| blob | d286e9abad55552d6818987c90c993ef929695e7 |
1 .file "powl.s"
3 // Copyright (C) 2000, 2001, Intel Corporation
4 // All rights reserved.
5 //
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.
8 //
9 // Redistribution and use in source and binary forms, with or without
10 // modification, are permitted provided that the following conditions are
11 // met:
12 //
13 // * Redistributions of source code must retain the above copyright
14 // notice, this list of conditions and the following disclaimer.
15 //
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.
19 //
20 // * The name of Intel Corporation may not be used to endorse or promote
21 // products derived from this software without specific prior written
22 // permission.
23 //
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.
35 //
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.
39 //
40 // *********************************************************************
41 //
42 // Function: powl(x,y), where
43 // y
44 // powl(x,y) = x , for double extended precision x and y values
45 //
46 // *********************************************************************
47 //
48 // History:
49 // 2/02/00 (Hand Optimized)
50 // 4/04/00 Unwind support added
51 // 8/15/00 Bundle added after call to __libm_error_support to properly
52 // set [the previously overwritten] GR_Parameter_RESULT.
53 // 1/22/01 Corrected results for powl(1,inf), powl(1,nan), and
54 // powl(snan,0) to be 1 per C99, not nan. Fixed many flag settings.
55 // 2/06/01 Call __libm_error support if over/underflow when y=2.
56 //
57 // *********************************************************************
58 //
59 // Resources Used:
60 //
61 // Floating-Point Registers:
62 // f8 (Input and Return Value)
63 // f9-f15,f32-f63,f99
64 //
65 // General Purpose Registers:
66 // Locals r32 - r61
67 // Parameters to __libm_error_support r62,r63,r64,r65
68 //
69 // Predicate Registers: p6-p15
70 //
71 // *********************************************************************
72 //
73 // Special Cases and IEEE special conditions:
74 //
75 // Denormal fault raised on denormal inputs
76 // Overflow exceptions raised when appropriate for pow
77 // Underflow exceptions raised when appropriate for pow
78 // (Error Handling Routine called for overflow and Underflow)
79 // Inexact raised when appropriate by algorithm
80 //
81 // 1. (anything) ** NatVal or (NatVal) ** anything is NatVal
82 // 2. X or Y unsupported or sNaN is qNaN/Invalid
83 // 3. (anything) ** 0 is 1
84 // 4. (anything) ** 1 is itself
85 // 5. (anything except 1) ** qNAN is qNAN
86 // 6. qNAN ** (anything except 0) is qNAN
87 // 7. +-(|x| > 1) ** +INF is +INF
88 // 8. +-(|x| > 1) ** -INF is +0
89 // 9. +-(|x| < 1) ** +INF is +0
90 // 10. +-(|x| < 1) ** -INF is +INF
91 // 11. +-1 ** +-INF is +1
92 // 12. +0 ** (+anything except 0, NAN) is +0
93 // 13. -0 ** (+anything except 0, NAN, odd integer) is +0
94 // 14. +0 ** (-anything except 0, NAN) is +INF/div_0
95 // 15. -0 ** (-anything except 0, NAN, odd integer) is +INF/div_0
96 // 16. -0 ** (odd integer) = -( +0 ** (odd integer) )
97 // 17. +INF ** (+anything except 0,NAN) is +INF
98 // 18. +INF ** (-anything except 0,NAN) is +0
99 // 19. -INF ** (anything except NAN) = -0 ** (-anything)
100 // 20. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer)
101 // 21. (-anything except 0 and inf) ** (non-integer) is qNAN/Invalid
102 // 22. X or Y denorm/unorm and denorm/unorm operand trap is enabled,
103 // generate denorm/unorm fault except if invalid or div_0 raised.
104 //
105 // *********************************************************************
106 //
107 // Algorithm
108 // =========
109 //
110 // Special Cases
111 //
112 // If Y = 2, return X*X.
113 // If Y = 0.5, return sqrt(X).
114 //
115 // Compute log(X) to extra precision.
116 //
117 // ker_log_80( X, logX_hi, logX_lo, Safe );
118 //
119 // ...logX_hi + logX_lo approximates log(X) to roughly 80
120 // ...significant bits of accuracy.
121 //
122 // Compute Y*log(X) to extra precision.
123 //
124 // P_hi := Y * logX_hi
125 // P_lo := Y * logX_hi - P_hi ...using FMA
126 // P_lo := Y * logX_lo + P_lo ...using FMA
127 //
128 // Compute exp(P_hi + P_lo)
129 //
130 // Flag := 2;
131 // Expo_Range := 2; (assuming double-extended power function)
132 // ker_exp_64( P_hi, P_lo, Flag, Expo_Range,
133 // Z_hi, Z_lo, scale, Safe )
134 //
135 // scale := sgn * scale
136 //
137 // If (Safe) then ...result will not over/underflow
138 // return scale*Z_hi + (scale*Z_lo)
139 // quickly
140 // Else
141 // take necessary precaution in computing
142 // scale*Z_hi + (scale*Z_lo)
143 // to set possible exceptions correctly.
144 // End If
145 //
146 // Case_Y_Special
147 //
148 // ...Follow the order of the case checks
149 //
150 // If Y is +-0, return +1 without raising any exception.
151 // If Y is +1, return X without raising any exception.
152 // If Y is qNaN, return Y without exception.
153 // If X is qNaN, return X without exception.
154 //
155 // At this point, X is real and Y is +-inf.
156 // Thus |X| can only be 1, strictly bigger than 1, or
157 // strictly less than 1.
158 //
159 // If |X| < 1, then
160 // return ( Y == +inf? +0 : +inf )
161 // elseif |X| > 1, then
162 // return ( Y == +inf? +0 : +inf )
163 // else
164 // goto Case_Invalid
165 //
166 // Case_X_Special
167 //
168 // ...Follow the order of the case checks
169 // ...Note that Y is real, finite, non-zero, and not +1.
170 //
171 // If X is qNaN, return X without exception.
172 //
173 // If X is +-0,
174 // return ( Y > 0 ? +0 : +inf )
175 //
176 // If X is +inf
177 // return ( Y > 0 ? +inf : +0 )
178 //
179 // If X is -inf
180 // return -0 ** -Y
181 // return ( Y > 0 ? +inf : +0 )
182 //
183 // Case_Invalid
184 //
185 // Return 0 * inf to generate a quiet NaN together
186 // with an invalid exception.
187 //
188 // Implementation
189 // ==============
190 //
191 // We describe the quick branch since this part is important
192 // in reaching the normal case efficiently.
193 //
194 // STAGE 1
195 // -------
196 // This stage contains two threads.
197 //
198 // Stage1.Thread1
199 //
200 // fclass.m X_excep, X_ok = X, (NatVal or s/qNaN) or
201 // +-0, +-infinity
202 //
203 // fclass.nm X_unsupp, X_supp = X, (NatVal or s/qNaN) or
204 // +-(0, unnorm, norm, infinity)
205 //
206 // X_norm := fnorm( X ) with traps disabled
207 //
208 // If (X_excep) goto Filtering (Step 2)
209 // If (X_unsupp) goto Filtering (Step 2)
210 //
211 // Stage1.Thread2
212 // ..............
213 //
214 // fclass.m Y_excep, Y_ok = Y, (NatVal or s/qNaN) or
215 // +-0, +-infinity
216 //
217 // fclass.nm Y_unsupp, Y_supp = Y, (NatVal or s/qNaN) or
218 // +-(0, unnorm, norm, infinity)
219 //
220 // Y_norm := fnorm( Y ) with traps disabled
221 //
222 // If (Y_excep) goto Filtering (Step 2)
223 // If (Y_unsupp) goto Filtering (Step 2)
224 //
225 //
226 // STAGE 2
227 // -------
228 // This stage contains two threads.
229 //
230 // Stage2.Thread1
231 // ..............
232 //
233 // Set X_lt_0 if X < 0 (using fcmp)
234 // sgn := +1.0
235 // If (X_lt_0) goto Filtering (Step 2)
236 //
237 // Stage2.Thread2
238 // ..............
239 //
240 // Set Y_is_1 if Y = +1 (using fcmp)
241 // If (Y_is_1) goto Filtering (Step 2)
242 //
243 // STAGE 3
244 // -------
245 // This stage contains two threads.
246 //
247 //
248 // Stage3.Thread1
249 // ..............
250 //
251 // X := fnorm(X) in prevailing traps
252 //
253 //
254 // Stage3.Thread2
255 // ..............
256 //
257 // Y := fnorm(Y) in prevailing traps
258 //
259 // STAGE 4
260 // -------
261 //
262 // Go to Case_Normal.
263 //
265 #include "libm_support.h"
267 #ifdef _LIBC
268 .rodata
269 #else
270 .data
271 #endif
273 // Inv_L, L_hi, L_lo
274 .align 64
275 Constants_exp_64_Arg:
276 ASM_TYPE_DIRECTIVE(Constants_exp_64_Arg,@object)
277 data4 0x5C17F0BC,0xB8AA3B29,0x0000400B,0x00000000
278 data4 0x00000000,0xB17217F4,0x00003FF2,0x00000000
279 data4 0xF278ECE6,0xF473DE6A,0x00003FD4,0x00000000
280 ASM_SIZE_DIRECTIVE(Constants_exp_64_Arg)
282 .align 64
283 Constants_exp_64_Exponents:
284 ASM_TYPE_DIRECTIVE(Constants_exp_64_Exponents,@object)
285 data4 0x0000007E,0x00000000,0xFFFFFF83,0xFFFFFFFF
286 data4 0x000003FE,0x00000000,0xFFFFFC03,0xFFFFFFFF
287 data4 0x00003FFE,0x00000000,0xFFFFC003,0xFFFFFFFF
288 data4 0x00003FFE,0x00000000,0xFFFFC003,0xFFFFFFFF
289 data4 0xFFFFFFE2,0xFFFFFFFF,0xFFFFFFC4,0xFFFFFFFF
290 data4 0xFFFFFFBA,0xFFFFFFFF,0xFFFFFFBA,0xFFFFFFFF
291 ASM_SIZE_DIRECTIVE(Constants_exp_64_Exponents)
293 .align 64
294 Constants_exp_64_A:
295 ASM_TYPE_DIRECTIVE(Constants_exp_64_A,@object)
296 // Reversed
297 data4 0xB1B736A0,0xAAAAAAAB,0x00003FFA,0x00000000
298 data4 0x90CD6327,0xAAAAAAAB,0x00003FFC,0x00000000
299 data4 0xFFFFFFFF,0xFFFFFFFF,0x00003FFD,0x00000000
300 ASM_SIZE_DIRECTIVE(Constants_exp_64_A)
302 .align 64
303 Constants_exp_64_P:
304 ASM_TYPE_DIRECTIVE(Constants_exp_64_P,@object)
305 // Reversed
306 data4 0x43914A8A,0xD00D6C81,0x00003FF2,0x00000000
307 data4 0x30304B30,0xB60BC4AC,0x00003FF5,0x00000000
308 data4 0x7474C518,0x88888888,0x00003FF8,0x00000000
309 data4 0x8DAE729D,0xAAAAAAAA,0x00003FFA,0x00000000
310 data4 0xAAAAAF61,0xAAAAAAAA,0x00003FFC,0x00000000
311 data4 0x000004C7,0x80000000,0x00003FFE,0x00000000
312 ASM_SIZE_DIRECTIVE(Constants_exp_64_P)
314 .align 64
315 Constants_exp_64_T1:
316 ASM_TYPE_DIRECTIVE(Constants_exp_64_T1,@object)
317 data4 0x3F800000,0x3F8164D2,0x3F82CD87,0x3F843A29
318 data4 0x3F85AAC3,0x3F871F62,0x3F88980F,0x3F8A14D5
319 data4 0x3F8B95C2,0x3F8D1ADF,0x3F8EA43A,0x3F9031DC
320 data4 0x3F91C3D3,0x3F935A2B,0x3F94F4F0,0x3F96942D
321 data4 0x3F9837F0,0x3F99E046,0x3F9B8D3A,0x3F9D3EDA
322 data4 0x3F9EF532,0x3FA0B051,0x3FA27043,0x3FA43516
323 data4 0x3FA5FED7,0x3FA7CD94,0x3FA9A15B,0x3FAB7A3A
324 data4 0x3FAD583F,0x3FAF3B79,0x3FB123F6,0x3FB311C4
325 data4 0x3FB504F3,0x3FB6FD92,0x3FB8FBAF,0x3FBAFF5B
326 data4 0x3FBD08A4,0x3FBF179A,0x3FC12C4D,0x3FC346CD
327 data4 0x3FC5672A,0x3FC78D75,0x3FC9B9BE,0x3FCBEC15
328 data4 0x3FCE248C,0x3FD06334,0x3FD2A81E,0x3FD4F35B
329 data4 0x3FD744FD,0x3FD99D16,0x3FDBFBB8,0x3FDE60F5
330 data4 0x3FE0CCDF,0x3FE33F89,0x3FE5B907,0x3FE8396A
331 data4 0x3FEAC0C7,0x3FED4F30,0x3FEFE4BA,0x3FF28177
332 data4 0x3FF5257D,0x3FF7D0DF,0x3FFA83B3,0x3FFD3E0C
333 ASM_SIZE_DIRECTIVE(Constants_exp_64_T1)
335 .align 64
336 Constants_exp_64_T2:
337 ASM_TYPE_DIRECTIVE(Constants_exp_64_T2,@object)
338 data4 0x3F800000,0x3F80058C,0x3F800B18,0x3F8010A4
