Sun, 01 Jan 2012 11:17:59 -0500
7125934: Add a fast unordered timestamp capability to Hotspot on x86/x64
Summary: Add rdtsc detection and inline generation.
Reviewed-by: kamg, dholmes
Contributed-by: karen.kinnear@oracle.com
1 /*
2 * Copyright (c) 1997, 2011, 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_x86.inline.hpp"
27 #include "memory/resourceArea.hpp"
28 #include "runtime/java.hpp"
29 #include "runtime/stubCodeGenerator.hpp"
30 #include "vm_version_x86.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
37 #ifdef TARGET_OS_FAMILY_windows
38 # include "os_windows.inline.hpp"
39 #endif
40 #ifdef TARGET_OS_FAMILY_bsd
41 # include "os_bsd.inline.hpp"
42 #endif
45 int VM_Version::_cpu;
46 int VM_Version::_model;
47 int VM_Version::_stepping;
48 int VM_Version::_cpuFeatures;
49 const char* VM_Version::_features_str = "";
50 VM_Version::CpuidInfo VM_Version::_cpuid_info = { 0, };
52 static BufferBlob* stub_blob;
53 static const int stub_size = 500;
55 extern "C" {
56 typedef void (*getPsrInfo_stub_t)(void*);
57 }
58 static getPsrInfo_stub_t getPsrInfo_stub = NULL;
61 class VM_Version_StubGenerator: public StubCodeGenerator {
62 public:
64 VM_Version_StubGenerator(CodeBuffer *c) : StubCodeGenerator(c) {}
66 address generate_getPsrInfo() {
67 // Flags to test CPU type.
68 const uint32_t EFL_AC = 0x40000;
69 const uint32_t EFL_ID = 0x200000;
70 // Values for when we don't have a CPUID instruction.
71 const int CPU_FAMILY_SHIFT = 8;
72 const uint32_t CPU_FAMILY_386 = (3 << CPU_FAMILY_SHIFT);
73 const uint32_t CPU_FAMILY_486 = (4 << CPU_FAMILY_SHIFT);
75 Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4;
76 Label ext_cpuid1, ext_cpuid5, ext_cpuid7, done;
78 StubCodeMark mark(this, "VM_Version", "getPsrInfo_stub");
79 # define __ _masm->
81 address start = __ pc();
83 //
84 // void getPsrInfo(VM_Version::CpuidInfo* cpuid_info);
85 //
86 // LP64: rcx and rdx are first and second argument registers on windows
88 __ push(rbp);
89 #ifdef _LP64
90 __ mov(rbp, c_rarg0); // cpuid_info address
91 #else
92 __ movptr(rbp, Address(rsp, 8)); // cpuid_info address
93 #endif
94 __ push(rbx);
95 __ push(rsi);
96 __ pushf(); // preserve rbx, and flags
97 __ pop(rax);
98 __ push(rax);
99 __ mov(rcx, rax);
100 //
101 // if we are unable to change the AC flag, we have a 386
102 //
103 __ xorl(rax, EFL_AC);
104 __ push(rax);
105 __ popf();
106 __ pushf();
107 __ pop(rax);
108 __ cmpptr(rax, rcx);
109 __ jccb(Assembler::notEqual, detect_486);
111 __ movl(rax, CPU_FAMILY_386);
112 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
113 __ jmp(done);
115 //
116 // If we are unable to change the ID flag, we have a 486 which does
117 // not support the "cpuid" instruction.
