1 /*
2 * Copyright (c) 2016, Intel Corporation.
3 * Intel Math Library (LIBM) Source Code
4 *
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "asm/assembler.hpp"
29 #include "asm/assembler.inline.hpp"
30 #include "macroAssembler_x86.hpp"
31 #include "runtime/stubRoutines.hpp"
32 #include "utilities/globalDefinitions.hpp"
33
34 /******************************************************************************/
35 // ALGORITHM DESCRIPTION - LOG10()
36 // ---------------------
37 //
38 // Let x=2^k * mx, mx in [1,2)
39 //
40 // Get B~1/mx based on the output of rcpss instruction (B0)
41 // B = int((B0*LH*2^7+0.5))/2^7
42 // LH is a short approximation for log10(e)
43 //
44 // Reduced argument: r=B*mx-LH (computed accurately in high and low parts)
45 //
46 // Result: k*log10(2) - log(B) + p(r)
47 // p(r) is a degree 7 polynomial
48 // -log(B) read from data table (high, low parts)
49 // Result is formed from high and low parts
50 //
51 // Special cases:
52 // log10(0) = -INF with divide-by-zero exception raised
53 // log10(1) = +0
54 // log10(x) = NaN with invalid exception raised if x < -0, including -INF
55 // log10(+INF) = +INF
56 //
57 /******************************************************************************/
58
59 #ifdef _LP64
60 // The 64 bit code is at most SSE2 compliant
61 ATTRIBUTE_ALIGNED(16) juint _HIGHSIGMASK_log10[] =
62 {
63 0xf8000000UL, 0xffffffffUL, 0x00000000UL, 0xffffe000UL
64 };
65
66 ATTRIBUTE_ALIGNED(16) juint _LOG10_E[] =
67 {
68 0x00000000UL, 0x3fdbc000UL, 0xbf2e4108UL, 0x3f5a7a6cUL
69 };
70
71 ATTRIBUTE_ALIGNED(16) juint _L_tbl_log10[] =
72 {
73 0x509f7800UL, 0x3fd34413UL, 0x1f12b358UL, 0x3d1fef31UL, 0x80333400UL,
74 0x3fd32418UL, 0xc671d9d0UL, 0xbcf542bfUL, 0x51195000UL, 0x3fd30442UL,
75 0x78a4b0c3UL, 0x3d18216aUL, 0x6fc79400UL, 0x3fd2e490UL, 0x80fa389dUL,
76 0xbc902869UL, 0x89d04000UL, 0x3fd2c502UL, 0x75c2f564UL, 0x3d040754UL,
77 0x4ddd1c00UL, 0x3fd2a598UL, 0xd219b2c3UL, 0xbcfa1d84UL, 0x6baa7c00UL,
78 0x3fd28651UL, 0xfd9abec1UL, 0x3d1be6d3UL, 0x94028800UL, 0x3fd2672dUL,
79 0xe289a455UL, 0xbd1ede5eUL, 0x78b86400UL, 0x3fd2482cUL, 0x6734d179UL,
80 0x3d1fe79bUL, 0xcca3c800UL, 0x3fd2294dUL, 0x981a40b8UL, 0xbced34eaUL,
81 0x439c5000UL, 0x3fd20a91UL, 0xcc392737UL, 0xbd1a9cc3UL, 0x92752c00UL,
82 0x3fd1ebf6UL, 0x03c9afe7UL, 0x3d1e98f8UL, 0x6ef8dc00UL, 0x3fd1cd7dUL,
83 0x71dae7f4UL, 0x3d08a86cUL, 0x8fe4dc00UL, 0x3fd1af25UL, 0xee9185a1UL,
84 0xbcff3412UL, 0xace59400UL, 0x3fd190eeUL, 0xc2cab353UL, 0x3cf17ed9UL,
85 0x7e925000UL, 0x3fd172d8UL, 0x6952c1b2UL, 0x3cf1521cUL, 0xbe694400UL,
86 0x3fd154e2UL, 0xcacb79caUL, 0xbd0bdc78UL, 0x26cbac00UL, 0x3fd1370dUL,
87 0xf71f4de1UL, 0xbd01f8beUL, 0x72fa0800UL, 0x3fd11957UL, 0x55bf910bUL,
88 0x3c946e2bUL, 0x5f106000UL, 0x3fd0fbc1UL, 0x39e639c1UL, 0x3d14a84bUL,
89 0xa802a800UL, 0x3fd0de4aUL, 0xd3f31d5dUL, 0xbd178385UL, 0x0b992000UL,
90 0x3fd0c0f3UL, 0x3843106fUL, 0xbd1f602fUL, 0x486ce800UL, 0x3fd0a3baUL,
91 0x8819497cUL, 0x3cef987aUL, 0x1de49400UL, 0x3fd086a0UL, 0x1caa0467UL,
92 0x3d0faec7UL, 0x4c30cc00UL, 0x3fd069a4UL, 0xa4424372UL, 0xbd1618fcUL,
93 0x94490000UL, 0x3fd04cc6UL, 0x946517d2UL, 0xbd18384bUL, 0xb7e84000UL,
94 0x3fd03006UL, 0xe0109c37UL, 0xbd19a6acUL, 0x798a0c00UL, 0x3fd01364UL,
95 0x5121e864UL, 0xbd164cf7UL, 0x38ce8000UL, 0x3fcfedbfUL, 0x46214d1aUL,
96 0xbcbbc402UL, 0xc8e62000UL, 0x3fcfb4efUL, 0xdab93203UL, 0x3d1e0176UL,
97 0x2cb02800UL, 0x3fcf7c5aUL, 0x2a2ea8e4UL, 0xbcfec86aUL, 0xeeeaa000UL,
98 0x3fcf43fdUL, 0xc18e49a4UL, 0x3cf110a8UL, 0x9bb6e800UL, 0x3fcf0bdaUL,
99 0x923cc9c0UL, 0xbd15ce99UL, 0xc093f000UL, 0x3fced3efUL, 0x4d4b51e9UL,
100 0x3d1a04c7UL, 0xec58f800UL, 0x3fce9c3cUL, 0x163cad59UL, 0x3cac8260UL,
