1 | /* Function log1p vectorized with AVX-512. |
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 | * 1+x = 2^k*(xh + xl) is computed in high-low parts; xh in [1, 2) |
23 | * Get short reciprocal approximation Rcp ~ 1/xh |
24 | * R = (Rcp*xh - 1.0) + Rcp*xl |
25 | * log1p(x) = k*log(2.0) - log(Rcp) + poly(R) |
26 | * log(Rcp) is tabulated |
27 | * |
28 | * |
29 | */ |
30 | |
31 | /* Offsets for data table __svml_dlog1p_data_internal_avx512 |
32 | */ |
33 | #define Log_tbl 0 |
34 | #define One 128 |
35 | #define SgnMask 192 |
36 | #define C075 256 |
37 | #define poly_coeff9 320 |
38 | #define poly_coeff8 384 |
39 | #define poly_coeff7 448 |
40 | #define poly_coeff6 512 |
41 | #define poly_coeff5 576 |
42 | #define poly_coeff4 640 |
43 | #define poly_coeff3 704 |
44 | #define poly_coeff2 768 |
45 | #define L2 832 |
46 | |
47 | #include <sysdep.h> |
48 | |
49 | .section .text.evex512, "ax" , @progbits |
50 | ENTRY(_ZGVeN8v_log1p_skx) |
51 | pushq %rbp |
52 | cfi_def_cfa_offset(16) |
53 | movq %rsp, %rbp |
54 | cfi_def_cfa(6, 16) |
55 | cfi_offset(6, -16) |
56 | andq $-64, %rsp |
57 | subq $192, %rsp |
58 | vmovups One+__svml_dlog1p_data_internal_avx512(%rip), %zmm7 |
59 | vmovups SgnMask+__svml_dlog1p_data_internal_avx512(%rip), %zmm14 |
60 | vmovaps %zmm0, %zmm9 |
61 | vaddpd {rn-sae}, %zmm9, %zmm7, %zmm11 |
62 | vandpd %zmm14, %zmm9, %zmm8 |
63 | |
64 | /* compute 1+x as high, low parts */ |
65 | vmaxpd {sae}, %zmm9, %zmm7, %zmm10 |
66 | vminpd {sae}, %zmm9, %zmm7, %zmm12 |
67 | |
68 | /* GetMant(x), normalized to [1, 2) for x>=0, NaN for x<0 */ |
69 | vgetmantpd $8, {sae}, %zmm11, %zmm6 |
70 | |
71 | /* GetExp(x) */ |
72 | vgetexppd {sae}, %zmm11, %zmm5 |
73 | vsubpd {rn-sae}, %zmm10, %zmm11, %zmm13 |
74 | |
75 | /* DblRcp ~ 1/Mantissa */ |
76 | vrcp14pd %zmm6, %zmm15 |
77 | |
78 | /* Start polynomial evaluation */ |
79 | vmovups poly_coeff9+__svml_dlog1p_data_internal_avx512(%rip), %zmm10 |
80 | vmovups poly_coeff7+__svml_dlog1p_data_internal_avx512(%rip), %zmm11 |
81 | |
82 | /* Xl */ |
83 | vsubpd {rn-sae}, %zmm13, %zmm12, %zmm2 |
84 | vxorpd %zmm14, %zmm5, %zmm3 |
85 | |
86 | /* round DblRcp to 4 fractional bits (RN mode, no Precision exception) */ |
87 | vrndscalepd $88, {sae}, %zmm15, %zmm4 |
88 | vmovups poly_coeff5+__svml_dlog1p_data_internal_avx512(%rip), %zmm12 |
89 | vmovups poly_coeff6+__svml_dlog1p_data_internal_avx512(%rip), %zmm14 |
90 | vmovups poly_coeff3+__svml_dlog1p_data_internal_avx512(%rip), %zmm13 |
91 | |
92 | /* Xl*2^(-Expon) */ |
93 | vscalefpd {rn-sae}, %zmm3, %zmm2, %zmm1 |
94 | |
95 | /* Reduced argument: R = DblRcp*(Mantissa+Xl) - 1 */ |
96 | vfmsub213pd {rn-sae}, %zmm7, %zmm4, %zmm6 |
97 | vmovups __svml_dlog1p_data_internal_avx512(%rip), %zmm3 |
98 | |
99 | /* |
100 | * Table lookup |
101 | * Prepare exponent correction: DblRcp<0.75? |
102 | */ |
103 | vmovups C075+__svml_dlog1p_data_internal_avx512(%rip), %zmm2 |
104 | |