118 //
119 __ bind(detect_486);
120 __ mov(rax, rcx);
121 __ xorl(rax, EFL_ID);
122 __ push(rax);
123 __ popf();
124 __ pushf();
125 __ pop(rax);
126 __ cmpptr(rcx, rax);
127 __ jccb(Assembler::notEqual, detect_586);
129 __ bind(cpu486);
130 __ movl(rax, CPU_FAMILY_486);
131 __ movl(Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())), rax);
132 __ jmp(done);
134 //
135 // At this point, we have a chip which supports the "cpuid" instruction
136 //
137 __ bind(detect_586);
138 __ xorl(rax, rax);
139 __ cpuid();
140 __ orl(rax, rax);
141 __ jcc(Assembler::equal, cpu486); // if cpuid doesn't support an input
142 // value of at least 1, we give up and
143 // assume a 486
144 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset())));
145 __ movl(Address(rsi, 0), rax);
146 __ movl(Address(rsi, 4), rbx);
147 __ movl(Address(rsi, 8), rcx);
148 __ movl(Address(rsi,12), rdx);
150 __ cmpl(rax, 0xa); // Is cpuid(0xB) supported?
151 __ jccb(Assembler::belowEqual, std_cpuid4);
153 //
154 // cpuid(0xB) Processor Topology
155 //
156 __ movl(rax, 0xb);
157 __ xorl(rcx, rcx); // Threads level
158 __ cpuid();
160 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB0_offset())));
161 __ movl(Address(rsi, 0), rax);
162 __ movl(Address(rsi, 4), rbx);
163 __ movl(Address(rsi, 8), rcx);
164 __ movl(Address(rsi,12), rdx);
166 __ movl(rax, 0xb);
167 __ movl(rcx, 1); // Cores level
168 __ cpuid();
169 __ push(rax);
170 __ andl(rax, 0x1f); // Determine if valid topology level
171 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level
172 __ andl(rax, 0xffff);
173 __ pop(rax);
174 __ jccb(Assembler::equal, std_cpuid4);
176 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB1_offset())));
177 __ movl(Address(rsi, 0), rax);
178 __ movl(Address(rsi, 4), rbx);
179 __ movl(Address(rsi, 8), rcx);
180 __ movl(Address(rsi,12), rdx);
182 __ movl(rax, 0xb);
183 __ movl(rcx, 2); // Packages level
184 __ cpuid();
185 __ push(rax);
186 __ andl(rax, 0x1f); // Determine if valid topology level
187 __ orl(rax, rbx); // eax[4:0] | ebx[0:15] == 0 indicates invalid level
188 __ andl(rax, 0xffff);
189 __ pop(rax);
190 __ jccb(Assembler::equal, std_cpuid4);
192 __ lea(rsi, Address(rbp, in_bytes(VM_Version::tpl_cpuidB2_offset())));
193 __ movl(Address(rsi, 0), rax);
194 __ movl(Address(rsi, 4), rbx);
195 __ movl(Address(rsi, 8), rcx);
196 __ movl(Address(rsi,12), rdx);
198 //
199 // cpuid(0x4) Deterministic cache params
200 //
201 __ bind(std_cpuid4);
202 __ movl(rax, 4);
203 __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x4) supported?
204 __ jccb(Assembler::greater, std_cpuid1);
206 __ xorl(rcx, rcx); // L1 cache
207 __ cpuid();
208 __ push(rax);
209 __ andl(rax, 0x1f); // Determine if valid cache parameters used
210 __ orl(rax, rax); // eax[4:0] == 0 indicates invalid cache
211 __ pop(rax);
212 __ jccb(Assembler::equal, std_cpuid1);
214 __ lea(rsi, Address(rbp, in_bytes(VM_Version::dcp_cpuid4_offset())));
215 __ movl(Address(rsi, 0), rax);
216 __ movl(Address(rsi, 4), rbx);
217 __ movl(Address(rsi, 8), rcx);
218 __ movl(Address(rsi,12), rdx);
220 //
221 // Standard cpuid(0x1)
222 //
223 __ bind(std_cpuid1);
224 __ movl(rax, 1);
225 __ cpuid();
226 __ lea(rsi, Address(rbp, in_bytes(VM_Version::std_cpuid1_offset())));
227 __ movl(Address(rsi, 0), rax);
228 __ movl(Address(rsi, 4), rbx);
229 __ movl(Address(rsi, 8), rcx);
230 __ movl(Address(rsi,12), rdx);
232 __ movl(rax, 0x80000000);
233 __ cpuid();
234 __ cmpl(rax, 0x80000000); // Is cpuid(0x80000001) supported?
