localedata: dz_BT, bo_CN: convert to UTF-8
[glibc.git] / sysdeps / ia64 / fpu / e_exp2.S
blob76e26f298c9a191120398090597cc8927424c26b
1 .file "exp2.s"
4 // Copyright (c) 2000 - 2005, Intel Corporation
5 // All rights reserved.
6 //
7 //
8 // Redistribution and use in source and binary forms, with or without
9 // modification, are permitted provided that the following conditions are
10 // met:
12 // * Redistributions of source code must retain the above copyright
13 // notice, this list of conditions and the following disclaimer.
15 // * Redistributions in binary form must reproduce the above copyright
16 // notice, this list of conditions and the following disclaimer in the
17 // documentation and/or other materials provided with the distribution.
19 // * The name of Intel Corporation may not be used to endorse or promote
20 // products derived from this software without specific prior written
21 // permission.
23 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
27 // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
28 // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29 // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
31 // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
32 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 // Intel Corporation is the author of this code, and requests that all
36 // problem reports or change requests be submitted to it directly at
37 // http://www.intel.com/software/products/opensource/libraries/num.htm.
39 // History
40 //==============================================================
41 // 08/25/00  Initial version
42 // 05/20/02  Cleaned up namespace and sf0 syntax
43 // 09/05/02  Improved performance
44 // 01/17/03  Fixed to call error support when x=1024.0
45 // 03/31/05  Reformatted delimiters between data tables
47 // API
48 //==============================================================
49 // double exp2(double)
51 // Overview of operation
52 //==============================================================
53 // Background
55 // Implementation
57 // Let x= (K + fh + fl + r), where
58 // K is an integer, fh= 0.b1 b2 b3 b4 b5,
59 // fl= 2^{-5}* 0.b6 b7 b8 b8 b10 (fh, fl >= 0),
60 // and |r|<2^{-11}
61 // Th is a table that stores 2^fh (32 entries) rounded to
62 // double extended precision (only mantissa is stored)
63 // Tl is a table that stores 2^fl (32 entries) rounded to
64 // double extended precision (only mantissa is stored)
66 // 2^x is approximated as
67 // 2^K * Th [ f ] * Tl [ f ] * (1+c1*r+c2*r^2+c3*r^3+c4*r^4)
69 // Note: We use the following trick to speed up conversion from FP to integer:
71 // Let  x = K + r, where K is an integer, and  |r| <= 0.5
72 // Let N be the number of significand bits for the FP format used
73 //   ( N=64 for double-extended, N=53 for double)
75 // Then let y = 1.5 * 2^(N-1)  +  x    for RN mode
76 //          K = y -  1.5 * 2^(N-1)
77 //          r  = x - K
79 // If we want to obtain the integer part and the first m fractional bits of x,
