res_func.s

优龙2410linux2.6.8内核源代码

wi:
	moveql	#2,%d0		|set byte count

	btstb	#7,STAG(%a6)	|check for extended denorm
	bne	int_dnrm	|branch if so

	fmovemx ETEMP(%a6),%fp0-%fp0
	fcmps	#0x46fffe00,%fp0
| 46fffe00 in sgl prec = 400d0000fffe000000000000 in ext prec
	fbge	wo_plrg
	fcmps	#0xc7000000,%fp0
| c7000000 in sgl prec = c00e00008000000000000000 in ext prec
	fble	wo_nlrg

|
| at this point, the answer is between the largest pos and neg values
|
	movel	USER_FPCR(%a6),%d1	|use user's rounding mode
	andil	#0x30,%d1
	fmovel	%d1,%fpcr
	fmovew	%fp0,L_SCR1(%a6)	|let the 040 perform conversion
	fmovel %fpsr,%d1
	orl	%d1,USER_FPSR(%a6)	|capture inex2/ainex if set
	bra	int_wrt

wo_plrg:
	movew	#0x7fff,L_SCR1(%a6)	|answer is largest positive int
	fbeq	int_wrt			|exact answer
	fcmps	#0x46ffff00,%fp0
| 46ffff00 in sgl prec = 400d0000ffff000000000000 in ext prec
	fbge	int_operr		|set operr
	bra	int_inx			|set inexact

wo_nlrg:
	movew	#0x8000,L_SCR1(%a6)
	fbeq	int_wrt			|exact answer
	fcmps	#0xc7000080,%fp0
| c7000080 in sgl prec = c00e00008000800000000000 in ext prec
	fblt	int_operr		|set operr
	bra	int_inx			|set inexact

|
| bi is used to handle a byte integer source specifier
|

bi:
	moveql	#1,%d0		|set byte count

	btstb	#7,STAG(%a6)	|check for extended denorm
	bne	int_dnrm	|branch if so

	fmovemx ETEMP(%a6),%fp0-%fp0
	fcmps	#0x42fe0000,%fp0
| 42fe0000 in sgl prec = 40050000fe00000000000000 in ext prec
	fbge	by_plrg
	fcmps	#0xc3000000,%fp0
| c3000000 in sgl prec = c00600008000000000000000 in ext prec
	fble	by_nlrg

|
| at this point, the answer is between the largest pos and neg values
|
	movel	USER_FPCR(%a6),%d1	|use user's rounding mode
	andil	#0x30,%d1
	fmovel	%d1,%fpcr
	fmoveb	%fp0,L_SCR1(%a6)	|let the 040 perform conversion
	fmovel %fpsr,%d1
	orl	%d1,USER_FPSR(%a6)	|capture inex2/ainex if set
	bra	int_wrt

by_plrg:
	moveb	#0x7f,L_SCR1(%a6)		|answer is largest positive int
	fbeq	int_wrt			|exact answer
	fcmps	#0x42ff0000,%fp0
| 42ff0000 in sgl prec = 40050000ff00000000000000 in ext prec
	fbge	int_operr		|set operr
	bra	int_inx			|set inexact

by_nlrg:
	moveb	#0x80,L_SCR1(%a6)
	fbeq	int_wrt			|exact answer
	fcmps	#0xc3008000,%fp0
| c3008000 in sgl prec = c00600008080000000000000 in ext prec
	fblt	int_operr		|set operr
	bra	int_inx			|set inexact

|
| Common integer routines
|
| int_drnrm---account for possible nonzero result for round up with positive
| operand and round down for negative answer.  In the first case (result = 1)
| byte-width (store in d0) of result must be honored.  In the second case,
| -1 in L_SCR1(a6) will cover all contingencies (FMOVE.B/W/L out).

int_dnrm:
	movel	#0,L_SCR1(%a6)	| initialize result to 0
	bfextu	FPCR_MODE(%a6){#2:#2},%d1	| d1 is the rounding mode
	cmpb	#2,%d1
	bmis	int_inx		| if RN or RZ, done
	bnes	int_rp		| if RP, continue below
	tstw	ETEMP(%a6)	| RM: store -1 in L_SCR1 if src is negative
	bpls	int_inx		| otherwise result is 0
	movel	#-1,L_SCR1(%a6)
	bras	int_inx
int_rp:
	tstw	ETEMP(%a6)	| RP: store +1 of proper width in L_SCR1 if
|				; source is greater than 0
	bmis	int_inx		| otherwise, result is 0
	lea	L_SCR1(%a6),%a1	| a1 is address of L_SCR1
	addal	%d0,%a1		| offset by destination width -1
	subal	#1,%a1
	bsetb	#0,(%a1)		| set low bit at a1 address
int_inx:
	oril	#inx2a_mask,USER_FPSR(%a6)
	bras	int_wrt
int_operr:
	fmovemx %fp0-%fp0,FPTEMP(%a6)	|FPTEMP must contain the extended
|				;precision source that needs to be
|				;converted to integer this is required
|				;if the operr exception is enabled.
|				;set operr/aiop (no inex2 on int ovfl)

