1/*
2 * Copyright (c) 2006, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25#include "precompiled.hpp"
26#include "runtime/sharedRuntime.hpp"
27#include "utilities/align.hpp"
28#include "utilities/copy.hpp"
29
30
31// Copy bytes; larger units are filled atomically if everything is aligned.
32void Copy::conjoint_memory_atomic(const void* from, void* to, size_t size) {
33 uintptr_t bits = (uintptr_t) from | (uintptr_t) to | (uintptr_t) size;
34
35 // (Note: We could improve performance by ignoring the low bits of size,
36 // and putting a short cleanup loop after each bulk copy loop.
37 // There are plenty of other ways to make this faster also,
38 // and it's a slippery slope. For now, let's keep this code simple
39 // since the simplicity helps clarify the atomicity semantics of
40 // this operation. There are also CPU-specific assembly versions
41 // which may or may not want to include such optimizations.)
42
43 if (bits % sizeof(jlong) == 0) {
44 Copy::conjoint_jlongs_atomic((const jlong*) from, (jlong*) to, size / sizeof(jlong));
45 } else if (bits % sizeof(jint) == 0) {
46 Copy::conjoint_jints_atomic((const jint*) from, (jint*) to, size / sizeof(jint));
47 } else if (bits % sizeof(jshort) == 0) {
48 Copy::conjoint_jshorts_atomic((const jshort*) from, (jshort*) to, size / sizeof(jshort));
49 } else {
50 // Not aligned, so no need to be atomic.
51 Copy::conjoint_jbytes((const void*) from, (void*) to, size);
52 }
53}
54
55class CopySwap : AllStatic {
56public:
57 /**
58 * Copy and optionally byte swap elements
59 *
60 * <swap> - true if elements should be byte swapped
61 *
62 * @param src address of source
63 * @param dst address of destination
64 * @param byte_count number of bytes to copy
65 * @param elem_size size of the elements to copy-swap
66 */
67 template<bool swap>
68 static void conjoint_swap_if_needed(const void* src, void* dst, size_t byte_count, size_t elem_size) {
69 assert(src != NULL, "address must not be NULL");
70 assert(dst != NULL, "address must not be NULL");
71 assert(elem_size == 2 || elem_size == 4 || elem_size == 8,
72 "incorrect element size: " SIZE_FORMAT, elem_size);
73 assert(is_aligned(byte_count, elem_size),
74 "byte_count " SIZE_FORMAT " must be multiple of element size " SIZE_FORMAT, byte_count, elem_size);
75
76 address src_end = (address)src + byte_count;
77
78 if (dst <= src || dst >= src_end) {
79 do_conjoint_swap<RIGHT,swap>(src, dst, byte_count, elem_size);
80 } else {
81 do_conjoint_swap<LEFT,swap>(src, dst, byte_count, elem_size);
82 }
83 }
84
85private:
86 /**
87 * Byte swap a 16-bit value
88 */
89 static uint16_t byte_swap(uint16_t x) {
90 return (x << 8) | (x >> 8);
91 }
92
93 /**
94 * Byte swap a 32-bit value
95 */
96 static uint32_t byte_swap(uint32_t x) {
97 uint16_t lo = (uint16_t)x;
98 uint16_t hi = (uint16_t)(x >> 16);
99
100 return ((uint32_t)byte_swap(lo) << 16) | (uint32_t)byte_swap(hi);
101 }
102
103 /**
104 * Byte swap a 64-bit value
105 */
106 static uint64_t byte_swap(uint64_t x) {
107 uint32_t lo = (uint32_t)x;
108 uint32_t hi = (uint32_t)(x >> 32);
109
110 return ((uint64_t)byte_swap(lo) << 32) | (uint64_t)byte_swap(hi);
111 }
112
113 enum CopyDirection {
114 RIGHT, // lower -> higher address
115 LEFT // higher -> lower address
116 };
117
118 /**
119 * Copy and byte swap elements
120 *
121 * <T> - type of element to copy
122 * <D> - copy direction
123 * <is_src_aligned> - true if src argument is aligned to element size
124 * <is_dst_aligned> - true if dst argument is aligned to element size
125 *
126 * @param src address of source
127 * @param dst address of destination
128 * @param byte_count number of bytes to copy
129 */
130 template <typename T, CopyDirection D, bool swap, bool is_src_aligned, bool is_dst_aligned>
131 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count) {
132 const char* cur_src;
133 char* cur_dst;
134
135 switch (D) {
136 case RIGHT:
137 cur_src = (const char*)src;
138 cur_dst = (char*)dst;
139 break;
140 case LEFT:
141 cur_src = (const char*)src + byte_count - sizeof(T);