101 0x9a907000UL, 0x3fce2d7dUL, 0x3fa93646UL, 0x3ce4a1c0UL, 0x37311000UL,
102 0x3fcdbf99UL, 0x32abd1fdUL, 0x3d07ea9dUL, 0x6744b800UL, 0x3fcd528cUL,
103 0x4dcbdfd4UL, 0xbd1b08e2UL, 0xe36de800UL, 0x3fcce653UL, 0x0b7b7f7fUL,
104 0xbd1b8f03UL, 0x77506800UL, 0x3fcc7aecUL, 0xa821c9fbUL, 0x3d13c163UL,
105 0x00ff8800UL, 0x3fcc1053UL, 0x536bca76UL, 0xbd074ee5UL, 0x70719800UL,
106 0x3fcba684UL, 0xd7da9b6bUL, 0xbd1fbf16UL, 0xc6f8d800UL, 0x3fcb3d7dUL,
107 0xe2220bb3UL, 0x3d1a295dUL, 0x16c15800UL, 0x3fcad53cUL, 0xe724911eUL,
108 0xbcf55822UL, 0x82533800UL, 0x3fca6dbcUL, 0x6d982371UL, 0x3cac567cUL,
109 0x3c19e800UL, 0x3fca06fcUL, 0x84d17d80UL, 0x3d1da204UL, 0x85ef8000UL,
110 0x3fc9a0f8UL, 0x54466a6aUL, 0xbd002204UL, 0xb0ac2000UL, 0x3fc93baeUL,
111 0xd601fd65UL, 0x3d18840cUL, 0x1bb9b000UL, 0x3fc8d71cUL, 0x7bf58766UL,
112 0xbd14f897UL, 0x34aae800UL, 0x3fc8733eUL, 0x3af6ac24UL, 0xbd0f5c45UL,
113 0x76d68000UL, 0x3fc81012UL, 0x4303e1a1UL, 0xbd1f9a80UL, 0x6af57800UL,
114 0x3fc7ad96UL, 0x43fbcb46UL, 0x3cf4c33eUL, 0xa6c51000UL, 0x3fc74bc7UL,
115 0x70f0eac5UL, 0xbd192e3bUL, 0xccab9800UL, 0x3fc6eaa3UL, 0xc0093dfeUL,
116 0xbd0faf15UL, 0x8b60b800UL, 0x3fc68a28UL, 0xde78d5fdUL, 0xbc9ea4eeUL,
117 0x9d987000UL, 0x3fc62a53UL, 0x962bea6eUL, 0xbd194084UL, 0xc9b0e800UL,
118 0x3fc5cb22UL, 0x888dd999UL, 0x3d1fe201UL, 0xe1634800UL, 0x3fc56c93UL,
119 0x16ada7adUL, 0x3d1b1188UL, 0xc176c000UL, 0x3fc50ea4UL, 0x4159b5b5UL,
120 0xbcf09c08UL, 0x51766000UL, 0x3fc4b153UL, 0x84393d23UL, 0xbcf6a89cUL,
121 0x83695000UL, 0x3fc4549dUL, 0x9f0b8bbbUL, 0x3d1c4b8cUL, 0x538d5800UL,
122 0x3fc3f881UL, 0xf49df747UL, 0x3cf89b99UL, 0xc8138000UL, 0x3fc39cfcUL,
123 0xd503b834UL, 0xbd13b99fUL, 0xf0df0800UL, 0x3fc3420dUL, 0xf011b386UL,
124 0xbd05d8beUL, 0xe7466800UL, 0x3fc2e7b2UL, 0xf39c7bc2UL, 0xbd1bb94eUL,
125 0xcdd62800UL, 0x3fc28de9UL, 0x05e6d69bUL, 0xbd10ed05UL, 0xd015d800UL,
126 0x3fc234b0UL, 0xe29b6c9dUL, 0xbd1ff967UL, 0x224ea800UL, 0x3fc1dc06UL,
127 0x727711fcUL, 0xbcffb30dUL, 0x01540000UL, 0x3fc183e8UL, 0x39786c5aUL,
128 0x3cc23f57UL, 0xb24d9800UL, 0x3fc12c54UL, 0xc905a342UL, 0x3d003a1dUL,
129 0x82835800UL, 0x3fc0d54aUL, 0x9b9920c0UL, 0x3d03b25aUL, 0xc72ac000UL,
130 0x3fc07ec7UL, 0x46f26a24UL, 0x3cf0fa41UL, 0xdd35d800UL, 0x3fc028caUL,
131 0x41d9d6dcUL, 0x3d034a65UL, 0x52474000UL, 0x3fbfa6a4UL, 0x44f66449UL,
132 0x3d19cad3UL, 0x2da3d000UL, 0x3fbefcb8UL, 0x67832999UL, 0x3d18400fUL,
133 0x32a10000UL, 0x3fbe53ceUL, 0x9c0e3b1aUL, 0xbcff62fdUL, 0x556b7000UL,
134 0x3fbdabe3UL, 0x02976913UL, 0xbcf8243bUL, 0x97e88000UL, 0x3fbd04f4UL,
135 0xec793797UL, 0x3d1c0578UL, 0x09647000UL, 0x3fbc5effUL, 0x05fc0565UL,
136 0xbd1d799eUL, 0xc6426000UL, 0x3fbbb9ffUL, 0x4625f5edUL, 0x3d1f5723UL,
137 0xf7afd000UL, 0x3fbb15f3UL, 0xdd5aae61UL, 0xbd1a7e1eUL, 0xd358b000UL,
138 0x3fba72d8UL, 0x3314e4d3UL, 0x3d17bc91UL, 0x9b1f5000UL, 0x3fb9d0abUL,
139 0x9a4d514bUL, 0x3cf18c9bUL, 0x9cd4e000UL, 0x3fb92f69UL, 0x7e4496abUL,
140 0x3cf1f96dUL, 0x31f4f000UL, 0x3fb88f10UL, 0xf56479e7UL, 0x3d165818UL,
141 0xbf628000UL, 0x3fb7ef9cUL, 0x26bf486dUL, 0xbd1113a6UL, 0xb526b000UL,
142 0x3fb7510cUL, 0x1a1c3384UL, 0x3ca9898dUL, 0x8e31e000UL, 0x3fb6b35dUL,
143 0xb3875361UL, 0xbd0661acUL, 0xd01de000UL, 0x3fb6168cUL, 0x2a7cacfaUL,
144 0xbd1bdf10UL, 0x0af23000UL, 0x3fb57a98UL, 0xff868816UL, 0x3cf046d0UL,
145 0xd8ea0000UL, 0x3fb4df7cUL, 0x1515fbe7UL, 0xbd1fd529UL, 0xde3b2000UL,
146 0x3fb44538UL, 0x6e59a132UL, 0x3d1faeeeUL, 0xc8df9000UL, 0x3fb3abc9UL,
147 0xf1322361UL, 0xbd198807UL, 0x505f1000UL, 0x3fb3132dUL, 0x0888e6abUL,
148 0x3d1e5380UL, 0x359bd000UL, 0x3fb27b61UL, 0xdfbcbb22UL, 0xbcfe2724UL,
149 0x429ee000UL, 0x3fb1e463UL, 0x6eb4c58cUL, 0xbcfe4dd6UL, 0x4a673000UL,
