diff options
Diffstat (limited to 'arch/m68k/ifpsp060/src/fpsp.S')
-rw-r--r-- | arch/m68k/ifpsp060/src/fpsp.S | 24785 |
1 files changed, 24785 insertions, 0 deletions
diff --git a/arch/m68k/ifpsp060/src/fpsp.S b/arch/m68k/ifpsp060/src/fpsp.S new file mode 100644 index 000000000000..3b597a9bbf43 --- /dev/null +++ b/arch/m68k/ifpsp060/src/fpsp.S @@ -0,0 +1,24785 @@ +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +MOTOROLA MICROPROCESSOR & MEMORY TECHNOLOGY GROUP +M68000 Hi-Performance Microprocessor Division +M68060 Software Package +Production Release P1.00 -- October 10, 1994 + +M68060 Software Package Copyright © 1993, 1994 Motorola Inc. All rights reserved. + +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 + +_060FPSP_TABLE: + +############################################################### + +# 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 + + space 56 + +############################################################### + global _fpsp_done +_fpsp_done: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_done,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_ovfl +_real_ovfl: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_ovfl,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_unfl +_real_unfl: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_unfl,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_inex +_real_inex: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_inex,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_bsun +_real_bsun: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_bsun,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_operr +_real_operr: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_operr,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_snan +_real_snan: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_snan,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_dz +_real_dz: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_dz,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_fline +_real_fline: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_fline,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_fpu_disabled +_real_fpu_disabled: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_fpu_dis,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_trap +_real_trap: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_trap,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_trace +_real_trace: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_trace,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _real_access +_real_access: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_access,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + +####################################### + + global _imem_read +_imem_read: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_imr,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _dmem_read +_dmem_read: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_dmr,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _dmem_write +_dmem_write: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_dmw,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _imem_read_word +_imem_read_word: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_irw,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _imem_read_long +_imem_read_long: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_irl,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _dmem_read_byte +_dmem_read_byte: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_drb,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _dmem_read_word +_dmem_read_word: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_drw,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _dmem_read_long +_dmem_read_long: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_drl,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _dmem_write_byte +_dmem_write_byte: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_dwb,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _dmem_write_word +_dmem_write_word: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_dww,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + + global _dmem_write_long +_dmem_write_long: + mov.l %d0,-(%sp) + mov.l (_060FPSP_TABLE-0x80+_off_dwl,%pc),%d0 + pea.l (_060FPSP_TABLE-0x80,%pc,%d0) + mov.l 0x4(%sp),%d0 + rtd &0x4 + +# +# 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 mantissalen, 64 # length of mantissa in bits + +set BYTE, 1 # len(byte) == 1 byte +set WORD, 2 # len(word) == 2 bytes +set LONG, 4 # len(longword) == 2 bytes + +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: + +#$# sub.l &24,%sp # make room for src/dst + + link.w %a6,&-LOCAL_SIZE # init stack frame + + fsave FP_SRC(%a6) # grab the "busy" frame + + movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1 + fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs + fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack + +# the FPIAR holds the "current PC" of the faulting instruction + mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6) + mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr + addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr + bsr.l _imem_read_long # fetch the instruction words + mov.l %d0,EXC_OPWORD(%a6) + +############################################################################## + + btst &0x5,EXC_CMDREG(%a6) # is instr an fmove out? + bne.w fovfl_out + + + lea FP_SRC(%a6),%a0 # pass: ptr to src op + bsr.l fix_skewed_ops # fix src op + +# since, I believe, only NORMs and DENORMs can come through here, +# maybe we can avoid the subroutine call. + lea FP_SRC(%a6),%a0 # pass: ptr to src op + bsr.l set_tag_x # tag the operand type + mov.b %d0,STAG(%a6) # maybe NORM,DENORM + +# bit five of the fp extension word separates the monadic and dyadic operations +# that can pass through fpsp_ovfl(). remember that fcmp, ftst, and fsincos +# will never take this exception. + btst &0x5,1+EXC_CMDREG(%a6) # is operation monadic or dyadic? + beq.b fovfl_extract # monadic + + bfextu EXC_CMDREG(%a6){&6:&3},%d0 # dyadic; load dst reg + bsr.l load_fpn2 # load dst into FP_DST + + lea FP_DST(%a6),%a0 # pass: ptr to dst op + bsr.l set_tag_x # tag the operand type + cmpi.b %d0,&UNNORM # is operand an UNNORM? + bne.b fovfl_op2_done # no + bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO +fovfl_op2_done: + mov.b %d0,DTAG(%a6) # save dst optype tag + +fovfl_extract: + +#$# mov.l FP_SRC_EX(%a6),TRAP_SRCOP_EX(%a6) +#$# mov.l FP_SRC_HI(%a6),TRAP_SRCOP_HI(%a6) +#$# mov.l FP_SRC_LO(%a6),TRAP_SRCOP_LO(%a6) +#$# mov.l FP_DST_EX(%a6),TRAP_DSTOP_EX(%a6) +#$# mov.l FP_DST_HI(%a6),TRAP_DSTOP_HI(%a6) +#$# mov.l FP_DST_LO(%a6),TRAP_DSTOP_LO(%a6) + + clr.l %d0 + mov.b FPCR_MODE(%a6),%d0 # pass rnd prec/mode + + mov.b 1+EXC_CMDREG(%a6),%d1 + andi.w &0x007f,%d1 # extract extension + + andi.l &0x00ff01ff,USER_FPSR(%a6) # zero all but accured field + + fmov.l &0x0,%fpcr # zero current control regs + fmov.l &0x0,%fpsr + + lea FP_SRC(%a6),%a0 + lea FP_DST(%a6),%a1 + +# maybe we can make these entry points ONLY the OVFL entry points of each routine. + mov.l (tbl_unsupp.l,%pc,%d1.w*4),%d1 # fetch routine addr + jsr (tbl_unsupp.l,%pc,%d1.l*1) + +# the operation has been