sysdeps/x86_64/fpu/multiarch/svml_s_log10f8_core_avx2.S

   1 /* Function log10f vectorized with AVX2.
   2    Copyright (C) 2021-2022 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  *    Get short reciprocal approximation Rcp ~ 1/mantissa(x)
  23  *    R = Rcp*x - 1.0
  24  *    log10(x) = k*log10(2.0) - log10(Rcp) + poly_approximation(R)
  25  *       log10(Rcp) is tabulated
  26  *
  27  *
  28  */
  29
  30 /* Offsets for data table __svml_slog10_data_internal
  31  */
  32 #define MinNorm                         0
  33 #define MaxNorm                         32
  34 #define L2H                             64
  35 #define L2L                             96
  36 #define iBrkValue                       128
  37 #define iOffExpoMask                    160
  38 #define One                             192
  39 #define sPoly                           224
  40 #define L2                              512
  41
  42 #include <sysdep.h>
  43
  44         .text
  45         .section .text.avx2,"ax",@progbits
  46 ENTRY(_ZGVdN8v_log10f_avx2)
  47         pushq     %rbp
  48         cfi_def_cfa_offset(16)
  49         movq      %rsp, %rbp
  50         cfi_def_cfa(6, 16)
  51         cfi_offset(6, -16)
  52         andq      $-32, %rsp
  53         subq      $96, %rsp
  54
  55 /* reduction: compute r,n */
  56         vmovups   iBrkValue+__svml_slog10_data_internal(%rip), %ymm4
  57         vmovups   sPoly+__svml_slog10_data_internal(%rip), %ymm15
  58         vmovups   sPoly+64+__svml_slog10_data_internal(%rip), %ymm9
  59         vmovups   sPoly+128+__svml_slog10_data_internal(%rip), %ymm10
  60         vmovups   sPoly+192+__svml_slog10_data_internal(%rip), %ymm12
  61         vpsubd    %ymm4, %ymm0, %ymm1
  62         vcmplt_oqps MinNorm+__svml_slog10_data_internal(%rip), %ymm0, %ymm5
  63         vcmpnle_uqps MaxNorm+__svml_slog10_data_internal(%rip), %ymm0, %ymm6
  64         vpand     iOffExpoMask+__svml_slog10_data_internal(%rip), %ymm1, %ymm3
  65         vpsrad    $23, %ymm1, %ymm2
  66         vpaddd    %ymm4, %ymm3, %ymm8
  67         vcvtdq2ps %ymm2, %ymm1
  68         vsubps    One+__svml_slog10_data_internal(%rip), %ymm8, %ymm13
  69         vmulps    L2L+__svml_slog10_data_internal(%rip), %ymm1, %ymm14
  70         vfmadd213ps sPoly+32+__svml_slog10_data_internal(%rip), %ymm13, %ymm15
  71         vfmadd213ps sPoly+96+__svml_slog10_data_internal(%rip), %ymm13, %ymm9
  72         vmulps    %ymm13, %ymm13, %ymm11
  73         vfmadd213ps sPoly+160+__svml_slog10_data_internal(%rip), %ymm13, %ymm10
  74         vfmadd213ps sPoly+224+__svml_slog10_data_internal(%rip), %ymm13, %ymm12
  75         vfmadd213ps %ymm9, %ymm11, %ymm15
  76         vfmadd213ps %ymm10, %ymm11, %ymm15
  77         vfmadd213ps %ymm12, %ymm11, %ymm15
  78         vfmadd213ps sPoly+256+__svml_slog10_data_internal(%rip), %ymm13, %ymm15
  79         vfmadd213ps %ymm14, %ymm13, %ymm15
  80         vorps     %ymm6, %ymm5, %ymm7
  81
  82 /* combine and get argument value range mask */
  83         vmovmskps %ymm7, %edx
  84         vfmadd132ps L2H+__svml_slog10_data_internal(%rip), %ymm15, %ymm1
  85         testl     %edx, %edx
  86
  87 /* Go to special inputs processing branch */
  88         jne       L(SPECIAL_VALUES_BRANCH)
  89                                 # LOE rbx r12 r13 r14 r15 edx ymm0 ymm1
  90
  91 /* Restore registers
  92  * and exit the function
  93  */
  94
  95 L(EXIT):
  96         vmovaps   %ymm1, %ymm0
  97         movq      %rbp, %rsp
  98         popq      %rbp
  99         cfi_def_cfa(7, 8)
 100         cfi_restore(6)
 101         ret
 102         cfi_def_cfa(6, 16)
 103         cfi_offset(6, -16)
 104
 105 /* Branch to process
 106  * special inputs
 107  */
 108
 109 L(SPECIAL_VALUES_BRANCH):
 110         vmovups   %ymm0, 32(%rsp)
 111         vmovups   %ymm1, 64(%rsp)
 112                                 # LOE rbx r12 r13 r14 r15 edx ymm1
 113
 114         xorl      %eax, %eax
 115                                 # LOE rbx r12 r13 r14 r15 eax edx
 116
 117         vzeroupper
 118         movq      %r12, 16(%rsp)
 119         /*  DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus)  */
 120         .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
 121         movl      %eax, %r12d
 122         movq      %r13, 8(%rsp)
 123         /*  DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus)  */
 124         .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
 125         movl      %edx, %r13d
 126         movq      %r14, (%rsp)
 127         /*  DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus)  */
 128         .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
 129                                 # LOE rbx r15 r12d r13d