80 // we can use the same trick, but with a constant of  1.5 * 2^(N-1-m):
82 // Let x = K + f + r
83 // f = 0.b_1 b_2 ... b_m
84 // |r| <= 2^(-m-1)
86 // Then let y = 1.5 * 2^(N-1-m)  +  x    for RN mode
87 //          (K+f) = y -  1.5 * 2^(N-1-m)
88 //          r  = x - K
91 // Special values
92 //==============================================================
93 // exp2(0)= 1
94 // exp2(+inf)= inf
95 // exp2(-inf)= 0
98 // Registers used
99 //==============================================================
100 // r2-r3, r14-r40
101 // f6-f15, f32-f45
102 // p6-p8, p12
106 GR_TBL_START        = r2
107 GR_LOG_TBL          = r3
109 GR_OF_LIMIT         = r14
110 GR_UF_LIMIT         = r15
111 GR_EXP_CORR         = r16
112 GR_F_low            = r17
113 GR_F_high           = r18
114 GR_K                = r19
115 GR_Flow_ADDR        = r20
117 GR_BIAS             = r21
118 GR_Fh               = r22
119 GR_Fh_ADDR          = r23
120 GR_EXPMAX           = r24
121 GR_EMIN             = r25
123 GR_ROUNDVAL         = r26
124 GR_MASK             = r27
125 GR_KF0              = r28
126 GR_MASK_low         = r29
127 GR_COEFF_START      = r30
129 GR_SAVE_B0          = r33
130 GR_SAVE_PFS         = r34
131 GR_SAVE_GP          = r35
132 GR_SAVE_SP          = r36
134 GR_Parameter_X      = r37
135 GR_Parameter_Y      = r38
136 GR_Parameter_RESULT = r39
137 GR_Parameter_TAG    = r40
140 FR_X                = f10
141 FR_Y                = f1
142 FR_RESULT           = f8
145 FR_COEFF1           = f6
146 FR_COEFF2           = f7
147 FR_R                = f9
149 FR_KF0              = f12
150 FR_COEFF3           = f13
151 FR_COEFF4           = f14
152 FR_UF_LIMIT         = f15
154 FR_OF_LIMIT         = f32
155 FR_EXPMIN           = f33
156 FR_ROUNDVAL         = f34
157 FR_KF               = f35
159 FR_2_TO_K           = f36
160 FR_T_low            = f37
161 FR_T_high           = f38
162 FR_P34              = f39
163 FR_R2               = f40
165 FR_P12              = f41
166 FR_T_low_K          = f42
167 FR_P14              = f43
168 FR_T                = f44
169 FR_P                = f45
172 // Data tables
173 //==============================================================
175 RODATA
177 .align 16
179 LOCAL_OBJECT_START(poly_coeffs)
181 data8 0x3fac6b08d704a0c0, 0x3f83b2ab6fba4e77 // C_3 and C_4
182 data8 0xb17217f7d1cf79ab, 0x00003ffe // C_1
183 data8 0xf5fdeffc162c7541, 0x00003ffc // C_2
184 LOCAL_OBJECT_END(poly_coeffs)
187 LOCAL_OBJECT_START(T_table)
189 // 2^{0.00000 b6 b7 b8 b9 b10}
190 data8 0x8000000000000000, 0x8016302f17467628
191 data8 0x802c6436d0e04f50, 0x80429c17d77c18ed
192 data8 0x8058d7d2d5e5f6b0, 0x806f17687707a7af
193 data8 0x80855ad965e88b83, 0x809ba2264dada76a
194 data8 0x80b1ed4fd999ab6c, 0x80c83c56b50cf77f
195 data8 0x80de8f3b8b85a0af, 0x80f4e5ff089f763e
196 data8 0x810b40a1d81406d4, 0x81219f24a5baa59d
197 data8 0x813801881d886f7b, 0x814e67cceb90502c
198 data8 0x8164d1f3bc030773, 0x817b3ffd3b2f2e47
199 data8 0x8191b1ea15813bfd, 0x81a827baf7838b78
200 data8 0x81bea1708dde6055, 0x81d51f0b8557ec1c
