1 | /* Function asinf 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 | * 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 32 |
33 | #define SmallNorm 64 |
34 | #define One 96 |
35 | #define Two 128 |
36 | #define sqrt_coeff 160 |
37 | #define poly_coeff 224 |
38 | #define Pi2H 384 |
39 | |
40 | #include <sysdep.h> |
41 | |
42 | .section .text.avx2, "ax" , @progbits |
43 | ENTRY(_ZGVdN8v_asinf_avx2) |
44 | pushq %rbp |
45 | cfi_def_cfa_offset(16) |
46 | movq %rsp, %rbp |
47 | cfi_def_cfa(6, 16) |
48 | cfi_offset(6, -16) |
49 | andq $-32, %rsp |
50 | subq $96, %rsp |
51 | vmovups __svml_sasin_data_internal(%rip), %ymm5 |
52 | vmovups OneHalf+__svml_sasin_data_internal(%rip), %ymm9 |
53 | vmovups One+__svml_sasin_data_internal(%rip), %ymm6 |
54 | vmovaps %ymm0, %ymm4 |
55 | |
56 | /* x = |arg| */ |
57 | vandps %ymm4, %ymm5, %ymm3 |
58 | |
59 | /* Y = 0.5 - 0.5*x */ |
60 | vmovaps %ymm9, %ymm12 |
61 | vfnmadd231ps %ymm3, %ymm9, %ymm12 |
62 | |
63 | /* x^2 */ |
64 | vmulps %ymm3, %ymm3, %ymm7 |
65 | vcmplt_oqps %ymm3, %ymm6, %ymm8 |
66 | |
67 | /* SQ ~ -2*sqrt(Y) */ |
68 | vcmplt_oqps SmallNorm+__svml_sasin_data_internal(%rip), %ymm12, %ymm10 |
69 | vminps %ymm12, %ymm7, %ymm1 |
70 | vaddps %ymm12, %ymm12, %ymm15 |
71 | vcmpnlt_uqps %ymm9, %ymm3, %ymm0 |
72 | vrsqrtps %ymm12, %ymm11 |
73 | vmovups poly_coeff+64+__svml_sasin_data_internal(%rip), %ymm7 |
74 | vmulps %ymm1, %ymm1, %ymm6 |
75 | vmovups sqrt_coeff+__svml_sasin_data_internal(%rip), %ymm9 |
76 | vfmadd213ps poly_coeff+96+__svml_sasin_data_internal(%rip), %ymm1, %ymm7 |
77 | vmovmskps %ymm8, %edx |
78 | |
79 | /* polynomial */ |
80 | vmovups poly_coeff+__svml_sasin_data_internal(%rip), %ymm8 |
81 | vandnps %ymm11, %ymm10, %ymm13 |
82 | vmulps %ymm13, %ymm13, %ymm14 |
83 | vfmadd213ps poly_coeff+32+__svml_sasin_data_internal(%rip), %ymm1, %ymm8 |
84 | vandnps %ymm4, %ymm5, %ymm2 |
85 | vmulps %ymm15, %ymm13, %ymm5 |
86 | vfmsub213ps Two+__svml_sasin_data_internal(%rip), %ymm14, %ymm15 |
87 | vfmadd213ps %ymm7, %ymm6, %ymm8 |
88 | vfmadd213ps sqrt_coeff+32+__svml_sasin_data_internal(%rip), %ymm15, %ymm9 |
89 | vmulps %ymm15, %ymm5, %ymm15 |
90 | vfmadd213ps poly_coeff+128+__svml_sasin_data_internal(%rip), %ymm1, %ymm8 |
91 | vfmsub213ps %ymm5, %ymm15, %ymm9 |
92 | vmulps %ymm8, %ymm1, %ymm1 |
93 | vblendvps %ymm0, %ymm9, %ymm3, %ymm3 |
94 | vfmadd213ps %ymm3, %ymm3, %ymm1 |
95 | vandps Pi2H+__svml_sasin_data_internal(%rip), %ymm0, %ymm0 |
96 | vaddps %ymm1, %ymm0, %ymm10 |
97 | vxorps %ymm2, %ymm10, %ymm0 |
98 | testl %edx, %edx |
99 | |
100 | /* Go to special inputs processing branch */ |
101 | jne L(SPECIAL_VALUES_BRANCH) |
102 | # LOE rbx r12 r13 r14 r15 edx ymm0 ymm4 |
