AuroraMiddleware/glibc
1
0
Fork 0
mirror of https://sourceware.org/git/glibc.git synced 2024-12-12 14:20:13 +00:00
Code Issues Packages Projects Releases Wiki Activity
c61b4d41c9
glibc/ports/sysdeps/ia64/fpu/s_erfcf.S
Ondřej Bílka 382466e04e Fix typos.
2013-08-30 18:08:59 +02:00

984 lines
29 KiB
ArmAsm
Raw Blame History

.file "erfcf.s"
// Copyright (c) 2002 - 2005, Intel Corporation
// All rights reserved.
//
// Contributed 2002 by the Intel Numerics Group, Intel Corporation
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// * Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote
// products derived from this software without specific prior written
// permission.
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
// Intel Corporation is the author of this code, and requests that all
// problem reports or change requests be submitted to it directly at
// http://www.intel.com/software/products/opensource/libraries/num.htm.
//
// History
//==============================================================
// 01/17/02 Initial version
// 05/20/02 Cleaned up namespace and sf0 syntax
// 02/06/03 Reordered header: .section, .global, .proc, .align
// 03/31/05 Reformatted delimiters between data tables
//
// API
//==============================================================
// float erfcf(float)
//
// Overview of operation
//==============================================================
// 1. 0 <= x <= 10.06
//
// erfcf(x) = P15(x) * exp( -x^2 )
//
// Comment:
//
// Let x(0)=0, x(i) = 2^(i), i=1,...3, x(4)= 10.06
//
// Let x(i)<= x < x(i+1).
// We can find i as exponent of argument x (let i = 0 for 0<= x < 2 )
//
// Let P15(x) - polynomial approximation of degree 15 for function
// erfcf(x) * exp( x^2) and x(i) <= x <= x(i+1), i = 0,1,2,3
// Polynomial coefficients we have in the table erfc_p_table.
//
// So we can find result for erfcf(x) as above.
// Algorithm description for exp function see below.
//
// 2. -4.4 <= x < 0
//
// erfcf(x) = 2.0 - erfcf(-x)
//
// 3. x > 10.06
//
// erfcf(x) ~=~ 0.0
//
// 4. x < -4.4
//
// erfcf(x) ~=~ 2.0
// Special values
//==============================================================
// erfcf(+0) = 1.0
// erfcf(-0) = 1.0
// erfcf(+qnan) = +qnan
// erfcf(-qnan) = -qnan
// erfcf(+snan) = +qnan
// erfcf(-snan) = -qnan
// erfcf(-inf) = 2.0
// erfcf(+inf) = +0
//==============================================================
// Take double exp(double) from libm_64.
//
// Overview of operation
//==============================================================
// Take the input x. w is "how many log2/128 in x?"
// w = x * 128/log2
// n = int(w)
// x = n log2/128 + r + delta
// n = 128M + index_1 + 2^4 index_2
// x = M log2 + (log2/128) index_1 + (log2/8) index_2 + r + delta
// exp(x) = 2^M 2^(index_1/128) 2^(index_2/8) exp(r) exp(delta)
// Construct 2^M
// Get 2^(index_1/128) from table_1;
// Get 2^(index_2/8) from table_2;
// Calculate exp(r) by series
// r = x - n (log2/128)_high
// delta = - n (log2/128)_low
// Calculate exp(delta) as 1 + delta
//
// Comment for erfcf:
//
// Let exp(r) = 1 + x + 0.5*x^2 + (1/6)*x^3
// Let delta = 0.