201 data8 0x81eba08c8ad4536f, 0x820225f44b55b33b
202 data8 0x8218af4373fc25eb, 0x822f3c7ab205c89a
203 data8 0x8245cd9ab2cec048, 0x825c62a423d13f0c
204 data8 0x8272fb97b2a5894c, 0x828998760d01faf3
205 data8 0x82a0393fe0bb0ca8, 0x82b6ddf5dbc35906
207 // 2^{0.b1 b2 b3 b4 b5}
208 data8 0x8000000000000000, 0x82cd8698ac2ba1d7
209 data8 0x85aac367cc487b14, 0x88980e8092da8527
210 data8 0x8b95c1e3ea8bd6e6, 0x8ea4398b45cd53c0
211 data8 0x91c3d373ab11c336, 0x94f4efa8fef70961
212 data8 0x9837f0518db8a96f, 0x9b8d39b9d54e5538
213 data8 0x9ef5326091a111ad, 0xa27043030c496818
214 data8 0xa5fed6a9b15138ea, 0xa9a15ab4ea7c0ef8
215 data8 0xad583eea42a14ac6, 0xb123f581d2ac258f
216 data8 0xb504f333f9de6484, 0xb8fbaf4762fb9ee9
217 data8 0xbd08a39f580c36be, 0xc12c4cca66709456
218 data8 0xc5672a115506dadd, 0xc9b9bd866e2f27a2
219 data8 0xce248c151f8480e3, 0xd2a81d91f12ae45a
220 data8 0xd744fccad69d6af4, 0xdbfbb797daf23755
221 data8 0xe0ccdeec2a94e111, 0xe5b906e77c8348a8
222 data8 0xeac0c6e7dd24392e, 0xefe4b99bdcdaf5cb
223 data8 0xf5257d152486cc2c, 0xfa83b2db722a033a
224 LOCAL_OBJECT_END(T_table)
228 .section .text
229 WEAK_LIBM_ENTRY(exp2)
232 {.mfi
233        alloc r32= ar.pfs, 1, 4, 4, 0
234        // will continue only for non-zero normal/denormal numbers
235        fclass.nm p12, p0= f8, 0x1b
236        // GR_TBL_START= pointer to C_1...C_4 followed by T_table
237        addl GR_TBL_START= @ltoff(poly_coeffs), gp
239 {.mlx
240        mov GR_OF_LIMIT= 0xffff + 10              // Exponent of overflow limit
241        movl GR_ROUNDVAL= 0x5a400000              // 1.5*2^(63-10) (SP)
245 // Form special constant 1.5*2^(63-10) to give integer part and first 10
246 // fractional bits of x
247 {.mfi
248        setf.s FR_ROUNDVAL= GR_ROUNDVAL           // Form special constant
249        fcmp.lt.s1 p6, p8= f8, f0                 // X<0 ?
250        nop.i 0
252 {.mfb
253        ld8 GR_COEFF_START= [ GR_TBL_START ]      // Load pointer to coeff table
254        nop.f 0
255  (p12) br.cond.spnt SPECIAL_exp2                 // Branch if nan, inf, zero
259 {.mlx
260        setf.exp FR_OF_LIMIT= GR_OF_LIMIT         // Set overflow limit
261        movl GR_UF_LIMIT= 0xc4866000              // (-2^10-51) = -1075
265 {.mfi
266        ldfpd FR_COEFF3, FR_COEFF4= [ GR_COEFF_START ], 16 // load C_3, C_4
267        fma.s0 f8= f8, f1, f0                     // normalize x
268        nop.i 0
272 {.mmi
273        setf.s FR_UF_LIMIT= GR_UF_LIMIT           // Set underflow limit
274        ldfe FR_COEFF1= [ GR_COEFF_START ], 16    // load C_1
275        mov GR_EXP_CORR= 0xffff-126
279 {.mfi
280        ldfe FR_COEFF2= [ GR_COEFF_START ], 16    // load C_2
281        fma.s1 FR_KF0= f8, f1, FR_ROUNDVAL        // y= x + 1.5*2^(63-10)
282        nop.i 0
286 {.mfi
287        mov GR_MASK= 1023
288        fms.s1 FR_KF= FR_KF0, f1, FR_ROUNDVAL     // (K+f)
289        mov GR_MASK_low= 31
293 {.mfi
294        getf.sig GR_KF0= FR_KF0                   // (K+f)*2^10= round_to_int(y)
295        fcmp.ge.s1 p12, p7= f8, FR_OF_LIMIT       // x >= overflow threshold ?
