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 // 2/16/00 Modified to conform to C9X
44 // 3/16/00 Improved speed
45 // 4/04/00 Unwind support added
46 // 5/30/00 Fixed bug when x double-extended denormal
47 // 8/15/00 Bundle added after call to __libm_error_support to properly
48 // set [the previously overwritten] GR_Parameter_RESULT.
51 //==============================================================
52 // float logbf( float x);
54 // Overview of operation
55 //==============================================================
56 // The logbf function extracts the exponent of x as an integer in
57 // floating-point format.
58 // logbf computes log2 of x as a float
60 // logbf is similar to ilogbf but differs in the following ways:
62 // ilogbf: returns INT_MAX
63 // logbf: returns +inf
64 // Nan returns FP_LOGBNAN (which is either INT_MAX or INT_MIN)
65 // ilogbf: returns INT_MAX (7fffffff)
66 // logbf: returns QNAN (quietized SNAN)
67 // 0 returns FP_ILOGB0 (which is either INT_MIN or -INT_MAX)
68 // ilogbf: returns -INT_MAX (80000001)
69 // logbf: returns -inf, raises the divide-by-zero exception,
70 // and calls libm_error_support to set domain error
73 //==============================================================
74 // general registers used:
77 // r38 -> r41 used as parameters to error path
79 // predicate registers used:
82 // floating-point registers used:
86 #include "libm_support.h"
89 // r40 is address of table of coefficients
95 GR_Parameter_RESULT = r40
96 GR_Parameter_TAG = r41
113 // qnan snan inf norm unorm 0 -+
117 alloc r32=ar.pfs,1,5,4,0
118 (p0) fclass.m.unc p8,p0 = f8, 0x0b
122 // r37 = exp(f8) - - 0xffff
124 // f8 = convert_to_fp (sig))
126 (p0) getf.exp r35 = f8
131 // qnan snan inf norm unorm 0 -+
135 (p0) mov r33 = 0xffff
136 (p0) mov r34 = 0x1ffff
137 (p0) fclass.m.unc p6,p0 = f8, 0xe3 ;;
141 (p0) and r36 = r35, r34
142 (p0) fclass.m.unc p7,p0 = f8, 0x07
143 (p8) br.cond.spnt L(LOGB_DENORM) ;;
147 (p0) sub r37 = r36, r33
149 (p6) br.cond.spnt L(LOGB_NAN_INF) ;;
153 (p0) setf.sig f9 = r37
155 (p7) br.cond.spnt L(LOGB_ZERO) ;;
160 (p0) fcvt.xf f10 = f9
166 (p0) fnorm.s f8 = f10
167 (p0) br.ret.sptk b0 ;;
171 // Form signexp of 2^64 in case need to scale denormal
172 // Check to see if double-extended denormal
174 (p0) mov r38 = 0x1003f
175 (p0) fclass.m.unc p8,p0 = f10, 0x0b
179 // Form 2^64 in case need to scale denormal
181 (p0) setf.exp f11 = r38
186 // If double-extended denormal add 64 to exponent bias for scaling
187 // If double-extended denormal form x * 2^64 which is normal
189 (p8) add r33 = 64, r33
190 (p8) fmpy f10 = f10, f11
194 // Logic is the same as normal path but use normalized input
196 (p0) getf.exp r35 = f10 ;;
202 (p0) and r36 = r35, r34 ;;
203 (p0) sub r37 = r36, r33
208 (p0) setf.sig f9 = r37
215 (p0) fcvt.xf f10 = f9
221 (p0) fnorm.s f8 = f10
222 (p0) br.ret.sptk b0 ;;
227 // X NAN or INFINITY, return f8 * f8
230 (p0) fma.s f8= f8,f8,f0
231 (p0) br.ret.sptk b0 ;;
237 // return -1.0/fabs(f8)=-inf, set divide-by-zero flag, call error support
240 (p0) fmerge.s f9 = f0,f8
246 (p0) fmerge.ns f10 = f0,f9
252 (p0) frcpa f10,p6 = f1,f10
257 ASM_SIZE_DIRECTIVE(logbf)
260 .proc __libm_error_region
264 add GR_Parameter_Y=-32,sp // Parameter 2 value
265 (p0) mov GR_Parameter_TAG = 152
266 .save ar.pfs,GR_SAVE_PFS
267 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
271 add sp=-64,sp // Create new stack
273 mov GR_SAVE_GP=gp // Save gp
276 stfs [GR_Parameter_Y] = FR_Y,16 // Store Parameter 2 on stack
277 add GR_Parameter_X = 16,sp // Parameter 1 address
279 mov GR_SAVE_B0=b0 // Save b0
283 stfs [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
284 add GR_Parameter_RESULT = 0,GR_Parameter_Y
285 nop.b 0 // Parameter 3 address
288 stfs [GR_Parameter_Y] = FR_RESULT // Store Parameter 3 on stack
289 add GR_Parameter_Y = -16,GR_Parameter_Y
290 br.call.sptk b0=__libm_error_support# // Call error handling function
295 add GR_Parameter_RESULT = 48,sp
298 ldfs f8 = [GR_Parameter_RESULT] // Get return result off stack
300 add sp = 64,sp // Restore stack pointer
301 mov b0 = GR_SAVE_B0 // Restore return address
304 mov gp = GR_SAVE_GP // Restore gp
305 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
306 br.ret.sptk b0 // Return
309 .endp __libm_error_region
310 ASM_SIZE_DIRECTIVE(__libm_error_region)
313 .type __libm_error_support#,@function
314 .global __libm_error_support#