1 | /* Function sincos 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 (_ZGVdN4vl8l8_sincos_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 = N*Pi + R |
32 | |
33 | Result calculation: |
34 | sin(arg) = sin(N*Pi + R) = (-1)^N * sin(R) |
35 | arg + Pi/2 = (N'*Pi + R') |
36 | cos(arg) = sin(arg+Pi/2) = sin(N'*Pi + R') = (-1)^N' * sin(R') |
37 | sin(R), sin(R') are approximated by corresponding polynomial. */ |
38 | |
39 | pushq %rbp |
40 | cfi_adjust_cfa_offset (8) |
41 | cfi_rel_offset (%rbp, 0) |
42 | movq %rsp, %rbp |
43 | cfi_def_cfa_register (%rbp) |
44 | andq $-64, %rsp |
45 | subq $448, %rsp |
46 | movq __svml_d_trig_data@GOTPCREL(%rip), %rax |
47 | vmovups %ymm14, 288(%rsp) |
48 | vmovups %ymm8, 352(%rsp) |
49 | vmovupd __dSignMask(%rax), %ymm6 |
50 | vmovupd __dInvPI(%rax), %ymm2 |
51 | vmovupd __dPI1_FMA(%rax), %ymm5 |
52 | vmovups %ymm9, 224(%rsp) |
53 | |
54 | /* ARGUMENT RANGE REDUCTION: |
55 | Absolute argument: X' = |X| */ |
56 | vandnpd %ymm0, %ymm6, %ymm1 |
57 | |
58 | /* SinY = X'*InvPi + RS : right shifter add */ |
59 | vfmadd213pd __dRShifter(%rax), %ymm1, %ymm2 |
60 | |
61 | /* SinSignRes = Y<<63 : shift LSB to MSB place for result sign */ |
62 | vpsllq $63, %ymm2, %ymm4 |
63 | |
64 | /* SinN = Y - RS : right shifter sub */ |
65 | vsubpd __dRShifter(%rax), %ymm2, %ymm2 |
66 | |
67 | /* SinR = X' - SinN*Pi1 */ |
68 | vmovdqa %ymm1, %ymm14 |
69 | vfnmadd231pd %ymm2, %ymm5, %ymm14 |
70 | |
71 | /* SinR = SinR - SinN*Pi1 */ |
72 | vfnmadd231pd __dPI2_FMA(%rax), %ymm2, %ymm14 |
73 | |
74 | /* Sine result sign: SinRSign = SignMask & SinR */ |
75 | vandpd %ymm14, %ymm6, %ymm7 |
76 | |
77 | /* Set SinRSign to 0.5 */ |
78 | vorpd __dOneHalf(%rax), %ymm7, %ymm3 |
79 | |
80 | /* CosN = SinN +(-)0.5 */ |
81 | vaddpd %ymm3, %ymm2, %ymm3 |
82 | |
83 | /* CosR = SinX - CosN*Pi1 */ |
84 | vmovdqa %ymm1, %ymm8 |
85 | vfnmadd231pd %ymm3, %ymm5, %ymm8 |
86 | vmovupd __dPI3_FMA(%rax), %ymm5 |
87 | vcmpnle_uqpd __dRangeVal(%rax), %ymm1, %ymm1 |
88 | |
89 | /* CosR = CosR - CosN*Pi2 */ |
90 | vfnmadd231pd __dPI2_FMA(%rax), %ymm3, %ymm8 |
91 | |
92 | /* SinR = SinR - SinN*Pi3 */ |
93 | vfnmadd213pd %ymm14, %ymm5, %ymm2 |
94 | |
95 | /* CosR = CosR - CosN*Pi3 */ |
96 | vfnmadd213pd %ymm8, %ymm5, %ymm3 |
97 | vmovupd __dC6(%rax), %ymm8 |
98 | |
99 | /* SinR2 = SinR^2 */ |
100 | vmulpd %ymm2, %ymm2, %ymm14 |
101 | |
102 | /* CosR2 = CosR^2 */ |
103 | vmulpd %ymm3, %ymm3, %ymm5 |
104 | |
105 | /* Grab SignX */ |
106 | vandpd %ymm0, %ymm6, %ymm9 |
