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 (hand-optimized)
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 sqrtl(x), where
52 // sqrtl(x) = |x, for double-extended precision x values
54 // ********************************************************************
58 // Floating-Point Registers: f8 (Input and Return Value)
61 // General Purpose Registers:
63 // r37-r40 (Used to pass arguments to error handling routine)
65 // Predicate Registers: p6, p7, p8
67 // ********************************************************************
69 // IEEE Special Conditions:
71 // All faults and exceptions should be raised correctly.
75 // sqrtl(negative) = QNaN and error handling is called
77 // ********************************************************************
81 // Modified Newton-Raphson Algorithm
83 // ********************************************************************
85 #include "libm_support.h"
92 GR_Parameter_RESULT = r39
93 GR_Parameter_TAG = r40
107 .type __sqrtl,@function
109 .global __ieee754_sqrtl
110 .type __ieee754_sqrtl,@function
114 alloc r32= ar.pfs,0,5,4,0
115 // exponent of +1/2 in r2
121 // y0 = 1/sqrt(a) in f7
127 // H0 = +1/2 * y0 in f9
128 (p6) fma.s1 f9=f12,f7,f0
134 (p6) fma.s1 f7=f8,f7,f0
143 fclass.m.unc p7,p8 = f8,0x3A
148 // d0 = 1/2 - S0 * H0 in f10
149 (p6) fnma.s1 f10=f7,f9,f12
159 // H1 = H0 + d0 * H0 in f9
160 (p6) fma.s1 f9=f10,f9,f9
165 // S1 = S0 + d0 * S0 in f7
166 (p6) fma.s1 f7=f10,f7,f7
171 // d1 = 1/2 - S1 * H1 in f10
172 (p6) fnma.s1 f10=f7,f9,f12
177 // H2 = H1 + d1 * H1 in f9
178 (p6) fma.s1 f9=f10,f9,f9
183 // S2 = S1 + d1 * S1 in f7
184 (p6) fma.s1 f7=f10,f7,f7
189 // d2 = 1/2 - S2 * H2 in f10
190 (p6) fnma.s1 f10=f7,f9,f12
195 // e2 = a - S2 * S2 in f12
196 (p6) fnma.s1 f12=f7,f7,f8
201 // S3 = S2 + d2 * S2 in f7
202 (p6) fma.s1 f7=f12,f9,f7
207 // H3 = H2 + d2 * H2 in f9
208 (p6) fma.s1 f9=f10,f9,f9
213 // e3 = a - S3 * S3 in f12
214 (p6) fnma.s1 f12=f7,f7,f8
219 // S = S3 + e3 * H3 in f7
220 (p6) fma.s0 f8=f12,f9,f7
221 (p6) br.ret.sptk b0 ;;
224 (p0) mov GR_Parameter_TAG = 48
226 (p8) br.ret.sptk b0 ;;
229 // This branch includes all those special values that are not negative,
230 // with the result equal to frcpa(x)
234 // END DOUBLE EXTENDED PRECISION MINIMUM LATENCY SQUARE ROOT ALGORITHM
236 ASM_SIZE_DIRECTIVE(sqrtl)
238 ASM_SIZE_DIRECTIVE(__sqrtl)
239 ASM_SIZE_DIRECTIVE(__ieee754_sqrtl)
242 .proc __libm_error_region
246 add GR_Parameter_Y=-32,sp // Parameter 2 value
248 .save ar.pfs,GR_SAVE_PFS
249 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
253 add sp=-64,sp // Create new stack
255 mov GR_SAVE_GP=gp // Save gp
258 stfe [GR_Parameter_Y] = FR_Y,16 // Save Parameter 2 on stack
259 add GR_Parameter_X = 16,sp // Parameter 1 address
261 mov GR_SAVE_B0=b0 // Save b0
265 stfe [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
266 add GR_Parameter_RESULT = 0,GR_Parameter_Y
267 nop.b 0 // Parameter 3 address
270 stfe [GR_Parameter_Y] = FR_RESULT // Store Parameter 3 on stack
271 add GR_Parameter_Y = -16,GR_Parameter_Y
272 br.call.sptk b0=__libm_error_support# // Call error handling function
277 add GR_Parameter_RESULT = 48,sp
280 ldfe f8 = [GR_Parameter_RESULT] // Get return result off stack
282 add sp = 64,sp // Restore stack pointer
283 mov b0 = GR_SAVE_B0 // Restore return address
286 mov gp = GR_SAVE_GP // Restore gp
287 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
288 br.ret.sptk b0 // Return
291 .endp __libm_error_region
292 ASM_SIZE_DIRECTIVE(__libm_error_region)
293 .type __libm_error_support#,@function
294 .global __libm_error_support#