235 __ jcc(Assembler::belowEqual, done);
236 __ cmpl(rax, 0x80000004); // Is cpuid(0x80000005) supported?
237 __ jccb(Assembler::belowEqual, ext_cpuid1);
238 __ cmpl(rax, 0x80000006); // Is cpuid(0x80000007) supported?
239 __ jccb(Assembler::belowEqual, ext_cpuid5);
240 __ cmpl(rax, 0x80000007); // Is cpuid(0x80000008) supported?
241 __ jccb(Assembler::belowEqual, ext_cpuid7);
242 //
243 // Extended cpuid(0x80000008)
244 //
245 __ movl(rax, 0x80000008);
246 __ cpuid();
247 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid8_offset())));
248 __ movl(Address(rsi, 0), rax);
249 __ movl(Address(rsi, 4), rbx);
250 __ movl(Address(rsi, 8), rcx);
251 __ movl(Address(rsi,12), rdx);
253 //
254 // Extended cpuid(0x80000007)
255 //
256 __ bind(ext_cpuid7);
257 __ movl(rax, 0x80000007);
258 __ cpuid();
259 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid7_offset())));
260 __ movl(Address(rsi, 0), rax);
261 __ movl(Address(rsi, 4), rbx);
262 __ movl(Address(rsi, 8), rcx);
263 __ movl(Address(rsi,12), rdx);
265 //
266 // Extended cpuid(0x80000005)
267 //
268 __ bind(ext_cpuid5);
269 __ movl(rax, 0x80000005);
270 __ cpuid();
271 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid5_offset())));
272 __ movl(Address(rsi, 0), rax);
273 __ movl(Address(rsi, 4), rbx);
274 __ movl(Address(rsi, 8), rcx);
275 __ movl(Address(rsi,12), rdx);
277 //
278 // Extended cpuid(0x80000001)
279 //
280 __ bind(ext_cpuid1);
281 __ movl(rax, 0x80000001);
282 __ cpuid();
283 __ lea(rsi, Address(rbp, in_bytes(VM_Version::ext_cpuid1_offset())));
284 __ movl(Address(rsi, 0), rax);
285 __ movl(Address(rsi, 4), rbx);
286 __ movl(Address(rsi, 8), rcx);
287 __ movl(Address(rsi,12), rdx);
289 //
290 // return
291 //
292 __ bind(done);
293 __ popf();
294 __ pop(rsi);
295 __ pop(rbx);
296 __ pop(rbp);
297 __ ret(0);
299 # undef __
301 return start;
302 };
303 };
306 void VM_Version::get_processor_features() {
308 _cpu = 4; // 486 by default
309 _model = 0;
310 _stepping = 0;
311 _cpuFeatures = 0;
312 _logical_processors_per_package = 1;
314 if (!Use486InstrsOnly) {
315 // Get raw processor info
316 getPsrInfo_stub(&_cpuid_info);
317 assert_is_initialized();
318 _cpu = extended_cpu_family();
319 _model = extended_cpu_model();
320 _stepping = cpu_stepping();
322 if (cpu_family() > 4) { // it supports CPUID
323 _cpuFeatures = feature_flags();
324 // Logical processors are only available on P4s and above,
325 // and only if hyperthreading is available.
326 _logical_processors_per_package = logical_processor_count();
327 }
328 }
330 _supports_cx8 = supports_cmpxchg8();
332 #ifdef _LP64
333 // OS should support SSE for x64 and hardware should support at least SSE2.
