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.
40 // *********************************************************************
43 // 2/02/00 Initial version
44 // 4/04/00 Unwind support added
45 // 8/15/00 Bundle added after call to __libm_error_support to properly
46 // set [the previously overwritten] GR_Parameter_RESULT.
48 // *********************************************************************
50 // Function: Combined sqrtf(x), where
52 // sqrtf(x) = |x, for single precision x values
54 // ********************************************************************
56 // Accuracy: Correctly Rounded
58 // ********************************************************************
62 // Floating-Point Registers: f8 (Input and Return Value)
65 // General Purpose Registers:
67 // r37-r40 (Used to pass arguments to error handling routine)
69 // Predicate Registers: p6, p7, p8
71 // ********************************************************************
73 // IEEE Special Conditions:
75 // All faults and exceptions should be raised correctly.
79 // sqrtf(negative) = QNaN and error handling is called
81 // ********************************************************************
85 // Modified Newton-Raphson Algorithm
87 // ********************************************************************
89 #include "libm_support.h"
97 GR_Parameter_RESULT = r39
98 GR_Parameter_TAG = r40
114 .type __sqrtf,@function
116 .global __ieee754_sqrtf
117 .type __ieee754_sqrtf,@function
121 // BEGIN SINGLE PRECISION MINIMUM LATENCY SQUARE ROOT ALGORITHM
122 alloc r32= ar.pfs,0,5,4,0
123 // exponent of +1/2 in r2
133 // y0 = 1/sqrt(a) in f7
134 fclass.m.unc p7,p8 = f8,0x3A
138 // Make a copy of x just in case
144 // H0 = 1/2 * y0 in f9
145 (p6) fma.s1 f9=f12,f7,f0
151 (p6) fma.s1 f7=f8,f7,f0
156 // d = 1/2 - S0 * H0 in f10
157 (p6) fnma.s1 f10=f7,f9,f12
162 // 3/2 = 1 + 1/2 in f12
163 (p6) fma.s1 f12=f12,f1,f1
168 // e = 1 + 3/2 * d in f12
169 (p6) fma.s1 f12=f12,f10,f1
174 // T0 = d * S0 in f11
175 (p6) fma.s1 f11=f10,f7,f0
180 // G0 = d * H0 in f10
181 (p6) fma.s1 f10=f10,f9,f0
186 // S1 = S0 + e * T0 in f7
187 (p6) fma.s.s1 f7=f12,f11,f7
192 // H1 = H0 + e * G0 in f12
193 (p6) fma.s1 f12=f12,f10,f9
198 // d1 = a - S1 * S1 in f9
199 (p6) fnma.s1 f9=f7,f7,f8
204 // S = S1 + d1 * H1 in f7
205 (p6) fma.s.s0 f8=f9,f12,f7
206 (p6) br.ret.sptk b0 ;;
207 // END SINGLE PRECISION MINIMUM LATENCY SQUARE ROOT ALGORITHM
211 (p8) br.ret.sptk b0 ;;
214 // This branch includes all those special values that are not negative,
215 // with the result equal to frcpa(x)
218 ASM_SIZE_DIRECTIVE(sqrtf)
220 ASM_SIZE_DIRECTIVE(__sqrtf)
221 ASM_SIZE_DIRECTIVE(__ieee754_sqrtf)
225 .proc __libm_error_region
229 add GR_Parameter_Y=-32,sp // Parameter 2 value
230 (p0) mov GR_Parameter_TAG = 50
231 .save ar.pfs,GR_SAVE_PFS
232 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
236 add sp=-64,sp // Create new stack
238 mov GR_SAVE_GP=gp // Save gp
241 stfs [GR_Parameter_Y] = FR_Y,16 // Store Parameter 2 on stack
242 add GR_Parameter_X = 16,sp // Parameter 1 address
244 mov GR_SAVE_B0=b0 // Save b0
248 stfs [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
249 add GR_Parameter_RESULT = 0,GR_Parameter_Y
250 nop.b 0 // Parameter 3 address
253 stfs [GR_Parameter_Y] = FR_RESULT // Store Parameter 3 on stack
254 add GR_Parameter_Y = -16,GR_Parameter_Y
255 br.call.sptk b0=__libm_error_support# // Call error handling function
260 add GR_Parameter_RESULT = 48,sp
263 ldfs f8 = [GR_Parameter_RESULT] // Get return result off stack
265 add sp = 64,sp // Restore stack pointer
266 mov b0 = GR_SAVE_B0 // Restore return address
269 mov gp = GR_SAVE_GP // Restore gp
270 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
271 br.ret.sptk b0 // Return
274 .endp __libm_error_region
275 ASM_SIZE_DIRECTIVE(__libm_error_region)
278 .type __libm_error_support#,@function
279 .global __libm_error_support#