Mon, 26 Sep 2011 10:24:05 -0700
7081933: Use zeroing elimination optimization for large array
Summary: Don't zero new typeArray during runtime call if the allocation is followed by arraycopy into it.
Reviewed-by: twisti
1 /*
2 * Copyright (c) 1997, 2010, 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 */
25 #include "precompiled.hpp"
26 #include "assembler_sparc.inline.hpp"
27 #include "memory/resourceArea.hpp"
28 #include "runtime/java.hpp"
29 #include "runtime/stubCodeGenerator.hpp"
30 #include "vm_version_sparc.hpp"
31 #ifdef TARGET_OS_FAMILY_linux
32 # include "os_linux.inline.hpp"
33 #endif
34 #ifdef TARGET_OS_FAMILY_solaris
35 # include "os_solaris.inline.hpp"
36 #endif
38 int VM_Version::_features = VM_Version::unknown_m;
39 const char* VM_Version::_features_str = "";
41 void VM_Version::initialize() {
42 _features = determine_features();
43 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes();
44 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes();
45 PrefetchFieldsAhead = prefetch_fields_ahead();
47 assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 1, "invalid value");
48 if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0;
49 if( AllocatePrefetchInstr > 1 ) AllocatePrefetchInstr = 0;
51 // Allocation prefetch settings
52 intx cache_line_size = prefetch_data_size();
53 if( cache_line_size > AllocatePrefetchStepSize )
54 AllocatePrefetchStepSize = cache_line_size;
56 assert(AllocatePrefetchLines > 0, "invalid value");
57 if( AllocatePrefetchLines < 1 ) // set valid value in product VM
58 AllocatePrefetchLines = 3;
59 assert(AllocateInstancePrefetchLines > 0, "invalid value");
60 if( AllocateInstancePrefetchLines < 1 ) // set valid value in product VM
61 AllocateInstancePrefetchLines = 1;
63 AllocatePrefetchDistance = allocate_prefetch_distance();
64 AllocatePrefetchStyle = allocate_prefetch_style();
66 assert((AllocatePrefetchDistance % AllocatePrefetchStepSize) == 0 &&
67 (AllocatePrefetchDistance > 0), "invalid value");
68 if ((AllocatePrefetchDistance % AllocatePrefetchStepSize) != 0 ||
69 (AllocatePrefetchDistance <= 0)) {
70 AllocatePrefetchDistance = AllocatePrefetchStepSize;
71 }
73 if (AllocatePrefetchStyle == 3 && !has_blk_init()) {
74 warning("BIS instructions are not available on this CPU");
75 FLAG_SET_DEFAULT(AllocatePrefetchStyle, 1);
76 }
78 if (has_v9()) {
79 assert(ArraycopySrcPrefetchDistance < 4096, "invalid value");
80 if (ArraycopySrcPrefetchDistance >= 4096)
81 ArraycopySrcPrefetchDistance = 4064;
82 assert(ArraycopyDstPrefetchDistance < 4096, "invalid value");
83 if (ArraycopyDstPrefetchDistance >= 4096)
84 ArraycopyDstPrefetchDistance = 4064;
85 } else {
86 if (ArraycopySrcPrefetchDistance > 0) {
87 warning("prefetch instructions are not available on this CPU");
88 FLAG_SET_DEFAULT(ArraycopySrcPrefetchDistance, 0);
89 }
90 if (ArraycopyDstPrefetchDistance > 0) {
91 warning("prefetch instructions are not available on this CPU");
92 FLAG_SET_DEFAULT(ArraycopyDstPrefetchDistance, 0);
93 }
94 }
96 UseSSE = 0; // Only on x86 and x64
98 _supports_cx8 = has_v9();
100 if (is_niagara()) {
101 // Indirect branch is the same cost as direct
102 if (FLAG_IS_DEFAULT(UseInlineCaches)) {
103 FLAG_SET_DEFAULT(UseInlineCaches, false);
104 }
105 // Align loops on a single instruction boundary.
106 if (FLAG_IS_DEFAULT(OptoLoopAlignment)) {
107 FLAG_SET_DEFAULT(OptoLoopAlignment, 4);
108 }
109 // When using CMS, we cannot use memset() in BOT updates because
110 // the sun4v/CMT version in libc_psr uses BIS which exposes
111 // "phantom zeros" to concurrent readers. See 6948537.
112 if (FLAG_IS_DEFAULT(UseMemSetInBOT) && UseConcMarkSweepGC) {
113 FLAG_SET_DEFAULT(UseMemSetInBOT, false);
114 }
115 #ifdef _LP64
116 // 32-bit oops don't make sense for the 64-bit VM on sparc
117 // since the 32-bit VM has the same registers and smaller objects.
