Thu, 03 Jan 2013 15:09:55 -0800
8005522: use fast-string instructions on x86 for zeroing
Summary: use 'rep stosb' instead of 'rep stosq' when fast-string operations are available.
Reviewed-by: twisti, roland
1 /*
2 * Copyright (c) 1997, 2012, 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 #ifndef CPU_X86_VM_VM_VERSION_X86_HPP
26 #define CPU_X86_VM_VM_VERSION_X86_HPP
28 #include "runtime/globals_extension.hpp"
29 #include "runtime/vm_version.hpp"
31 class VM_Version : public Abstract_VM_Version {
32 public:
33 // cpuid result register layouts. These are all unions of a uint32_t
34 // (in case anyone wants access to the register as a whole) and a bitfield.
36 union StdCpuid1Eax {
37 uint32_t value;
38 struct {
39 uint32_t stepping : 4,
40 model : 4,
41 family : 4,
42 proc_type : 2,
43 : 2,
44 ext_model : 4,
45 ext_family : 8,
46 : 4;
47 } bits;
48 };
50 union StdCpuid1Ebx { // example, unused
51 uint32_t value;
52 struct {
53 uint32_t brand_id : 8,
54 clflush_size : 8,
55 threads_per_cpu : 8,
56 apic_id : 8;
57 } bits;
58 };
60 union StdCpuid1Ecx {
61 uint32_t value;
62 struct {
63 uint32_t sse3 : 1,
64 : 2,
65 monitor : 1,
66 : 1,
67 vmx : 1,
68 : 1,
69 est : 1,
70 : 1,
71 ssse3 : 1,
72 cid : 1,
73 : 2,
74 cmpxchg16: 1,
75 : 4,
76 dca : 1,
77 sse4_1 : 1,
78 sse4_2 : 1,
79 : 2,
80 popcnt : 1,
81 : 1,
82 aes : 1,
83 : 1,
84 osxsave : 1,
85 avx : 1,
86 : 3;
87 } bits;
88 };
90 union StdCpuid1Edx {
91 uint32_t value;
92 struct {
93 uint32_t : 4,
94 tsc : 1,
95 : 3,
96 cmpxchg8 : 1,
97 : 6,
98 cmov : 1,
99 : 3,
100 clflush : 1,
101 : 3,
102 mmx : 1,
103 fxsr : 1,
104 sse : 1,
105 sse2 : 1,
106 : 1,
107 ht : 1,
108 : 3;
109 } bits;
110 };
112 union DcpCpuid4Eax {
113 uint32_t value;
114 struct {
115 uint32_t cache_type : 5,
116 : 21,
117 cores_per_cpu : 6;
118 } bits;
119 };
121 union DcpCpuid4Ebx {
122 uint32_t value;
123 struct {
124 uint32_t L1_line_size : 12,
125 partitions : 10,
126 associativity : 10;
127 } bits;
128 };
130 union TplCpuidBEbx {
131 uint32_t value;
132 struct {
133 uint32_t logical_cpus : 16,
134 : 16;
135 } bits;
136 };
138 union ExtCpuid1Ecx {
139 uint32_t value;
140 struct {
141 uint32_t LahfSahf : 1,
142 CmpLegacy : 1,
143 : 4,
144 lzcnt : 1,
145 sse4a : 1,
146 misalignsse : 1,
147 prefetchw : 1,
148 : 22;
149 } bits;
150 };
152 union ExtCpuid1Edx {
153 uint32_t value;
154 struct {
155 uint32_t : 22,
156 mmx_amd : 1,
157 mmx : 1,
158 fxsr : 1,
159 : 4,
160 long_mode : 1,
161 tdnow2 : 1,
162 tdnow : 1;
163 } bits;
164 };
166 union ExtCpuid5Ex {
167 uint32_t value;
168 struct {
169 uint32_t L1_line_size : 8,