105 | /* Prepare table index */ |
106 | vpsrlq $48, %zmm4, %zmm0 |
107 | vfmadd231pd {rn-sae}, %zmm4, %zmm1, %zmm6 |
108 | vmovups poly_coeff8+__svml_dlog1p_data_internal_avx512(%rip), %zmm1 |
109 | vcmppd $17, {sae}, %zmm2, %zmm4, %k1 |
110 | vcmppd $4, {sae}, %zmm6, %zmm6, %k0 |
111 | vfmadd231pd {rn-sae}, %zmm6, %zmm10, %zmm1 |
112 | vmovups poly_coeff4+__svml_dlog1p_data_internal_avx512(%rip), %zmm10 |
113 | vfmadd231pd {rn-sae}, %zmm6, %zmm11, %zmm14 |
114 | vmovups L2+__svml_dlog1p_data_internal_avx512(%rip), %zmm4 |
115 | vpermt2pd Log_tbl+64+__svml_dlog1p_data_internal_avx512(%rip), %zmm0, %zmm3 |
116 | |
117 | /* add 1 to Expon if DblRcp<0.75 */ |
118 | vaddpd {rn-sae}, %zmm7, %zmm5, %zmm5{%k1} |
119 | |
120 | /* R^2 */ |
121 | vmulpd {rn-sae}, %zmm6, %zmm6, %zmm0 |
122 | vfmadd231pd {rn-sae}, %zmm6, %zmm12, %zmm10 |
123 | vmovups poly_coeff2+__svml_dlog1p_data_internal_avx512(%rip), %zmm12 |
124 | vmulpd {rn-sae}, %zmm0, %zmm0, %zmm15 |
125 | vfmadd231pd {rn-sae}, %zmm6, %zmm13, %zmm12 |
126 | vfmadd213pd {rn-sae}, %zmm14, %zmm0, %zmm1 |
127 | kmovw %k0, %edx |
128 | vfmadd213pd {rn-sae}, %zmm12, %zmm0, %zmm10 |
129 | |
130 | /* polynomial */ |
131 | vfmadd213pd {rn-sae}, %zmm10, %zmm15, %zmm1 |
132 | vfmadd213pd {rn-sae}, %zmm6, %zmm0, %zmm1 |
133 | vaddpd {rn-sae}, %zmm1, %zmm3, %zmm6 |
134 | vfmadd213pd {rn-sae}, %zmm6, %zmm4, %zmm5 |
135 | vorpd %zmm8, %zmm5, %zmm0 |
136 | testl %edx, %edx |
137 | |
138 | /* Go to special inputs processing branch */ |
139 | jne L(SPECIAL_VALUES_BRANCH) |
140 | # LOE rbx r12 r13 r14 r15 edx zmm0 zmm9 |
141 | |
142 | /* Restore registers |
143 | * and exit the function |
144 | */ |
145 | |
146 | L(EXIT): |
147 | movq %rbp, %rsp |
148 | popq %rbp |
149 | cfi_def_cfa(7, 8) |
150 | cfi_restore(6) |
151 | ret |
152 | cfi_def_cfa(6, 16) |
153 | cfi_offset(6, -16) |
154 | |
155 | /* Branch to process |
156 | * special inputs |
157 | */ |
158 | |
159 | L(SPECIAL_VALUES_BRANCH): |
160 | vmovups %zmm9, 64(%rsp) |
161 | vmovups %zmm0, 128(%rsp) |
162 | # LOE rbx r12 r13 r14 r15 edx zmm0 |
163 | |
164 | xorl %eax, %eax |
165 | # LOE rbx r12 r13 r14 r15 eax edx |
166 | |
167 | vzeroupper |
168 | movq %r12, 16(%rsp) |
169 | /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -176; DW_OP_plus) */ |
170 | .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 0xff, 0xff, 0xff, 0x22 |
171 | movl %eax, %r12d |
172 | movq %r13, 8(%rsp) |
173 | /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -184; DW_OP_plus) */ |
174 | .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 0xff, 0xff, 0xff, 0x22 |
175 | movl %edx, %r13d |
176 | movq %r14, (%rsp) |
177 | /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -192; DW_OP_plus) */ |
178 | .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 0xff, 0xff, 0xff, 0x22 |
179 | # LOE rbx r15 r12d r13d |
180 | |
181 | /* Range mask |
182 | * bits check |
183 | */ |
184 | |
