1 /* Function atan2 vectorized with SSE4.
2 Copyright (C) 2021-2023 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 https://www.gnu.org/licenses/. */
20 * ALGORITHM DESCRIPTION:
21 * For 0.0 <= x <= 7.0/16.0: atan(x) = atan(0.0) + atan(s), where s=(x-0.0)/(1.0+0.0*x)
22 * For 7.0/16.0 <= x <= 11.0/16.0: atan(x) = atan(0.5) + atan(s), where s=(x-0.5)/(1.0+0.5*x)
23 * For 11.0/16.0 <= x <= 19.0/16.0: atan(x) = atan(1.0) + atan(s), where s=(x-1.0)/(1.0+1.0*x)
24 * For 19.0/16.0 <= x <= 39.0/16.0: atan(x) = atan(1.5) + atan(s), where s=(x-1.5)/(1.0+1.5*x)
25 * For 39.0/16.0 <= x <= inf : atan(x) = atan(inf) + atan(s), where s=-1.0/x
26 * Where atan(s) ~= s+s^3*Poly11(s^2) on interval |s|<7.0/0.16.
31 /* Offsets for data table __svml_datan2_data_internal
55 #define dSIGN_MASK 352
56 #define iCHK_WORK_SUB 368
57 #define iCHK_WORK_CMP 384
63 .section .text.sse4, "ax", @progbits
64 ENTRY(_ZGVbN2vv_atan2_sse4)
66 cfi_def_cfa_offset(96)
70 * #define NO_VECTOR_ZERO_ATAN2_ARGS
74 * The end of declarations
77 * Cannot be replaced by VQRCP(D, dR0, dB);
78 * Argument Absolute values
80 movups dABS_MASK+__svml_datan2_data_internal(%rip), %xmm1
89 movups dSIGN_MASK+__svml_datan2_data_internal(%rip), %xmm5
95 * 1) If y<x then a= y, b=x, PIO2=0
96 * 2) If y>x then a=-x, b=y, PIO2=Pi/2
102 movups dPIO2+__svml_datan2_data_internal(%rip), %xmm6
111 movq iCHK_WORK_SUB+__svml_datan2_data_internal(%rip), %xmm2
114 /* Check if y and x are on main path. */
115 pshufd $221, %xmm9, %xmm3
117 pshufd $221, %xmm1, %xmm13
121 movq iCHK_WORK_CMP+__svml_datan2_data_internal(%rip), %xmm12
122 movdqa %xmm13, %xmm14
123 pcmpgtd %xmm12, %xmm4
124 pcmpeqd %xmm12, %xmm3
125 pcmpgtd %xmm12, %xmm14
126 pcmpeqd %xmm12, %xmm13
133 /* P = A19*R2 + A18 */
134 movups dA19+__svml_datan2_data_internal(%rip), %xmm15
137 addpd dA18+__svml_datan2_data_internal(%rip), %xmm15
141 addpd dA17+__svml_datan2_data_internal(%rip), %xmm15
145 addpd dA16+__svml_datan2_data_internal(%rip), %xmm15
149 addpd dA15+__svml_datan2_data_internal(%rip), %xmm15
153 addpd dA14+__svml_datan2_data_internal(%rip), %xmm15
157 addpd dA13+__svml_datan2_data_internal(%rip), %xmm15
161 addpd dA12+__svml_datan2_data_internal(%rip), %xmm15
165 addpd dA11+__svml_datan2_data_internal(%rip), %xmm15
169 addpd dA10+__svml_datan2_data_internal(%rip), %xmm15
173 addpd dA09+__svml_datan2_data_internal(%rip), %xmm15
177 addpd dA08+__svml_datan2_data_internal(%rip), %xmm15
181 addpd dA07+__svml_datan2_data_internal(%rip), %xmm15
185 addpd dA06+__svml_datan2_data_internal(%rip), %xmm15
189 addpd dA05+__svml_datan2_data_internal(%rip), %xmm15
193 addpd dA04+__svml_datan2_data_internal(%rip), %xmm15
197 addpd dA03+__svml_datan2_data_internal(%rip), %xmm15
201 addpd dA02+__svml_datan2_data_internal(%rip), %xmm15
205 addpd dA01+__svml_datan2_data_internal(%rip), %xmm15
212 * dP=(R+R*dP) + dPIO2
