sysdeps/x86_64/fpu/multiarch/svml_s_asinf4_core_sse4.S

   1 /* Function asinf vectorized with SSE4.
   2    Copyright (C) 2021 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 /*
  20  * ALGORITHM DESCRIPTION:
  21  *
  22  *      SelMask = (|x| >= 0.5) ? 1 : 0;
  23  *      R = SelMask ? sqrt(0.5 - 0.5*|x|) : |x|
  24  *      asin(x) = (SelMask ? (Pi/2 - 2*Poly(R)) : Poly(R))*(-1)^sign(x)
  25  *
  26  *
  27  */
  28
  29 /* Offsets for data table __svml_sasin_data_internal
  30  */
  31 #define AbsMask                         0
  32 #define OneHalf                         16
  33 #define SmallNorm                       32
  34 #define One                             48
  35 #define Two                             64
  36 #define sqrt_coeff                      80
  37 #define poly_coeff                      112
  38 #define Pi2H                            192
  39
  40 #include <sysdep.h>
  41
  42         .text
  43         .section .text.sse4,"ax",@progbits
  44 ENTRY(_ZGVbN4v_asinf_sse4)
  45         subq      $72, %rsp
  46         cfi_def_cfa_offset(80)
  47         movaps    %xmm0, %xmm2
  48         movups    __svml_sasin_data_internal(%rip), %xmm1
  49         movups    OneHalf+__svml_sasin_data_internal(%rip), %xmm5
  50
  51 /* x = |arg| */
  52         movaps    %xmm1, %xmm0
  53         andps     %xmm2, %xmm0
  54
  55 /* Y = 0.5 - 0.5*x */
  56         movaps    %xmm5, %xmm3
  57         mulps     %xmm0, %xmm3
  58         movaps    %xmm5, %xmm8
  59
  60 /* x^2 */
  61         movaps    %xmm0, %xmm14
  62         movaps    %xmm0, %xmm15
  63         mulps     %xmm0, %xmm14
  64         subps     %xmm3, %xmm8
  65         cmpnltps  %xmm5, %xmm15
  66
  67 /* SQ ~ -2*sqrt(Y) */
  68         rsqrtps   %xmm8, %xmm6
  69         minps     %xmm8, %xmm14
  70         movaps    %xmm8, %xmm9
  71         movaps    %xmm14, %xmm10
  72         cmpltps   SmallNorm+__svml_sasin_data_internal(%rip), %xmm9
  73         mulps     %xmm14, %xmm10
  74         addps     %xmm8, %xmm8
  75         andnps    %xmm6, %xmm9
  76         movaps    %xmm15, %xmm3
  77         movaps    %xmm9, %xmm7
  78         andnps    %xmm0, %xmm3
  79         mulps     %xmm9, %xmm7
  80         andnps    %xmm2, %xmm1
  81         mulps     %xmm8, %xmm9
  82         mulps     %xmm7, %xmm8
  83
  84 /* polynomial */
  85         movups    poly_coeff+__svml_sasin_data_internal(%rip), %xmm11
  86         mulps     %xmm14, %xmm11
  87         subps     Two+__svml_sasin_data_internal(%rip), %xmm8
  88         movups    poly_coeff+32+__svml_sasin_data_internal(%rip), %xmm12
  89         mulps     %xmm14, %xmm12
  90         addps     poly_coeff+16+__svml_sasin_data_internal(%rip), %xmm11
  91         mulps     %xmm10, %xmm11
  92         addps     poly_coeff+48+__svml_sasin_data_internal(%rip), %xmm12
  93         movups    sqrt_coeff+__svml_sasin_data_internal(%rip), %xmm13
  94         addps     %xmm11, %xmm12
  95         mulps     %xmm8, %xmm13
  96         mulps     %xmm9, %xmm8
  97         mulps     %xmm14, %xmm12
  98         addps     sqrt_coeff+16+__svml_sasin_data_internal(%rip), %xmm13
  99         addps     poly_coeff+64+__svml_sasin_data_internal(%rip), %xmm12
 100         mulps     %xmm8, %xmm13
 101         mulps     %xmm12, %xmm14
 102         subps     %xmm9, %xmm13
 103         andps     %xmm15, %xmm13
 104         orps      %xmm13, %xmm3
 105         mulps     %xmm3, %xmm14
 106         movups    One+__svml_sasin_data_internal(%rip), %xmm4
 107         addps     %xmm14, %xmm3
 108         cmpltps   %xmm0, %xmm4
 109         movups    Pi2H+__svml_sasin_data_internal(%rip), %xmm0
 110         andps     %xmm15, %xmm0
 111         movmskps  %xmm4, %edx
 112         addps     %xmm3, %xmm0
 113         pxor      %xmm1, %xmm0
 114         testl     %edx, %edx
 115
 116 /* Go to special inputs processing branch */
 117         jne       L(SPECIAL_VALUES_BRANCH)
 118                                 # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm2
 119
 120 /* Restore registers
 121  * and exit the function
 122  */
 123
 124 L(EXIT):
 125         addq      $72, %rsp
 126         cfi_def_cfa_offset(8)
 127         ret
 128         cfi_def_cfa_offset(80)
 129
 130 /* Branch to process
 131  * special inputs
 132  */
 133
 134 L(SPECIAL_VALUES_BRANCH):
 135         movups    %xmm2, 32(%rsp)