339 data4 0x3F801630,0x3F801BBD,0x3F80214A,0x3F8026D7
340 data4 0x3F802C64,0x3F8031F2,0x3F803780,0x3F803D0E
341 data4 0x3F80429C,0x3F80482B,0x3F804DB9,0x3F805349
342 data4 0x3F8058D8,0x3F805E67,0x3F8063F7,0x3F806987
343 data4 0x3F806F17,0x3F8074A8,0x3F807A39,0x3F807FCA
344 data4 0x3F80855B,0x3F808AEC,0x3F80907E,0x3F809610
345 data4 0x3F809BA2,0x3F80A135,0x3F80A6C7,0x3F80AC5A
346 data4 0x3F80B1ED,0x3F80B781,0x3F80BD14,0x3F80C2A8
347 data4 0x3F80C83C,0x3F80CDD1,0x3F80D365,0x3F80D8FA
348 data4 0x3F80DE8F,0x3F80E425,0x3F80E9BA,0x3F80EF50
349 data4 0x3F80F4E6,0x3F80FA7C,0x3F810013,0x3F8105AA
350 data4 0x3F810B41,0x3F8110D8,0x3F81166F,0x3F811C07
351 data4 0x3F81219F,0x3F812737,0x3F812CD0,0x3F813269
352 data4 0x3F813802,0x3F813D9B,0x3F814334,0x3F8148CE
353 data4 0x3F814E68,0x3F815402,0x3F81599C,0x3F815F37
354 ASM_SIZE_DIRECTIVE(Constants_exp_64_T2)
356 .align 64
357 Constants_exp_64_W1:
358 ASM_TYPE_DIRECTIVE(Constants_exp_64_W1,@object)
359 data4 0x00000000,0x00000000,0x171EC4B4,0xBE384454
360 data4 0x4AA72766,0xBE694741,0xD42518F8,0xBE5D32B6
361 data4 0x3A319149,0x3E68D96D,0x62415F36,0xBE68F4DA
362 data4 0xC9C86A3B,0xBE6DDA2F,0xF49228FE,0x3E6B2E50
363 data4 0x1188B886,0xBE49C0C2,0x1A4C2F1F,0x3E64BFC2
364 data4 0x2CB98B54,0xBE6A2FBB,0x9A55D329,0x3E5DC5DE
365 data4 0x39A7AACE,0x3E696490,0x5C66DBA5,0x3E54728B
366 data4 0xBA1C7D7D,0xBE62B0DB,0x09F1AF5F,0x3E576E04
367 data4 0x1A0DD6A1,0x3E612500,0x795FBDEF,0xBE66A419
368 data4 0xE1BD41FC,0xBE5CDE8C,0xEA54964F,0xBE621376
369 data4 0x476E76EE,0x3E6370BE,0x3427EB92,0x3E390D1A
370 data4 0x2BF82BF8,0x3E1336DE,0xD0F7BD9E,0xBE5FF1CB
371 data4 0x0CEB09DD,0xBE60A355,0x0980F30D,0xBE5CA37E
372 data4 0x4C082D25,0xBE5C541B,0x3B467D29,0xBE5BBECA
373 data4 0xB9D946C5,0xBE400D8A,0x07ED374A,0xBE5E2A08
374 data4 0x365C8B0A,0xBE66CB28,0xD3403BCA,0x3E3AAD5B
375 data4 0xC7EA21E0,0x3E526055,0xE72880D6,0xBE442C75
376 data4 0x85222A43,0x3E58B2BB,0x522C42BF,0xBE5AAB79
377 data4 0x469DC2BC,0xBE605CB4,0xA48C40DC,0xBE589FA7
378 data4 0x1AA42614,0xBE51C214,0xC37293F4,0xBE48D087
379 data4 0xA2D673E0,0x3E367A1C,0x114F7A38,0xBE51BEBB
380 data4 0x661A4B48,0xBE6348E5,0x1D3B9962,0xBDF52643
381 data4 0x35A78A53,0x3E3A3B5E,0x1CECD788,0xBE46C46C
382 data4 0x7857D689,0xBE60B7EC,0xD14F1AD7,0xBE594D3D
383 data4 0x4C9A8F60,0xBE4F9C30,0x02DFF9D2,0xBE521873
384 data4 0x55E6D68F,0xBE5E4C88,0x667F3DC4,0xBE62140F
385 data4 0x3BF88747,0xBE36961B,0xC96EC6AA,0x3E602861
386 data4 0xD57FD718,0xBE3B5151,0xFC4A627B,0x3E561CD0
387 data4 0xCA913FEA,0xBE3A5217,0x9A5D193A,0x3E40A3CC
388 data4 0x10A9C312,0xBE5AB713,0xC5F57719,0x3E4FDADB
389 data4 0xDBDF59D5,0x3E361428,0x61B4180D,0x3E5DB5DB
390 data4 0x7408D856,0xBE42AD5F,0x31B2B707,0x3E2A3148
391 ASM_SIZE_DIRECTIVE(Constants_exp_64_W1)
393 .align 64
394 Constants_exp_64_W2:
395 ASM_TYPE_DIRECTIVE(Constants_exp_64_W2,@object)
396 data4 0x00000000,0x00000000,0x37A3D7A2,0xBE641F25
397 data4 0xAD028C40,0xBE68DD57,0xF212B1B6,0xBE5C77D8
398 data4 0x1BA5B070,0x3E57878F,0x2ECAE6FE,0xBE55A36A
399 data4 0x569DFA3B,0xBE620608,0xA6D300A3,0xBE53B50E
400 data4 0x223F8F2C,0x3E5B5EF2,0xD6DE0DF4,0xBE56A0D9
401 data4 0xEAE28F51,0xBE64EEF3,0x367EA80B,0xBE5E5AE2
402 data4 0x5FCBC02D,0x3E47CB1A,0x9BDAFEB7,0xBE656BA0
403 data4 0x805AFEE7,0x3E6E70C6,0xA3415EBA,0xBE6E0509
404 data4 0x49BFF529,0xBE56856B,0x00508651,0x3E66DD33
405 data4 0xC114BC13,0x3E51165F,0xC453290F,0x3E53333D
406 data4 0x05539FDA,0x3E6A072B,0x7C0A7696,0xBE47CD87
407 data4 0xEB05C6D9,0xBE668BF4,0x6AE86C93,0xBE67C3E3
408 data4 0xD0B3E84B,0xBE533904,0x556B53CE,0x3E63E8D9
409 data4 0x63A98DC8,0x3E212C89,0x032A7A22,0xBE33138F
410 data4 0xBC584008,0x3E530FA9,0xCCB93C97,0xBE6ADF82
411 data4 0x8370EA39,0x3E5F9113,0xFB6A05D8,0x3E5443A4
412 data4 0x181FEE7A,0x3E63DACD,0xF0F67DEC,0xBE62B29D
413 data4 0x3DDE6307,0x3E65C483,0xD40A24C1,0x3E5BF030
414 data4 0x14E437BE,0x3E658B8F,0xED98B6C7,0xBE631C29
415 data4 0x04CF7C71,0x3E6335D2,0xE954A79D,0x3E529EED
416 data4 0xF64A2FB8,0x3E5D9257,0x854ED06C,0xBE6BED1B
417 data4 0xD71405CB,0x3E5096F6,0xACB9FDF5,0xBE3D4893
418 data4 0x01B68349,0xBDFEB158,0xC6A463B9,0x3E628D35
419 data4 0xADE45917,0xBE559725,0x042FC476,0xBE68C29C
420 data4 0x01E511FA,0xBE67593B,0x398801ED,0xBE4A4313
421 data4 0xDA7C3300,0x3E699571,0x08062A9E,0x3E5349BE
422 data4 0x755BB28E,0x3E5229C4,0x77A1F80D,0x3E67E426
423 data4 0x6B69C352,0xBE52B33F,0x084DA57F,0xBE6B3550
424 data4 0xD1D09A20,0xBE6DB03F,0x2161B2C1,0xBE60CBC4
425 data4 0x78A2B771,0x3E56ED9C,0x9D0FA795,0xBE508E31
426 data4 0xFD1A54E9,0xBE59482A,0xB07FD23E,0xBE2A17CE
427 data4 0x17365712,0x3E68BF5C,0xB3785569,0x3E3956F9
428 ASM_SIZE_DIRECTIVE(Constants_exp_64_W2)
430 .align 64
431 Constants_log_80_P:
432 ASM_TYPE_DIRECTIVE(Constants_log_80_P,@object)
433 // 1/2, P_8, P_7, ..., P_1
434 data4 0x00000000, 0x80000000, 0x00003FFE, 0x00000000
435 data4 0x3B1042BC, 0xCCCE8B88, 0x0000BFFB, 0x00000000
436 data4 0xCADC2149, 0xE38997B7, 0x00003FFB, 0x00000000
437 data4 0xB1ACB090, 0xFFFFFFFE, 0x0000BFFB, 0x00000000
438 data4 0x06481C81, 0x92492498, 0x00003FFC, 0x00000000
439 data4 0xAAAAB0EF, 0xAAAAAAAA, 0x0000BFFC, 0x00000000
440 data4 0xCCC91416, 0xCCCCCCCC, 0x00003FFC, 0x00000000
441 data4 0x00000000, 0x80000000, 0x0000BFFD, 0x00000000
442 data4 0xAAAAAAAB, 0xAAAAAAAA, 0x00003FFD
443 ASM_SIZE_DIRECTIVE(Constants_log_80_P)
445 .align 64
446 Constants_log_80_Q:
447 ASM_TYPE_DIRECTIVE(Constants_log_80_Q,@object)
448 // log2_hi, log2_lo, Q_6, Q_5, Q_4, Q_3, Q_2, Q_1
449 data4 0x00000000,0xB1721800,0x00003FFE,0x00000000
450 data4 0x4361C4C6,0x82E30865,0x0000BFE2,0x00000000
451 data4 0xA51BE0AF,0x92492453,0x00003FFC,0x00000000
452 data4 0xA0CFD29F,0xAAAAAB73,0x0000BFFC,0x00000000
453 data4 0xCCCE3872,0xCCCCCCCC,0x00003FFC,0x00000000
454 data4 0xFFFFB4FB,0xFFFFFFFF,0x0000BFFC,0x00000000
455 data4 0xAAAAAAAB,0xAAAAAAAA,0x00003FFD,0x00000000
456 data4 0x00000000,0x80000000,0x0000BFFE,0x00000000
457 ASM_SIZE_DIRECTIVE(Constants_log_80_Q)
459 .align 64
460 Constants_log_80_Z_G_H_h1:
461 ASM_TYPE_DIRECTIVE(Constants_log_80_Z_G_H_h1,@object)
462 // Z1 - 16 bit fixed, G1 and H1 IEEE single, h1 IEEE double
463 data4 0x00008000,0x3F800000,0x00000000,0x00000000
464 data4 0x00000000,0x00000000,0x00000000,0x00000000
465 data4 0x00007879,0x3F70F0F0,0x3D785196,0x00000000
466 data4 0xEBA0E0D1,0x8B1D330B,0x00003FDA,0x00000000
467 data4 0x000071C8,0x3F638E38,0x3DF13843,0x00000000
468 data4 0x9EADD553,0xE2AF365E,0x00003FE2,0x00000000
469 data4 0x00006BCB,0x3F579430,0x3E2FF9A0,0x00000000
470 data4 0x752F34A2,0xF585FEC3,0x0000BFE3,0x00000000
471 data4 0x00006667,0x3F4CCCC8,0x3E647FD6,0x00000000
472 data4 0x893B03F3,0xF3546435,0x00003FE2,0x00000000
473 data4 0x00006187,0x3F430C30,0x3E8B3AE7,0x00000000
474 data4 0x39CDD2AC,0xBABA62E0,0x00003FE4,0x00000000
475 data4 0x00005D18,0x3F3A2E88,0x3EA30C68,0x00000000
476 data4 0x457978A1,0x8718789F,0x00003FE2,0x00000000
477 data4 0x0000590C,0x3F321640,0x3EB9CEC8,0x00000000
478 data4 0x3185E56A,0x9442DF96,0x0000BFE4,0x00000000
479 data4 0x00005556,0x3F2AAAA8,0x3ECF9927,0x00000000
480 data4 0x2BBE2CBD,0xCBF9A4BF,0x00003FE4,0x00000000
481 data4 0x000051EC,0x3F23D708,0x3EE47FC5,0x00000000
482 data4 0x852D5935,0xF3537535,0x00003FE3,0x00000000
483 data4 0x00004EC5,0x3F1D89D8,0x3EF8947D,0x00000000
484 data4 0x46CDF32F,0xA1F1E699,0x0000BFDF,0x00000000
485 data4 0x00004BDB,0x3F17B420,0x3F05F3A1,0x00000000
486 data4 0xD8484CE3,0x84A61856,0x00003FE4,0x00000000
487 data4 0x00004925,0x3F124920,0x3F0F4303,0x00000000
488 data4 0xFF28821B,0xC7DD97E0,0x0000BFE2,0x00000000
489 data4 0x0000469F,0x3F0D3DC8,0x3F183EBF,0x00000000
490 data4 0xEF1FD32F,0xD3C4A887,0x00003FE3,0x00000000
491 data4 0x00004445,0x3F088888,0x3F20EC80,0x00000000
492 data4 0x464C76DA,0x84672BE6,0x00003FE5,0x00000000
493 data4 0x00004211,0x3F042108,0x3F29516A,0x00000000
494 data4 0x18835FB9,0x9A43A511,0x0000BFE5,0x00000000
495 ASM_SIZE_DIRECTIVE(Constants_log_80_Z_G_H_h1)
497 .align 64
498 Constants_log_80_Z_G_H_h2:
499 ASM_TYPE_DIRECTIVE(Constants_log_80_Z_G_H_h2,@object)
500 // Z2 - 16 bit fixed, G2 and H2 IEEE single, h2 IEEE double
501 data4 0x00008000,0x3F800000,0x00000000,0x00000000
502 data4 0x00000000,0x00000000,0x00000000,0x00000000
503 data4 0x00007F81,0x3F7F00F8,0x3B7F875D,0x00000000
504 data4 0x211398BF,0xAD08B116,0x00003FDB,0x00000000
505 data4 0x00007F02,0x3F7E03F8,0x3BFF015B,0x00000000
506 data4 0xC376958E,0xB106790F,0x00003FDE,0x00000000
507 data4 0x00007E85,0x3F7D08E0,0x3C3EE393,0x00000000
508 data4 0x79A7679A,0xFD03F242,0x0000BFDA,0x00000000
509 data4 0x00007E08,0x3F7C0FC0,0x3C7E0586,0x00000000
510 data4 0x05E7AE08,0xF03F81C3,0x0000BFDF,0x00000000
511 data4 0x00007D8D,0x3F7B1880,0x3C9E75D2,0x00000000
512 data4 0x049EB22F,0xD1B87D3C,0x00003FDE,0x00000000
513 data4 0x00007D12,0x3F7A2328,0x3CBDC97A,0x00000000
514 data4 0x3A9E81E0,0xFABC8B95,0x00003FDF,0x00000000
515 data4 0x00007C98,0x3F792FB0,0x3CDCFE47,0x00000000
516 data4 0x7C4B5443,0xF5F3653F,0x00003FDF,0x00000000
517 data4 0x00007C20,0x3F783E08,0x3CFC15D0,0x00000000
518 data4 0xF65A1773,0xE78AB204,0x00003FE0,0x00000000
519 data4 0x00007BA8,0x3F774E38,0x3D0D874D,0x00000000
520 data4 0x7B8EF695,0xDB7CBFFF,0x0000BFE0,0x00000000
521 data4 0x00007B31,0x3F766038,0x3D1CF49B,0x00000000
522 data4 0xCF773FB3,0xC0241AEA,0x0000BFE0,0x00000000
523 data4 0x00007ABB,0x3F757400,0x3D2C531D,0x00000000
524 data4 0xC9539FDF,0xFC8F4D48,0x00003FE1,0x00000000
525 data4 0x00007A45,0x3F748988,0x3D3BA322,0x00000000
526 data4 0x954665C2,0x9CD035FB,0x0000BFE1,0x00000000
527 data4 0x000079D1,0x3F73A0D0,0x3D4AE46F,0x00000000
528 data4 0xDD367A30,0xEC9017C7,0x00003FE1,0x00000000
529 data4 0x0000795D,0x3F72B9D0,0x3D5A1756,0x00000000
530 data4 0xCB11189C,0xEE6625D3,0x0000BFE1,0x00000000
531 data4 0x000078EB,0x3F71D488,0x3D693B9D,0x00000000
532 data4 0xBE11C424,0xA49C8DB5,0x0000BFE0,0x00000000
533 ASM_SIZE_DIRECTIVE(Constants_log_80_Z_G_H_h2)
535 .align 64
536 Constants_log_80_h3_G_H:
537 ASM_TYPE_DIRECTIVE(Constants_log_80_h3_G_H,@object)
538 // h3 IEEE double extended, H3 and G3 IEEE single
539 data4 0x112666B0,0xAAACAAB1,0x00003FD3,0x3F7FFC00
540 data4 0x9B7FAD21,0x90051030,0x00003FD8,0x3F7FF400
541 data4 0xF4D783C4,0xA6B46F46,0x00003FDA,0x3F7FEC00
542 data4 0x11C6DDCA,0xDA148D88,0x0000BFD8,0x3F7FE400
543 data4 0xCA964D95,0xCE65C1D8,0x0000BFD8,0x3F7FDC00
544 data4 0x23412D13,0x883838EE,0x0000BFDB,0x3F7FD400
545 data4 0x983ED687,0xB7E5CFA1,0x00003FDB,0x3F7FCC08
546 data4 0xE3C3930B,0xDBE23B16,0x0000BFD9,0x3F7FC408
547 data4 0x48AA4DFC,0x9B92F1FC,0x0000BFDC,0x3F7FBC10
