1.file "ceill.s"
2
3
4// Copyright (c) 2000 - 2003, Intel Corporation
5// All rights reserved.
6//
7//
8// Redistribution and use in source and binary forms, with or without
9// modification, are permitted provided that the following conditions are
10// met:
11//
12// * Redistributions of source code must retain the above copyright
13// notice, this list of conditions and the following disclaimer.
14//
15// * Redistributions in binary form must reproduce the above copyright
16// notice, this list of conditions and the following disclaimer in the
17// documentation and/or other materials provided with the distribution.
18//
19// * The name of Intel Corporation may not be used to endorse or promote
20// products derived from this software without specific prior written
21// permission.
22
23// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
27// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
28// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
31// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
32// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34//
35// Intel Corporation is the author of this code, and requests that all
36// problem reports or change requests be submitted to it directly at
37// http://www.intel.com/software/products/opensource/libraries/num.htm.
38//
39// History
40//==============================================================
41// 02/02/00 Initial version
42// 06/13/00 Improved speed
43// 06/27/00 Eliminated incorrect invalid flag setting
44// 05/20/02 Cleaned up namespace and sf0 syntax
45// 01/28/03 Improved performance
46//==============================================================
47
48// API
49//==============================================================
50// long double ceill(long double x)
51//==============================================================
52
53// general input registers:
54// r14 - r19
55
56rSignexp   = r14
57rExp       = r15
58rExpMask   = r16
59rBigexp    = r17
60rM1        = r18
61rSignexpM1 = r19
62
63// floating-point registers:
64// f8 - f13
65
66fXInt      = f9
67fNormX     = f10
68fTmp       = f11
69fAdj       = f12
70fPreResult = f13
71
72// predicate registers used:
73// p6 - p10
74
75// Overview of operation
76//==============================================================
77// long double ceill(long double x)
78// Return an integer value (represented as a long double) that is the smallest
79// value not less than x
80// This is x rounded toward +infinity to an integral value.
81// Inexact is set if x != ceill(x)
82//==============================================================
83
84// double_extended
85// if the exponent is > 1003e => 3F(true) = 63(decimal)
86// we have a significand of 64 bits 1.63-bits.
87// If we multiply by 2^63, we no longer have a fractional part
88// So input is an integer value already.
89
90// double
91// if the exponent is >= 10033 => 34(true) = 52(decimal)
92// 34 + 3ff = 433
93// we have a significand of 53 bits 1.52-bits. (implicit 1)
94// If we multiply by 2^52, we no longer have a fractional part
95// So input is an integer value already.
96
97// single
98// if the exponent is > 10016 => 17(true) = 23(decimal)
99// we have a significand of 24 bits 1.23-bits. (implicit 1)
100// If we multiply by 2^23, we no longer have a fractional part
101// So input is an integer value already.
102
103
104.section .text
105GLOBAL_LIBM_ENTRY(ceill)
106
107{ .mfi
108      getf.exp         rSignexp  = f8        // Get signexp, recompute if unorm
109      fclass.m         p7,p0 = f8, 0x0b      // Test x unorm
110      addl             rBigexp = 0x1003e, r0 // Set exponent at which is integer
111}
112{ .mfi
113      mov              rM1 = -1              // Set all ones
114      fcvt.fx.trunc.s1 fXInt  = f8           // Convert to int in significand
115      mov              rExpMask    = 0x1FFFF // Form exponent mask
116}
117;;
118
119{ .mfi
120      mov              rSignexpM1  = 0x2FFFF // Form signexp of -1
121      fcmp.lt.s1       p8,p9 = f8, f0        // Test x < 0
122      nop.i            0
123}
124{ .mfb
125      setf.sig         fTmp = rM1            // Make const for setting inexact
126      fnorm.s1         fNormX  = f8          // Normalize input
127(p7)  br.cond.spnt     CEIL_UNORM            // Branch if x unorm
128}
129;;
130
131CEIL_COMMON:
132// Return here from CEIL_UNORM
133{ .mfi
134      nop.m            0
135      fclass.m         p6,p0 = f8, 0x1e7     // Test x natval, nan, inf, 0
136      nop.i            0
137}
138;;
139
140.pred.rel "mutex",p8,p9
141{ .mfi
142      nop.m            0
143(p8)  fma.s1           fAdj = f0, f0, f0     // If x < 0, adjustment is 0
144      nop.i            0
145}
146{ .mfi
147      nop.m            0
148(p9)  fma.s1           fAdj = f1, f1, f0     // If x > 0, adjustment is +1
149      nop.i            0
150}
151;;
152
153{ .mfi
154      nop.m            0
155      fcvt.xf          fPreResult = fXInt    // trunc(x)
156      nop.i            0
157}
158{ .mfb
159      nop.m            0
160(p6)  fma.s0           f8 = f8, f1, f0       // Result if x natval, nan, inf, 0
161(p6)  br.ret.spnt      b0                    // Exit if x natval, nan, inf, 0
162}
163;;
164
165{ .mmi
166      and              rExp = rSignexp, rExpMask // Get biased exponent
167;;
168      cmp.ge           p7,p6 = rExp, rBigexp  // Is |x| >= 2^63?
169(p8)  cmp.lt.unc       p10,p0 = rSignexp, rSignexpM1 // Is -1 < x < 0?
170}
171;;
172
173// If -1 < x < 0, we turn off p6 and compute result as -0
174{ .mfi
175(p10) cmp.ne           p6,p0 = r0,r0
176(p10) fmerge.s         f8 = fNormX, f0
177      nop.i            0
178}
179;;
180
181.pred.rel "mutex",p6,p7
182{ .mfi
183      nop.m            0
184(p6)  fma.s0           f8 = fPreResult, f1, fAdj // Result if !int, |x| < 2^63
185      nop.i            0
186}
187{ .mfi
188      nop.m            0
189(p7)  fma.s0           f8 = fNormX, f1, f0    // Result, if |x| >= 2^63
190(p10) cmp.eq           p6,p0 = r0,r0          // If -1 < x < 0, turn on p6 again
191}
192;;
193
194{ .mfi
195      nop.m            0
196(p6)  fcmp.eq.unc.s1   p8, p9 = fPreResult, fNormX // Is trunc(x) = x ?
197      nop.i            0
198}
199;;
200
201{ .mfi
202      nop.m            0
203(p9)  fmpy.s0          fTmp = fTmp, fTmp      // Dummy to set inexact
204      nop.i            0
205}
206{ .mfb
207      nop.m            0
208(p8)  fma.s0           f8 = fNormX, f1, f0    // If x int, result normalized x
209      br.ret.sptk      b0                     // Exit main path, 0 < |x| < 2^63
210}
211;;
212
213
214CEIL_UNORM:
215// Here if x unorm
216{ .mfb
217      getf.exp         rSignexp  = fNormX     // Get signexp, recompute if unorm
218      fcmp.eq.s0       p7,p0 = f8, f0         // Dummy op to set denormal flag
219      br.cond.sptk     CEIL_COMMON            // Return to main path
220}
221;;
222
223GLOBAL_LIBM_END(ceill)
224libm_alias_ldouble_other (ceil, ceil)
225