1 /* FMA/AVX2 version of IEEE 754 expf.
2 Copyright (C) 2017 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, see
17 <http://www.gnu.org/licenses/>. */
21 /* Short algorithm description:
23 Let K = 64 (table size).
24 e^x = 2^(x/log(2)) = 2^n * T[j] * (1 + P(y))
26 x = m*log(2)/K + y, y in [0.0..log(2)/K]
27 m = n*K + j, m,n,j - signed integer, j in [0..K-1]
28 values of 2^(j/K) are tabulated as T[j].
30 P(y) is a minimax polynomial approximation of expf(x)-1
31 on small interval [0.0..log(2)/K].
33 P(y) = P3*y*y*y*y + P2*y*y*y + P1*y*y + P0*y, calculated as
34 z = y*y; P(y) = (P3*z + P1)*z + (P2*z + P0)*y
40 expf(x) = 1 for subnormals
41 for finite argument, only expf(0)=1 is exact
42 expf(x) overflows if x>88.7228317260742190
43 expf(x) underflows if x<-103.972076416015620
46 .section .text.fma,"ax",@progbits
47 ENTRY(__ieee754_expf_fma)
48 /* Input: single precision x in %xmm0 */
49 vcvtss2sd %xmm0, %xmm0, %xmm1 /* Convert x to double precision */
50 vmovd %xmm0, %ecx /* Copy x */
51 vmovsd L(DP_KLN2)(%rip), %xmm2 /* DP K/log(2) */
52 vfmadd213sd L(DP_RD)(%rip), %xmm1, %xmm2 /* DP x*K/log(2)+RD */
53 vmovsd L(DP_P2)(%rip), %xmm3 /* DP P2 */
54 movl %ecx, %eax /* x */
55 andl $0x7fffffff, %ecx /* |x| */
56 lea L(DP_T)(%rip), %rsi /* address of table T[j] */
57 vmovsd L(DP_P3)(%rip), %xmm4 /* DP P3 */
59 cmpl $0x42ad496b, %ecx /* |x|<125*log(2) ? */
60 jae L(special_paths_fma)
62 /* Here if |x|<125*log(2) */
63 cmpl $0x31800000, %ecx /* |x|<2^(-28) ? */
66 /* Main path: here if 2^(-28)<=|x|<125*log(2) */
67 /* %xmm2 = SP x*K/log(2)+RS */
69 vsubsd L(DP_RD)(%rip), %xmm2, %xmm2 /* DP t=round(x*K/log(2)) */
70 movl %eax, %edx /* n*K+j with trash */
71 andl $0x3f, %eax /* bits of j */
72 vmovsd (%rsi,%rax,8), %xmm5 /* T[j] */
73 andl $0xffffffc0, %edx /* bits of n */
75 vfmadd132sd L(DP_NLN2K)(%rip), %xmm1, %xmm2 /* DP y=x-t*log(2)/K */
76 vmulsd %xmm2, %xmm2, %xmm6 /* DP z=y*y */
79 vfmadd213sd L(DP_P1)(%rip), %xmm6, %xmm4 /* DP P3*z + P1 */
80 vfmadd213sd L(DP_P0)(%rip), %xmm6, %xmm3 /* DP P2*z+P0 */
82 addl $0x1fc0, %edx /* bits of n + SP exponent bias */
83 shll $17, %edx /* SP 2^n */
84 vmovd %edx, %xmm1 /* SP 2^n */
86 vmulsd %xmm6, %xmm4, %xmm4 /* DP (P3*z+P1)*z */
88 vfmadd213sd %xmm4, %xmm3, %xmm2 /* DP P(Y) (P2*z+P0)*y */
89 vfmadd213sd %xmm5, %xmm5, %xmm2 /* DP T[j]*(P(y)+1) */
90 vcvtsd2ss %xmm2, %xmm2, %xmm0 /* SP T[j]*(P(y)+1) */
91 vmulss %xmm1, %xmm0, %xmm0 /* SP result=2^n*(T[j]*(P(y)+1)) */
96 /* Here if 0<=|x|<2^(-28) */
97 vaddss L(SP_ONE)(%rip), %xmm0, %xmm0 /* 1.0 + x */
98 /* Return 1.0 with inexact raised, except for x==0 */
102 L(special_paths_fma):
103 /* Here if 125*log(2)<=|x| */
