1 | /* Function logf vectorized with SSE4. |
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_s_logf_data.h" |
21 | |
22 | .section .text.sse4, "ax" , @progbits |
23 | ENTRY (_ZGVbN4v_logf_sse4) |
24 | /* |
25 | ALGORITHM DESCRIPTION: |
26 | |
27 | log(x) = exponent_x*log(2) + log(mantissa_x), if mantissa_x<4/3 |
28 | log(x) = (exponent_x+1)*log(2) + log(0.5*mantissa_x), if mantissa_x>4/3 |
29 | |
30 | R = mantissa_x - 1, if mantissa_x<4/3 |
31 | R = 0.5*mantissa_x - 1, if mantissa_x>4/3 |
32 | |R|< 1/3 |
33 | |
34 | log(1+R) is approximated as a polynomial: degree 9 for 1-ulp, |
35 | degree 7 for 4-ulp, degree 3 for half-precision. */ |
36 | |
37 | pushq %rbp |
38 | cfi_adjust_cfa_offset (8) |
39 | cfi_rel_offset (%rbp, 0) |
40 | movq %rsp, %rbp |
41 | cfi_def_cfa_register (%rbp) |
42 | andq $-64, %rsp |
43 | subq $320, %rsp |
44 | |
45 | /* reduction: compute r,n */ |
46 | movaps %xmm0, %xmm2 |
47 | |
48 | /* check for working range, |
49 | set special argument mask (denormals/zero/Inf/NaN) */ |
50 | movq __svml_slog_data@GOTPCREL(%rip), %rax |
51 | movdqu _iHiDelta(%rax), %xmm1 |
52 | movdqu _iLoRange(%rax), %xmm4 |
53 | paddd %xmm0, %xmm1 |
54 | movdqu _iBrkValue(%rax), %xmm3 |
55 | pcmpgtd %xmm1, %xmm4 |
56 | movdqu _iOffExpoMask(%rax), %xmm1 |
57 | psubd %xmm3, %xmm2 |
58 | pand %xmm2, %xmm1 |
59 | |
60 | /* exponent_x (mantissa_x<4/3) or exponent_x+1 (mantissa_x>4/3) */ |
61 | psrad $23, %xmm2 |
62 | paddd %xmm3, %xmm1 |
63 | movups _sPoly_7(%rax), %xmm5 |
64 | |
65 | /* mantissa_x (mantissa_x<4/3), or 0.5*mantissa_x (mantissa_x>4/3) */ |
66 | cvtdq2ps %xmm2, %xmm6 |
67 | |
68 | /* reduced argument R */ |
69 | subps _sOne(%rax), %xmm1 |
70 | movmskps %xmm4, %ecx |
71 | |
72 | /* final reconstruction: |
73 | add exponent_value*log2 to polynomial result */ |
74 | mulps _sLn2(%rax), %xmm6 |
75 | |
76 | /* polynomial evaluation starts here */ |
77 | mulps %xmm1, %xmm5 |
78 | addps _sPoly_6(%rax), %xmm5 |
79 | mulps %xmm1, %xmm5 |
80 | addps _sPoly_5(%rax), %xmm5 |
81 | mulps %xmm1, %xmm5 |
82 | addps _sPoly_4(%rax), %xmm5 |
83 | mulps %xmm1, %xmm5 |
84 | addps _sPoly_3(%rax), %xmm5 |
85 | mulps %xmm1, %xmm5 |
86 | addps _sPoly_2(%rax), %xmm5 |
87 | mulps %xmm1, %xmm5 |
88 | addps _sPoly_1(%rax), %xmm5 |
89 | mulps %xmm1, %xmm5 |
90 | |
91 | /* polynomial evaluation end */ |
92 | mulps %xmm1, %xmm5 |
93 | addps %xmm5, %xmm1 |
94 | addps %xmm6, %xmm1 |
95 | testl %ecx, %ecx |
96 | jne .LBL_1_3 |
