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
// 02/08/01 Corrected behavior for all rounding modes.
44
// 05/20/02 Cleaned up namespace and sf0 syntax
45
// 01/20/03 Improved performance
46
//==============================================================
49
//==============================================================
50
// float rintf(float x)
51
//==============================================================
53
// general input registers:
65
// floating-point registers:
72
// predicate registers used:
75
// Overview of operation
76
//==============================================================
77
// float rintf(float x)
78
// Return an integer value (represented as a float) that is x
79
// rounded to integer in current rounding mode
80
// Inexact is set if x != rint(x)
81
//==============================================================
84
// if the exponent is > 1003e => 3F(true) = 63(decimal)
85
// we have a significand of 64 bits 1.63-bits.
86
// If we multiply by 2^63, we no longer have a fractional part
87
// So input is an integer value already.
90
// if the exponent is >= 10033 => 34(true) = 52(decimal)
92
// we have a significand of 53 bits 1.52-bits. (implicit 1)
93
// If we multiply by 2^52, we no longer have a fractional part
94
// So input is an integer value already.
97
// if the exponent is > 10016 => 17(true) = 23(decimal)
98
// we have a significand of 24 bits 1.23-bits. (implicit 1)
99
// If we multiply by 2^23, we no longer have a fractional part
100
// So input is an integer value already.
103
GLOBAL_IEEE754_ENTRY(rintf)
106
getf.exp rSignexp = f8 // Get signexp, recompute if unorm
107
fclass.m p7,p0 = f8, 0x0b // Test x unorm
108
addl rBigexp = 0x10016, r0 // Set exponent at which is integer
111
mov rM1 = -1 // Set all ones
112
fcvt.fx.s1 fXInt = f8 // Convert to int in significand
113
mov rExpMask = 0x1FFFF // Form exponent mask
118
mov rFpsr = ar40 // Read fpsr -- check rc.s0
119
fclass.m p6,p0 = f8, 0x1e3 // Test x natval, nan, inf
123
setf.sig fTmp = rM1 // Make const for setting inexact
124
fnorm.s1 fNormX = f8 // Normalize input
125
(p7) br.cond.spnt RINT_UNORM // Branch if x unorm
131
// Return here from RINT_UNORM
133
and rExp = rSignexp, rExpMask // Get biased exponent
134
(p6) fma.s.s0 f8 = f8, f1, f0 // Result if x natval, nan, inf
135
(p6) br.ret.spnt b0 // Exit if x natval, nan, inf
140
mov rRcs0Mask = 0x0c00 // Mask for rc.s0
141
fcvt.xf f8 = fXInt // Result assume |x| < 2^23
142
cmp.ge p7,p8 = rExp, rBigexp // Is |x| >= 2^23?
146
// We must correct result if |x| >= 2^23
149
(p7) fma.s.s0 f8 = fNormX, f1, f0 // If |x| >= 2^23, result x
156
fcmp.eq.unc.s1 p0, p9 = f8, fNormX // Is result = x ?
161
(p8) fmerge.s f8 = fNormX, f8 // Make sure sign rint(x) = sign x
167
(p8) and rRcs0 = rFpsr, rRcs0Mask // Get rounding mode for sf0
173
// If |x| < 2^23 we must test for other rounding modes
175
(p8) cmp.ne.unc p10,p0 = rRcs0, r0 // Test for other rounding modes
176
(p9) fmpy.s0 fTmp = fTmp, fTmp // Dummy to set inexact
181
(p10) br.cond.spnt RINT_NOT_ROUND_NEAREST // Branch if not round nearest
182
br.ret.sptk b0 // Exit main path if round nearest
191
getf.exp rSignexp = fNormX // Get signexp, recompute if unorm
192
fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag
193
br.cond.sptk RINT_COMMON // Return to main path
197
RINT_NOT_ROUND_NEAREST:
198
// Here if not round to nearest, and |x| < 2^23
199
// Set rounding mode of s2 to that of s0, and repeat the conversion using s2
209
fcvt.fx.s2 fXInt = fNormX // Convert to int in significand
216
fcvt.xf f8 = fXInt // Expected result
221
// Be sure sign of result = sign of input. Fixes cases where result is 0.
224
fmerge.s f8 = fNormX, f8
225
br.ret.sptk b0 // Exit main path
229
GLOBAL_IEEE754_END(rintf)