//==============================================================
//
// Registers used
//==============================================================
// Floating Point registers used:
// f8, input
// f6,f7,f9 -> f11, f32 -> f92
// General registers used:
// r14 -> r22,r32 -> r50
// Predicate registers used:
// p6 -> p15
// Assembly macros
//==============================================================
EXP_AD_TB1 = r14
exp_GR_sig_inv_ln2 = r15
exp_TB1_size = r16
exp_GR_rshf_2to56 = r17
exp_GR_exp_2tom56 = r18
exp_GR_rshf = r33
EXP_AD_TB2 = r34
EXP_AD_P = r35
exp_GR_N = r36
exp_GR_index_1 = r37
exp_GR_index_2_16 = r38
exp_GR_biased_M = r39
EXP_AD_T1 = r40
EXP_AD_T2 = r41
exp_TB2_size = r42
// GR for erfcf(x)
//==============================================================
GR_IndxPlusBias = r19
GR_ExpMask = r20
GR_BIAS = r21
GR_ShftPi_bias = r22
GR_P_POINT_1 = r43
GR_P_POINT_2 = r44
GR_P_POINT_3 = r45
GR_P_POINT_4 = r46
GR_ShftPi = r47
GR_EpsNorm = r48
GR_05 = r49
GR_1_by_6 = r50
// GR for __libm_support call
//==============================================================
GR_SAVE_B0 = r43
GR_SAVE_PFS = r44
GR_SAVE_GP = r45
GR_SAVE_SP = r46
GR_Parameter_X = r47
GR_Parameter_Y = r48
GR_Parameter_RESULT = r49
GR_Parameter_TAG = r50
// FR for exp(-x^2)
//==============================================================
FR_X = f10
FR_Y = f1
FR_RESULT = f8
EXP_2TOM56 = f6
EXP_INV_LN2_2TO63 = f7
EXP_W_2TO56_RSH = f9
exp_ln2_by_128_hi = f11
EXP_RSHF_2TO56 = f32
exp_ln2_by_128_lo = f33
EXP_RSHF = f34
EXP_Nfloat = f35
exp_r = f36
exp_rsq = f37
EXP_2M = f38
exp_S1 = f39
exp_T1 = f40
exp_P = f41
exp_S = f42
EXP_NORM_f8 = f43
exp_S2 = f44
exp_T2 = f45
// FR for erfcf(x)
//==============================================================
FR_AbsArg = f46
FR_Tmp = f47
FR_Tmp1 = f48
FR_Tmpf = f49
FR_NormX = f50
FR_A15 = f51
FR_A14 = f52
FR_A13 = f53
FR_A12 = f54
FR_A11 = f55
FR_A10 = f56
FR_A9 = f57
FR_A8 = f58
FR_A7 = f59
FR_A6 = f60
FR_A5 = f61
FR_A4 = f62
FR_A3 = f63
FR_A2 = f64
FR_A1 = f65
FR_A0 = f66
FR_P15_0_1 = f67
FR_P15_1_1 = f68
FR_P15_1_2 = f69
FR_P15_2_1 = f70
FR_P15_2_2 = f71
FR_P15_3_1 = f72
FR_P15_3_2 = f73
FR_P15_4_1 = f74
FR_P15_4_2 = f75
FR_P15_7_1 = f76
FR_P15_7_2 = f77
FR_P15_8_1 = f78
FR_P15_9_1 = f79
FR_P15_9_2 = f80
FR_P15_13_1 = f81
FR_P15_14_1 = f82
FR_P15_14_2 = f83
FR_2 = f84
FR_05 = f85
FR_1_by_6 = f86
FR_Pol = f87
FR_Exp = f88
FR_POS_ARG_ASYMP = f89
FR_NEG_ARG_ASYMP = f90
FR_UnfBound = f91
FR_EpsNorm = f92
// Data tables
//==============================================================
RODATA
.align 16
// ************* DO NOT CHANGE ORDER OF THESE TABLES ********************
// double-extended 1/ln(2)
// 3fff b8aa 3b29 5c17 f0bb be87fed0691d3e88
// 3fff b8aa 3b29 5c17 f0bc
// For speed the significand will be loaded directly with a movl and setf.sig
// and the exponent will be bias+63 instead of bias+0. Thus subsequent
// computations need to scale appropriately.
// The constant 128/ln(2) is needed for the computation of w. This is also
// obtained by scaling the computations.
//
// Two shifting constants are loaded directly with movl and setf.d.
// 1. EXP_RSHF_2TO56 = 1.1000..00 * 2^(63-7)
// This constant is added to x*1/ln2 to shift the integer part of
// x*128/ln2 into the rightmost bits of the significand.
// The result of this fma is EXP_W_2TO56_RSH.
// 2. EXP_RSHF = 1.1000..00 * 2^(63)
// This constant is subtracted from EXP_W_2TO56_RSH * 2^(-56) to give
// the integer part of w, n, as a floating-point number.
// The result of this fms is EXP_Nfloat.