185 | L(RANGEMASK_CHECK): |
186 | btl %r12d, %r13d |
187 | |
188 | /* Call scalar math function */ |
189 | jc L(SCALAR_MATH_CALL) |
190 | # LOE rbx r15 r12d r13d |
191 | |
192 | /* Special inputs |
193 | * processing loop |
194 | */ |
195 | |
196 | L(SPECIAL_VALUES_LOOP): |
197 | incl %r12d |
198 | cmpl $8, %r12d |
199 | |
200 | /* Check bits in range mask */ |
201 | jl L(RANGEMASK_CHECK) |
202 | # LOE rbx r15 r12d r13d |
203 | |
204 | movq 16(%rsp), %r12 |
205 | cfi_restore(12) |
206 | movq 8(%rsp), %r13 |
207 | cfi_restore(13) |
208 | movq (%rsp), %r14 |
209 | cfi_restore(14) |
210 | vmovups 128(%rsp), %zmm0 |
211 | |
212 | /* Go to exit */ |
213 | jmp L(EXIT) |
214 | /* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -176; DW_OP_plus) */ |
215 | .cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x50, 0xff, 0xff, 0xff, 0x22 |
216 | /* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -184; DW_OP_plus) */ |
217 | .cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x48, 0xff, 0xff, 0xff, 0x22 |
218 | /* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -192; DW_OP_plus) */ |
219 | .cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0x40, 0xff, 0xff, 0xff, 0x22 |
220 | # LOE rbx r12 r13 r14 r15 zmm0 |
221 | |
222 | /* Scalar math function call |
223 | * to process special input |
224 | */ |
225 | |
226 | L(SCALAR_MATH_CALL): |
227 | movl %r12d, %r14d |
228 | vmovsd 64(%rsp, %r14, 8), %xmm0 |
229 | call log1p@PLT |
230 | # LOE rbx r14 r15 r12d r13d xmm0 |
231 | |
232 | vmovsd %xmm0, 128(%rsp, %r14, 8) |
233 | |
234 | /* Process special inputs in loop */ |
235 | jmp L(SPECIAL_VALUES_LOOP) |
236 | # LOE rbx r15 r12d r13d |
237 | END(_ZGVeN8v_log1p_skx) |
238 | |
239 | .section .rodata, "a" |
240 | .align 64 |
241 | |
242 | #ifdef __svml_dlog1p_data_internal_avx512_typedef |
243 | typedef unsigned int VUINT32; |
244 | typedef struct { |
245 | __declspec(align(64)) VUINT32 Log_tbl[16][2]; |
246 | __declspec(align(64)) VUINT32 One[8][2]; |
247 | __declspec(align(64)) VUINT32 SgnMask[8][2]; |
248 | __declspec(align(64)) VUINT32 C075[8][2]; |
249 | __declspec(align(64)) VUINT32 poly_coeff9[8][2]; |
250 | __declspec(align(64)) VUINT32 poly_coeff8[8][2]; |
251 | __declspec(align(64)) VUINT32 poly_coeff7[8][2]; |
252 | __declspec(align(64)) VUINT32 poly_coeff6[8][2]; |
253 | __declspec(align(64)) VUINT32 poly_coeff5[8][2]; |
254 | __declspec(align(64)) VUINT32 poly_coeff4[8][2]; |
255 | __declspec(align(64)) VUINT32 poly_coeff3[8][2]; |
256 | __declspec(align(64)) VUINT32 poly_coeff2[8][2]; |
257 | __declspec(align(64)) VUINT32 L2[8][2]; |
258 | } __svml_dlog1p_data_internal_avx512; |
259 | #endif |
260 | __svml_dlog1p_data_internal_avx512: |
261 | /* Log_tbl */ |
262 | .quad 0x0000000000000000 |
263 | .quad 0xbfaf0a30c01162a6 |
264 | .quad 0xbfbe27076e2af2e6 |
265 | .quad 0xbfc5ff3070a793d4 |
266 | .quad 0xbfcc8ff7c79a9a22 |
267 | .quad 0xbfd1675cababa60e |