150 0x3fb14e31UL, 0x4ce1ac9bUL, 0x3d1ba691UL, 0x28b96000UL, 0x3fb0b8c9UL,
151 0x8c7813b8UL, 0xbd0b3872UL, 0xc1f08000UL, 0x3fb02428UL, 0xc2bc8c2cUL,
152 0x3cb5ea6bUL, 0x05a1a000UL, 0x3faf209cUL, 0x72e8f18eUL, 0xbce8df84UL,
153 0xc0b5e000UL, 0x3fadfa6dUL, 0x9fdef436UL, 0x3d087364UL, 0xaf416000UL,
154 0x3facd5c2UL, 0x1068c3a9UL, 0x3d0827e7UL, 0xdb356000UL, 0x3fabb296UL,
155 0x120a34d3UL, 0x3d101a9fUL, 0x5dfea000UL, 0x3faa90e6UL, 0xdaded264UL,
156 0xbd14c392UL, 0x6034c000UL, 0x3fa970adUL, 0x1c9d06a9UL, 0xbd1b705eUL,
157 0x194c6000UL, 0x3fa851e8UL, 0x83996ad9UL, 0xbd0117bcUL, 0xcf4ac000UL,
158 0x3fa73492UL, 0xb1a94a62UL, 0xbca5ea42UL, 0xd67b4000UL, 0x3fa618a9UL,
159 0x75aed8caUL, 0xbd07119bUL, 0x9126c000UL, 0x3fa4fe29UL, 0x5291d533UL,
160 0x3d12658fUL, 0x6f4d4000UL, 0x3fa3e50eUL, 0xcd2c5cd9UL, 0x3d1d5c70UL,
161 0xee608000UL, 0x3fa2cd54UL, 0xd1008489UL, 0x3d1a4802UL, 0x9900e000UL,
162 0x3fa1b6f9UL, 0x54fb5598UL, 0xbd16593fUL, 0x06bb6000UL, 0x3fa0a1f9UL,
163 0x64ef57b4UL, 0xbd17636bUL, 0xb7940000UL, 0x3f9f1c9fUL, 0xee6a4737UL,
164 0x3cb5d479UL, 0x91aa0000UL, 0x3f9cf7f5UL, 0x3a16373cUL, 0x3d087114UL,
165 0x156b8000UL, 0x3f9ad5edUL, 0x836c554aUL, 0x3c6900b0UL, 0xd4764000UL,
166 0x3f98b67fUL, 0xed12f17bUL, 0xbcffc974UL, 0x77dec000UL, 0x3f9699a7UL,
167 0x232ce7eaUL, 0x3d1e35bbUL, 0xbfbf4000UL, 0x3f947f5dUL, 0xd84ffa6eUL,
168 0x3d0e0a49UL, 0x82c7c000UL, 0x3f92679cUL, 0x8d170e90UL, 0xbd14d9f2UL,
169 0xadd20000UL, 0x3f90525dUL, 0x86d9f88eUL, 0x3cdeb986UL, 0x86f10000UL,
170 0x3f8c7f36UL, 0xb9e0a517UL, 0x3ce29faaUL, 0xb75c8000UL, 0x3f885e9eUL,
171 0x542568cbUL, 0xbd1f7bdbUL, 0x46b30000UL, 0x3f8442e8UL, 0xb954e7d9UL,
172 0x3d1e5287UL, 0xb7e60000UL, 0x3f802c07UL, 0x22da0b17UL, 0xbd19fb27UL,
173 0x6c8b0000UL, 0x3f7833e3UL, 0x821271efUL, 0xbd190f96UL, 0x29910000UL,
174 0x3f701936UL, 0xbc3491a5UL, 0xbd1bcf45UL, 0x354a0000UL, 0x3f600fe3UL,
175 0xc0ff520aUL, 0xbd19d71cUL, 0x00000000UL, 0x00000000UL, 0x00000000UL,
176 0x00000000UL
177 };
178
179 ATTRIBUTE_ALIGNED(16) juint _log2_log10[] =
180 {
181 0x509f7800UL, 0x3f934413UL, 0x1f12b358UL, 0x3cdfef31UL
182 };
183
184 ATTRIBUTE_ALIGNED(16) juint _coeff_log10[] =
185 {
186 0xc1a5f12eUL, 0x40358874UL, 0x64d4ef0dUL, 0xc0089309UL, 0x385593b1UL,
187 0xc025c917UL, 0xdc963467UL, 0x3ffc6a02UL, 0x7f9d3aa1UL, 0x4016ab9fUL,
188 0xdc77b115UL, 0xbff27af2UL
189 };
190
191 // Registers:
192 // input: xmm0
193 // scratch: xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
194 // rax, rdx, rcx, tmp - r11
195
196 // Code generated by Intel C compiler for LIBM library
197
198 void MacroAssembler::fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register eax, Register ecx, Register edx, Register r11) {
199 Label L_2TAG_PACKET_0_0_2, L_2TAG_PACKET_1_0_2, L_2TAG_PACKET_2_0_2, L_2TAG_PACKET_3_0_2;
200 Label L_2TAG_PACKET_4_0_2, L_2TAG_PACKET_5_0_2, L_2TAG_PACKET_6_0_2, L_2TAG_PACKET_7_0_2;
201 Label L_2TAG_PACKET_8_0_2, L_2TAG_PACKET_9_0_2, B1_2, B1_3, B1_4, B1_5, start;
202
203 assert_different_registers(r11, eax, ecx, edx);
204
205 address HIGHSIGMASK = (address)_HIGHSIGMASK_log10;
206 address LOG10_E = (address)_LOG10_E;
207 address L_tbl = (address)_L_tbl_log10;
208 address log2 = (address)_log2_log10;
209 address coeff = (address)_coeff_log10;
210
211 bind(start);
212 subq(rsp, 24);
213 movsd(Address(rsp, 0), xmm0);
214
215 bind(B1_2);
216 xorpd(xmm2, xmm2);
217 movl(eax, 16368);
218 pinsrw(xmm2, eax, 3);
219 movl(ecx, 1054736384);
220 movdl(xmm7, ecx);
221 xorpd(xmm3, xmm3);
222 movl(edx, 30704);
223 pinsrw(xmm3, edx, 3);
224 movdqu(xmm1, xmm0);
225 movl(edx, 32768);
226 movdl(xmm4, edx);