170 L1_tag_lines : 8,
171 L1_assoc : 8,
172 L1_size : 8;
173 } bits;
174 };
176 union ExtCpuid7Edx {
177 uint32_t value;
178 struct {
179 uint32_t : 8,
180 tsc_invariance : 1,
181 : 23;
182 } bits;
183 };
185 union ExtCpuid8Ecx {
186 uint32_t value;
187 struct {
188 uint32_t cores_per_cpu : 8,
189 : 24;
190 } bits;
191 };
193 union SefCpuid7Eax {
194 uint32_t value;
195 };
197 union SefCpuid7Ebx {
198 uint32_t value;
199 struct {
200 uint32_t fsgsbase : 1,
201 : 2,
202 bmi1 : 1,
203 : 1,
204 avx2 : 1,
205 : 2,
206 bmi2 : 1,
207 erms : 1,
208 : 22;
209 } bits;
210 };
212 union XemXcr0Eax {
213 uint32_t value;
214 struct {
215 uint32_t x87 : 1,
216 sse : 1,
217 ymm : 1,
218 : 29;
219 } bits;
220 };
222 protected:
223 static int _cpu;
224 static int _model;
225 static int _stepping;
226 static int _cpuFeatures; // features returned by the "cpuid" instruction
227 // 0 if this instruction is not available
228 static const char* _features_str;
230 enum {
231 CPU_CX8 = (1 << 0), // next bits are from cpuid 1 (EDX)
232 CPU_CMOV = (1 << 1),
233 CPU_FXSR = (1 << 2),
234 CPU_HT = (1 << 3),
235 CPU_MMX = (1 << 4),
236 CPU_3DNOW_PREFETCH = (1 << 5), // Processor supports 3dnow prefetch and prefetchw instructions
237 // may not necessarily support other 3dnow instructions
238 CPU_SSE = (1 << 6),
239 CPU_SSE2 = (1 << 7),
240 CPU_SSE3 = (1 << 8), // SSE3 comes from cpuid 1 (ECX)
241 CPU_SSSE3 = (1 << 9),
242 CPU_SSE4A = (1 << 10),
243 CPU_SSE4_1 = (1 << 11),
244 CPU_SSE4_2 = (1 << 12),
245 CPU_POPCNT = (1 << 13),
246 CPU_LZCNT = (1 << 14),
247 CPU_TSC = (1 << 15),
248 CPU_TSCINV = (1 << 16),
249 CPU_AVX = (1 << 17),
250 CPU_AVX2 = (1 << 18),
251 CPU_AES = (1 << 19),
252 CPU_ERMS = (1 << 20) // enhanced 'rep movsb/stosb' instructions
253 } cpuFeatureFlags;
255 enum {
256 // AMD
257 CPU_FAMILY_AMD_11H = 0x11,
258 // Intel
259 CPU_FAMILY_INTEL_CORE = 6,
260 CPU_MODEL_NEHALEM = 0x1e,
261 CPU_MODEL_NEHALEM_EP = 0x1a,
262 CPU_MODEL_NEHALEM_EX = 0x2e,
263 CPU_MODEL_WESTMERE = 0x25,
264 CPU_MODEL_WESTMERE_EP = 0x2c,
265 CPU_MODEL_WESTMERE_EX = 0x2f,
266 CPU_MODEL_SANDYBRIDGE = 0x2a,
267 CPU_MODEL_SANDYBRIDGE_EP = 0x2d,
268 CPU_MODEL_IVYBRIDGE_EP = 0x3a
269 } cpuExtendedFamily;
271 // cpuid information block. All info derived from executing cpuid with
272 // various function numbers is stored here. Intel and AMD info is
273 // merged in this block: accessor methods disentangle it.
274 //
275 // The info block is laid out in subblocks of 4 dwords corresponding to
276 // eax, ebx, ecx and edx, whether or not they contain anything useful.