296        add GR_LOG_TBL= 256, GR_COEFF_START       // Pointer to high T_table
300 {.mmi
301        and GR_F_low= GR_KF0, GR_MASK_low         // f_low
302        and GR_F_high= GR_MASK, GR_KF0            // f_high*32
303        shr GR_K= GR_KF0, 10                      // K
307 {.mmi
308        shladd GR_Flow_ADDR= GR_F_low, 3, GR_COEFF_START // address of 2^{f_low}
309        add GR_BIAS= GR_K, GR_EXP_CORR            // K= bias-2*63
310        shr GR_Fh= GR_F_high, 5                   // f_high
314 {.mfi
315        setf.exp FR_2_TO_K= GR_BIAS               // 2^{K-126}
316        fnma.s1 FR_R= FR_KF, f1, f8               // r= x - (K+f)
317        shladd GR_Fh_ADDR= GR_Fh, 3, GR_LOG_TBL   // address of 2^{f_high}
319 {.mlx
320        ldf8 FR_T_low= [ GR_Flow_ADDR ]           // load T_low= 2^{f_low}
321        movl GR_EMIN= 0xc47f8000                  // EMIN= -1022
325 {.mfi
326        ldf8 FR_T_high= [ GR_Fh_ADDR ]            // load T_high= 2^{f_high}
327  (p7)  fcmp.lt.s1 p12, p7= f8, FR_UF_LIMIT       // x<underflow threshold ?
328        nop.i 0
332 {.mfi
333        setf.s FR_EXPMIN= GR_EMIN                 // FR_EXPMIN= EMIN
334        fma.s1 FR_P34= FR_COEFF4, FR_R, FR_COEFF3 // P34= C_3+C_4*r
335        nop.i 0
337 {.mfb
338        nop.m 0
339        fma.s1 FR_R2= FR_R, FR_R, f0              // r*r
340  (p12) br.cond.spnt OUT_RANGE_exp2
344 {.mfi
345        nop.m 0
346        fma.s1 FR_P12= FR_COEFF2, FR_R, FR_COEFF1 // P12= C_1+C_2*r
347        nop.i 0
351 {.mfi
352        nop.m 0
353        fma.s1 FR_T_low_K= FR_T_low, FR_2_TO_K, f0 // T= 2^{K-126}*T_low
354        nop.i 0
358 {.mfi
359        nop.m 0
360        fma.s1 FR_P14= FR_R2, FR_P34, FR_P12       // P14= P12+r2*P34
361        nop.i 0
365 {.mfi
366        nop.m 0
367        fma.s1 FR_T= FR_T_low_K, FR_T_high, f0     // T= T*T_high
368        nop.i 0
372 {.mfi
373        nop.m 0
374        fcmp.lt.s0 p6, p8= f8, FR_EXPMIN           // underflow (x<EMIN) ?
375        nop.i 0
379 {.mfi
380        nop.m 0
381        fma.s1 FR_P= FR_P14, FR_R, f0              // P= P14*r
382        nop.i 0
386 {.mfb
387        nop.m 0
388        fma.d.s0 f8= FR_P, FR_T, FR_T              // result= T+T*P
389  (p8)  br.ret.sptk b0                             // return
393 {.mfb
394  (p6)  mov GR_Parameter_TAG= 162
395        nop.f 0
396  (p6)  br.cond.sptk __libm_error_region
401 SPECIAL_exp2:
402 {.mfi
403        nop.m 0
404        fclass.m p6, p0= f8, 0x22                  // x= -Infinity ?
405        nop.i 0
409 {.mfi
410        nop.m 0
411        fclass.m p7, p0= f8, 0x21                  // x= +Infinity ?
412        nop.i 0
416 {.mfi
417        nop.m 0
418        fclass.m p8, p0= f8, 0x7                   // x= +/-Zero ?
