cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

srem_mod.S (12287B)


      1|
      2|	srem_mod.sa 3.1 12/10/90
      3|
      4|      The entry point sMOD computes the floating point MOD of the
      5|      input values X and Y. The entry point sREM computes the floating
      6|      point (IEEE) REM of the input values X and Y.
      7|
      8|      INPUT
      9|      -----
     10|      Double-extended value Y is pointed to by address in register
     11|      A0. Double-extended value X is located in -12(A0). The values
     12|      of X and Y are both nonzero and finite; although either or both
     13|      of them can be denormalized. The special cases of zeros, NaNs,
     14|      and infinities are handled elsewhere.
     15|
     16|      OUTPUT
     17|      ------
     18|      FREM(X,Y) or FMOD(X,Y), depending on entry point.
     19|
     20|       ALGORITHM
     21|       ---------
     22|
     23|       Step 1.  Save and strip signs of X and Y: signX := sign(X),
     24|                signY := sign(Y), X := |X|, Y := |Y|,
     25|                signQ := signX EOR signY. Record whether MOD or REM
     26|                is requested.
     27|
     28|       Step 2.  Set L := expo(X)-expo(Y), k := 0, Q := 0.
     29|                If (L < 0) then
     30|                   R := X, go to Step 4.
     31|                else
     32|                   R := 2^(-L)X, j := L.
     33|                endif
     34|
     35|       Step 3.  Perform MOD(X,Y)
     36|            3.1 If R = Y, go to Step 9.
     37|            3.2 If R > Y, then { R := R - Y, Q := Q + 1}
     38|            3.3 If j = 0, go to Step 4.
     39|            3.4 k := k + 1, j := j - 1, Q := 2Q, R := 2R. Go to
     40|                Step 3.1.
     41|
     42|       Step 4.  At this point, R = X - QY = MOD(X,Y). Set
     43|                Last_Subtract := false (used in Step 7 below). If
     44|                MOD is requested, go to Step 6.
     45|
     46|       Step 5.  R = MOD(X,Y), but REM(X,Y) is requested.
     47|            5.1 If R < Y/2, then R = MOD(X,Y) = REM(X,Y). Go to
     48|                Step 6.
     49|            5.2 If R > Y/2, then { set Last_Subtract := true,
     50|                Q := Q + 1, Y := signY*Y }. Go to Step 6.
     51|            5.3 This is the tricky case of R = Y/2. If Q is odd,
     52|                then { Q := Q + 1, signX := -signX }.
     53|
     54|       Step 6.  R := signX*R.
     55|
     56|       Step 7.  If Last_Subtract = true, R := R - Y.
     57|
     58|       Step 8.  Return signQ, last 7 bits of Q, and R as required.
     59|
     60|       Step 9.  At this point, R = 2^(-j)*X - Q Y = Y. Thus,
     61|                X = 2^(j)*(Q+1)Y. set Q := 2^(j)*(Q+1),
     62|                R := 0. Return signQ, last 7 bits of Q, and R.
     63|
     64|
     65
     66|		Copyright (C) Motorola, Inc. 1990
     67|			All Rights Reserved
     68|
     69|       For details on the license for this file, please see the
     70|       file, README, in this same directory.
