sysdeps/x86_64/fpu/multiarch/svml_s_sincosf16_core_avx512.S

   1 /* Function sincosf vectorized with AVX-512. KNL and SKX versions.
   2    Copyright (C) 2014-2016 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    <http://www.gnu.org/licenses/>.  */
  18
  19 #include <sysdep.h>
  20 #include "svml_s_trig_data.h"
  21 #include "svml_s_wrapper_impl.h"
  22
  23 /*
  24    ALGORITHM DESCRIPTION:
  25
  26      1) Range reduction to [-Pi/4; +Pi/4] interval
  27         a) Grab sign from source argument and save it.
  28         b) Remove sign using AND operation
  29         c) Getting octant Y by 2/Pi multiplication
  30         d) Add "Right Shifter" value
  31         e) Treat obtained value as integer S for destination sign setting.
  32            SS = ((S-S&1)&2)<<30; For sin part
  33            SC = ((S+S&1)&2)<<30; For cos part
  34         f) Change destination sign if source sign is negative
  35            using XOR operation.
  36         g) Subtract "Right Shifter" (0x4B000000) value
  37         h) Subtract Y*(PI/2) from X argument, where PI/2 divided to 4 parts:
  38            X = X - Y*PI1 - Y*PI2 - Y*PI3 - Y*PI4;
  39      2) Polynomial (minimax for sin within  [-Pi/4; +Pi/4] interval)
  40         a) Calculate X^2 = X * X
  41         b) Calculate 2 polynomials for sin and cos:
  42            RS = X * ( A0 + X^2 * (A1 + x^2 * (A2 + x^2 * (A3))));
  43            RC = B0 + X^2 * (B1 + x^2 * (B2 + x^2 * (B3 + x^2 * (B4))));
  44         c) Swap RS & RC if if first bit of obtained value after
  45            Right Shifting is set to 1. Using And, Andnot & Or operations.
  46      3) Destination sign setting
  47         a) Set shifted destination sign using XOR operation:
  48            R1 = XOR( RS, SS );
  49            R2 = XOR( RC, SC ).  */
  50
  51         .text
  52 ENTRY (_ZGVeN16vvv_sincosf_knl)
  53 #ifndef HAVE_AVX512_ASM_SUPPORT
  54 WRAPPER_IMPL_AVX512_fFF _ZGVdN8vvv_sincosf
  55 #else
  56         pushq     %rbp
  57         cfi_adjust_cfa_offset (8)
  58         cfi_rel_offset (%rbp, 0)
  59         movq      %rsp, %rbp
  60         cfi_def_cfa_register (%rbp)
  61         andq      $-64, %rsp
  62         subq      $1344, %rsp
  63         movq      __svml_s_trig_data@GOTPCREL(%rip), %rax
  64         vmovaps   %zmm0, %zmm2
  65         movl      $-1, %edx
  66         vmovups __sAbsMask(%rax), %zmm0
  67         vmovups __sInvPI(%rax), %zmm3
  68
  69 /* Absolute argument computation */
  70         vpandd    %zmm0, %zmm2, %zmm1
  71         vmovups __sPI1_FMA(%rax), %zmm5
  72         vmovups __sSignMask(%rax), %zmm9
  73         vpandnd   %zmm2, %zmm0, %zmm0
  74
  75 /* h) Subtract Y*(PI/2) from X argument, where PI/2 divided to 3 parts:
  76       X = X - Y*PI1 - Y*PI2 - Y*PI3 */
  77         vmovaps   %zmm1, %zmm6
  78         vmovaps   %zmm1, %zmm8
  79
  80 /* c) Getting octant Y by 2/Pi multiplication
  81    d) Add "Right Shifter" value */
  82         vfmadd213ps __sRShifter(%rax), %zmm1, %zmm3
  83         vmovups __sPI3_FMA(%rax), %zmm7
