mirror of
https://github.com/AuxXxilium/linux_dsm_epyc7002.git
synced 2024-12-15 23:26:39 +07:00
e00d82d07f
Acked-by: Alan Cox <alan@redhat.com> Signed-off-by: Matt Waddel <Matt.Waddel@freescale.com> Cc: Roman Zippel <zippel@linux-m68k.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
371 lines
8.7 KiB
ArmAsm
371 lines
8.7 KiB
ArmAsm
|
|
|
| scale.sa 3.3 7/30/91
|
|
|
|
|
| The entry point sSCALE computes the destination operand
|
|
| scaled by the source operand. If the absolute value of
|
|
| the source operand is (>= 2^14) an overflow or underflow
|
|
| is returned.
|
|
|
|
|
| The entry point sscale is called from do_func to emulate
|
|
| the fscale unimplemented instruction.
|
|
|
|
|
| Input: Double-extended destination operand in FPTEMP,
|
|
| double-extended source operand in ETEMP.
|
|
|
|
|
| Output: The function returns scale(X,Y) to fp0.
|
|
|
|
|
| Modifies: fp0.
|
|
|
|
|
| Algorithm:
|
|
|
|
|
| Copyright (C) Motorola, Inc. 1990
|
|
| All Rights Reserved
|
|
|
|
|
| For details on the license for this file, please see the
|
|
| file, README, in this same directory.
|
|
|
|
|SCALE idnt 2,1 | Motorola 040 Floating Point Software Package
|
|
|
|
|section 8
|
|
|
|
#include "fpsp.h"
|
|
|
|
|xref t_ovfl2
|
|
|xref t_unfl
|
|
|xref round
|
|
|xref t_resdnrm
|
|
|
|
SRC_BNDS: .short 0x3fff,0x400c
|
|
|
|
|
|
|
| This entry point is used by the unimplemented instruction exception
|
|
| handler.
|
|
|
|
|
|
|
|
|
|
|
| FSCALE
|
|
|
|
|
.global sscale
|
|
sscale:
|
|
fmovel #0,%fpcr |clr user enabled exc
|
|
clrl %d1
|
|
movew FPTEMP(%a6),%d1 |get dest exponent
|
|
smi L_SCR1(%a6) |use L_SCR1 to hold sign
|
|
andil #0x7fff,%d1 |strip sign
|
|
movew ETEMP(%a6),%d0 |check src bounds
|
|
andiw #0x7fff,%d0 |clr sign bit
|
|
cmp2w SRC_BNDS,%d0
|
|
bccs src_in
|
|
cmpiw #0x400c,%d0 |test for too large
|
|
bge src_out
|
|
|
|
|
| The source input is below 1, so we check for denormalized numbers
|
|
| and set unfl.
|
|
|
|
|
src_small:
|
|
moveb DTAG(%a6),%d0
|
|
andib #0xe0,%d0
|
|
tstb %d0
|
|
beqs no_denorm
|
|
st STORE_FLG(%a6) |dest already contains result
|
|
orl #unfl_mask,USER_FPSR(%a6) |set UNFL
|
|
den_done:
|
|
leal FPTEMP(%a6),%a0
|
|
bra t_resdnrm
|
|
no_denorm:
|
|
fmovel USER_FPCR(%a6),%FPCR
|
|
fmovex FPTEMP(%a6),%fp0 |simply return dest
|
|
rts
|
|
|
|
|
|
|
|
|
| Source is within 2^14 range. To perform the int operation,
|
|
| move it to d0.
|
|
|
|
|
src_in:
|
|
fmovex ETEMP(%a6),%fp0 |move in src for int
|
|
fmovel #rz_mode,%fpcr |force rz for src conversion
|
|
fmovel %fp0,%d0 |int src to d0
|
|
fmovel #0,%FPSR |clr status from above
|
|
tstw ETEMP(%a6) |check src sign
|
|
blt src_neg
|
|
|
|
|
| Source is positive. Add the src to the dest exponent.
|
|
| The result can be denormalized, if src = 0, or overflow,
|
|
| if the result of the add sets a bit in the upper word.
