1 | /* Function log10f vectorized with AVX2. |
2 | Copyright (C) 2021-2024 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 | /* |
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 | .section .text.avx2, "ax" , @progbits |
45 | ENTRY(_ZGVdN8v_log10f_avx2) |
46 | pushq %rbp |
47 | cfi_def_cfa_offset(16) |
48 | movq %rsp, %rbp |
49 | cfi_def_cfa(6, 16) |
50 | cfi_offset(6, -16) |
51 | andq $-32, %rsp |
52 | subq $96, %rsp |
53 | |
54 | /* reduction: compute r, n */ |
55 | vmovups iBrkValue+__svml_slog10_data_internal(%rip), %ymm4 |
56 | vmovups sPoly+__svml_slog10_data_internal(%rip), %ymm15 |
57 | vmovups sPoly+64+__svml_slog10_data_internal(%rip), %ymm9 |
58 | vmovups sPoly+128+__svml_slog10_data_internal(%rip), %ymm10 |
59 | vmovups sPoly+192+__svml_slog10_data_internal(%rip), %ymm12 |
60 | vpsubd %ymm4, %ymm0, %ymm1 |
61 | vcmplt_oqps MinNorm+__svml_slog10_data_internal(%rip), %ymm0, %ymm5 |
62 | vcmpnle_uqps MaxNorm+__svml_slog10_data_internal(%rip), %ymm0, %ymm6 |
63 | vpand iOffExpoMask+__svml_slog10_data_internal(%rip), %ymm1, %ymm3 |
64 | vpsrad $23, %ymm1, %ymm2 |
65 | vpaddd %ymm4, %ymm3, %ymm8 |
66 | vcvtdq2ps %ymm2, %ymm1 |
67 | vsubps One+__svml_slog10_data_internal(%rip), %ymm8, %ymm13 |
68 | vmulps L2L+__svml_slog10_data_internal(%rip), %ymm1, %ymm14 |
69 | vfmadd213ps sPoly+32+__svml_slog10_data_internal(%rip), %ymm13, %ymm15 |
70 | vfmadd213ps sPoly+96+__svml_slog10_data_internal(%rip), %ymm13, %ymm9 |
71 | vmulps %ymm13, %ymm13, %ymm11 |
72 | vfmadd213ps sPoly+160+__svml_slog10_data_internal(%rip), %ymm13, %ymm10 |
73 | vfmadd213ps sPoly+224+__svml_slog10_data_internal(%rip), %ymm13, %ymm12 |
74 | vfmadd213ps %ymm9, %ymm11, %ymm15 |
75 | vfmadd213ps %ymm10, %ymm11, %ymm15 |
76 | vfmadd213ps %ymm12, %ymm11, %ymm15 |
77 | vfmadd213ps sPoly+256+__svml_slog10_data_internal(%rip), %ymm13, %ymm15 |
78 | vfmadd213ps %ymm14, %ymm13, %ymm15 |
79 | vorps %ymm6, %ymm5, %ymm7 |
80 | |
81 | /* combine and get argument value range mask */ |
82 | vmovmskps %ymm7, %edx |
83 | vfmadd132ps L2H+__svml_slog10_data_internal(%rip), %ymm15, %ymm1 |
84 | testl %edx, %edx |
85 | |
86 | /* Go to special inputs processing branch */ |
87 | jne L(SPECIAL_VALUES_BRANCH) |
88 | # LOE rbx r12 r13 r14 r15 edx ymm0 ymm1 |
89 | |
90 | /* Restore registers |
91 | * and exit the function |
92 | */ |
93 | |
94 | L(EXIT): |
95 | vmovaps %ymm1, %ymm0 |
96 | movq %rbp, %rsp |
97 | popq %rbp |
98 | cfi_def_cfa(7, 8) |
99 | cfi_restore(6) |
100 | ret |
101 | cfi_def_cfa(6, 16) |
102 | cfi_offset(6, -16) |
103 | |
104 | /* Branch to process |
105 | * special inputs |
106 | */ |
107 | |
108 | L(SPECIAL_VALUES_BRANCH): |
109 | vmovups %ymm0, 32(%rsp) |
110 | vmovups %ymm1, 64(%rsp) |
111 | # LOE rbx r12 r13 r14 r15 edx ymm1 |
112 | |
113 | xorl %eax, %eax |
114 | # LOE rbx r12 r13 r14 r15 eax edx |
115 | |
116 | vzeroupper |
117 | movq %r12, 16(%rsp) |
118 | /* 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) */ |
119 | .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22 |
120 | movl %eax, %r12d |
121 | movq %r13, 8(%rsp) |
122 | /* 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) */ |
123 | .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22 |
124 | movl %edx, %r13d |
125 | movq %r14, (%rsp) |
126 | /* 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) */ |
127 | .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22 |
128 | # LOE rbx r15 r12d r13d |
129 | |
130 | /* Range mask |
131 | * bits check |
132 | */ |
133 | |
134 | L(RANGEMASK_CHECK): |
135 | btl %r12d, %r13d |
136 | |
137 | /* Call scalar math function */ |
138 | jc L(SCALAR_MATH_CALL) |
139 | # LOE rbx r15 r12d r13d |
140 | |
141 | /* Special inputs |
142 | * processing loop |
143 | */ |
144 | |
145 | L(SPECIAL_VALUES_LOOP): |
146 | incl %r12d |
147 | cmpl $8, %r12d |
148 | |
149 | /* Check bits in range mask */ |
150 | jl L(RANGEMASK_CHECK) |
151 | # LOE rbx r15 r12d r13d |