142 cur_dst = (char*)dst + byte_count - sizeof(T);
143 break;
144 }
145
146 for (size_t i = 0; i < byte_count / sizeof(T); i++) {
147 T tmp;
148
149 if (is_src_aligned) {
150 tmp = *(T*)cur_src;
151 } else {
152 memcpy(&tmp, cur_src, sizeof(T));
153 }
154
155 if (swap) {
156 tmp = byte_swap(tmp);
157 }
158
159 if (is_dst_aligned) {
160 *(T*)cur_dst = tmp;
161 } else {
162 memcpy(cur_dst, &tmp, sizeof(T));
163 }
164
165 switch (D) {
166 case RIGHT:
167 cur_src += sizeof(T);
168 cur_dst += sizeof(T);
169 break;
170 case LEFT:
171 cur_src -= sizeof(T);
172 cur_dst -= sizeof(T);
173 break;
174 }
175 }
176 }
177
178 /**
179 * Copy and byte swap elements
180 *
181 * <T> - type of element to copy
182 * <D> - copy direction
183 * <swap> - true if elements should be byte swapped
184 *
185 * @param src address of source
186 * @param dst address of destination
187 * @param byte_count number of bytes to copy
188 */
189 template <typename T, CopyDirection direction, bool swap>
190 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count) {
191 if (is_aligned(src, sizeof(T))) {
192 if (is_aligned(dst, sizeof(T))) {
193 do_conjoint_swap<T,direction,swap,true,true>(src, dst, byte_count);
194 } else {
195 do_conjoint_swap<T,direction,swap,true,false>(src, dst, byte_count);
196 }
197 } else {
198 if (is_aligned(dst, sizeof(T))) {
199 do_conjoint_swap<T,direction,swap,false,true>(src, dst, byte_count);
200 } else {
201 do_conjoint_swap<T,direction,swap,false,false>(src, dst, byte_count);
202 }
203 }
204 }
205
206
207 /**
208 * Copy and byte swap elements
209 *
210 * <D> - copy direction
211 * <swap> - true if elements should be byte swapped
212 *
213 * @param src address of source
214 * @param dst address of destination
215 * @param byte_count number of bytes to copy
216 * @param elem_size size of the elements to copy-swap
217 */
218 template <CopyDirection D, bool swap>
219 static void do_conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size) {
220 switch (elem_size) {
221 case 2: do_conjoint_swap<uint16_t,D,swap>(src, dst, byte_count); break;
222 case 4: do_conjoint_swap<uint32_t,D,swap>(src, dst, byte_count); break;
223 case 8: do_conjoint_swap<uint64_t,D,swap>(src, dst, byte_count); break;
224 default: guarantee(false, "do_conjoint_swap: Invalid elem_size " SIZE_FORMAT "\n", elem_size);
225 }
226 }
227};
228
229void Copy::conjoint_copy(const void* src, void* dst, size_t byte_count, size_t elem_size) {
230 CopySwap::conjoint_swap_if_needed<false>(src, dst, byte_count, elem_size);
231}
232
233void Copy::conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size) {
234 CopySwap::conjoint_swap_if_needed<true>(src, dst, byte_count, elem_size);
235}
236
237// Fill bytes; larger units are filled atomically if everything is aligned.
238void Copy::fill_to_memory_atomic(void* to, size_t size, jubyte value) {
239 address dst = (address) to;
240 uintptr_t bits = (uintptr_t) to | (uintptr_t) size;
241 if (bits % sizeof(jlong) == 0) {
242 jlong fill = (julong)( (jubyte)value ); // zero-extend
243 if (fill != 0) {
244 fill += fill << 8;
245 fill += fill << 16;
246 fill += fill << 32;
247 }
248 //Copy::fill_to_jlongs_atomic((jlong*) dst, size / sizeof(jlong));
249 for (uintptr_t off = 0; off < size; off += sizeof(jlong)) {
250 *(jlong*)(dst + off) = fill;
251 }
252 } else if (bits % sizeof(jint) == 0) {
253 jint fill = (juint)( (jubyte)value ); // zero-extend
254 if (fill != 0) {
255 fill += fill << 8;
256 fill += fill << 16;
257 }
258 //Copy::fill_to_jints_atomic((jint*) dst, size / sizeof(jint));
259 for (uintptr_t off = 0; off < size; off += sizeof(jint)) {
260 *(jint*)(dst + off) = fill;
261 }
262 } else if (bits % sizeof(jshort) == 0) {
263 jshort fill = (jushort)( (jubyte)value ); // zero-extend
264 fill += fill << 8;
265 //Copy::fill_to_jshorts_atomic((jshort*) dst, size / sizeof(jshort));
266 for (uintptr_t off = 0; off < size; off += sizeof(jshort)) {
267 *(jshort*)(dst + off) = fill;
268 }
269 } else {
270 // Not aligned, so no need to be atomic.
271 Copy::fill_to_bytes(dst, size, value);
272 }
273}
274