277 struct CpuidInfo {
278 // cpuid function 0
279 uint32_t std_max_function;
280 uint32_t std_vendor_name_0;
281 uint32_t std_vendor_name_1;
282 uint32_t std_vendor_name_2;
284 // cpuid function 1
285 StdCpuid1Eax std_cpuid1_eax;
286 StdCpuid1Ebx std_cpuid1_ebx;
287 StdCpuid1Ecx std_cpuid1_ecx;
288 StdCpuid1Edx std_cpuid1_edx;
290 // cpuid function 4 (deterministic cache parameters)
291 DcpCpuid4Eax dcp_cpuid4_eax;
292 DcpCpuid4Ebx dcp_cpuid4_ebx;
293 uint32_t dcp_cpuid4_ecx; // unused currently
294 uint32_t dcp_cpuid4_edx; // unused currently
296 // cpuid function 7 (structured extended features)
297 SefCpuid7Eax sef_cpuid7_eax;
298 SefCpuid7Ebx sef_cpuid7_ebx;
299 uint32_t sef_cpuid7_ecx; // unused currently
300 uint32_t sef_cpuid7_edx; // unused currently
302 // cpuid function 0xB (processor topology)
303 // ecx = 0
304 uint32_t tpl_cpuidB0_eax;
305 TplCpuidBEbx tpl_cpuidB0_ebx;
306 uint32_t tpl_cpuidB0_ecx; // unused currently
307 uint32_t tpl_cpuidB0_edx; // unused currently
309 // ecx = 1
310 uint32_t tpl_cpuidB1_eax;
311 TplCpuidBEbx tpl_cpuidB1_ebx;
312 uint32_t tpl_cpuidB1_ecx; // unused currently
313 uint32_t tpl_cpuidB1_edx; // unused currently
315 // ecx = 2
316 uint32_t tpl_cpuidB2_eax;
317 TplCpuidBEbx tpl_cpuidB2_ebx;
318 uint32_t tpl_cpuidB2_ecx; // unused currently
319 uint32_t tpl_cpuidB2_edx; // unused currently
321 // cpuid function 0x80000000 // example, unused
322 uint32_t ext_max_function;
323 uint32_t ext_vendor_name_0;
324 uint32_t ext_vendor_name_1;
325 uint32_t ext_vendor_name_2;
327 // cpuid function 0x80000001
328 uint32_t ext_cpuid1_eax; // reserved
329 uint32_t ext_cpuid1_ebx; // reserved
330 ExtCpuid1Ecx ext_cpuid1_ecx;
331 ExtCpuid1Edx ext_cpuid1_edx;
333 // cpuid functions 0x80000002 thru 0x80000004: example, unused
334 uint32_t proc_name_0, proc_name_1, proc_name_2, proc_name_3;
335 uint32_t proc_name_4, proc_name_5, proc_name_6, proc_name_7;
336 uint32_t proc_name_8, proc_name_9, proc_name_10,proc_name_11;
338 // cpuid function 0x80000005 // AMD L1, Intel reserved
339 uint32_t ext_cpuid5_eax; // unused currently
340 uint32_t ext_cpuid5_ebx; // reserved
341 ExtCpuid5Ex ext_cpuid5_ecx; // L1 data cache info (AMD)
342 ExtCpuid5Ex ext_cpuid5_edx; // L1 instruction cache info (AMD)
344 // cpuid function 0x80000007
345 uint32_t ext_cpuid7_eax; // reserved
346 uint32_t ext_cpuid7_ebx; // reserved
347 uint32_t ext_cpuid7_ecx; // reserved
348 ExtCpuid7Edx ext_cpuid7_edx; // tscinv
350 // cpuid function 0x80000008
351 uint32_t ext_cpuid8_eax; // unused currently
352 uint32_t ext_cpuid8_ebx; // reserved
353 ExtCpuid8Ecx ext_cpuid8_ecx;
354 uint32_t ext_cpuid8_edx; // reserved
356 // extended control register XCR0 (the XFEATURE_ENABLED_MASK register)
357 XemXcr0Eax xem_xcr0_eax;
358 uint32_t xem_xcr0_edx; // reserved
359 };
361 // The actual cpuid info block
362 static CpuidInfo _cpuid_info;
364 // Extractors and predicates
365 static uint32_t extended_cpu_family() {
366 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.family;
367 result += _cpuid_info.std_cpuid1_eax.bits.ext_family;
368 return result;
369 }
371 static uint32_t extended_cpu_model() {
372 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.model;
373 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4;
374 return result;
375 }
377 static uint32_t cpu_stepping() {
378 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping;
379 return result;
380 }
382 static uint logical_processor_count() {
383 uint result = threads_per_core();
384 return result;
385 }
387 static uint32_t feature_flags() {
388 uint32_t result = 0;
389 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0)
390 result |= CPU_CX8;
391 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0)
392 result |= CPU_CMOV;
393 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || (is_amd() &&
394 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0))
395 result |= CPU_FXSR;
396 // HT flag is set for multi-core processors also.