103 | |
104 | /* Restore registers |
105 | * and exit the function |
106 | */ |
107 | |
108 | L(EXIT): |
109 | movq %rbp, %rsp |
110 | popq %rbp |
111 | cfi_def_cfa(7, 8) |
112 | cfi_restore(6) |
113 | ret |
114 | cfi_def_cfa(6, 16) |
115 | cfi_offset(6, -16) |
116 | |
117 | /* Branch to process |
118 | * special inputs |
119 | */ |
120 | |
121 | L(SPECIAL_VALUES_BRANCH): |
122 | vmovups %ymm4, 32(%rsp) |
123 | vmovups %ymm0, 64(%rsp) |
124 | # LOE rbx r12 r13 r14 r15 edx ymm0 |
125 | |
126 | xorl %eax, %eax |
127 | # LOE rbx r12 r13 r14 r15 eax edx |
128 | |
129 | vzeroupper |
130 | movq %r12, 16(%rsp) |
131 | /* 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) */ |
132 | .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22 |
133 | movl %eax, %r12d |
134 | movq %r13, 8(%rsp) |
135 | /* 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) */ |
136 | .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22 |
137 | movl %edx, %r13d |
138 | movq %r14, (%rsp) |
139 | /* 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) */ |
140 | .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22 |
141 | # LOE rbx r15 r12d r13d |
142 | |
143 | /* Range mask |
144 | * bits check |
145 | */ |
146 | |
147 | L(RANGEMASK_CHECK): |
148 | btl %r12d, %r13d |
149 | |
150 | /* Call scalar math function */ |
151 | jc L(SCALAR_MATH_CALL) |
152 | # LOE rbx r15 r12d r13d |
153 | |
154 | /* Special inputs |
155 | * processing loop |
156 | */ |
157 | |
158 | L(SPECIAL_VALUES_LOOP): |
159 | incl %r12d |
160 | cmpl $8, %r12d |
161 | |
162 | /* Check bits in range mask */ |
163 | jl L(RANGEMASK_CHECK) |
164 | # LOE rbx r15 r12d r13d |
165 | |
166 | movq 16(%rsp), %r12 |
167 | cfi_restore(12) |
168 | movq 8(%rsp), %r13 |
169 | cfi_restore(13) |
170 | movq (%rsp), %r14 |
171 | cfi_restore(14) |
172 | vmovups 64(%rsp), %ymm0 |
173 | |
174 | /* Go to exit */ |
175 | jmp L(EXIT) |
176 | /* 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) */ |
177 | .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22 |
178 | /* 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) */ |
179 | .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22 |
180 | /* 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) */ |
181 | .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22 |
182 | # LOE rbx r12 r13 r14 r15 ymm0 |
183 | |
184 | /* Scalar math function call |
185 | * to process special input |
186 | */ |
187 | |
188 | L(SCALAR_MATH_CALL): |
189 | movl %r12d, %r14d |
190 | vmovss 32(%rsp, %r14, 4), %xmm0 |
191 | call asinf@PLT |
192 | # LOE rbx r14 r15 r12d r13d xmm0 |
193 | |
194 | vmovss %xmm0, 64(%rsp, %r14, 4) |
195 | |
