1 /*
2 * Copyright (c) 2003, 2025, 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 #ifndef SHARE_UTILITIES_COPY_HPP
26 #define SHARE_UTILITIES_COPY_HPP
27
28 #include "oops/oopsHierarchy.hpp"
29 #include "runtime/atomicAccess.hpp"
30 #include "runtime/globals.hpp"
31 #include "utilities/align.hpp"
32 #include "utilities/bytes.hpp"
33 #include "utilities/debug.hpp"
34 #include "utilities/macros.hpp"
35
36 // Assembly code for platforms that need it.
37 extern "C" {
38 void _Copy_conjoint_words(const HeapWord* from, HeapWord* to, size_t count);
39 void _Copy_disjoint_words(const HeapWord* from, HeapWord* to, size_t count);
40
41 void _Copy_conjoint_jshorts_atomic(const jshort* from, jshort* to, size_t count);
42 void _Copy_conjoint_jints_atomic (const jint* from, jint* to, size_t count);
43 void _Copy_conjoint_jlongs_atomic (const jlong* from, jlong* to, size_t count);
44
45 void _Copy_arrayof_conjoint_bytes (const HeapWord* from, HeapWord* to, size_t count);
46 void _Copy_arrayof_conjoint_jshorts(const HeapWord* from, HeapWord* to, size_t count);
47 void _Copy_arrayof_conjoint_jints (const HeapWord* from, HeapWord* to, size_t count);
48 void _Copy_arrayof_conjoint_jlongs (const HeapWord* from, HeapWord* to, size_t count);
49 }
50
51 class Copy : AllStatic {
52 public:
53 // Block copy methods have four attributes. We don't define all possibilities.
54 // alignment: aligned to BytesPerLong
55 // arrayof: arraycopy operation with both operands aligned on the same
56 // boundary as the first element of an array of the copy unit.
57 // This is currently a HeapWord boundary on all platforms, except
58 // for long and double arrays, which are aligned on an 8-byte
59 // boundary on all platforms.
60 // arraycopy operations are implicitly atomic on each array element.
61 // overlap: disjoint or conjoint.
62 // copy unit: bytes or words (i.e., HeapWords) or oops (i.e., pointers).
63 // atomicity: atomic or non-atomic on the copy unit.
64 //
65 // Names are constructed thusly:
66 //
67 // [ 'aligned_' | 'arrayof_' ]
68 // ('conjoint_' | 'disjoint_')
69 // ('words' | 'bytes' | 'jshorts' | 'jints' | 'jlongs' | 'oops')
70 // [ '_atomic' ]
71 //
72 // Except in the arrayof case, whatever the alignment is, we assume we can copy
73 // whole alignment units. E.g., if BytesPerLong is 2x word alignment, an odd
74 // count may copy an extra word. In the arrayof case, we are allowed to copy
75 // only the number of copy units specified.
76 //
77 // All callees check count for 0.
78 //
79
80 // HeapWords
81
82 // Word-aligned words, conjoint, not atomic on each word
83 static void conjoint_words(const HeapWord* from, HeapWord* to, size_t count) {
84 assert_params_ok(from, to, HeapWordSize);
85 pd_conjoint_words(from, to, count);
86 }
87
88 // Word-aligned words, disjoint, not atomic on each word
89 static void disjoint_words(const HeapWord* from, HeapWord* to, size_t count) {
90 assert_params_ok(from, to, HeapWordSize);
91 assert_disjoint(from, to, count);
92 pd_disjoint_words(from, to, count);
93 }
94
95 // Word-aligned words, disjoint, atomic on each word
96 static void disjoint_words_atomic(const HeapWord* from, HeapWord* to, size_t count) {
97 assert_params_ok(from, to, HeapWordSize);
98 assert_disjoint(from, to, count);
99 pd_disjoint_words_atomic(from, to, count);
100 }
101
102 // Object-aligned words, conjoint, not atomic on each word
103 static void aligned_conjoint_words(const HeapWord* from, HeapWord* to, size_t count) {
104 assert_params_aligned(from, to);
105 pd_aligned_conjoint_words(from, to, count);
106 }
107
108 // Object-aligned words, disjoint, not atomic on each word
109 static void aligned_disjoint_words(const HeapWord* from, HeapWord* to, size_t count) {
110 assert_params_aligned(from, to);
111 assert_disjoint(from, to, count);
112 pd_aligned_disjoint_words(from, to, count);
113 }
114
115 // bytes, jshorts, jints, jlongs, oops
116
117 // bytes, conjoint, not atomic on each byte (not that it matters)
118 static void conjoint_jbytes(const void* from, void* to, size_t count) {
119 pd_conjoint_bytes(from, to, count);
120 }
121
122 // bytes, conjoint, atomic on each byte (not that it matters)
123 static void conjoint_jbytes_atomic(const void* from, void* to, size_t count) {
124 pd_conjoint_bytes(from, to, count);
125 }
126
127 // jshorts, conjoint, atomic on each jshort
128 static void conjoint_jshorts_atomic(const jshort* from, jshort* to, size_t count) {
129 assert_params_ok(from, to, BytesPerShort);
130 pd_conjoint_jshorts_atomic(from, to, count);
131 }
132
133 // jints, conjoint, atomic on each jint
134 static void conjoint_jints_atomic(const jint* from, jint* to, size_t count) {
135 assert_params_ok(from, to, BytesPerInt);
136 pd_conjoint_jints_atomic(from, to, count);
137 }
138
139 // jlongs, conjoint, atomic on each jlong
140 static void conjoint_jlongs_atomic(const jlong* from, jlong* to, size_t count) {
141 assert_params_ok(from, to, BytesPerLong);
142 pd_conjoint_jlongs_atomic(from, to, count);
143 }
144
145 // oops, conjoint, atomic on each oop
146 static void conjoint_oops_atomic(const oop* from, oop* to, size_t count) {
147 assert_params_ok(from, to, BytesPerHeapOop);
148 pd_conjoint_oops_atomic(from, to, count);
149 }
150
151 // overloaded for UseCompressedOops
152 static void conjoint_oops_atomic(const narrowOop* from, narrowOop* to, size_t count) {
153 assert(sizeof(narrowOop) == sizeof(jint), "this cast is wrong");
154 assert_params_ok(from, to, BytesPerInt);
155 pd_conjoint_jints_atomic((const jint*)from, (jint*)to, count);
156 }
157
158 // Copy a span of memory. If the span is an integral number of aligned
159 // longs, words, or ints, copy those units atomically.