  84
  85 /* g) Subtract "Right Shifter" (0x4B000000) value */
  86         vsubps __sRShifter(%rax), %zmm3, %zmm12
  87
  88 /* e) Treat obtained value as integer S for destination sign setting */
  89         vpslld    $31, %zmm3, %zmm13
  90         vmovups __sA7_FMA(%rax), %zmm14
  91         vfnmadd231ps %zmm12, %zmm5, %zmm6
  92
  93 /* 2) Polynomial (minimax for sin within  [-Pi/4; +Pi/4] interval)
  94       a) Calculate X^2 = X * X
  95       b) Calculate 2 polynomials for sin and cos:
  96          RS = X * ( A0 + X^2 * (A1 + x^2 * (A2 + x^2 * (A3))));
  97          RC = B0 + X^2 * (B1 + x^2 * (B2 + x^2 * (B3 + x^2 * (B4)))) */
  98         vmovaps   %zmm14, %zmm15
  99         vmovups __sA9_FMA(%rax), %zmm3
 100         vcmpps    $22, __sRangeReductionVal(%rax), %zmm1, %k1
 101         vpbroadcastd %edx, %zmm1{%k1}{z}
 102         vfnmadd231ps __sPI2_FMA(%rax), %zmm12, %zmm6
 103         vptestmd  %zmm1, %zmm1, %k0
 104         vpandd    %zmm6, %zmm9, %zmm11
 105         kmovw     %k0, %ecx
 106         vpxord __sOneHalf(%rax), %zmm11, %zmm4
 107
 108 /* Result sign calculations */
 109         vpternlogd $150, %zmm13, %zmm9, %zmm11
 110
 111 /* Add correction term 0.5 for cos() part */
 112         vaddps    %zmm4, %zmm12, %zmm10
 113         vfnmadd213ps %zmm6, %zmm7, %zmm12
 114         vfnmadd231ps %zmm10, %zmm5, %zmm8
 115         vpxord    %zmm13, %zmm12, %zmm13
 116         vmulps    %zmm13, %zmm13, %zmm12
 117         vfnmadd231ps __sPI2_FMA(%rax), %zmm10, %zmm8
 118         vfmadd231ps __sA9_FMA(%rax), %zmm12, %zmm15
 119         vfnmadd213ps %zmm8, %zmm7, %zmm10
 120         vfmadd213ps __sA5_FMA(%rax), %zmm12, %zmm15
 121         vpxord    %zmm11, %zmm10, %zmm5
 122         vmulps    %zmm5, %zmm5, %zmm4
 123         vfmadd213ps __sA3(%rax), %zmm12, %zmm15
 124         vfmadd213ps %zmm14, %zmm4, %zmm3
 125         vmulps    %zmm12, %zmm15, %zmm14
 126         vfmadd213ps __sA5_FMA(%rax), %zmm4, %zmm3
 127         vfmadd213ps %zmm13, %zmm13, %zmm14
 128         vfmadd213ps __sA3(%rax), %zmm4, %zmm3
 129         vpxord    %zmm0, %zmm14, %zmm0
 130         vmulps    %zmm4, %zmm3, %zmm3
 131         vfmadd213ps %zmm5, %zmm5, %zmm3
 132         testl     %ecx, %ecx
 133         jne       .LBL_1_3
 134
 135 .LBL_1_2:
 136         cfi_remember_state
 137         vmovups   %zmm0, (%rdi)
 138         vmovups   %zmm3, (%rsi)
 139         movq      %rbp, %rsp
 140         cfi_def_cfa_register (%rsp)
 141         popq      %rbp
 142         cfi_adjust_cfa_offset (-8)
 143         cfi_restore (%rbp)
 144         ret
 145
 146 .LBL_1_3:
 147         cfi_restore_state
 148         vmovups   %zmm2, 1152(%rsp)
 149         vmovups   %zmm0, 1216(%rsp)
 150         vmovups   %zmm3, 1280(%rsp)
 151         je        .LBL_1_2
 152
 153         xorb      %dl, %dl
 154         kmovw     %k4, 1048(%rsp)
 155         xorl      %eax, %eax
 156         kmovw     %k5, 1040(%rsp)
 157         kmovw     %k6, 1032(%rsp)
 158         kmovw     %k7, 1024(%rsp)
 159         vmovups   %zmm16, 960(%rsp)
