# #
# ALGORITHM *********************************************************** #
# Get the correct <ea> which is the value on the exception stack #
# frame w/ maybe a correction factor if the <ea> is -(an) or (an)+. #
# Then, fetch the operand from memory. If the fetch fails, exit #
# through facc_in_x(). #
# If the packed operand is a ZERO,NAN, or INF, convert it to #
# its binary representation here. Else, call decbin() which will #
# convert the packed value to an extended precision binary value. #
# #
#########################################################################
# the stacked <ea> for packed is correct except for -(An).
# the base reg must be updated for both -(An) and (An)+. global get_packed
get_packed:
mov.l &0xc,%d0 # packed is 12 bytes
bsr.l _dcalc_ea # fetch <ea>; correct An
lea FP_SRC(%a6),%a1 # pass: ptr to super dst
mov.l &0xc,%d0 # pass: 12 bytes
bsr.l _dmem_read # read packed operand
tst.l %d1 # did dfetch fail?
bne.l facc_in_x # yes
# The packed operand is an INF or a NAN if the exponent field is all ones.
bfextu FP_SRC(%a6){&1:&15},%d0 # get exp
cmpi.w %d0,&0x7fff # INF or NAN?
bne.b gp_try_zero # no
rts # operand is an INF or NAN
# The packed operand is a zero if the mantissa is all zero, else it's
# a normal packed op.
gp_try_zero:
mov.b 3+FP_SRC(%a6),%d0 # get byte 4
andi.b &0x0f,%d0 # clear all but last nybble
bne.b gp_not_spec # not a zero
tst.l FP_SRC_HI(%a6) # is lw 2 zero?
bne.b gp_not_spec # not a zero
tst.l FP_SRC_LO(%a6) # is lw 3 zero?
bne.b gp_not_spec # not a zero
rts # operand is a ZERO
gp_not_spec:
lea FP_SRC(%a6),%a0 # pass: ptr to packed op
bsr.l decbin # convert to extended
fmovm.x &0x80,FP_SRC(%a6) # make this the srcop
rts
#########################################################################
# decbin(): Converts normalized packed bcd value pointed to by register #
# a0 to extended-precision value in fp0. #
# #
# INPUT *************************************************************** #
# a0 = pointer to normalized packed bcd value #
# #
# OUTPUT ************************************************************** #
# fp0 = exact fp representation of the packed bcd value. #
# #
# ALGORITHM *********************************************************** #
# Expected is a normal bcd (i.e. non-exceptional; all inf, zero, #
# and NaN operands are dispatched without entering this routine) #
# value in 68881/882 format at location (a0). #
# #
# A1. Convert the bcd exponent to binary by successive adds and #
# muls. Set the sign according to SE. Subtract 16 to compensate #
# for the mantissa which is to be interpreted as 17 integer #
# digits, rather than 1 integer and 16 fraction digits. #
# Note: this operation can never overflow. #
# #
# A2. Convert the bcd mantissa to binary by successive #
# adds and muls in FP0. Set the sign according to SM. #
# The mantissa digits will be converted with the decimal point #
# assumed following the least-significant digit. #
# Note: this operation can never overflow. #
# #
# A3. Count the number of leading/trailing zeros in the #
# bcd string. If SE is positive, count the leading zeros; #
# if negative, count the trailing zeros. Set the adjusted #
# exponent equal to the exponent from A1 and the zero count #
# added if SM = 1 and subtracted if SM = 0. Scale the #
# mantissa the equivalent of forcing in the bcd value: #
# #
# SM = 0 a non-zero digit in the integer position #
# SM = 1 a non-zero digit in Mant0, lsd of the fraction #
# #
# this will insure that any value, regardless of its #
# representation (ex. 0.1E2, 1E1, 10E0, 100E-1), is converted #
# consistently. #
# #
# A4. Calculate the factor 10^exp in FP1 using a table of #
# 10^(2^n) values. To reduce the error in forming factors #
# greater than 10^27, a directed rounding scheme is used with #
# tables rounded to RN, RM, and RP, according to the table #
# in the comments of the pwrten section. #
# #
# A5. Form the final binary number by scaling the mantissa by #
# the exponent factor. This is done by multiplying the #
# mantissa in FP0 by the factor in FP1 if the adjusted #
# exponent sign is positive, and dividing FP0 by FP1 if #
# it is negative. #
# #
# Clean up and return. Check if the final mul or div was inexact. #
# If so, set INEX1 in USER_FPSR. #
# #
#########################################################################
#
# PTENRN, PTENRM, and PTENRP are arrays of powers of 10 rounded
# to nearest, minus, and plus, respectively. The tables include
# 10**{1,2,4,8,16,32,64,128,256,512,1024,2048,4096}. No rounding
# is required until the power is greater than 27, however, all
# tables include the first 5 for ease of indexing.
#
RTABLE:
byte 0,0,0,0
byte 2,3,2,3
byte 2,3,3,2
byte 3,2,2,3
set FNIBS,7 set FSTRT,0
set ESTRT,4 set EDIGITS,2
global decbin
decbin:
mov.l 0x0(%a0),FP_SCR0_EX(%a6) # make a copy of input
mov.l 0x4(%a0),FP_SCR0_HI(%a6) # so we don't alter it
mov.l 0x8(%a0),FP_SCR0_LO(%a6)
lea FP_SCR0(%a6),%a0
movm.l &0x3c00,-(%sp) # save d2-d5
fmovm.x &0x1,-(%sp) # save fp1
#
# Calculate exponent:
# 1. Copy bcd value in memory for use as a working copy.
# 2. Calculate absolute value of exponent in d1 by mul and add.
# 3. Correct for exponent sign.
# 4. Subtract 16 to compensate for interpreting the mant as all integer digits.
# (i.e., all digits assumed left of the decimal point.)
#
# Register usage:
#
# calc_e:
# (*) d0: temp digit storage
# (*) d1: accumulator for binary exponent
# (*) d2: digit count
# (*) d3: offset pointer
# ( ) d4: first word of bcd
# ( ) a0: pointer to working bcd value
# ( ) a6: pointer to original bcd value
# (*) FP_SCR1: working copy of original bcd value
# (*) L_SCR1: copy of original exponent word
#
calc_e:
mov.l &EDIGITS,%d2 # # of nibbles (digits) in fraction part
mov.l &ESTRT,%d3 # counter to pick up digits
mov.l (%a0),%d4 # get first word of bcd clr.l %d1 # zero d1 for accumulator
e_gd:
mulu.l &0xa,%d1 # mul partial product by one digit place
bfextu %d4{%d3:&4},%d0 # get the digit and zero extend into d0
add.l %d0,%d1 # d1 = d1 + d0
addq.b &4,%d3 # advance d3 to the next digit
dbf.w %d2,e_gd # if we have used all 3 digits, exit loop
btst &30,%d4 # get SE
beq.b e_pos # don't negate if pos
neg.l %d1 # negate before subtracting
e_pos: sub.l &16,%d1 # sub to compensate for shift of mant
bge.b e_save # if still pos, do not neg
neg.l %d1 # now negative, make pos and set SE
or.l &0x40000000,%d4 # set SE in d4,
or.l &0x40000000,(%a0) # and in working bcd
e_save:
mov.l %d1,-(%sp) # save exp on stack
#
#
# Calculate mantissa:
# 1. Calculate absolute value of mantissa in fp0 by mul and add.
# 2. Correct for mantissa sign.
# (i.e., all digits assumed left of the decimal point.)
#
# Register usage:
#
# calc_m:
# (*) d0: temp digit storage
# (*) d1: lword counter
# (*) d2: digit count
# (*) d3: offset pointer
# ( ) d4: words 2 and 3 of bcd
# ( ) a0: pointer to working bcd value
# ( ) a6: pointer to original bcd value
# (*) fp0: mantissa accumulator
# ( ) FP_SCR1: working copy of original bcd value
# ( ) L_SCR1: copy of original exponent word
#
calc_m:
mov.l &1,%d1 # word counter, init to 1
fmov.s &0x00000000,%fp0 # accumulator
#
#
# Since the packed number has a long word between the first & second parts,
# get the integer digit then skip down & get the rest of the
# mantissa. We will unroll the loop once.
#
bfextu (%a0){&28:&4},%d0 # integer part is ls digit in long word
fadd.b %d0,%fp0 # add digit to sum in fp0
#
#
# Get the rest of the mantissa.
#
loadlw:
mov.l (%a0,%d1.L*4),%d4 # load mantissa lonqword into d4
mov.l &FSTRT,%d3 # counter to pick up digits
mov.l &FNIBS,%d2 # reset number of digits per a0 ptr
md2b:
fmul.s &0x41200000,%fp0 # fp0 = fp0 * 10
bfextu %d4{%d3:&4},%d0 # get the digit and zero extend
fadd.b %d0,%fp0 # fp0 = fp0 + digit
#
#
# If all the digits (8) in that long word have been converted (d2=0),
# then inc d1 (=2) to point to the next long word and reset d3 to 0
# to initialize the digit offset, and set d2 to 7 for the digit count;
# else continue with this long word.
#
addq.b &4,%d3 # advance d3 to the next digit
dbf.w %d2,md2b # check for last digit in this lw
nextlw:
addq.l &1,%d1 # inc lw pointer in mantissa
cmp.l %d1,&2 # test for last lw
ble.b loadlw # if not, get last one
#
# Check the sign of the mant and make the value in fp0 the same sign.
#
m_sign:
btst &31,(%a0) # test sign of the mantissa
beq.b ap_st_z # if clear, go to append/strip zeros
fneg.x %fp0 # if set, negate fp0
#
# Append/strip zeros:
#
# For adjusted exponents which have an absolute value greater than 27*,
# this routine calculates the amount needed to normalize the mantissa
# for the adjusted exponent. That number is subtracted from the exp
# if the exp was positive, and added if it was negative. The purpose
# of this is to reduce the value of the exponent and the possibility
# of error in calculation of pwrten.
#
# 1. Branch on the sign of the adjusted exponent.
# 2p.(positive exp)
# 2. Check M16 and the digits in lwords 2 and 3 in descending order.
# 3. Add one for each zero encountered until a non-zero digit.
# 4. Subtract the count from the exp.
# 5. Check if the exp has crossed zero in #3 above; make the exp abs
# and set SE.
# 6. Multiply the mantissa by 10**count.
# 2n.(negative exp)
# 2. Check the digits in lwords 3 and 2 in descending order.
# 3. Add one for each zero encountered until a non-zero digit.
# 4. Add the count to the exp.
# 5. Check if the exp has crossed zero in #3 above; clear SE.
# 6. Divide the mantissa by 10**count.
#
# *Why 27? If the adjusted exponent is within -28 < expA < 28, than
# any adjustment due to append/strip zeros will drive the resultane
# exponent towards zero. Since all pwrten constants with a power
# of 27 or less are exact, there is no need to use this routine to
# attempt to lessen the resultant exponent.
#
# Register usage:
#
# ap_st_z:
# (*) d0: temp digit storage
# (*) d1: zero count
# (*) d2: digit count
# (*) d3: offset pointer
# ( ) d4: first word of bcd
# (*) d5: lword counter
# ( ) a0: pointer to working bcd value
# ( ) FP_SCR1: working copy of original bcd value
# ( ) L_SCR1: copy of original exponent word
#
#
# First check the absolute value of the exponent to see if this
# routine is necessary. If so, then check the sign of the exponent
# and do append (+) or strip (-) zeros accordingly.
# This section handles a positive adjusted exponent.
#
ap_st_z:
mov.l (%sp),%d1 # load expA for range test
cmp.l %d1,&27 # test is with 27
ble.w pwrten # if abs(expA) <28, skip ap/st zeros
btst &30,(%a0) # check sign of exp
bne.b ap_st_n # if neg, go to neg side clr.l %d1 # zero count reg
mov.l (%a0),%d4 # load lword 1 to d4
bfextu %d4{&28:&4},%d0 # get M16 in d0
bne.b ap_p_fx # if M16 is non-zero, go fix exp
addq.l &1,%d1 # inc zero count
mov.l &1,%d5 # init lword counter
mov.l (%a0,%d5.L*4),%d4 # get lword 2 to d4
bne.b ap_p_cl # if lw 2 is zero, skip it
addq.l &8,%d1 # and inc count by 8
addq.l &1,%d5 # inc lword counter
mov.l (%a0,%d5.L*4),%d4 # get lword 3 to d4
ap_p_cl: clr.l %d3 # init offset reg
mov.l &7,%d2 # init digit counter
ap_p_gd:
bfextu %d4{%d3:&4},%d0 # get digit
bne.b ap_p_fx # if non-zero, go to fix exp
addq.l &4,%d3 # point to next digit
addq.l &1,%d1 # inc digit counter
dbf.w %d2,ap_p_gd # get next digit
ap_p_fx:
mov.l %d1,%d0 # copy counter to d2
mov.l (%sp),%d1 # get adjusted exp from memory sub.l %d0,%d1 # subtract count from exp
bge.b ap_p_fm # if still pos, go to pwrten
neg.l %d1 # now its neg; get abs
mov.l (%a0),%d4 # load lword 1 to d4
or.l &0x40000000,%d4 # and set SE in d4
or.l &0x40000000,(%a0) # and in memory
#
# Calculate the mantissa multiplier to compensate for the striping of
# zeros from the mantissa.
#
ap_p_fm:
lea.l PTENRN(%pc),%a1 # get address of power-of-ten table clr.l %d3 # init table index
fmov.s &0x3f800000,%fp1 # init fp1 to 1
mov.l &3,%d2 # init d2 to count bits in counter
ap_p_el:
asr.l &1,%d0 # shift lsb into carry
bcc.b ap_p_en # if 1, mul fp1 by pwrten factor
fmul.x (%a1,%d3),%fp1 # mul by 10**(d3_bit_no)
ap_p_en:
add.l &12,%d3 # inc d3 to next rtable entry
tst.l %d0 # check if d0 is zero
bne.b ap_p_el # if not, get next bit
fmul.x %fp1,%fp0 # mul mantissa by 10**(no_bits_shifted)
bra.b pwrten # go calc pwrten
#
# This section handles a negative adjusted exponent.
#
ap_st_n: clr.l %d1 # clr counter
mov.l &2,%d5 # set up d5 to point to lword 3
mov.l (%a0,%d5.L*4),%d4 # get lword 3
bne.b ap_n_cl # if not zero, check digits sub.l &1,%d5 # dec d5 to point to lword 2
addq.l &8,%d1 # inc counter by 8
mov.l (%a0,%d5.L*4),%d4 # get lword 2
ap_n_cl:
mov.l &28,%d3 # point to last digit
mov.l &7,%d2 # init digit counter
ap_n_gd:
bfextu %d4{%d3:&4},%d0 # get digit
bne.b ap_n_fx # if non-zero, go to exp fix
subq.l &4,%d3 # point to previous digit
addq.l &1,%d1 # inc digit counter
dbf.w %d2,ap_n_gd # get next digit
ap_n_fx:
mov.l %d1,%d0 # copy counter to d0
mov.l (%sp),%d1 # get adjusted exp from memory sub.l %d0,%d1 # subtract count from exp
bgt.b ap_n_fm # if still pos, go fix mantissa
neg.l %d1 # take abs of exp and clr SE
mov.l (%a0),%d4 # load lword 1 to d4
and.l &0xbfffffff,%d4 # and clr SE in d4
and.l &0xbfffffff,(%a0) # and in memory
#
# Calculate the mantissa multiplier to compensate for the appending of
# zeros to the mantissa.
#
ap_n_fm:
lea.l PTENRN(%pc),%a1 # get address of power-of-ten table clr.l %d3 # init table index
fmov.s &0x3f800000,%fp1 # init fp1 to 1
mov.l &3,%d2 # init d2 to count bits in counter
ap_n_el:
asr.l &1,%d0 # shift lsb into carry
bcc.b ap_n_en # if 1, mul fp1 by pwrten factor
fmul.x (%a1,%d3),%fp1 # mul by 10**(d3_bit_no)
ap_n_en:
add.l &12,%d3 # inc d3 to next rtable entry
tst.l %d0 # check if d0 is zero
bne.b ap_n_el # if not, get next bit
fdiv.x %fp1,%fp0 # div mantissa by 10**(no_bits_shifted)
