4 // Copyright (c) 2000 - 2003, Intel Corporation
5 // All rights reserved.
7 // Contributed 2000 by the Intel Numerics Group, Intel Corporation
9 // Redistribution and use in source and binary forms, with or without
10 // modification, are permitted provided that the following conditions are
13 // * Redistributions of source code must retain the above copyright
14 // notice, this list of conditions and the following disclaimer.
16 // * Redistributions in binary form must reproduce the above copyright
17 // notice, this list of conditions and the following disclaimer in the
18 // documentation and/or other materials provided with the distribution.
20 // * The name of Intel Corporation may not be used to endorse or promote
21 // products derived from this software without specific prior written
24 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
27 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
28 // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
29 // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
30 // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
31 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
32 // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
33 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 // Intel Corporation is the author of this code, and requests that all
37 // problem reports or change requests be submitted to it directly at
38 // http://www.intel.com/software/products/opensource/libraries/num.htm.
41 //==============================================================
42 // 02/02/00 Initial version
43 // 03/22/00 Updated to improve performance
44 // 06/13/00 Improved speed, fixed setting of inexact flag
45 // 06/27/00 Eliminated incorrect invalid flag setting
46 // 02/07/01 Corrected sign of zero result in round to -inf mode
47 // 05/20/02 Cleaned up namespace and sf0 syntax
48 // 01/28/03 Improved performance
49 //==============================================================
52 //==============================================================
53 // double floor(double x)
54 //==============================================================
56 // general input registers:
65 // floating-point registers:
74 // predicate registers used:
77 // Overview of operation
78 //==============================================================
79 // double floor(double x)
80 // Return an integer value (represented as a double) that is the largest
81 // value not greater than x
82 // This is x rounded toward -infinity to an integral value.
83 // Inexact is set if x != floor(x)
84 //==============================================================
87 // if the exponent is > 1003e => 3F(true) = 63(decimal)
88 // we have a significand of 64 bits 1.63-bits.
89 // If we multiply by 2^63, we no longer have a fractional part
90 // So input is an integer value already.
93 // if the exponent is >= 10033 => 34(true) = 52(decimal)
95 // we have a significand of 53 bits 1.52-bits. (implicit 1)
96 // If we multiply by 2^52, we no longer have a fractional part
97 // So input is an integer value already.
100 // if the exponent is > 10016 => 17(true) = 23(decimal)
101 // we have a significand of 24 bits 1.23-bits. (implicit 1)
102 // If we multiply by 2^23, we no longer have a fractional part
103 // So input is an integer value already.
107 GLOBAL_IEEE754_ENTRY(floor)
110 getf.exp rSignexp = f8 // Get signexp, recompute if unorm
111 fclass.m p7,p0 = f8, 0x0b // Test x unorm
112 addl rBigexp = 0x10033, r0 // Set exponent at which is integer
115 mov rM1 = -1 // Set all ones
116 fcvt.fx.trunc.s1 fXInt = f8 // Convert to int in significand
117 mov rExpMask = 0x1FFFF // Form exponent mask
123 fcmp.lt.s1 p8,p9 = f8, f0 // Test x < 0
127 setf.sig fTmp = rM1 // Make const for setting inexact
128 fnorm.s1 fNormX = f8 // Normalize input
129 (p7) br.cond.spnt FLOOR_UNORM // Branch if x unorm
134 // Return here from FLOOR_UNORM
137 fclass.m p6,p0 = f8, 0x1e7 // Test x natval, nan, inf, 0
142 .pred.rel "mutex",p8,p9
145 (p8) fnma.s1 fAdj = f1, f1, f0 // If x < 0, adjustment is -1
150 (p9) fma.s1 fAdj = f0, f0, f0 // If x > 0, adjustment is 0
157 fcvt.xf fPreResult = fXInt // trunc(x)
162 (p6) fma.d.s0 f8 = f8, f1, f0 // Result if x natval, nan, inf, 0
163 (p6) br.ret.spnt b0 // Exit if x natval, nan, inf, 0
168 and rExp = rSignexp, rExpMask // Get biased exponent
170 cmp.ge p7,p6 = rExp, rBigexp // Is |x| >= 2^52?
177 (p6) fma.d.s0 f8 = fPreResult, f1, fAdj // Result if !int, |x| < 2^52
182 (p7) fma.d.s0 f8 = fNormX, f1, f0 // Result, if |x| >= 2^52
189 (p6) fcmp.eq.unc.s1 p8, p9 = fPreResult, fNormX // Is trunc(x) = x ?
196 (p9) fmpy.s0 fTmp = fTmp, fTmp // Dummy to set inexact
201 (p8) fma.d.s0 f8 = fNormX, f1, f0 // If x int, result normalized x
202 br.ret.sptk b0 // Exit main path, 0 < |x| < 2^52
210 getf.exp rSignexp = fNormX // Get signexp, recompute if unorm
211 fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag
212 br.cond.sptk FLOOR_COMMON // Return to main path
216 GLOBAL_IEEE754_END(floor)