 160         vmovups   %zmm17, 896(%rsp)
 161         vmovups   %zmm18, 832(%rsp)
 162         vmovups   %zmm19, 768(%rsp)
 163         vmovups   %zmm20, 704(%rsp)
 164         vmovups   %zmm21, 640(%rsp)
 165         vmovups   %zmm22, 576(%rsp)
 166         vmovups   %zmm23, 512(%rsp)
 167         vmovups   %zmm24, 448(%rsp)
 168         vmovups   %zmm25, 384(%rsp)
 169         vmovups   %zmm26, 320(%rsp)
 170         vmovups   %zmm27, 256(%rsp)
 171         vmovups   %zmm28, 192(%rsp)
 172         vmovups   %zmm29, 128(%rsp)
 173         vmovups   %zmm30, 64(%rsp)
 174         vmovups   %zmm31, (%rsp)
 175         movq      %rsi, 1056(%rsp)
 176         movq      %r12, 1096(%rsp)
 177         cfi_offset_rel_rsp (12, 1096)
 178         movb      %dl, %r12b
 179         movq      %r13, 1088(%rsp)
 180         cfi_offset_rel_rsp (13, 1088)
 181         movl      %eax, %r13d
 182         movq      %r14, 1080(%rsp)
 183         cfi_offset_rel_rsp (14, 1080)
 184         movl      %ecx, %r14d
 185         movq      %r15, 1072(%rsp)
 186         cfi_offset_rel_rsp (15, 1072)
 187         movq      %rbx, 1064(%rsp)
 188         movq      %rdi, %rbx
 189         cfi_remember_state
 190
 191 .LBL_1_6:
 192         btl       %r13d, %r14d
 193         jc        .LBL_1_13
 194
 195 .LBL_1_7:
 196         lea       1(%r13), %esi
 197         btl       %esi, %r14d
 198         jc        .LBL_1_10
 199
 200 .LBL_1_8:
 201         addb      $1, %r12b
 202         addl      $2, %r13d
 203         cmpb      $16, %r12b
 204         jb        .LBL_1_6
 205
 206         movq      %rbx, %rdi
 207         kmovw     1048(%rsp), %k4
 208         movq      1056(%rsp), %rsi
 209         kmovw     1040(%rsp), %k5
 210         movq      1096(%rsp), %r12
 211         cfi_restore (%r12)
 212         kmovw     1032(%rsp), %k6
 213         movq      1088(%rsp), %r13
 214         cfi_restore (%r13)
 215         kmovw     1024(%rsp), %k7
 216         vmovups   960(%rsp), %zmm16
 217         vmovups   896(%rsp), %zmm17
 218         vmovups   832(%rsp), %zmm18
 219         vmovups   768(%rsp), %zmm19
 220         vmovups   704(%rsp), %zmm20
 221         vmovups   640(%rsp), %zmm21
 222         vmovups   576(%rsp), %zmm22
 223         vmovups   512(%rsp), %zmm23
 224         vmovups   448(%rsp), %zmm24
 225         vmovups   384(%rsp), %zmm25
 226         vmovups   320(%rsp), %zmm26
 227         vmovups   256(%rsp), %zmm27
 228         vmovups   192(%rsp), %zmm28
 229         vmovups   128(%rsp), %zmm29
 230         vmovups   64(%rsp), %zmm30
 231         vmovups   (%rsp), %zmm31
 232         movq      1080(%rsp), %r14
 233         cfi_restore (%r14)
 234         movq      1072(%rsp), %r15
 235         cfi_restore (%r15)
 236         movq      1064(%rsp), %rbx
 237         vmovups   1216(%rsp), %zmm0
 238         vmovups   1280(%rsp), %zmm3
 239         jmp       .LBL_1_2
 240
 241 .LBL_1_10:
 242         cfi_restore_state
 243         movzbl    %r12b, %r15d
 244         vmovss    1156(%rsp,%r15,8), %xmm0
 245