#
#
# Calculate power-of-ten factor from adjusted and shifted exponent.
#
# Register usage:
#
# pwrten:
# (*) d0: temp
# ( ) d1: exponent
# (*) d2: {FPCR[6:5],SM,SE} as index in RTABLE; temp
# (*) d3: FPCR work copy
# ( ) d4: first word of bcd
# (*) a1: RTABLE pointer
# calc_p:
# (*) d0: temp
# ( ) d1: exponent
# (*) d3: PWRTxx table index
# ( ) a0: pointer to working copy of bcd
# (*) a1: PWRTxx pointer
# (*) fp1: power-of-ten accumulator
#
# Pwrten calculates the exponent factor in the selected rounding mode
# according to the following table:
#
# Sign of Mant Sign of Exp Rounding Mode PWRTEN Rounding Mode
#
# ANY ANY RN RN
#
# + + RP RP
# - + RP RM
# + - RP RM
# - - RP RP
#
# + + RM RM
# - + RM RP
# + - RM RP
# - - RM RM
#
# + + RZ RM
# - + RZ RM
# + - RZ RP
# - - RZ RP
#
#
pwrten:
mov.l USER_FPCR(%a6),%d3 # get user's FPCR
bfextu %d3{&26:&2},%d2 # isolate rounding mode bits
mov.l (%a0),%d4 # reload 1st bcd word to d4
asl.l &2,%d2 # format d2 to be
bfextu %d4{&0:&2},%d0 # {FPCR[6],FPCR[5],SM,SE}
add.l %d0,%d2 # in d2 as index into RTABLE
lea.l RTABLE(%pc),%a1 # load rtable base
mov.b (%a1,%d2),%d0 # load new rounding bits from table clr.l %d3 # clear d3 to force no exc and extended
bfins %d0,%d3{&26:&2} # stuff new rounding bits in FPCR
fmov.l %d3,%fpcr # write new FPCR
asr.l &1,%d0 # write correct PTENxx table
bcc.b not_rp # to a1
lea.l PTENRP(%pc),%a1 # it is RP
bra.b calc_p # go to init section
not_rp:
asr.l &1,%d0 # keep checking
bcc.b not_rm
lea.l PTENRM(%pc),%a1 # it is RM
bra.b calc_p # go to init section
not_rm:
lea.l PTENRN(%pc),%a1 # it is RN
calc_p:
mov.l %d1,%d0 # copy exp to d0;use d0
bpl.b no_neg # if exp is negative,
neg.l %d0 # invert it
or.l &0x40000000,(%a0) # and set SE bit
no_neg: clr.l %d3 # table index
fmov.s &0x3f800000,%fp1 # init fp1 to 1
e_loop:
asr.l &1,%d0 # shift next bit into carry
bcc.b e_next # if zero, skip the mul
fmul.x (%a1,%d3),%fp1 # mul by 10**(d3_bit_no)
e_next:
add.l &12,%d3 # inc d3 to next rtable entry
tst.l %d0 # check if d0 is zero
bne.b e_loop # not zero, continue shifting
#
#
# Check the sign of the adjusted exp and make the value in fp0 the
# same sign. If the exp was pos then multiply fp1*fp0;
# else divide fp0/fp1.
#
# Register Usage:
# norm:
# ( ) a0: pointer to working bcd value
# (*) fp0: mantissa accumulator
# ( ) fp1: scaling factor - 10**(abs(exp))
#
pnorm:
btst &30,(%a0) # test the sign of the exponent
beq.b mul # if clear, go to multiply
div:
fdiv.x %fp1,%fp0 # exp is negative, so divide mant by exp
bra.b end_dec
mul:
fmul.x %fp1,%fp0 # exp is positive, so multiply by exp
#
#
# Clean up and return with result in fp0.
#
# If the final mul/div in decbin incurred an inex exception,
# it will be inex2, but will be reported as inex1 by get_op.
#
end_dec:
fmov.l %fpsr,%d0 # get status register
bclr &inex2_bit+8,%d0 # test for inex2 and clear it
beq.b no_exc # skip this if no exc
ori.w &inx1a_mask,2+USER_FPSR(%a6) # set INEX1/AINEX
no_exc:
add.l &0x4,%sp # clear 1 lw param
fmovm.x (%sp)+,&0x40 # restore fp1
movm.l (%sp)+,&0x3c # restore d2-d5
fmov.l &0x0,%fpcr
fmov.l &0x0,%fpsr
rts
#########################################################################
# bindec(): Converts an input in extended precision format to bcd format#
# #
# INPUT *************************************************************** #
# a0 = pointer to the input extended precision value in memory. #
# the input may be either normalized, unnormalized, or #
# denormalized. #
# d0 = contains the k-factor sign-extended to 32-bits. #
# #
# OUTPUT ************************************************************** #
# FP_SCR0(a6) = bcd format result on the stack. #
# #
# ALGORITHM *********************************************************** #
# #
# A1. Set RM and size ext; Set SIGMA = sign of input. #
# The k-factor is saved for use in d7. Clear the #
# BINDEC_FLG for separating normalized/denormalized #
# input. If input is unnormalized or denormalized, #
# normalize it. #
# #
# A2. Set X = abs(input). #
# #
# A3. Compute ILOG. #
# ILOG is the log base 10 of the input value. It is #
# approximated by adding e + 0.f when the original #
# value is viewed as 2^^e * 1.f in extended precision. #
# This value is stored in d6. #
# #
# A4. Clr INEX bit. #
# The operation in A3 above may have set INEX2. #
# #
# A5. Set ICTR = 0; #
# ICTR is a flag used in A13. It must be set before the #
# loop entry A6. #
# #
# A6. Calculate LEN. #
# LEN is the number of digits to be displayed. The #
# k-factor can dictate either the total number of digits, #
# if it is a positive number, or the number of digits #
# after the decimal point which are to be included as #
# significant. See the 68882 manual for examples. #
# If LEN is computed to be greater than 17, set OPERR in #
# USER_FPSR. LEN is stored in d4. #
# #
# A7. Calculate SCALE. #
# SCALE is equal to 10^ISCALE, where ISCALE is the number #
# of decimal places needed to insure LEN integer digits #
# in the output before conversion to bcd. LAMBDA is the #
# sign of ISCALE, used in A9. Fp1 contains #
# 10^^(abs(ISCALE)) using a rounding mode which is a #
# function of the original rounding mode and the signs #
# of ISCALE and X. A table is given in the code. #
# #
# A8. Clr INEX; Force RZ. #
# The operation in A3 above may have set INEX2. #
# RZ mode is forced for the scaling operation to insure #
# only one rounding error. The grs bits are collected in #
# the INEX flag for use in A10. #
# #
# A9. Scale X -> Y. #
# The mantissa is scaled to the desired number of #
# significant digits. The excess digits are collected #
# in INEX2. #
# #
# A10. Or in INEX. #
# If INEX is set, round error occurred. This is #
# compensated for by 'or-ing' in the INEX2 flag to #
# the lsb of Y. #
# #
# A11. Restore original FPCR; setsize ext. #
# Perform FINT operation in the user's rounding mode. #
# Keep the size to extended. #
# #
# A12. Calculate YINT = FINT(Y) according to user's rounding #
# mode. The FPSP routine sintd0 is used. The output #
# is in fp0. #
# #
# A13. Check for LEN digits. #
# If the int operation results in more than LEN digits, #
# or less than LEN -1 digits, adjust ILOG and repeat from #
# A6. This test occurs only on the first pass. If the #
# result is exactly 10^LEN, decrement ILOG and divide #
# the mantissa by 10. #
# #
# A14. Convert the mantissa to bcd. #
# The binstr routine is used to convert the LEN digit #
# mantissa to bcd in memory. The input to binstr is #
# to be a fraction; i.e. (mantissa)/10^LEN and adjusted #
# such that the decimal point is to the left of bit 63. #
# The bcd digits are stored in the correct position in #
# the final string area in memory. #
# #
# A15. Convert the exponent to bcd. #
# As in A14 above, the exp is converted to bcd and the #
# digits are stored in the final string. #
# Test the length of the final exponent string. If the #
# length is 4, set operr. #
# #
# A16. Write sign bits to final string. #
# #
#########################################################################
set BINDEC_FLG, EXC_TEMP # DENORM flag
# Constants in extended precision
PLOG2:
long 0x3FFD0000,0x9A209A84,0xFBCFF798,0x00000000
PLOG2UP1:
long 0x3FFD0000,0x9A209A84,0xFBCFF799,0x00000000
# Constants in single precision
FONE:
long 0x3F800000,0x00000000,0x00000000,0x00000000
FTWO:
long 0x40000000,0x00000000,0x00000000,0x00000000
FTEN:
long 0x41200000,0x00000000,0x00000000,0x00000000
F4933:
long 0x459A2800,0x00000000,0x00000000,0x00000000
# A1. Set RM and size ext. Set SIGMA = sign input;
# The k-factor is saved for use in d7. Clear BINDEC_FLG for
# separating normalized/denormalized input. If the input
# is a denormalized number, set the BINDEC_FLG memory word
# to signal denorm. If the input is unnormalized, normalize
# the input and test for denormalized result.
#
fmov.l &rm_mode*0x10,%fpcr # set RM and ext
mov.l (%a0),L_SCR2(%a6) # save exponent for sign check
mov.l %d0,%d7 # move k-factor to d7
clr.b BINDEC_FLG(%a6) # clr norm/denorm flag
cmpi.b STAG(%a6),&DENORM # is input a DENORM?