|
|
|
|
|
src_pos:
|
|
tstw %d1 |check for denorm
|
|
beq dst_dnrm
|
|
addl %d0,%d1 |add src to dest exp
|
|
beqs denorm |if zero, result is denorm
|
|
cmpil #0x7fff,%d1 |test for overflow
|
|
bges ovfl
|
|
tstb L_SCR1(%a6)
|
|
beqs spos_pos
|
|
orw #0x8000,%d1
|
|
spos_pos:
|
|
movew %d1,FPTEMP(%a6) |result in FPTEMP
|
|
fmovel USER_FPCR(%a6),%FPCR
|
|
fmovex FPTEMP(%a6),%fp0 |write result to fp0
|
|
rts
|
|
ovfl:
|
|
tstb L_SCR1(%a6)
|
|
beqs sovl_pos
|
|
orw #0x8000,%d1
|
|
sovl_pos:
|
|
movew FPTEMP(%a6),ETEMP(%a6) |result in ETEMP
|
|
movel FPTEMP_HI(%a6),ETEMP_HI(%a6)
|
|
movel FPTEMP_LO(%a6),ETEMP_LO(%a6)
|
|
bra t_ovfl2
|
|
|
|
denorm:
|
|
tstb L_SCR1(%a6)
|
|
beqs den_pos
|
|
orw #0x8000,%d1
|
|
den_pos:
|
|
tstl FPTEMP_HI(%a6) |check j bit
|
|
blts nden_exit |if set, not denorm
|
|
movew %d1,ETEMP(%a6) |input expected in ETEMP
|
|
movel FPTEMP_HI(%a6),ETEMP_HI(%a6)
|
|
movel FPTEMP_LO(%a6),ETEMP_LO(%a6)
|
|
orl #unfl_bit,USER_FPSR(%a6) |set unfl
|
|
leal ETEMP(%a6),%a0
|
|
bra t_resdnrm
|
|
nden_exit:
|
|
movew %d1,FPTEMP(%a6) |result in FPTEMP
|
|
fmovel USER_FPCR(%a6),%FPCR
|
|
fmovex FPTEMP(%a6),%fp0 |write result to fp0
|
|
rts
|
|
|
|
|
|
|
| Source is negative. Add the src to the dest exponent.
|
|
| (The result exponent will be reduced). The result can be
|
|
| denormalized.
|
|
|
|
|
src_neg:
|
|
addl %d0,%d1 |add src to dest
|
|
beqs denorm |if zero, result is denorm
|
|
blts fix_dnrm |if negative, result is
|
|
| ;needing denormalization
|
|
tstb L_SCR1(%a6)
|
|
beqs sneg_pos
|
|
orw #0x8000,%d1
|
|
sneg_pos:
|
|
movew %d1,FPTEMP(%a6) |result in FPTEMP
|
|
fmovel USER_FPCR(%a6),%FPCR
|
|
fmovex FPTEMP(%a6),%fp0 |write result to fp0
|
|
rts
|
|
|
|
|
|
|
|
|
| The result exponent is below denorm value. Test for catastrophic
|
|
| underflow and force zero if true. If not, try to shift the
|
|
| mantissa right until a zero exponent exists.
|
|
|
|
|
fix_dnrm:
|
|
cmpiw #0xffc0,%d1 |lower bound for normalization
|
|
blt fix_unfl |if lower, catastrophic unfl
|
|
movew %d1,%d0 |use d0 for exp
|
|
movel %d2,-(%a7) |free d2 for norm
|
|
movel FPTEMP_HI(%a6),%d1
|
|
movel FPTEMP_LO(%a6),%d2
|
|
clrl L_SCR2(%a6)
|
|
fix_loop:
|
|
addw #1,%d0 |drive d0 to 0
|
|
lsrl #1,%d1 |while shifting the
|
|
roxrl #1,%d2 |mantissa to the right
|
|
bccs no_carry
|
|
st L_SCR2(%a6) |use L_SCR2 to capture inex
|
|
no_carry:
|
|
tstw %d0 |it is finished when
|
|
blts fix_loop |d0 is zero or the mantissa
|
|
tstb L_SCR2(%a6)
|
|
beqs tst_zero
|
|
orl #unfl_inx_mask,USER_FPSR(%a6)
|
|
| ;set unfl, aunfl, ainex
|
|
|
|
|
| Test for zero. If zero, simply use fmove to return +/- zero
|
|
| to the fpu.