 246         call      sinf@PLT
 247
 248         vmovss    %xmm0, 1220(%rsp,%r15,8)
 249         vmovss    1156(%rsp,%r15,8), %xmm0
 250
 251         call      cosf@PLT
 252
 253         vmovss    %xmm0, 1284(%rsp,%r15,8)
 254         jmp       .LBL_1_8
 255
 256 .LBL_1_13:
 257         movzbl    %r12b, %r15d
 258         vmovss    1152(%rsp,%r15,8), %xmm0
 259
 260         call      sinf@PLT
 261
 262         vmovss    %xmm0, 1216(%rsp,%r15,8)
 263         vmovss    1152(%rsp,%r15,8), %xmm0
 264
 265         call      cosf@PLT
 266
 267         vmovss    %xmm0, 1280(%rsp,%r15,8)
 268         jmp       .LBL_1_7
 269 #endif
 270 END (_ZGVeN16vvv_sincosf_knl)
 271
 272 ENTRY (_ZGVeN16vvv_sincosf_skx)
 273 #ifndef HAVE_AVX512_ASM_SUPPORT
 274 WRAPPER_IMPL_AVX512_fFF _ZGVdN8vvv_sincosf
 275 #else
 276         pushq     %rbp
 277         cfi_adjust_cfa_offset (8)
 278         cfi_rel_offset (%rbp, 0)
 279         movq      %rsp, %rbp
 280         cfi_def_cfa_register (%rbp)
 281         andq      $-64, %rsp
 282         subq      $1344, %rsp
 283         movq      __svml_s_trig_data@GOTPCREL(%rip), %rax
 284         vmovaps   %zmm0, %zmm4
 285         vmovups __sAbsMask(%rax), %zmm3
 286         vmovups __sInvPI(%rax), %zmm5
 287         vmovups __sRShifter(%rax), %zmm6
 288         vmovups __sPI1_FMA(%rax), %zmm9
 289         vmovups __sPI2_FMA(%rax), %zmm10
 290         vmovups __sSignMask(%rax), %zmm14
 291         vmovups __sOneHalf(%rax), %zmm7
 292         vmovups __sPI3_FMA(%rax), %zmm12
 293
 294 /* Absolute argument computation */
 295         vandps    %zmm3, %zmm4, %zmm2
 296
 297 /* c) Getting octant Y by 2/Pi multiplication
 298    d) Add "Right Shifter" value */
 299         vfmadd213ps %zmm6, %zmm2, %zmm5
 300         vcmpps    $18, __sRangeReductionVal(%rax), %zmm2, %k1
 301
 302 /* e) Treat obtained value as integer S for destination sign setting */
 303         vpslld    $31, %zmm5, %zmm0
 304
 305 /* g) Subtract "Right Shifter" (0x4B000000) value */
 306         vsubps    %zmm6, %zmm5, %zmm5
 307         vmovups __sA3(%rax), %zmm6
 308
 309 /* h) Subtract Y*(PI/2) from X argument, where PI/2 divided to 3 parts:
 310       X = X - Y*PI1 - Y*PI2 - Y*PI3 */
 311         vmovaps   %zmm2, %zmm11
 312         vfnmadd231ps %zmm5, %zmm9, %zmm11
 313         vfnmadd231ps %zmm5, %zmm10, %zmm11
 314         vandps    %zmm11, %zmm14, %zmm1
 315         vxorps    %zmm1, %zmm7, %zmm8
 316
 317 /* Result sign calculations */
 318         vpternlogd $150, %zmm0, %zmm14, %zmm1
 319         vmovups   .L_2il0floatpacket.13(%rip), %zmm14
 320
 321 /* Add correction term 0.5 for cos() part */
 322         vaddps    %zmm8, %zmm5, %zmm15
 323         vfnmadd213ps %zmm11, %zmm12, %zmm5
 324         vandnps   %zmm4, %zmm3, %zmm11
 325         vmovups __sA7_FMA(%rax), %zmm3
 326         vmovaps   %zmm2, %zmm13
 327         vfnmadd231ps %zmm15, %zmm9, %zmm13
 328         vxorps    %zmm0, %zmm5, %zmm9
 329         vmovups __sA5_FMA(%rax), %zmm0
 330         vfnmadd231ps %zmm15, %zmm10, %zmm13