bne.w A2_str # no; input is a NORM
#
# Normalize the denorm
#
un_de_norm:
mov.w (%a0),%d0
and.w &0x7fff,%d0 # strip sign of normalized exp
mov.l 4(%a0),%d1
mov.l 8(%a0),%d2
norm_loop: sub.w &1,%d0
lsl.l &1,%d2
roxl.l &1,%d1
tst.l %d1
bge.b norm_loop
#
# Test if the normalized input is denormalized
#
tst.w %d0
bgt.b pos_exp # if greater than zero, it is a norm st BINDEC_FLG(%a6) # set flag for denorm
pos_exp:
and.w &0x7fff,%d0 # strip sign of normalized exp
mov.w %d0,(%a0)
mov.l %d1,4(%a0)
mov.l %d2,8(%a0)
# A2. Set X = abs(input).
#
A2_str:
mov.l (%a0),FP_SCR1(%a6) # move input to work space
mov.l 4(%a0),FP_SCR1+4(%a6) # move input to work space
mov.l 8(%a0),FP_SCR1+8(%a6) # move input to work space
and.l &0x7fffffff,FP_SCR1(%a6) # create abs(X)
# A3. Compute ILOG.
# ILOG is the log base 10 of the input value. It is approx-
# imated by adding e + 0.f when the original value is viewed
# as 2^^e * 1.f in extended precision. This value is stored
# in d6.
#
# Register usage:
# Input/Output
# d0: k-factor/exponent
# d2: x/x
# d3: x/x
# d4: x/x
# d5: x/x
# d6: x/ILOG
# d7: k-factor/Unchanged
# a0: ptr for original operand/final result
# a1: x/x
# a2: x/x
# fp0: x/float(ILOG)
# fp1: x/x
# fp2: x/x
# F_SCR1:x/x
# F_SCR2:Abs(X)/Abs(X) with $3fff exponent
# L_SCR1:x/x
# L_SCR2:first word of X packed/Unchanged
tst.b BINDEC_FLG(%a6) # check for denorm
beq.b A3_cont # if clr, continue with norm
mov.l &-4933,%d6 # force ILOG = -4933
bra.b A4_str
A3_cont:
mov.w FP_SCR1(%a6),%d0 # move exp to d0
mov.w &0x3fff,FP_SCR1(%a6) # replace exponent with 0x3fff
fmov.x FP_SCR1(%a6),%fp0 # now fp0 has 1.f sub.w &0x3fff,%d0 # strip off bias
fadd.w %d0,%fp0 # add in exp
fsub.s FONE(%pc),%fp0 # subtract off 1.0
fbge.w pos_res # if pos, branch
fmul.x PLOG2UP1(%pc),%fp0 # if neg, mul by LOG2UP1
fmov.l %fp0,%d6 # put ILOG in d6 as a lword
bra.b A4_str # go move out ILOG
pos_res:
fmul.x PLOG2(%pc),%fp0 # if pos, mul by LOG2
fmov.l %fp0,%d6 # put ILOG in d6 as a lword
# A4. Clr INEX bit.
# The operation in A3 above may have set INEX2.
A4_str:
fmov.l &0,%fpsr # zero all of fpsr - nothing needed
# A5. Set ICTR = 0;
# ICTR is a flag used in A13. It must be set before the
# loop entry A6. The lower word of d5 is used for ICTR.
clr.w %d5 # clear ICTR
# A6. Calculate LEN.
# LEN is the number of digits to be displayed. The k-factor
# can dictate either the total number of digits, if it is
# a positive number, or the number of digits after the
# original decimal point which are to be included as
# significant. See the 68882 manual for examples.
# If LEN is computed to be greater than 17, set OPERR in
# USER_FPSR. LEN is stored in d4.
#
# Register usage:
# Input/Output
# d0: exponent/Unchanged
# d2: x/x/scratch
# d3: x/x
# d4: exc picture/LEN
# d5: ICTR/Unchanged
# d6: ILOG/Unchanged
# d7: k-factor/Unchanged
# a0: ptr for original operand/final result
# a1: x/x
# a2: x/x
# fp0: float(ILOG)/Unchanged
# fp1: x/x
# fp2: x/x
# F_SCR1:x/x
# F_SCR2:Abs(X) with $3fff exponent/Unchanged
# L_SCR1:x/x
# L_SCR2:first word of X packed/Unchanged
A6_str:
tst.l %d7 # branch on sign of k
ble.b k_neg # if k <= 0, LEN = ILOG + 1 - k
mov.l %d7,%d4 # if k > 0, LEN = k
bra.b len_ck # skip to LEN check
k_neg:
mov.l %d6,%d4 # first load ILOG to d4 sub.l %d7,%d4 # subtract off k
addq.l &1,%d4 # add in the 1
len_ck:
tst.l %d4 # LEN check: branch on sign of LEN
ble.b LEN_ng # if neg, set LEN = 1
cmp.l %d4,&17 # test if LEN > 17
ble.b A7_str # if not, forget it
mov.l &17,%d4 # set max LEN = 17
tst.l %d7 # if negative, never set OPERR
ble.b A7_str # if positive, continue
or.l &opaop_mask,USER_FPSR(%a6) # set OPERR & AIOP in USER_FPSR
bra.b A7_str # finished here
LEN_ng:
mov.l &1,%d4 # min LEN is 1
# A7. Calculate SCALE.
# SCALE is equal to 10^ISCALE, where ISCALE is the number
# of decimal places needed to insure LEN integer digits
# in the output before conversion to bcd. LAMBDA is the sign
# of ISCALE, used in A9. Fp1 contains 10^^(abs(ISCALE)) using
# the rounding mode as given in the following table (see
# Coonen, p. 7.23 as ref.; however, the SCALE variable is
# of opposite sign in bindec.sa from Coonen).
#
# Initial USE
# FPCR[6:5] LAMBDA SIGN(X) FPCR[6:5]
# ----------------------------------------------
# RN 00 0 0 00/0 RN
# RN 00 0 1 00/0 RN
# RN 00 1 0 00/0 RN
# RN 00 1 1 00/0 RN
# RZ 01 0 0 11/3 RP
# RZ 01 0 1 11/3 RP
# RZ 01 1 0 10/2 RM
# RZ 01 1 1 10/2 RM
# RM 10 0 0 11/3 RP
# RM 10 0 1 10/2 RM
# RM 10 1 0 10/2 RM
# RM 10 1 1 11/3 RP
# RP 11 0 0 10/2 RM
# RP 11 0 1 11/3 RP
# RP 11 1 0 11/3 RP
# RP 11 1 1 10/2 RM
#
# Register usage:
# Input/Output
# d0: exponent/scratch - final is 0
# d2: x/0 or 24 for A9
# d3: x/scratch - offset ptr into PTENRM array
# d4: LEN/Unchanged
# d5: 0/ICTR:LAMBDA
# d6: ILOG/ILOG or k if ((k<=0)&(ILOG<k))
# d7: k-factor/Unchanged
# a0: ptr for original operand/final result
# a1: x/ptr to PTENRM array
# a2: x/x
# fp0: float(ILOG)/Unchanged
# fp1: x/10^ISCALE
# fp2: x/x
# F_SCR1:x/x
# F_SCR2:Abs(X) with $3fff exponent/Unchanged
# L_SCR1:x/x
# L_SCR2:first word of X packed/Unchanged
A7_str:
tst.l %d7 # test sign of k
bgt.b k_pos # if pos and > 0, skip this
cmp.l %d7,%d6 # test k - ILOG
blt.b k_pos # if ILOG >= k, skip this
mov.l %d7,%d6 # if ((k<0) & (ILOG < k)) ILOG = k
k_pos:
mov.l %d6,%d0 # calc ILOG + 1 - LEN in d0
addq.l &1,%d0 # add the 1 sub.l %d4,%d0 # sub off LEN
swap %d5 # use upper word of d5 for LAMBDA clr.w %d5 # set it zero initially clr.w %d2 # set up d2 for very small case
tst.l %d0 # test sign of ISCALE
bge.b iscale # if pos, skip next inst
addq.w &1,%d5 # if neg, set LAMBDA true
cmp.l %d0,&0xffffecd4 # test iscale <= -4908
bgt.b no_inf # if false, skip rest
add.l &24,%d0 # add in 24 to iscale
mov.l &24,%d2 # put 24 in d2 for A9
no_inf:
neg.l %d0 # and take abs of ISCALE
iscale:
fmov.s FONE(%pc),%fp1 # init fp1 to 1
bfextu USER_FPCR(%a6){&26:&2},%d1 # get initial rmode bits
lsl.w &1,%d1 # put them in bits 2:1
add.w %d5,%d1 # add in LAMBDA
lsl.w &1,%d1 # put them in bits 3:1
tst.l L_SCR2(%a6) # test sign of original x
bge.b x_pos # if pos, don't set bit 0
addq.l &1,%d1 # if neg, set bit 0
x_pos:
lea.l RBDTBL(%pc),%a2 # load rbdtbl base
mov.b (%a2,%d1),%d3 # load d3 with new rmode
lsl.l &4,%d3 # put bits in proper position
fmov.l %d3,%fpcr # load bits into fpu
lsr.l &4,%d3 # put bits in proper position
tst.b %d3 # decode new rmode for pten table
bne.b not_rn # if zero, it is RN
lea.l PTENRN(%pc),%a1 # load a1 with RN table base
bra.b rmode # exit decode
not_rn:
lsr.b &1,%d3 # get lsb in carry
bcc.b not_rp2 # if carry clear, it is RM
lea.l PTENRP(%pc),%a1 # load a1 with RP table base
bra.b rmode # exit decode
not_rp2:
lea.l PTENRM(%pc),%a1 # load a1 with RM table base
rmode: clr.l %d3 # clr table index
e_loop2:
lsr.l &1,%d0 # shift next bit into carry
bcc.b e_next2 # if zero, skip the mul
fmul.x (%a1,%d3),%fp1 # mul by 10**(d3_bit_no)
e_next2:
