1 /* Function cos vectorized with AVX2.
2 Copyright (C) 2014-2019 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/>. */
20 #include "svml_d_trig_data.h"
23 ENTRY (_ZGVdN4v_cos_avx2)
25 /* ALGORITHM DESCRIPTION:
27 ( low accuracy ( < 4ulp ) or enhanced performance
28 ( half of correct mantissa ) implementation )
30 Argument representation:
31 arg + Pi/2 = (N*Pi + R)
34 cos(arg) = sin(arg+Pi/2) = sin(N*Pi + R) = (-1)^N * sin(R)
35 sin(R) is approximated by corresponding polynomial
38 cfi_adjust_cfa_offset (8)
39 cfi_rel_offset (%rbp, 0)
41 cfi_def_cfa_register (%rbp)
44 movq __svml_d_trig_data@GOTPCREL(%rip), %rax
46 vmovupd __dInvPI(%rax), %ymm4
47 vmovupd __dRShifter(%rax), %ymm5
50 ARGUMENT RANGE REDUCTION:
51 Add Pi/2 to argument: X' = X+Pi/2
53 vaddpd __dHalfPI(%rax), %ymm1, %ymm7
55 /* Get absolute argument value: X' = |X'| */
56 vandpd __dAbsMask(%rax), %ymm7, %ymm2
58 /* Y = X'*InvPi + RS : right shifter add */
59 vfmadd213pd %ymm5, %ymm4, %ymm7
60 vmovupd __dC7(%rax), %ymm4
62 /* Check for large arguments path */
63 vcmpnle_uqpd __dRangeVal(%rax), %ymm2, %ymm3
65 /* N = Y - RS : right shifter sub */
66 vsubpd %ymm5, %ymm7, %ymm6
67 vmovupd __dPI1_FMA(%rax), %ymm2
69 /* SignRes = Y<<63 : shift LSB to MSB place for result sign */
70 vpsllq $63, %ymm7, %ymm7
73 vsubpd __dOneHalf(%rax), %ymm6, %ymm0
78 vfnmadd231pd %ymm0, %ymm2, %ymm3
81 vfnmadd231pd __dPI2_FMA(%rax), %ymm0, %ymm3
84 vfnmadd132pd __dPI3_FMA(%rax), %ymm3, %ymm0
86 /* POLYNOMIAL APPROXIMATION: R2 = R*R */
87 vmulpd %ymm0, %ymm0, %ymm5
88 vfmadd213pd __dC6(%rax), %ymm5, %ymm4
89 vfmadd213pd __dC5(%rax), %ymm5, %ymm4
90 vfmadd213pd __dC4(%rax), %ymm5, %ymm4
92 /* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))) */
93 vfmadd213pd __dC3(%rax), %ymm5, %ymm4
95 /* Poly = R+R*(R2*(C1+R2*(C2+R2*Poly))) */
96 vfmadd213pd __dC2(%rax), %ymm5, %ymm4
97 vfmadd213pd __dC1(%rax), %ymm5, %ymm4
98 vmulpd %ymm5, %ymm4, %ymm6
99 vfmadd213pd %ymm0, %ymm0, %ymm6
103 Final sign setting: Res = Poly^SignRes */
104 vxorpd %ymm7, %ymm6, %ymm0
111 cfi_def_cfa_register (%rsp)
113 cfi_adjust_cfa_offset (-8)
119 vmovupd %ymm1, 320(%rsp)
120 vmovupd %ymm0, 384(%rsp)
125 vmovups %ymm8, 224(%rsp)
126 vmovups %ymm9, 192(%rsp)
127 vmovups %ymm10, 160(%rsp)
128 vmovups %ymm11, 128(%rsp)
129 vmovups %ymm12, 96(%rsp)
130 vmovups %ymm13, 64(%rsp)
131 vmovups %ymm14, 32(%rsp)
132 vmovups %ymm15, (%rsp)
136 cfi_offset_rel_rsp (12, 296)
139 cfi_offset_rel_rsp (13, 288)
142 cfi_offset_rel_rsp (14, 280)
145 cfi_offset_rel_rsp (15, 272)
163 vmovups 224(%rsp), %ymm8
164 vmovups 192(%rsp), %ymm9
165 vmovups 160(%rsp), %ymm10
166 vmovups 128(%rsp), %ymm11
167 vmovups 96(%rsp), %ymm12
168 vmovups 64(%rsp), %ymm13
169 vmovups 32(%rsp), %ymm14
170 vmovups (%rsp), %ymm15
171 vmovupd 384(%rsp), %ymm0
188 vmovsd 328(%rsp,%r15), %xmm0
193 vmovsd %xmm0, 392(%rsp,%r15)
199 vmovsd 320(%rsp,%r15), %xmm0
204 vmovsd %xmm0, 384(%rsp,%r15)
207 END (_ZGVdN4v_cos_avx2)