 331         vmulps    %zmm9, %zmm9, %zmm8
 332         vfnmadd213ps %zmm13, %zmm12, %zmm15
 333         vmovups __sA9_FMA(%rax), %zmm12
 334         vxorps    %zmm1, %zmm15, %zmm1
 335         vmulps    %zmm1, %zmm1, %zmm13
 336
 337 /* 2) Polynomial (minimax for sin within  [-Pi/4; +Pi/4] interval)
 338       a) Calculate X^2 = X * X
 339       b) Calculate 2 polynomials for sin and cos:
 340          RS = X * ( A0 + X^2 * (A1 + x^2 * (A2 + x^2 * (A3))));
 341          RC = B0 + X^2 * (B1 + x^2 * (B2 + x^2 * (B3 + x^2 * (B4)))) */
 342         vmovaps   %zmm12, %zmm7
 343         vfmadd213ps %zmm3, %zmm8, %zmm7
 344         vfmadd213ps %zmm3, %zmm13, %zmm12
 345         vfmadd213ps %zmm0, %zmm8, %zmm7
 346         vfmadd213ps %zmm0, %zmm13, %zmm12
 347         vfmadd213ps %zmm6, %zmm8, %zmm7
 348         vfmadd213ps %zmm6, %zmm13, %zmm12
 349         vmulps    %zmm8, %zmm7, %zmm10
 350         vmulps    %zmm13, %zmm12, %zmm3
 351         vfmadd213ps %zmm9, %zmm9, %zmm10
 352         vfmadd213ps %zmm1, %zmm1, %zmm3
 353         vxorps    %zmm11, %zmm10, %zmm0
 354         vpandnd   %zmm2, %zmm2, %zmm14{%k1}
 355         vptestmd  %zmm14, %zmm14, %k0
 356         kmovw     %k0, %ecx
 357         testl     %ecx, %ecx
 358         jne       .LBL_2_3
 359
 360 .LBL_2_2:
 361         cfi_remember_state
 362         vmovups   %zmm0, (%rdi)
 363         vmovups   %zmm3, (%rsi)
 364         movq      %rbp, %rsp
 365         cfi_def_cfa_register (%rsp)
 366         popq      %rbp
 367         cfi_adjust_cfa_offset (-8)
 368         cfi_restore (%rbp)
 369         ret
 370
 371 .LBL_2_3:
 372         cfi_restore_state
 373         vmovups   %zmm4, 1152(%rsp)
 374         vmovups   %zmm0, 1216(%rsp)
 375         vmovups   %zmm3, 1280(%rsp)
 376         je        .LBL_2_2
 377
 378         xorb      %dl, %dl
 379         xorl      %eax, %eax
 380         kmovw     %k4, 1048(%rsp)
 381         kmovw     %k5, 1040(%rsp)
 382         kmovw     %k6, 1032(%rsp)
 383         kmovw     %k7, 1024(%rsp)
 384         vmovups   %zmm16, 960(%rsp)
 385         vmovups   %zmm17, 896(%rsp)
 386         vmovups   %zmm18, 832(%rsp)
 387         vmovups   %zmm19, 768(%rsp)
 388         vmovups   %zmm20, 704(%rsp)
 389         vmovups   %zmm21, 640(%rsp)
 390         vmovups   %zmm22, 576(%rsp)
 391         vmovups   %zmm23, 512(%rsp)
 392         vmovups   %zmm24, 448(%rsp)
 393         vmovups   %zmm25, 384(%rsp)
 394         vmovups   %zmm26, 320(%rsp)
 395         vmovups   %zmm27, 256(%rsp)
 396         vmovups   %zmm28, 192(%rsp)
 397         vmovups   %zmm29, 128(%rsp)
 398         vmovups   %zmm30, 64(%rsp)
 399         vmovups   %zmm31, (%rsp)
 400         movq      %rsi, 1056(%rsp)
 401         movq      %r12, 1096(%rsp)
 402         cfi_offset_rel_rsp (12, 1096)
 403         movb      %dl, %r12b
 404         movq      %r13, 1088(%rsp)
 405         cfi_offset_rel_rsp (13, 1088)
 406         movl      %eax, %r13d
 407         movq      %r14, 1080(%rsp)
 408         cfi_offset_rel_rsp (14, 1080)
 409         movl      %ecx, %r14d