217 /* if x<0, dPI = Pi, else dPI =0 */
218 movups dZERO+__svml_datan2_data_internal(%rip), %xmm3
222 andps __svml_datan2_data_internal(%rip), %xmm2
229 /* Special branch for fast (vector) processing of zero arguments */
232 /* Go to auxilary branch */
234 # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm1 xmm2 xmm3 xmm4 xmm5 xmm6 xmm7 xmm8 xmm9 xmm10 xmm11
236 /* Return from auxilary branch
237 * for out of main path inputs
240 L(AUX_BRANCH_RETURN):
242 * Special branch for fast (vector) processing of zero arguments
243 * The end of implementation
247 /* Go to special inputs processing branch */
248 jne L(SPECIAL_VALUES_BRANCH)
249 # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm10 xmm11
252 * and exit the function
257 cfi_def_cfa_offset(8)
259 cfi_def_cfa_offset(96)
265 L(SPECIAL_VALUES_BRANCH):
266 movups %xmm10, 32(%rsp)
267 movups %xmm11, 48(%rsp)
268 movups %xmm0, 64(%rsp)
269 # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0
279 # LOE rbx rbp r15 r12d r13d
288 /* Call scalar math function */
289 jc L(SCALAR_MATH_CALL)
290 # LOE rbx rbp r15 r12d r13d
296 L(SPECIAL_VALUES_LOOP):
300 /* Check bits in range mask */
301 jl L(RANGEMASK_CHECK)
302 # LOE rbx rbp r15 r12d r13d
310 movups 64(%rsp), %xmm0
317 # LOE rbx rbp r12 r13 r14 r15 xmm0
319 /* Scalar math fucntion call
320 * to process special input
325 movsd 32(%rsp, %r14, 8), %xmm0
326 movsd 48(%rsp, %r14, 8), %xmm1
328 # LOE rbx rbp r14 r15 r12d r13d xmm0
330 movsd %xmm0, 64(%rsp, %r14, 8)
332 /* Process special inputs in loop */
333 jmp L(SPECIAL_VALUES_LOOP)
337 # LOE rbx rbp r15 r12d r13d
340 * for out of main path inputs
344 /* Check if both X & Y are not NaNs: iXYnotNAN */
345 movaps %xmm11, %xmm13
346 movaps %xmm10, %xmm12
347 cmpordpd %xmm11, %xmm13
348 cmpordpd %xmm10, %xmm12
350 /* Check if at least on of Y or Y is zero: iAXAYZERO */
355 * Path for zero arguments (at least one of both)
356 * Check if both args are zeros (den. is zero)
361 pshufd $221, %xmm9, %xmm1
362 pshufd $221, %xmm13, %xmm9
364 /* Check if at least on of Y or Y is zero and not NaN: iAXAYZEROnotNAN */
367 /* Exclude from previous callout mask zero (and not NaN) arguments */
371 /* Set sPIO2 to zero if den. is zero */
376 /* Res = sign(Y)*(X<0)?(PIO2+PI):PIO2 */
377 pshufd $221, %xmm3, %xmm3
379 pshufd $221, %xmm11, %xmm5
382 pshufd $80, %xmm3, %xmm6
387 movmskps %xmm14, %edx
389 /* Merge results from main and spec path */
390 pshufd $80, %xmm1, %xmm2
399 /* Return to main vector processing path */
400 jmp L(AUX_BRANCH_RETURN)
401 # LOE rbx rbp r12 r13 r14 r15 eax edx xmm0 xmm10 xmm11
402 END(_ZGVbN2vv_atan2_sse4)
404 .section .rodata, "a"
407 #ifdef __svml_datan2_data_internal_typedef
408 typedef unsigned int VUINT32;
410 __declspec(align(16)) VUINT32 dPI[2][2];
411 __declspec(align(16)) VUINT32 dPIO2[2][2];
412 __declspec(align(16)) VUINT32 dA19[2][2];
413 __declspec(align(16)) VUINT32 dA18[2][2];