548 data4 0xCE9C8F7E,0x9A8CEB15,0x0000BFD9,0x3F7FB410
549 data4 0x0DECE74A,0x8C220879,0x00003FDC,0x3F7FAC18
550 data4 0x2F053150,0xB25CA912,0x0000BFDA,0x3F7FA420
551 data4 0xD9A5BE20,0xA5876555,0x00003FDB,0x3F7F9C20
552 data4 0x2053F087,0xC919BB6E,0x00003FD9,0x3F7F9428
553 data4 0x041E9A77,0xB70BDA79,0x00003FDC,0x3F7F8C30
554 data4 0xEA1C9C30,0xF18A5C08,0x00003FDA,0x3F7F8438
555 data4 0x796D89E5,0xA3790D84,0x0000BFDD,0x3F7F7C40
556 data4 0xA2915A3A,0xE1852369,0x0000BFDD,0x3F7F7448
557 data4 0xA39ED868,0xD803858F,0x00003FDC,0x3F7F6C50
558 data4 0x9417EBB7,0xB2EEE356,0x0000BFDD,0x3F7F6458
559 data4 0x9BB0D07F,0xED5C1F8A,0x0000BFDC,0x3F7F5C68
560 data4 0xE87C740A,0xD6D201A0,0x0000BFDD,0x3F7F5470
561 data4 0x1CA74025,0xE8DEBF5E,0x00003FDC,0x3F7F4C78
562 data4 0x1F34A7EB,0x9A995A97,0x0000BFDC,0x3F7F4488
563 data4 0x359EED97,0x9CB0F742,0x0000BFDA,0x3F7F3C90
564 data4 0xBBC6A1C8,0xD6F833C2,0x0000BFDD,0x3F7F34A0
565 data4 0xE71090EC,0xE1F68F2A,0x00003FDC,0x3F7F2CA8
566 data4 0xC160A74F,0xD1881CF1,0x0000BFDB,0x3F7F24B8
567 data4 0xD78CB5A4,0x9AD05AE2,0x00003FD6,0x3F7F1CC8
568 data4 0x9A77DC4B,0xE658CB8E,0x0000BFDD,0x3F7F14D8
569 data4 0x6BD6D312,0xBA281296,0x00003FDC,0x3F7F0CE0
570 data4 0xF95210D0,0xB478BBEB,0x0000BFDB,0x3F7F04F0
571 data4 0x38800100,0x39400480,0x39A00640,0x39E00C41 // H's start here
572 data4 0x3A100A21,0x3A300F22,0x3A4FF51C,0x3A6FFC1D
573 data4 0x3A87F20B,0x3A97F68B,0x3AA7EB86,0x3AB7E101
574 data4 0x3AC7E701,0x3AD7DD7B,0x3AE7D474,0x3AF7CBED
575 data4 0x3B03E1F3,0x3B0BDE2F,0x3B13DAAA,0x3B1BD766
576 data4 0x3B23CC5C,0x3B2BC997,0x3B33C711,0x3B3BBCC6
577 data4 0x3B43BAC0,0x3B4BB0F4,0x3B53AF6D,0x3B5BA620
578 data4 0x3B639D12,0x3B6B9444,0x3B7393BC,0x3B7B8B6D
579 ASM_SIZE_DIRECTIVE(Constants_log_80_h3_G_H)
581 .align 64
582 Constant_half:
583 ASM_TYPE_DIRECTIVE(Constant_half,@object)
584 data4 0x00000000,0x80000000,0x00003FFE
585 ASM_SIZE_DIRECTIVE(Constant_half)
587 GR_Expo_Range = r32
588 GR_Flag = r33
589 GR_Table_Ptr = r34
591 GR_Table_Ptr1 = r35
592 GR_BIAS = r35
594 GR_Index1 = r36
595 GR_sign_mask = r36
597 GR_Index2 = r37
598 GR_Expo_X = r37
600 GR_signif_Z = r38
601 GR_M = r38
603 GR_X_0 = r39
604 GR_Mask = r39
606 GR_X_1 = r40
607 GR_W1_ptr = r40
609 GR_W2_ptr = r41
610 GR_X_2 = r41
612 GR_Z_1 = r42
613 GR_M2 = r42
615 GR_M1 = r43
616 GR_Z_2 = r43
618 GR_N = r44
619 GR_k = r44
621 GR_Big_Pos_Exp = r45
624 GR_BIAS_p_k = r47
625 GR_BIASed_exp_y = r47
627 GR_Big_Neg_Exp = r48
628 GR_Index3 = r48
629 GR_temp = r48
631 GR_vsm_expo = r49
632 GR_y_sign = r49
634 GR_T1_ptr = r50
635 GR_T2_ptr = r51
636 GR_N_fix = r52
637 GR_exp_y = r53
638 GR_signif_y = r54
639 GR_exp_and_sign_y = r55
640 GR_low_order_bit = r56
641 GR_get_exp_mask = r57
642 GR_exponent_zero = r58
644 // ** Registers for unwind support
646 GR_SAVE_PFS = r59
647 GR_SAVE_B0 = r60
648 GR_SAVE_GP = r61
649 GR_Parameter_X = r62
650 GR_Parameter_Y = r63
651 GR_Parameter_RESULT = r64
652 GR_Parameter_TAG = r65
654 FR_X = f8
655 FR_Y = f9
656 FR_RESULT = f99
658 // **
660 FR_Input_X = f8
661 FR_Output = f8
662 FR_Input_Y = f9
664 FR_Neg = f10
665 FR_P_hi = f10
666 FR_X = f10
668 FR_Half = f11
669 FR_h_3 = f11
670 FR_poly_hi = f11
672 FR_Sgn = f12
674 FR_Neg_X = f13
675 FR_half_W = f13
677 FR_X_cor = f14
678 FR_P_lo = f14
680 FR_W = f15
682 FR_X_lo = f32
684 FR_S = f33
685 FR_W3 = f33
687 FR_Y_hi = f34
688 FR_logx_hi = f34
690 FR_Z = f35
691 FR_logx_lo = f35
692 FR_GS_hi = f35
693 FR_Y_lo = f35
695 FR_r_cor = f36
696 FR_Scale = f36
698 FR_G_1 = f37
699 FR_G = f37
700 FR_Wsq = f37
701 FR_L_Inv = f37
702 FR_temp = f37
704 FR_H_1 = f38
705 FR_H = f38
706 FR_W4 = f38
707 FR_float_N = f38
709 FR_h = f39
710 FR_h_1 = f39
711 FR_N = f39
712 FR_P_7 = f39
714 FR_G_2 = f40
715 FR_P_8 = f40
716 FR_L_hi = f40
718 FR_H_2 = f41
719 FR_L_lo = f41
720 FR_A_1 = f41
722 FR_h_2 = f42
723 FR_P_6 = f42
725 FR_abs_W = f43
726 FR_W1 = f43
728 FR_G_3 = f44
729 FR_P_8 = f44
730 FR_T1 = f44
732 FR_log2_hi = f45
733 FR_W2 = f45
735 FR_GS_lo = f46
736 FR_T2 = f46
738 FR_W_1_p1 = f47
739 FR_H_3 = f47
741 FR_float_N = f48
743 FR_P_4 = f49
744 FR_A_2 = f49
746 FR_Q_4 = f50
747 FR_r4 = f50
749 FR_Q_3 = f51
750 FR_A_3 = f51
752 FR_Q_2 = f52
753 FR_P_2 = f52
755 FR_Q_1 = f53
756 FR_P_1 = f53
757 FR_T = f53
759 FR_Wp1 = f54
760 FR_Q_5 = f54
761 FR_P_3 = f54
763 FR_Q_6 = f55
765 FR_log2_lo = f56
766 FR_Two = f56
768 FR_Big = f57
770 FR_neg_2_mK = f58
771 FR_NBig = f58
773 FR_r = f59
775 FR_poly_lo = f60
777 FR_poly = f61
779 FR_P_5 = f62
781 FR_rsq = f63
783 FR_Result = f99
784 FR_Result_small = f100
785 FR_Result_big = f101
787 .section .text
788 .proc powl#
789 .global powl#
790 .align 64
792 powl:
793 { .mfi
794 alloc GR_Expo_Range = ar.pfs,0,30,4,0
795 (p0) fclass.m.unc p7, p13 = FR_Input_Y, 0x1E7
796 nop.i 0
797 }
798 { .mfi
799 (p0) getf.exp GR_exp_and_sign_y = FR_Input_Y
800 //
801 // Save State
802 //
803 (p0) fclass.m.unc p6, p12 = FR_Input_X, 0x1E7
804 nop.i 0
805 };;
806 { .mfi
807 (p0) getf.sig GR_signif_y = FR_Input_Y
808 (p0) fcmp.eq.unc.s1 p12, p13 = FR_Input_X, f1
809 nop.i 0
810 }
811 { .mfi
812 nop.m 999
813 //
814 // Check for y = 1
815 // Identify EM unsupporteds.
816 // Load FR_half = .5
817 //
818 (p0) fadd.s1 FR_Two = f1, f1
819 //
820 // Load 1/2 in GP register
821 //
822 nop.i 0
823 }
824 ;;
826 { .mmi
827 nop.m 999
828 (p0) addl GR_Table_Ptr = @ltoff(Constant_half#), gp
829 nop.i 999
830 }
831 ;;
833 { .mmi
834 ld8 GR_Table_Ptr = [GR_Table_Ptr]
835 nop.m 999
836 nop.i 999
837 }
838 ;;
840 { .mlx
841 (p0) ldfe FR_Half =[GR_Table_Ptr],0
842 (p0) movl GR_get_exp_mask = 0x1FFFF ;;
843 }
845 { .mfi
846 nop.m 999
847 (p0) fclass.nm.unc p9, p15 = FR_Input_Y, 0x1FF
848 //
849 // Create FR_Two = 2
850 // Get exp and significand of Y
851 // Crate Masks
852 // sgn = 1
853 //
854 (p0) and GR_exp_y = GR_get_exp_mask,GR_exp_and_sign_y
855 }
856 { .mlx
857 nop.m 999
858 (p0) movl GR_exponent_zero = 0xFFFF ;;
859 }
860 { .mfi
861 nop.m 999
862 (p0) mov FR_Sgn = f1
863 nop.i 999
864 }
865 { .mfi
866 nop.m 999
867 (p0) fcmp.eq.unc.s1 p10, p11 = FR_Input_Y, f1
868 nop.i 999 ;;
869 }
870 { .mfb
871 nop.m 999
872 //
873 // Identify NatVals, NaNs, Infs, and Zeros.
874 // Load Half
875 //
876 (p0) fclass.nm.unc p8, p14 = FR_Input_X, 0x1FF
877 //
878 // Remove sign bit from exponent of y.
879 // Check for x = 1
880 //
881 (p6) br.cond.spnt L(POWL_64_SPECIAL) ;;
882 }
883 { .mib
884 nop.m 999
885 nop.i 999
886 (p7) br.cond.spnt L(POWL_64_SPECIAL) ;;
887 }
888 { .mib
889 nop.m 999
890 nop.i 999
891 (p8) br.cond.spnt L(POWL_64_UNSUPPORT) ;;
892 }
893 { .mib
894 nop.m 999
895 nop.i 999
896 (p9) br.cond.spnt L(POWL_64_UNSUPPORT) ;;
897 }
898 { .mfi
899 (p0) cmp.lt.unc p9, p0 = GR_exp_y,GR_exponent_zero
900 (p0) fcmp.lt.unc.s1 p6, p13 = FR_Input_X, f0
901 //
902 // Branch on Infs, Nans, Zeros, and Natvals
903 // Check to see that exponent < 0
904 //
905 (p0) sub GR_exp_y = GR_exp_y,GR_exponent_zero
906 }
907 // x not zero, is y ==2?
908 { .mfi
909 nop.m 999
910 (p11) fcmp.eq.unc.s1 p7, p14 = FR_Input_Y, FR_Two
911 nop.i 999 ;;
912 }
913 { .mfb
914 nop.m 999
915 (p9) fcmp.lt.unc.s1 p9, p0 = FR_Input_X, f0
916 (p7) br.cond.spnt L(POWL_64_SQUARE) ;; // Branch if x not zero and y=2
917 }
918 { .mfi
919 nop.m 999
920 (p6) fmerge.ns FR_Neg_X = FR_Input_X, FR_Input_X
921 nop.i 999 ;;
922 }
923 { .mfi
924 nop.m 999
925 (p10) fmpy.s0 FR_Result = FR_Input_X, f1
926 //
927 // For y = 1, compute result = x
928 // For x = 1, compute 1
929 // When Y is one return X and possible raise
930 // denormal operand exception.
931 // Remove exponent BIAS
932 //
933 (p6) shl GR_exp_and_sign_y= GR_signif_y,GR_exp_y ;;
934 }
935 { .mfi
936 (p9) or GR_exp_and_sign_y = 0xF,GR_signif_y
937 (p12) fma.s0 FR_Result = FR_Input_Y, f0, f1
938 nop.i 999 ;;
939 }
940 { .mii
941 nop.m 999
942 (p6) extr.u GR_exp_y = GR_exp_and_sign_y,63,1 ;;
943 (p6) cmp.ne.unc p9, p0 = GR_exp_y, r0
944 }
945 { .mii
946 nop.m 999
947 //
948 // Both predicates can be set.
949 // Don't consider y's < 1.
950 //
951 (p6) shl GR_signif_y= GR_exp_and_sign_y,1 ;;
952 //
953 // Is shift off integer part of y.
954 // Get y's even or odd bit.
955 //
956 (p6) cmp.ne.unc p8, p0 = GR_signif_y, r0
957 }
958 { .mib
959 nop.m 999
960 nop.i 999
961 //
962 // Is the fractional part of the y = 0?
963 // Is the integer even or odd.
964 //
965 (p10) br.cond.spnt L(POWL_64_RETURN) ;;
966 }
967 { .mib
968 nop.m 999
969 nop.i 999
970 (p12) br.cond.spnt L(POWL_64_RETURN) ;;
971 }
972 { .mib
973 nop.m 999
974 nop.i 999
975 (p8) br.cond.spnt L(POWL_64_XNEG) ;;
976 }
977 { .mfi
978 nop.m 999
979 (p9) fmerge.ns FR_Sgn = FR_Sgn, FR_Sgn
980 nop.i 999
981 }
982 { .mfi
983 nop.m 999
984 (p0) fcmp.eq.unc.s0 p11, p0 = FR_Input_Y, FR_Half
985 nop.i 999 ;;
986 }
987 //
988 // Raise possible denormal operand exception for both
989 // X and Y.
990 //
991 { .mfb
992 nop.m 999
993 //
994 // Branch for (x < 0) and Y not an integer.
995 //
996 (p0) fcmp.eq.unc.s0 p12, p0 = FR_Input_X, f1
997 //
998 // For x < 0 and y integer, make x positive
999 // For x < 0 and y odd integer,, set sign = -1.
1000 //
1001 (p11) br.cond.spnt L(POWL_64_SQRT) ;;
1002 }
1003 { .mmf
1004 (p0) cmp.eq.unc p15, p14 = r0, r0
1005 nop.m 999
1006 (p13) fnorm.s1 FR_Z = FR_Input_X ;;
1007 }
1008 { .mfi
1009 nop.m 999
1010 (p6) fnorm.s1 FR_Z = FR_Neg_X
1011 nop.i 999
1012 }
1013 ;;
1015 //
1016 // Branch to embedded sqrt(x)
1017 //
1018 //
1019 // Computes ln( x ) to extra precision
1020 // Input FR 1: FR_X
1021 // Output FR 2: FR_Y_hi
1022 // Output FR 3: FR_Y_lo
1023 // Output PR 1: PR_Safe
1024 //
1026 { .mmi
1027 nop.m 999
1028 (p0) addl GR_Table_Ptr = @ltoff(Constants_log_80_Z_G_H_h1#), gp
1029 nop.i 999
1030 }
1031 ;;
1033 { .mmi
1034 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1035 nop.m 999
1036 nop.i 999
1037 }
1038 ;;
1041 { .mlx
1042 nop.m 999
1043 (p0) movl GR_BIAS = 0x000000000000FFFF ;;
1044 }
1045 { .mfi
1046 nop.m 999
1047 (p0) fsub.s1 FR_W = FR_Z, f1
1048 nop.i 999 ;;
1049 }
1050 //
1051 // Z = Norm(X) - both + and - case
1052 // Set Safe = True
1053 //
1054 { .mmb
1055 (p0) getf.sig GR_signif_Z = FR_Z
1056 (p0) getf.exp GR_N = FR_Z
1057 nop.b 999 ;;
1058 }
1059 { .mii
1060 nop.m 999
1061 //
1062 // Get significand of Z
1063 // W = Z - 1
1064 //
1065 (p0) extr.u GR_Index1 = GR_signif_Z, 59, 4 ;;
1066 //
1067 // Index1 = High order 4 bits of Z
1068 // X_0 = High order 15 bit of Z
1069 //
1070 (p0) shl GR_Index1 = GR_Index1,5 ;;
1071 }
1072 { .mfi
1073 nop.m 999
1074 //
1075 // Add offset to Index1 ptr.