107 | |
108 | /* Update CosRSign and CosSignRes signs */ |
109 | vxorpd %ymm6, %ymm7, %ymm6 |
110 | vxorpd %ymm6, %ymm4, %ymm7 |
111 | |
112 | /* Update sign SinSignRes */ |
113 | vxorpd %ymm9, %ymm4, %ymm6 |
114 | |
115 | /* Polynomial approximation */ |
116 | vmovupd __dC7(%rax), %ymm4 |
117 | vmovdqa %ymm8, %ymm9 |
118 | vfmadd231pd __dC7(%rax), %ymm14, %ymm9 |
119 | vfmadd213pd %ymm8, %ymm5, %ymm4 |
120 | vfmadd213pd __dC5(%rax), %ymm14, %ymm9 |
121 | vfmadd213pd __dC5(%rax), %ymm5, %ymm4 |
122 | vfmadd213pd __dC4(%rax), %ymm14, %ymm9 |
123 | vfmadd213pd __dC4(%rax), %ymm5, %ymm4 |
124 | |
125 | /* SinPoly = C3 + SinR2*(C4 + SinR2*(C5 + SinR2*(C6 + SinR2*C7))) */ |
126 | vfmadd213pd __dC3(%rax), %ymm14, %ymm9 |
127 | |
128 | /* CosPoly = C3 + CosR2*(C4 + CosR2*(C5 + CosR2*(C6 + CosR2*C7))) */ |
129 | vfmadd213pd __dC3(%rax), %ymm5, %ymm4 |
130 | |
131 | /* SinPoly = C2 + SinR2*SinPoly */ |
132 | vfmadd213pd __dC2(%rax), %ymm14, %ymm9 |
133 | |
134 | /* CosPoly = C2 + CosR2*CosPoly */ |
135 | vfmadd213pd __dC2(%rax), %ymm5, %ymm4 |
136 | |
137 | /* SinPoly = C1 + SinR2*SinPoly */ |
138 | vfmadd213pd __dC1(%rax), %ymm14, %ymm9 |
139 | |
140 | /* CosPoly = C1 + CosR2*CosPoly */ |
141 | vfmadd213pd __dC1(%rax), %ymm5, %ymm4 |
142 | |
143 | /* SinPoly = SinR2*SinPoly */ |
144 | vmulpd %ymm14, %ymm9, %ymm8 |
145 | |
146 | /* CosPoly = CosR2*CosPoly */ |
147 | vmulpd %ymm5, %ymm4, %ymm4 |
148 | |
149 | /* SinPoly = SinR*SinPoly */ |
150 | vfmadd213pd %ymm2, %ymm2, %ymm8 |
151 | |
152 | /* CosPoly = CosR*CosPoly */ |
153 | vfmadd213pd %ymm3, %ymm3, %ymm4 |
154 | vmovmskpd %ymm1, %ecx |
155 | |
156 | /* Final reconstruction |
157 | Update Sin result's sign */ |
158 | vxorpd %ymm6, %ymm8, %ymm3 |
159 | |
160 | /* Update Cos result's sign */ |
161 | vxorpd %ymm7, %ymm4, %ymm2 |
162 | testl %ecx, %ecx |
163 | jne .LBL_1_3 |
164 | |
165 | .LBL_1_2: |
166 | cfi_remember_state |
167 | vmovups 352(%rsp), %ymm8 |
168 | vmovups 224(%rsp), %ymm9 |
169 | vmovups 288(%rsp), %ymm14 |
170 | vmovupd %ymm2, (%rsi) |
171 | vmovdqa %ymm3, (%rdi) |
172 | movq %rbp, %rsp |
173 | cfi_def_cfa_register (%rsp) |
174 | popq %rbp |
175 | cfi_adjust_cfa_offset (-8) |
176 | cfi_restore (%rbp) |
177 | ret |
178 | |
179 | .LBL_1_3: |
180 | cfi_restore_state |
181 | vmovupd %ymm0, 256(%rsp) |
182 | vmovupd %ymm3, 320(%rsp) |
183 | vmovupd %ymm2, 384(%rsp) |
184 | je .LBL_1_2 |
185 | |
186 | xorb %dl, %dl |
187 | xorl %eax, %eax |
188 | vmovups %ymm10, 128(%rsp) |