emulated. the result is in fp0. +# the EXOP, if an exception occurred, is in fp1. +# we must save the default result regardless of whether +# traps are enabled or disabled. + bfextu EXC_CMDREG(%a6){&6:&3},%d0 + bsr.l store_fpreg + +# the exceptional possibilities we have left ourselves with are ONLY overflow +# and inexact. and, the inexact is such that overflow occurred and was disabled +# but inexact was enabled. + btst &ovfl_bit,FPCR_ENABLE(%a6) + bne.b fovfl_ovfl_on + + btst &inex2_bit,FPCR_ENABLE(%a6) + bne.b fovfl_inex_on + + fmovm.x EXC_FPREGS(%a6),&0xc0 # restore fp0-fp1 + fmovm.l USER_FPCR(%a6),%fpcr,%fpsr,%fpiar # restore ctrl regs + movm.l EXC_DREGS(%a6),&0x0303 # restore d0-d1/a0-a1 + + unlk %a6 +#$# add.l &24,%sp + bra.l _fpsp_done + +# 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 + + mov.w &0xe005,2+FP_SRC(%a6) # save exc status + + 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 + + frestore FP_SRC(%a6) # do this after fmovm,other f<op>s! + + unlk %a6 + + bra.l _real_ovfl + +# overflow occurred but is disabled. meanwhile, inexact is enabled. therefore, +# we must jump to real_inex(). +fovfl_inex_on: + + fmovm.x &0x40,FP_SRC(%a6) # save EXOP (fp1) to stack + + mov.b &0xc4,1+EXC_VOFF(%a6) # vector offset = 0xc4 + mov.w &0xe001,2+FP_SRC(%a6) # save exc status + + 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 + + frestore FP_SRC(%a6) # do this after fmovm,other f<op>s! + + unlk %a6 + + bra.l _real_inex + +######################################################################## +fovfl_out: + + +#$# mov.l FP_SRC_EX(%a6),TRAP_SRCOP_EX(%a6) +#$# mov.l FP_SRC_HI(%a6),TRAP_SRCOP_HI(%a6) +#$# mov.l FP_SRC_LO(%a6),TRAP_SRCOP_LO(%a6) + +# the src operand is definitely a NORM(!), so tag it as such + mov.b &NORM,STAG(%a6) # set src optype tag + + clr.l %d0 + mov.b FPCR_MODE(%a6),%d0 # pass rnd prec/mode + + and.l &0xffff00ff,USER_FPSR(%a6) # zero all but accured field + + fmov.l &0x0,%fpcr # zero current control regs + fmov.l &0x0,%fpsr + + lea FP_SRC(%a6),%a0 # pass ptr to src operand + + bsr.l fout + + btst &ovfl_bit,FPCR_ENABLE(%a6) + bne.w fovfl_ovfl_on + + btst &inex2_bit,FPCR_ENABLE(%a6) + bne.w fovfl_inex_on + + fmovm.x EXC_FPREGS(%a6),&0xc0 # restore fp0-fp1 + fmovm.l USER_FPCR(%a6),%fpcr,%fpsr,%fpiar # restore ctrl regs + movm.l EXC_DREGS(%a6),&0x0303 # restore d0-d1/a0-a1 + + unlk %a6 +#$# add.l &24,%sp + + 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: + +#$# sub.l &24,%sp # make room for src/dst + + link.w %a6,&-LOCAL_SIZE # init stack frame + + fsave FP_SRC(%a6) # grab the "busy" frame + + movm.l &0x0303,EXC_DREGS(%a6) # save d0-d1/a0-a1 + fmovm.l %fpcr,%fpsr,%fpiar,USER_FPCR(%a6) # save ctrl regs + fmovm.x &0xc0,EXC_FPREGS(%a6) # save fp0-fp1 on stack + +# the FPIAR holds the "current PC" of the faulting instruction + mov.l USER_FPIAR(%a6),EXC_EXTWPTR(%a6) + mov.l EXC_EXTWPTR(%a6),%a0 # fetch instruction addr + addq.l &0x4,EXC_EXTWPTR(%a6) # incr instruction ptr + bsr.l _imem_read_long # fetch the instruction words + mov.l %d0,EXC_OPWORD(%a6) + +############################################################################## + + btst &0x5,EXC_CMDREG(%a6) # is instr an fmove out? + bne.w funfl_out + + + lea FP_SRC(%a6),%a0 # pass: ptr to src op + bsr.l fix_skewed_ops # fix src op + + lea FP_SRC(%a6),%a0 # pass: ptr to src op + bsr.l set_tag_x # tag the operand type + mov.b %d0,STAG(%a6) # maybe NORM,DENORM + +# bit five of the fp ext word separates the monadic and dyadic operations +# that can pass through fpsp_unfl(). remember that fcmp, and ftst +# will never take this exception. + btst &0x5,1+EXC_CMDREG(%a6) # is op