1076 //
1077 (p0) fabs FR_abs_W = FR_W
1078 //
1079 // BIAS = 0x000...FFFF
1080 // Adjust Index1 ptr ( x 32) .
1081 //
1082 (p0) add GR_Index1 = GR_Index1,GR_Table_Ptr
1083 }
1084 { .mmi
1085 nop.m 999 ;;
1086 (p0) ld2 GR_Z_1 =[GR_Index1],4
1087 (p0) extr.u GR_X_0 = GR_signif_Z, 49, 15
1088 }
1089 ;;
1091 { .mmi
1092 nop.m 999
1093 (p0) addl GR_Table_Ptr = @ltoff(Constants_log_80_Z_G_H_h2#), gp
1094 nop.i 999
1095 }
1096 ;;
1098 { .mmi
1099 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1100 nop.m 999
1101 nop.i 999
1102 }
1103 ;;
1106 { .mmi
1107 (p0) ldfs FR_G_1 = [GR_Index1],4 ;;
1108 (p0) ldfs FR_H_1 = [GR_Index1],8
1109 nop.i 999 ;;
1110 }
1111 //
1112 // Adjust Index2 (x 32).
1113 //
1114 { .mfi
1115 (p0) ldfe FR_h_1 = [GR_Index1],0
1116 nop.f 999
1117 (p0) pmpyshr2.u GR_X_1 = GR_X_0,GR_Z_1,15 ;;
1118 }
1119 { .mmi
1120 nop.m 999 ;;
1121 //
1122 // load Z_1 from Index1
1123 // abs_W = |W|
1124 // Point to Table2
1125 //
1126 (p0) getf.exp GR_M = FR_abs_W
1127 //
1128 // M = M - BIAS
1129 // Load G_1
1130 // N = exponent of Z
1131 //
1132 nop.i 999;;
1133 }
1134 { .mmi
1135 nop.m 999
1136 nop.m 999
1137 nop.i 999;;
1138 }
1139 { .mmi
1140 nop.m 999
1141 nop.m 999
1142 nop.i 999;;
1143 }
1144 { .mmi
1145 nop.m 999
1146 nop.m 999
1147 (p0) extr.u GR_Index2 = GR_X_1, 6, 4 ;;
1148 }
1149 { .mii
1150 nop.m 999
1151 //
1152 // Extract Index2
1153 // Load H_1
1154 // Is -8 > M ?
1155 //
1156 (p0) shl GR_Index2=GR_Index2,5 ;;
1157 (p0) add GR_Index2 = GR_Index2, GR_Table_Ptr
1158 }
1159 //
1160 // M = exponent of abs_W
1161 // X_1 = X_0 * Z_1
1162 //
1163 { .mii
1164 (p0) sub GR_M = GR_M, GR_BIAS
1165 nop.i 999 ;;
1166 (p0) cmp.gt.unc p7, p14 = -8, GR_M
1167 }
1168 { .mib
1169 nop.m 999
1170 nop.i 999
1171 (p7) br.cond.spnt L(LOGL80_NEAR) ;;
1172 }
1173 //
1174 // Load h_1
1175 // Possible branch out.
1176 // Add offset of table to Index2
1177 //
1178 { .mfi
1179 (p0) ld2 GR_Z_2 =[GR_Index2],4
1180 (p0) fmerge.se FR_S = f1,FR_Z
1181 (p0) sub GR_N = GR_N, GR_BIAS
1182 }
1183 ;;
1185 { .mmi
1186 nop.m 999
1187 (p0) addl GR_Table_Ptr = @ltoff(Constants_log_80_h3_G_H#), gp
1188 nop.i 999
1189 }
1190 ;;
1192 { .mmi
1193 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1194 nop.m 999
1195 nop.i 999
1196 }
1197 ;;
1199 //
1200 // load Z_2
1201 // N - BIAS
1202 // Point to Table 3.
1203 // S = merging of Z and 1.0
1204 //
1205 { .mmi
1206 (p0) ldfs FR_G_2 = [GR_Index2],4
1207 (p0) setf.sig FR_float_N = GR_N
1208 (p0) add GR_Table_Ptr1 = 0x200,GR_Table_Ptr ;;
1209 }
1210 //
1211 // load G_2
1212 // X_2 = X_1 * Z_2
1213 // Add offset to Table 2 ptr.
1214 // float_N = significand of N
1215 //
1216 { .mmi
1217 (p0) ldfs FR_H_2 = [GR_Index2],8 ;;
1218 //
1219 // load H_2
1220 // G = G * G_2
1221 //
1222 (p0) ldfe FR_h_2 = [GR_Index2],0
1223 (p0) pmpyshr2.u GR_X_2 = GR_X_1,GR_Z_2,15 ;;
1224 }
1225 { .mmi
1226 nop.m 999
1227 nop.m 999
1228 nop.i 999;;
1229 }
1230 { .mmi
1231 nop.m 999
1232 nop.m 999
1233 nop.i 999;;
1234 }
1235 { .mmi
1236 nop.m 999
1237 nop.m 999
1238 nop.i 999;;
1239 }
1240 { .mii
1241 nop.m 999
1242 nop.i 999 ;;
1243 (p0) extr.u GR_Index3 = GR_X_2, 1, 5 ;;
1244 }
1245 { .mfi
1246 (p0) shladd GR_Table_Ptr1 = GR_Index3,2,GR_Table_Ptr1
1247 nop.f 999
1248 //
1249 // h = h_1 + h_2
1250 // Adjust Index3
1251 //
1252 (p0) shladd GR_Index3 = GR_Index3,4,GR_Table_Ptr ;;
1253 }
1254 { .mmb
1255 nop.m 999
1256 (p0) ldfe FR_h_3 = [GR_Index3],12
1257 nop.b 999 ;;
1258 }
1259 { .mmf
1260 (p0) ldfs FR_H_3 = [GR_Table_Ptr1],0
1261 //
1262 // float_N = Make N a fp number
1263 // Load h_3
1264 // Get pointer to Q table.
1265 //
1266 (p0) ldfs FR_G_3 = [GR_Index3],0
1267 (p0) fmpy.s1 FR_G = FR_G_1, FR_G_2
1268 }
1269 ;;
1271 { .mmi
1272 nop.m 999
1273 (p0) addl GR_Table_Ptr = @ltoff(Constants_log_80_Q#), gp
1274 nop.i 999
1275 }
1276 ;;
1278 { .mmi
1279 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1280 nop.m 999
1281 nop.i 999
1282 }
1283 ;;
1287 { .mfi
1288 (p0) ldfe FR_log2_hi = [GR_Table_Ptr],16
1289 (p0) fadd.s1 FR_H = FR_H_1, FR_H_2
1290 nop.i 999 ;;
1291 }
1292 { .mmf
1293 nop.m 999
1294 //
1295 // G = G_1 * G_2 * G_3
1296 //
1297 (p0) ldfe FR_log2_lo = [GR_Table_Ptr],16
1298 //
1299 // load h_2
1300 // H = H_1 + H_2
1301 // Get Index3
1302 //
1303 (p0) fadd.s1 FR_h = FR_h_1, FR_h_2 ;;
1304 }
1305 //
1306 // Load log2_lo part
1307 // r = G*S -1
1308 //
1309 { .mfi
1310 (p0) ldfe FR_Q_6 = [GR_Table_Ptr],16
1311 //
1312 // Load H_3
1313 //
1314 (p0) fcvt.xf FR_float_N = FR_float_N
1315 nop.i 999 ;;
1316 }
1317 //
1318 // Load Q_6
1319 //
1320 { .mmi
1321 (p0) ldfe FR_Q_5 = [GR_Table_Ptr],16 ;;
1322 (p0) ldfe FR_Q_4 = [GR_Table_Ptr],16
1323 nop.i 999 ;;
1324 }
1325 { .mmi
1326 (p0) ldfe FR_Q_3 = [GR_Table_Ptr],16 ;;
1327 (p0) ldfe FR_Q_2 = [GR_Table_Ptr],16
1328 nop.i 999 ;;
1329 }
1330 { .mmf
1331 nop.m 999
1332 //
1333 // poly_lo = Q_5 + r * Q_6
1334 // Load Q_2
1335 // rsq = r * r
1336 //
1337 (p0) ldfe FR_Q_1 = [GR_Table_Ptr],16
1338 //
1339 // h = h_1 + h_2 + h_3
1340 // H = H_1 + H_2 + H_3
1341 // Load G_3.
1342 // Begin Loading Q's - load log2_hi part
1343 //
1344 (p0) fmpy.s1 FR_G = FR_G, FR_G_3
1345 }
1346 { .mfi
1347 nop.m 999
1348 (p0) fadd.s1 FR_H = FR_H, FR_H_3
1349 nop.i 999
1350 }
1351 ;;
1353 //
1354 // Y_lo = poly + Y_lo
1355 //
1357 { .mmi
1358 nop.m 999
1359 (p0) addl GR_Table_Ptr = @ltoff(Constants_exp_64_Arg#), gp
1360 nop.i 999
1361 }
1362 ;;
1364 { .mmi
1365 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1366 nop.m 999
1367 nop.i 999
1368 }
1369 ;;
1372 { .mfi
1373 nop.m 999
1374 (p0) fadd.s1 FR_h = FR_h, FR_h_3
1375 nop.i 999 ;;
1376 }
1377 { .mfi
1378 nop.m 999
1379 //
1380 // Load Q_5
1381 //
1382 (p0) fmpy.s1 FR_GS_hi = FR_G, FR_S
1383 nop.i 999
1384 }
1385 { .mfi
1386 nop.m 999
1387 (p0) fms.s1 FR_r = FR_G, FR_S, f1
1388 nop.i 999 ;;
1389 }
1390 { .mfi
1391 nop.m 999
1392 (p0) fma.s1 FR_poly_lo = FR_r, FR_Q_6, FR_Q_5
1393 nop.i 999
1394 }
1395 { .mfi
1396 nop.m 999
1397 //
1398 // GS_hi = G*S
1399 // Load Q_4
1400 //
1401 (p0) fsub.s1 FR_r_cor = FR_GS_hi, f1
1402 nop.i 999 ;;
1403 }
1404 { .mfi
1405 nop.m 999
1406 (p0) fms.s1 FR_GS_lo = FR_G, FR_S, FR_GS_hi
1407 nop.i 999
1408 }
1409 { .mfi
1410 nop.m 999
1411 (p0) fma.s1 FR_poly = FR_r, FR_Q_2, FR_Q_1
1412 nop.i 999 ;;
1413 }
1414 { .mfi
1415 nop.m 999
1416 //
1417 // Load Q_3
1418 // r_cor = GS_hi -1
1419 // GS_lo = G*S - GS_hi
1420 //
1421 (p0) fmpy.s1 FR_rsq = FR_r, FR_r
1422 nop.i 999
1423 }
1424 { .mfi
1425 nop.m 999
1426 (p0) fma.s1 FR_G = FR_float_N, FR_log2_hi, FR_H
1427 nop.i 999 ;;
1428 }
1429 { .mfi
1430 nop.m 999
1431 //
1432 // poly = poly_hi + rsq * poly_lo
1433 // Tbl = float_N*log2_hi + H
1434 //
1435 (p0) fma.s1 FR_Y_lo = FR_float_N, FR_log2_lo, FR_h
1436 nop.i 999 ;;
1437 }
1438 { .mfi
1439 nop.m 999
1440 //
1441 // r_cor = r_cor - r
1442 // poly_hi = r * Q_2 + Q_1
1443 //
1444 (p0) fma.s1 FR_poly_lo = FR_r, FR_poly_lo, FR_Q_4
1445 nop.i 999
1446 }
1447 { .mfi
1448 nop.m 999
1449 //
1450 // Load Q_1
1451 //
1452 (p0) fsub.s1 FR_r_cor = FR_r_cor, FR_r
1453 nop.i 999 ;;
1454 }
1455 { .mfi
1456 nop.m 999
1457 //
1458 // Y_lo = float_N*log2_lo + h
1459 //
1460 (p0) fadd.s1 FR_Y_hi = FR_G, FR_r
1461 nop.i 999 ;;
1462 }
1463 { .mfi
1464 nop.m 999
1465 //
1466 // poly_lo = Q_4 + r * poly_lo;;
1467 // r_cor = r_cor + GS_lo;;
1468 //
1469 (p0) fma.s1 FR_poly_lo = FR_r, FR_poly_lo, FR_Q_3
1470 nop.i 999
1471 }
1472 { .mfi
1473 nop.m 999
1474 (p0) fadd.s1 FR_r_cor = FR_r_cor, FR_GS_lo
1475 nop.i 999 ;;
1476 }
1477 { .mfi
1478 nop.m 999
1479 (p0) fadd.s1 FR_r_cor = FR_r_cor, FR_Y_lo
1480 nop.i 999
1481 }
1482 { .mfi
1483 nop.m 999
1484 //
1485 // poly_lo = Q_3 + r * poly_lo;;
1486 //
1487 (p0) fma.s1 FR_poly = FR_rsq, FR_poly_lo, FR_poly
1488 nop.i 999 ;;
1489 }
1490 { .mfi
1491 nop.m 999
1492 (p0) fsub.s1 FR_Y_lo = FR_G, FR_Y_hi
1493 nop.i 999
1494 }
1495 { .mmi
1496 (p0) ldfe FR_L_Inv = [GR_Table_Ptr],16 ;;
1497 (p0) ldfe FR_L_hi = [GR_Table_Ptr],16
1498 nop.i 999 ;;
1499 }
1500 { .mfi
1501 (p0) ldfe FR_L_lo = [GR_Table_Ptr],16
1502 nop.f 999
1503 nop.i 999 ;;
1504 }
1505 { .mfi
1506 nop.m 999
1507 //
1508 // Y_hi = Tbl + r
1509 // r_cor = r_cor + Y_lo
1510 //
1511 (p0) fma.s1 FR_poly = FR_rsq, FR_poly, FR_r_cor
1512 nop.i 999 ;;
1513 }
1514 { .mfi
1515 nop.m 999
1516 // Y_lo = Tbl - Y_hi
1517 // poly = rsq * poly + r_cor
1518 //
1519 (p0) fadd.s1 FR_Y_lo = FR_Y_lo, FR_r
1520 nop.i 999 ;;
1521 }
1522 { .mfb
1523 nop.m 999
1524 //
1525 // Y_lo = Y_lo + r
1526 //
1527 (p0) fadd.s1 FR_Y_lo = FR_Y_lo, FR_poly
1528 //
1529 // Load L_Inv
1530 // Load L_hi
1531 // Load L_lo
1532 // all long before they are needed.