189 | vmovups %ymm11, 96(%rsp) |
190 | vmovups %ymm12, 64(%rsp) |
191 | vmovups %ymm13, 32(%rsp) |
192 | vmovups %ymm15, (%rsp) |
193 | movq %rsi, 160(%rsp) |
194 | movq %r12, 200(%rsp) |
195 | cfi_offset_rel_rsp (12, 200) |
196 | movb %dl, %r12b |
197 | movq %r13, 192(%rsp) |
198 | cfi_offset_rel_rsp (13, 192) |
199 | movl %eax, %r13d |
200 | movq %r14, 184(%rsp) |
201 | cfi_offset_rel_rsp (14, 184) |
202 | movl %ecx, %r14d |
203 | movq %r15, 176(%rsp) |
204 | cfi_offset_rel_rsp (15, 176) |
205 | movq %rbx, 168(%rsp) |
206 | movq %rdi, %rbx |
207 | cfi_remember_state |
208 | |
209 | .LBL_1_6: |
210 | btl %r13d, %r14d |
211 | jc .LBL_1_13 |
212 | |
213 | .LBL_1_7: |
214 | lea 1(%r13), %esi |
215 | btl %esi, %r14d |
216 | jc .LBL_1_10 |
217 | |
218 | .LBL_1_8: |
219 | incb %r12b |
220 | addl $2, %r13d |
221 | cmpb $16, %r12b |
222 | jb .LBL_1_6 |
223 | |
224 | vmovups 128(%rsp), %ymm10 |
225 | movq %rbx, %rdi |
226 | vmovups 96(%rsp), %ymm11 |
227 | vmovups 64(%rsp), %ymm12 |
228 | vmovups 32(%rsp), %ymm13 |
229 | vmovups (%rsp), %ymm15 |
230 | vmovupd 320(%rsp), %ymm3 |
231 | vmovupd 384(%rsp), %ymm2 |
232 | movq 160(%rsp), %rsi |
233 | movq 200(%rsp), %r12 |
234 | cfi_restore (%r12) |
235 | movq 192(%rsp), %r13 |
236 | cfi_restore (%r13) |
237 | movq 184(%rsp), %r14 |
238 | cfi_restore (%r14) |
239 | movq 176(%rsp), %r15 |
240 | cfi_restore (%r15) |
241 | movq 168(%rsp), %rbx |
242 | jmp .LBL_1_2 |
243 | |
244 | .LBL_1_10: |
245 | cfi_restore_state |
246 | movzbl %r12b, %r15d |
247 | shlq $4, %r15 |
248 | vmovsd 264(%rsp,%r15), %xmm0 |
249 | vzeroupper |
250 | |
251 | call JUMPTARGET(sin) |
252 | |
253 | vmovsd %xmm0, 328(%rsp,%r15) |
254 | vmovsd 264(%rsp,%r15), %xmm0 |
255 | |
256 | call JUMPTARGET(cos) |
257 | |
258 | vmovsd %xmm0, 392(%rsp,%r15) |
259 | jmp .LBL_1_8 |
260 | |
261 | .LBL_1_13: |
262 | movzbl %r12b, %r15d |
263 | shlq $4, %r15 |
264 | vmovsd 256(%rsp,%r15), %xmm0 |
265 | vzeroupper |
266 | |
267 | call JUMPTARGET(sin) |
268 | |
269 | vmovsd %xmm0, 320(%rsp,%r15) |
270 | vmovsd 256(%rsp,%r15), %xmm0 |
271 | |
272 | call JUMPTARGET(cos) |
273 | |
274 | vmovsd %xmm0, 384(%rsp,%r15) |
275 | jmp .LBL_1_7 |
276 | |
277 | END (_ZGVdN4vl8l8_sincos_avx2) |
278 | libmvec_hidden_def(_ZGVdN4vl8l8_sincos_avx2) |
279 | |
280 | /* vvv version implemented with wrapper to vl8l8 variant. */ |
281 | ENTRY (_ZGVdN4vvv_sincos_avx2) |
282 | #ifndef __ILP32__ |
283 | pushq %rbp |
284 | cfi_adjust_cfa_offset (8) |
285 | cfi_rel_offset (%rbp, 0) |