196 | /* Process special inputs in loop */ |
197 | jmp L(SPECIAL_VALUES_LOOP) |
198 | # LOE rbx r15 r12d r13d |
199 | END(_ZGVdN8v_asinf_avx2) |
200 | |
201 | .section .rodata, "a" |
202 | .align 32 |
203 | |
204 | #ifdef __svml_sasin_data_internal_typedef |
205 | typedef unsigned int VUINT32; |
206 | typedef struct { |
207 | __declspec(align(32)) VUINT32 AbsMask[8][1]; |
208 | __declspec(align(32)) VUINT32 OneHalf[8][1]; |
209 | __declspec(align(32)) VUINT32 SmallNorm[8][1]; |
210 | __declspec(align(32)) VUINT32 One[8][1]; |
211 | __declspec(align(32)) VUINT32 Two[8][1]; |
212 | __declspec(align(32)) VUINT32 sqrt_coeff[2][8][1]; |
213 | __declspec(align(32)) VUINT32 poly_coeff[5][8][1]; |
214 | __declspec(align(32)) VUINT32 Pi2H[8][1]; |
215 | } __svml_sasin_data_internal; |
216 | #endif |
217 | __svml_sasin_data_internal: |
218 | /* AbsMask */ |
219 | .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff |
220 | /* OneHalf */ |
221 | .align 32 |
222 | .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 |
223 | /* SmallNorm */ |
224 | .align 32 |
225 | .long 0x2f800000, 0x2f800000, 0x2f800000, 0x2f800000, 0x2f800000, 0x2f800000, 0x2f800000, 0x2f800000 |
226 | /* One */ |
227 | .align 32 |
228 | .long 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 |
229 | /* Two */ |
230 | .align 32 |
231 | .long 0x40000000, 0x40000000, 0x40000000, 0x40000000, 0x40000000, 0x40000000, 0x40000000, 0x40000000 |
232 | /* sqrt_coeff[2] */ |
233 | .align 32 |
234 | .long 0xbdC00004, 0xbdC00004, 0xbdC00004, 0xbdC00004, 0xbdC00004, 0xbdC00004, 0xbdC00004, 0xbdC00004 /* sqrt_coeff2 */ |
235 | .long 0x3e800001, 0x3e800001, 0x3e800001, 0x3e800001, 0x3e800001, 0x3e800001, 0x3e800001, 0x3e800001 /* sqrt_coeff1 */ |
236 | /* poly_coeff[5] */ |
237 | .align 32 |
238 | .long 0x3d2EDC07, 0x3d2EDC07, 0x3d2EDC07, 0x3d2EDC07, 0x3d2EDC07, 0x3d2EDC07, 0x3d2EDC07, 0x3d2EDC07 /* poly_coeff5 */ |
239 | .long 0x3CC32A6B, 0x3CC32A6B, 0x3CC32A6B, 0x3CC32A6B, 0x3CC32A6B, 0x3CC32A6B, 0x3CC32A6B, 0x3CC32A6B /* poly_coeff4 */ |
240 | .long 0x3d3A9AB4, 0x3d3A9AB4, 0x3d3A9AB4, 0x3d3A9AB4, 0x3d3A9AB4, 0x3d3A9AB4, 0x3d3A9AB4, 0x3d3A9AB4 /* poly_coeff3 */ |
241 | .long 0x3d997C12, 0x3d997C12, 0x3d997C12, 0x3d997C12, 0x3d997C12, 0x3d997C12, 0x3d997C12, 0x3d997C12 /* poly_coeff2 */ |
242 | .long 0x3e2AAAFF, 0x3e2AAAFF, 0x3e2AAAFF, 0x3e2AAAFF, 0x3e2AAAFF, 0x3e2AAAFF, 0x3e2AAAFF, 0x3e2AAAFF /* poly_coeff1 */ |
243 | /* Pi2H */ |
244 | .align 32 |
245 | .long 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB, 0x3fc90FDB |
246 | .align 32 |
247 | .type __svml_sasin_data_internal, @object |
248 | .size __svml_sasin_data_internal, .-__svml_sasin_data_internal |
249 | |