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.
40 //*********************************************************************
43 // 02/02/00 Initial version
44 // 04/04/00 Unwind support added
45 // 08/15/00 Bundle added after call to __libm_error_support to properly
46 // set [the previously overwritten] GR_Parameter_RESULT.
47 // 05/20/02 Cleaned up namespace and sf0 syntax
48 // 02/10/03 Reordered header: .section, .global, .proc, .align
50 //*********************************************************************
52 // Function: Combined sqrtf(x), where
54 // sqrtf(x) = |x, for single precision x values
56 //********************************************************************
58 // Accuracy: Correctly Rounded
60 //********************************************************************
64 // Floating-Point Registers: f8 (Input and Return Value)
67 // General Purpose Registers:
69 // r37-r40 (Used to pass arguments to error handling routine)
71 // Predicate Registers: p6, p7, p8
73 //********************************************************************
75 // IEEE Special Conditions:
77 // All faults and exceptions should be raised correctly.
81 // sqrtf(negative) = QNaN and error handling is called
83 //********************************************************************
87 // Modified Newton-Raphson Algorithm
89 //********************************************************************
98 GR_Parameter_RESULT = r39
99 GR_Parameter_TAG = r40
107 GLOBAL_IEEE754_ENTRY(sqrtf)
109 // BEGIN SINGLE PRECISION MINIMUM LATENCY SQUARE ROOT ALGORITHM
110 alloc r32= ar.pfs,0,5,4,0
111 // exponent of +1/2 in r2
121 // y0 = 1/sqrt(a) in f7
122 fclass.m.unc p7,p8 = f8,0x3A
126 // Make a copy of x just in case
132 // H0 = 1/2 * y0 in f9
133 (p6) fma.s1 f9=f12,f7,f0
139 (p6) fma.s1 f7=f8,f7,f0
144 // d = 1/2 - S0 * H0 in f10
145 (p6) fnma.s1 f10=f7,f9,f12
150 // 3/2 = 1 + 1/2 in f12
151 (p6) fma.s1 f12=f12,f1,f1
156 // e = 1 + 3/2 * d in f12
157 (p6) fma.s1 f12=f12,f10,f1
162 // T0 = d * S0 in f11
163 (p6) fma.s1 f11=f10,f7,f0
168 // G0 = d * H0 in f10
169 (p6) fma.s1 f10=f10,f9,f0
174 // S1 = S0 + e * T0 in f7
175 (p6) fma.s.s1 f7=f12,f11,f7
180 // H1 = H0 + e * G0 in f12
181 (p6) fma.s1 f12=f12,f10,f9
186 // d1 = a - S1 * S1 in f9
187 (p6) fnma.s1 f9=f7,f7,f8
192 // S = S1 + d1 * H1 in f7
193 (p6) fma.s.s0 f8=f9,f12,f7
194 (p6) br.ret.sptk b0 ;;
195 // END SINGLE PRECISION MINIMUM LATENCY SQUARE ROOT ALGORITHM
199 (p8) br.ret.sptk b0 ;;
202 // This branch includes all those special values that are not negative,
203 // with the result equal to frcpa(x)
205 GLOBAL_IEEE754_END(sqrtf)
208 LOCAL_LIBM_ENTRY(__libm_error_region)
211 add GR_Parameter_Y=-32,sp // Parameter 2 value
212 mov GR_Parameter_TAG = 50
213 .save ar.pfs,GR_SAVE_PFS
214 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
218 add sp=-64,sp // Create new stack
220 mov GR_SAVE_GP=gp // Save gp
223 stfs [GR_Parameter_Y] = FR_Y,16 // Store Parameter 2 on stack
224 add GR_Parameter_X = 16,sp // Parameter 1 address
226 mov GR_SAVE_B0=b0 // Save b0
230 stfs [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
231 add GR_Parameter_RESULT = 0,GR_Parameter_Y
232 nop.b 0 // Parameter 3 address
235 stfs [GR_Parameter_Y] = FR_RESULT // Store Parameter 3 on stack
236 add GR_Parameter_Y = -16,GR_Parameter_Y
237 br.call.sptk b0=__libm_error_support# // Call error handling function
242 add GR_Parameter_RESULT = 48,sp
245 ldfs f8 = [GR_Parameter_RESULT] // Get return result off stack
247 add sp = 64,sp // Restore stack pointer
248 mov b0 = GR_SAVE_B0 // Restore return address
251 mov gp = GR_SAVE_GP // Restore gp
252 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
253 br.ret.sptk b0 // Return
256 LOCAL_LIBM_END(__libm_error_region)
259 .type __libm_error_support#,@function
260 .global __libm_error_support#