|
|
|
|
|
tst_zero:
|
|
clrw FPTEMP_EX(%a6)
|
|
tstb L_SCR1(%a6) |test for sign
|
|
beqs tst_con
|
|
orw #0x8000,FPTEMP_EX(%a6) |set sign bit
|
|
tst_con:
|
|
movel %d1,FPTEMP_HI(%a6)
|
|
movel %d2,FPTEMP_LO(%a6)
|
|
movel (%a7)+,%d2
|
|
tstl %d1
|
|
bnes not_zero
|
|
tstl FPTEMP_LO(%a6)
|
|
bnes not_zero
|
|
|
|
|
| Result is zero. Check for rounding mode to set lsb. If the
|
|
| mode is rp, and the zero is positive, return smallest denorm.
|
|
| If the mode is rm, and the zero is negative, return smallest
|
|
| negative denorm.
|
|
|
|
|
btstb #5,FPCR_MODE(%a6) |test if rm or rp
|
|
beqs no_dir
|
|
btstb #4,FPCR_MODE(%a6) |check which one
|
|
beqs zer_rm
|
|
zer_rp:
|
|
tstb L_SCR1(%a6) |check sign
|
|
bnes no_dir |if set, neg op, no inc
|
|
movel #1,FPTEMP_LO(%a6) |set lsb
|
|
bras sm_dnrm
|
|
zer_rm:
|
|
tstb L_SCR1(%a6) |check sign
|
|
beqs no_dir |if clr, neg op, no inc
|
|
movel #1,FPTEMP_LO(%a6) |set lsb
|
|
orl #neg_mask,USER_FPSR(%a6) |set N
|
|
bras sm_dnrm
|
|
no_dir:
|
|
fmovel USER_FPCR(%a6),%FPCR
|
|
fmovex FPTEMP(%a6),%fp0 |use fmove to set cc's
|
|
rts
|
|
|
|
|
|
|
| The rounding mode changed the zero to a smallest denorm. Call
|
|
| t_resdnrm with exceptional operand in ETEMP.
|
|
|
|
|
sm_dnrm:
|
|
movel FPTEMP_EX(%a6),ETEMP_EX(%a6)
|
|
movel FPTEMP_HI(%a6),ETEMP_HI(%a6)
|
|
movel FPTEMP_LO(%a6),ETEMP_LO(%a6)
|
|
leal ETEMP(%a6),%a0
|
|
bra t_resdnrm
|
|
|
|
|
|
|
| Result is still denormalized.
|
|
|
|
|
not_zero:
|
|
orl #unfl_mask,USER_FPSR(%a6) |set unfl
|
|
tstb L_SCR1(%a6) |check for sign
|
|
beqs fix_exit
|
|
orl #neg_mask,USER_FPSR(%a6) |set N
|
|
fix_exit:
|
|
bras sm_dnrm
|
|
|
|
|
|
|
|
|
| The result has underflowed to zero. Return zero and set
|
|
| unfl, aunfl, and ainex.
|
|
|
|
|
fix_unfl:
|
|
orl #unfl_inx_mask,USER_FPSR(%a6)
|
|
btstb #5,FPCR_MODE(%a6) |test if rm or rp
|
|
beqs no_dir2
|
|
btstb #4,FPCR_MODE(%a6) |check which one
|
|
beqs zer_rm2
|
|
zer_rp2:
|
|
tstb L_SCR1(%a6) |check sign
|
|
bnes no_dir2 |if set, neg op, no inc
|
|
clrl FPTEMP_EX(%a6)
|
|
clrl FPTEMP_HI(%a6)
|
|
movel #1,FPTEMP_LO(%a6) |set lsb
|
|
bras sm_dnrm |return smallest denorm
|
|
zer_rm2:
|
|
tstb L_SCR1(%a6) |check sign
|
|
beqs no_dir2 |if clr, neg op, no inc
|
|
movew #0x8000,FPTEMP_EX(%a6)
|
|
clrl FPTEMP_HI(%a6)
|
|
movel #1,FPTEMP_LO(%a6) |set lsb
|
|
orl #neg_mask,USER_FPSR(%a6) |set N
|
|
bra sm_dnrm |return smallest denorm
|
|
|
|
no_dir2:
|
|
tstb L_SCR1(%a6)
|
|
bges pos_zero
|
|
neg_zero:
|
|
clrl FP_SCR1(%a6) |clear the exceptional operand
|
|
clrl FP_SCR1+4(%a6) |for gen_except.