152 | |
153 | movq 16(%rsp), %r12 |
154 | cfi_restore(12) |
155 | movq 8(%rsp), %r13 |
156 | cfi_restore(13) |
157 | movq (%rsp), %r14 |
158 | cfi_restore(14) |
159 | vmovups 64(%rsp), %ymm1 |
160 | |
161 | /* Go to exit */ |
162 | jmp L(EXIT) |
163 | /* 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) */ |
164 | .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22 |
165 | /* 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) */ |
166 | .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22 |
167 | /* 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) */ |
168 | .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22 |
169 | # LOE rbx r12 r13 r14 r15 ymm1 |
170 | |
171 | /* Scalar math function call |
172 | * to process special input |
173 | */ |
174 | |
175 | L(SCALAR_MATH_CALL): |
176 | movl %r12d, %r14d |
177 | vmovss 32(%rsp, %r14, 4), %xmm0 |
178 | call log10f@PLT |
179 | # LOE rbx r14 r15 r12d r13d xmm0 |
180 | |
181 | vmovss %xmm0, 64(%rsp, %r14, 4) |
182 | |
183 | /* Process special inputs in loop */ |
184 | jmp L(SPECIAL_VALUES_LOOP) |
185 | # LOE rbx r15 r12d r13d |
186 | END(_ZGVdN8v_log10f_avx2) |
187 | |
188 | .section .rodata, "a" |
189 | .align 32 |
190 | |
191 | #ifdef __svml_slog10_data_internal_typedef |
192 | typedef unsigned int VUINT32; |
193 | typedef struct { |
194 | __declspec(align(32)) VUINT32 MinNorm[8][1]; |
195 | __declspec(align(32)) VUINT32 MaxNorm[8][1]; |
196 | __declspec(align(32)) VUINT32 L2H[8][1]; |
197 | __declspec(align(32)) VUINT32 L2L[8][1]; |
198 | __declspec(align(32)) VUINT32 iBrkValue[8][1]; |
199 | __declspec(align(32)) VUINT32 iOffExpoMask[8][1]; |
200 | __declspec(align(32)) VUINT32 One[8][1]; |
201 | __declspec(align(32)) VUINT32 sPoly[9][8][1]; |
202 | __declspec(align(32)) VUINT32 L2[8][1]; |
203 | } __svml_slog10_data_internal; |
204 | #endif |
205 | __svml_slog10_data_internal: |
206 | /* MinNorm */ |
207 | .long 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000, 0x00800000 |
208 | /* MaxNorm */ |
209 | .align 32 |
210 | .long 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff, 0x7f7fffff |
211 | /* L2H */ |
212 | .align 32 |
213 | .long 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100, 0x3e9a2100 |
214 | /* L2L */ |
215 | .align 32 |
216 | .long 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600, 0xb64AF600 |
217 | /* iBrkValue = SP 2/3 */ |
218 | .align 32 |
219 | .long 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab, 0x3f2aaaab |
220 | /* iOffExpoMask = SP significand mask */ |
221 | .align 32 |
222 | .long 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff |
223 | /* sOne = SP 1.0 */ |
224 | .align 32 |
225 | .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 |
226 | /* spoly[9] */ |
227 | .align 32 |
228 | .long 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4, 0x3d8063B4 /* coeff9 */ |
229 | .long 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073, 0xbd890073 /* coeff8 */ |
230 | .long 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317, 0x3d775317 /* coeff7 */ |
231 | .long 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27, 0xbd91FB27 /* coeff6 */ |
232 | .long 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96, 0x3dB20B96 /* coeff5 */ |
233 | .long 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20, 0xbdDE6E20 /* coeff4 */ |
234 | .long 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5, 0x3e143CE5 /* coeff3 */ |
235 | .long 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5, 0xbe5E5BC5 /* coeff2 */ |
236 | .long 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9, 0x3eDE5BD9 /* coeff1 */ |
237 | /* L2 */ |
238 | .align 32 |
239 | .long 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b, 0x3e9a209b |
240 | .align 32 |
241 | .type __svml_slog10_data_internal, @object |
242 | .size __svml_slog10_data_internal, .-__svml_slog10_data_internal |
243 | |