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 // 10/25/00 Initial version
43 // 06/14/01 Changed cmp to an equivalent form
44 // 05/20/02 Cleaned up namespace and sf0 syntax
45 // 01/20/03 Improved performance and reduced code size
46 // 04/18/03 Eliminate possible WAW dependency warning
47 // 09/03/03 Improved performance
48 //==============================================================
51 //==============================================================
52 // float roundf(float x)
53 //==============================================================
55 // general input registers:
64 // floating-point registers:
73 // predicate registers used:
76 // Overview of operation
77 //==============================================================
78 // float roundf(float x)
79 // Return an integer value (represented as a float) that is x
80 // rounded to nearest integer, halfway cases rounded away from
82 // if x>0 result = trunc(x+0.5)
83 // if x<0 result = trunc(x-0.5)
85 //==============================================================
88 // if the exponent is > 1003e => 3F(true) = 63(decimal)
89 // we have a significand of 64 bits 1.63-bits.
90 // If we multiply by 2^63, we no longer have a fractional part
91 // So input is an integer value already.
94 // if the exponent is >= 10033 => 34(true) = 52(decimal)
96 // we have a significand of 53 bits 1.52-bits. (implicit 1)
97 // If we multiply by 2^52, we no longer have a fractional part
98 // So input is an integer value already.
101 // if the exponent is > 10016 => 17(true) = 23(decimal)
102 // we have a significand of 24 bits 1.23-bits. (implicit 1)
103 // If we multiply by 2^23, we no longer have a fractional part
104 // So input is an integer value already.
108 GLOBAL_LIBM_ENTRY(roundf)
111 getf.exp rSignexp = f8 // Get signexp, recompute if unorm
112 fcvt.fx.trunc.s1 fXtruncInt = f8 // Convert to int in significand
113 addl rBigexp = 0x10016, r0 // Set exponent at which is integer
116 mov rExpHalf = 0x0FFFE // Form sign and exponent of 0.5
117 fnorm.s1 fNormX = f8 // Normalize input
118 mov rExpMask = 0x1FFFF // Form exponent mask
123 setf.exp fHalf = rExpHalf // Form 0.5
124 fclass.m p7,p0 = f8, 0x0b // Test x unorm
131 fclass.m p6,p0 = f8, 0x1e3 // Test x natval, nan, inf
132 (p7) br.cond.spnt ROUND_UNORM // Branch if x unorm
137 // Return here from ROUND_UNORM
140 fcmp.lt.s1 p8,p9 = f8, f0 // Test if x < 0
141 (p6) br.cond.spnt ROUND_SPECIAL // Exit if x natval, nan, inf
147 fcvt.xf f8 = fXtruncInt // Pre-Result if 0.5 <= |x| < 2^23
153 and rExp = rSignexp, rExpMask // Get biased exponent
154 fmerge.s fInc = fNormX, f1 // Form increment if |rem| >= 0.5
160 cmp.lt p6,p0 = rExp, rExpHalf // Is |x| < 0.5?
161 cmp.ge p7,p0 = rExp, rBigexp // Is |x| >= 2^23?
162 cmp.lt p10,p0 = rExp, rExpHalf // Is |x| < 0.5?
166 // We must correct result if |x| < 0.5, or |x| >= 2^23
167 .pred.rel "mutex",p6,p7
170 (p6) fmerge.s f8 = fNormX, f0 // If |x| < 0.5, result sgn(x)*0
174 (p7) cmp.eq p10,p0 = r0, r0 // Also turn on p10 if |x| >= 2^23
175 (p7) fma.s.s0 f8 = fNormX, f1, f0 // If |x| >= 2^23, result x
176 (p10) br.ret.spnt b0 // Exit |x| < 0.5 or |x| >= 2^23
180 // Here if 0.5 <= |x| < 2^23
183 (p9) fms.s1 fRem = fNormX, f1, f8 // Get remainder = x - trunc(x)
188 (p8) fms.s1 fRem = f8, f1, fNormX // Get remainder = trunc(x) - x
195 fcmp.ge.s1 p9,p0 = fRem, fHalf // Test |rem| >= 0.5
200 // If x < 0 and remainder <= -0.5, then subtract 1 from result
201 // If x > 0 and remainder >= +0.5, then add 1 to result
204 (p9) fma.s.s0 f8 = f8, f1, fInc
211 // Here if x natval, nan, inf
214 fma.s.s0 f8 = f8, f1, f0
222 getf.exp rSignexp = fNormX // Get signexp, recompute if unorm
223 fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag
228 fcvt.fx.trunc.s1 fXtruncInt = fNormX // Convert to int in significand
229 br.cond.sptk ROUND_COMMON // Return to main path
233 GLOBAL_LIBM_END(roundf)