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 // double logb( double x);
54 // Overview of operation
55 //==============================================================
56 // The logb function extracts the exponent of x as an integer in
57 // floating-point format.
58 // logb computes log2 of x as a double
60 // logb is similar to ilogb but differs in the following ways:
62 // ilogb: returns INT_MAX
64 // Nan returns FP_LOGBNAN (which is either INT_MAX or INT_MIN)
65 // ilogb: returns INT_MAX (7fffffff)
66 // logb: returns QNAN (quietized SNAN)
67 // 0 returns FP_ILOGB0 (which is either INT_MIN or -INT_MAX)
68 // ilogb: returns -INT_MAX (80000001)
69 // logb: 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:
81 // floating-point registers used:
85 #include "libm_support.h"
93 GR_Parameter_RESULT = r40
105 // qnan snan inf norm unorm 0 -+
109 alloc r32=ar.pfs,1,5,4,0
110 (p0) fclass.m.unc p8,p0 = f8, 0x0b
114 // r37 = exp(f8) - - 0xffff
116 // f8 = convert_to_fp (sig))
118 (p0) getf.exp r35 = f8
123 // qnan snan inf norm unorm 0 -+
127 (p0) mov r33 = 0xffff
128 (p0) mov r34 = 0x1ffff
129 (p0) fclass.m.unc p6,p0 = f8, 0xe3 ;;
133 (p0) and r36 = r35, r34
134 (p0) fclass.m.unc p7,p0 = f8, 0x07
135 (p8) br.cond.spnt L(LOGB_DENORM) ;;
139 (p0) sub r37 = r36, r33
141 (p6) br.cond.spnt L(LOGB_NAN_INF) ;;
145 (p0) setf.sig f9 = r37
147 (p7) br.cond.spnt L(LOGB_ZERO) ;;
152 (p0) fcvt.xf f10 = f9
158 (p0) fnorm.d f8 = f10
159 (p0) br.ret.sptk b0 ;;
163 // Form signexp of 2^64 in case need to scale denormal
164 // Check to see if double-extended denormal
166 (p0) mov r38 = 0x1003f
167 (p0) fclass.m.unc p8,p0 = f10, 0x0b
171 // Form 2^64 in case need to scale denormal
173 (p0) setf.exp f11 = r38
178 // If double-extended denormal add 64 to exponent bias for scaling
179 // If double-extended denormal form x * 2^64 which is normal
181 (p8) add r33 = 64, r33
182 (p8) fmpy f10 = f10, f11
186 // Logic is the same as normal path but use normalized input
188 (p0) getf.exp r35 = f10 ;;
194 (p0) and r36 = r35, r34 ;;
195 (p0) sub r37 = r36, r33
200 (p0) setf.sig f9 = r37
207 (p0) fcvt.xf f10 = f9
213 (p0) fnorm.d f8 = f10
214 (p0) br.ret.sptk b0 ;;
219 // X NAN or INFINITY, return f8 * f8
222 (p0) fma.d f8= f8,f8,f0
223 (p0) br.ret.sptk b0 ;;
227 ASM_SIZE_DIRECTIVE(logb)
229 // Stack operations when calling error support.
230 // (1) (2) (3) (call) (4)
231 // sp -> + psp -> + psp -> + sp -> +
233 // | | <- GR_Y R3 ->| <- GR_RESULT | -> f8
235 // | <-GR_Y Y2->| Y2 ->| <- GR_Y |
237 // | | <- GR_X X1 ->| |
239 // sp-64 -> + sp -> + sp -> + +
240 // save ar.pfs save b0 restore gp
241 // save gp restore ar.pfs
245 .proc __libm_error_region
252 // f9 = 1.0/f10 = -1.0/-|f8|
255 mov r41 = 151 // Error code
256 (p0) fmerge.s f9 = f0,f8
264 fmerge.ns f10 = f0,f9
271 add GR_Parameter_Y=-32,sp // Parameter 2 value
273 .save ar.pfs,GR_SAVE_PFS
274 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
278 add sp=-64,sp // Create new stack
280 mov GR_SAVE_GP=gp // Save gp
286 stfd [GR_Parameter_Y] = f0,16 // STORE Parameter 2 on stack
287 add GR_Parameter_X = 16,sp // Parameter 1 address
289 mov GR_SAVE_B0=b0 // Save b0
295 stfd [GR_Parameter_X] = f8 // STORE Parameter 1 on stack
296 add GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address
300 stfd [GR_Parameter_Y] = f9 // Store Parameter 3 on stack
301 add GR_Parameter_Y = -16,GR_Parameter_Y
302 br.call.sptk b0=__libm_error_support# // Call error handling function
307 add GR_Parameter_RESULT = 48,sp
312 ldfd f8 = [GR_Parameter_RESULT] // Get return result off stack
314 add sp = 64,sp // Restore stack pointer
315 mov b0 = GR_SAVE_B0 // Restore return address
318 mov gp = GR_SAVE_GP // Restore gp
319 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
323 .endp __libm_error_region
324 ASM_SIZE_DIRECTIVE(__libm_error_region)
326 .type __libm_error_support#,@function
327 .global __libm_error_support#