414 __declspec(align(16)) VUINT32 dA17[2][2];
415 __declspec(align(16)) VUINT32 dA16[2][2];
416 __declspec(align(16)) VUINT32 dA15[2][2];
417 __declspec(align(16)) VUINT32 dA14[2][2];
418 __declspec(align(16)) VUINT32 dA13[2][2];
419 __declspec(align(16)) VUINT32 dA12[2][2];
420 __declspec(align(16)) VUINT32 dA11[2][2];
421 __declspec(align(16)) VUINT32 dA10[2][2];
422 __declspec(align(16)) VUINT32 dA09[2][2];
423 __declspec(align(16)) VUINT32 dA08[2][2];
424 __declspec(align(16)) VUINT32 dA07[2][2];
425 __declspec(align(16)) VUINT32 dA06[2][2];
426 __declspec(align(16)) VUINT32 dA05[2][2];
427 __declspec(align(16)) VUINT32 dA04[2][2];
428 __declspec(align(16)) VUINT32 dA03[2][2];
429 __declspec(align(16)) VUINT32 dA02[2][2];
430 __declspec(align(16)) VUINT32 dA01[2][2];
431 __declspec(align(16)) VUINT32 dA00[2][2];
432 __declspec(align(16)) VUINT32 dSIGN_MASK[2][2];
433 __declspec(align(16)) VUINT32 iCHK_WORK_SUB[4][1];
434 __declspec(align(16)) VUINT32 iCHK_WORK_CMP[4][1];
435 __declspec(align(16)) VUINT32 dABS_MASK[2][2];
436 __declspec(align(16)) VUINT32 dZERO[2][2];
437 } __svml_datan2_data_internal;
439 __svml_datan2_data_internal:
440 .quad 0x400921FB54442D18, 0x400921FB54442D18 // dPI
442 .quad 0x3FF921FB54442D18, 0x3FF921FB54442D18 // dPIO2
444 .quad 0xBEF4FDB537ABC7A3, 0xBEF4FDB537ABC7A3 // dA19
446 .quad 0x3F2CED0A36665209, 0x3F2CED0A36665209 // dA18
448 .quad 0xBF52E67C93954C23, 0xBF52E67C93954C23 // dA17
450 .quad 0x3F6F5A1DAE82AFB3, 0x3F6F5A1DAE82AFB3 // dA16
452 .quad 0xBF82B2EC618E4BAD, 0xBF82B2EC618E4BAD // dA15
454 .quad 0x3F914F4C661116A5, 0x3F914F4C661116A5 // dA14
456 .quad 0xBF9A5E83B081F69C, 0xBF9A5E83B081F69C // dA13
458 .quad 0x3FA169980CB6AD4F, 0x3FA169980CB6AD4F // dA12
460 .quad 0xBFA4EFA2E563C1BC, 0xBFA4EFA2E563C1BC // dA11
462 .quad 0x3FA7EC0FBC50683B, 0x3FA7EC0FBC50683B // dA10
464 .quad 0xBFAAD261EAA09954, 0xBFAAD261EAA09954 // dA09
466 .quad 0x3FAE1749BD612DCF, 0x3FAE1749BD612DCF // dA08
468 .quad 0xBFB11084009435E0, 0xBFB11084009435E0 // dA07
470 .quad 0x3FB3B12A49295651, 0x3FB3B12A49295651 // dA06
472 .quad 0xBFB745D009BADA94, 0xBFB745D009BADA94 // dA05
474 .quad 0x3FBC71C707F7D5B5, 0x3FBC71C707F7D5B5 // dA04
476 .quad 0xBFC2492491EE55C7, 0xBFC2492491EE55C7 // dA03
478 .quad 0x3FC999999997EE34, 0x3FC999999997EE34 // dA02
480 .quad 0xBFD55555555553C5, 0xBFD55555555553C5 // dA01
482 .quad 0x3FF0000000000000, 0x3FF0000000000000 // dA00
484 .quad 0x8000000000000000, 0x8000000000000000 // dSIGN_MASK
486 .long 0x80300000, 0x80300000, 0x80300000, 0x80300000 // iCHK_WORK_SUB
488 .long 0xfdd00000, 0xfdd00000, 0xfdd00000, 0xfdd00000 // iCHK_WORK_CMP
490 .quad 0x7fffffffffffffff, 0x7fffffffffffffff // dABS_MASK
492 .quad 0x0000000000000000, 0x0000000000000000 // dZERO
494 .type __svml_datan2_data_internal, @object
495 .size __svml_datan2_data_internal, .-__svml_datan2_data_internal