1994-10-26 10:48:18 +03:00
|
|
|
* $NetBSD: x_unfl.sa,v 1.3 1994/10/26 07:50:30 cgd Exp $
|
|
|
|
|
1994-07-05 21:50:24 +04:00
|
|
|
* MOTOROLA MICROPROCESSOR & MEMORY TECHNOLOGY GROUP
|
|
|
|
* M68000 Hi-Performance Microprocessor Division
|
|
|
|
* M68040 Software Package
|
|
|
|
*
|
|
|
|
* M68040 Software Package Copyright (c) 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.
|
|
|
|
|
|
|
|
*
|
|
|
|
* x_unfl.sa 3.4 7/1/91
|
|
|
|
*
|
|
|
|
* fpsp_unfl --- FPSP handler for underflow exception
|
|
|
|
*
|
|
|
|
* Trap disabled results
|
|
|
|
* For 881/2 compatibility, sw must denormalize the intermediate
|
|
|
|
* result, then store the result. Denormalization is accomplished
|
|
|
|
* by taking the intermediate result (which is always normalized) and
|
|
|
|
* shifting the mantissa right while incrementing the exponent until
|
|
|
|
* it is equal to the denormalized exponent for the destination
|
|
|
|
* format. After denormalizatoin, the result is rounded to the
|
|
|
|
* destination format.
|
|
|
|
*
|
|
|
|
* Trap enabled results
|
|
|
|
* All trap disabled code applies. In addition the exceptional
|
|
|
|
* operand needs to made available to the user with a bias of $6000
|
|
|
|
* added to the exponent.
|
|
|
|
*
|
|
|
|
|
|
|
|
X_UNFL IDNT 2,1 Motorola 040 Floating Point Software Package
|
|
|
|
|
|
|
|
section 8
|
|
|
|
|
|
|
|
include fpsp.h
|
|
|
|
|
|
|
|
xref denorm
|
|
|
|
xref round
|
|
|
|
xref store
|
|
|
|
xref g_rndpr
|
|
|
|
xref g_opcls
|
|
|
|
xref g_dfmtou
|
|
|
|
xref real_unfl
|
|
|
|
xref real_inex
|
|
|
|
xref fpsp_done
|
|
|
|
xref b1238_fix
|
|
|
|
|
|
|
|
xdef fpsp_unfl
|
|
|
|
fpsp_unfl:
|
|
|
|
link a6,#-LOCAL_SIZE
|
|
|
|
fsave -(a7)
|
|
|
|
movem.l d0-d1/a0-a1,USER_DA(a6)
|
|
|
|
fmovem.x fp0-fp3,USER_FP0(a6)
|
|
|
|
fmovem.l fpcr/fpsr/fpiar,USER_FPCR(a6)
|
|
|
|
|
|
|
|
*
|
|
|
|
bsr.l unf_res ;denormalize, round & store interm op
|
|
|
|
*
|
|
|
|
* If underflow exceptions are not enabled, check for inexact
|
|
|
|
* exception
|
|
|
|
*
|
|
|
|
btst.b #unfl_bit,FPCR_ENABLE(a6)
|
|
|
|
beq.b ck_inex
|
|
|
|
|
|
|
|
btst.b #E3,E_BYTE(a6)
|
|
|
|
beq.b no_e3_1
|
|
|
|
*
|
|
|
|
* Clear dirty bit on dest resister in the frame before branching
|
|
|
|
* to b1238_fix.
|
|
|
|
*
|
|
|
|
bfextu CMDREG3B(a6){6:3},d0 ;get dest reg no
|
|
|
|
bclr.b d0,FPR_DIRTY_BITS(a6) ;clr dest dirty bit
|
|
|
|
bsr.l b1238_fix ;test for bug1238 case
|
|
|
|
move.l USER_FPSR(a6),FPSR_SHADOW(a6)
|
|
|
|
or.l #sx_mask,E_BYTE(a6)
|
|
|
|
no_e3_1:
|
|
|
|
movem.l USER_DA(a6),d0-d1/a0-a1
|
|
|
|
fmovem.x USER_FP0(a6),fp0-fp3
|
|
|
|
fmovem.l USER_FPCR(a6),fpcr/fpsr/fpiar
|
|
|
|
frestore (a7)+
|
|
|
|
unlk a6
|
|
|
|
bra.l real_unfl
|
|
|
|
*
|
|
|
|
* It is possible to have either inex2 or inex1 exceptions with the
|
|
|
|
* unfl. If the inex enable bit is set in the FPCR, and either
|
|
|
|
* inex2 or inex1 occured, we must clean up and branch to the
|
|
|
|
* real inex handler.