LOCAL_OBJECT_START(exp_table_1)
data4 0x4120f5c3, 0x408ccccd //POS_ARG_ASYMP = 10.06, NEG_ARG_ASYMP = 4.4
data4 0x41131Cdf, 0x00800000 //UnfBound ~=~ 9.1, EpsNorm ~=~ 1.1754944e-38
//
data8 0xb17217f7d1cf79ab , 0x00003ff7 // ln2/128 hi
data8 0xc9e3b39803f2f6af , 0x00003fb7 // ln2/128 lo
//
// Table 1 is 2^(index_1/128) where
// index_1 goes from 0 to 15
//
data8 0x8000000000000000 , 0x00003FFF
data8 0x80B1ED4FD999AB6C , 0x00003FFF
data8 0x8164D1F3BC030773 , 0x00003FFF
data8 0x8218AF4373FC25EC , 0x00003FFF
data8 0x82CD8698AC2BA1D7 , 0x00003FFF
data8 0x8383594EEFB6EE37 , 0x00003FFF
data8 0x843A28C3ACDE4046 , 0x00003FFF
data8 0x84F1F656379C1A29 , 0x00003FFF
data8 0x85AAC367CC487B15 , 0x00003FFF
data8 0x8664915B923FBA04 , 0x00003FFF
data8 0x871F61969E8D1010 , 0x00003FFF
data8 0x87DB357FF698D792 , 0x00003FFF
data8 0x88980E8092DA8527 , 0x00003FFF
data8 0x8955EE03618E5FDD , 0x00003FFF
data8 0x8A14D575496EFD9A , 0x00003FFF
data8 0x8AD4C6452C728924 , 0x00003FFF
LOCAL_OBJECT_END(exp_table_1)
// Table 2 is 2^(index_1/8) where
// index_2 goes from 0 to 7
LOCAL_OBJECT_START(exp_table_2)
data8 0x8000000000000000 , 0x00003FFF
data8 0x8B95C1E3EA8BD6E7 , 0x00003FFF
data8 0x9837F0518DB8A96F , 0x00003FFF
data8 0xA5FED6A9B15138EA , 0x00003FFF
data8 0xB504F333F9DE6484 , 0x00003FFF
data8 0xC5672A115506DADD , 0x00003FFF
data8 0xD744FCCAD69D6AF4 , 0x00003FFF
data8 0xEAC0C6E7DD24392F , 0x00003FFF
LOCAL_OBJECT_END(exp_table_2)
LOCAL_OBJECT_START(erfc_p_table)
// Pol_0
data8 0xBEA3260C63CB0446 //A15 = -5.70673541831883454676e-07
data8 0x3EE63D6178077654 //A14 = +1.06047480138940182343e-05
data8 0xBF18646BC5FC70A7 //A13 = -9.30491237309283694347e-05
data8 0x3F40F92F909117FE //A12 = +5.17986512144075019133e-04
data8 0xBF611344289DE1E6 //A11 = -2.08438217390159994419e-03
data8 0x3F7AF9FE6AD16DC0 //A10 = +6.58606893292862351928e-03
data8 0xBF91D219E196CBA7 //A9 = -1.74030345858217321001e-02
data8 0x3FA4AFDDA355854C //A8 = +4.04042493708041968315e-02
data8 0xBFB5D465BB7025AE //A7 = -8.52721769916999425445e-02
data8 0x3FC54C15A95B717D //A6 = +1.66384418195672549029e-01
data8 0xBFD340A75B4B1AB5 //A5 = -3.00821150926292166899e-01
data8 0x3FDFFFC0BFCD247F //A4 = +4.99984919839853542841e-01
data8 0xBFE81270C361852B //A3 = -7.52251035312075583309e-01
data8 0x3FEFFFFFC67295FC //A2 = +9.99999892800303301771e-01
data8 0xBFF20DD74F8CD2BF //A1 = -1.12837916445020868099e+00
data8 0x3FEFFFFFFFFE7C1D //A0 = +9.99999999988975570714e-01
// Pol_1
data8 0xBDE8EC4BDD953B56 //A15 = -1.81338928934942767144e-10
data8 0x3E43607F269E2A1C //A14 = +9.02309090272196442358e-09
data8 0xBE8C4D9E69C10E02 //A13 = -2.10875261143659275328e-07
data8 0x3EC9CF2F84566725 //A12 = +3.07671055805877356583e-06
data8 0xBF007980B1B46A4D //A11 = -3.14228438702169818945e-05
data8 0x3F2F4C3AD6DEF24A //A10 = +2.38783056770846320260e-04
data8 0xBF56F5129F8D30FA //A9 = -1.40120333363130546426e-03
data8 0x3F7AA6C7ABFC38EE //A8 = +6.50671002200751820429e-03
data8 0xBF98E7522CB84BEF //A7 = -2.43199195666185511109e-02
data8 0x3FB2F68EB1C3D073 //A6 = +7.40746673580490638637e-02
data8 0xBFC7C16055AC6385 //A5 = -1.85588876564704611769e-01