     71
     72SREM_MOD:    |idnt    2,1 | Motorola 040 Floating Point Software Package
     73
     74	|section    8
     75
     76#include "fpsp.h"
     77
     78	.set	Mod_Flag,L_SCR3
     79	.set	SignY,FP_SCR3+4
     80	.set	SignX,FP_SCR3+8
     81	.set	SignQ,FP_SCR3+12
     82	.set	Sc_Flag,FP_SCR4
     83
     84	.set	Y,FP_SCR1
     85	.set	Y_Hi,Y+4
     86	.set	Y_Lo,Y+8
     87
     88	.set	R,FP_SCR2
     89	.set	R_Hi,R+4
     90	.set	R_Lo,R+8
     91
     92
     93Scale:     .long	0x00010000,0x80000000,0x00000000,0x00000000
     94
     95	|xref	t_avoid_unsupp
     96
     97        .global        smod
     98smod:
     99
    100   movel               #0,Mod_Flag(%a6)
    101   bras                Mod_Rem
    102
    103        .global        srem
    104srem:
    105
    106   movel               #1,Mod_Flag(%a6)
    107
    108Mod_Rem:
    109|..Save sign of X and Y
    110   moveml              %d2-%d7,-(%a7)     | ...save data registers
    111   movew               (%a0),%d3
    112   movew               %d3,SignY(%a6)
    113   andil               #0x00007FFF,%d3   | ...Y := |Y|
    114
    115|
    116   movel               4(%a0),%d4
    117   movel               8(%a0),%d5        | ...(D3,D4,D5) is |Y|
    118
    119   tstl                %d3
    120   bnes                Y_Normal
    121
    122   movel               #0x00003FFE,%d3	| ...$3FFD + 1
    123   tstl                %d4
    124   bnes                HiY_not0
    125
    126HiY_0:
    127   movel               %d5,%d4
    128   clrl                %d5
    129   subil               #32,%d3
    130   clrl                %d6
    131   bfffo                %d4{#0:#32},%d6
    132   lsll                %d6,%d4
    133   subl                %d6,%d3           | ...(D3,D4,D5) is normalized
    134|                                       ...with bias $7FFD
    135   bras                Chk_X
    136
    137HiY_not0:
    138   clrl                %d6
    139   bfffo                %d4{#0:#32},%d6
    140   subl                %d6,%d3
    141   lsll                %d6,%d4
    142   movel               %d5,%d7           | ...a copy of D5
    143   lsll                %d6,%d5
    144   negl                %d6
    145   addil               #32,%d6
    146   lsrl                %d6,%d7
    147   orl                 %d7,%d4           | ...(D3,D4,D5) normalized
    148|                                       ...with bias $7FFD
    149   bras                Chk_X
    150
    151Y_Normal:
    152   addil               #0x00003FFE,%d3   | ...(D3,D4,D5) normalized
    153|                                       ...with bias $7FFD
    154
    155Chk_X:
    156   movew               -12(%a0),%d0
    157   movew               %d0,SignX(%a6)
    158   movew               SignY(%a6),%d1
    159   eorl                %d0,%d1
    160   andil               #0x00008000,%d1
    161   movew               %d1,SignQ(%a6)	| ...sign(Q) obtained
    162   andil               #0x00007FFF,%d0
    163   movel               -8(%a0),%d1
    164   movel               -4(%a0),%d2       | ...(D0,D1,D2) is |X|
    165   tstl                %d0
    166   bnes                X_Normal
    167   movel               #0x00003FFE,%d0
    168   tstl                %d1
    169   bnes                HiX_not0
    170
    171HiX_0:
    172   movel               %d2,%d1
    173   clrl                %d2
    174   subil               #32,%d0
    175   clrl                %d6
    176   bfffo                %d1{#0:#32},%d6
    177   lsll                %d6,%d1
    178   subl                %d6,%d0           | ...(D0,D1,D2) is normalized
    179|                                       ...with bias $7FFD
    180   bras                Init
    181
    182HiX_not0:
    183   clrl                %d6
    184   bfffo                %d1{#0:#32},%d6
    185   subl                %d6,%d0
    186   lsll                %d6,%d1
    187   movel               %d2,%d7           | ...a copy of D2
    188   lsll                %d6,%d2
    189   negl                %d6
    190   addil               #32,%d6