286 | movq %rsp, %rbp |
287 | cfi_def_cfa_register (%rbp) |
288 | andq $-32, %rsp |
289 | subq $128, %rsp |
290 | vmovdqu %ymm1, 64(%rsp) |
291 | lea (%rsp), %rdi |
292 | vmovdqu %ymm2, 96(%rdi) |
293 | lea 32(%rsp), %rsi |
294 | call HIDDEN_JUMPTARGET(_ZGVdN4vl8l8_sincos_avx2) |
295 | movq 64(%rsp), %rdx |
296 | movq 96(%rsp), %rsi |
297 | movq 72(%rsp), %r8 |
298 | movq 104(%rsp), %r10 |
299 | movq (%rsp), %rax |
300 | movq 32(%rsp), %rcx |
301 | movq 8(%rsp), %rdi |
302 | movq 40(%rsp), %r9 |
303 | movq %rax, (%rdx) |
304 | movq %rcx, (%rsi) |
305 | movq 80(%rsp), %rax |
306 | movq 112(%rsp), %rcx |
307 | movq %rdi, (%r8) |
308 | movq %r9, (%r10) |
309 | movq 88(%rsp), %rdi |
310 | movq 120(%rsp), %r9 |
311 | movq 16(%rsp), %r11 |
312 | movq 48(%rsp), %rdx |
313 | movq 24(%rsp), %rsi |
314 | movq 56(%rsp), %r8 |
315 | movq %r11, (%rax) |
316 | movq %rdx, (%rcx) |
317 | movq %rsi, (%rdi) |
318 | movq %r8, (%r9) |
319 | movq %rbp, %rsp |
320 | cfi_def_cfa_register (%rsp) |
321 | popq %rbp |
322 | cfi_adjust_cfa_offset (-8) |
323 | cfi_restore (%rbp) |
324 | ret |
325 | #else |
326 | leal 8(%rsp), %r10d |
327 | .cfi_def_cfa 10, 0 |
328 | andl $-32, %esp |
329 | pushq -8(%r10d) |
330 | pushq %rbp |
331 | .cfi_escape 0x10,0x6,0x2,0x76,0 |
332 | movl %esp, %ebp |
333 | pushq %r10 |
334 | .cfi_escape 0xf,0x3,0x76,0x78,0x6 |
335 | leal -48(%rbp), %esi |
336 | leal -80(%rbp), %edi |
337 | subl $104, %esp |
338 | vmovaps %xmm1, -96(%ebp) |
339 | vmovaps %xmm2, -112(%ebp) |
340 | call HIDDEN_JUMPTARGET(_ZGVdN4vl8l8_sincos_avx2) |
341 | movl -96(%ebp), %eax |
342 | vmovsd -80(%ebp), %xmm0 |
343 | vmovsd %xmm0, (%eax) |
344 | movl -92(%ebp), %eax |
345 | vmovsd -72(%ebp), %xmm0 |
346 | vmovsd %xmm0, (%eax) |
347 | movl -88(%ebp), %eax |
348 | vmovsd -64(%ebp), %xmm0 |
349 | vmovsd %xmm0, (%eax) |
350 | movl -84(%ebp), %eax |
351 | vmovsd -56(%ebp), %xmm0 |
352 | vmovsd %xmm0, (%eax) |
353 | movl -112(%ebp), %eax |
354 | vmovsd -48(%ebp), %xmm0 |
355 | vmovsd %xmm0, (%eax) |
356 | movl -108(%ebp), %eax |
357 | vmovsd -40(%ebp), %xmm0 |
358 | vmovsd %xmm0, (%eax) |
359 | movl -104(%ebp), %eax |
360 | vmovsd -32(%ebp), %xmm0 |
361 | vmovsd %xmm0, (%eax) |
362 | movl -100(%ebp), %eax |
363 | vmovsd -24(%ebp), %xmm0 |
364 | vmovsd %xmm0, (%eax) |
365 | addl $104, %esp |
366 | popq %r10 |
367 | .cfi_def_cfa 10, 0 |
368 | popq %rbp |
369 | leal -8(%r10), %esp |
370 | .cfi_def_cfa 7, 8 |
371 | ret |
372 | #endif |
373 | END (_ZGVdN4vvv_sincos_avx2) |
374 | |