data8 0x3FD8A707AEF5A440 //A4 = +3.85194702967570635211e-01
data8 0xBFE547BFE39AE2EA //A3 = -6.65008492032112467310e-01
data8 0x3FEE7C91BDF13578 //A2 = +9.52706213932898128515e-01
data8 0xBFF1CB5B61F8C589 //A1 = -1.11214769621105541214e+00
data8 0x3FEFEA56BC81FD37 //A0 = +9.97355812243688815239e-01
// Pol_2
data8 0xBD302724A12F46E0 //A15 = -5.73866382814058809406e-14
data8 0x3D98889B75D3102E //A14 = +5.57829983681360947356e-12
data8 0xBDF16EA15074A1E9 //A13 = -2.53671153922423457844e-10
data8 0x3E3EC6E688CFEE5F //A12 = +7.16581828336436419561e-09
data8 0xBE82E5ED44C52609 //A11 = -1.40802202239825487803e-07
data8 0x3EC120BE5CE42353 //A10 = +2.04180535157522081699e-06
data8 0xBEF7B8B0311A1911 //A9 = -2.26225266204633600888e-05
data8 0x3F29A281F43FC238 //A8 = +1.95577968156184077632e-04
data8 0xBF55E19858B3B7A4 //A7 = -1.33552434527526534043e-03
data8 0x3F7DAC8C3D12E5FD //A6 = +7.24463253680473816303e-03
data8 0xBF9FF9C04613FB47 //A5 = -3.12261622211693854028e-02
data8 0x3FBB3D5DBF9D9366 //A4 = +1.06405123978743883370e-01
data8 0xBFD224DE9F62C258 //A3 = -2.83500342989133623476e-01
data8 0x3FE28A95CB8C6D3E //A2 = +5.79417131000276437708e-01
data8 0xBFEC21205D358672 //A1 = -8.79043752717008257224e-01
data8 0x3FEDAE44D5EDFE5B //A0 = +9.27523057776805771830e-01
// Pol_3
data8 0xBCA3BCA734AC82F1 //A15 = -1.36952437983096410260e-16
data8 0x3D16740DC3990612 //A14 = +1.99425676175410093285e-14
data8 0xBD77F4353812C46A //A13 = -1.36162367755616790260e-12
data8 0x3DCFD0BE13C73DB4 //A12 = +5.78718761040355136007e-11
data8 0xBE1D728DF71189B4 //A11 = -1.71406885583934105120e-09
data8 0x3E64252C8CB710B5 //A10 = +3.75233795940731111303e-08
data8 0xBEA514B93180F33D //A9 = -6.28261292774310809962e-07
data8 0x3EE1381118CC7151 //A8 = +8.21066421390821904504e-06
data8 0xBF1634404FB0FA72 //A7 = -8.47019436358372148764e-05
data8 0x3F46B2CBBCF0EB32 //A6 = +6.92700845213200923490e-04
data8 0xBF725C2B445E6D81 //A5 = -4.48243046949004063741e-03
data8 0x3F974E7CFA4D89D9 //A4 = +2.27603462002522228717e-02
data8 0xBFB6D7BAC2E342D1 //A3 = -8.92292714882032736443e-02
data8 0x3FD0D156AD9CE2A6 //A2 = +2.62777013343603696631e-01
data8 0xBFE1C228572AADB0 //A1 = -5.54950876471982857725e-01
data8 0x3FE8A739F48B9A3B //A0 = +7.70413377406675619766e-01
LOCAL_OBJECT_END(erfc_p_table)
.section .text
GLOBAL_LIBM_ENTRY(erfcf)
// Form index i for table erfc_p_table as exponent of x
// We use i + bias in real calculations
{ .mlx
getf.exp GR_IndxPlusBias = f8 // (sign + exp + bias) of x
movl exp_GR_sig_inv_ln2 = 0xb8aa3b295c17f0bc //signif.of 1/ln2
}
{ .mlx
addl EXP_AD_TB1 = @ltoff(exp_table_1), gp
movl exp_GR_rshf_2to56 = 0x4768000000000000 // 1.100 2^(63+56)
}
;;
// Form argument EXP_NORM_f8 for exp(-x^2)
{ .mfi
ld8 EXP_AD_TB1 = [EXP_AD_TB1]
fcmp.ge.s1 p6,p7 = f8, f0 // p6: x >= 0 ,p7: x<0
mov GR_BIAS = 0x0FFFF
}
{ .mfi
mov exp_GR_exp_2tom56 = 0xffff-56
fnma.s1 EXP_NORM_f8 = f8, f8, f0 // -x^2
mov GR_ExpMask = 0x1ffff
}
;;
// Form two constants we need
// 1/ln2 * 2^63 to compute w = x * 1/ln2 * 128
// 1.1000..000 * 2^(63+63-7) to right shift int(w) into the significand
// p9: x = 0,+inf,-inf,nan,unnorm.