    191   lsrl                %d6,%d7
    192   orl                 %d7,%d1           | ...(D0,D1,D2) normalized
    193|                                       ...with bias $7FFD
    194   bras                Init
    195
    196X_Normal:
    197   addil               #0x00003FFE,%d0   | ...(D0,D1,D2) normalized
    198|                                       ...with bias $7FFD
    199
    200Init:
    201|
    202   movel               %d3,L_SCR1(%a6)   | ...save biased expo(Y)
    203   movel		%d0,L_SCR2(%a6)	|save d0
    204   subl                %d3,%d0           | ...L := expo(X)-expo(Y)
    205|   Move.L               D0,L            ...D0 is j
    206   clrl                %d6              | ...D6 := carry <- 0
    207   clrl                %d3              | ...D3 is Q
    208   moveal              #0,%a1           | ...A1 is k; j+k=L, Q=0
    209
    210|..(Carry,D1,D2) is R
    211   tstl                %d0
    212   bges                Mod_Loop
    213
    214|..expo(X) < expo(Y). Thus X = mod(X,Y)
    215|
    216   movel		L_SCR2(%a6),%d0	|restore d0
    217   bra                Get_Mod
    218
    219|..At this point  R = 2^(-L)X; Q = 0; k = 0; and  k+j = L
    220
    221
    222Mod_Loop:
    223   tstl                %d6              | ...test carry bit
    224   bgts                R_GT_Y
    225
    226|..At this point carry = 0, R = (D1,D2), Y = (D4,D5)
    227   cmpl                %d4,%d1           | ...compare hi(R) and hi(Y)
    228   bnes                R_NE_Y
    229   cmpl                %d5,%d2           | ...compare lo(R) and lo(Y)
    230   bnes                R_NE_Y
    231
    232|..At this point, R = Y
    233   bra                Rem_is_0
    234
    235R_NE_Y:
    236|..use the borrow of the previous compare
    237   bcss                R_LT_Y          | ...borrow is set iff R < Y
    238
    239R_GT_Y:
    240|..If Carry is set, then Y < (Carry,D1,D2) < 2Y. Otherwise, Carry = 0
    241|..and Y < (D1,D2) < 2Y. Either way, perform R - Y
    242   subl                %d5,%d2           | ...lo(R) - lo(Y)
    243   subxl               %d4,%d1           | ...hi(R) - hi(Y)
    244   clrl                %d6              | ...clear carry
    245   addql               #1,%d3           | ...Q := Q + 1
    246
    247R_LT_Y:
    248|..At this point, Carry=0, R < Y. R = 2^(k-L)X - QY; k+j = L; j >= 0.
    249   tstl                %d0              | ...see if j = 0.
    250   beqs                PostLoop
    251
    252   addl                %d3,%d3           | ...Q := 2Q
    253   addl                %d2,%d2           | ...lo(R) = 2lo(R)
    254   roxll               #1,%d1           | ...hi(R) = 2hi(R) + carry
    255   scs                  %d6              | ...set Carry if 2(R) overflows
    256   addql               #1,%a1           | ...k := k+1
    257   subql               #1,%d0           | ...j := j - 1
    258|..At this point, R=(Carry,D1,D2) = 2^(k-L)X - QY, j+k=L, j >= 0, R < 2Y.
    259
    260   bras                Mod_Loop
    261
    262PostLoop:
    263|..k = L, j = 0, Carry = 0, R = (D1,D2) = X - QY, R < Y.
    264
    265|..normalize R.
    266   movel               L_SCR1(%a6),%d0           | ...new biased expo of R
    267   tstl                %d1
    268   bnes                HiR_not0
    269
    270HiR_0:
    271   movel               %d2,%d1
    272   clrl                %d2
    273   subil               #32,%d0
    274   clrl                %d6
    275   bfffo                %d1{#0:#32},%d6
    276   lsll                %d6,%d1
    277   subl                %d6,%d0           | ...(D0,D1,D2) is normalized
    278|                                       ...with bias $7FFD
    279   bras                Get_Mod
    280
    281HiR_not0:
    282   clrl                %d6
    283   bfffo                %d1{#0:#32},%d6
    284   bmis                Get_Mod         | ...already normalized
    285   subl                %d6,%d0
    286   lsll                %d6,%d1
    287   movel               %d2,%d7           | ...a copy of D2
    288   lsll                %d6,%d2