397 if (threads_per_core() > 1)
398 result |= CPU_HT;
399 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || (is_amd() &&
400 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0))
401 result |= CPU_MMX;
402 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0)
403 result |= CPU_SSE;
404 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0)
405 result |= CPU_SSE2;
406 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0)
407 result |= CPU_SSE3;
408 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0)
409 result |= CPU_SSSE3;
410 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_1 != 0)
411 result |= CPU_SSE4_1;
412 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0)
413 result |= CPU_SSE4_2;
414 if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0)
415 result |= CPU_POPCNT;
416 if (_cpuid_info.std_cpuid1_ecx.bits.avx != 0 &&
417 _cpuid_info.std_cpuid1_ecx.bits.osxsave != 0 &&
418 _cpuid_info.xem_xcr0_eax.bits.sse != 0 &&
419 _cpuid_info.xem_xcr0_eax.bits.ymm != 0) {
420 result |= CPU_AVX;
421 if (_cpuid_info.sef_cpuid7_ebx.bits.avx2 != 0)
422 result |= CPU_AVX2;
423 }
424 if (_cpuid_info.std_cpuid1_edx.bits.tsc != 0)
425 result |= CPU_TSC;
426 if (_cpuid_info.ext_cpuid7_edx.bits.tsc_invariance != 0)
427 result |= CPU_TSCINV;
428 if (_cpuid_info.std_cpuid1_ecx.bits.aes != 0)
429 result |= CPU_AES;
430 if (_cpuid_info.sef_cpuid7_ebx.bits.erms != 0)
431 result |= CPU_ERMS;
433 // AMD features.
434 if (is_amd()) {
435 if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) ||
436 (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0))
437 result |= CPU_3DNOW_PREFETCH;
438 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0)
439 result |= CPU_LZCNT;
440 if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0)
441 result |= CPU_SSE4A;
442 }
444 return result;
445 }
447 static void get_processor_features();
449 public:
450 // Offsets for cpuid asm stub
451 static ByteSize std_cpuid0_offset() { return byte_offset_of(CpuidInfo, std_max_function); }
452 static ByteSize std_cpuid1_offset() { return byte_offset_of(CpuidInfo, std_cpuid1_eax); }
453 static ByteSize dcp_cpuid4_offset() { return byte_offset_of(CpuidInfo, dcp_cpuid4_eax); }
454 static ByteSize sef_cpuid7_offset() { return byte_offset_of(CpuidInfo, sef_cpuid7_eax); }
455 static ByteSize ext_cpuid1_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1_eax); }
456 static ByteSize ext_cpuid5_offset() { return byte_offset_of(CpuidInfo, ext_cpuid5_eax); }
457 static ByteSize ext_cpuid7_offset() { return byte_offset_of(CpuidInfo, ext_cpuid7_eax); }
458 static ByteSize ext_cpuid8_offset() { return byte_offset_of(CpuidInfo, ext_cpuid8_eax); }
459 static ByteSize tpl_cpuidB0_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB0_eax); }
460 static ByteSize tpl_cpuidB1_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB1_eax); }
461 static ByteSize tpl_cpuidB2_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB2_eax); }
462 static ByteSize xem_xcr0_offset() { return byte_offset_of(CpuidInfo, xem_xcr0_eax); }
464 // Initialization
465 static void initialize();
467 // Asserts
468 static void assert_is_initialized() {
469 assert(_cpuid_info.std_cpuid1_eax.bits.family != 0, "VM_Version not initialized");
470 }
472 //
473 // Processor family:
474 // 3 - 386
475 // 4 - 486
476 // 5 - Pentium
477 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon,
478 // Pentium M, Core Solo, Core Duo, Core2 Duo
479 // family 6 model: 9, 13, 14, 15
480 // 0x0f - Pentium 4, Opteron
481 //
482 // Note: The cpu family should be used to select between
483 // instruction sequences which are valid on all Intel
484 // processors. Use the feature test functions below to
485 // determine whether a particular instruction is supported.