monadic or dyadic? + beq.b funfl_extract # monadic + +# now, what's left that's not dyadic is fsincos. we can distinguish it +# from all dyadics by the '0110xxx pattern + btst &0x4,1+EXC_CMDREG(%a6) # is op an fsincos? + bne.b funfl_extract # yes + + bfextu EXC_CMDREG(%a6){&6:&3},%d0 # dyadic; load dst reg + bsr.l load_fpn2 # load dst into FP_DST + + lea FP_DST(%a6),%a0 # pass: ptr to dst op + bsr.l set_tag_x # tag the operand type + cmpi.b %d0,&UNNORM # is operand an UNNORM? + bne.b funfl_op2_done # no + bsr.l unnorm_fix # yes; convert to NORM,DENORM,or ZERO +funfl_op2_done: + mov.b %d0,DTAG(%a6) # save dst optype tag + +funfl_extract: + +#$# mov.l FP_SRC_EX(%a6),TRAP_SRCOP_EX(%a6) +#$# mov.l FP_SRC_HI(%a6),TRAP_SRCOP_HI(%a6) +#$# mov.l FP_SRC_LO(%a6),TRAP_SRCOP_LO(%a6) +#$# mov.l FP_DST_EX(%a6),TRAP_DSTOP_EX(%a6) +#$# mov.l FP_DST_HI(%a6),TRAP_DSTOP_HI(%a6) +#$# mov.l FP_DST_LO(%a6),TRAP_DSTOP_LO(%a6) + + clr.l %d0 + mov.b FPCR_MODE(%a6),%d0 # pass rnd prec/mode + + mov.b 1+EXC_CMDREG(%a6),%d1 + andi.w &0x007f,%d1 # extract extension + + andi.l &0x00ff01ff,USER_FPSR(%a6) + + fmov.l &0x0,%fpcr # zero current control regs + fmov.l &0x0,%fpsr + + lea FP_SRC(%a6),%a0 + lea FP_DST(%a6),%a1 + +# maybe we can make these entry points ONLY the OVFL entry points of each routine. + mov.l (tbl_unsupp.l,%pc,%d1.w*4),%d1 # fetch routine addr + jsr (tbl_unsupp.l,%pc,%d1.l*1) + + bfextu EXC_CMDREG(%a6){&6:&3},%d0 + bsr.l store_fpreg + +# The `060 FPU multiplier hardware is such that if the result of a +# multiply operation is the smallest possible normalized number +# (0x00000000_80000000_00000000), then the machine will take an +# underflow exception. Since this is incorrect, we need to check +# if our emulation, after re-doing the operation, decided that +# no underflow was called for. We do these checks only in +# funfl_{unfl,inex}_on() because w/ both exceptions disabled, this +# special case will simply exit gracefully with the correct result. + +# the exceptional possibilities we have left ourselves with are ONLY overflow +# and inexact. and, the inexact is such that overflow occurred and was disabled +# but inexact was enabled. + btst &unfl_bit,FPCR_ENABLE(%a6) + bne.b funfl_unfl_on + +funfl_chkinex: + btst &inex2_bit,FPCR_ENABLE(%a6) + bne.b funfl_inex_on + +funfl_exit: + fmovm.x EXC_FPREGS(%a6),&0xc0 # restore fp0-fp1 + fmovm.l USER_FPCR(%a6),%fpcr,%fpsr,%fpiar # restore ctrl regs + movm.l EXC_DREGS(%a6),&0x0303 # restore d0-d1/a0-a1 + + unlk %a6 +#$# add.l &24,%sp + bra.l _fpsp_done + +# 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 + + mov.w &0xe003,2+FP_SRC(%a6) # save exc status + + 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 + + frestore FP_SRC(%a6) # do this after fmovm,other f<op>s! + + unlk %a6 + + bra.l _real_unfl + +# undeflow 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 + + 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 + + frestore FP_SRC(%a6) # do this after fmovm,other f<op>s! + + unlk %a6 + + bra.l _real_inex + +####################################################################### +funfl_out: + + +#$# mov.l FP_SRC_EX(%a6),TRAP_SRCOP_EX(%a6) +#$# mov.l FP_SRC_HI(%a6),TRAP_SRCOP_HI(%a6) +#$# mov.l FP_SRC_LO(%a6),TRAP_SRCOP_LO(%a6) + +# the src operand is definitely a NORM(!), so tag it as such + mov.b &NORM,STAG(%a6) # set src optype tag + + clr.l %d0 + mov.b FPCR_MODE(%a6),%d0 # pass rnd prec/mode + + and.l &0xffff00ff,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 # pass ptr to src operand + + bsr.l fout + + btst &unfl_bit,FPCR_ENABLE(%a6) + bne.w funfl_unfl_on2 + + btst &inex2_bit,FPCR_ENABLE(%a6) |