419        nop.i 0
421 {.mfb
422        nop.m 0
423  (p6)  mov f8= f0                                 // exp2(-Infinity)= 0
424  (p6)  br.ret.spnt b0
428 {.mfb
429        nop.m 0
430        nop.f 0
431  (p7)  br.ret.spnt b0                             // exp2(+Infinity)= +Infinity
435 {.mfb
436        nop.m 0
437  (p8)  mov f8= f1                                 // exp2(+/-0)= 1
438  (p8)  br.ret.spnt b0
442 {.mfb
443        nop.m 0
444        fma.d.s0 f8= f8, f1, f0                    // Remaining cases: NaNs
445        br.ret.sptk b0
450 OUT_RANGE_exp2:
452 // overflow: p8= 1
454 {.mii
455  (p8)  mov GR_EXPMAX= 0x1fffe
456        nop.i 0
457        nop.i 0
461 {.mmb
462  (p8)  mov GR_Parameter_TAG= 161
463  (p8)  setf.exp FR_R= GR_EXPMAX
464        nop.b 999
468 {.mfi
469        nop.m 999
470  (p8)  fma.d.s0 f8= FR_R, FR_R, f0                // Create overflow
471        nop.i 999
473 // underflow: p6= 1
474 {.mii
475  (p6)  mov GR_Parameter_TAG= 162
476  (p6)  mov GR_EXPMAX= 1
477        nop.i 0
481 {.mmb
482        nop.m 0
483  (p6)  setf.exp FR_R= GR_EXPMAX
484        nop.b 999
488 {.mfb
489        nop.m 999
490  (p6)  fma.d.s0 f8= FR_R, FR_R, f0                // Create underflow
491        nop.b 0
495 WEAK_LIBM_END(exp2)
496 libm_alias_double_other (__exp2, exp2)
497 #ifdef SHARED
498 .symver exp2,exp2@@GLIBC_2.29
499 .weak __exp2_compat
500 .set __exp2_compat,__exp2
501 .symver __exp2_compat,exp2@GLIBC_2.2
502 #endif
505 LOCAL_LIBM_ENTRY(__libm_error_region)
507 .prologue
508 {.mfi
509        add GR_Parameter_Y= -32, sp                // Parameter 2 value
510        nop.f 0
511 .save ar.pfs, GR_SAVE_PFS
512        mov GR_SAVE_PFS= ar.pfs                    // Save ar.pfs
515 {.mfi
516 .fframe 64
517        add sp= -64, sp                            // Create new stack
518        nop.f 0
519        mov GR_SAVE_GP= gp                         // Save gp
523 {.mmi
524        stfd [ GR_Parameter_Y ]= FR_Y, 16          // STORE Parameter 2 on stack
525        add GR_Parameter_X= 16, sp                 // Parameter 1 address
526 .save b0, GR_SAVE_B0
527        mov GR_SAVE_B0= b0                         // Save b0
531 .body
532 {.mib
533        stfd [ GR_Parameter_X ]= FR_X              // STORE Parameter 1 on stack
534        add GR_Parameter_RESULT= 0, GR_Parameter_Y // Parameter 3 address
535        nop.b 0
537 {.mib
538        stfd [ GR_Parameter_Y ]= FR_RESULT         // STORE Parameter 3 on stack
539        add GR_Parameter_Y= -16, GR_Parameter_Y
540        br.call.sptk b0= __libm_error_support#    // Call error handling function
544 {.mmi
545        add GR_Parameter_RESULT= 48, sp
546        nop.m 0
547        nop.i 0
551 {.mmi
552        ldfd f8= [ GR_Parameter_RESULT ]          // Get return result off stack
553 .restore sp
554        add sp= 64, sp                            // Restore stack pointer
555        mov b0= GR_SAVE_B0                        // Restore return address
559 {.mib
560        mov gp= GR_SAVE_GP                        // Restore gp
561        mov ar.pfs= GR_SAVE_PFS                   // Restore ar.pfs
562        br.ret.sptk b0                            // Return
567 LOCAL_LIBM_END(__libm_error_region)
569 .type __libm_error_support#, @function
570 .global __libm_error_support#