|
|
|
|
*
|
|
|
|
ck_inex:
|
|
|
|
move.b FPCR_ENABLE(a6),d0
|
|
|
|
and.b FPSR_EXCEPT(a6),d0
|
|
|
|
andi.b #$3,d0
|
|
|
|
beq.b unfl_done
|
|
|
|
|
|
|
|
*
|
|
|
|
* Inexact enabled and reported, and we must take an inexact exception
|
|
|
|
*
|
|
|
|
take_inex:
|
|
|
|
btst.b #E3,E_BYTE(a6)
|
|
|
|
beq.b no_e3_2
|
|
|
|
*
|
|
|
|
* Clear dirty bit on dest resister in the frame before branching
|
|
|
|
* to b1238_fix.
|
|
|
|
*
|
|
|
|
bfextu CMDREG3B(a6){6:3},d0 ;get dest reg no
|
|
|
|
bclr.b d0,FPR_DIRTY_BITS(a6) ;clr dest dirty bit
|
|
|
|
bsr.l b1238_fix ;test for bug1238 case
|
|
|
|
move.l USER_FPSR(a6),FPSR_SHADOW(a6)
|
|
|
|
or.l #sx_mask,E_BYTE(a6)
|
|
|
|
no_e3_2:
|
|
|
|
move.b #INEX_VEC,EXC_VEC+1(a6)
|
|
|
|
movem.l USER_DA(a6),d0-d1/a0-a1
|
|
|
|
fmovem.x USER_FP0(a6),fp0-fp3
|
|
|
|
fmovem.l USER_FPCR(a6),fpcr/fpsr/fpiar
|
|
|
|
frestore (a7)+
|
|
|
|
unlk a6
|
|
|
|
bra.l real_inex
|
|
|
|
|
|
|
|
unfl_done:
|
|
|
|
bclr.b #E3,E_BYTE(a6)
|
|
|
|
beq.b e1_set ;if set then branch
|
|
|
|
*
|
|
|
|
* Clear dirty bit on dest resister in the frame before branching
|
|
|
|
* to b1238_fix.
|
|
|
|
*
|
|
|
|
bfextu CMDREG3B(a6){6:3},d0 ;get dest reg no
|
|
|
|
bclr.b d0,FPR_DIRTY_BITS(a6) ;clr dest dirty bit
|
|
|
|
bsr.l b1238_fix ;test for bug1238 case
|
|
|
|
move.l USER_FPSR(a6),FPSR_SHADOW(a6)
|
|
|
|
or.l #sx_mask,E_BYTE(a6)
|
|
|
|
movem.l USER_DA(a6),d0-d1/a0-a1
|
|
|
|
fmovem.x USER_FP0(a6),fp0-fp3
|
|
|
|
fmovem.l USER_FPCR(a6),fpcr/fpsr/fpiar
|
|
|
|
frestore (a7)+
|
|
|
|
unlk a6
|
|
|
|
bra.l fpsp_done
|
|
|
|
e1_set:
|
|
|
|
movem.l USER_DA(a6),d0-d1/a0-a1
|
|
|
|
fmovem.x USER_FP0(a6),fp0-fp3
|
|
|
|
fmovem.l USER_FPCR(a6),fpcr/fpsr/fpiar
|
|
|
|
unlk a6
|
|
|
|
bra.l fpsp_done
|
|
|
|
*
|
|
|
|
* unf_res --- underflow result calculation
|
|
|
|
*
|
|
|
|
unf_res:
|
|
|
|
bsr.l g_rndpr ;returns RND_PREC in d0 0=ext,
|
|
|
|
* ;1=sgl, 2=dbl
|
|
|
|
* ;we need the RND_PREC in the
|
|
|
|
* ;upper word for round
|
1994-07-05 21:56:52 +04:00
|
|
|
clr.w -(a7)
|
1994-07-05 21:50:24 +04:00
|
|
|
move.w d0,-(a7) ;copy RND_PREC to stack
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* If the exception bit set is E3, the exceptional operand from the
|
|
|
|
* fpu is in WBTEMP; else it is in FPTEMP.
|
|
|
|
*
|
|
|
|
btst.b #E3,E_BYTE(a6)
|
|
|
|
beq.b unf_E1
|
|
|
|
unf_E3:
|
|
|
|
lea WBTEMP(a6),a0 ;a0 now points to operand
|
|
|
|
*
|
|
|
|
* Test for fsgldiv and fsglmul. If the inst was one of these, then
|
|
|
|
* force the precision to extended for the denorm routine. Use
|
|
|
|
* the user's precision for the round routine.