334 if (!VM_Version::supports_sse2()) {
335 vm_exit_during_initialization("Unknown x64 processor: SSE2 not supported");
336 }
337 // in 64 bit the use of SSE2 is the minimum
338 if (UseSSE < 2) UseSSE = 2;
339 #endif
341 #ifdef AMD64
342 // flush_icache_stub have to be generated first.
343 // That is why Icache line size is hard coded in ICache class,
344 // see icache_x86.hpp. It is also the reason why we can't use
345 // clflush instruction in 32-bit VM since it could be running
346 // on CPU which does not support it.
347 //
348 // The only thing we can do is to verify that flushed
349 // ICache::line_size has correct value.
350 guarantee(_cpuid_info.std_cpuid1_edx.bits.clflush != 0, "clflush is not supported");
351 // clflush_size is size in quadwords (8 bytes).
352 guarantee(_cpuid_info.std_cpuid1_ebx.bits.clflush_size == 8, "such clflush size is not supported");
353 #endif
355 // If the OS doesn't support SSE, we can't use this feature even if the HW does
356 if (!os::supports_sse())
357 _cpuFeatures &= ~(CPU_SSE|CPU_SSE2|CPU_SSE3|CPU_SSSE3|CPU_SSE4A|CPU_SSE4_1|CPU_SSE4_2);
359 if (UseSSE < 4) {
360 _cpuFeatures &= ~CPU_SSE4_1;
361 _cpuFeatures &= ~CPU_SSE4_2;
362 }
364 if (UseSSE < 3) {
365 _cpuFeatures &= ~CPU_SSE3;
366 _cpuFeatures &= ~CPU_SSSE3;
367 _cpuFeatures &= ~CPU_SSE4A;
368 }
370 if (UseSSE < 2)
371 _cpuFeatures &= ~CPU_SSE2;
373 if (UseSSE < 1)
374 _cpuFeatures &= ~CPU_SSE;
376 if (logical_processors_per_package() == 1) {
377 // HT processor could be installed on a system which doesn't support HT.
378 _cpuFeatures &= ~CPU_HT;
379 }
381 char buf[256];
382 jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
383 cores_per_cpu(), threads_per_core(),
384 cpu_family(), _model, _stepping,
385 (supports_cmov() ? ", cmov" : ""),
386 (supports_cmpxchg8() ? ", cx8" : ""),
387 (supports_fxsr() ? ", fxsr" : ""),
388 (supports_mmx() ? ", mmx" : ""),
389 (supports_sse() ? ", sse" : ""),
390 (supports_sse2() ? ", sse2" : ""),
391 (supports_sse3() ? ", sse3" : ""),
392 (supports_ssse3()? ", ssse3": ""),
393 (supports_sse4_1() ? ", sse4.1" : ""),
394 (supports_sse4_2() ? ", sse4.2" : ""),
395 (supports_popcnt() ? ", popcnt" : ""),
396 (supports_mmx_ext() ? ", mmxext" : ""),
397 (supports_3dnow_prefetch() ? ", 3dnowpref" : ""),
398 (supports_lzcnt() ? ", lzcnt": ""),
399 (supports_sse4a() ? ", sse4a": ""),
400 (supports_ht() ? ", ht": ""),
401 (supports_tsc() ? ", tsc": ""),
402 (supports_tscinv_bit() ? ", tscinvbit": ""),
403 (supports_tscinv() ? ", tscinv": ""));
404 _features_str = strdup(buf);
406 // UseSSE is set to the smaller of what hardware supports and what
407 // the command line requires. I.e., you cannot set UseSSE to 2 on
408 // older Pentiums which do not support it.
409 if( UseSSE > 4 ) UseSSE=4;
410 if( UseSSE < 0 ) UseSSE=0;
411 if( !supports_sse4_1() ) // Drop to 3 if no SSE4 support
412 UseSSE = MIN2((intx)3,UseSSE);
413 if( !supports_sse3() ) // Drop to 2 if no SSE3 support
414 UseSSE = MIN2((intx)2,UseSSE);
415 if( !supports_sse2() ) // Drop to 1 if no SSE2 support
416 UseSSE = MIN2((intx)1,UseSSE);
417 if( !supports_sse () ) // Drop to 0 if no SSE support
418 UseSSE = 0;
420 // On new cpus instructions which update whole XMM register should be used
421 // to prevent partial register stall due to dependencies on high half.
