Lines Matching refs:CAN
634 # can handle this case. #
674 # since, I believe, only NORMs and DENORMs can come through here,
675 # maybe we can avoid the subroutine call.
681 # that can pass through fpsp_ovfl(). remember that fcmp, ftst, and fsincos
720 # maybe we can make these entry points ONLY the OVFL entry points of each routine.
874 # can handle this case. #
919 # that can pass through fpsp_unfl(). remember that fcmp, and ftst
924 # now, what's left that's not dyadic is fsincos. we can distinguish it
963 # maybe we can make these entry points ONLY the OVFL entry points of each routine.
1150 # Two main instruction types can enter here: (1) DENORM or UNNORM #
1151 # unimplemented data types. These can be either opclass 0,2 or 3 #
1524 # the src can ONLY be a DENORM or an UNNORM! so, don't make any big subroutine
1597 # somewhere on the stack until we can copy it to its final resting place.
1599 # for fp0/fp1 which have already been restored. that way, we can write
1674 short tbl_fu_out - tbl_fu_out # BSUN can't happen
1675 short tbl_fu_out - tbl_fu_out # SNAN can't happen
1679 short tbl_fu_out - tbl_fu_out # DZ can't happen
1731 # underflow can happen for extended precision. extended precision opclass
1765 # somewhere on the stack until we can copy it to its final resting place
2197 # can store the default result where the exception frame was.
2237 # can store the default result where the exception frame was.
2277 # can store the default result where the exception frame was.
2431 # decode and emulate the instruction. No FP exceptions can be pending #
2432 # as a result of this operation emulation. A Trace exception can be #
2438 # fmovm_dynamic() can't handle this. A normal exit is made through #
2508 # which can all use f<op>.{x,p}
2619 # OPERR : all reg-reg or mem-reg operations that can normally operr
2958 # which can use most addressing modes.
2990 # the instruction is an fmovm.x dynamic which can use many addressing
2991 # modes and thus can have several different total instruction lengths.
3142 # denorm operand in the sgl or dbl format. NANs also become skewed, but can't
3164 # byte, word, and long destination format operations can pass
3169 # although packed opclass three operations can take operand error
3360 # byte, word, long, and packed destination format operations can pass
3663 # operations that can pass through fpsp_inex(). remember that fcmp and ftst
3725 # byte, word, and long destination format operations can pass
3726 # through here. so can double and single precision.
3727 # although packed opclass three operations can take inexact
4069 # (for the '060, only two types can take this exception)
5192 #--R' = SGN*R, S=R*R. THIS CAN BE REWRITTEN AS
5245 #--S=R*R AND S'=SGN*S. THIS CAN BE REWRITTEN AS
6380 #--FETCH F AND SAVING OF REGISTERS CAN BE ALL HIDED UNDER THE
6527 #--THIS CAN BE RE-WRITTEN AS
6840 # the comparisons in Steps 1.1 and 1.3 can be performed #
6884 # d) It is important to estimate how large can |R| be #
6942 # 6.4. Although one can argue that the inexact flag #
7002 # However, it is conceivable |X| can be small very often #
7081 # the best we can. For |X| >= 2^(-16312), the #
8127 # can be performed by a multiplication. #
8359 #--LOG(1+U) CAN BE VERY EFFICIENT.
8601 #--TAKEN BECAUSE 1+Z CAN HAVE 67 BITS OF INFORMATION AND WE MUST
9717 # The input operands X and Y can be either normalized or #
12051 # Norms can be emulated w/ a regular fmove instruction. For #
12825 # norms/denorms into ext/sgl/dbl precisions. Extended precision can be #
12859 # precision selected is extended. so...we can not get an underflow
13446 # Double and single precision can overflow and underflow. First, #
13482 # precision selected is extended. so...we can not get an underflow
13863 # If you have a 2 DENORMs, then you can just force the j-bit to a one
16597 # For the IEEE miscellaneous tests, all but fdbf and fdbt can set bsun. #
18192 # a lookup table such that we can determine the number of bytes #
18201 # so that the move can occur outside of this routine. This special #
19730 # LONG: can be either NORM or ZERO...
19743 # WORD: can be either NORM or ZERO...
19756 # BYTE: can be either NORM or ZERO...
19769 # SGL: can be either NORM, DENORM, ZERO, INF, QNAN or SNAN but not UNNORM
19771 # separate SNANs and DENORMs so they can be loaded w/ special care.
19772 # all others can simply be moved "in" using fmove.
19788 fmov.s (%a0), %fp0 # no, so can load it regular
19825 # -number can't fault #
19861 # -number can't fault #
19897 # -number can't fault #
19933 # -number can't fault #
20015 # -number can't fault #
20104 # -number can't fault #
20136 # -number can't fault #
20188 # it's b,w,l,s,d,x, or p in size. b,w,l can be stored to either a data #
20192 # For sgl or dbl precision, overflow or underflow can occur. If #
20204 # access error can be handled properly. #
20463 # NORMs(in range) can be stored out by a simple "fmov.s"
20464 # Unnormalized inputs can come through this point.
20701 # NORMs(in range) can be stored out by a simple "fmov.d"
20702 # Unnormalized inputs can come through this point.
21074 # According to the index value in d1 which can range from zero #
21359 # register update can be undone. #
21423 # register update can be undone. #
21629 # value in d0. The FP number can be DENORM or SNAN so we have to be #
21723 # to see if (threshold - exponent) is > 65 in which case we can
22759 # can inex1 also be set along with unfl and inex2???
22819 # can inex1 also be set along with unfl and inex2???
22854 # The default overflow result can be determined by the sign of #
23021 # Note: this operation can never overflow. #
23027 # Note: this operation can never overflow. #
23507 # k-factor can dictate either the total number of digits, #
23734 # can dictate either the total number of digits, if it is
23909 # to overflow. Only a negative iscale can cause this, so
23950 # since the input operand is a DENORM, we can't multiply it directly.
Indexes created Sat Feb 21 01:20:28 UTC 2026