|
|
|
|
*
|
|
|
|
move.w CMDREG3B(a6),d1 ;check for fsgldiv or fsglmul
|
|
|
|
andi.w #$7f,d1
|
|
|
|
cmpi.w #$30,d1 ;check for sgldiv
|
|
|
|
beq.b unf_sgl
|
|
|
|
cmpi.w #$33,d1 ;check for sglmul
|
|
|
|
bne.b unf_cont ;if not, use fpcr prec in round
|
|
|
|
unf_sgl:
|
|
|
|
clr.l d0
|
|
|
|
move.w #$1,(a7) ;override g_rndpr precision
|
|
|
|
* ;force single
|
|
|
|
bra.b unf_cont
|
|
|
|
unf_E1:
|
|
|
|
lea FPTEMP(a6),a0 ;a0 now points to operand
|
|
|
|
unf_cont:
|
|
|
|
bclr.b #sign_bit,LOCAL_EX(a0) ;clear sign bit
|
|
|
|
sne LOCAL_SGN(a0) ;store sign
|
|
|
|
|
|
|
|
bsr.l denorm ;returns denorm, a0 points to it
|
|
|
|
*
|
|
|
|
* WARNING:
|
|
|
|
* ;d0 has guard,round sticky bit
|
|
|
|
* ;make sure that it is not corrupted
|
|
|
|
* ;before it reaches the round subroutine
|
|
|
|
* ;also ensure that a0 isn't corrupted
|
|
|
|
|
|
|
|
*
|
|
|
|
* Set up d1 for round subroutine d1 contains the PREC/MODE
|
|
|
|
* information respectively on upper/lower register halves.
|
|
|
|
*
|
|
|
|
bfextu FPCR_MODE(a6){2:2},d1 ;get mode from FPCR
|
|
|
|
* ;mode in lower d1
|
|
|
|
add.l (a7)+,d1 ;merge PREC/MODE
|
|
|
|
*
|
|
|
|
* WARNING: a0 and d0 are assumed to be intact between the denorm and
|
|
|
|
* round subroutines. All code between these two subroutines
|
|
|
|
* must not corrupt a0 and d0.
|
|
|
|
*
|
|
|
|
*
|
|
|
|
* Perform Round
|
|
|
|
* Input: a0 points to input operand
|
|
|
|
* d0{31:29} has guard, round, sticky
|
|
|
|
* d1{01:00} has rounding mode
|
|
|
|
* d1{17:16} has rounding precision
|
|
|
|
* Output: a0 points to rounded operand
|
|
|
|
*
|
|
|
|
|
|
|
|
bsr.l round ;returns rounded denorm at (a0)
|
|
|
|
*
|
|
|
|
* Differentiate between store to memory vs. store to register
|
|
|
|
*
|
|
|
|
unf_store:
|
|
|
|
bsr.l g_opcls ;returns opclass in d0{2:0}
|
|
|
|
cmpi.b #$3,d0
|
|
|
|
bne.b not_opc011
|
|
|
|
*
|
|
|
|
* At this point, a store to memory is pending
|
|
|
|
*
|
|
|
|
opc011:
|
|
|
|
bsr.l g_dfmtou
|
|
|
|
tst.b d0
|
|
|
|
beq.b ext_opc011 ;If extended, do not subtract
|
|
|
|
* ;If destination format is sgl/dbl,
|
|
|
|
tst.b LOCAL_HI(a0) ;If rounded result is normal,don't
|
|
|
|
* ;subtract
|
|
|
|
bmi.b ext_opc011
|
|
|
|
subq.w #1,LOCAL_EX(a0) ;account for denorm bias vs.
|
|
|
|
* ;normalized bias
|
|
|
|
* ; normalized denormalized
|
|
|
|
* ;single $7f $7e
|
|
|
|
* ;double $3ff $3fe
|
|
|
|
*
|
|
|
|
ext_opc011:
|
|
|
|
bsr.l store ;stores to memory
|
|
|
|
bra.b unf_done ;finish up
|
|
|
|
|
|
|
|
*
|
|
|
|
* At this point, a store to a float register is pending
|
|
|
|
*
|
|
|
|
not_opc011:
|
|
|
|
bsr.l store ;stores to float register
|
|
|
|
* ;a0 is not corrupted on a store to a
|
|
|
|
* ;float register.
|
|
|
|
*
|
|
|
|
* Set the condition codes according to result
|
|
|
|
*
|
|
|
|
tst.l LOCAL_HI(a0) ;check upper mantissa
|
|
|
|
bne.b ck_sgn
|
|
|
|
tst.l LOCAL_LO(a0) ;check lower mantissa
|
|
|
|
bne.b ck_sgn
|
|
|
|
bset.b #z_bit,FPSR_CC(a6) ;set condition codes if zero
|
|
|
|
ck_sgn:
|
|
|
|
btst.b #sign_bit,LOCAL_EX(a0) ;check the sign bit
|
|
|
|
beq.b unf_done
|
|
|
|
bset.b #neg_bit,FPSR_CC(a6)
|
|
|
|
|
|
|
|
*
|
|
|
|
* Finish.
|
|
|
|
*
|
|
|
|
unf_done:
|
|
|
|
btst.b #inex2_bit,FPSR_EXCEPT(a6)
|
|
|
|
beq.b no_aunfl
|
|
|
|
bset.b #aunfl_bit,FPSR_AEXCEPT(a6)
|
|
|
|
no_aunfl:
|
|
|
|
rts
|
|
|
|
|
|
|
|
end
|