1533 // They are used in LOGL_RETURN PATH
1534 //
1535 br.cond.sptk L(LOGL_RETURN) ;;
1536 }
1537 L(LOGL80_NEAR):
1538 //
1539 // Branch LOGL80_NEAR
1540 //
1542 { .mmi
1543 nop.m 999
1544 (p0) addl GR_Table_Ptr = @ltoff(Constants_log_80_P#), gp
1545 nop.i 999
1546 }
1547 ;;
1549 { .mmi
1550 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1551 nop.m 999
1552 nop.i 999
1553 }
1554 ;;
1556 { .mfi
1557 nop.m 999
1558 (p0) fmpy.s1 FR_Wsq = FR_W, FR_W
1559 (p0) add GR_Table_Ptr1 = 0x50,GR_Table_Ptr
1560 }
1561 //
1562 // Adjust ptr to 1/2
1563 // Adjust Ptr1 to P_4
1564 //
1565 { .mmi
1566 (p0) ldfe FR_Half = [GR_Table_Ptr],16 ;;
1567 (p0) ldfe FR_P_4 = [GR_Table_Ptr1],16
1568 nop.i 999
1569 }
1570 //
1571 // Load 1/2
1572 //
1573 { .mmi
1574 (p0) ldfe FR_P_8 = [GR_Table_Ptr],16 ;;
1575 (p0) ldfe FR_P_3 = [GR_Table_Ptr1],16
1576 nop.i 999
1577 }
1578 { .mmi
1579 (p0) ldfe FR_P_7 = [GR_Table_Ptr],16 ;;
1580 (p0) ldfe FR_P_2 = [GR_Table_Ptr1],16
1581 nop.i 999
1582 }
1583 //
1584 // Load P_7
1585 // half_W = .5 * W
1586 // Load P_3
1587 //
1588 { .mmi
1589 (p0) ldfe FR_P_6 = [GR_Table_Ptr],16 ;;
1590 (p0) ldfe FR_P_1 = [GR_Table_Ptr1],16
1591 nop.i 999 ;;
1592 }
1593 //
1594 // Load P_6
1595 // Wsq = w * w
1596 // poly = w*P_4 + P_3
1597 // Load P_2
1598 //
1599 { .mfi
1600 (p0) ldfe FR_P_5 = [GR_Table_Ptr],16
1601 //
1602 // Load P_5
1603 // poly_lo = w * P_8 + P_7
1604 // Y_hi = w - (1/2)w*w
1605 // Load P_1
1606 //
1607 (p0) fmpy.s1 FR_W4 = FR_Wsq, FR_Wsq
1608 nop.i 999
1609 }
1610 { .mfi
1611 nop.m 999
1612 (p0) fmpy.s1 FR_W3 = FR_Wsq, FR_W
1613 nop.i 999
1614 }
1615 ;;
1617 //
1618 // Y_lo = W3 * poly + Y_lo
1619 //
1621 { .mmi
1622 nop.m 999
1623 (p0) addl GR_Table_Ptr = @ltoff(Constants_exp_64_Arg#), gp
1624 nop.i 999
1625 }
1626 ;;
1628 { .mmi
1629 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1630 nop.m 999
1631 nop.i 999
1632 }
1633 ;;
1636 { .mmi
1637 (p0) ldfe FR_L_Inv = [GR_Table_Ptr],16 ;;
1638 (p0) ldfe FR_L_hi = [GR_Table_Ptr],16
1639 nop.i 999 ;;
1640 }
1641 { .mfi
1642 (p0) ldfe FR_L_lo = [GR_Table_Ptr],16
1643 //
1644 // Load P_8
1645 // Load P_4
1646 //
1647 (p0) fmpy.s1 FR_half_W = FR_Half, FR_W
1648 nop.i 999 ;;
1649 }
1650 { .mfi
1651 nop.m 999
1652 (p0) fma.s1 FR_poly_lo = FR_W, FR_P_8,FR_P_7
1653 nop.i 999
1654 }
1655 { .mfi
1656 nop.m 999
1657 (p0) fma.s1 FR_poly = FR_W, FR_P_4, FR_P_3
1658 nop.i 999 ;;
1659 }
1660 { .mfi
1661 nop.m 999
1662 (p0) fnma.s1 FR_Y_hi = FR_W, FR_half_W, FR_W
1663 nop.i 999 ;;
1664 }
1665 { .mfi
1666 nop.m 999
1667 //
1668 // W4 = Wsq * Wsq
1669 // poly = w *poly + P_2
1670 //
1671 (p0) fma.s1 FR_poly_lo = FR_W, FR_poly_lo, FR_P_6
1672 nop.i 999
1673 }
1674 { .mfi
1675 nop.m 999
1676 (p0) fma.s1 FR_poly = FR_W, FR_poly, FR_P_2
1677 nop.i 999 ;;
1678 }
1679 { .mfi
1680 nop.m 999
1681 (p0) fsub.s1 FR_Y_lo = FR_W, FR_Y_hi
1682 nop.i 999 ;;
1683 }
1684 { .mfi
1685 nop.m 999
1686 //
1687 // poly = w * poly + P_1
1688 // w3 = wsq * w
1689 //
1690 (p0) fma.s1 FR_poly_lo = FR_W, FR_poly_lo, FR_P_5
1691 nop.i 999
1692 }
1693 { .mfi
1694 nop.m 999
1695 //
1696 // poly_lo = w * poly_lo + P_6
1697 // Y_lo = W - Y_hi
1698 //
1699 (p0) fma.s1 FR_poly = FR_W, FR_poly, FR_P_1
1700 nop.i 999 ;;
1701 }
1702 { .mfi
1703 nop.m 999
1704 (p0) fnma.s1 FR_Y_lo = FR_W, FR_half_W, FR_Y_lo
1705 nop.i 999 ;;
1706 }
1707 { .mfi
1708 nop.m 999
1709 //
1710 // poly_lo = w * poly_lo +
1711 // Y_lo = Y_lo - w * (1/2)w
1712 //
1713 (p0) fma.s1 FR_poly = FR_poly_lo, FR_W4, FR_poly
1714 nop.i 999 ;;
1715 }
1716 { .mfi
1717 nop.m 999
1718 //
1719 // Y_lo = (W-Y_hi) - w * (1/2)w
1720 // poly = W4* poly_lo + poly
1721 //
1722 (p0) fma.s1 FR_Y_lo = FR_poly, FR_W3, FR_Y_lo
1723 nop.i 999 ;;
1724 }
1725 L(LOGL_RETURN):
1726 { .mfi
1727 (p0) add GR_Expo_Range = 0x2,r0
1728 //
1729 // Load L_Inv
1730 // Load L_hi
1731 // Load L_lo
1732 // all long before they are needed.
1733 //
1734 //
1735 // kernel_log_80 computed ln(X)
1736 // and return logX_hi and logX_lo as results.
1737 // PR_pow_Safe set as well.
1738 //
1739 (p0) fmpy.s1 FR_X_lo = FR_Input_Y, FR_logx_lo
1740 //
1741 // Compute Y * (logX_hi + logX_lo)
1742 // P_hi -> X
1743 // P_lo -> X_cor
1744 // (Manipulate names so that inputs are in
1745 // the place kernel_exp expects them)
1746 // Set GR_Flag to 2
1747 // Set GR_Expo_Range to Double
1748 //
1749 // This function computes exp( x + x_cor)
1750 // Input FR 1: FR_X
1751 // Input FR 2: FR_X_cor
1752 // Input GR 1: GR_Flag
1753 // Input GR 2: GR_Expo_Range
1754 // Output FR 3: FR_Y_hi
1755 // Output FR 4: FR_Y_lo
1756 // Output FR 5: FR_Scale
1757 // Output PR 1: PR_Safe
1758 //
1759 (p0) cmp.eq.unc p15, p0 = r0, r0
1760 }
1761 ;;
1763 { .mmi
1764 (p0) addl GR_W1_ptr = @ltoff(Constants_exp_64_W1#), gp
1765 (p0) addl GR_W2_ptr = @ltoff(Constants_exp_64_W2#), gp
1766 (p0) add GR_Flag = 0x2,r0
1767 }
1768 ;;
1770 { .mmi
1771 ld8 GR_W1_ptr = [GR_W1_ptr]
1772 ld8 GR_W2_ptr = [GR_W2_ptr]
1773 (p0) cmp.ne.unc p7, p0 = 0x1, GR_Flag
1774 }
1775 ;;
1777 { .mlx
1778 nop.m 999
1779 (p0) movl GR_Mask = 0x1FFFF ;;
1780 }
1783 { .mlx
1784 nop.m 999
1785 (p0) movl GR_BIAS = 0x0FFFF ;;
1786 }
1787 { .mfi
1788 nop.m 999
1789 //
1790 // X_lo = Y * logX_lo
1791 //
1792 (p0) fma.s1 FR_P_hi = FR_Input_Y, FR_logx_hi,FR_X_lo
1793 nop.i 999 ;;
1794 }
1795 { .mfi
1796 nop.m 999
1797 //
1798 // Set Safe=True
1799 // Flag is always 2 for this routine
1800 //
1801 (p0) fmpy.s1 FR_float_N = FR_X, FR_L_Inv
1802 nop.i 999
1803 }
1804 { .mfi
1805 nop.m 999
1806 //
1807 // X_hi = Y * logX_hi + X_lo
1808 // Set GR_Flag = 2 for exp(x + xcor)
1809 //
1810 (p0) fms.s1 FR_P_lo= FR_Input_Y, FR_logx_hi, FR_P_hi
1811 nop.i 999 ;;
1812 }
1813 { .mmi
1814 nop.m 999 ;;
1815 (p0) getf.exp GR_Expo_X = FR_X
1816 nop.i 999 ;;
1817 }
1818 { .mfi
1819 (p0) and GR_Expo_X = GR_Expo_X, GR_Mask
1820 //
1821 // Calculate unBIASed exponent of X
1822 // Point to Table of W1s
1823 // Point to Table of W2s
1824 //
1825 (p0) fcvt.fx.s1 FR_N = FR_float_N
1826 nop.i 999 ;;
1827 }
1828 { .mfi
1829 nop.m 999
1830 (p0) fadd.s1 FR_P_lo = FR_P_lo, FR_X_lo
1831 //
1832 // Float_N = X * L_Inv
1833 // Create exponent BIAS
1834 // Get BIASed exponent of X
1835 //
1836 (p0) sub GR_Expo_X = GR_Expo_X, GR_BIAS ;;
1837 }
1838 { .mib
1839 (p0) cmp.gt.unc p9, p0 = -6, GR_Expo_X
1840 nop.i 999
1841 //
1842 // N = fcvt.fx(float_N)
1843 // If -6 > Expo_X, set P9
1844 //
1845 (p9) br.cond.spnt L(EXPL_SMALL)
1846 }
1847 ;;
1849 //
1850 // If expo_X < -6 goto exp_small
1851 //
1852 { .mmi
1853 nop.m 999
1854 (p0) addl GR_T1_ptr = @ltoff(Constants_exp_64_T1#), gp
1855 (p0) cmp.lt.unc p10, p0 = 14, GR_Expo_X
1856 }
1857 ;;
1859 { .mmi
1860 ld8 GR_T1_ptr = [GR_T1_ptr]
1861 nop.m 999
1862 nop.i 999
1863 }
1864 ;;
1866 { .mib
1867 nop.m 999
1868 nop.i 999
1869 //
1870 // If 14 < Expo_X, set P10
1871 // Create pointer to T1 table
1872 //
1873 (p10) br.cond.spnt L(EXPL_HUGE) ;;
1874 }
1877 { .mmi
1878 (p0) addl GR_Table_Ptr = @ltoff(Constants_exp_64_Exponents#), gp
1879 (p0) addl GR_T2_ptr = @ltoff(Constants_exp_64_T2#), gp
1880 nop.i 999
1881 }
1882 ;;
1884 { .mmi
1885 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1886 ld8 GR_T2_ptr = [GR_T2_ptr]
1887 nop.i 999
1888 }
1889 ;;
1892 { .mmi
1893 (p0) shladd GR_Table_Ptr = GR_Expo_Range,4,GR_Table_Ptr ;;
1894 //
1895 // Adjust T1_ptr by x 4 for single-precision values
1896 // Adjust T2_ptr by x 4 for single-precision values
1897 //
1898 (p0) ld8 GR_Big_Pos_Exp = [GR_Table_Ptr],8
1899 nop.i 999 ;;
1900 }
1901 //
1902 // Load double W1
1903 // Load +max exponent
1904 //
1905 { .mfi
1906 (p0) ld8 GR_Big_Neg_Exp = [GR_Table_Ptr],0
1907 //
1908 // If 14 < Expo_X, goto exp_huge
1909 //
1910 (p0) fcvt.xf FR_float_N = FR_N
1911 nop.i 999
1912 }
1913 ;;
1915 //
1916 // Load double W2
1917 // Load -max exponent
1918 // Load ptr to A's
1919 //
1921 { .mmi
1922 (p0) getf.sig GR_N_fix = FR_N
1923 (p0) addl GR_Table_Ptr = @ltoff(Constants_exp_64_A#), gp
1924 nop.i 999
1925 }
1926 ;;
1928 { .mmi
1929 ld8 GR_Table_Ptr = [GR_Table_Ptr]
1930 nop.m 999
1931 nop.i 999
1932 }
1933 ;;
1935 //
1936 // Load single T1
1937 // Load single T2
1938 // W_1_p1 = W_1 + 1
1939 //
1940 { .mmi
1941 (p0) ldfe FR_A_3 = [GR_Table_Ptr],16 ;;
1942 //
1943 // Load A_3
1944 // if k > big_pos_exp, set p14 and Safe=False
1945 //
1946 (p0) ldfe FR_A_2 = [GR_Table_Ptr],16
1947 (p0) extr.u GR_M1 = GR_N_fix, 6, 6
1948 }
1949 { .mmi
1950 nop.m 999 ;;
1951 (p0) shladd GR_W1_ptr = GR_M1,3,GR_W1_ptr
1952 //
1953 // float_N = fcvt.xf(N)
1954 // N_fix = significand of N
1955 // Create pointer to T2 table
1956 //
1957 (p0) extr.u GR_M2 = GR_N_fix, 0, 6
1958 }
1959 //
1960 // r = r + X_cor
1961 // Adjust W1_ptr by x 8 for double-precision values
1962 // Adjust W2_ptr by x 8 for double-precision values
1963 // Adjust Table_ptr by Expo_Rangex16
1964 //
1965 { .mmi
1966 (p0) shladd GR_T1_ptr = GR_M1,2,GR_T1_ptr ;;
1967 (p0) ldfd FR_W1 = [GR_W1_ptr],0
1968 (p0) shladd GR_W2_ptr = GR_M2,3,GR_W2_ptr
1969 }
1970 //
1971 // Load ptr to A's
1972 //
1973 { .mfi
1974 (p0) ldfs FR_T1 = [GR_T1_ptr],0
1975 (p0) fnma.s1 FR_r = FR_L_hi, FR_float_N, FR_X
1976 (p0) shladd GR_T2_ptr = GR_M2,2,GR_T2_ptr ;;
1977 }
1978 { .mmi
1979 (p0) ldfd FR_W2 = [GR_W2_ptr],0
1980 (p0) ldfs FR_T2 = [GR_T2_ptr],0
1981 //
1982 // r = x - L_hi * float_N
1983 // M2 = extr.u(N_fix,0,6)
1984 // M1 = extr.u(N_fix,6,6)
1985 //
1986 (p0) extr GR_k = GR_N_fix, 12, 52 ;;
1987 }
1988 //
1989 // Load A_1
1990 // poly = A_3 * r + A_2
1991 // rsq = r*r
1992 //
1993 { .mii
1994 (p0) add GR_BIAS_p_k = GR_BIAS, GR_k
1995 (p0) cmp.gt.unc p14,p15 = GR_k,GR_Big_Pos_Exp ;;
1996 (p15) cmp.lt p14,p15 = GR_k,GR_Big_Neg_Exp
1997 }
1998 //
1999 // BIAS_p_K = BIAS + k
2000 // T = T1 * T2
2001 //
2002 { .mfi
2003 (p0) setf.exp FR_Scale = GR_BIAS_p_k
2004 nop.f 999
2005 nop.i 999 ;;
2006 }
2007 { .mfi
2008 nop.m 999
2009 (p0) fnma.s1 FR_r = FR_L_lo, FR_float_N, FR_r
2010 nop.i 999
2011 }
2012 //
2013 // W = W_1_p1 * W2 + W1
2014 //
2015 { .mfi
2016 (p0) ldfe FR_A_1 = [GR_Table_Ptr],16
2017 nop.f 999
2018 nop.i 999 ;;
2019 }
2020 { .mfi
2021 nop.m 999
2022 (p0) fadd.s1 FR_W_1_p1 = FR_W1, f1
2023 nop.i 999 ;;
2024 }
2025 { .mfi
2026 nop.m 999
2027 //
2028 // k = extr.u(N_fix,0,6)
2029 // r = r - N * L_lo
2030 // Load ptr to Table of exponent thresholds.
