1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
|
.file "sqrt.s"
// Copyright (C) 2000, 2001, Intel Corporation
// All rights reserved.
//
// Contributed 2/2/2000 by John Harrison, Ted Kubaska, Bob Norin, Shane Story,
// and Ping Tak Peter Tang of the Computational Software Lab, Intel Corporation.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// * Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote
// products derived from this software without specific prior written
// permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
// Intel Corporation is the author of this code, and requests that all
// problem reports or change requests be submitted to it directly at
// http://developer.intel.com/opensource.
//
// ********************************************************************
// History
// ********************************************************************
// 2/02/00 Initial version
// 4/04/00 Unwind support added
// 8/15/00 Bundle added after call to __libm_error_support to properly
// set [the previously overwritten] GR_Parameter_RESULT.
//
// ********************************************************************
//
// Function: Combined sqrt(x), where
// _
// sqrt(x) = |x, for double precision x values
//
// ********************************************************************
//
// Accuracy: Correctly Rounded
//
// ********************************************************************
//
// Resources Used:
//
// Floating-Point Registers: f8 (Input and Return Value)
// f7 -f14
//
// General Purpose Registers:
// r32-r36 (Locals)
// r37-r40 (Used to pass arguments to error handling routine)
//
// Predicate Registers: p6, p7, p8
//
// *********************************************************************
//
// IEEE Special Conditions:
//
// All faults and exceptions should be raised correctly.
// sqrt(QNaN) = QNaN
// sqrt(SNaN) = QNaN
// sqrt(+/-0) = +/-0
// sqrt(negative) = QNaN and error handling is called
//
// *********************************************************************
//
// Implementation:
//
// Modified Newton-Raphson Algorithm
//
// *********************************************************************
#include "libm_support.h"
GR_SAVE_PFS = r33
GR_SAVE_B0 = r34
GR_SAVE_GP = r35
GR_Parameter_X = r37
GR_Parameter_Y = r38
GR_Parameter_RESULT = r39
.section .text
.proc sqrt#
.global sqrt#
.align 64
sqrt:
#ifdef _LIBC
.global __sqrt
.type __sqrt,@function
__sqrt:
.global __ieee754_sqrt
.type __ieee754_sqrt,@function
__ieee754_sqrt:
#endif
{ .mfi
alloc r32= ar.pfs,0,5,4,0
frsqrta.s0 f7,p6=f8
nop.i 0
} { .mlx
// BEGIN DOUBLE PRECISION MINIMUM LATENCY SQUARE ROOT ALGORITHM
nop.m 0
// exponent of +1/2 in r2
movl r2 = 0x0fffe;;
} { .mmi
// +1/2 in f9
setf.exp f9 = r2
nop.m 0
nop.i 0
} { .mlx
nop.m 0
// 3/2 in r3
movl r3=0x3fc00000;;
} { .mfi
setf.s f10=r3
// Step (1)
// y0 = 1/sqrt(a) in f7
fclass.m.unc p7,p8 = f8,0x3A
nop.i 0;;
} { .mlx
nop.m 0
// 5/2 in r2
movl r2 = 0x40200000
} { .mlx
nop.m 0
// 63/8 in r3
movl r3 = 0x40fc0000;;
} { .mfi
setf.s f11=r2
// Step (2)
// h = +1/2 * y0 in f6
(p6) fma.s1 f6=f9,f7,f0
nop.i 0
} { .mfi
setf.s f12=r3
// Step (3)
// g = a * y0 in f7
(p6) fma.s1 f7=f8,f7,f0
nop.i 0
} { .mfi
nop.m 0
mov f15 = f8
nop.i 0;;
} { .mlx
nop.m 0
// 231/16 in r2
movl r2 = 0x41670000;;
} { .mfi
setf.s f13=r2
// Step (4)
// e = 1/2 - g * h in f9
(p6) fnma.s1 f9=f7,f6,f9
nop.i 0
} { .mlx
nop.m 0
// 35/8 in r3
movl r3 = 0x408c0000;;