160 // The largest atomic transfer unit is 8 bytes, or the largest power
161 // of two which divides all of from, to, and size, whichever is smaller.
162 static void conjoint_memory_atomic(const void* from, void* to, size_t size);
163
164 // bytes, conjoint array, atomic on each byte (not that it matters)
165 static void arrayof_conjoint_jbytes(const HeapWord* from, HeapWord* to, size_t count) {
166 pd_arrayof_conjoint_bytes(from, to, count);
167 }
168
169 // jshorts, conjoint array, atomic on each jshort
170 static void arrayof_conjoint_jshorts(const HeapWord* from, HeapWord* to, size_t count) {
171 assert_params_ok(from, to, BytesPerShort);
172 pd_arrayof_conjoint_jshorts(from, to, count);
173 }
174
175 // jints, conjoint array, atomic on each jint
176 static void arrayof_conjoint_jints(const HeapWord* from, HeapWord* to, size_t count) {
177 assert_params_ok(from, to, BytesPerInt);
178 pd_arrayof_conjoint_jints(from, to, count);
179 }
180
181 // jlongs, conjoint array, atomic on each jlong
182 static void arrayof_conjoint_jlongs(const HeapWord* from, HeapWord* to, size_t count) {
183 assert_params_ok(from, to, BytesPerLong);
184 pd_arrayof_conjoint_jlongs(from, to, count);
185 }
186
187 // oops, conjoint array, atomic on each oop
188 static void arrayof_conjoint_oops(const HeapWord* from, HeapWord* to, size_t count) {
189 assert_params_ok(from, to, BytesPerHeapOop);
190 pd_arrayof_conjoint_oops(from, to, count);
191 }
192
193 // Known overlap methods
194
195 // Copy word-aligned words from higher to lower addresses, not atomic on each word
196 inline static void conjoint_words_to_lower(const HeapWord* from, HeapWord* to, size_t byte_count) {
197 // byte_count is in bytes to check its alignment
198 assert_params_ok(from, to, HeapWordSize);
199 assert_byte_count_ok(byte_count, HeapWordSize);
200
201 size_t count = align_up(byte_count, HeapWordSize) >> LogHeapWordSize;
202 assert(to <= from || from + count <= to, "do not overwrite source data");
203
204 while (count-- > 0) {
205 *to++ = *from++;
206 }
207 }
208
209 // Copy word-aligned words from lower to higher addresses, not atomic on each word
210 inline static void conjoint_words_to_higher(const HeapWord* from, HeapWord* to, size_t byte_count) {
211 // byte_count is in bytes to check its alignment
212 assert_params_ok(from, to, HeapWordSize);
213 assert_byte_count_ok(byte_count, HeapWordSize);
214 if (byte_count == 0) return;
215
216 size_t count = align_up(byte_count, HeapWordSize) >> LogHeapWordSize;
217 assert(from <= to || to + count <= from, "do not overwrite source data");
218
219 from += count - 1;
220 to += count - 1;
221 while (count-- > 0) {
222 *to-- = *from--;
223 }
224 }
225
226 /**
227 * Copy elements
228 *
229 * @param src address of source
230 * @param dst address of destination
231 * @param byte_count number of bytes to copy
232 * @param elem_size size of the elements to copy-swap
233 */
234 static void conjoint_copy(const void* src, void* dst, size_t byte_count, size_t elem_size);
235
236 /**
237 * Copy and *unconditionally* byte swap elements
238 *
239 * @param src address of source
240 * @param dst address of destination
241 * @param byte_count number of bytes to copy
242 * @param elem_size size of the elements to copy-swap
243 */
244 static void conjoint_swap(const void* src, void* dst, size_t byte_count, size_t elem_size);
245
246 /**
247 * Copy and byte swap elements from the specified endian to the native (cpu) endian if needed (if they differ)
248 *
249 * @param src address of source
250 * @param dst address of destination
251 * @param byte_count number of bytes to copy
252 * @param elem_size size of the elements to copy-swap
253 */
254 template <Endian::Order endian>
255 static void conjoint_swap_if_needed(const void* src, void* dst, size_t byte_count, size_t elem_size) {
256 if (Endian::NATIVE != endian) {
257 conjoint_swap(src, dst, byte_count, elem_size);
258 } else {
259 conjoint_copy(src, dst, byte_count, elem_size);
260 }
261 }
262
263 // Fill methods
264
265 // Fill word-aligned words, not atomic on each word
266 // set_words
267 static void fill_to_words(HeapWord* to, size_t count, juint value = 0) {
268 assert_params_ok(to, HeapWordSize);
269 pd_fill_to_words(to, count, value);
270 }
271
272 static void fill_to_aligned_words(HeapWord* to, size_t count, juint value = 0) {
273 assert_params_aligned(to);
274 pd_fill_to_aligned_words(to, count, value);
275 }
276
277 // Fill bytes
278 static void fill_to_bytes(void* to, size_t count, jubyte value = 0) {
279 pd_fill_to_bytes(to, count, value);
280 }
281
282 // Fill a span of memory. If the span is an integral number of aligned
283 // longs, words, or ints, store to those units atomically.