2031 //
2032 (p0) fadd.s1 FR_r = FR_r, FR_X_cor
2033 nop.i 999
2034 }
2035 { .mfi
2036 nop.m 999
2037 (p0) fmpy.s1 FR_T = FR_T1, FR_T2
2038 nop.i 999 ;;
2039 }
2040 { .mfi
2041 nop.m 999
2042 //
2043 // if k < big_neg_exp, set p14 and Safe=False
2044 // Load A_2
2045 //
2046 (p0) fma.s1 FR_W = FR_W2, FR_W_1_p1, FR_W1
2047 nop.i 999 ;;
2048 }
2049 { .mfi
2050 nop.m 999
2051 (p0) fma.s1 FR_poly = FR_r, FR_A_3, FR_A_2
2052 nop.i 999
2053 }
2054 { .mfi
2055 nop.m 999
2056 (p0) fmpy.s1 FR_rsq = FR_r, FR_r
2057 nop.i 999 ;;
2058 }
2059 { .mfi
2060 nop.m 999
2061 (p0) mov FR_Y_hi = FR_T
2062 nop.i 999 ;;
2063 }
2064 { .mfi
2065 nop.m 999
2066 //
2067 // Scale = set_exp(BIAS_p_k)
2068 // poly = r * poly + A_1
2069 //
2070 (p0) fadd.s1 FR_Wp1 = FR_W, f1
2071 nop.i 999 ;;
2072 }
2073 { .mfi
2074 nop.m 999
2075 (p0) fma.s1 FR_poly = FR_r, FR_poly, FR_A_1
2076 nop.i 999 ;;
2077 }
2078 { .mfi
2079 nop.m 999
2080 (p0) fma.s1 FR_poly = FR_rsq, FR_poly,FR_r
2081 nop.i 999 ;;
2082 }
2083 { .mfi
2084 nop.m 999
2085 //
2086 // Wp1 = W + 1
2087 // poly = rsq * poly + rk
2088 //
2089 (p0) fma.s1 FR_Y_lo = FR_Wp1, FR_poly, FR_W
2090 nop.i 999 ;;
2091 }
2092 { .mfb
2093 nop.m 999
2094 //
2095 // Y_lo = poly * Wp1 + W
2096 // Y_hi = T
2097 //
2098 (p0) fmpy.s1 FR_Y_lo = FR_Y_lo, FR_T
2099 //
2100 // Y_lo = T * Y_lo
2101 //
2102 (p0) br.cond.sptk L(EXPL_RETURN) ;;
2103 }
2105 L(EXPL_SMALL):
2107 //
2108 // r4 = rsq * rsq
2109 //
2111 { .mmi
2112 nop.m 999
2113 (p0) addl GR_Table_Ptr1 = @ltoff(Constants_exp_64_P), gp
2114 nop.i 999
2115 }
2116 ;;
2118 { .mmi
2119 ld8 GR_Table_Ptr1 = [GR_Table_Ptr1]
2120 nop.m 999
2121 nop.i 999
2122 }
2123 ;;
2125 { .mmf
2126 nop.m 999
2127 (p0) ldfe FR_P_6 = [GR_Table_Ptr1],16
2128 //
2129 // Return
2130 //
2131 (p0) fadd.s1 FR_r = FR_X,f0 ;;
2132 }
2134 { .mmi
2135 nop.m 999
2136 (p0) addl GR_Table_Ptr = @ltoff(Constants_exp_64_Exponents#), gp
2137 nop.i 999
2138 }
2139 ;;
2141 { .mmi
2142 ld8 GR_Table_Ptr = [GR_Table_Ptr]
2143 (p0) ldfe FR_P_5 = [GR_Table_Ptr1],16
2144 nop.i 999
2145 }
2146 ;;
2148 //
2149 // Is input very small?
2150 // Load P_5
2151 //
2152 { .mii
2153 (p0) ldfe FR_P_4 = [GR_Table_Ptr1],16
2154 (p0) add GR_Table_Ptr = 0x040,GR_Table_Ptr ;;
2155 (p0) shladd GR_Table_Ptr = GR_Expo_Range,3,GR_Table_Ptr ;;
2156 }
2157 { .mmb
2158 (p0) ldfe FR_P_3 = [GR_Table_Ptr1],16
2159 //
2160 // Adjust ptr.
2161 //
2162 (p0) ld8 GR_vsm_expo = [GR_Table_Ptr],0
2163 nop.b 999 ;;
2164 }
2165 { .mfi
2166 nop.m 999
2167 //
2168 // r = X (don't seem to need X_Cor)
2169 // Load the threshold exponents
2170 //
2171 (p0) fmpy.s1 FR_rsq = FR_r, FR_r
2172 nop.i 999 ;;
2173 }
2174 //
2175 // Load the negative integer
2176 // Load P_5
2177 //
2178 { .mfi
2179 (p0) cmp.lt.unc p12, p0 = GR_Expo_X, GR_vsm_expo
2180 nop.f 999
2181 nop.i 999 ;;
2182 }
2183 { .mfb
2184 nop.m 999
2185 //
2186 // rsq = r * r
2187 // Offset into exponents
2188 //
2189 (p0) fmpy.s1 FR_r4 = FR_rsq, FR_rsq
2190 (p12) br.cond.spnt L(EXPL_VERY_SMALL) ;;
2191 }
2192 { .mfi
2193 (p0) ldfe FR_P_2 = [GR_Table_Ptr1],16
2194 //
2195 // Load p4,p3,p2,p1
2196 //
2197 (p0) fma.s1 FR_poly_lo = FR_P_6, FR_r, FR_P_5
2198 //
2199 // Y_lo = r4 * poly_lo + poly_hi
2200 // Scale = 1.0
2201 //
2202 (p0) add GR_temp = 0x1,r0 ;;
2203 }
2204 { .mmf
2205 nop.m 999
2206 (p0) ldfe FR_P_1 = [GR_Table_Ptr1],0
2207 (p0) mov FR_Scale = f1
2208 }
2209 //
2210 // Begin creating lsb to perturb final result
2211 //
2212 { .mfi
2213 (p0) setf.sig FR_temp = GR_temp
2214 (p0) mov FR_Y_hi = f1
2215 nop.i 999 ;;
2216 }
2217 { .mfi
2218 nop.m 999
2219 //
2220 // poly_lo = p_5 + p_6 * r
2221 // poly_hi = p_1 + p_2 * r
2222 //
2223 (p0) fma.s1 FR_poly_lo = FR_poly_lo, FR_r, FR_P_4
2224 nop.i 999 ;;
2225 }
2226 { .mfi
2227 nop.m 999
2228 //
2229 // poly_lo = p_4 + poly_lo * r
2230 // poly_hi = r + poly_hi * rsq
2231 //
2232 (p0) fma.s1 FR_poly_lo = FR_poly_lo, FR_r, FR_P_3
2233 nop.i 999
2234 }
2235 { .mfi
2236 nop.m 999
2237 (p0) fma.s1 FR_poly_hi = FR_P_2, FR_r, FR_P_1
2238 nop.i 999 ;;
2239 }
2240 { .mfi
2241 nop.m 999
2242 (p0) fma.s1 FR_poly_hi = FR_poly_hi, FR_rsq, FR_r
2243 nop.i 999 ;;
2244 }
2245 { .mfi
2246 nop.m 999
2247 //
2248 // poly_lo = p_3 + poly_lo * r
2249 // Y_hi = 1, always
2250 //
2251 (p0) fma.s1 FR_Y_lo = FR_poly_lo, FR_r4, FR_poly_hi
2252 nop.i 999 ;;
2253 }
2254 { .mfi
2255 nop.m 999
2256 //
2257 // Set lsb in fp register
2258 //
2259 (p0) for FR_temp = FR_Y_lo,FR_temp
2260 nop.i 999 ;;
2261 }
2262 { .mfb
2263 nop.m 999
2264 //
2265 // Toggle on last bit of Y_lo
2266 //
2267 (p0) fmerge.se FR_Y_lo = FR_Y_lo,FR_temp
2268 //
2269 // Set lsb of Y_lo to 1
2270 //
2271 (p0) br.cond.sptk L(EXPL_RETURN) ;;
2272 }
2273 L(EXPL_VERY_SMALL):
2274 { .mfi
2275 nop.m 999
2276 (p0) mov FR_Y_lo = FR_r
2277 (p0) cmp.eq.unc p15, p0 = r0, r0
2278 }
2279 { .mfi
2280 nop.m 999
2281 (p0) mov FR_Scale = f1
2282 nop.i 999
2283 };;
2284 { .mfb
2285 nop.m 999
2286 (p0) mov FR_Y_hi = f1
2287 //
2288 // If flag_not_1,
2289 // Y_hi = 1.0
2290 // Y_lo = X + X_cor
2291 // PR_Safe = true
2292 //
2293 (p0) br.cond.sptk L(EXPL_RETURN) ;;
2294 }
2295 L(EXPL_HUGE):
2296 { .mfi
2297 nop.m 999
2298 //
2299 // Return for flag=2
2300 //
2301 (p0) fcmp.gt.unc.s1 p12, p13 = FR_X, f0
2302 (p0) cmp.eq.unc p14, p15 = r0, r0 ;;
2303 }
2304 { .mlx
2305 nop.m 999
2306 //
2307 // Set Safe to false
2308 // Is x > 0
2309 //
2310 (p12) movl GR_Mask = 0x15DC0 ;;
2311 }
2312 { .mlx
2313 (p12) setf.exp FR_Y_hi = GR_Mask
2314 (p13) movl GR_Mask = 0xA240 ;;
2315 }
2316 { .mlx
2317 (p13) setf.exp FR_Y_hi = GR_Mask
2318 //
2319 // x > 0: Create mask for Y_hi = 2**(24,000)
2320 // x <= 0: Create mask for Y_hi = 2**(-24,000)
2321 //
2322 (p13) movl GR_temp = 0xA1DC ;;
2323 }
2324 { .mfi
2325 (p13) setf.exp FR_Y_lo = GR_temp
2326 //
2327 // x < =0: Create mask for 2**(-24,100)
2328 // x <= 0: Y_lo = w**(-24,100)
2329 //
2330 (p12) mov FR_Y_lo = f1
2331 nop.i 999 ;;
2332 }
2333 { .mfi
2334 nop.m 999
2335 (p12) mov FR_Scale = FR_Y_hi
2336 nop.i 999 ;;
2337 }
2338 { .mfi
2339 nop.m 999
2340 //
2341 // x > 0: Y_lo = 1.0
2342 // x > 0: Scale = 2**(24,000)
2343 //
2344 (p13) mov FR_Scale = FR_Y_hi
2345 nop.i 999 ;;
2346 }
2347 L(EXPL_RETURN):
2348 { .mfi
2349 nop.m 999
2350 //
2351 // Scale = 2**(24,000)
2352 //
2353 //
2354 // exp(y *ln(x)) almost complete
2355 // FR_Scale is Scale
2356 // f34 is Z_hi
2357 // f35 is Z_lo
2358 //
2359 (p0) fmpy.s1 FR_Sgn = FR_Scale, FR_Sgn
2360 nop.i 999 ;;
2361 }
2362 { .mfi
2363 nop.m 999
2364 //
2365 // sgn * scale
2366 //
2367 (p0) fmpy.s1 FR_Y_lo = FR_Y_lo,FR_Sgn
2368 nop.i 999 ;;
2369 }
2370 { .mfb
2371 nop.m 999
2372 //
2373 // Z_lo * (sgn * scale)
2374 //
2375 (p0) fma.s0 FR_Result = FR_Y_hi, FR_Sgn, FR_Y_lo
2376 //
2377 // Z_hi * (sgn * scale) + Z_lo
2378 //
2379 (p15) br.cond.sptk L(POWL_64_RETURN) ;;
2380 }
2381 { .mfi
2382 nop.m 999
2383 (p0) fsetc.s3 0x7F,0x01
2384 nop.i 999
2385 }
2386 { .mlx
2387 nop.m 999
2388 //
2389 // Z_hi * (sgn * scale) + Z_lo with wre & td
2390 // Z_hi * (sgn * scale) + Z_lo with fz & td
2391 //
2392 (p0) movl GR_T1_ptr = 0x00000000013FFF ;;
2393 }
2394 { .mfi
2395 nop.m 999
2396 (p0) fma.s3 FR_Result_small = FR_Y_hi, FR_Sgn, FR_Y_lo
2397 nop.i 999
2398 }
2399 { .mfi
2400 nop.m 999
2401 (p0) fsetc.s3 0x7F,0x40
2402 nop.i 999 ;;
2403 }
2404 { .mfi
2405 nop.m 999
2406 //
2407 // Return if no danger of over of underflow.
2408 //
2409 (p0) fsetc.s2 0x7F,0x42
2410 nop.i 999;;
2411 }
2412 { .mfi
2413 nop.m 999
2414 //
2415 // S0 user supplied status
2416 // S2 user supplied status + WRE + TD (Overflows)
2417 // S3 user supplied status + FZ + TD (Underflows)
2418 //
2419 (p0) fma.s2 FR_Result_big = FR_Y_hi, FR_Sgn, FR_Y_lo
2420 nop.i 999 ;;
2421 }
2422 //
2423 // S0 user supplied status
2424 // S2 user supplied status + WRE + TD (Overflows)
2425 // S3 user supplied status + FZ + TD (Underflows)
2426 //
2427 //
2428 // If (Safe) is true, then
2429 // Compute result using user supplied status field.
2430 // No overflow or underflow here, but perhaps inexact.
2431 // Return
2432 // Else
2433 // Determine if overflow or underflow was raised.
2434 // Fetch +/- overflow threshold for IEEE single, double,
2435 // double extended
2436 //
2437 { .mfi
2438 (p0) setf.exp FR_Big = GR_T1_ptr
2439 (p0) fsetc.s2 0x7F,0x40
2440 nop.i 999 ;;
2441 }
2442 { .mfi
2443 nop.m 999
2444 (p0) fclass.m.unc p11, p0 = FR_Result_small, 0x00F
2445 nop.i 999 ;;
2446 }
2447 { .mfi
2448 nop.m 999
2449 (p0) fmerge.ns FR_NBig = FR_Big, FR_Big
2450 nop.i 999
2451 }
2452 { .mfi
2453 nop.m 999
2454 //
2455 // Create largest double exponent + 1.
2456 // Create smallest double exponent - 1.
2457 // Identify denormals
2458 //
2459 (p0) fcmp.ge.unc.s1 p8, p0 = FR_Result_big , FR_Big
2460 nop.i 999 ;;
2461 }
2462 { .mii
2463 nop.m 999
2464 nop.i 999 ;;
2465 //
2466 // fcmp: resultS2 <= - overflow threshold
2467 // fclass: resultS3 is denorm/unorm/0
2468 //
2469 (p8) mov GR_Parameter_TAG = 18 ;;
2470 }
2471 { .mfb
2472 nop.m 999
2473 //
2474 // fcmp: resultS2 >= + overflow threshold
2475 //
2476 (p0) fcmp.le.unc.s1 p9, p0 = FR_Result_big, FR_NBig
2477 (p8) br.cond.spnt __libm_error_region ;;
2478 }
2479 { .mii
2480 nop.m 999
2481 nop.i 999 ;;
2482 (p9) mov GR_Parameter_TAG = 18
2483 }
2484 { .mib
2485 nop.m 999
2486 nop.i 999
2487 (p9) br.cond.spnt __libm_error_region ;;
2488 }
2489 //
2490 // Report that pow overflowed - either +Inf, or -Inf
2491 //
2492 { .mmb
2493 (p11) mov GR_Parameter_TAG = 19
2494 nop.m 999
2495 (p11) br.cond.spnt __libm_error_region ;;
2496 }
2497 { .mib
2498 nop.m 999
2499 nop.i 999
2500 //
2501 // Report that pow underflowed
2502 //
2503 (p0) br.cond.sptk L(POWL_64_RETURN) ;;
2504 }
2507 L(POWL_64_SQUARE):
2508 // Here if x not zero and y=2.