	oril	#opaop_mask,USER_FPSR(%a6)
|				;fall through to perform int_wrt
int_wrt:
	movel	EXC_EA(%a6),%a1	|load destination address
	tstl	%a1		|check to see if it is a dest register
	beqs	wrt_dn		|write data register
	lea	L_SCR1(%a6),%a0	|point to supervisor source address
	bsrl	mem_write
	bra	mvouti_end

wrt_dn:
	movel	%d0,-(%sp)	|d0 currently contains the size to write
	bsrl	get_fline	|get_fline returns Dn in d0
	andiw	#0x7,%d0		|isolate register
	movel	(%sp)+,%d1	|get size
	cmpil	#4,%d1		|most frequent case
	beqs	sz_long
	cmpil	#2,%d1
	bnes	sz_con
	orl	#8,%d0		|add 'word' size to register#
	bras	sz_con
sz_long:
	orl	#0x10,%d0		|add 'long' size to register#
sz_con:
	movel	%d0,%d1		|reg_dest expects size:reg in d1
	bsrl	reg_dest	|load proper data register
	bra	mvouti_end
xp:
	lea	ETEMP(%a6),%a0
	bclrb	#sign_bit,LOCAL_EX(%a0)
	sne	LOCAL_SGN(%a0)
	btstb	#7,STAG(%a6)	|check for extended denorm
	bne	xdnrm
	clrl	%d0
	bras	do_fp		|do normal case
sgp:
	lea	ETEMP(%a6),%a0
	bclrb	#sign_bit,LOCAL_EX(%a0)
	sne	LOCAL_SGN(%a0)
	btstb	#7,STAG(%a6)	|check for extended denorm
	bne	sp_catas	|branch if so
	movew	LOCAL_EX(%a0),%d0
	lea	sp_bnds,%a1
	cmpw	(%a1),%d0
	blt	sp_under
	cmpw	2(%a1),%d0
	bgt	sp_over
	movel	#1,%d0		|set destination format to single
	bras	do_fp		|do normal case
dp:
	lea	ETEMP(%a6),%a0
	bclrb	#sign_bit,LOCAL_EX(%a0)
	sne	LOCAL_SGN(%a0)

	btstb	#7,STAG(%a6)	|check for extended denorm
	bne	dp_catas	|branch if so

	movew	LOCAL_EX(%a0),%d0
	lea	dp_bnds,%a1

	cmpw	(%a1),%d0
	blt	dp_under
	cmpw	2(%a1),%d0
	bgt	dp_over

	movel	#2,%d0		|set destination format to double
|				;fall through to do_fp
|
do_fp:
	bfextu	FPCR_MODE(%a6){#2:#2},%d1	|rnd mode in d1
	swap	%d0			|rnd prec in upper word
	addl	%d0,%d1			|d1 has PREC/MODE info

	clrl	%d0			|clear g,r,s

	bsrl	round			|round

	movel	%a0,%a1
	movel	EXC_EA(%a6),%a0

	bfextu	CMDREG1B(%a6){#3:#3},%d1	|extract destination format
|					;at this point only the dest
|					;formats sgl, dbl, ext are
|					;possible
	cmpb	#2,%d1
	bgts	ddbl			|double=5, extended=2, single=1
	bnes	dsgl
|					;fall through to dext
dext:
	bsrl	dest_ext
	bra	mvout_end
dsgl:
	bsrl	dest_sgl
	bra	mvout_end
ddbl:
	bsrl	dest_dbl
	bra	mvout_end

|
| Handle possible denorm or catastrophic underflow cases here
|
xdnrm:
	bsr	set_xop		|initialize WBTEMP
	bsetb	#wbtemp15_bit,WB_BYTE(%a6) |set wbtemp15

	movel	%a0,%a1
	movel	EXC_EA(%a6),%a0	|a0 has the destination pointer
	bsrl	dest_ext	|store to memory
	bsetb	#unfl_bit,FPSR_EXCEPT(%a6)
	bra	mvout_end

sp_under:
	bsetb	#etemp15_bit,STAG(%a6)