268 | .quad 0xbfd4618bc21c5ec2 |
269 | .quad 0xbfd739d7f6bbd007 |
270 | .quad 0x3fd269621134db92 |
271 | .quad 0x3fcf991c6cb3b379 |
272 | .quad 0x3fca93ed3c8ad9e3 |
273 | .quad 0x3fc5bf406b543db2 |
274 | .quad 0x3fc1178e8227e47c |
275 | .quad 0x3fb9335e5d594989 |
276 | .quad 0x3fb08598b59e3a07 |
277 | .quad 0x3fa0415d89e74444 |
278 | /* One */ |
279 | .align 64 |
280 | .quad 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000, 0x3ff0000000000000 |
281 | /* SgnMask */ |
282 | .align 64 |
283 | .quad 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000 |
284 | /* C075 0.75 */ |
285 | .align 64 |
286 | .quad 0x3fe8000000000000, 0x3fe8000000000000, 0x3fe8000000000000, 0x3fe8000000000000, 0x3fe8000000000000, 0x3fe8000000000000, 0x3fe8000000000000, 0x3fe8000000000000 |
287 | /* poly_coeff9 */ |
288 | .align 64 |
289 | .quad 0x3fbC81CD309D7C70, 0x3fbC81CD309D7C70, 0x3fbC81CD309D7C70, 0x3fbC81CD309D7C70, 0x3fbC81CD309D7C70, 0x3fbC81CD309D7C70, 0x3fbC81CD309D7C70, 0x3fbC81CD309D7C70 |
290 | /* poly_coeff8 */ |
291 | .align 64 |
292 | .quad 0xbfc007357E93AF62, 0xbfc007357E93AF62, 0xbfc007357E93AF62, 0xbfc007357E93AF62, 0xbfc007357E93AF62, 0xbfc007357E93AF62, 0xbfc007357E93AF62, 0xbfc007357E93AF62 |
293 | /* poly_coeff7 */ |
294 | .align 64 |
295 | .quad 0x3fc249229CEE81EF, 0x3fc249229CEE81EF, 0x3fc249229CEE81EF, 0x3fc249229CEE81EF, 0x3fc249229CEE81EF, 0x3fc249229CEE81EF, 0x3fc249229CEE81EF, 0x3fc249229CEE81EF |
296 | /* poly_coeff6 */ |
297 | .align 64 |
298 | .quad 0xbfc55553FB28DB06, 0xbfc55553FB28DB06, 0xbfc55553FB28DB06, 0xbfc55553FB28DB06, 0xbfc55553FB28DB06, 0xbfc55553FB28DB06, 0xbfc55553FB28DB06, 0xbfc55553FB28DB06 |
299 | /* poly_coeff5 */ |
300 | .align 64 |
301 | .quad 0x3fc9999999CC9F5C, 0x3fc9999999CC9F5C, 0x3fc9999999CC9F5C, 0x3fc9999999CC9F5C, 0x3fc9999999CC9F5C, 0x3fc9999999CC9F5C, 0x3fc9999999CC9F5C, 0x3fc9999999CC9F5C |
302 | /* poly_coeff4 */ |
303 | .align 64 |
304 | .quad 0xbfd00000000C05BD, 0xbfd00000000C05BD, 0xbfd00000000C05BD, 0xbfd00000000C05BD, 0xbfd00000000C05BD, 0xbfd00000000C05BD, 0xbfd00000000C05BD, 0xbfd00000000C05BD |
305 | /* poly_coeff3 */ |
306 | .align 64 |
307 | .quad 0x3fd5555555555466, 0x3fd5555555555466, 0x3fd5555555555466, 0x3fd5555555555466, 0x3fd5555555555466, 0x3fd5555555555466, 0x3fd5555555555466, 0x3fd5555555555466 |
308 | /* poly_coeff2 */ |
309 | .align 64 |
310 | .quad 0xbfdFFFFFFFFFFFC6, 0xbfdFFFFFFFFFFFC6, 0xbfdFFFFFFFFFFFC6, 0xbfdFFFFFFFFFFFC6, 0xbfdFFFFFFFFFFFC6, 0xbfdFFFFFFFFFFFC6, 0xbfdFFFFFFFFFFFC6, 0xbfdFFFFFFFFFFFC6 |
311 | /* L2 = log(2) */ |
312 | .align 64 |
313 | .quad 0x3fe62E42FEFA39EF, 0x3fe62E42FEFA39EF, 0x3fe62E42FEFA39EF, 0x3fe62E42FEFA39EF, 0x3fe62E42FEFA39EF, 0x3fe62E42FEFA39EF, 0x3fe62E42FEFA39EF, 0x3fe62E42FEFA39EF |
314 | .align 64 |
315 | .type __svml_dlog1p_data_internal_avx512, @object |
316 | .size __svml_dlog1p_data_internal_avx512, .-__svml_dlog1p_data_internal_avx512 |
317 | |