227 movdqu(xmm5, ExternalAddress(HIGHSIGMASK)); //0xf8000000UL, 0xffffffffUL, 0x00000000UL, 0xffffe000UL
228 pextrw(eax, xmm0, 3);
229 por(xmm0, xmm2);
230 movl(ecx, 16352);
231 psrlq(xmm0, 27);
232 movdqu(xmm2, ExternalAddress(LOG10_E)); //0x00000000UL, 0x3fdbc000UL, 0xbf2e4108UL, 0x3f5a7a6cUL
233 psrld(xmm0, 2);
234 rcpps(xmm0, xmm0);
235 psllq(xmm1, 12);
236 pshufd(xmm6, xmm5, 78);
237 psrlq(xmm1, 12);
238 subl(eax, 16);
239 cmpl(eax, 32736);
240 jcc(Assembler::aboveEqual, L_2TAG_PACKET_0_0_2);
241
242 bind(L_2TAG_PACKET_1_0_2);
243 mulss(xmm0, xmm7);
244 por(xmm1, xmm3);
245 lea(r11, ExternalAddress(L_tbl));
246 andpd(xmm5, xmm1);
247 paddd(xmm0, xmm4);
248 subsd(xmm1, xmm5);
249 movdl(edx, xmm0);
250 psllq(xmm0, 29);
251 andpd(xmm0, xmm6);
252 andl(eax, 32752);
253 subl(eax, ecx);
254 cvtsi2sdl(xmm7, eax);
255 mulpd(xmm5, xmm0);
256 mulsd(xmm1, xmm0);
257 movq(xmm6, ExternalAddress(log2)); //0x509f7800UL, 0x3f934413UL, 0x1f12b358UL, 0x3cdfef31UL
258 movdqu(xmm3, ExternalAddress(coeff)); //0xc1a5f12eUL, 0x40358874UL, 0x64d4ef0dUL, 0xc0089309UL
259 subsd(xmm5, xmm2);
260 andl(edx, 16711680);
261 shrl(edx, 12);
262 movdqu(xmm0, Address(r11, rdx, Address::times_1, -1504));
263 movdqu(xmm4, ExternalAddress(16 + coeff)); //0x385593b1UL, 0xc025c917UL, 0xdc963467UL, 0x3ffc6a02UL
264 addsd(xmm1, xmm5);
265 movdqu(xmm2, ExternalAddress(32 + coeff)); //0x7f9d3aa1UL, 0x4016ab9fUL, 0xdc77b115UL, 0xbff27af2UL
266 mulsd(xmm6, xmm7);
267 pshufd(xmm5, xmm1, 68);
268 mulsd(xmm7, ExternalAddress(8 + log2)); //0x1f12b358UL, 0x3cdfef31UL
269 mulsd(xmm3, xmm1);
270 addsd(xmm0, xmm6);
271 mulpd(xmm4, xmm5);
272 movq(xmm6, ExternalAddress(8 + LOG10_E)); //0xbf2e4108UL, 0x3f5a7a6cUL
273 mulpd(xmm5, xmm5);
274 addpd(xmm4, xmm2);
275 mulpd(xmm3, xmm5);
276 pshufd(xmm2, xmm0, 228);
277 addsd(xmm0, xmm1);
278 mulsd(xmm4, xmm1);
279 subsd(xmm2, xmm0);
280 mulsd(xmm6, xmm1);
281 addsd(xmm1, xmm2);
282 pshufd(xmm2, xmm0, 238);
283 mulsd(xmm5, xmm5);
284 addsd(xmm7, xmm2);
285 addsd(xmm1, xmm6);
286 addpd(xmm4, xmm3);
287 addsd(xmm1, xmm7);
288 mulpd(xmm4, xmm5);
289 addsd(xmm1, xmm4);
290 pshufd(xmm5, xmm4, 238);
291 addsd(xmm1, xmm5);
292 addsd(xmm0, xmm1);
293 jmp(B1_5);
294
295 bind(L_2TAG_PACKET_0_0_2);
296 movq(xmm0, Address(rsp, 0));
297 movq(xmm1, Address(rsp, 0));
298 addl(eax, 16);
299 cmpl(eax, 32768);
300 jcc(Assembler::aboveEqual, L_2TAG_PACKET_2_0_2);
301 cmpl(eax, 16);
302 jcc(Assembler::below, L_2TAG_PACKET_3_0_2);
303
304 bind(L_2TAG_PACKET_4_0_2);
305 addsd(xmm0, xmm0);
306 jmp(B1_5);
307
308 bind(L_2TAG_PACKET_5_0_2);
309 jcc(Assembler::above, L_2TAG_PACKET_4_0_2);
310 cmpl(edx, 0);
311 jcc(Assembler::above, L_2TAG_PACKET_4_0_2);
312 jmp(L_2TAG_PACKET_6_0_2);
313
314 bind(L_2TAG_PACKET_3_0_2);
315 xorpd(xmm1, xmm1);
316 addsd(xmm1, xmm0);
317 movdl(edx, xmm1);
318 psrlq(xmm1, 32);
319 movdl(ecx, xmm1);
320 orl(edx, ecx);
321 cmpl(edx, 0);
322 jcc(Assembler::equal, L_2TAG_PACKET_7_0_2);
323 xorpd(xmm1, xmm1);
324 movl(eax, 18416);
325 pinsrw(xmm1, eax, 3);
326 mulsd(xmm0, xmm1);
327 xorpd(xmm2, xmm2);
328 movl(eax, 16368);
329 pinsrw(xmm2, eax, 3);
330 movdqu(xmm1, xmm0);
331 pextrw(eax, xmm0, 3);
332 por(xmm0, xmm2);
333 movl(ecx, 18416);
334 psrlq(xmm0, 27);
335 movdqu(xmm2, ExternalAddress(LOG10_E)); //0x00000000UL, 0x3fdbc000UL, 0xbf2e4108UL, 0x3f5a7a6cUL
336 psrld(xmm0, 2);
337 rcpps(xmm0, xmm0);
338 psllq(xmm1, 12);
339 pshufd(xmm6, xmm5, 78);
340 psrlq(xmm1, 12);
341 jmp(L_2TAG_PACKET_1_0_2);
342
343 bind(L_2TAG_PACKET_2_0_2);
344 movdl(edx, xmm1);
345 psrlq(xmm1, 32);
346 movdl(ecx, xmm1);
347 addl(ecx, ecx);
348 cmpl(ecx, -2097152);
349 jcc(Assembler::aboveEqual, L_2TAG_PACKET_5_0_2);