	cmpw	4(%a1),%d0
	blts	sp_catas	|catastrophic underflow case

	movel	#1,%d0		|load in round precision
	movel	#sgl_thresh,%d1	|load in single denorm threshold
	bsrl	dpspdnrm	|expects d1 to have the proper
|				;denorm threshold
	bsrl	dest_sgl	|stores value to destination
	bsetb	#unfl_bit,FPSR_EXCEPT(%a6)
	bra	mvout_end	|exit

dp_under:
	bsetb	#etemp15_bit,STAG(%a6)

	cmpw	4(%a1),%d0
	blts	dp_catas	|catastrophic underflow case

	movel	#dbl_thresh,%d1	|load in double precision threshold
	movel	#2,%d0
	bsrl	dpspdnrm	|expects d1 to have proper
|				;denorm threshold
|				;expects d0 to have round precision
	bsrl	dest_dbl	|store value to destination
	bsetb	#unfl_bit,FPSR_EXCEPT(%a6)
	bra	mvout_end	|exit

|
| Handle catastrophic underflow cases here
|
sp_catas:
| Temp fix for z bit set in unf_sub
	movel	USER_FPSR(%a6),-(%a7)

	movel	#1,%d0		|set round precision to sgl

	bsrl	unf_sub		|a0 points to result

	movel	(%a7)+,USER_FPSR(%a6)

	movel	#1,%d0
	subw	%d0,LOCAL_EX(%a0) |account for difference between
|				;denorm/norm bias

	movel	%a0,%a1		|a1 has the operand input
	movel	EXC_EA(%a6),%a0	|a0 has the destination pointer

	bsrl	dest_sgl	|store the result
	oril	#unfinx_mask,USER_FPSR(%a6)
	bra	mvout_end

dp_catas:
| Temp fix for z bit set in unf_sub
	movel	USER_FPSR(%a6),-(%a7)

	movel	#2,%d0		|set round precision to dbl
	bsrl	unf_sub		|a0 points to result

	movel	(%a7)+,USER_FPSR(%a6)

	movel	#1,%d0
	subw	%d0,LOCAL_EX(%a0) |account for difference between
|				;denorm/norm bias

	movel	%a0,%a1		|a1 has the operand input
	movel	EXC_EA(%a6),%a0	|a0 has the destination pointer

	bsrl	dest_dbl	|store the result
	oril	#unfinx_mask,USER_FPSR(%a6)
	bra	mvout_end

|
| Handle catastrophic overflow cases here
|
sp_over:
| Temp fix for z bit set in unf_sub
	movel	USER_FPSR(%a6),-(%a7)

	movel	#1,%d0
	leal	FP_SCR1(%a6),%a0	|use FP_SCR1 for creating result
	movel	ETEMP_EX(%a6),(%a0)
	movel	ETEMP_HI(%a6),4(%a0)
	movel	ETEMP_LO(%a6),8(%a0)
	bsrl	ovf_res

	movel	(%a7)+,USER_FPSR(%a6)

	movel	%a0,%a1
	movel	EXC_EA(%a6),%a0
	bsrl	dest_sgl
	orl	#ovfinx_mask,USER_FPSR(%a6)
	bra	mvout_end

dp_over:
| Temp fix for z bit set in ovf_res
	movel	USER_FPSR(%a6),-(%a7)

	movel	#2,%d0
	leal	FP_SCR1(%a6),%a0	|use FP_SCR1 for creating result
	movel	ETEMP_EX(%a6),(%a0)
	movel	ETEMP_HI(%a6),4(%a0)
	movel	ETEMP_LO(%a6),8(%a0)
	bsrl	ovf_res

	movel	(%a7)+,USER_FPSR(%a6)

	movel	%a0,%a1
	movel	EXC_EA(%a6),%a0
	bsrl	dest_dbl
	orl	#ovfinx_mask,USER_FPSR(%a6)
	bra	mvout_end

|
|	DPSPDNRM
|
| This subroutine takes an extended normalized number and denormalizes
| it to the given round precision. This subroutine also decrements
| the input operand's exponent by 1 to account for the fact that
| dest_sgl or dest_dbl expects a normalized number's bias.
|
| Input: a0  points to a normalized number in internal extended format
|	 d0  is the round precision (=1 for sgl; =2 for dbl)
|	 d1  is the single precision or double precision
|	     denorm threshold
|
| Output: (In the format for dest_sgl or dest_dbl)
|	 a0   points to the destination
|	 a1   points to the operand
|
| Exceptions: Reports inexact 2 exception by setting USER_FPSR bits
|
dpspdnrm:
	movel	%d0,-(%a7)	|save round precision
	clrl	%d0		|clear initial g,r,s
	bsrl	dnrm_lp		|careful with d0, it's needed by round