422 //
423 // UseXmmLoadAndClearUpper == true --> movsd(xmm, mem)
424 // UseXmmLoadAndClearUpper == false --> movlpd(xmm, mem)
425 // UseXmmRegToRegMoveAll == true --> movaps(xmm, xmm), movapd(xmm, xmm).
426 // UseXmmRegToRegMoveAll == false --> movss(xmm, xmm), movsd(xmm, xmm).
428 if( is_amd() ) { // AMD cpus specific settings
429 if( supports_sse2() && FLAG_IS_DEFAULT(UseAddressNop) ) {
430 // Use it on new AMD cpus starting from Opteron.
431 UseAddressNop = true;
432 }
433 if( supports_sse2() && FLAG_IS_DEFAULT(UseNewLongLShift) ) {
434 // Use it on new AMD cpus starting from Opteron.
435 UseNewLongLShift = true;
436 }
437 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
438 if( supports_sse4a() ) {
439 UseXmmLoadAndClearUpper = true; // use movsd only on '10h' Opteron
440 } else {
441 UseXmmLoadAndClearUpper = false;
442 }
443 }
444 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
445 if( supports_sse4a() ) {
446 UseXmmRegToRegMoveAll = true; // use movaps, movapd only on '10h'
447 } else {
448 UseXmmRegToRegMoveAll = false;
449 }
450 }
451 if( FLAG_IS_DEFAULT(UseXmmI2F) ) {
452 if( supports_sse4a() ) {
453 UseXmmI2F = true;
454 } else {
455 UseXmmI2F = false;
456 }
457 }
458 if( FLAG_IS_DEFAULT(UseXmmI2D) ) {
459 if( supports_sse4a() ) {
460 UseXmmI2D = true;
461 } else {
462 UseXmmI2D = false;
463 }
464 }
465 if( FLAG_IS_DEFAULT(UseSSE42Intrinsics) ) {
466 if( supports_sse4_2() && UseSSE >= 4 ) {
467 UseSSE42Intrinsics = true;
468 }
469 }
471 // Use count leading zeros count instruction if available.
472 if (supports_lzcnt()) {
473 if (FLAG_IS_DEFAULT(UseCountLeadingZerosInstruction)) {
474 UseCountLeadingZerosInstruction = true;
475 }
476 }
478 // some defaults for AMD family 15h
479 if ( cpu_family() == 0x15 ) {
480 // On family 15h processors default is no sw prefetch
481 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
482 AllocatePrefetchStyle = 0;
483 }
484 // Also, if some other prefetch style is specified, default instruction type is PREFETCHW
485 if (FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
486 AllocatePrefetchInstr = 3;
487 }
488 // On family 15h processors use XMM and UnalignedLoadStores for Array Copy
489 if( FLAG_IS_DEFAULT(UseXMMForArrayCopy) ) {
490 UseXMMForArrayCopy = true;
491 }
492 if( FLAG_IS_DEFAULT(UseUnalignedLoadStores) && UseXMMForArrayCopy ) {
493 UseUnalignedLoadStores = true;
494 }
495 }
497 }
499 if( is_intel() ) { // Intel cpus specific settings
500 if( FLAG_IS_DEFAULT(UseStoreImmI16) ) {
501 UseStoreImmI16 = false; // don't use it on Intel cpus
502 }
503 if( cpu_family() == 6 || cpu_family() == 15 ) {
504 if( FLAG_IS_DEFAULT(UseAddressNop) ) {
505 // Use it on all Intel cpus starting from PentiumPro
506 UseAddressNop = true;
507 }
508 }
509 if( FLAG_IS_DEFAULT(UseXmmLoadAndClearUpper) ) {
510 UseXmmLoadAndClearUpper = true; // use movsd on all Intel cpus
511 }
512 if( FLAG_IS_DEFAULT(UseXmmRegToRegMoveAll) ) {
513 if( supports_sse3() ) {
514 UseXmmRegToRegMoveAll = true; // use movaps, movapd on new Intel cpus
515 } else {
516 UseXmmRegToRegMoveAll = false;
517 }
518 }
519 if( cpu_family() == 6 && supports_sse3() ) { // New Intel cpus
520 #ifdef COMPILER2
521 if( FLAG_IS_DEFAULT(MaxLoopPad) ) {
522 // For new Intel cpus do the next optimization:
523 // don't align the beginning of a loop if there are enough instructions
524 // left (NumberOfLoopInstrToAlign defined in c2_globals.hpp)
525 // in current fetch line (OptoLoopAlignment) or the padding
526 // is big (> MaxLoopPad).