350 orl(edx, ecx);
351 cmpl(edx, 0);
352 jcc(Assembler::equal, L_2TAG_PACKET_7_0_2);
353
354 bind(L_2TAG_PACKET_6_0_2);
355 xorpd(xmm1, xmm1);
356 xorpd(xmm0, xmm0);
357 movl(eax, 32752);
358 pinsrw(xmm1, eax, 3);
359 mulsd(xmm0, xmm1);
360 movl(Address(rsp, 16), 9);
361 jmp(L_2TAG_PACKET_8_0_2);
362
363 bind(L_2TAG_PACKET_7_0_2);
364 xorpd(xmm1, xmm1);
365 xorpd(xmm0, xmm0);
366 movl(eax, 49136);
367 pinsrw(xmm0, eax, 3);
368 divsd(xmm0, xmm1);
369 movl(Address(rsp, 16), 8);
370
371 bind(L_2TAG_PACKET_8_0_2);
372 movq(Address(rsp, 8), xmm0);
373
374 bind(B1_3);
375 movq(xmm0, Address(rsp, 8));
376
377 bind(L_2TAG_PACKET_9_0_2);
378
379 bind(B1_5);
380 addq(rsp, 24);
381
382 }
383 #else
384 // The 32 bit code is at most SSE2 compliant
385 ATTRIBUTE_ALIGNED(16) juint _static_const_table_log10[] =
386 {
387 0x509f7800UL, 0x3fd34413UL, 0x1f12b358UL, 0x3d1fef31UL, 0x80333400UL,
388 0x3fd32418UL, 0xc671d9d0UL, 0xbcf542bfUL, 0x51195000UL, 0x3fd30442UL,
389 0x78a4b0c3UL, 0x3d18216aUL, 0x6fc79400UL, 0x3fd2e490UL, 0x80fa389dUL,
390 0xbc902869UL, 0x89d04000UL, 0x3fd2c502UL, 0x75c2f564UL, 0x3d040754UL,
391 0x4ddd1c00UL, 0x3fd2a598UL, 0xd219b2c3UL, 0xbcfa1d84UL, 0x6baa7c00UL,
392 0x3fd28651UL, 0xfd9abec1UL, 0x3d1be6d3UL, 0x94028800UL, 0x3fd2672dUL,
393 0xe289a455UL, 0xbd1ede5eUL, 0x78b86400UL, 0x3fd2482cUL, 0x6734d179UL,
394 0x3d1fe79bUL, 0xcca3c800UL, 0x3fd2294dUL, 0x981a40b8UL, 0xbced34eaUL,
395 0x439c5000UL, 0x3fd20a91UL, 0xcc392737UL, 0xbd1a9cc3UL, 0x92752c00UL,
396 0x3fd1ebf6UL, 0x03c9afe7UL, 0x3d1e98f8UL, 0x6ef8dc00UL, 0x3fd1cd7dUL,
397 0x71dae7f4UL, 0x3d08a86cUL, 0x8fe4dc00UL, 0x3fd1af25UL, 0xee9185a1UL,
398 0xbcff3412UL, 0xace59400UL, 0x3fd190eeUL, 0xc2cab353UL, 0x3cf17ed9UL,
399 0x7e925000UL, 0x3fd172d8UL, 0x6952c1b2UL, 0x3cf1521cUL, 0xbe694400UL,
400 0x3fd154e2UL, 0xcacb79caUL, 0xbd0bdc78UL, 0x26cbac00UL, 0x3fd1370dUL,
401 0xf71f4de1UL, 0xbd01f8beUL, 0x72fa0800UL, 0x3fd11957UL, 0x55bf910bUL,
402 0x3c946e2bUL, 0x5f106000UL, 0x3fd0fbc1UL, 0x39e639c1UL, 0x3d14a84bUL,
403 0xa802a800UL, 0x3fd0de4aUL, 0xd3f31d5dUL, 0xbd178385UL, 0x0b992000UL,
404 0x3fd0c0f3UL, 0x3843106fUL, 0xbd1f602fUL, 0x486ce800UL, 0x3fd0a3baUL,
405 0x8819497cUL, 0x3cef987aUL, 0x1de49400UL, 0x3fd086a0UL, 0x1caa0467UL,
406 0x3d0faec7UL, 0x4c30cc00UL, 0x3fd069a4UL, 0xa4424372UL, 0xbd1618fcUL,
407 0x94490000UL, 0x3fd04cc6UL, 0x946517d2UL, 0xbd18384bUL, 0xb7e84000UL,
408 0x3fd03006UL, 0xe0109c37UL, 0xbd19a6acUL, 0x798a0c00UL, 0x3fd01364UL,
409 0x5121e864UL, 0xbd164cf7UL, 0x38ce8000UL, 0x3fcfedbfUL, 0x46214d1aUL,
410 0xbcbbc402UL, 0xc8e62000UL, 0x3fcfb4efUL, 0xdab93203UL, 0x3d1e0176UL,
411 0x2cb02800UL, 0x3fcf7c5aUL, 0x2a2ea8e4UL, 0xbcfec86aUL, 0xeeeaa000UL,
412 0x3fcf43fdUL, 0xc18e49a4UL, 0x3cf110a8UL, 0x9bb6e800UL, 0x3fcf0bdaUL,
413 0x923cc9c0UL, 0xbd15ce99UL, 0xc093f000UL, 0x3fced3efUL, 0x4d4b51e9UL,
414 0x3d1a04c7UL, 0xec58f800UL, 0x3fce9c3cUL, 0x163cad59UL, 0x3cac8260UL,
415 0x9a907000UL, 0x3fce2d7dUL, 0x3fa93646UL, 0x3ce4a1c0UL, 0x37311000UL,
416 0x3fcdbf99UL, 0x32abd1fdUL, 0x3d07ea9dUL, 0x6744b800UL, 0x3fcd528cUL,
417 0x4dcbdfd4UL, 0xbd1b08e2UL, 0xe36de800UL, 0x3fcce653UL, 0x0b7b7f7fUL,
418 0xbd1b8f03UL, 0x77506800UL, 0x3fcc7aecUL, 0xa821c9fbUL, 0x3d13c163UL,
419 0x00ff8800UL, 0x3fcc1053UL, 0x536bca76UL, 0xbd074ee5UL, 0x70719800UL,
420 0x3fcba684UL, 0xd7da9b6bUL, 0xbd1fbf16UL, 0xc6f8d800UL, 0x3fcb3d7dUL,
421 0xe2220bb3UL, 0x3d1a295dUL, 0x16c15800UL, 0x3fcad53cUL, 0xe724911eUL,
422 0xbcf55822UL, 0x82533800UL, 0x3fca6dbcUL, 0x6d982371UL, 0x3cac567cUL,
423 0x3c19e800UL, 0x3fca06fcUL, 0x84d17d80UL, 0x3d1da204UL, 0x85ef8000UL,
424 0x3fc9a0f8UL, 0x54466a6aUL, 0xbd002204UL, 0xb0ac2000UL, 0x3fc93baeUL,