add.l &12,%d3 # inc d3 to next pwrten table entry
tst.l %d0 # test if ISCALE is zero
bne.b e_loop2 # if not, loop
# A8. Clr INEX; Force RZ.
# The operation in A3 above may have set INEX2.
# RZ mode is forced for the scaling operation to insure
# only one rounding error. The grs bits are collected in
# the INEX flag for use in A10.
#
# Register usage:
# Input/Output
# A9. Scale X -> Y.
# The mantissa is scaled to the desired number of significant
# digits. The excess digits are collected in INEX2. If mul,
# Check d2 for excess 10 exponential value. If not zero,
# the iscale value would have caused the pwrten calculation
# to overflow. Only a negative iscale can cause this, so
# multiply by 10^(d2), which is now only allowed to be 24,
# with a multiply by 10^8 and 10^16, which is exact since
# 10^24 is exact. If the input was denormalized, we must
# create a busy stack frame with the mul command and the
# two operands, and allow the fpu to complete the multiply.
#
# Register usage:
# Input/Output
# d0: FPCR with RZ mode/Unchanged
# d2: 0 or 24/unchanged
# d3: x/x
# d4: LEN/Unchanged
# d5: ICTR:LAMBDA
# d6: ILOG/Unchanged
# d7: k-factor/Unchanged
# a0: ptr for original operand/final result
# a1: ptr to PTENRM array/Unchanged
# a2: x/x
# fp0: float(ILOG)/X adjusted for SCALE (Y)
# fp1: 10^ISCALE/Unchanged
# fp2: x/x
# F_SCR1:x/x
# F_SCR2:Abs(X) with $3fff exponent/Unchanged
# L_SCR1:x/x
# L_SCR2:first word of X packed/Unchanged
A9_str:
fmov.x (%a0),%fp0 # load X from memory
fabs.x %fp0 # use abs(X)
tst.w %d5 # LAMBDA is in lower word of d5
bne.b sc_mul # if neg (LAMBDA = 1), scale by mul
fdiv.x %fp1,%fp0 # calculate X / SCALE -> Y to fp0
bra.w A10_st # branch to A10
sc_mul:
tst.b BINDEC_FLG(%a6) # check for denorm
beq.w A9_norm # if norm, continue with mul
# for DENORM, we must calculate:
# fp0 = input_op * 10^ISCALE * 10^24
# since the input operand is a DENORM, we can't multiply it directly.
# so, we do the multiplication of the exponents and mantissas separately.
# in this way, we avoid underflow on intermediate stages of the
# multiplication and guarantee a result without exception.
fmovm.x &0x2,-(%sp) # save 10^ISCALE to stack
andi.w &0x8000,(%sp) # keep sign
or.w %d3,(%sp) # insert new exponent
andi.w &0x7fff,(%a0) # clear sign bit on DENORM again
mov.l 0x8(%a0),-(%sp) # put input op mantissa on stk
mov.l 0x4(%a0),-(%sp)
mov.l &0x3fff0000,-(%sp) # force exp to zero
fmovm.x (%sp)+,&0x80 # load normalized DENORM into fp0
fmul.x (%sp)+,%fp0
# fmul.x 36(%a1),%fp0 # multiply fp0 by 10^8
# fmul.x 48(%a1),%fp0 # multiply fp0 by 10^16
mov.l 36+8(%a1),-(%sp) # get 10^8 mantissa
mov.l 36+4(%a1),-(%sp)
mov.l &0x3fff0000,-(%sp) # force exp to zero
mov.l 48+8(%a1),-(%sp) # get 10^16 mantissa
mov.l 48+4(%a1),-(%sp)
mov.l &0x3fff0000,-(%sp)# force exp to zero
fmul.x (%sp)+,%fp0 # multiply fp0 by 10^8
fmul.x (%sp)+,%fp0 # multiply fp0 by 10^16
bra.b A10_st
sc_mul_err:
bra.b sc_mul_err
A9_norm:
tst.w %d2 # test for small exp case
beq.b A9_con # if zero, continue as normal
fmul.x 36(%a1),%fp0 # multiply fp0 by 10^8
fmul.x 48(%a1),%fp0 # multiply fp0 by 10^16
A9_con:
fmul.x %fp1,%fp0 # calculate X * SCALE -> Y to fp0
# A10. Or in INEX.
# If INEX is set, round error occurred. This is compensated
# for by 'or-ing' in the INEX2 flag to the lsb of Y.
#
# Register usage:
# Input/Output
# d0: FPCR with RZ mode/FPSR with INEX2 isolated
# d2: x/x
# d3: x/x
# d4: LEN/Unchanged
# d5: ICTR:LAMBDA
# d6: ILOG/Unchanged
# d7: k-factor/Unchanged
# a0: ptr for original operand/final result
# a1: ptr to PTENxx array/Unchanged
# a2: x/ptr to FP_SCR1(a6)
# fp0: Y/Y with lsb adjusted
# fp1: 10^ISCALE/Unchanged
# fp2: x/x
A10_st:
fmov.l %fpsr,%d0 # get FPSR
fmov.x %fp0,FP_SCR1(%a6) # move Y to memory
lea.l FP_SCR1(%a6),%a2 # load a2 with ptr to FP_SCR1
btst &9,%d0 # check if INEX2 set
beq.b A11_st # if clear, skip rest
or.l &1,8(%a2) # or in 1 to lsb of mantissa
fmov.x FP_SCR1(%a6),%fp0 # write adjusted Y back to fpu
# A11. Restore original FPCR; setsize ext.
# Perform FINT operation in the user's rounding mode. Keep
# the size to extended. The sintdo entry point in the sint
# routine expects the FPCR value to be in USER_FPCR for
# mode and precision. The original FPCR is saved in L_SCR1.
A11_st:
mov.l USER_FPCR(%a6),L_SCR1(%a6) # save it for later
and.l &0x00000030,USER_FPCR(%a6) # setsize to ext,
# ;block exceptions
# A12. Calculate YINT = FINT(Y) according to user's rounding mode.
# The FPSP routine sintd0 is used. The output is in fp0.
#
# Register usage:
# Input/Output
# d0: FPSR with AINEX cleared/FPCR with sizeset to ext
# d2: x/x/scratch
# d3: x/x
# d4: LEN/Unchanged
# d5: ICTR:LAMBDA/Unchanged
# d6: ILOG/Unchanged
# d7: k-factor/Unchanged
# a0: ptr for original operand/src ptr for sintdo
# a1: ptr to PTENxx array/Unchanged
# a2: ptr to FP_SCR1(a6)/Unchanged
# a6: temp pointer to FP_SCR1(a6) - orig value saved and restored
# fp0: Y/YINT
# fp1: 10^ISCALE/Unchanged
# fp2: x/x
# F_SCR1:x/x
# F_SCR2:Y adjusted for inex/Y with original exponent
# L_SCR1:x/original USER_FPCR
# L_SCR2:first word of X packed/Unchanged
A12_st:
movm.l &0xc0c0,-(%sp) # save regs used by sintd0 {%d0-%d1/%a0-%a1}
mov.l L_SCR1(%a6),-(%sp)
mov.l L_SCR2(%a6),-(%sp)
lea.l FP_SCR1(%a6),%a0 # a0 is ptr to FP_SCR1(a6)
fmov.x %fp0,(%a0) # move Y to memory at FP_SCR1(a6)
tst.l L_SCR2(%a6) # test sign of original operand
bge.b do_fint12 # if pos, use Y
or.l &0x80000000,(%a0) # if neg, use -Y
do_fint12:
mov.l USER_FPSR(%a6),-(%sp)
