Lines Matching refs:THE
9 THE SOFTWARE is provided on an "AS IS" basis and without warranty.
20 ARISING OF THE USE OR INABILITY TO USE THE SOFTWARE.
5045 #--THIS IS THE USUAL CASE, |X| <= 15 PI.
5046 #--THE ARGUMENT REDUCTION IS DONE BY TABLE LOOK UP.
5059 # A1 IS THE ADDRESS OF N*PIBY2
5074 #--LET J BE THE LEAST SIG. BIT OF D0, LET SGN := (-1)**J.
5127 #--LET J BE THE LEAST SIG. BIT OF D0, LET SGN := (-1)**J.
5192 #--IF |X| > 15PI, WE USE THE GENERAL ARGUMENT REDUCTION.
5256 #--THIS IS THE USUAL CASE, |X| <= 15 PI.
5257 #--THE ARGUMENT REDUCTION IS DONE BY TABLE LOOK UP.
5462 #--WHEN REDUCEX IS USED, THE CODE WILL INEVITABLY BE SLOW.
5464 #--THE REMAINDER INSTRUCTION WHICH IS NOW IN SOFTWARE.
5519 #--FIND THE REMAINDER OF (R,r) W.R.T. 2**L * (PI/2). L IS SO CHOSEN
5536 #--FLOATING POINT FORMAT, THE TWO FMOVE'S FMOVE.L FP <--> N
5537 #--WILL BE TOO INEFFICIENT. THE WAY AROUND IT IS THAT
5539 #--US THE DESIRED VALUE IN FLOATING POINT.
5545 fadd.s TWOTO63(%a6),%fp2 # THE FRACTIONAL PART OF FP1 IS ROUNDED
5775 #--THIS IS THE USUAL CASE, |X| <= 15 PI.
5776 #--THE ARGUMENT REDUCTION IS DONE BY TABLE LOOK UP.
5876 #--IF |X| > 15PI, WE USE THE GENERAL ARGUMENT REDUCTION.
5893 #--WHEN REDUCEX IS USED, THE CODE WILL INEVITABLY BE SLOW.
5895 #--THE REMAINDER INSTRUCTION WHICH IS NOW IN SOFTWARE.
5950 #--FIND THE REMAINDER OF (R,r) W.R.T. 2**L * (PI/2). L IS SO CHOSEN
5967 #--FLOATING POINT FORMAT, THE TWO FMOVE'S FMOVE.L FP <--> N
5968 #--WILL BE TOO INEFFICIENT. THE WAY AROUND IT IS THAT
5970 #--US THE DESIRED VALUE IN FLOATING POINT.
5976 fadd.s TWOTO63(%a6),%fp2 # THE FRACTIONAL PART OF FP1 IS ROUNDED
6258 #--THE MOST LIKELY CASE, |X| IN [1/16, 16). WE USE TABLE TECHNIQUE
6259 #--THE IDEA IS ATAN(X) = ATAN(F) + ATAN( [X-F] / [1+XF] ).
6263 #--TRUE THAT A DIVIDE IS NOW NEEDED, BUT THE APPROXIMATION FOR
6264 #--ATAN(U) IS A VERY SHORT POLYNOMIAL AND THE INDEXING TO
6265 #--FETCH F AND SAVING OF REGISTERS CAN BE ALL HIDED UNDER THE
6266 #--DIVIDE. IN THE END THIS METHOD IS MUCH FASTER THAN A TRADITIONAL
6267 #--ONE. NOTE ALSO THAT THE TRADITIONAL SCHEME THAT APPROXIMATE
6274 #--THAT IS IT MATCHES THE EXPONENT AND FIRST 5 BITS OF X, THE
6291 #--WHILE THE DIVISION IS TAKING ITS TIME, WE FETCH ATAN(|F|)
6296 mov.l %d1,%d2 # THE EXP AND 16 BITS OF X
6301 add.l %d2,%d1 # THE 7 BITS IDENTIFYING F
6314 #--BUT ALAS, THE DIVIDE IS STILL CRANKING!
6318 #--THE POLYNOMIAL MAY LOOK STRANGE, BUT IS NEVERTHELESS CORRECT.
6319 #--THE NATURAL FORM IS U + U*V*(A1 + V*(A2 + V*A3))
6321 #--THE REASON FOR THIS REARRANGEMENT IS TO MAKE THE INDEPENDENT
6540 #--THIS IS THE USUAL CASE, |X| < 1
6633 #--THIS IS THE USUAL CASE, |X| < 1
7153 #--TO FULLY UTILIZE THE PIPELINE, WE COMPUTE S = R*R
7316 #--TO FULLY UTILIZE THE PIPELINE, WE COMPUTE S = R*R
7623 #--THIS IS THE USUAL CASE, |X| < 16380 LOG2
7736 #--THIS IS THE USUAL CASE, |X| < 16380 LOG2
7864 #--THIS IS THE USUAL CASE
8237 #--THIS SHOULD BE THE USUAL CASE, X NOT VERY CLOSE TO 1
8241 #--THE IDEA IS THAT LOG(X) = K*LOG2 + LOG(Y)
8245 #--ALSO NOTE THAT THE VALUE 1/F IS STORED IN A TABLE SO THAT NO
8256 #--WHILE THE CONVERSION IS GOING ON, WE GET F AND ADDRESS OF 1/F
8260 or.l &0x01000000,FFRAC(%a6) # GET F: ATTACH A 1 AT THE EIGHTH BIT
8265 asr.l &4,%d1 # SHIFTED 20, D0 IS THE DISPLACEMENT
8266 add.l %d1,%a0 # A0 IS THE ADDRESS FOR 1/F
8327 #--REGISTERS SAVED: FPCR, FP1. FP0 CONTAINS THE INPUT.
8485 #--HERE WE USE THE USUAL TABLE DRIVEN APPROACH. CARE HAS TO BE
8487 #--PRESERVE ALL THE INFORMATION. BECAUSE 1+Z IS IN [1/2,3/2],
8607 #--THIS IS THE USUAL CASE, |X| < 1
9022 #--SUMMARY: a1 IS ADDRESS FOR THE LEADING PORTION OF 2^(J/64),
9119 #--SUMMARY: a1 IS ADDRESS FOR THE LEADING PORTION OF 2^(J/64),
9150 #--FP0 IS R. THE FOLLOWING CODE COMPUTES