 136         movups    %xmm0, 48(%rsp)
 137                                 # LOE rbx rbp r12 r13 r14 r15 edx
 138
 139         xorl      %eax, %eax
 140         movq      %r12, 16(%rsp)
 141         cfi_offset(12, -64)
 142         movl      %eax, %r12d
 143         movq      %r13, 8(%rsp)
 144         cfi_offset(13, -72)
 145         movl      %edx, %r13d
 146         movq      %r14, (%rsp)
 147         cfi_offset(14, -80)
 148                                 # LOE rbx rbp r15 r12d r13d
 149
 150 /* Range mask
 151  * bits check
 152  */
 153
 154 L(RANGEMASK_CHECK):
 155         btl       %r12d, %r13d
 156
 157 /* Call scalar math function */
 158         jc        L(SCALAR_MATH_CALL)
 159                                 # LOE rbx rbp r15 r12d r13d
 160
 161 /* Special inputs
 162  * processing loop
 163  */
 164
 165 L(SPECIAL_VALUES_LOOP):
 166         incl      %r12d
 167         cmpl      $4, %r12d
 168
 169 /* Check bits in range mask */
 170         jl        L(RANGEMASK_CHECK)
 171                                 # LOE rbx rbp r15 r12d r13d
 172
 173         movq      16(%rsp), %r12
 174         cfi_restore(12)
 175         movq      8(%rsp), %r13
 176         cfi_restore(13)
 177         movq      (%rsp), %r14
 178         cfi_restore(14)
 179         movups    48(%rsp), %xmm0
 180
 181 /* Go to exit */
 182         jmp       L(EXIT)
 183         cfi_offset(12, -64)
 184         cfi_offset(13, -72)
 185         cfi_offset(14, -80)
 186                                 # LOE rbx rbp r12 r13 r14 r15 xmm0
 187
 188 /* Scalar math fucntion call
 189  * to process special input
 190  */
 191
 192 L(SCALAR_MATH_CALL):
 193         movl      %r12d, %r14d
 194         movss     32(%rsp,%r14,4), %xmm0
 195         call      asinf@PLT
 196                                 # LOE rbx rbp r14 r15 r12d r13d xmm0
 197
 198         movss     %xmm0, 48(%rsp,%r14,4)
 199
 200 /* Process special inputs in loop */
 201         jmp       L(SPECIAL_VALUES_LOOP)
 202                                 # LOE rbx rbp r15 r12d r13d
 203 END(_ZGVbN4v_asinf_sse4)
 204
 205         .section .rodata, "a"
 206         .align 16
 207
 208 #ifdef __svml_sasin_data_internal_typedef
 209 typedef unsigned int VUINT32;
 210 typedef struct {
 211         __declspec(align(16)) VUINT32 AbsMask[4][1];
 212         __declspec(align(16)) VUINT32 OneHalf[4][1];
 213         __declspec(align(16)) VUINT32 SmallNorm[4][1];
 214         __declspec(align(16)) VUINT32 One[4][1];
 215         __declspec(align(16)) VUINT32 Two[4][1];
 216         __declspec(align(16)) VUINT32 sqrt_coeff[2][4][1];
 217         __declspec(align(16)) VUINT32 poly_coeff[5][4][1];
 218         __declspec(align(16)) VUINT32 Pi2H[4][1];
 219 } __svml_sasin_data_internal;
 220 #endif
 221 __svml_sasin_data_internal:
 222         /*== AbsMask ==*/
 223         .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff
 224         /*== OneHalf ==*/
 225         .align 16
 226         .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000
 227         /*== SmallNorm ==*/
 228         .align 16
 229         .long 0x2f800000, 0x2f800000, 0x2f800000, 0x2f800000
 230         /*== One ==*/
 231         .align 16
 232         .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000
 233         /*== Two ==*/
 234         .align 16
 235         .long 0x40000000, 0x40000000, 0x40000000, 0x40000000
 236         /*== sqrt_coeff[2] ==*/
 237         .align 16
 238         .long 0xbdC00004, 0xbdC00004, 0xbdC00004, 0xbdC00004 /* sqrt_coeff2 */
 239         .long 0x3e800001, 0x3e800001, 0x3e800001, 0x3e800001 /* sqrt_coeff1 */
 240         /*== poly_coeff[5] ==*/
 241         .align 16
 242         .long 0x3d2EDC07, 0x3d2EDC07, 0x3d2EDC07, 0x3d2EDC07 /* poly_coeff5 */
 243         .long 0x3CC32A6B, 0x3CC32A6B, 0x3CC32A6B, 0x3CC32A6B /* poly_coeff4 */
 244         .long 0x3d3A9AB4, 0x3d3A9AB4, 0x3d3A9AB4, 0x3d3A9AB4 /* poly_coeff3 */
 245         .long 0x3d997C12, 0x3d997C12, 0x3d997C12, 0x3d997C12 /* poly_coeff2 */
 246         .long 0x3e2AAAFF, 0x3e2AAAFF, 0x3e2AAAFF, 0x3e2AAAFF /* poly_coeff1 */
 247         /*== Pi2H ==*/
 248         .align 16
 249         .long 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB
 250         .align 16
 251         .type   __svml_sasin_data_internal,@object
 252         .size   __svml_sasin_data_internal,.-__svml_sasin_data_internal