# bsr sintdo # sint routine returns int in fp0
# A13. Check for LEN digits.
# If the int operation results in more than LEN digits,
# or less than LEN -1 digits, adjust ILOG and repeat from
# A6. This test occurs only on the first pass. If the
# result is exactly 10^LEN, decrement ILOG and divide
# the mantissa by 10. The calculation of 10^LEN cannot
# be inexact, since all powers of ten up to 10^27 are exact
# in extended precision, so the use of a previous power-of-ten
# table will introduce no error.
#
#
# Register usage:
# Input/Output
# d0: FPCR with sizeset to ext/scratch final = 0
# d2: x/x
# d3: x/scratch final = x
# d4: LEN/LEN adjusted
# d5: ICTR:LAMBDA/LAMBDA:ICTR
# d6: ILOG/ILOG adjusted
# d7: k-factor/Unchanged
# a0: pointer into memory for packed bcd string formation
# a1: ptr to PTENxx array/Unchanged
# a2: ptr to FP_SCR1(a6)/Unchanged
# fp0: int portion of Y/abs(YINT) adjusted
# fp1: 10^ISCALE/Unchanged
# fp2: x/10^LEN
# F_SCR1:x/x
# F_SCR2:Y with original exponent/Unchanged
# L_SCR1:original USER_FPCR/Unchanged
# L_SCR2:first word of X packed/Unchanged
A13_st:
swap %d5 # put ICTR in lower word of d5
tst.w %d5 # check if ICTR = 0
bne not_zr # if non-zero, go to second test
#
# Compute 10^(LEN-1)
#
fmov.s FONE(%pc),%fp2 # init fp2 to 1.0
mov.l %d4,%d0 # put LEN in d0
subq.l &1,%d0 # d0 = LEN -1 clr.l %d3 # clr table index
l_loop:
lsr.l &1,%d0 # shift next bit into carry
bcc.b l_next # if zero, skip the mul
fmul.x (%a1,%d3),%fp2 # mul by 10**(d3_bit_no)
l_next:
add.l &12,%d3 # inc d3 to next pwrten table entry
tst.l %d0 # test if LEN is zero
bne.b l_loop # if not, loop
#
# 10^LEN-1 is computed for this test and A14. If the input was
# denormalized, check only the case in which YINT > 10^LEN.
#
tst.b BINDEC_FLG(%a6) # check if input was norm
beq.b A13_con # if norm, continue with checking
fabs.x %fp0 # take abs of YINT
bra test_2
#
# Compare abs(YINT) to 10^(LEN-1) and 10^LEN
#
A13_con:
fabs.x %fp0 # take abs of YINT
fcmp.x %fp0,%fp2 # compare abs(YINT) with 10^(LEN-1)
fbge.w test_2 # if greater, do next test
subq.l &1,%d6 # subtract 1 from ILOG
mov.w &1,%d5 # set ICTR
fmov.l &rm_mode*0x10,%fpcr # set rmode to RM
fmul.s FTEN(%pc),%fp2 # compute 10^LEN
bra.w A6_str # return to A6 and recompute YINT
test_2:
fmul.s FTEN(%pc),%fp2 # compute 10^LEN
fcmp.x %fp0,%fp2 # compare abs(YINT) with 10^LEN
fblt.w A14_st # if less, all is ok, go to A14
fbgt.w fix_ex # if greater, fix and redo
fdiv.s FTEN(%pc),%fp0 # if equal, divide by 10
addq.l &1,%d6 # and inc ILOG
bra.b A14_st # and continue elsewhere
fix_ex:
addq.l &1,%d6 # increment ILOG by 1
mov.w &1,%d5 # set ICTR
fmov.l &rm_mode*0x10,%fpcr # set rmode to RM
bra.w A6_str # return to A6 and recompute YINT
#
# Since ICTR <> 0, we have already been through one adjustment,
# and shouldn't have another; this is to check if abs(YINT) = 10^LEN
# 10^LEN is again computed using whatever table is in a1 since the
# value calculated cannot be inexact.
#
not_zr:
fmov.s FONE(%pc),%fp2 # init fp2 to 1.0
mov.l %d4,%d0 # put LEN in d0 clr.l %d3 # clr table index
z_loop:
lsr.l &1,%d0 # shift next bit into carry
bcc.b z_next # if zero, skip the mul
fmul.x (%a1,%d3),%fp2 # mul by 10**(d3_bit_no)
z_next:
add.l &12,%d3 # inc d3 to next pwrten table entry
tst.l %d0 # test if LEN is zero
bne.b z_loop # if not, loop
fabs.x %fp0 # get abs(YINT)
fcmp.x %fp0,%fp2 # check if abs(YINT) = 10^LEN
fbneq.w A14_st # if not, skip this
fdiv.s FTEN(%pc),%fp0 # divide abs(YINT) by 10
addq.l &1,%d6 # and inc ILOG by 1
addq.l &1,%d4 # and inc LEN
fmul.s FTEN(%pc),%fp2 # if LEN++, the get 10^^LEN
# A14. Convert the mantissa to bcd.
# The binstr routine is used to convert the LEN digit
# mantissa to bcd in memory. The input to binstr is
# to be a fraction; i.e. (mantissa)/10^LEN and adjusted
# such that the decimal point is to the left of bit 63.
# The bcd digits are stored in the correct position in
# the final string area in memory.
#
#
# Register usage:
# Input/Output
# d0: x/LEN call to binstr - final is 0
# d1: x/0
# d2: x/ms 32-bits of mant of abs(YINT)
# d3: x/ls 32-bits of mant of abs(YINT)
# d4: LEN/Unchanged
# d5: ICTR:LAMBDA/LAMBDA:ICTR
# d6: ILOG
# d7: k-factor/Unchanged
# a0: pointer into memory for packed bcd string formation
# /ptr to first mantissa byte in result string
# a1: ptr to PTENxx array/Unchanged
# a2: ptr to FP_SCR1(a6)/Unchanged
# fp0: int portion of Y/abs(YINT) adjusted
# fp1: 10^ISCALE/Unchanged
# fp2: 10^LEN/Unchanged
# F_SCR1:x/Work area for final result
# F_SCR2:Y with original exponent/Unchanged
# L_SCR1:original USER_FPCR/Unchanged
# L_SCR2:first word of X packed/Unchanged
A14_st:
fmov.l &rz_mode*0x10,%fpcr # force rz for conversion
fdiv.x %fp2,%fp0 # divide abs(YINT) by 10^LEN
lea.l FP_SCR0(%a6),%a0
fmov.x %fp0,(%a0) # move abs(YINT)/10^LEN to memory
mov.l 4(%a0),%d2 # move 2nd word of FP_RES to d2
mov.l 8(%a0),%d3 # move 3rd word of FP_RES to d3 clr.l 4(%a0) # zero word 2 of FP_RES clr.l 8(%a0) # zero word 3 of FP_RES
mov.l (%a0),%d0 # move exponent to d0
swap %d0 # put exponent in lower word
beq.b no_sft # if zero, don't shift sub.l &0x3ffd,%d0 # sub bias less 2 to make fract
tst.l %d0 # check if > 1
bgt.b no_sft # if so, don't shift
neg.l %d0 # make exp positive
m_loop:
lsr.l &1,%d2 # shift d2:d3 right, add 0s
roxr.l &1,%d3 # the number of places
dbf.w %d0,m_loop # given in d0
no_sft:
tst.l %d2 # check for mantissa of zero
bne.b no_zr # if not, go on
tst.l %d3 # continue zero check
beq.b zer_m # if zero, go directly to binstr
no_zr: clr.l %d1 # put zero in d1 for addx
add.l &0x00000080,%d3 # inc at bit 7
addx.l %d1,%d2 # continue inc
and.l &0xffffff80,%d3 # strip off lsb not used by 882
zer_m:
mov.l %d4,%d0 # put LEN in d0 for binstr call
addq.l &3,%a0 # a0 points to M16 byte in result
bsr binstr # call binstr to convert mant
# A15. Convert the exponent to bcd.
# As in A14 above, the exp is converted to bcd and the
# digits are stored in the final string.
#
# Digits are stored in L_SCR1(a6) on return from BINDEC as:
#
# 32 16 15 0
# -----------------------------------------
# | 0 | e3 | e2 | e1 | e4 | X | X | X |
# -----------------------------------------
#
# And are moved into their proper places in FP_SCR0. If digit e4
# is non-zero, OPERR is signaled. In all cases, all 4 digits are
# written as specified in the 881/882 manual for packed decimal.
#
# Register usage:
# Input/Output
# d0: x/LEN call to binstr - final is 0
# d1: x/scratch (0);shift count for final exponent packing
# d2: x/ms 32-bits of exp fraction/scratch
# d3: x/ls 32-bits of exp fraction
# d4: LEN/Unchanged
# d5: ICTR:LAMBDA/LAMBDA:ICTR
# d6: ILOG
# d7: k-factor/Unchanged
# a0: ptr to result string/ptr to L_SCR1(a6)
# a1: ptr to PTENxx array/Unchanged
# a2: ptr to FP_SCR1(a6)/Unchanged
# fp0: abs(YINT) adjusted/float(ILOG)
# fp1: 10^ISCALE/Unchanged
# fp2: 10^LEN/Unchanged
# F_SCR1:Work area for final result/BCD result
# F_SCR2:Y with original exponent/ILOG/10^4
# L_SCR1:original USER_FPCR/Exponent digits on return from binstr
# L_SCR2:first word of X packed/Unchanged
A15_st:
tst.b BINDEC_FLG(%a6) # check for denorm
beq.b not_denorm
ftest.x %fp0 # test for zero
fbeq.w den_zero # if zero, use k-factor or 4933
fmov.l %d6,%fp0 # float ILOG
fabs.x %fp0 # get abs of ILOG
bra.b convrt
den_zero:
tst.l %d7 # check sign of the k-factor
blt.b use_ilog # if negative, use ILOG
fmov.s F4933(%pc),%fp0 # force exponent to 4933
bra.b convrt # do it
use_ilog:
fmov.l %d6,%fp0 # float ILOG
fabs.x %fp0 # get abs of ILOG
bra.b convrt
not_denorm:
ftest.x %fp0 # test for zero
fbneq.w not_zero # if zero, force exponent
fmov.s FONE(%pc),%fp0 # force exponent to 1
bra.b convrt # do it
not_zero:
fmov.l %d6,%fp0 # float ILOG
fabs.x %fp0 # get abs of ILOG
convrt:
fdiv.x 24(%a1),%fp0 # compute ILOG/10^4
fmov.x %fp0,FP_SCR1(%a6) # store fp0 in memory
mov.l 4(%a2),%d2 # move word 2 to d2
mov.l 8(%a2),%d3 # move word 3 to d3
mov.w (%a2),%d0 # move exp to d0
beq.b x_loop_fin # if zero, skip the shift sub.w &0x3ffd,%d0 # subtract off bias
neg.w %d0 # make exp positive
x_loop:
lsr.l &1,%d2 # shift d2:d3 right
roxr.l &1,%d3 # the number of places
dbf.w %d0,x_loop # given in d0
x_loop_fin: clr.l %d1 # put zero in d1 for addx
add.l &0x00000080,%d3 # inc at bit 6
addx.l %d1,%d2 # continue inc
and.l &0xffffff80,%d3 # strip off lsb not used by 882
mov.l &4,%d0 # put 4 in d0 for binstr call
lea.l L_SCR1(%a6),%a0 # a0 is ptr to L_SCR1 for exp digits
bsr binstr # call binstr to convert exp
mov.l L_SCR1(%a6),%d0 # load L_SCR1 lword to d0
mov.l &12,%d1 # use d1 for shift count
lsr.l %d1,%d0 # shift d0 right by 12
bfins %d0,FP_SCR0(%a6){&4:&12} # put e3:e2:e1 in FP_SCR0
lsr.l %d1,%d0 # shift d0 right by 12
bfins %d0,FP_SCR0(%a6){&16:&4} # put e4 in FP_SCR0
tst.b %d0 # check if e4 is zero
beq.b A16_st # if zero, skip rest
or.l &opaop_mask,USER_FPSR(%a6) # set OPERR & AIOP in USER_FPSR
# A16. Write sign bits to final string.
# Sigma is bit 31 of initial value; RHO is bit 31 of d6 (ILOG).
#
# Register usage:
# Input/Output
# d0: x/scratch - final is x
# d2: x/x
# d3: x/x
# d4: LEN/Unchanged
# d5: ICTR:LAMBDA/LAMBDA:ICTR
# d6: ILOG/ILOG adjusted
# d7: k-factor/Unchanged
# a0: ptr to L_SCR1(a6)/Unchanged
# a1: ptr to PTENxx array/Unchanged
# a2: ptr to FP_SCR1(a6)/Unchanged
# fp0: float(ILOG)/Unchanged
# fp1: 10^ISCALE/Unchanged
# fp2: 10^LEN/Unchanged
# F_SCR1:BCD result with correct signs
# F_SCR2:ILOG/10^4
# L_SCR1:Exponent digits on return from binstr
# L_SCR2:first word of X packed/Unchanged
A16_st: clr.l %d0 # clr d0 for collection of signs
and.b &0x0f,FP_SCR0(%a6) # clear first nibble of FP_SCR0
tst.l L_SCR2(%a6) # check sign of original mantissa
bge.b mant_p # if pos, don't set SM
mov.l &2,%d0 # move 2 in to d0 for SM
mant_p:
tst.l %d6 # check sign of ILOG
bge.b wr_sgn # if pos, don't set SE
addq.l &1,%d0 # set bit 0 in d0 for SE
wr_sgn:
bfins %d0,FP_SCR0(%a6){&0:&2} # insert SM and SE into FP_SCR0
global PTENRN
PTENRN:
long 0x40020000,0xA0000000,0x00000000 # 10 ^ 1
long 0x40050000,0xC8000000,0x00000000 # 10 ^ 2
long 0x400C0000,0x9C400000,0x00000000 # 10 ^ 4
long 0x40190000,0xBEBC2000,0x00000000 # 10 ^ 8
long 0x40340000,0x8E1BC9BF,0x04000000 # 10 ^ 16
long 0x40690000,0x9DC5ADA8,0x2B70B59E # 10 ^ 32
long 0x40D30000,0xC2781F49,0xFFCFA6D5 # 10 ^ 64
long 0x41A80000,0x93BA47C9,0x80E98CE0 # 10 ^ 128
long 0x43510000,0xAA7EEBFB,0x9DF9DE8E # 10 ^ 256
long 0x46A30000,0xE319A0AE,0xA60E91C7 # 10 ^ 512
long 0x4D480000,0xC9767586,0x81750C17 # 10 ^ 1024
long 0x5A920000,0x9E8B3B5D,0xC53D5DE5 # 10 ^ 2048
long 0x75250000,0xC4605202,0x8A20979B # 10 ^ 4096
global PTENRP
PTENRP:
long 0x40020000,0xA0000000,0x00000000 # 10 ^ 1
long 0x40050000,0xC8000000,0x00000000 # 10 ^ 2
long 0x400C0000,0x9C400000,0x00000000 # 10 ^ 4
long 0x40190000,0xBEBC2000,0x00000000 # 10 ^ 8
long 0x40340000,0x8E1BC9BF,0x04000000 # 10 ^ 16
long 0x40690000,0x9DC5ADA8,0x2B70B59E # 10 ^ 32
long 0x40D30000,0xC2781F49,0xFFCFA6D6 # 10 ^ 64
long 0x41A80000,0x93BA47C9,0x80E98CE0 # 10 ^ 128
long 0x43510000,0xAA7EEBFB,0x9DF9DE8E # 10 ^ 256
long 0x46A30000,0xE319A0AE,0xA60E91C7 # 10 ^ 512
long 0x4D480000,0xC9767586,0x81750C18 # 10 ^ 1024
long 0x5A920000,0x9E8B3B5D,0xC53D5DE5 # 10 ^ 2048
long 0x75250000,0xC4605202,0x8A20979B # 10 ^ 4096
global PTENRM
PTENRM:
long 0x40020000,0xA0000000,0x00000000 # 10 ^ 1
long 0x40050000,0xC8000000,0x00000000 # 10 ^ 2
long 0x400C0000,0x9C400000,0x00000000 # 10 ^ 4
long 0x40190000,0xBEBC2000,0x00000000 # 10 ^ 8
long 0x40340000,0x8E1BC9BF,0x04000000 # 10 ^ 16
long 0x40690000,0x9DC5ADA8,0x2B70B59D # 10 ^ 32
long 0x40D30000,0xC2781F49,0xFFCFA6D5 # 10 ^ 64
long 0x41A80000,0x93BA47C9,0x80E98CDF # 10 ^ 128
long 0x43510000,0xAA7EEBFB,0x9DF9DE8D # 10 ^ 256
long 0x46A30000,0xE319A0AE,0xA60E91C6 # 10 ^ 512
long 0x4D480000,0xC9767586,0x81750C17 # 10 ^ 1024
long 0x5A920000,0x9E8B3B5D,0xC53D5DE4 # 10 ^ 2048
long 0x75250000,0xC4605202,0x8A20979A # 10 ^ 4096
#########################################################################
# binstr(): Converts a 64-bit binary integer to bcd. #
# #
# INPUT *************************************************************** #
# d2:d3 = 64-bit binary integer #
# d0 = desired length (LEN) #
# a0 = pointer to start in memory for bcd characters #
# (This pointer must point to byte 4 of the first #
# lword of the packed decimal memory string.) #
# #
# OUTPUT ************************************************************** #
# a0 = pointer to LEN bcd digits representing the 64-bit integer. #
# #
# ALGORITHM *********************************************************** #
# The 64-bit binary is assumed to have a decimal point before #
# bit 63. The fraction is multiplied by 10 using a mul by 2 #
# shift and a mul by 8 shift. The bits shifted out of the #
# msb form a decimal digit. This process is iterated until #
# LEN digits are formed. #
# #
# A1. Init d7 to 1. D7 is the byte digit counter, and if 1, the #
# digit formed will be assumed the least significant. This is #
# to force the first byte formed to have a 0 in the upper 4 bits. #
# #
# A2. Beginning of the loop: #
# Copy the fraction in d2:d3 to d4:d5. #
# #
# A3. Multiply the fraction in d2:d3 by 8 using bit-field #
# extracts and shifts. The three msbs from d2 will go into d1. #
# #
# A4. Multiply the fraction in d4:d5 by 2 using shifts. The msb #
# will be collected by the carry. #
# #
# A5. Add using the carry the 64-bit quantities in d2:d3 and d4:d5 #
# into d2:d3. D1 will contain the bcd digit formed. #
# #
# A6. Test d7. If zero, the digit formed is the ms digit. If non- #
# zero, it is the ls digit. Put the digit in its place in the #
# upper word of d0. If it is the ls digit, write the word #
# from d0 to memory. #
# #
# A7. Decrement d6 (LEN counter) and repeat the loop until zero. #
# #
#########################################################################
# Implementation Notes:
#
# The registers are used as follows:
#
# d0: LEN counter
# d1: temp used to form the digit
# d2: upper 32-bits of fraction for mul by 8
# d3: lower 32-bits of fraction for mul by 8
# d4: upper 32-bits of fraction for mul by 2
# d5: lower 32-bits of fraction for mul by 2
# d6: temp for bit-field extracts
# d7: byte digit formation word;digit count {0,1}
# a0: pointer into memory for packed bcd string formation
#
global binstr
binstr:
movm.l &0xff00,-(%sp) # {%d0-%d7}
#
# A1: Init d7
#
mov.l &1,%d7 # init d7 for second digit
subq.l &1,%d0 # for dbf d0 would have LEN+1 passes
#
# A2. Copy d2:d3 to d4:d5. Start loop.
#
loop:
mov.l %d2,%d4 # copy the fraction before muls
mov.l %d3,%d5 # to d4:d5
#
# A3. Multiply d2:d3 by 8; extract msbs into d1.
#
bfextu %d2{&0:&3},%d1 # copy 3 msbs of d2 into d1
asl.l &3,%d2 # shift d2 left by 3 places
bfextu %d3{&0:&3},%d6 # copy 3 msbs of d3 into d6
asl.l &3,%d3 # shift d3 left by 3 places
or.l %d6,%d2 # or in msbs from d3 into d2
#
# A4. Multiply d4:d5 by 2; add carry out to d1.
#
asl.l &1,%d5 # mul d5 by 2
roxl.l &1,%d4 # mul d4 by 2
swap %d6 # put 0 in d6 lower word
addx.w %d6,%d1 # add in extend from mul by 2
#
# A5. Add mul by 8 to mul by 2. D1 contains the digit formed.
#
add.l %d5,%d3 # add lower 32 bits
nop # ERRATA FIX #13 (Rev. 1.2 6/6/90)
addx.l %d4,%d2 # add with extend upper 32 bits
nop # ERRATA FIX #13 (Rev. 1.2 6/6/90)
addx.w %d6,%d1 # add in extend from add to d1
swap %d6 # with d6 = 0; put 0 in upper word
#
# A6. Test d7 and branch.
#
tst.w %d7 # if zero, store digit & to loop
beq.b first_d # if non-zero, form byte & write
sec_d:
swap %d7 # bring first digit to word d7b
asl.w &4,%d7 # first digit in upper 4 bits d7b
add.w %d1,%d7 # add in ls digit to d7b
mov.b %d7,(%a0)+ # store d7b byte in memory
swap %d7 # put LEN counter in word d7a clr.w %d7 # set d7a to signal no digits done
dbf.w %d0,loop # do loop some more!
bra.b end_bstr # finished, so exit
first_d:
swap %d7 # put digit word in d7b
mov.w %d1,%d7 # put new digit in d7b
swap %d7 # put LEN counter in word d7a
addq.w &1,%d7 # set d7a to signal first digit done
dbf.w %d0,loop # do loop some more!
swap %d7 # put last digit in string
lsl.w &4,%d7 # move it to upper 4 bits
mov.b %d7,(%a0)+ # store it in memory string
#
# Clean up and return with result in fp0.
#
end_bstr:
movm.l (%sp)+,&0xff # {%d0-%d7}
rts
#########################################################################
# XDEF **************************************************************** #
# facc_in_b(): dmem_read_byte failed #
# facc_in_w(): dmem_read_word failed #
# facc_in_l(): dmem_read_long failed #
# facc_in_d(): dmem_read of dbl prec failed #
# facc_in_x(): dmem_read of ext prec failed #
# #
# facc_out_b(): dmem_write_byte failed #
# facc_out_w(): dmem_write_word failed #
# facc_out_l(): dmem_write_long failed #
# facc_out_d(): dmem_write of dbl prec failed #
# facc_out_x(): dmem_write of ext prec failed #
# #
# XREF **************************************************************** #
# _real_access() - exit through access error handler #
# #
# INPUT *************************************************************** #
# None #
# #
# OUTPUT ************************************************************** #
# None #
# #
# ALGORITHM *********************************************************** #
# Flow jumps here when an FP data fetch call gets an error #
# result. This means the operating system wants an access error frame #
# made out of the current exception stack frame. #
# So, we first call restore() which makes sure that any updated #
# -(an)+ register gets returned to its pre-exception value and then #
# we change the stack to an access error stack frame. #
# #
#########################################################################
facc_in_b:
movq.l &0x1,%d0 # one byte
bsr.w restore # fix An
mov.w &0x0121,EXC_VOFF(%a6) # set FSLW
bra.w facc_finish
facc_in_w:
movq.l &0x2,%d0 # two bytes
bsr.w restore # fix An
mov.w &0x0141,EXC_VOFF(%a6) # set FSLW
bra.b facc_finish
facc_in_l:
movq.l &0x4,%d0 # four bytes
bsr.w restore # fix An
mov.w &0x0101,EXC_VOFF(%a6) # set FSLW
bra.b facc_finish
# here's where we actually create the access error frame from the
# current exception stack frame.
facc_finish:
mov.l USER_FPIAR(%a6),EXC_PC(%a6) # store current PC
# if the effective addressing mode was predecrement or postincrement,
# the emulation has already changed its value to the correct post-
# instruction value. but since we're exiting to the access error
# handler, then AN must be returned to its pre-instruction value.
# we do that here.
restore:
mov.b EXC_OPWORD+0x1(%a6),%d1
andi.b &0x38,%d1 # extract opmode
cmpi.b %d1,&0x18 # postinc?
beq.w rest_inc
cmpi.b %d1,&0x20 # predec?
beq.w rest_dec
rts
rest_inc:
mov.b EXC_OPWORD+0x1(%a6),%d1
andi.w &0x0007,%d1 # fetch An
tbl_rest_inc:
short ri_a0 - tbl_rest_inc
short ri_a1 - tbl_rest_inc
short ri_a2 - tbl_rest_inc
short ri_a3 - tbl_rest_inc
short ri_a4 - tbl_rest_inc
short ri_a5 - tbl_rest_inc
short ri_a6 - tbl_rest_inc
short ri_a7 - tbl_rest_inc
ri_a0: sub.l %d0,EXC_DREGS+0x8(%a6) # fix stacked a0
rts
ri_a1: sub.l %d0,EXC_DREGS+0xc(%a6) # fix stacked a1
rts
ri_a2: sub.l %d0,%a2 # fix a2
rts
ri_a3: sub.l %d0,%a3 # fix a3
rts
ri_a4: sub.l %d0,%a4 # fix a4
rts
ri_a5: sub.l %d0,%a5 # fix a5
rts
ri_a6: sub.l %d0,(%a6) # fix stacked a6
rts
# if it's a fmove out instruction, we don't have to fix a7
# because we hadn't changed it yet. if it's an opclass two
# instruction (data moved in) and the exception was in supervisor
# mode, then also also wasn't updated. if it was user mode, then
# restore the correct a7 which is in the USP currently.
ri_a7:
cmpi.b EXC_VOFF(%a6),&0x30 # move in or out?
bne.b ri_a7_done # out
# need to invert adjustment value if the <ea> was predec
rest_dec:
neg.l %d0
bra.b rest_inc
Messung V0.5 in Prozent
¤ Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.0.547Bemerkung:
(vorverarbeitet am 2026-04-29)
¤
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.