// p10: x!= 0,+inf,-inf,nan,unnorm.
{ .mfi
setf.sig EXP_INV_LN2_2TO63 = exp_GR_sig_inv_ln2 // Form 1/ln2*2^63
fclass.m p9,p10 = f8,0xef
shl GR_ShftPi_bias = GR_BIAS, 7
}
{ .mfi
setf.d EXP_RSHF_2TO56 = exp_GR_rshf_2to56 //Const 1.10*2^(63+56)
nop.f 0
and GR_IndxPlusBias = GR_IndxPlusBias, GR_ExpMask // i + bias
}
;;
{ .mfi
alloc r32 = ar.pfs, 0, 15, 4, 0
(p6) fma.s1 FR_AbsArg = f1, f0, f8 // |x| if x >= 0
cmp.lt p15,p0 = GR_IndxPlusBias, GR_BIAS//p15: i < 0 (for |x|<1)
}
{ .mlx
setf.exp EXP_2TOM56 = exp_GR_exp_2tom56 //2^-56 for scaling Nfloat
movl exp_GR_rshf = 0x43e8000000000000 //1.10 2^63,right shift.
}
;;
{ .mfi
ldfps FR_POS_ARG_ASYMP, FR_NEG_ARG_ASYMP = [EXP_AD_TB1],8
nop.f 0
(p15) mov GR_IndxPlusBias = GR_BIAS //Let i = 0 if i < 0
}
{ .mlx
mov GR_P_POINT_3 = 0x1A0
movl GR_05 = 0x3fe0000000000000
}
;;
// Form shift GR_ShftPi from the beginning of erfc_p_table
// to the polynomial with number i
{ .mfi
ldfps FR_UnfBound, FR_EpsNorm = [EXP_AD_TB1],8
nop.f 0
shl GR_ShftPi = GR_IndxPlusBias, 7
}
{ .mfi
setf.d EXP_RSHF = exp_GR_rshf // Form right shift 1.100 * 2^63
(p7) fms.s1 FR_AbsArg = f1, f0, f8 // |x| if x < 0
mov exp_TB1_size = 0x100
}
;;
// Form pointer GR_P_POINT_3 to the beginning of erfc_p_table
{ .mfi
setf.d FR_05 = GR_05
nop.f 0
sub GR_ShftPi = GR_ShftPi,GR_ShftPi_bias
}
{ .mfb
add GR_P_POINT_3 = GR_P_POINT_3, EXP_AD_TB1
nop.f 0
(p9) br.cond.spnt SPECIAL // For x = 0,+inf,-inf,nan,unnorm
}
;;
{ .mfi
add GR_P_POINT_1 = GR_P_POINT_3, GR_ShftPi
nop.f 0
add GR_P_POINT_2 = GR_P_POINT_3, GR_ShftPi
}
{ .mfi
ldfe exp_ln2_by_128_hi = [EXP_AD_TB1],16
fma.s1 FR_NormX = f8,f1,f0
add GR_P_POINT_3 = GR_P_POINT_3, GR_ShftPi
}
;;
// Load coefficients for polynomial P15(x)
{ .mfi
ldfpd FR_A15, FR_A14 = [GR_P_POINT_1], 16
nop.f 0
add GR_P_POINT_3 = 0x30, GR_P_POINT_3
}
{ .mfi
ldfe exp_ln2_by_128_lo = [EXP_AD_TB1], 16
nop.f 0
add GR_P_POINT_2 = 0x20, GR_P_POINT_2
}
;;
// Now EXP_AD_TB1 points to the beginning of table 1
{ .mlx
ldfpd FR_A13, FR_A12 = [GR_P_POINT_1]
movl GR_1_by_6 = 0x3FC5555555555555
}
{ .mfi
add GR_P_POINT_4 = 0x30, GR_P_POINT_2
nop.f 0
nop.i 0
}
;;
{ .mfi
ldfpd FR_A11, FR_A10 = [GR_P_POINT_2]
fma.s1 FR_2 = f1, f1, f1
mov exp_TB2_size = 0x80
}
{ .mfi
ldfpd FR_A9, FR_A8 = [GR_P_POINT_3],16
nop.f 0
add GR_P_POINT_1 = 0x60 ,GR_P_POINT_1
}
;;
// W = X * Inv_log2_by_128
// By adding 1.10...0*2^63 we shift and get round_int(W) in significand.