425 0xd601fd65UL, 0x3d18840cUL, 0x1bb9b000UL, 0x3fc8d71cUL, 0x7bf58766UL,
426 0xbd14f897UL, 0x34aae800UL, 0x3fc8733eUL, 0x3af6ac24UL, 0xbd0f5c45UL,
427 0x76d68000UL, 0x3fc81012UL, 0x4303e1a1UL, 0xbd1f9a80UL, 0x6af57800UL,
428 0x3fc7ad96UL, 0x43fbcb46UL, 0x3cf4c33eUL, 0xa6c51000UL, 0x3fc74bc7UL,
429 0x70f0eac5UL, 0xbd192e3bUL, 0xccab9800UL, 0x3fc6eaa3UL, 0xc0093dfeUL,
430 0xbd0faf15UL, 0x8b60b800UL, 0x3fc68a28UL, 0xde78d5fdUL, 0xbc9ea4eeUL,
431 0x9d987000UL, 0x3fc62a53UL, 0x962bea6eUL, 0xbd194084UL, 0xc9b0e800UL,
432 0x3fc5cb22UL, 0x888dd999UL, 0x3d1fe201UL, 0xe1634800UL, 0x3fc56c93UL,
433 0x16ada7adUL, 0x3d1b1188UL, 0xc176c000UL, 0x3fc50ea4UL, 0x4159b5b5UL,
434 0xbcf09c08UL, 0x51766000UL, 0x3fc4b153UL, 0x84393d23UL, 0xbcf6a89cUL,
435 0x83695000UL, 0x3fc4549dUL, 0x9f0b8bbbUL, 0x3d1c4b8cUL, 0x538d5800UL,
436 0x3fc3f881UL, 0xf49df747UL, 0x3cf89b99UL, 0xc8138000UL, 0x3fc39cfcUL,
437 0xd503b834UL, 0xbd13b99fUL, 0xf0df0800UL, 0x3fc3420dUL, 0xf011b386UL,
438 0xbd05d8beUL, 0xe7466800UL, 0x3fc2e7b2UL, 0xf39c7bc2UL, 0xbd1bb94eUL,
439 0xcdd62800UL, 0x3fc28de9UL, 0x05e6d69bUL, 0xbd10ed05UL, 0xd015d800UL,
440 0x3fc234b0UL, 0xe29b6c9dUL, 0xbd1ff967UL, 0x224ea800UL, 0x3fc1dc06UL,
441 0x727711fcUL, 0xbcffb30dUL, 0x01540000UL, 0x3fc183e8UL, 0x39786c5aUL,
442 0x3cc23f57UL, 0xb24d9800UL, 0x3fc12c54UL, 0xc905a342UL, 0x3d003a1dUL,
443 0x82835800UL, 0x3fc0d54aUL, 0x9b9920c0UL, 0x3d03b25aUL, 0xc72ac000UL,
444 0x3fc07ec7UL, 0x46f26a24UL, 0x3cf0fa41UL, 0xdd35d800UL, 0x3fc028caUL,
445 0x41d9d6dcUL, 0x3d034a65UL, 0x52474000UL, 0x3fbfa6a4UL, 0x44f66449UL,
446 0x3d19cad3UL, 0x2da3d000UL, 0x3fbefcb8UL, 0x67832999UL, 0x3d18400fUL,
447 0x32a10000UL, 0x3fbe53ceUL, 0x9c0e3b1aUL, 0xbcff62fdUL, 0x556b7000UL,
448 0x3fbdabe3UL, 0x02976913UL, 0xbcf8243bUL, 0x97e88000UL, 0x3fbd04f4UL,
449 0xec793797UL, 0x3d1c0578UL, 0x09647000UL, 0x3fbc5effUL, 0x05fc0565UL,
450 0xbd1d799eUL, 0xc6426000UL, 0x3fbbb9ffUL, 0x4625f5edUL, 0x3d1f5723UL,
451 0xf7afd000UL, 0x3fbb15f3UL, 0xdd5aae61UL, 0xbd1a7e1eUL, 0xd358b000UL,
452 0x3fba72d8UL, 0x3314e4d3UL, 0x3d17bc91UL, 0x9b1f5000UL, 0x3fb9d0abUL,
453 0x9a4d514bUL, 0x3cf18c9bUL, 0x9cd4e000UL, 0x3fb92f69UL, 0x7e4496abUL,
454 0x3cf1f96dUL, 0x31f4f000UL, 0x3fb88f10UL, 0xf56479e7UL, 0x3d165818UL,
455 0xbf628000UL, 0x3fb7ef9cUL, 0x26bf486dUL, 0xbd1113a6UL, 0xb526b000UL,
456 0x3fb7510cUL, 0x1a1c3384UL, 0x3ca9898dUL, 0x8e31e000UL, 0x3fb6b35dUL,
457 0xb3875361UL, 0xbd0661acUL, 0xd01de000UL, 0x3fb6168cUL, 0x2a7cacfaUL,
458 0xbd1bdf10UL, 0x0af23000UL, 0x3fb57a98UL, 0xff868816UL, 0x3cf046d0UL,
459 0xd8ea0000UL, 0x3fb4df7cUL, 0x1515fbe7UL, 0xbd1fd529UL, 0xde3b2000UL,
460 0x3fb44538UL, 0x6e59a132UL, 0x3d1faeeeUL, 0xc8df9000UL, 0x3fb3abc9UL,
461 0xf1322361UL, 0xbd198807UL, 0x505f1000UL, 0x3fb3132dUL, 0x0888e6abUL,
462 0x3d1e5380UL, 0x359bd000UL, 0x3fb27b61UL, 0xdfbcbb22UL, 0xbcfe2724UL,
463 0x429ee000UL, 0x3fb1e463UL, 0x6eb4c58cUL, 0xbcfe4dd6UL, 0x4a673000UL,
464 0x3fb14e31UL, 0x4ce1ac9bUL, 0x3d1ba691UL, 0x28b96000UL, 0x3fb0b8c9UL,
465 0x8c7813b8UL, 0xbd0b3872UL, 0xc1f08000UL, 0x3fb02428UL, 0xc2bc8c2cUL,
466 0x3cb5ea6bUL, 0x05a1a000UL, 0x3faf209cUL, 0x72e8f18eUL, 0xbce8df84UL,
467 0xc0b5e000UL, 0x3fadfa6dUL, 0x9fdef436UL, 0x3d087364UL, 0xaf416000UL,
468 0x3facd5c2UL, 0x1068c3a9UL, 0x3d0827e7UL, 0xdb356000UL, 0x3fabb296UL,
469 0x120a34d3UL, 0x3d101a9fUL, 0x5dfea000UL, 0x3faa90e6UL, 0xdaded264UL,
470 0xbd14c392UL, 0x6034c000UL, 0x3fa970adUL, 0x1c9d06a9UL, 0xbd1b705eUL,
471 0x194c6000UL, 0x3fa851e8UL, 0x83996ad9UL, 0xbd0117bcUL, 0xcf4ac000UL,
472 0x3fa73492UL, 0xb1a94a62UL, 0xbca5ea42UL, 0xd67b4000UL, 0x3fa618a9UL,
473 0x75aed8caUL, 0xbd07119bUL, 0x9126c000UL, 0x3fa4fe29UL, 0x5291d533UL,