 130
 131 /* Range mask
 132  * bits check
 133  */
 134
 135 L(RANGEMASK_CHECK):
 136         btl       %r12d, %r13d
 137
 138 /* Call scalar math function */
 139         jc        L(SCALAR_MATH_CALL)
 140                                 # LOE rbx r15 r12d r13d
 141
 142 /* Special inputs
 143  * processing loop
 144  */
 145
 146 L(SPECIAL_VALUES_LOOP):
 147         incl      %r12d
 148         cmpl      $8, %r12d
 149
 150 /* Check bits in range mask */
 151         jl        L(RANGEMASK_CHECK)
 152                                 # LOE rbx r15 r12d r13d
 153
 154         movq      16(%rsp), %r12
 155         cfi_restore(12)
 156         movq      8(%rsp), %r13
 157         cfi_restore(13)
 158         movq      (%rsp), %r14
 159         cfi_restore(14)
 160         vmovups   64(%rsp), %ymm1
 161
 162 /* Go to exit */
 163         jmp       L(EXIT)
 164         /*  DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus)  */
 165         .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
 166         /*  DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus)  */
 167         .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
 168         /*  DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus)  */
 169         .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
 170                                 # LOE rbx r12 r13 r14 r15 ymm1
 171
 172 /* Scalar math fucntion call
 173  * to process special input
 174  */
 175
 176 L(SCALAR_MATH_CALL):
 177         movl      %r12d, %r14d
 178         movss     32(%rsp,%r14,4), %xmm0
 179         call      log10f@PLT
 180                                 # LOE rbx r14 r15 r12d r13d xmm0
 181
 182         movss     %xmm0, 64(%rsp,%r14,4)
 183
 184 /* Process special inputs in loop */
 185         jmp       L(SPECIAL_VALUES_LOOP)
 186                                 # LOE rbx r15 r12d r13d
 187 END(_ZGVdN8v_log10f_avx2)
 188
 189         .section .rodata, "a"
 190         .align 32
 191
 192 #ifdef __svml_slog10_data_internal_typedef
 193 typedef unsigned int VUINT32;
 194 typedef struct {
 195         __declspec(align(32)) VUINT32 MinNorm[8][1];
 196         __declspec(align(32)) VUINT32 MaxNorm[8][1];
 197         __declspec(align(32)) VUINT32 L2H[8][1];
 198         __declspec(align(32)) VUINT32 L2L[8][1];
 199         __declspec(align(32)) VUINT32 iBrkValue[8][1];
 200         __declspec(align(32)) VUINT32 iOffExpoMask[8][1];
 201         __declspec(align(32)) VUINT32 One[8][1];
 202         __declspec(align(32)) VUINT32 sPoly[9][8][1];
 203         __declspec(align(32)) VUINT32 L2[8][1];
 204 } __svml_slog10_data_internal;
 205 #endif
 206 __svml_slog10_data_internal:
 207         /*== MinNorm ==*/
 208         .long 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000
 209         /*== MaxNorm ==*/
 210         .align 32
 211         .long 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff
 212         /*== L2H ==*/
 213         .align 32
 214         .long 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100
 215         /*== L2L ==*/
 216         .align 32
 217         .long 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600
 218         /*== iBrkValue = SP 2/3 ==*/
 219         .align 32
 220         .long 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab
 221         /*== iOffExpoMask = SP significand mask ==*/
 222         .align 32
 223         .long 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff
 224         /*== sOne = SP 1.0 ==*/
 225         .align 32
 226         .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000
 227         /*== spoly[9] ==*/
 228         .align 32
 229         .long 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4 /* coeff9 */
 230         .long 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073 /* coeff8 */
 231         .long 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317 /* coeff7 */
 232         .long 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27 /* coeff6 */
 233         .long 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96 /* coeff5 */
 234         .long 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20 /* coeff4 */
 235         .long 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5 /* coeff3 */
 236         .long 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5 /* coeff2 */
 237         .long 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9 /* coeff1 */
 238         /*== L2 ==*/
 239         .align 32
 240         .long 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b
 241         .align 32
 242         .type   __svml_slog10_data_internal,@object
 243         .size   __svml_slog10_data_internal,.-__svml_slog10_data_internal