    289   negl                %d6
    290   addil               #32,%d6
    291   lsrl                %d6,%d7
    292   orl                 %d7,%d1           | ...(D0,D1,D2) normalized
    293
    294|
    295Get_Mod:
    296   cmpil		#0x000041FE,%d0
    297   bges		No_Scale
    298Do_Scale:
    299   movew		%d0,R(%a6)
    300   clrw		R+2(%a6)
    301   movel		%d1,R_Hi(%a6)
    302   movel		%d2,R_Lo(%a6)
    303   movel		L_SCR1(%a6),%d6
    304   movew		%d6,Y(%a6)
    305   clrw		Y+2(%a6)
    306   movel		%d4,Y_Hi(%a6)
    307   movel		%d5,Y_Lo(%a6)
    308   fmovex		R(%a6),%fp0		| ...no exception
    309   movel		#1,Sc_Flag(%a6)
    310   bras		ModOrRem
    311No_Scale:
    312   movel		%d1,R_Hi(%a6)
    313   movel		%d2,R_Lo(%a6)
    314   subil		#0x3FFE,%d0
    315   movew		%d0,R(%a6)
    316   clrw		R+2(%a6)
    317   movel		L_SCR1(%a6),%d6
    318   subil		#0x3FFE,%d6
    319   movel		%d6,L_SCR1(%a6)
    320   fmovex		R(%a6),%fp0
    321   movew		%d6,Y(%a6)
    322   movel		%d4,Y_Hi(%a6)
    323   movel		%d5,Y_Lo(%a6)
    324   movel		#0,Sc_Flag(%a6)
    325
    326|
    327
    328
    329ModOrRem:
    330   movel               Mod_Flag(%a6),%d6
    331   beqs                Fix_Sign
    332
    333   movel               L_SCR1(%a6),%d6           | ...new biased expo(Y)
    334   subql               #1,%d6           | ...biased expo(Y/2)
    335   cmpl                %d6,%d0
    336   blts                Fix_Sign
    337   bgts                Last_Sub
    338
    339   cmpl                %d4,%d1
    340   bnes                Not_EQ
    341   cmpl                %d5,%d2
    342   bnes                Not_EQ
    343   bra                Tie_Case
    344
    345Not_EQ:
    346   bcss                Fix_Sign
    347
    348Last_Sub:
    349|
    350   fsubx		Y(%a6),%fp0		| ...no exceptions
    351   addql               #1,%d3           | ...Q := Q + 1
    352
    353|
    354
    355Fix_Sign:
    356|..Get sign of X
    357   movew               SignX(%a6),%d6
    358   bges		Get_Q
    359   fnegx		%fp0
    360
    361|..Get Q
    362|
    363Get_Q:
    364   clrl		%d6
    365   movew               SignQ(%a6),%d6        | ...D6 is sign(Q)
    366   movel               #8,%d7
    367   lsrl                %d7,%d6
    368   andil               #0x0000007F,%d3   | ...7 bits of Q
    369   orl                 %d6,%d3           | ...sign and bits of Q
    370   swap                 %d3
    371   fmovel              %fpsr,%d6
    372   andil               #0xFF00FFFF,%d6
    373   orl                 %d3,%d6
    374   fmovel              %d6,%fpsr         | ...put Q in fpsr
    375
    376|
    377Restore:
    378   moveml              (%a7)+,%d2-%d7
    379   fmovel              USER_FPCR(%a6),%fpcr
    380   movel               Sc_Flag(%a6),%d0
    381   beqs                Finish
    382   fmulx		Scale(%pc),%fp0	| ...may cause underflow
    383   bra			t_avoid_unsupp	|check for denorm as a
    384|					;result of the scaling
    385
    386Finish:
    387	fmovex		%fp0,%fp0		|capture exceptions & round
    388	rts
    389
    390Rem_is_0:
    391|..R = 2^(-j)X - Q Y = Y, thus R = 0 and quotient = 2^j (Q+1)
    392   addql               #1,%d3
    393   cmpil               #8,%d0           | ...D0 is j
    394   bges                Q_Big
    395
    396   lsll                %d0,%d3
    397   bras                Set_R_0
    398
    399Q_Big:
    400   clrl                %d3
    401
    402Set_R_0:
    403   fmoves		#0x00000000,%fp0
    404   movel		#0,Sc_Flag(%a6)
    405   bra                Fix_Sign
    406
    407Tie_Case:
    408|..Check parity of Q
    409   movel               %d3,%d6
    410   andil               #0x00000001,%d6
    411   tstl                %d6
    412   beq                Fix_Sign	| ...Q is even
    413
    414|..Q is odd, Q := Q + 1, signX := -signX
    415   addql               #1,%d3
    416   movew               SignX(%a6),%d6
    417   eoril               #0x00008000,%d6
    418   movew               %d6,SignX(%a6)
    419   bra                Fix_Sign
    420
    421   |end