Lines Matching refs:WE
5084 #--THEN WE RETURN SGN*SIN(R). SGN*SIN(R) IS COMPUTED BY
5137 #--THEN WE RETURN SGN*COS(R). SGN*COS(R) IS COMPUTED BY
5201 #--IF |X| > 15PI, WE USE THE GENERAL ARGUMENT REDUCTION.
5544 #--WE MUST NOW FIND INT(FP2). SINCE WE NEED THIS VALUE IN
5573 #--We are now ready to perform (R+r) - N*P1 - N*P2, P1 = 2**(L) * Piby2_1 and
5581 #--we want P+p = W+w but |p| <= half ulp of P
5582 #--Then, we need to compute A := R-P and a := r-p
5592 #--Now we need to normalize (A,a) to "new (R,r)" where R+r = A+a but
5885 #--IF |X| > 15PI, WE USE THE GENERAL ARGUMENT REDUCTION.
5975 #--WE MUST NOW FIND INT(FP2). SINCE WE NEED THIS VALUE IN
6004 #--We are now ready to perform (R+r) - N*P1 - N*P2, P1 = 2**(L) * Piby2_1 and
6012 #--we want P+p = W+w but |p| <= half ulp of P
6013 #--Then, we need to compute A := R-P and a := r-p
6023 #--Now we need to normalize (A,a) to "new (R,r)" where R+r = A+a but
6267 #--THE MOST LIKELY CASE, |X| IN [1/16, 16). WE USE TABLE TECHNIQUE
6270 #--A TABLE, ALL WE NEED IS TO APPROXIMATE ATAN(U) WHERE
6281 #--NOW WE SEE X AS +-2^K * 1.BBBBBBB....B <- 1. + 63 BITS
6282 #--WE CHOSE F TO BE +-2^K * 1.BBBB1
6286 #-- -ATAN(|F|), WE NEED TO STORE ONLY ATAN(|F|).
6300 #--WHILE THE DIVISION IS TAKING ITS TIME, WE FETCH ATAN(|F|)
6325 #--U IN FP0, WE ARE NOW READY TO COMPUTE ATAN(U) AS
6329 #--WHAT WE HAVE HERE IS MERELY A1 = A3, A2 = A1/A3, A3 = A2/A3.
6826 # Notes: If AdjFlag = 0, we have X = Mlog2 + Jlog2/64 + R, #
6846 # Note also that we use the FMOVEM instruction to move X #
6974 # inefficiency, we stay away from denormalized numbers #
6975 # the best we can. For |X| >= 2^(-16312), the #
7160 #--WE NOW COMPUTE EXP(R)-1 BY A POLYNOMIAL
7162 #--TO FULLY USE THE PIPELINE, WE COMPUTE S = R*R
7323 #--WE NOW COMPUTE EXP(R)-1 BY A POLYNOMIAL
7325 #--TO FULLY USE THE PIPELINE, WE COMPUTE S = R*R
7552 # here, we build the result in a tmp location so as not to disturb the input
7605 # overflow. Thus, we calculate sinh(X) as follows: #
7719 # Thus, we calculate sinh(X) as follows: #
7842 # 6. (|X| >= 50 log2) Tanh(X) = +-1 (round to nearest). Thus, we #
8249 #--WE DEFINE F = 1.XXXXXX1, I.E. FIRST 7 BITS OF Y AND ATTACH A 1.
8265 #--WHILE THE CONVERSION IS GOING ON, WE GET F AND ADDRESS OF 1/F
8494 #--HERE WE USE THE USUAL TABLE DRIVEN APPROACH. CARE HAS TO BE
8495 #--TAKEN BECAUSE 1+Z CAN HAVE 67 BITS OF INFORMATION AND WE MUST
8500 #--ON RETURNING TO LP1CONT1, WE MUST HAVE K IN FP1, ADDRESS OF
9251 # we may need to read the fpsr for the last fp op in t_catch2().
9276 # the multiply factor that we're trying to create should be a denorm
9277 # for the multiply to work. therefore, we're going to actually do a
9280 # later. we don't do this with the DENORMs above because this method
9284 subi.l &-0x3fff,%d0 # how many should we shift?
9332 # The source input is below 1, so we check for denormalized numbers
10034 # For t_ovfl_sc() we take special care not to lose the INEX2 bit. #
10046 # dst op is a DENORM. we have to normalize the mantissa to see if the
10048 # dst so we don't screw up the version passed to us.
10132 # occurred and has been logged. Now we need to see if an inexact #
10165 # 060FPLSP should be inexact. So, if inexact is enabled, then we create #
10168 # Must check, too, if the result was zero, in which case we just #
10248 # values are already in fp0 and fp1 so we do nothing here.
Indexes created Sat Oct 25 10:09:55 GMT 2025