118 Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes);
119 #endif // _LP64
120 #ifdef COMPILER2
121 // Indirect branch is the same cost as direct
122 if (FLAG_IS_DEFAULT(UseJumpTables)) {
123 FLAG_SET_DEFAULT(UseJumpTables, true);
124 }
125 // Single-issue, so entry and loop tops are
126 // aligned on a single instruction boundary
127 if (FLAG_IS_DEFAULT(InteriorEntryAlignment)) {
128 FLAG_SET_DEFAULT(InteriorEntryAlignment, 4);
129 }
130 if (is_niagara_plus()) {
131 if (has_blk_init() && UseTLAB &&
132 FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
133 // Use BIS instruction for TLAB allocation prefetch.
134 FLAG_SET_ERGO(intx, AllocatePrefetchInstr, 1);
135 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
136 FLAG_SET_ERGO(intx, AllocatePrefetchStyle, 3);
137 }
138 if (FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
139 // Use smaller prefetch distance with BIS
140 FLAG_SET_DEFAULT(AllocatePrefetchDistance, 64);
141 }
142 }
143 if (is_T4()) {
144 // Double number of prefetched cache lines on T4
145 // since L2 cache line size is smaller (32 bytes).
146 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) {
147 FLAG_SET_ERGO(intx, AllocatePrefetchLines, AllocatePrefetchLines*2);
148 }
149 if (FLAG_IS_DEFAULT(AllocateInstancePrefetchLines)) {
150 FLAG_SET_ERGO(intx, AllocateInstancePrefetchLines, AllocateInstancePrefetchLines*2);
151 }
152 }
153 if (AllocatePrefetchStyle != 3 && FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
154 // Use different prefetch distance without BIS
155 FLAG_SET_DEFAULT(AllocatePrefetchDistance, 256);
156 }
157 if (AllocatePrefetchInstr == 1) {
158 // Need a space at the end of TLAB for BIS since it
159 // will fault when accessing memory outside of heap.
161 // +1 for rounding up to next cache line, +1 to be safe
162 int lines = AllocatePrefetchLines + 2;
163 int step_size = AllocatePrefetchStepSize;
164 int distance = AllocatePrefetchDistance;
165 _reserve_for_allocation_prefetch = (distance + step_size*lines)/(int)HeapWordSize;
166 }
167 }
168 #endif
169 }
171 // Use hardware population count instruction if available.
172 if (has_hardware_popc()) {
173 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
174 FLAG_SET_DEFAULT(UsePopCountInstruction, true);
175 }
176 } else if (UsePopCountInstruction) {
177 warning("POPC instruction is not available on this CPU");
178 FLAG_SET_DEFAULT(UsePopCountInstruction, false);
179 }
181 // T4 and newer Sparc cpus have new compare and branch instruction.
182 if (has_cbcond()) {
183 if (FLAG_IS_DEFAULT(UseCBCond)) {
184 FLAG_SET_DEFAULT(UseCBCond, true);
185 }
186 } else if (UseCBCond) {
187 warning("CBCOND instruction is not available on this CPU");
188 FLAG_SET_DEFAULT(UseCBCond, false);
189 }
191 assert(BlockZeroingLowLimit > 0, "invalid value");
192 if (has_block_zeroing()) {
193 if (FLAG_IS_DEFAULT(UseBlockZeroing)) {
194 FLAG_SET_DEFAULT(UseBlockZeroing, true);
195 }
196 } else if (UseBlockZeroing) {
197 warning("BIS zeroing instructions are not available on this CPU");
198 FLAG_SET_DEFAULT(UseBlockZeroing, false);
199 }
201 assert(BlockCopyLowLimit > 0, "invalid value");
202 if (has_block_zeroing()) { // has_blk_init() && is_T4(): core's local L2 cache
203 if (FLAG_IS_DEFAULT(UseBlockCopy)) {
204 FLAG_SET_DEFAULT(UseBlockCopy, true);
205 }
206 } else if (UseBlockCopy) {
207 warning("BIS instructions are not available or expensive on this CPU");
208 FLAG_SET_DEFAULT(UseBlockCopy, false);
209 }
211 #ifdef COMPILER2
212 // T4 and newer Sparc cpus have fast RDPC.
213 if (has_fast_rdpc() && FLAG_IS_DEFAULT(UseRDPCForConstantTableBase)) {
214 // FLAG_SET_DEFAULT(UseRDPCForConstantTableBase, true);
215 }
217 // Currently not supported anywhere.