2509 // Must call __libm_error_support for overflow or underflow
2510 //
2511 // S0 user supplied status
2512 // S2 user supplied status + WRE + TD (Overflows)
2513 // S3 user supplied status + FZ + TD (Underflows)
2514 //
2515 { .mfi
2516 nop.m 999
2517 (p0) fma.s0 FR_Result = FR_Input_X, FR_Input_X, f0
2518 nop.i 999
2519 }
2520 { .mfi
2521 nop.m 999
2522 (p0) fsetc.s3 0x7F,0x01
2523 nop.i 999
2524 }
2525 { .mlx
2526 nop.m 999
2527 (p0) movl GR_T1_ptr = 0x00000000013FFF ;;
2528 }
2529 { .mfi
2530 nop.m 999
2531 (p0) fma.s3 FR_Result_small = FR_Input_X, FR_Input_X, f0
2532 nop.i 999
2533 }
2534 { .mfi
2535 nop.m 999
2536 (p0) fsetc.s3 0x7F,0x40
2537 nop.i 999 ;;
2538 }
2539 { .mfi
2540 nop.m 999
2541 //
2542 // Return if no danger of over of underflow.
2543 //
2544 (p0) fsetc.s2 0x7F,0x42
2545 nop.i 999;;
2546 }
2547 { .mfi
2548 nop.m 999
2549 (p0) fma.s2 FR_Result_big = FR_Input_X, FR_Input_X, f0
2550 nop.i 999 ;;
2551 }
2552 //
2553 // S0 user supplied status
2554 // S2 user supplied status + WRE + TD (Overflows)
2555 // S3 user supplied status + FZ + TD (Underflows)
2556 //
2557 //
2558 // If (Safe) is true, then
2559 // Compute result using user supplied status field.
2560 // No overflow or underflow here, but perhaps inexact.
2561 // Return
2562 // Else
2563 // Determine if overflow or underflow was raised.
2564 // Fetch +/- overflow threshold for IEEE single, double,
2565 // double extended
2566 //
2567 { .mfi
2568 (p0) setf.exp FR_Big = GR_T1_ptr
2569 (p0) fsetc.s2 0x7F,0x40
2570 nop.i 999 ;;
2571 }
2572 { .mfi
2573 nop.m 999
2574 (p0) fclass.m.unc p11, p0 = FR_Result_small, 0x00F
2575 nop.i 999 ;;
2576 }
2577 { .mfi
2578 nop.m 999
2579 (p0) fmerge.ns FR_NBig = FR_Big, FR_Big
2580 nop.i 999
2581 }
2582 { .mfi
2583 nop.m 999
2584 //
2585 // Create largest double exponent + 1.
2586 // Create smallest double exponent - 1.
2587 // Identify denormals
2588 //
2589 (p0) fcmp.ge.unc.s1 p8, p0 = FR_Result_big , FR_Big
2590 nop.i 999 ;;
2591 }
2592 { .mii
2593 nop.m 999
2594 nop.i 999 ;;
2595 //
2596 // fcmp: resultS2 <= - overflow threshold
2597 // fclass: resultS3 is denorm/unorm/0
2598 //
2599 (p8) mov GR_Parameter_TAG = 18 ;;
2600 }
2601 { .mfb
2602 nop.m 999
2603 //
2604 // fcmp: resultS2 >= + overflow threshold
2605 //
2606 (p0) fcmp.le.unc.s1 p9, p0 = FR_Result_big, FR_NBig
2607 (p8) br.cond.spnt __libm_error_region ;;
2608 }
2609 { .mii
2610 nop.m 999
2611 nop.i 999 ;;
2612 (p9) mov GR_Parameter_TAG = 18
2613 }
2614 { .mib
2615 nop.m 999
2616 nop.i 999
2617 (p9) br.cond.spnt __libm_error_region ;;
2618 }
2619 //
2620 // Report that pow overflowed - either +Inf, or -Inf
2621 //
2622 { .mmb
2623 (p11) mov GR_Parameter_TAG = 19
2624 nop.m 999
2625 (p11) br.cond.spnt __libm_error_region ;;
2626 }
2627 { .mib
2628 nop.m 999
2629 nop.i 999
2630 //
2631 // Report that pow underflowed
2632 //
2633 (p0) br.cond.sptk L(POWL_64_RETURN) ;;
2634 }
2639 L(POWL_64_SPECIAL):
2640 { .mfi
2641 nop.m 999
2642 (p0) fcmp.eq.s1 p15, p0 = FR_Input_X, f1 // Is x=+1
2643 nop.i 999 ;;
2644 }
2645 { .mfi
2646 nop.m 999
2647 (p0) fclass.m.unc p14, p0 = FR_Input_Y, 0x023
2648 nop.i 999 ;;
2649 }
2651 { .mfi
2652 nop.m 999
2653 (p15) fcmp.eq.unc.s0 p6,p0 = FR_Input_Y, f0 // If x=1, flag invalid if y=SNaN
2654 nop.i 999
2655 }
2656 { .mfb
2657 nop.m 999
2658 (p15) fmpy.s0 FR_Result = f1,f1 // If x=1, result=1
2659 (p15) br.cond.spnt L(POWL_64_RETURN) ;; // Exit if x=1
2660 }
2662 { .mfi
2663 nop.m 999
2664 (p0) fclass.m.unc p13, p0 = FR_Input_X, 0x023
2665 nop.i 999 ;;
2666 }
2667 { .mfi
2668 nop.m 999
2669 (p0) fclass.m.unc p8, p0 = FR_Input_X, 0x143
2670 nop.i 999
2671 }
2672 { .mfi
2673 nop.m 999
2674 (p0) fclass.m.unc p9, p0 = FR_Input_Y, 0x143
2675 nop.i 999 ;;
2676 }
2677 { .mfi
2678 nop.m 999
2679 (p0) fclass.m.unc p10, p0 = FR_Input_X, 0x083
2680 nop.i 999
2681 }
2682 { .mfi
2683 nop.m 999
2684 (p0) fclass.m.unc p11, p0 = FR_Input_Y, 0x083
2685 nop.i 999 ;;
2686 }
2687 { .mfi
2688 nop.m 999
2689 (p0) fclass.m.unc p6, p0 = FR_Input_Y, 0x007
2690 nop.i 999
2691 }
2692 { .mfi
2693 nop.m 999
2694 (p0) fcmp.eq.unc.s1 p7, p0 = FR_Input_Y, f1
2695 nop.i 999 ;;
2696 }
2697 { .mfi
2698 nop.m 999
2699 //
2700 // set p13 if x +/- Inf
2701 // set p14 if y +/- Inf
2702 // set p8 if x Natval or +/-SNaN
2703 // set p9 if y Natval or +/-SNaN
2704 // set p10 if x QNaN
2705 // set p11 if y QNaNs
2706 // set p6 if y is +/-0
2707 // set p7 if y is 1
2708 //
2709 (p8) fmpy.s0 FR_Result = FR_Input_Y, FR_Input_X
2710 (p6) cmp.ne p8,p0 = r0,r0 ;; // Don't exit if x=snan, y=0 ==> result=+1
2711 }
2712 { .mfb
2713 nop.m 999
2714 (p9) fmpy.s0 FR_Result = FR_Input_Y, FR_Input_X
2715 (p8) br.cond.spnt L(POWL_64_RETURN) ;;
2716 }
2717 { .mfb
2718 nop.m 999
2719 (p10) fmpy.s0 FR_Result = FR_Input_X, f0
2720 (p9) br.cond.spnt L(POWL_64_RETURN) ;;
2721 }
2722 { .mfi
2723 nop.m 999
2724 //
2725 // Produce result for SNaN and NatVals and return
2726 //
2727 (p6) fclass.m.unc p15, p0 = FR_Input_X,0x007
2728 nop.i 999
2729 }
2730 { .mfi
2731 nop.m 999
2732 //
2733 // If Y +/- 0, set p15 if x +/- 0
2734 //
2735 (p6) fclass.m.unc p8, p0 = FR_Input_X,0x0C3
2736 nop.i 999 ;;
2737 }
2739 { .mfi
2740 nop.m 999
2741 (p6) fcmp.eq.s0 p9,p0 = FR_Input_X, f0 // If y=0, flag if x denormal
2742 nop.i 999
2743 }
2744 { .mfi
2745 nop.m 999
2746 (p6) fadd.s0 FR_Result = f1, f0
2747 nop.i 999 ;;
2748 }
2749 { .mfi
2750 nop.m 999
2751 //
2752 // Set p8 if y = +/-0 and X is a QNaN/SNaN
2753 // If y = +/-0, let result = 1.0
2754 //
2755 (p7) fmpy.s0 FR_Result = FR_Input_X,f1
2756 //
2757 // If y == 1, result = x * 1
2758 //
2759 (p15) mov GR_Parameter_TAG = 20
2760 }
2761 { .mib
2762 nop.m 999
2763 nop.i 999
2764 (p15) br.cond.spnt __libm_error_region ;;
2765 }
2766 { .mib
2767 nop.m 999
2768 //
2769 // If x and y are both zero, result = 1.0 and call error
2770 // support.
2771 //
2772 (p8) mov GR_Parameter_TAG = 23
2773 (p8) br.cond.spnt __libm_error_region ;;
2774 }
2775 { .mib
2776 nop.m 999
2777 nop.i 999
2778 //
2779 // If y = +/-0 and x is a QNaN, result = 1.0 and call error
2780 // support.
2781 //
2782 (p6) br.cond.spnt L(POWL_64_RETURN) ;;
2783 }
2785 // If x=0, y=-inf, go to the X_IS_ZERO path
2786 { .mfb
2787 nop.m 999
2788 (p14) fcmp.eq.unc.s1 p0,p14 = FR_Input_X,f0
2789 (p7) br.cond.spnt L(POWL_64_RETURN) ;;
2790 }
2792 { .mfi
2793 nop.m 999
2794 //
2795 // Produce all results for x**0 and x**1
2796 // Let all the result x ** 0 == 1 and return
2797 // Let all x ** 1 == x and return
2798 //
2799 (p10) fmpy.s0 FR_Result = FR_Input_Y,FR_Input_X
2800 nop.i 999 ;;
2801 }
2802 { .mfb
2803 nop.m 999
2804 (p11) fmpy.s0 FR_Result = FR_Input_Y,FR_Input_X
2805 (p10) br.cond.spnt L(POWL_64_RETURN) ;;
2806 }
2807 { .mib
2808 nop.m 999
2809 nop.i 999
2810 (p11) br.cond.spnt L(POWL_64_RETURN) ;;
2811 }
2812 { .mib
2813 nop.m 999
2814 nop.i 999
2815 //
2816 // Return result for x or y QNaN input with QNaN result
2817 //
2818 (p14) br.cond.spnt L(POWL_64_Y_IS_INF) ;;
2819 }
2820 { .mib
2821 nop.m 999
2822 nop.i 999
2823 (p13) br.cond.spnt L(POWL_64_X_IS_INF) ;;
2824 }
2825 L(POWL_64_X_IS_ZERO):
2826 { .mmb
2827 (p0) getf.sig GR_signif_y = FR_Input_Y
2828 (p0) getf.exp GR_BIASed_exp_y = FR_Input_Y
2829 nop.b 999 ;;
2830 }
2831 { .mlx
2832 nop.m 999
2833 (p0) movl GR_Mask = 0x1FFFF
2834 }
2835 { .mlx
2836 nop.m 999
2837 (p0) movl GR_y_sign = 0x20000 ;;
2838 }
2839 //
2840 // Get BIASed exp and significand of y
2841 //
2842 { .mfi
2843 (p0) and GR_exp_y = GR_Mask,GR_BIASed_exp_y
2844 nop.f 999
2845 (p0) and GR_y_sign = GR_y_sign,GR_BIASed_exp_y
2846 }
2847 { .mlx
2848 nop.m 999
2849 (p0) movl GR_BIAS = 0xFFFF ;;
2850 }
2851 { .mfi
2852 (p0) cmp.lt.unc p9, p8 = GR_exp_y,GR_BIAS
2853 nop.f 999
2854 //
2855 // Maybe y is < 1 already, so
2856 // can never be an integer.
2857 // Remove sign bit from exponent.
2858 //
2859 (p0) sub GR_exp_y = GR_exp_y,GR_BIAS ;;
2860 }
2861 { .mii
2862 nop.m 999
2863 nop.i 999 ;;
2864 //
2865 // Remove exponent BIAS
2866 //
2867 (p8) shl GR_exp_y= GR_signif_y,GR_exp_y ;;
2868 }
2869 { .mfi
2870 (p9) or GR_exp_y= 0xF,GR_signif_y
2871 nop.f 999
2872 nop.i 999 ;;
2873 }
2874 { .mii
2875 nop.m 999
2876 //
2877 // Shift significand of y looking for nonzero bits
2878 // For y > 1, shift signif_y exp_y bits to the left
2879 // For y < 1, turn on 4 low order bits of significand of y
2880 // so that the fraction will always be non-zero
2881 //
2882 (p0) shl GR_signif_y= GR_exp_y,1 ;;
2883 (p0) extr.u GR_low_order_bit = GR_exp_y,63,1
2884 }
2885 //
2886 // Integer part of y shifted off.
2887 // Get y's low even or odd bit - y might not be an int.
2888 //
2889 { .mii
2890 (p0) cmp.eq.unc p13,p0 = GR_signif_y, r0
2891 (p0) cmp.eq.unc p8,p9 = GR_y_sign, r0 ;;
2892 //
2893 // Is y an int?
2894 // Is y positive
2895 //
2896 (p13) cmp.ne.unc p13,p0 = GR_low_order_bit, r0 ;;
2897 }
2898 //
2899 // Is y and int and odd?
2900 //
2901 { .mfb
2902 (p13) cmp.eq.unc p13,p14 = GR_y_sign, r0
2903 (p8) fcmp.eq.s0 p12,p0 = FR_Input_Y, f0 // If x=0 and y>0 flag if y denormal
2904 nop.b 999 ;;
2905 }
2906 { .mfb
2907 nop.m 999
2908 //
2909 // Is y and int and odd and positive?
2910 //
2911 (p13) mov FR_Result = FR_Input_X
2912 (p13) br.cond.sptk L(POWL_64_RETURN) ;;
2913 }
2914 { .mfi
2915 nop.m 999
2916 //
2917 // Return +/-0 when x=+/-0 and y is and odd pos. int
2918 //
2919 (p14) frcpa.s0 FR_Result, p10 = f1, FR_Input_X
2920 (p14) mov GR_Parameter_TAG = 21
2921 }
2922 { .mib
2923 nop.m 999
2924 nop.i 999
2925 (p14) br.cond.spnt __libm_error_region ;;
2926 }
2928 { .mfb
2929 nop.m 999
2930 //
2931 // Return +/-0 when x=+/-Inf and y is and odd neg int
2932 // and raise dz exception
2933 //
2934 (p8) mov FR_Result = f0
2935 (p8) br.cond.sptk L(POWL_64_RETURN) ;;
2936 }
2937 { .mfi
2938 nop.m 999
2939 //
2940 // Return +0 when x=+/-0 and y > 0 and not odd.
2941 //
2942 (p9) frcpa.s0 FR_Result, p10 = f1,f0
2943 (p9) mov GR_Parameter_TAG = 21
2944 }
2945 { .mib
2946 nop.m 999
2947 nop.i 999
2948 (p9) br.cond.sptk __libm_error_region ;;
2949 }
2950 L(POWL_64_X_IS_INF):
2951 { .mfi
2952 (p0) getf.exp GR_exp_y = FR_Input_Y
2953 (p0) fclass.m.unc p13, p0 = FR_Input_X,0x022
2954 (p0) mov GR_Mask = 0x1FFFF ;;
2955 }
2957 { .mfi
2958 (p0) getf.sig GR_signif_y = FR_Input_Y
2959 (p0) fcmp.eq.s0 p9,p0 = FR_Input_Y, f0 // Flag if y denormal
2960 nop.i 999 ;;
2961 }
2963 //
2964 // Get exp and significand of y
2965 // Create exponent mask and sign mask
2966 //
2967 { .mlx
2968 (p0) and GR_low_order_bit = GR_Mask,GR_exp_y
2969 (p0) movl GR_BIAS = 0xFFFF
2970 }
2971 { .mmi
2972 nop.m 999 ;;
2973 //
2974 // Remove sign bit from exponent.