474 0x3d12658fUL, 0x6f4d4000UL, 0x3fa3e50eUL, 0xcd2c5cd9UL, 0x3d1d5c70UL,
475 0xee608000UL, 0x3fa2cd54UL, 0xd1008489UL, 0x3d1a4802UL, 0x9900e000UL,
476 0x3fa1b6f9UL, 0x54fb5598UL, 0xbd16593fUL, 0x06bb6000UL, 0x3fa0a1f9UL,
477 0x64ef57b4UL, 0xbd17636bUL, 0xb7940000UL, 0x3f9f1c9fUL, 0xee6a4737UL,
478 0x3cb5d479UL, 0x91aa0000UL, 0x3f9cf7f5UL, 0x3a16373cUL, 0x3d087114UL,
479 0x156b8000UL, 0x3f9ad5edUL, 0x836c554aUL, 0x3c6900b0UL, 0xd4764000UL,
480 0x3f98b67fUL, 0xed12f17bUL, 0xbcffc974UL, 0x77dec000UL, 0x3f9699a7UL,
481 0x232ce7eaUL, 0x3d1e35bbUL, 0xbfbf4000UL, 0x3f947f5dUL, 0xd84ffa6eUL,
482 0x3d0e0a49UL, 0x82c7c000UL, 0x3f92679cUL, 0x8d170e90UL, 0xbd14d9f2UL,
483 0xadd20000UL, 0x3f90525dUL, 0x86d9f88eUL, 0x3cdeb986UL, 0x86f10000UL,
484 0x3f8c7f36UL, 0xb9e0a517UL, 0x3ce29faaUL, 0xb75c8000UL, 0x3f885e9eUL,
485 0x542568cbUL, 0xbd1f7bdbUL, 0x46b30000UL, 0x3f8442e8UL, 0xb954e7d9UL,
486 0x3d1e5287UL, 0xb7e60000UL, 0x3f802c07UL, 0x22da0b17UL, 0xbd19fb27UL,
487 0x6c8b0000UL, 0x3f7833e3UL, 0x821271efUL, 0xbd190f96UL, 0x29910000UL,
488 0x3f701936UL, 0xbc3491a5UL, 0xbd1bcf45UL, 0x354a0000UL, 0x3f600fe3UL,
489 0xc0ff520aUL, 0xbd19d71cUL, 0x00000000UL, 0x00000000UL, 0x00000000UL,
490 0x00000000UL, 0x509f7800UL, 0x3f934413UL, 0x1f12b358UL, 0x3cdfef31UL,
491 0xc1a5f12eUL, 0x40358874UL, 0x64d4ef0dUL, 0xc0089309UL, 0x385593b1UL,
492 0xc025c917UL, 0xdc963467UL, 0x3ffc6a02UL, 0x7f9d3aa1UL, 0x4016ab9fUL,
493 0xdc77b115UL, 0xbff27af2UL, 0xf8000000UL, 0xffffffffUL, 0x00000000UL,
494 0xffffe000UL, 0x00000000UL, 0x3fdbc000UL, 0xbf2e4108UL, 0x3f5a7a6cUL
495 };
496 //registers,
497 // input: xmm0
498 // scratch: xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7
499 // rax, rdx, rcx, rbx (tmp)
500
501 void MacroAssembler::fast_log10(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3, XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7, Register eax, Register ecx, Register edx, Register tmp) {
502
503 Label L_2TAG_PACKET_0_0_2, L_2TAG_PACKET_1_0_2, L_2TAG_PACKET_2_0_2, L_2TAG_PACKET_3_0_2;
504 Label L_2TAG_PACKET_4_0_2, L_2TAG_PACKET_5_0_2, L_2TAG_PACKET_6_0_2, L_2TAG_PACKET_7_0_2;
505 Label L_2TAG_PACKET_8_0_2, L_2TAG_PACKET_9_0_2, L_2TAG_PACKET_10_0_2, start;
506
507 assert_different_registers(tmp, eax, ecx, edx);
508
509 address static_const_table_log10 = (address)_static_const_table_log10;
510
511 bind(start);
512 subl(rsp, 104);
513 movl(Address(rsp, 40), tmp);
514 lea(tmp, ExternalAddress(static_const_table_log10));
515 xorpd(xmm2, xmm2);
516 movl(eax, 16368);
517 pinsrw(xmm2, eax, 3);
518 movl(ecx, 1054736384);
519 movdl(xmm7, ecx);
520 xorpd(xmm3, xmm3);
521 movl(edx, 30704);
522 pinsrw(xmm3, edx, 3);
523 movsd(xmm0, Address(rsp, 112));
524 movdqu(xmm1, xmm0);
525 movl(edx, 32768);
526 movdl(xmm4, edx);
527 movdqu(xmm5, Address(tmp, 2128)); //0x3ffc6a02UL, 0x7f9d3aa1UL, 0x4016ab9fUL, 0xdc77b115UL
528 pextrw(eax, xmm0, 3);
529 por(xmm0, xmm2);
530 movl(ecx, 16352);
531 psllq(xmm0, 5);
532 movsd(xmm2, Address(tmp, 2144)); //0xbff27af2UL, 0xf8000000UL, 0xffffffffUL, 0x00000000UL
533 psrlq(xmm0, 34);
534 rcpss(xmm0, xmm0);
535 psllq(xmm1, 12);
536 pshufd(xmm6, xmm5, 78);
537 psrlq(xmm1, 12);
538 subl(eax, 16);
539 cmpl(eax, 32736);
540 jcc(Assembler::aboveEqual, L_2TAG_PACKET_0_0_2);
541
542 bind(L_2TAG_PACKET_1_0_2);
543 mulss(xmm0, xmm7);
544 por(xmm1, xmm3);
545 andpd(xmm5, xmm1);
546 paddd(xmm0, xmm4);
547 subsd(xmm1, xmm5);
548 movdl(edx, xmm0);
549 psllq(xmm0, 29);
550 andpd(xmm0, xmm6);
551 andl(eax, 32752);
552 subl(eax, ecx);
553 cvtsi2sdl(xmm7, eax);
554 mulpd(xmm5, xmm0);
555 mulsd(xmm1, xmm0);
556 movsd(xmm6, Address(tmp, 2064)); //0xbd19d71cUL, 0x00000000UL, 0x00000000UL, 0x00000000UL
