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

ssinh.S (3410B)


      1|
      2|	ssinh.sa 3.1 12/10/90
      3|
      4|       The entry point sSinh computes the hyperbolic sine of
      5|       an input argument; sSinhd does the same except for denormalized
      6|       input.
      7|
      8|       Input: Double-extended number X in location pointed to
      9|		by address register a0.
     10|
     11|       Output: The value sinh(X) returned in floating-point register Fp0.
     12|
     13|       Accuracy and Monotonicity: The returned result is within 3 ulps in
     14|               64 significant bit, i.e. within 0.5001 ulp to 53 bits if the
     15|               result is subsequently rounded to double precision. The
     16|               result is provably monotonic in double precision.
     17|
     18|       Speed: The program sSINH takes approximately 280 cycles.
     19|
     20|       Algorithm:
     21|
     22|       SINH
     23|       1. If |X| > 16380 log2, go to 3.
     24|
     25|       2. (|X| <= 16380 log2) Sinh(X) is obtained by the formulae
     26|               y = |X|, sgn = sign(X), and z = expm1(Y),
     27|               sinh(X) = sgn*(1/2)*( z + z/(1+z) ).
     28|          Exit.
     29|
     30|       3. If |X| > 16480 log2, go to 5.
     31|
     32|       4. (16380 log2 < |X| <= 16480 log2)
     33|               sinh(X) = sign(X) * exp(|X|)/2.
     34|          However, invoking exp(|X|) may cause premature overflow.
     35|          Thus, we calculate sinh(X) as follows:
     36|             Y       := |X|
     37|             sgn     := sign(X)
     38|             sgnFact := sgn * 2**(16380)
     39|             Y'      := Y - 16381 log2
     40|             sinh(X) := sgnFact * exp(Y').
     41|          Exit.
     42|
     43|       5. (|X| > 16480 log2) sinh(X) must overflow. Return
     44|          sign(X)*Huge*Huge to generate overflow and an infinity with
     45|          the appropriate sign. Huge is the largest finite number in
     46|          extended format. Exit.
     47|
     48
     49|		Copyright (C) Motorola, Inc. 1990
     50|			All Rights Reserved
     51|
     52|       For details on the license for this file, please see the
     53|       file, README, in this same directory.
     54
     55|SSINH	idnt	2,1 | Motorola 040 Floating Point Software Package
     56
     57	|section	8
     58
     59T1:	.long 0x40C62D38,0xD3D64634 | ... 16381 LOG2 LEAD
     60T2:	.long 0x3D6F90AE,0xB1E75CC7 | ... 16381 LOG2 TRAIL
     61
     62	|xref	t_frcinx
     63	|xref	t_ovfl
     64	|xref	t_extdnrm
     65	|xref	setox
     66	|xref	setoxm1
     67
     68	.global	ssinhd
     69ssinhd:
     70|--SINH(X) = X FOR DENORMALIZED X
     71
     72	bra	t_extdnrm
     73
     74	.global	ssinh
     75ssinh:
     76	fmovex	(%a0),%fp0	| ...LOAD INPUT
     77
     78	movel	(%a0),%d0
     79	movew	4(%a0),%d0
     80	movel	%d0,%a1		| save a copy of original (compacted) operand
     81	andl	#0x7FFFFFFF,%d0
     82	cmpl	#0x400CB167,%d0
     83	bgts	SINHBIG
     84
     85|--THIS IS THE USUAL CASE, |X| < 16380 LOG2
     86|--Y = |X|, Z = EXPM1(Y), SINH(X) = SIGN(X)*(1/2)*( Z + Z/(1+Z) )
     87
     88	fabsx	%fp0		| ...Y = |X|
     89
     90	moveml	%a1/%d1,-(%sp)
     91	fmovemx %fp0-%fp0,(%a0)
     92	clrl	%d1
     93	bsr	setoxm1		| ...FP0 IS Z = EXPM1(Y)
     94	fmovel	#0,%fpcr
     95	moveml	(%sp)+,%a1/%d1
     96
     97	fmovex	%fp0,%fp1
     98	fadds	#0x3F800000,%fp1	| ...1+Z
     99	fmovex	%fp0,-(%sp)
    100	fdivx	%fp1,%fp0		| ...Z/(1+Z)
    101	movel	%a1,%d0
    102	andl	#0x80000000,%d0
    103	orl	#0x3F000000,%d0
    104	faddx	(%sp)+,%fp0
    105	movel	%d0,-(%sp)
    106
    107	fmovel	%d1,%fpcr
    108	fmuls	(%sp)+,%fp0	|last fp inst - possible exceptions set
    109
    110	bra	t_frcinx
    111
    112SINHBIG:
    113	cmpl	#0x400CB2B3,%d0
    114	bgt	t_ovfl
    115	fabsx	%fp0
    116	fsubd	T1(%pc),%fp0	| ...(|X|-16381LOG2_LEAD)
    117	movel	#0,-(%sp)
    118	movel	#0x80000000,-(%sp)
    119	movel	%a1,%d0
    120	andl	#0x80000000,%d0
    121	orl	#0x7FFB0000,%d0
    122	movel	%d0,-(%sp)	| ...EXTENDED FMT
    123	fsubd	T2(%pc),%fp0	| ...|X| - 16381 LOG2, ACCURATE
    124
    125	movel	%d1,-(%sp)
    126	clrl	%d1
    127	fmovemx %fp0-%fp0,(%a0)
    128	bsr	setox
    129	fmovel	(%sp)+,%fpcr
    130
    131	fmulx	(%sp)+,%fp0	|possible exception
    132	bra	t_frcinx
    133
    134	|end