	bfextu	FPCR_MODE(%a6){#2:#2},%d1 |get rounding mode
	swap	%d1
	movew	2(%a7),%d1	|set rounding precision
	swap	%d1		|at this point d1 has PREC/MODE info
	bsrl	round		|round result, sets the inex bit in
|				;USER_FPSR if needed

	movew	#1,%d0
	subw	%d0,LOCAL_EX(%a0) |account for difference in denorm
|				;vs norm bias

	movel	%a0,%a1		|a1 has the operand input
	movel	EXC_EA(%a6),%a0	|a0 has the destination pointer
	addw	#4,%a7		|pop stack
	rts
|
| SET_XOP initialized WBTEMP with the value pointed to by a0
| input: a0 points to input operand in the internal extended format
|
set_xop:
	movel	LOCAL_EX(%a0),WBTEMP_EX(%a6)
	movel	LOCAL_HI(%a0),WBTEMP_HI(%a6)
	movel	LOCAL_LO(%a0),WBTEMP_LO(%a6)
	bfclr	WBTEMP_SGN(%a6){#0:#8}
	beqs	sxop
	bsetb	#sign_bit,WBTEMP_EX(%a6)
sxop:
	bfclr	STAG(%a6){#5:#4}	|clear wbtm66,wbtm1,wbtm0,sbit
	rts
|
|	P_MOVE
|
p_movet:
	.long	p_move
	.long	p_movez
	.long	p_movei
	.long	p_moven
	.long	p_move
p_regd:
	.long	p_dyd0
	.long	p_dyd1
	.long	p_dyd2
	.long	p_dyd3
	.long	p_dyd4
	.long	p_dyd5
	.long	p_dyd6
	.long	p_dyd7

pack_out:
	leal	p_movet,%a0	|load jmp table address
	movew	STAG(%a6),%d0	|get source tag
	bfextu	%d0{#16:#3},%d0	|isolate source bits
	movel	(%a0,%d0.w*4),%a0	|load a0 with routine label for tag
	jmp	(%a0)		|go to the routine

p_write:
	movel	#0x0c,%d0	|get byte count
	movel	EXC_EA(%a6),%a1	|get the destination address
	bsr	mem_write	|write the user's destination
	moveb	#0,CU_SAVEPC(%a6) |set the cu save pc to all 0's

|
| Also note that the dtag must be set to norm here - this is because
| the 040 uses the dtag to execute the correct microcode.
|
        bfclr    DTAG(%a6){#0:#3}  |set dtag to norm

	rts

| Notes on handling of special case (zero, inf, and nan) inputs:
|	1. Operr is not signalled if the k-factor is greater than 18.
|	2. Per the manual, status bits are not set.
|

p_move:
	movew	CMDREG1B(%a6),%d0
	btstl	#kfact_bit,%d0	|test for dynamic k-factor
	beqs	statick		|if clear, k-factor is static
dynamick:
	bfextu	%d0{#25:#3},%d0	|isolate register for dynamic k-factor
	lea	p_regd,%a0
	movel	%a0@(%d0:l:4),%a0
	jmp	(%a0)
statick:
	andiw	#0x007f,%d0	|get k-factor
	bfexts	%d0{#25:#7},%d0	|sign extend d0 for bindec
	leal	ETEMP(%a6),%a0	|a0 will point to the packed decimal
	bsrl	bindec		|perform the convert; data at a6
	leal	FP_SCR1(%a6),%a0	|load a0 with result address
	bral	p_write
p_movez:
	leal	ETEMP(%a6),%a0	|a0 will point to the packed decimal
	clrw	2(%a0)		|clear lower word of exp
	clrl	4(%a0)		|load second lword of ZERO
	clrl	8(%a0)		|load third lword of ZERO
	bra	p_write		|go write results
p_movei:
	fmovel	#0,%FPSR		|clear aiop
	leal	ETEMP(%a6),%a0	|a0 will point to the packed decimal
	clrw	2(%a0)		|clear lower word of exp
	bra	p_write		|go write the result
p_moven:
	leal	ETEMP(%a6),%a0	|a0 will point to the packed decimal
	clrw	2(%a0)		|clear lower word of exp
	bra	p_write		|go write the result

|
| Routines to read the dynamic k-factor from Dn.
|
p_dyd0:
	movel	USER_D0(%a6),%d0
	bras	statick
p_dyd1:
	movel	USER_D1(%a6),%d0
	bras	statick
p_dyd2:
	movel	%d2,%d0
	bras	statick
p_dyd3:
	movel	%d3,%d0
	bras	statick
p_dyd4:
	movel	%d4,%d0
	bras	statick
p_dyd5:
	movel	%d5,%d0
	bras	statick
p_dyd6:
	movel	%d6,%d0
	bra	statick
p_dyd7:
	movel	%d7,%d0
	bra	statick

	|end