104 shrl $31, %eax /* Get sign bit of x, and depending on it: */
105 lea L(SP_RANGE)(%rip), %rdx /* load over/underflow bound */
106 cmpl (%rdx,%rax,4), %ecx /* |x|<under/overflow bound ? */
107 jbe L(near_under_or_overflow_fma)
109 /* Here if |x|>under/overflow bound */
110 cmpl $0x7f800000, %ecx /* |x| is finite ? */
111 jae L(arg_inf_or_nan_fma)
113 /* Here if |x|>under/overflow bound, and x is finite */
114 testl %eax, %eax /* sign of x nonzero ? */
115 je L(res_overflow_fma)
117 /* Here if -inf<x<underflow bound (x<0) */
118 vmovss L(SP_SMALL)(%rip), %xmm0/* load small value 2^(-100) */
119 vmulss %xmm0, %xmm0, %xmm0 /* Return underflowed result (zero or subnormal) */
124 /* Here if overflow bound<x<inf (x>0) */
125 vmovss L(SP_LARGE)(%rip), %xmm0/* load large value 2^100 */
126 vmulss %xmm0, %xmm0, %xmm0 /* Return overflowed result (Inf or max normal) */
130 L(arg_inf_or_nan_fma):
131 /* Here if |x| is Inf or NAN */
132 jne L(arg_nan_fma) /* |x| is Inf ? */
134 /* Here if |x| is Inf */
135 lea L(SP_INF_0)(%rip), %rdx /* depending on sign of x: */
136 vmovss (%rdx,%rax,4), %xmm0 /* return zero or Inf */
141 /* Here if |x| is NaN */
142 vaddss %xmm0, %xmm0, %xmm0 /* Return x+x (raise invalid) */
146 L(near_under_or_overflow_fma):
147 /* Here if 125*log(2)<=|x|<under/overflow bound */
148 vmovd %xmm2, %eax /* bits of n*K+j with trash */
149 vsubsd L(DP_RD)(%rip), %xmm2, %xmm2 /* DP t=round(x*K/log(2)) */
150 movl %eax, %edx /* n*K+j with trash */
151 andl $0x3f, %eax /* bits of j */
152 vmulsd L(DP_NLN2K)(%rip),%xmm2, %xmm2/* DP -t*log(2)/K */
153 andl $0xffffffc0, %edx /* bits of n */
154 vaddsd %xmm1, %xmm2, %xmm0 /* DP y=x-t*log(2)/K */
155 vmulsd %xmm0, %xmm0, %xmm2 /* DP z=y*y */
156 addl $0xffc0, %edx /* bits of n + DP exponent bias */
157 vfmadd213sd L(DP_P0)(%rip), %xmm2, %xmm3/* DP P2*z+P0 */
158 shlq $46, %rdx /* DP 2^n */
159 vfmadd213sd L(DP_P1)(%rip), %xmm2, %xmm4/* DP P3*z+P1 */
160 vmovq %rdx, %xmm1 /* DP 2^n */
161 vmulsd %xmm2, %xmm4, %xmm4 /* DP (P3*z+P1)*z */
162 vfmadd213sd %xmm4, %xmm3, %xmm0 /* DP (P2*z+P0)*y */
163 vmovsd (%rsi,%rax,8), %xmm2
164 vfmadd213sd %xmm2, %xmm2, %xmm0 /* DP T[j]*(P(y)+1) */
165 vmulsd %xmm1, %xmm0, %xmm0 /* DP result=2^n*(T[j]*(P(y)+1)) */
166 vcvtsd2ss %xmm0, %xmm0, %xmm0 /* convert result to single precision */
168 END(__ieee754_expf_fma)
170 .section .rodata.cst8,"aM",@progbits,8
172 L(DP_RD): /* double precision 2^52+2^51 */
173 .long 0x00000000, 0x43380000
174 .type L(DP_RD), @object
175 ASM_SIZE_DIRECTIVE(L(DP_RD))
177 #define __ieee754_expf __ieee754_expf_sse2
180 #define strong_alias(ignored1, ignored2)
182 #include <sysdeps/x86_64/fpu/e_expf.S>