97 | |
98 | .LBL_1_2: |
99 | cfi_remember_state |
100 | movdqa %xmm1, %xmm0 |
101 | movq %rbp, %rsp |
102 | cfi_def_cfa_register (%rsp) |
103 | popq %rbp |
104 | cfi_adjust_cfa_offset (-8) |
105 | cfi_restore (%rbp) |
106 | ret |
107 | |
108 | .LBL_1_3: |
109 | cfi_restore_state |
110 | movups %xmm0, 192(%rsp) |
111 | movups %xmm1, 256(%rsp) |
112 | je .LBL_1_2 |
113 | |
114 | xorb %dl, %dl |
115 | xorl %eax, %eax |
116 | movups %xmm8, 112(%rsp) |
117 | movups %xmm9, 96(%rsp) |
118 | movups %xmm10, 80(%rsp) |
119 | movups %xmm11, 64(%rsp) |
120 | movups %xmm12, 48(%rsp) |
121 | movups %xmm13, 32(%rsp) |
122 | movups %xmm14, 16(%rsp) |
123 | movups %xmm15, (%rsp) |
124 | movq %rsi, 136(%rsp) |
125 | movq %rdi, 128(%rsp) |
126 | movq %r12, 168(%rsp) |
127 | cfi_offset_rel_rsp (12, 168) |
128 | movb %dl, %r12b |
129 | movq %r13, 160(%rsp) |
130 | cfi_offset_rel_rsp (13, 160) |
131 | movl %ecx, %r13d |
132 | movq %r14, 152(%rsp) |
133 | cfi_offset_rel_rsp (14, 152) |
134 | movl %eax, %r14d |
135 | movq %r15, 144(%rsp) |
136 | cfi_offset_rel_rsp (15, 144) |
137 | cfi_remember_state |
138 | |
139 | .LBL_1_6: |
140 | btl %r14d, %r13d |
141 | jc .LBL_1_12 |
142 | |
143 | .LBL_1_7: |
144 | lea 1(%r14), %esi |
145 | btl %esi, %r13d |
146 | jc .LBL_1_10 |
147 | |
148 | .LBL_1_8: |
149 | incb %r12b |
150 | addl $2, %r14d |
151 | cmpb $16, %r12b |
152 | jb .LBL_1_6 |
153 | |
154 | movups 112(%rsp), %xmm8 |
155 | movups 96(%rsp), %xmm9 |
156 | movups 80(%rsp), %xmm10 |
157 | movups 64(%rsp), %xmm11 |
158 | movups 48(%rsp), %xmm12 |
159 | movups 32(%rsp), %xmm13 |
160 | movups 16(%rsp), %xmm14 |
161 | movups (%rsp), %xmm15 |
162 | movq 136(%rsp), %rsi |
163 | movq 128(%rsp), %rdi |
164 | movq 168(%rsp), %r12 |
165 | cfi_restore (%r12) |
166 | movq 160(%rsp), %r13 |
167 | cfi_restore (%r13) |
168 | movq 152(%rsp), %r14 |
169 | cfi_restore (%r14) |
170 | movq 144(%rsp), %r15 |
171 | cfi_restore (%r15) |
172 | movups 256(%rsp), %xmm1 |
173 | jmp .LBL_1_2 |
174 | |
175 | .LBL_1_10: |
176 | cfi_restore_state |
177 | movzbl %r12b, %r15d |
178 | movss 196(%rsp,%r15,8), %xmm0 |
179 | |
180 | call JUMPTARGET(logf) |
181 | |
182 | movss %xmm0, 260(%rsp,%r15,8) |
183 | jmp .LBL_1_8 |
184 | |
185 | .LBL_1_12: |
186 | movzbl %r12b, %r15d |
187 | movss 192(%rsp,%r15,8), %xmm0 |
188 | |
189 | call JUMPTARGET(logf) |
190 | |
191 | movss %xmm0, 256(%rsp,%r15,8) |
192 | jmp .LBL_1_7 |
193 | |
194 | END (_ZGVbN4v_logf_sse4) |
195 | |