|
|
clrl FP_SCR1+8(%a6)
|
|
fmoves #0x80000000,%fp0
|
|
rts
|
|
pos_zero:
|
|
clrl FP_SCR1(%a6) |clear the exceptional operand
|
|
clrl FP_SCR1+4(%a6) |for gen_except.
|
|
clrl FP_SCR1+8(%a6)
|
|
fmoves #0x00000000,%fp0
|
|
rts
|
|
|
|
|
|
|
| The destination is a denormalized number. It must be handled
|
|
| by first shifting the bits in the mantissa until it is normalized,
|
|
| then adding the remainder of the source to the exponent.
|
|
|
|
|
dst_dnrm:
|
|
moveml %d2/%d3,-(%a7)
|
|
movew FPTEMP_EX(%a6),%d1
|
|
movel FPTEMP_HI(%a6),%d2
|
|
movel FPTEMP_LO(%a6),%d3
|
|
dst_loop:
|
|
tstl %d2 |test for normalized result
|
|
blts dst_norm |exit loop if so
|
|
tstl %d0 |otherwise, test shift count
|
|
beqs dst_fin |if zero, shifting is done
|
|
subil #1,%d0 |dec src
|
|
lsll #1,%d3
|
|
roxll #1,%d2
|
|
bras dst_loop
|
|
|
|
|
| Destination became normalized. Simply add the remaining
|
|
| portion of the src to the exponent.
|
|
|
|
|
dst_norm:
|
|
addw %d0,%d1 |dst is normalized; add src
|
|
tstb L_SCR1(%a6)
|
|
beqs dnrm_pos
|
|
orl #0x8000,%d1
|
|
dnrm_pos:
|
|
movemw %d1,FPTEMP_EX(%a6)
|
|
moveml %d2,FPTEMP_HI(%a6)
|
|
moveml %d3,FPTEMP_LO(%a6)
|
|
fmovel USER_FPCR(%a6),%FPCR
|
|
fmovex FPTEMP(%a6),%fp0
|
|
moveml (%a7)+,%d2/%d3
|
|
rts
|
|
|
|
|
|
|
| Destination remained denormalized. Call t_excdnrm with
|
|
| exceptional operand in ETEMP.
|
|
|
|
|
dst_fin:
|
|
tstb L_SCR1(%a6) |check for sign
|
|
beqs dst_exit
|
|
orl #neg_mask,USER_FPSR(%a6) |set N
|
|
orl #0x8000,%d1
|
|
dst_exit:
|
|
movemw %d1,ETEMP_EX(%a6)
|
|
moveml %d2,ETEMP_HI(%a6)
|
|
moveml %d3,ETEMP_LO(%a6)
|
|
orl #unfl_mask,USER_FPSR(%a6) |set unfl
|
|
moveml (%a7)+,%d2/%d3
|
|
leal ETEMP(%a6),%a0
|
|
bra t_resdnrm
|
|
|
|
|
|
|
| Source is outside of 2^14 range. Test the sign and branch
|
|
| to the appropriate exception handler.
|
|
|
|
|
src_out:
|
|
tstb L_SCR1(%a6)
|
|
beqs scro_pos
|
|
orl #0x8000,%d1
|
|
scro_pos:
|
|
movel FPTEMP_HI(%a6),ETEMP_HI(%a6)
|
|
movel FPTEMP_LO(%a6),ETEMP_LO(%a6)
|
|
tstw ETEMP(%a6)
|
|
blts res_neg
|
|
res_pos:
|
|
movew %d1,ETEMP(%a6) |result in ETEMP
|
|
bra t_ovfl2
|
|
res_neg:
|
|
movew %d1,ETEMP(%a6) |result in ETEMP
|
|
leal ETEMP(%a6),%a0
|
|
bra t_unfl
|
|
|end
|