} { .mfi
setf.s f14=r3
// Step (5)
// S = 3/2 + 5/2 * e in f10
(p6) fma.s1 f10=f11,f9,f10
nop.i 0
} { .mfi
nop.m 0
// Step (6)
// e2 = e * e in f11
(p6) fma.s1 f11=f9,f9,f0
nop.i 0;;
} { .mfi
nop.m 0
// Step (7)
// t = 63/8 + 231/16 * e in f12
(p6) fma.s1 f12=f13,f9,f12
nop.i 0;;
} { .mfi
nop.m 0
// Step (8)
// S1 = e + e2 * S in f10
(p6) fma.s1 f10=f11,f10,f9
nop.i 0
} { .mfi
nop.m 0
// Step (9)
// e4 = e2 * e2 in f11
(p6) fma.s1 f11=f11,f11,f0
nop.i 0;;
} { .mfi
nop.m 0
// Step (10)
// t1 = 35/8 + e * t in f9
(p6) fma.s1 f9=f9,f12,f14
nop.i 0;;
} { .mfi
nop.m 0
// Step (11)
// G = g + S1 * g in f12
(p6) fma.s1 f12=f10,f7,f7
nop.i 0
} { .mfi
nop.m 0
// Step (12)
// E = g * e4 in f7
(p6) fma.s1 f7=f7,f11,f0
nop.i 0;;
} { .mfi
nop.m 0
// Step (13)
// u = S1 + e4 * t1 in f10
(p6) fma.s1 f10=f11,f9,f10
nop.i 0;;
} { .mfi
nop.m 0
// Step (14)
// g1 = G + t1 * E in f7
(p6) fma.d.s1 f7=f9,f7,f12
nop.i 0;;
} { .mfi
nop.m 0
// Step (15)
// h1 = h + u * h in f6
(p6) fma.s1 f6=f10,f6,f6
nop.i 0;;
} { .mfi
nop.m 0
// Step (16)
// d = a - g1 * g1 in f9
(p6) fnma.s1 f9=f7,f7,f8
nop.i 0;;
} { .mfb
nop.m 0
// Step (17)
// g2 = g1 + d * h1 in f7
(p6) fma.d.s0 f8=f9,f6,f7
(p6) br.ret.sptk b0 ;;
}
{ .mfb
nop.m 0
(p0) mov f8 = f7
(p8) br.ret.sptk b0 ;;
}
{ .mfb
(p7) mov r40 = 49
nop.f 0
(p7) br.cond.sptk __libm_error_region ;;
}
// END DOUBLE PRECISION MINIMUM LATENCY SQUARE ROOT ALGORITHM
.endp sqrt#
ASM_SIZE_DIRECTIVE(sqrt)
#ifdef _LIBC
ASM_SIZE_DIRECTIVE(__sqrt)
ASM_SIZE_DIRECTIVE(__ieee754_sqrt)
#endif
// Stack operations when calling error support.
// (1) (2) (3) (call) (4)
// sp -> + psp -> + psp -> + sp -> +
// | | | |
// | | <- GR_Y R3 ->| <- GR_RESULT | -> f8
// | | | |
// | <-GR_Y Y2->| Y2 ->| <- GR_Y |
// | | | |
// | | <- GR_X X1 ->| |
// | | | |
// sp-64 -> + sp -> + sp -> + +
// save ar.pfs save b0 restore gp
// save gp restore ar.pfs
.proc __libm_error_region
__libm_error_region:
//
// This branch includes all those special values that are not negative,
// with the result equal to frcpa(x)
//
.prologue
// We are distinguishing between over(under)flow and letting
// __libm_error_support set ERANGE or do anything else needed.
// (1)
{ .mfi
add GR_Parameter_Y=-32,sp // Parameter 2 value
nop.f 0
.save ar.pfs,GR_SAVE_PFS
mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
}
{ .mfi
.fframe 64
add sp=-64,sp // Create new stack
nop.f 0
mov GR_SAVE_GP=gp // Save gp
};;
// (2)
{ .mmi
stfd [GR_Parameter_Y] = f0,16 // STORE Parameter 2 on stack
add GR_Parameter_X = 16,sp // Parameter 1 address
.save b0, GR_SAVE_B0
mov GR_SAVE_B0=b0 // Save b0
};;
.body
// (3)
{ .mib
stfd [GR_Parameter_X] = f15 // STORE Parameter 1 on stack
add GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address
nop.b 0
}
{ .mib
stfd [GR_Parameter_Y] = f8 // STORE Parameter 3 on stack
add GR_Parameter_Y = -16,GR_Parameter_Y
br.call.sptk b0=__libm_error_support# // Call error handling function
};;
{ .mmi
nop.m 0
nop.m 0
add GR_Parameter_RESULT = 48,sp
};;
// (4)
{ .mmi
ldfd f8 = [GR_Parameter_RESULT] // Get return result off stack
.restore sp
add sp = 64,sp // Restore stack pointer
mov b0 = GR_SAVE_B0 // Restore return address
};;
{ .mib
mov gp = GR_SAVE_GP // Restore gp
mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
br.ret.sptk b0 // Return
};;
.endp __libm_error_region
ASM_SIZE_DIRECTIVE(__libm_error_region)
.type __libm_error_support#,@function
.global __libm_error_support#
|