527 // Set MaxLoopPad to 11 for new Intel cpus to reduce number of
528 // generated NOP instructions. 11 is the largest size of one
529 // address NOP instruction '0F 1F' (see Assembler::nop(i)).
530 MaxLoopPad = 11;
531 }
532 #endif // COMPILER2
533 if( FLAG_IS_DEFAULT(UseXMMForArrayCopy) ) {
534 UseXMMForArrayCopy = true; // use SSE2 movq on new Intel cpus
535 }
536 if( supports_sse4_2() && supports_ht() ) { // Newest Intel cpus
537 if( FLAG_IS_DEFAULT(UseUnalignedLoadStores) && UseXMMForArrayCopy ) {
538 UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
539 }
540 }
541 if( supports_sse4_2() && UseSSE >= 4 ) {
542 if( FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
543 UseSSE42Intrinsics = true;
544 }
545 }
546 }
547 }
549 // Use population count instruction if available.
550 if (supports_popcnt()) {
551 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
552 UsePopCountInstruction = true;
553 }
554 }
556 #ifdef COMPILER2
557 if (UseFPUForSpilling) {
558 if (UseSSE < 2) {
559 // Only supported with SSE2+
560 FLAG_SET_DEFAULT(UseFPUForSpilling, false);
561 }
562 }
563 #endif
565 assert(0 <= ReadPrefetchInstr && ReadPrefetchInstr <= 3, "invalid value");
566 assert(0 <= AllocatePrefetchInstr && AllocatePrefetchInstr <= 3, "invalid value");
568 // set valid Prefetch instruction
569 if( ReadPrefetchInstr < 0 ) ReadPrefetchInstr = 0;
570 if( ReadPrefetchInstr > 3 ) ReadPrefetchInstr = 3;
571 if( ReadPrefetchInstr == 3 && !supports_3dnow_prefetch() ) ReadPrefetchInstr = 0;
572 if( !supports_sse() && supports_3dnow_prefetch() ) ReadPrefetchInstr = 3;
574 if( AllocatePrefetchInstr < 0 ) AllocatePrefetchInstr = 0;
575 if( AllocatePrefetchInstr > 3 ) AllocatePrefetchInstr = 3;
576 if( AllocatePrefetchInstr == 3 && !supports_3dnow_prefetch() ) AllocatePrefetchInstr=0;
577 if( !supports_sse() && supports_3dnow_prefetch() ) AllocatePrefetchInstr = 3;
579 // Allocation prefetch settings
580 intx cache_line_size = prefetch_data_size();
581 if( cache_line_size > AllocatePrefetchStepSize )
582 AllocatePrefetchStepSize = cache_line_size;
584 assert(AllocatePrefetchLines > 0, "invalid value");
585 if( AllocatePrefetchLines < 1 ) // set valid value in product VM
586 AllocatePrefetchLines = 3;
587 assert(AllocateInstancePrefetchLines > 0, "invalid value");
588 if( AllocateInstancePrefetchLines < 1 ) // set valid value in product VM
589 AllocateInstancePrefetchLines = 1;
591 AllocatePrefetchDistance = allocate_prefetch_distance();
592 AllocatePrefetchStyle = allocate_prefetch_style();
594 if( is_intel() && cpu_family() == 6 && supports_sse3() ) {
595 if( AllocatePrefetchStyle == 2 ) { // watermark prefetching on Core
596 #ifdef _LP64