284 // The largest atomic transfer unit is 8 bytes, or the largest power
285 // of two which divides both to and size, whichever is smaller.
286 static void fill_to_memory_atomic(void* to, size_t size, jubyte value = 0);
287
288 // Zero-fill methods
289
290 // Zero word-aligned words, not atomic on each word
291 static void zero_to_words(HeapWord* to, size_t count) {
292 assert_params_ok(to, HeapWordSize);
293 pd_zero_to_words(to, count);
294 }
295
296 // Zero bytes
297 static void zero_to_bytes(void* to, size_t count) {
298 pd_zero_to_bytes(to, count);
299 }
300
301 protected:
302 inline static void shared_disjoint_words_atomic(const HeapWord* from,
303 HeapWord* to, size_t count) {
304
305 switch (count) {
306 case 8: AtomicAccess::store(&to[7], AtomicAccess::load(&from[7]));
307 case 7: AtomicAccess::store(&to[6], AtomicAccess::load(&from[6]));
308 case 6: AtomicAccess::store(&to[5], AtomicAccess::load(&from[5]));
309 case 5: AtomicAccess::store(&to[4], AtomicAccess::load(&from[4]));
310 case 4: AtomicAccess::store(&to[3], AtomicAccess::load(&from[3]));
311 case 3: AtomicAccess::store(&to[2], AtomicAccess::load(&from[2]));
312 case 2: AtomicAccess::store(&to[1], AtomicAccess::load(&from[1]));
313 case 1: AtomicAccess::store(&to[0], AtomicAccess::load(&from[0]));
314 case 0: break;
315 default:
316 while (count-- > 0) {
317 AtomicAccess::store(to++, AtomicAccess::load(from++));
318 }
319 break;
320 }
321 }
322
323 private:
324 static bool params_disjoint(const HeapWord* from, HeapWord* to, size_t count) {
325 if (from < to) {
326 return pointer_delta(to, from) >= count;
327 }
328 return pointer_delta(from, to) >= count;
329 }
330
331 // These methods raise a fatal if they detect a problem.
332
333 static void assert_disjoint(const HeapWord* from, HeapWord* to, size_t count) {
334 assert(params_disjoint(from, to, count), "source and dest overlap");
335 }
336
337 static void assert_params_ok(const void* from, void* to, intptr_t alignment) {
338 assert(is_aligned(from, alignment), "must be aligned: " PTR_FORMAT, p2i(from));
339 assert(is_aligned(to, alignment), "must be aligned: " PTR_FORMAT, p2i(to));
340 }
341
342 static void assert_params_ok(HeapWord* to, intptr_t alignment) {
343 assert(is_aligned(to, alignment), "must be aligned: " PTR_FORMAT, p2i(to));
344 }
345
346 static void assert_params_aligned(const HeapWord* from, HeapWord* to) {
347 assert(is_aligned(from, BytesPerLong), "must be aligned: " PTR_FORMAT, p2i(from));
348 assert(is_aligned(to, BytesPerLong), "must be aligned: " PTR_FORMAT, p2i(to));
349 }
350
351 static void assert_params_aligned(HeapWord* to) {
352 assert(is_aligned(to, BytesPerLong), "must be aligned: " PTR_FORMAT, p2i(to));
353 }
354
355 static void assert_byte_count_ok(size_t byte_count, size_t unit_size) {
356 assert(is_aligned(byte_count, unit_size), "byte count must be aligned");
357 }
358
359 // Platform dependent implementations of the above methods.
360 #include CPU_HEADER(copy)
361
362 };
363
364 #endif // SHARE_UTILITIES_COPY_HPP