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