557 movdqu(xmm3, Address(tmp, 2080)); //0x00000000UL, 0x509f7800UL, 0x3f934413UL, 0x1f12b358UL
558 subsd(xmm5, xmm2);
559 andl(edx, 16711680);
560 shrl(edx, 12);
561 movdqu(xmm0, Address(tmp, edx, Address::times_1, -1504));
562 movdqu(xmm4, Address(tmp, 2096)); //0x3cdfef31UL, 0xc1a5f12eUL, 0x40358874UL, 0x64d4ef0dUL
563 addsd(xmm1, xmm5);
564 movdqu(xmm2, Address(tmp, 2112)); //0xc0089309UL, 0x385593b1UL, 0xc025c917UL, 0xdc963467UL
565 mulsd(xmm6, xmm7);
566 pshufd(xmm5, xmm1, 68);
567 mulsd(xmm7, Address(tmp, 2072)); //0x00000000UL, 0x00000000UL, 0x00000000UL, 0x509f7800UL
568 mulsd(xmm3, xmm1);
569 addsd(xmm0, xmm6);
570 mulpd(xmm4, xmm5);
571 movsd(xmm6, Address(tmp, 2152)); //0xffffffffUL, 0x00000000UL, 0xffffe000UL, 0x00000000UL
572 mulpd(xmm5, xmm5);
573 addpd(xmm4, xmm2);
574 mulpd(xmm3, xmm5);
575 pshufd(xmm2, xmm0, 228);
576 addsd(xmm0, xmm1);
577 mulsd(xmm4, xmm1);
578 subsd(xmm2, xmm0);
579 mulsd(xmm6, xmm1);
580 addsd(xmm1, xmm2);
581 pshufd(xmm2, xmm0, 238);
582 mulsd(xmm5, xmm5);
583 addsd(xmm7, xmm2);
584 addsd(xmm1, xmm6);
585 addpd(xmm4, xmm3);
586 addsd(xmm1, xmm7);
587 mulpd(xmm4, xmm5);
588 addsd(xmm1, xmm4);
589 pshufd(xmm5, xmm4, 238);
590 addsd(xmm1, xmm5);
591 addsd(xmm0, xmm1);
592 jmp(L_2TAG_PACKET_2_0_2);
593
594 bind(L_2TAG_PACKET_0_0_2);
595 movsd(xmm0, Address(rsp, 112)); //0xbcfa1d84UL, 0x6baa7c00UL, 0x3fd28651UL, 0xfd9abec1UL
596 movdqu(xmm1, xmm0);
597 addl(eax, 16);
598 cmpl(eax, 32768);
599 jcc(Assembler::aboveEqual, L_2TAG_PACKET_3_0_2);
600 cmpl(eax, 16);
601 jcc(Assembler::below, L_2TAG_PACKET_4_0_2);
602
603 bind(L_2TAG_PACKET_5_0_2);
604 addsd(xmm0, xmm0);
605 jmp(L_2TAG_PACKET_2_0_2);
606
607 bind(L_2TAG_PACKET_6_0_2);
608 jcc(Assembler::above, L_2TAG_PACKET_5_0_2);
609 cmpl(edx, 0);
610 jcc(Assembler::above, L_2TAG_PACKET_5_0_2);
611 jmp(L_2TAG_PACKET_7_0_2);
612
613 bind(L_2TAG_PACKET_3_0_2);
614 movdl(edx, xmm1);
615 psrlq(xmm1, 32);
616 movdl(ecx, xmm1);
617 addl(ecx, ecx);
618 cmpl(ecx, -2097152);
619 jcc(Assembler::aboveEqual, L_2TAG_PACKET_6_0_2);
620 orl(edx, ecx);
621 cmpl(edx, 0);
622 jcc(Assembler::equal, L_2TAG_PACKET_8_0_2);
623
624 bind(L_2TAG_PACKET_7_0_2);
625 xorpd(xmm1, xmm1);
626 xorpd(xmm0, xmm0);
627 movl(eax, 32752);
628 pinsrw(xmm1, eax, 3);
629 movl(edx, 9);
630 mulsd(xmm0, xmm1);
631
632 bind(L_2TAG_PACKET_9_0_2);
633 movsd(Address(rsp, 0), xmm0);
634 movsd(xmm0, Address(rsp, 112)); //0xbcfa1d84UL, 0x6baa7c00UL, 0x3fd28651UL, 0xfd9abec1UL
635 fld_d(Address(rsp, 0));
636 jmp(L_2TAG_PACKET_10_0_2);
637
638 bind(L_2TAG_PACKET_8_0_2);
639 xorpd(xmm1, xmm1);
640 xorpd(xmm0, xmm0);
641 movl(eax, 49136);
642 pinsrw(xmm0, eax, 3);
643 divsd(xmm0, xmm1);
644 movl(edx, 8);
645 jmp(L_2TAG_PACKET_9_0_2);
646
647 bind(L_2TAG_PACKET_4_0_2);
648 movdl(edx, xmm1);
649 psrlq(xmm1, 32);
650 movdl(ecx, xmm1);
651 orl(edx, ecx);
652 cmpl(edx, 0);
653 jcc(Assembler::equal, L_2TAG_PACKET_8_0_2);
654 xorpd(xmm1, xmm1);
655 movl(eax, 18416);
656 pinsrw(xmm1, eax, 3);
657 mulsd(xmm0, xmm1);
658 xorpd(xmm2, xmm2);
659 movl(eax, 16368);
660 pinsrw(xmm2, eax, 3);
661 movdqu(xmm1, xmm0);
662 pextrw(eax, xmm0, 3);
663 por(xmm0, xmm2);
664 movl(ecx, 18416);
665 psllq(xmm0, 5);
666 movsd(xmm2, Address(tmp, 2144)); //0xbff27af2UL, 0xf8000000UL, 0xffffffffUL, 0x00000000UL
667 psrlq(xmm0, 34);
668 rcpss(xmm0, xmm0);
669 psllq(xmm1, 12);
670 pshufd(xmm6, xmm5, 78);
671 psrlq(xmm1, 12);
672 jmp(L_2TAG_PACKET_1_0_2);
673
674 bind(L_2TAG_PACKET_2_0_2);
675 movsd(Address(rsp, 24), xmm0);
676 fld_d(Address(rsp, 24));
677
678 bind(L_2TAG_PACKET_10_0_2);
679 movl(tmp, Address(rsp, 40));
680
681 }
682 #endif