 410         movq      %r15, 1072(%rsp)
 411         cfi_offset_rel_rsp (15, 1072)
 412         movq      %rbx, 1064(%rsp)
 413         movq      %rdi, %rbx
 414         cfi_remember_state
 415
 416 .LBL_2_6:
 417         btl       %r13d, %r14d
 418         jc        .LBL_2_13
 419
 420 .LBL_2_7:
 421         lea       1(%r13), %esi
 422         btl       %esi, %r14d
 423         jc        .LBL_2_10
 424
 425 .LBL_2_8:
 426         incb      %r12b
 427         addl      $2, %r13d
 428         cmpb      $16, %r12b
 429         jb        .LBL_2_6
 430
 431         kmovw     1048(%rsp), %k4
 432         movq      %rbx, %rdi
 433         kmovw     1040(%rsp), %k5
 434         kmovw     1032(%rsp), %k6
 435         kmovw     1024(%rsp), %k7
 436         vmovups   960(%rsp), %zmm16
 437         vmovups   896(%rsp), %zmm17
 438         vmovups   832(%rsp), %zmm18
 439         vmovups   768(%rsp), %zmm19
 440         vmovups   704(%rsp), %zmm20
 441         vmovups   640(%rsp), %zmm21
 442         vmovups   576(%rsp), %zmm22
 443         vmovups   512(%rsp), %zmm23
 444         vmovups   448(%rsp), %zmm24
 445         vmovups   384(%rsp), %zmm25
 446         vmovups   320(%rsp), %zmm26
 447         vmovups   256(%rsp), %zmm27
 448         vmovups   192(%rsp), %zmm28
 449         vmovups   128(%rsp), %zmm29
 450         vmovups   64(%rsp), %zmm30
 451         vmovups   (%rsp), %zmm31
 452         vmovups   1216(%rsp), %zmm0
 453         vmovups   1280(%rsp), %zmm3
 454         movq      1056(%rsp), %rsi
 455         movq      1096(%rsp), %r12
 456         cfi_restore (%r12)
 457         movq      1088(%rsp), %r13
 458         cfi_restore (%r13)
 459         movq      1080(%rsp), %r14
 460         cfi_restore (%r14)
 461         movq      1072(%rsp), %r15
 462         cfi_restore (%r15)
 463         movq      1064(%rsp), %rbx
 464         jmp       .LBL_2_2
 465
 466 .LBL_2_10:
 467         cfi_restore_state
 468         movzbl    %r12b, %r15d
 469         vmovss    1156(%rsp,%r15,8), %xmm0
 470         vzeroupper
 471         vmovss    1156(%rsp,%r15,8), %xmm0
 472
 473         call      sinf@PLT
 474
 475         vmovss    %xmm0, 1220(%rsp,%r15,8)
 476         vmovss    1156(%rsp,%r15,8), %xmm0
 477
 478         call      cosf@PLT
 479
 480         vmovss    %xmm0, 1284(%rsp,%r15,8)
 481         jmp       .LBL_2_8
 482
 483 .LBL_2_13:
 484         movzbl    %r12b, %r15d
 485         vmovss    1152(%rsp,%r15,8), %xmm0
 486         vzeroupper
 487         vmovss    1152(%rsp,%r15,8), %xmm0
 488
 489         call      sinf@PLT
 490
 491         vmovss    %xmm0, 1216(%rsp,%r15,8)
 492         vmovss    1152(%rsp,%r15,8), %xmm0
 493
 494         call      cosf@PLT
 495
 496         vmovss    %xmm0, 1280(%rsp,%r15,8)
 497         jmp       .LBL_2_7
 498 #endif
 499 END (_ZGVeN16vvv_sincosf_skx)
 500
 501         .section .rodata, "a"
 502 .L_2il0floatpacket.13:
 503         .long   0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff
 504         .type   .L_2il0floatpacket.13,@object