3 // Copyright (C) 2000, 2001, Intel Corporation
4 // All rights reserved.
6 // Contributed 2/2/2000 by John Harrison, Ted Kubaska, Bob Norin, Shane Story,
7 // and Ping Tak Peter Tang of the Computational Software Lab, 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://developer.intel.com/opensource.
41 //==============================================================
42 // 2/02/00: Initial version
43 // 5/24/00 Fixed case of 2^63 - 1 + 0.5 (0x1007dffffffffffffffff)
44 // 2/08/01 Corrected behavior for all rounding modes.
47 //==============================================================
48 // long double rintl(long double x)
50 #include "libm_support.h"
53 // general registers used:
57 rint_GR_exponent = r16
62 rint_GR_rcs0_mask = r21
65 // predicate registers used:
68 // floating-point registers used:
73 RINT_FLOAT_INT_f8 = f12
75 RINT_SIGNED_FLOAT_INT_f8 = f14
77 // Overview of operation
78 //==============================================================
80 // long double rintl(long double x)
81 // Return an integer value (represented as a long double) that is x rounded to integer in current
83 // Inexact is set if x != rintl(x)
84 // *******************************************************************************
86 // Set denormal flag for denormal input and
87 // and take denormal fault if necessary.
89 // Is the input an integer value already?
92 // if the exponent is >= 1003e => 3F(true) = 63(decimal)
93 // we have a significand of 64 bits 1.63-bits.
94 // If we multiply by 2^63, we no longer have a fractional part
95 // So input is an integer value already.
98 // if the exponent is >= 10033 => 34(true) = 52(decimal)
100 // we have a significand of 53 bits 1.52-bits. (implicit 1)
101 // If we multiply by 2^52, we no longer have a fractional part
102 // So input is an integer value already.
105 // if the exponent is >= 10016 => 17(true) = 23(decimal)
106 // we have a significand of 53 bits 1.52-bits. (implicit 1)
107 // If we multiply by 2^52, we no longer have a fractional part
108 // So input is an integer value already.
110 // If x is NAN, ZERO, or INFINITY, then return
112 // qnan snan inf norm unorm 0 -+
113 // 1 1 1 0 0 1 11 0xe7
127 .type __rintl,@function
132 mov rint_GR_fpsr = ar40 // Read the fpsr--need to check rc.s0
133 fcvt.fx.s1 RINT_INT_f8 = f8
134 addl rint_GR_10033 = 0x1003e, r0
137 mov rint_GR_FFFF = -1
138 fnorm.s1 RINT_NORM_f8 = f8
139 mov rint_GR_17ones = 0x1FFFF
144 setf.sig RINT_FFFF = rint_GR_FFFF
145 fclass.m.unc p6,p0 = f8, 0xe7
146 mov rint_GR_rcs0_mask = 0x0c00
153 (p6) br.ret.spnt b0 // Exit if x nan, inf, zero
159 fcvt.xf RINT_FLOAT_INT_f8 = RINT_INT_f8
165 getf.exp rint_GR_signexp = RINT_NORM_f8
166 fcmp.eq.s0 p8,p0 = f8,f0 // Dummy op to set denormal
175 and rint_GR_exponent = rint_GR_signexp, rint_GR_17ones
180 cmp.ge.unc p7,p6 = rint_GR_exponent, rint_GR_10033
181 and rint_GR_rcs0 = rint_GR_rcs0_mask, rint_GR_fpsr
186 // Check to see if s0 rounding mode is round to nearest. If not then set s2
187 // rounding mode to that of s0 and repeat conversions.
188 // Must merge the original sign for cases where the result is zero or the input
189 // is the largest that still has a fraction (0x1007dfffffffffff)
192 cmp.ne p11,p0 = rint_GR_rcs0, r0
193 (p6) fmerge.s RINT_SIGNED_FLOAT_INT_f8 = f8, RINT_FLOAT_INT_f8
194 (p11) br.cond.spnt L(RINT_NOT_ROUND_NEAREST) // Branch if not round to nearest
200 (p6) fcmp.eq.unc.s1 p0,p8 = RINT_FLOAT_INT_f8, RINT_NORM_f8
205 (p7) fnorm.s0 f8 = f8
212 (p6) fnorm f8 = RINT_SIGNED_FLOAT_INT_f8
219 (p8) fmpy.s0 RINT_INEXACT = RINT_FFFF,RINT_FFFF // Dummy to set inexact
224 L(RINT_NOT_ROUND_NEAREST):
225 // Set rounding mode of s2 to that of s0
227 mov rint_GR_rcs0 = r0 // Clear so we don't come back here
235 fcvt.fx.s2 RINT_INT_f8 = f8
242 fcvt.xf RINT_FLOAT_INT_f8 = RINT_INT_f8
243 br.cond.sptk L(RINT_COMMON)
249 ASM_SIZE_DIRECTIVE(rintl)
251 ASM_SIZE_DIRECTIVE(__rintl)