// We actually add 1.10...0*2^56 to X * Inv_log2 to do the same thing.
{ .mfi
ldfpd FR_A7, FR_A6 = [GR_P_POINT_3]
fma.s1 EXP_W_2TO56_RSH = EXP_NORM_f8,EXP_INV_LN2_2TO63,EXP_RSHF_2TO56
add EXP_AD_TB2 = exp_TB1_size, EXP_AD_TB1
}
{ .mfi
ldfpd FR_A5, FR_A4 = [GR_P_POINT_4], 16
nop.f 0
nop.i 0
}
;;
{ .mfi
ldfpd FR_A3, FR_A2 = [GR_P_POINT_4]
fmerge.s FR_X = f8,f8
nop.i 0
}
{ .mfi
ldfpd FR_A1, FR_A0 = [GR_P_POINT_1]
nop.f 0
nop.i 0
}
;;
//p14: x < - NEG_ARG_ASYMP = -4.4 -> erfcf(x) ~=~ 2.0
{ .mfi
setf.d FR_1_by_6 = GR_1_by_6
(p7) fcmp.gt.unc.s1 p14,p0 = FR_AbsArg, FR_NEG_ARG_ASYMP //p7: x < 0
nop.i 0
}
;;
//p15: x > POS_ARG_ASYMP = 10.06 -> erfcf(x) ~=~ 0.0
{ .mfi
nop.m 0
(p6) fcmp.gt.unc.s1 p15,p0 = FR_AbsArg, FR_POS_ARG_ASYMP //p6: x > 0
nop.i 0
}
;;
{ .mfi
nop.m 0
fcmp.le.s1 p8,p0 = FR_NormX, FR_UnfBound // p8: x <= UnfBound
nop.i 0
}
{ .mfb
nop.m 0
(p14) fnma.s.s0 FR_RESULT = FR_EpsNorm, FR_EpsNorm, FR_2//y = 2 if x <-4.4
(p14) br.ret.spnt b0
}
;;
// Nfloat = round_int(W)
// The signficand of EXP_W_2TO56_RSH contains the rounded integer part of W,
// as a twos complement number in the lower bits (that is, it may be negative).
// That twos complement number (called N) is put into exp_GR_N.
// Since EXP_W_2TO56_RSH is scaled by 2^56, it must be multiplied by 2^-56
// before the shift constant 1.10000 * 2^63 is subtracted to yield EXP_Nfloat.
