3 // Copyright (c) 2000, 2001, Intel Corporation
4 // All rights reserved.
6 // Contributed 2/2/2000 by John Harrison, Cristina Iordache, Ted Kubaska,
7 // Bob Norin, Shane Story, and Ping Tak Peter Tang of the
8 // Computational Software Lab, Intel Corporation.
10 // WARRANTY DISCLAIMER
12 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
13 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
14 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
15 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
16 // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
17 // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
18 // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
19 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
20 // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
21 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
22 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
24 // Intel Corporation is the author of this code, and requests that all
25 // problem reports or change requests be submitted to it directly at
26 // http://developer.intel.com/opensource.
28 // *********************************************************************
31 // 2/02/00 hand-optimized
32 // 4/04/00 Unwind support added
33 // 6/20/00 new version
34 // 8/15/00 Bundle added after call to __libm_error_support to properly
35 // set [the previously overwritten] GR_Parameter_RESULT.
37 // *********************************************************************
39 // Function: hypot(x,y) = |(x^2 + y^2) = for double precision values
41 // Also provides cabs functionality.
43 // *********************************************************************
47 // Floating-Point Registers: f8 (Input and Return Value)
51 // General Purpose Registers:
52 // r2,r3,r29 (Scratch)
54 // r37-r40 (Used to pass arguments to error handling routine)
56 // Predicate Registers: p6 - p10
58 // *********************************************************************
60 // IEEE Special Conditions:
62 // All faults and exceptions should be raised correctly.
63 // Overflow can occur.
64 // hypot(Infinity and anything) = +Infinity
65 // hypot(QNaN and anything) = QNaN
66 // hypot(SNaN and anything ) = QNaN
68 // *********************************************************************
71 // x2 = x * x in double-extended
72 // y2 = y * y in double-extended
73 // temp = x2 + y2 in double-extended
74 // sqrt(temp) rounded to double
76 // *********************************************************************
78 #include "libm_support.h"
85 GR_Parameter_RESULT = r38
86 GR_Parameter_TAG = r39
107 .global __ieee754_hypot
111 alloc r32= ar.pfs,0,4,4,0
123 mov r29=0x41d68;; //000
128 // Check if x is an Inf - if so return Inf even
129 // if y is a NaN (C9X)
130 fclass.m.unc p7, p6 = f8, 0x023
135 // if possible overflow, copy f8 to f32
136 // set Denormal, if necessary
138 fma.d.s0 f32=f8,f1,f0
143 // Check if y is an Inf - if so return Inf even
144 // if x is a NaN (C9X)
145 fclass.m.unc p8, p9 = f9, 0x023
151 // For x=inf, multiply y by 1 to raise invalid on y an SNaN
152 // (p7) fma.s0 f9=f9,f1,f0
153 // copy f9 to f33; set Denormal, if necessary
154 fma.d.s0 f33=f9,f1,f0
161 (p6) fclass.m p6,p0=f9,0x7
173 (p9) fclass.m p9,p0=f8,0x7
175 mov r3=0x4167;; //0000
180 fma.s1 f12=f10,f1,f11
186 (p9) fclass.m p8,p0=f9,0x3f
200 (p6) fclass.m p7,p0=f8,0x3f
213 // if f8=Infinity or f9=Zero, return |f8|
214 (p7) fmerge.s f8=f0,f32
219 // if f9=Infinity or f8=Zero, return |f9|
220 (p8) fmerge.s f8=f0,f33
221 (p8) br.ret.spnt b0;;
235 // Identify Natvals, Infs, NaNs, and Zeros
237 fclass.m.unc p7, p0 = f12, 0x1E7
241 // get exponent of x^2+y^2
243 // if special case, set f8
245 (p7) br.ret.spnt b0;;
252 (p6) fma.s1 f14=f12,f8,f0
258 (p6) fma.s1 f15=f8,f7,f0
279 (p6) fnma.s1 f7=f14,f15,f7
285 // P67=231/16+429/16*d
286 (p6) fma.s1 f10=f34,f7,f10
292 (p6) fma.s1 f9=f13,f7,f9
298 (p6) fma.s1 f11=f6,f7,f11
304 (p6) fma.s1 f13=f7,f7,f0
311 (p6) fma.s1 f10=f10,f13,f9
317 (p6) fma.s1 f11=f11,f7,f1
323 (p6) fma.s1 f13=f13,f7,f0
330 (p6) fma.s1 f15=f7,f14,f0
334 // Is x^2 + y^2 well less than the overflow
336 (p6) cmp.lt.unc p7, p8 = r3,r2
338 (p6) fma.s1 f10=f13,f10,f11
345 fma.d.s0 f8=f10,f15,f14
346 // No overflow in this case
347 (p7) br.ret.sptk b0;;
352 (p8) fsetc.s2 0x7F,0x42
353 // Possible overflow path, must detect by
354 // Setting widest range exponent with prevailing
361 // bias+0x400 (bias+EMAX+1)
364 (p8) fma.d.s2 f12=f10,f15,f14
368 (p8) setf.exp f11 = r2
369 (p8) fsetc.s2 0x7F,0x40
370 // Restore Original Mode in S2
375 (p8) fcmp.lt.unc.s1 p9, p10 = f12, f11
380 mov GR_Parameter_TAG = 46
382 (p9) br.ret.sptk b0;;
385 ASM_SIZE_DIRECTIVE(hypot)
387 .proc __libm_error_region
391 add GR_Parameter_Y=-32,sp // Parameter 2 value
393 .save ar.pfs,GR_SAVE_PFS
394 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
398 add sp=-64,sp // Create new stack
400 mov GR_SAVE_GP=gp // Save gp
403 stfd [GR_Parameter_Y] = FR_Y,16 // Save Parameter 2 on stack
404 add GR_Parameter_X = 16,sp // Parameter 1 address
406 mov GR_SAVE_B0=b0 // Save b0
410 stfd [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
411 add GR_Parameter_RESULT = 0,GR_Parameter_Y
412 nop.b 0 // Parameter 3 address
415 stfd [GR_Parameter_Y] = FR_RESULT // Store Parameter 3 on stack
416 add GR_Parameter_Y = -16,GR_Parameter_Y
417 br.call.sptk b0=__libm_error_support# // Call error handling function
422 add GR_Parameter_RESULT = 48,sp
425 ldfd f8 = [GR_Parameter_RESULT] // Get return result off stack
427 add sp = 64,sp // Restore stack pointer
428 mov b0 = GR_SAVE_B0 // Restore return address
431 mov gp = GR_SAVE_GP // Restore gp
432 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
433 br.ret.sptk b0 // Return
435 .endp __libm_error_region
436 ASM_SIZE_DIRECTIVE(__libm_error_region)
437 .type __libm_error_support#,@function
438 .global __libm_error_support#