2975 //
2976 (p0) cmp.lt.unc p9, p8 = GR_low_order_bit,GR_BIAS
2977 //
2978 // Maybe y is < 1 already, so
2979 // isn't an int.
2980 //
2981 (p0) sub GR_low_order_bit = GR_low_order_bit,GR_BIAS
2982 }
2983 { .mlx
2984 nop.m 999
2985 (p0) movl GR_sign_mask = 0x20000 ;;
2986 }
2987 { .mfi
2988 (p0) and GR_sign_mask = GR_sign_mask,GR_exp_y
2989 //
2990 // Return +Inf when x=+/-0 and y < 0 and not odd and raise
2991 // divide-by-zero exception.
2992 //
2993 (p0) fclass.m.unc p11, p0 = FR_Input_X,0x021
2994 nop.i 999 ;;
2995 }
2996 { .mmi
2997 nop.m 999 ;;
2998 //
2999 // Is shift off integer part of y.
3000 // Get y's even or odd bit - y might not be an int.
3001 //
3002 (p11) cmp.eq.unc p11,p12 = GR_sign_mask, r0
3003 //
3004 // Remove exponent BIAS
3005 //
3006 (p8) shl GR_exp_y = GR_signif_y,GR_low_order_bit ;;
3007 }
3008 { .mfi
3009 (p9) or GR_exp_y = 0xF,GR_signif_y
3010 //
3011 // Is y positive or negative when x is +Inf?
3012 // Is y and int when x = -Inf
3013 //
3014 (p11) mov FR_Result = FR_Input_X
3015 nop.i 999 ;;
3016 }
3017 { .mfi
3018 nop.m 999
3019 (p12) mov FR_Result = f0
3020 nop.i 999 ;;
3021 }
3022 { .mii
3023 nop.m 999
3024 //
3025 // Shift signficand looking for nonzero bits
3026 // For y non-ints, upset the significand.
3027 //
3028 (p0) shl GR_signif_y = GR_exp_y,1 ;;
3029 (p13) cmp.eq.unc p13,p0 = GR_signif_y, r0
3030 }
3031 { .mii
3032 nop.m 999
3033 (p0) extr.u GR_low_order_bit = GR_exp_y,63,1 ;;
3034 (p13) cmp.ne.unc p13,p0 = GR_low_order_bit, r0
3035 }
3036 { .mib
3037 nop.m 999
3038 nop.i 999
3039 (p11) br.cond.sptk L(POWL_64_RETURN) ;;
3040 }
3041 { .mib
3042 nop.m 999
3043 nop.i 999
3044 (p12) br.cond.sptk L(POWL_64_RETURN) ;;
3045 }
3046 //
3047 // Return Inf for y > 0
3048 // Return +0 for y < 0
3049 // Is y even or odd?
3050 //
3051 { .mii
3052 (p13) cmp.eq.unc p13,p10 = GR_sign_mask, r0
3053 (p0) cmp.eq.unc p8,p9 = GR_sign_mask, r0 ;;
3054 nop.i 999
3055 }
3056 { .mfi
3057 nop.m 999
3058 //
3059 // For x = -inf, y is and int, positive
3060 // and odd
3061 // Is y positive in general?
3062 //
3063 (p13) mov FR_Result = FR_Input_X
3064 nop.i 999 ;;
3065 }
3066 { .mfb
3067 nop.m 999
3068 (p10) fmerge.ns FR_Result = f0, f0
3069 (p13) br.cond.sptk L(POWL_64_RETURN) ;;
3070 }
3071 { .mib
3072 nop.m 999
3073 nop.i 999
3074 (p10) br.cond.sptk L(POWL_64_RETURN) ;;
3075 }
3076 { .mfi
3077 nop.m 999
3078 //
3079 // Return -Inf for x = -inf and y > 0 and odd int.
3080 // Return -0 for x = -inf and y < 0 and odd int.
3081 //
3082 (p8) fmerge.ns FR_Result = FR_Input_X, FR_Input_X
3083 nop.i 999 ;;
3084 }
3085 { .mfb
3086 nop.m 999
3087 (p9) mov FR_Result = f0
3088 (p8) br.cond.sptk L(POWL_64_RETURN) ;;
3089 }
3090 { .mib
3091 nop.m 999
3092 nop.i 999
3093 (p9) br.cond.sptk L(POWL_64_RETURN) ;;
3094 }
3095 L(POWL_64_Y_IS_INF):
3096 { .mfi
3097 nop.m 999
3098 //
3099 // Return Inf for x = -inf and y > 0 not an odd int.
3100 // Return +0 for x = -inf and y < 0 and not an odd int.
3101 //
3102 (p0) fclass.m.unc p8, p0 = FR_Input_Y, 0x021
3103 nop.i 999
3104 }
3105 { .mfi
3106 nop.m 999
3107 (p0) fclass.m.unc p9, p0 = FR_Input_Y, 0x022
3108 nop.i 999 ;;
3109 }
3110 { .mfi
3111 nop.m 999
3112 (p0) fabs FR_X = FR_Input_X
3113 nop.i 999 ;;
3114 }
3116 { .mfi
3117 nop.m 999
3118 (p0) fcmp.eq.s0 p10,p0 = FR_Input_X, f0 // flag if x denormal
3119 nop.i 999 ;;
3120 }
3122 { .mfi
3123 nop.m 999
3124 //
3125 // Find y = +/- Inf
3126 // Compute |x|
3127 //
3128 (p8) fcmp.lt.unc.s1 p6, p0 = FR_X, f1
3129 nop.i 999
3130 }
3131 { .mfi
3132 nop.m 999
3133 (p8) fcmp.gt.unc.s1 p7, p0 = FR_X, f1
3134 nop.i 999 ;;
3135 }
3136 { .mfi
3137 nop.m 999
3138 (p9) fcmp.lt.unc.s1 p12, p0 = FR_X, f1
3139 nop.i 999
3140 }
3141 { .mfi
3142 nop.m 999
3143 (p9) fcmp.gt.unc.s1 p13, p0 = FR_X, f1
3144 nop.i 999 ;;
3145 }
3146 { .mfi
3147 nop.m 999
3148 //
3149 // For y = +Inf and |x| < 1 returns 0
3150 // For y = +Inf and |x| > 1 returns Inf
3151 // For y = -Inf and |x| < 1 returns Inf
3152 // For y = -Inf and |x| > 1 returns 0
3153 //
3154 (p6) mov FR_Result = f0
3155 nop.i 999 ;;
3156 }
3157 { .mfi
3158 nop.m 999
3159 (p7) mov FR_Result = FR_Input_Y
3160 nop.i 999 ;;
3161 }
3162 { .mfi
3163 nop.m 999
3164 (p12) fmpy.s0 FR_Result = FR_Input_Y, FR_Input_Y
3165 nop.i 999 ;;
3166 }
3167 { .mfb
3168 nop.m 999
3169 (p13) mov FR_Result = f0
3170 //
3171 // Produce x ** +/- Inf results
3172 //
3173 (p6) br.cond.spnt L(POWL_64_RETURN) ;;
3174 }
3175 { .mib
3176 nop.m 999
3177 nop.i 999
3178 (p7) br.cond.spnt L(POWL_64_RETURN) ;;
3179 }
3180 { .mib
3181 nop.m 999
3182 nop.i 999
3183 (p12) br.cond.spnt L(POWL_64_RETURN) ;;
3184 }
3185 { .mib
3186 nop.m 999
3187 nop.i 999
3188 (p13) br.cond.spnt L(POWL_64_RETURN) ;;
3189 }
3190 { .mfb
3191 nop.m 999
3192 //
3193 // +/-1 ** +/-Inf, result is +1
3194 //
3195 (p0) fmpy.s0 FR_Result = f1,f1
3196 (p0) br.cond.sptk L(POWL_64_RETURN) ;;
3197 }
3198 L(POWL_64_UNSUPPORT):
3199 { .mfb
3200 nop.m 999
3201 //
3202 // Return NaN and raise invalid
3203 //
3204 (p0) fmpy.s0 FR_Result = FR_Input_X,f0
3205 //
3206 // Raise exceptions for specific
3207 // values - pseudo NaN and
3208 // infinities.
3209 //
3210 (p0) br.cond.sptk L(POWL_64_RETURN) ;;
3211 }
3212 L(POWL_64_XNEG):
3213 { .mfi
3214 nop.m 999
3215 (p0) frcpa.s0 FR_Result, p8 = f0, f0
3216 //
3217 // Raise invalid for x < 0 and
3218 // y not an integer and
3219 //
3220 (p0) mov GR_Parameter_TAG = 22
3221 }
3222 { .mib
3223 nop.m 999
3224 nop.i 999
3225 (p0) br.cond.sptk __libm_error_region ;;
3226 }
3227 L(POWL_64_SQRT):
3228 { .mfi
3229 nop.m 999
3230 (p0) frsqrta.s0 FR_Result,p10 = FR_Input_X
3231 nop.i 999 ;;
3232 }
3233 { .mfi
3234 nop.m 999
3235 (p10) fma.s1 f62=FR_Half,FR_Input_X,f0
3236 nop.i 999 ;;
3237 }
3238 { .mfi
3239 nop.m 999
3240 //
3241 // Step (2)
3242 // h = 1/2 * a in f9
3243 //
3244 (p10) fma.s1 f63=FR_Result,FR_Result,f0
3245 nop.i 999 ;;
3246 }
3247 { .mfi
3248 nop.m 999
3249 //
3250 // Step (3)
3251 // t1 = y0 * y0 in f10
3252 //
3253 (p10) fnma.s1 f32=f63,f62,f11
3254 nop.i 999 ;;
3255 }
3256 { .mfi
3257 nop.m 999
3258 //
3259 // Step (4)
3260 // t2 = 1/2 - t1 * h in f10
3261 //
3262 (p10) fma.s1 f33=f32,FR_Result,FR_Result
3263 nop.i 999 ;;
3264 }
3265 { .mfi
3266 nop.m 999
3267 //
3268 // Step (5)
3269 // y1 = y0 + t2 * y0 in f13
3270 //
3271 (p10) fma.s1 f34=f33,f62,f0
3272 nop.i 999 ;;
3273 }
3274 { .mfi
3275 nop.m 999
3276 //
3277 // Step (6)
3278 // t3 = y1 * h in f10
3279 //
3280 (p10) fnma.s1 f35=f34,f33,f11
3281 nop.i 999 ;;
3282 }
3283 { .mfi
3284 nop.m 999
3285 //
3286 // Step (7)
3287 // t4 = 1/2 - t3 * y1 in f10
3288 //
3289 (p10) fma.s1 f63=f35,f33,f33
3290 nop.i 999 ;;
3291 }
3292 { .mfi
3293 nop.m 999
3294 //
3295 // Step (8)
3296 // y2 = y1 + t4 * y1 in f13
3297 //
3298 (p10) fma.s1 f32=FR_Input_X,f63,f0
3299 nop.i 999
3300 }
3301 { .mfi
3302 nop.m 999
3303 //
3304 // Step (9)
3305 // S = a * y2 in f10
3306 //
3307 (p10) fma.s1 FR_Result=f63,f62,f0
3308 nop.i 999 ;;
3309 }
3310 { .mfi
3311 nop.m 999
3312 //
3313 // Step (10)
3314 // t5 = y2 * h in f9
3315 //
3316 (p10) fma.s1 f33=f11,f63,f0
3317 nop.i 999 ;;
3318 }
3319 { .mfi
3320 nop.m 999
3321 //
3322 // Step (11)
3323 // H = 1/2 * y2 in f11
3324 //
3325 (p10) fnma.s1 f34=f32,f32,f8
3326 nop.i 999
3327 }
3328 { .mfi
3329 nop.m 999
3330 //
3331 // Step (12)
3332 // d = a - S * S in f12
3333 //
3334 (p10) fnma.s1 f35=FR_Result,f63,f11
3335 nop.i 999 ;;
3336 }
3337 { .mfi
3338 nop.m 999
3339 //
3340 // Step (13)
3341 // t6 = 1/2 - t5 * y2 in f7
3342 //
3343 (p10) fma.s1 f62=f33,f34,f32
3344 nop.i 999
3345 }
3346 { .mfi
3347 nop.m 999
3348 //
3349 // Step (14)
3350 // S1 = S + d * H in f13
3351 //
3352 (p10) fma.s1 f63=f33,f35,f33
3353 nop.i 999 ;;
3354 }
3355 { .mfi
3356 nop.m 999
3357 //
3358 // Step (15)
3359 // H1 = H + t6 * h in f7
3360 //
3361 (p10) fnma.s1 f32=f62,f62,FR_Input_X
3362 nop.i 999 ;;
3363 }
3364 { .mfb
3365 nop.m 999
3366 //
3367 // Step (16)
3368 // d1 = a - S1 * S1
3369 //
3370 (p10) fma.s0 FR_Result=f32,f63,f62
3371 //
3372 // Step (17)
3373 // R = S1 + d1 * H1
3374 //
3375 (p10) br.cond.sptk L(POWL_64_RETURN) ;;
3376 }
3377 { .mib
3378 nop.m 999
3379 nop.i 999
3380 //
3381 // Do the Newton-Raphson iteration from the EAS.
3382 //
3383 (p0) br.cond.sptk L(POWL_64_RETURN) ;;
3384 }
3385 //
3386 // Take care of the degenerate cases.
3387 //
3389 L(POWL_64_RETURN):
3390 { .mfb
3391 nop.m 999
3392 (p0) mov FR_Output = FR_Result
3393 (p0) br.ret.sptk b0 ;;
3394 }
3395 .endp powl
3396 ASM_SIZE_DIRECTIVE(powl)
3398 .proc __libm_error_region
3399 __libm_error_region:
3400 .prologue
3401 { .mfi
3402 add GR_Parameter_Y=-32,sp // Parameter 2 value
3403 nop.f 0
3404 .save ar.pfs,GR_SAVE_PFS
3405 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
3406 }
3407 { .mfi
3408 .fframe 64
3409 add sp=-64,sp // Create new stack
3410 nop.f 0
3411 mov GR_SAVE_GP=gp // Save gp
3412 };;
3413 { .mmi
3414 stfe [GR_Parameter_Y] = FR_Y,16 // Save Parameter 2 on stack
3415 add GR_Parameter_X = 16,sp // Parameter 1 address
3416 .save b0, GR_SAVE_B0
3417 mov GR_SAVE_B0=b0 // Save b0
3418 };;
3419 .body
3420 { .mib
3421 stfe [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
3422 add GR_Parameter_RESULT = 0,GR_Parameter_Y
3423 nop.b 0 // Parameter 3 address
3424 }
3425 { .mib
3426 stfe [GR_Parameter_Y] = FR_RESULT // Store Parameter 3 on stack
3427 add GR_Parameter_Y = -16,GR_Parameter_Y
3428 br.call.sptk b0=__libm_error_support# // Call error handling function
3429 };;
3430 { .mmi
3431 nop.m 0
3432 nop.m 0
3433 add GR_Parameter_RESULT = 48,sp
3434 };;
3435 { .mmi
3436 ldfe f8 = [GR_Parameter_RESULT] // Get return result off stack
3437 .restore sp
3438 add sp = 64,sp // Restore stack pointer
3439 mov b0 = GR_SAVE_B0 // Restore return address
3440 };;
3441 { .mib
3442 mov gp = GR_SAVE_GP // Restore gp
3443 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
3444 br.ret.sptk b0 // Return
3445 };;
3447 .endp __libm_error_region
3448 ASM_SIZE_DIRECTIVE(__libm_error_region)
3449 .type __libm_error_support#,@function
3450 .global __libm_error_support#