// Thus, EXP_Nfloat contains the floating point version of N
{ .mfi
nop.m 0
fms.s1 EXP_Nfloat = EXP_W_2TO56_RSH, EXP_2TOM56, EXP_RSHF
nop.i 0
}
{ .mfb
(p15) mov GR_Parameter_TAG = 209
(p15) fma.s.s0 FR_RESULT = FR_EpsNorm,FR_EpsNorm,f0 //Result.for x>10.06
(p15) br.cond.spnt __libm_error_region
}
;;
// Now we can calculate polynomial P15(x)
{ .mfi
nop.m 0
fma.s1 FR_P15_1_1 = FR_AbsArg, FR_AbsArg, f0 // x ^2
nop.i 0
}
{ .mfi
nop.m 0
fma.s1 FR_P15_0_1 = FR_A15, FR_AbsArg, FR_A14
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 FR_P15_1_2 = FR_A13, FR_AbsArg, FR_A12
nop.i 0
}
;;
{ .mfi
getf.sig exp_GR_N = EXP_W_2TO56_RSH
fma.s1 FR_P15_2_1 = FR_A9, FR_AbsArg, FR_A8
nop.i 0
}
{ .mfi
nop.m 0
fma.s1 FR_P15_2_2 = FR_A11, FR_AbsArg, FR_A10
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 FR_P15_3_1 = FR_A5, FR_AbsArg, FR_A4
nop.i 0
}
{ .mfi
nop.m 0
fma.s1 FR_P15_3_2 = FR_A7, FR_AbsArg, FR_A6
nop.i 0
}
;;
// exp_GR_index_1 has index_1
// exp_GR_index_2_16 has index_2 * 16
// exp_GR_biased_M has M
// exp_GR_index_1_16 has index_1 * 16
// r2 has true M
{ .mfi
and exp_GR_index_1 = 0x0f, exp_GR_N
fma.s1 FR_P15_4_1 = FR_A1, FR_AbsArg, FR_A0
shr r2 = exp_GR_N, 0x7
}
{ .mfi
and exp_GR_index_2_16 = 0x70, exp_GR_N
fma.s1 FR_P15_4_2 = FR_A3, FR_AbsArg, FR_A2
nop.i 0
}
;;
// EXP_AD_T1 has address of T1
// EXP_AD_T2 has address if T2
{ .mfi
add EXP_AD_T2 = EXP_AD_TB2, exp_GR_index_2_16
nop.f 0
shladd EXP_AD_T1 = exp_GR_index_1, 4, EXP_AD_TB1
}
{ .mfi
addl exp_GR_biased_M = 0xffff, r2
fnma.s1 exp_r = EXP_Nfloat, exp_ln2_by_128_hi, EXP_NORM_f8
nop.i 0
}
;;
// Create Scale = 2^M
// r = x - Nfloat * ln2_by_128_hi
{ .mfi
setf.exp EXP_2M = exp_GR_biased_M
fma.s1 FR_P15_7_1 = FR_P15_0_1, FR_P15_1_1, FR_P15_1_2
nop.i 0
}
{ .mfi
ldfe exp_T2 = [EXP_AD_T2]
nop.f 0
nop.i 0
}
;;
// Load T1 and T2
{ .mfi
ldfe exp_T1 = [EXP_AD_T1]
fma.s1 FR_P15_7_2 = FR_P15_1_1, FR_P15_1_1, f0 // x^4
nop.i 0
}
{ .mfi
nop.m 0
fma.s1 FR_P15_8_1 = FR_P15_1_1, FR_P15_2_2, FR_P15_2_1
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 FR_P15_9_1 = FR_P15_1_1, FR_P15_4_2, FR_P15_4_1
nop.i 0
}
{ .mfi
nop.m 0
fma.s1 FR_P15_9_2 = FR_P15_1_1, FR_P15_3_2, FR_P15_3_1
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 exp_P = FR_1_by_6, exp_r, FR_05
nop.i 0
}
{ .mfi
nop.m 0
fma.s1 exp_rsq = exp_r, exp_r, f0
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 FR_P15_13_1 = FR_P15_7_2, FR_P15_7_1, FR_P15_8_1
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 FR_P15_14_1 = FR_P15_7_2, FR_P15_9_2, FR_P15_9_1
nop.i 0
}
{ .mfi
nop.m 0
fma.s1 FR_P15_14_2 = FR_P15_7_2, FR_P15_7_2, f0 // x^8
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 exp_P = exp_P, exp_rsq, exp_r
nop.i 0
}
{ .mfi
nop.m 0
fma.s1 exp_S1 = EXP_2M, exp_T2, f0
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 FR_Pol = FR_P15_14_2, FR_P15_13_1, FR_P15_14_1 // P15(x)
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 exp_S = exp_S1, exp_T1, f0
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s1 FR_Exp = exp_S, exp_P, exp_S // exp(-x^2)
