1 | /* Function cos vectorized with AVX2. |
2 | Copyright (C) 2014-2024 Free Software Foundation, Inc. |
3 | This file is part of the GNU C Library. |
4 | |
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. |
9 | |
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. |
14 | |
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 | <https://www.gnu.org/licenses/>. */ |
18 | |
19 | #include <sysdep.h> |
20 | #include "svml_d_trig_data.h" |
21 | |
22 | .section .text.avx2, "ax" , @progbits |
23 | ENTRY (_ZGVdN4v_cos_avx2) |
24 | |
25 | /* ALGORITHM DESCRIPTION: |
26 | |
27 | ( low accuracy ( < 4ulp ) or enhanced performance |
28 | ( half of correct mantissa ) implementation ) |
29 | |
30 | Argument representation: |
31 | arg + Pi/2 = (N*Pi + R) |
32 | |
33 | Result calculation: |
34 | cos(arg) = sin(arg+Pi/2) = sin(N*Pi + R) = (-1)^N * sin(R) |
35 | sin(R) is approximated by corresponding polynomial |
36 | */ |
37 | pushq %rbp |
38 | cfi_adjust_cfa_offset (8) |
39 | cfi_rel_offset (%rbp, 0) |
40 | movq %rsp, %rbp |
41 | cfi_def_cfa_register (%rbp) |
42 | andq $-64, %rsp |
43 | subq $448, %rsp |
44 | movq __svml_d_trig_data@GOTPCREL(%rip), %rax |
45 | vmovapd %ymm0, %ymm1 |
46 | vmovupd __dInvPI(%rax), %ymm4 |
47 | vmovupd __dRShifter(%rax), %ymm5 |
48 | |
49 | /* |
50 | ARGUMENT RANGE REDUCTION: |
51 | Add Pi/2 to argument: X' = X+Pi/2 |
52 | */ |
53 | vaddpd __dHalfPI(%rax), %ymm1, %ymm7 |
54 | |
55 | /* Get absolute argument value: X' = |X'| */ |
56 | vandpd __dAbsMask(%rax), %ymm7, %ymm2 |
57 | |
58 | /* Y = X'*InvPi + RS : right shifter add */ |
59 | vfmadd213pd %ymm5, %ymm4, %ymm7 |
60 | vmovupd __dC7(%rax), %ymm4 |
61 | |
62 | /* Check for large arguments path */ |
63 | vcmpnle_uqpd __dRangeVal(%rax), %ymm2, %ymm3 |
64 | |
65 | /* N = Y - RS : right shifter sub */ |
66 | vsubpd %ymm5, %ymm7, %ymm6 |
67 | vmovupd __dPI1_FMA(%rax), %ymm2 |
68 | |
69 | /* SignRes = Y<<63 : shift LSB to MSB place for result sign */ |
70 | vpsllq $63, %ymm7, %ymm7 |
71 | |
72 | /* N = N - 0.5 */ |
73 | vsubpd __dOneHalf(%rax), %ymm6, %ymm0 |
74 | vmovmskpd %ymm3, %ecx |
75 | |
76 | /* R = X - N*Pi1 */ |
77 | vmovapd %ymm1, %ymm3 |
78 | vfnmadd231pd %ymm0, %ymm2, %ymm3 |
79 | |
80 | /* R = R - N*Pi2 */ |
81 | vfnmadd231pd __dPI2_FMA(%rax), %ymm0, %ymm3 |
82 | |
83 | /* R = R - N*Pi3 */ |
84 | vfnmadd132pd __dPI3_FMA(%rax), %ymm3, %ymm0 |
85 | |
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 |
91 | |
92 | /* Poly = C3+R2*(C4+R2*(C5+R2*(C6+R2*C7))) */ |
93 | vfmadd213pd __dC3(%rax), %ymm5, %ymm4 |
94 | |
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 |
100 | |
101 | /* |
102 | RECONSTRUCTION: |
103 | Final sign setting: Res = Poly^SignRes */ |
104 | vxorpd %ymm7, %ymm6, %ymm0 |
105 | testl %ecx, %ecx |
106 | jne .LBL_1_3 |
107 | |
108 | .LBL_1_2: |
109 | cfi_remember_state |
110 | movq %rbp, %rsp |
111 | cfi_def_cfa_register (%rsp) |
112 | popq %rbp |
113 | cfi_adjust_cfa_offset (-8) |
114 | cfi_restore (%rbp) |
115 | ret |
116 | |
117 | .LBL_1_3: |
118 | cfi_restore_state |
119 | vmovupd %ymm1, 320(%rsp) |
120 | vmovupd %ymm0, 384(%rsp) |
121 | je .LBL_1_2 |
122 | |
123 | xorb %dl, %dl |
124 | xorl %eax, %eax |
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) |
133 | movq %rsi, 264(%rsp) |
134 | movq %rdi, 256(%rsp) |
135 | movq %r12, 296(%rsp) |
136 | cfi_offset_rel_rsp (12, 296) |
137 | movb %dl, %r12b |
138 | movq %r13, 288(%rsp) |
139 | cfi_offset_rel_rsp (13, 288) |
140 | movl %ecx, %r13d |
141 | movq %r14, 280(%rsp) |
142 | cfi_offset_rel_rsp (14, 280) |
143 | movl %eax, %r14d |
144 | movq %r15, 272(%rsp) |
145 | cfi_offset_rel_rsp (15, 272) |
146 | cfi_remember_state |
147 | |
148 | .LBL_1_6: |
149 | btl %r14d, %r13d |
150 | jc .LBL_1_12 |
151 | |
152 | .LBL_1_7: |
153 | lea 1(%r14), %esi |
154 | btl %esi, %r13d |
155 | jc .LBL_1_10 |
156 | |
157 | .LBL_1_8: |
158 | incb %r12b |
159 | addl $2, %r14d |
160 | cmpb $16, %r12b |
161 | jb .LBL_1_6 |
162 | |
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 |
172 | movq 264(%rsp), %rsi |
173 | movq 256(%rsp), %rdi |
174 | movq 296(%rsp), %r12 |
175 | cfi_restore (%r12) |
176 | movq 288(%rsp), %r13 |
177 | cfi_restore (%r13) |
178 | movq 280(%rsp), %r14 |
179 | cfi_restore (%r14) |
180 | movq 272(%rsp), %r15 |
181 | cfi_restore (%r15) |
182 | jmp .LBL_1_2 |
183 | |
184 | .LBL_1_10: |
185 | cfi_restore_state |
186 | movzbl %r12b, %r15d |
187 | shlq $4, %r15 |
188 | vmovsd 328(%rsp,%r15), %xmm0 |
189 | vzeroupper |
190 | |
191 | call JUMPTARGET(cos) |
192 | |
193 | vmovsd %xmm0, 392(%rsp,%r15) |
194 | jmp .LBL_1_8 |
195 | |
196 | .LBL_1_12: |
197 | movzbl %r12b, %r15d |
198 | shlq $4, %r15 |
199 | vmovsd 320(%rsp,%r15), %xmm0 |
200 | vzeroupper |
201 | |
202 | call JUMPTARGET(cos) |
203 | |
204 | vmovsd %xmm0, 384(%rsp,%r15) |
205 | jmp .LBL_1_7 |
206 | |
207 | END (_ZGVdN4v_cos_avx2) |
208 | |