597 AllocatePrefetchDistance = 384;
598 #else
599 AllocatePrefetchDistance = 320;
600 #endif
601 }
602 if( supports_sse4_2() && supports_ht() ) { // Nehalem based cpus
603 AllocatePrefetchDistance = 192;
604 AllocatePrefetchLines = 4;
605 #ifdef COMPILER2
606 if (AggressiveOpts && FLAG_IS_DEFAULT(UseFPUForSpilling)) {
607 FLAG_SET_DEFAULT(UseFPUForSpilling, true);
608 }
609 #endif
610 }
611 }
612 assert(AllocatePrefetchDistance % AllocatePrefetchStepSize == 0, "invalid value");
614 #ifdef _LP64
615 // Prefetch settings
616 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes();
617 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes();
618 PrefetchFieldsAhead = prefetch_fields_ahead();
619 #endif
621 #ifndef PRODUCT
622 if (PrintMiscellaneous && Verbose) {
623 tty->print_cr("Logical CPUs per core: %u",
624 logical_processors_per_package());
625 tty->print_cr("UseSSE=%d",UseSSE);
626 tty->print("Allocation");
627 if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow_prefetch()) {
628 tty->print_cr(": no prefetching");
629 } else {
630 tty->print(" prefetching: ");
631 if (UseSSE == 0 && supports_3dnow_prefetch()) {
632 tty->print("PREFETCHW");
633 } else if (UseSSE >= 1) {
634 if (AllocatePrefetchInstr == 0) {
635 tty->print("PREFETCHNTA");
636 } else if (AllocatePrefetchInstr == 1) {
637 tty->print("PREFETCHT0");
638 } else if (AllocatePrefetchInstr == 2) {
639 tty->print("PREFETCHT2");
640 } else if (AllocatePrefetchInstr == 3) {
641 tty->print("PREFETCHW");
642 }
643 }
644 if (AllocatePrefetchLines > 1) {
645 tty->print_cr(" at distance %d, %d lines of %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize);
646 } else {
647 tty->print_cr(" at distance %d, one line of %d bytes", AllocatePrefetchDistance, AllocatePrefetchStepSize);
648 }
649 }
651 if (PrefetchCopyIntervalInBytes > 0) {
652 tty->print_cr("PrefetchCopyIntervalInBytes %d", PrefetchCopyIntervalInBytes);
653 }
654 if (PrefetchScanIntervalInBytes > 0) {
655 tty->print_cr("PrefetchScanIntervalInBytes %d", PrefetchScanIntervalInBytes);
656 }
657 if (PrefetchFieldsAhead > 0) {
658 tty->print_cr("PrefetchFieldsAhead %d", PrefetchFieldsAhead);
659 }
660 }
661 #endif // !PRODUCT
662 }
664 void VM_Version::initialize() {
665 ResourceMark rm;
666 // Making this stub must be FIRST use of assembler
668 stub_blob = BufferBlob::create("getPsrInfo_stub", stub_size);
669 if (stub_blob == NULL) {
670 vm_exit_during_initialization("Unable to allocate getPsrInfo_stub");
671 }
672 CodeBuffer c(stub_blob);
673 VM_Version_StubGenerator g(&c);
674 getPsrInfo_stub = CAST_TO_FN_PTR(getPsrInfo_stub_t,
675 g.generate_getPsrInfo());
677 get_processor_features();
678 }