nop.i 0
}
;;
{ .mfi
nop.m 0
fma.s.s0 FR_Tmpf = f8, f1, f0 // Flag d
nop.i 0
}
;;
//p6: result for 0 < x < = POS_ARG_ASYMP
//p7: result for - NEG_ARG_ASYMP <= x < 0
//p8: exit for - NEG_ARG_ASYMP <= x <= UnfBound, x!=0
.pred.rel "mutex",p6,p7
{ .mfi
nop.m 0
(p6) fma.s.s0 f8 = FR_Exp, FR_Pol, f0
nop.i 0
}
{ .mfb
mov GR_Parameter_TAG = 209
(p7) fnma.s.s0 f8 = FR_Exp, FR_Pol, FR_2
(p8) br.ret.sptk b0
}
;;
//p10: branch for UnfBound < x < = POS_ARG_ASYMP
{ .mfb
nop.m 0
nop.f 0
(p10) br.cond.spnt __libm_error_region
}
;;
//Only via (p9) br.cond.spnt SPECIAL for x = 0,+inf,-inf,nan,unnorm
SPECIAL:
{ .mfi
nop.m 0
fclass.m.unc p10,p0 = f8,0x07 // p10: x = 0
nop.i 0
}
;;
{ .mfi
nop.m 0
fclass.m.unc p11,p0 = f8,0x21 // p11: x = +inf
nop.i 0
}
;;
{ .mfi
nop.m 0
fclass.m.unc p12,p0 = f8,0x22 // p12 x = -inf
nop.i 0
}
{ .mfb
nop.m 0
(p10) fma.s.s0 f8 = f1, f1, f0
(p10) br.ret.sptk b0 // Quick exit for x = 0
}
;;
{ .mfi
nop.m 0
fclass.m.unc p13,p0 = f8,0xc3 // p13: x = nan
nop.i 0
}
{ .mfb
nop.m 0
(p11) fma.s.s0 f8 = f0, f1, f0
(p11) br.ret.spnt b0 // Quick exit for x = +inf
}
;;
{ .mfi
nop.m 0
fclass.m.unc p14,p0 = f8,0x0b // P14: x = unnormalized
nop.i 0
}
{ .mfb
nop.m 0
(p12) fma.s.s0 f8 = f1, f1, f1
(p12) br.ret.spnt b0 // Quick exit for x = -inf
}
;;
{ .mfb
nop.m 0
(p13) fma.s.s0 f8 = f8, f1, f0
(p13) br.ret.sptk b0 // Quick exit for x = nan
}
;;
{ .mfb
nop.m 0
(p14) fnma.s.s0 f8 = f8, f1, f1
(p14) br.ret.sptk b0 // Quick exit for x = unnormalized
}
;;
GLOBAL_LIBM_END(erfcf)
// Call via (p10) br.cond.spnt __libm_error_region
// for UnfBound < x < = POS_ARG_ASYMP
// and
//
// call via (p15) br.cond.spnt __libm_error_region
// for x > POS_ARG_ASYMP
LOCAL_LIBM_ENTRY(__libm_error_region)
.prologue
{ .mfi
add GR_Parameter_Y=-32,sp // Parameter 2 value
nop.f 0
.save ar.pfs,GR_SAVE_PFS
mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
}
{ .mfi
.fframe 64
add sp=-64,sp // Create new stack
nop.f 0
mov GR_SAVE_GP=gp // Save gp
};;
{ .mmi
stfs [GR_Parameter_Y] = FR_Y,16 // STORE Parameter 2 on stack
add GR_Parameter_X = 16,sp // Parameter 1 address
.save b0, GR_SAVE_B0
mov GR_SAVE_B0=b0 // Save b0
};;
.body
{ .mib
stfs [GR_Parameter_X] = FR_X // STORE Parameter 1 on stack
add GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address
nop.b 0
}
{ .mib
stfs [GR_Parameter_Y] = FR_RESULT // STORE Parameter 3 on stack
add GR_Parameter_Y = -16,GR_Parameter_Y
br.call.sptk b0=__libm_error_support# // Call error handling function
};;
{ .mmi
nop.m 0
nop.m 0
add GR_Parameter_RESULT = 48,sp
};;
{ .mmi
ldfs f8 = [GR_Parameter_RESULT] // Get return result off stack
.restore sp
add sp = 64,sp // Restore stack pointer
mov b0 = GR_SAVE_B0 // Restore return address
};;
{ .mib
mov gp = GR_SAVE_GP // Restore gp
mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
br.ret.sptk b0 // Return
};;
LOCAL_LIBM_END(__libm_error_region)
.type __libm_error_support#,@function
.global __libm_error_support#
Reference in New Issue View Git Blame Copy Permalink