218 FLAG_SET_DEFAULT(UseFPUForSpilling, false);
220 assert((InteriorEntryAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
221 #endif
223 assert((CodeEntryAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
224 assert((OptoLoopAlignment % relocInfo::addr_unit()) == 0, "alignment is not a multiple of NOP size");
226 char buf[512];
227 jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
228 (has_v9() ? ", v9" : (has_v8() ? ", v8" : "")),
229 (has_hardware_popc() ? ", popc" : ""),
230 (has_vis1() ? ", vis1" : ""),
231 (has_vis2() ? ", vis2" : ""),
232 (has_vis3() ? ", vis3" : ""),
233 (has_blk_init() ? ", blk_init" : ""),
234 (has_cbcond() ? ", cbcond" : ""),
235 (is_ultra3() ? ", ultra3" : ""),
236 (is_sun4v() ? ", sun4v" : ""),
237 (is_niagara_plus() ? ", niagara_plus" : (is_niagara() ? ", niagara" : "")),
238 (is_sparc64() ? ", sparc64" : ""),
239 (!has_hardware_mul32() ? ", no-mul32" : ""),
240 (!has_hardware_div32() ? ", no-div32" : ""),
241 (!has_hardware_fsmuld() ? ", no-fsmuld" : ""));
243 // buf is started with ", " or is empty
244 _features_str = strdup(strlen(buf) > 2 ? buf + 2 : buf);
246 // UseVIS is set to the smallest of what hardware supports and what
247 // the command line requires. I.e., you cannot set UseVIS to 3 on
248 // older UltraSparc which do not support it.
249 if (UseVIS > 3) UseVIS=3;
250 if (UseVIS < 0) UseVIS=0;
251 if (!has_vis3()) // Drop to 2 if no VIS3 support
252 UseVIS = MIN2((intx)2,UseVIS);
253 if (!has_vis2()) // Drop to 1 if no VIS2 support
254 UseVIS = MIN2((intx)1,UseVIS);
255 if (!has_vis1()) // Drop to 0 if no VIS1 support
256 UseVIS = 0;
258 #ifndef PRODUCT
259 if (PrintMiscellaneous && Verbose) {
260 tty->print("Allocation");
261 if (AllocatePrefetchStyle <= 0) {
262 tty->print_cr(": no prefetching");
263 } else {
264 tty->print(" prefetching: ");
265 if (AllocatePrefetchInstr == 0) {
266 tty->print("PREFETCH");
267 } else if (AllocatePrefetchInstr == 1) {
268 tty->print("BIS");
269 }
270 if (AllocatePrefetchLines > 1) {
271 tty->print_cr(" at distance %d, %d lines of %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize);
272 } else {
273 tty->print_cr(" at distance %d, one line of %d bytes", AllocatePrefetchDistance, AllocatePrefetchStepSize);
274 }
275 }
276 if (PrefetchCopyIntervalInBytes > 0) {
277 tty->print_cr("PrefetchCopyIntervalInBytes %d", PrefetchCopyIntervalInBytes);
278 }
279 if (PrefetchScanIntervalInBytes > 0) {
280 tty->print_cr("PrefetchScanIntervalInBytes %d", PrefetchScanIntervalInBytes);
281 }
282 if (PrefetchFieldsAhead > 0) {
283 tty->print_cr("PrefetchFieldsAhead %d", PrefetchFieldsAhead);
284 }
285 }
286 #endif // PRODUCT
287 }
289 void VM_Version::print_features() {
290 tty->print_cr("Version:%s", cpu_features());
291 }
293 int VM_Version::determine_features() {
294 if (UseV8InstrsOnly) {
295 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-V8");)
296 return generic_v8_m;
297 }
299 int features = platform_features(unknown_m); // platform_features() is os_arch specific
301 if (features == unknown_m) {
302 features = generic_v9_m;
303 warning("Cannot recognize SPARC version. Default to V9");
304 }
306 assert(is_T_family(features) == is_niagara(features), "Niagara should be T series");
307 if (UseNiagaraInstrs) { // Force code generation for Niagara
308 if (is_T_family(features)) {
309 // Happy to accomodate...
310 } else {
311 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-Niagara");)
312 features |= T_family_m;
313 }
314 } else {
315 if (is_T_family(features) && !FLAG_IS_DEFAULT(UseNiagaraInstrs)) {
316 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-Not-Niagara");)
317 features &= ~(T_family_m | T1_model_m);
318 } else {
319 // Happy to accomodate...
320 }
321 }
323 return features;
324 }
326 static int saved_features = 0;
328 void VM_Version::allow_all() {
329 saved_features = _features;
330 _features = all_features_m;
331 }
333 void VM_Version::revert() {
334 _features = saved_features;
335 }
337 unsigned int VM_Version::calc_parallel_worker_threads() {
338 unsigned int result;
339 if (is_niagara_plus()) {
340 result = nof_parallel_worker_threads(5, 16, 8);
341 } else {
342 result = nof_parallel_worker_threads(5, 8, 8);
343 }
344 return result;
345 }