486 //
487 static int cpu_family() { return _cpu;}
488 static bool is_P6() { return cpu_family() >= 6; }
489 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA'
490 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG'
492 static bool supports_processor_topology() {
493 return (_cpuid_info.std_max_function >= 0xB) &&
494 // eax[4:0] | ebx[0:15] == 0 indicates invalid topology level.
495 // Some cpus have max cpuid >= 0xB but do not support processor topology.
496 (((_cpuid_info.tpl_cpuidB0_eax & 0x1f) | _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus) != 0);
497 }
499 static uint cores_per_cpu() {
500 uint result = 1;
501 if (is_intel()) {
502 if (supports_processor_topology()) {
503 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus /
504 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
505 } else {
506 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1);
507 }
508 } else if (is_amd()) {
509 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1);
510 }
511 return result;
512 }
514 static uint threads_per_core() {
515 uint result = 1;
516 if (is_intel() && supports_processor_topology()) {
517 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
518 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) {
519 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu /
520 cores_per_cpu();
521 }
522 return result;
523 }
525 static intx prefetch_data_size() {
526 intx result = 0;
527 if (is_intel()) {
528 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1);
529 } else if (is_amd()) {
530 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size;
531 }
532 if (result < 32) // not defined ?
533 result = 32; // 32 bytes by default on x86 and other x64
534 return result;
535 }
537 //
538 // Feature identification
539 //
540 static bool supports_cpuid() { return _cpuFeatures != 0; }
541 static bool supports_cmpxchg8() { return (_cpuFeatures & CPU_CX8) != 0; }
542 static bool supports_cmov() { return (_cpuFeatures & CPU_CMOV) != 0; }
543 static bool supports_fxsr() { return (_cpuFeatures & CPU_FXSR) != 0; }
544 static bool supports_ht() { return (_cpuFeatures & CPU_HT) != 0; }
545 static bool supports_mmx() { return (_cpuFeatures & CPU_MMX) != 0; }
546 static bool supports_sse() { return (_cpuFeatures & CPU_SSE) != 0; }
547 static bool supports_sse2() { return (_cpuFeatures & CPU_SSE2) != 0; }
548 static bool supports_sse3() { return (_cpuFeatures & CPU_SSE3) != 0; }
549 static bool supports_ssse3() { return (_cpuFeatures & CPU_SSSE3)!= 0; }
550 static bool supports_sse4_1() { return (_cpuFeatures & CPU_SSE4_1) != 0; }
551 static bool supports_sse4_2() { return (_cpuFeatures & CPU_SSE4_2) != 0; }
552 static bool supports_popcnt() { return (_cpuFeatures & CPU_POPCNT) != 0; }
553 static bool supports_avx() { return (_cpuFeatures & CPU_AVX) != 0; }
554 static bool supports_avx2() { return (_cpuFeatures & CPU_AVX2) != 0; }
555 static bool supports_tsc() { return (_cpuFeatures & CPU_TSC) != 0; }
556 static bool supports_aes() { return (_cpuFeatures & CPU_AES) != 0; }
557 static bool supports_erms() { return (_cpuFeatures & CPU_ERMS) != 0; }
559 // Intel features
560 static bool is_intel_family_core() { return is_intel() &&
561 extended_cpu_family() == CPU_FAMILY_INTEL_CORE; }
563 static bool is_intel_tsc_synched_at_init() {
564 if (is_intel_family_core()) {
565 uint32_t ext_model = extended_cpu_model();
566 if (ext_model == CPU_MODEL_NEHALEM_EP ||
567 ext_model == CPU_MODEL_WESTMERE_EP ||
568 ext_model == CPU_MODEL_SANDYBRIDGE_EP ||
569 ext_model == CPU_MODEL_IVYBRIDGE_EP) {
570 // <= 2-socket invariant tsc support. EX versions are usually used
571 // in > 2-socket systems and likely don't synchronize tscs at
572 // initialization.
573 // Code that uses tsc values must be prepared for them to arbitrarily
574 // jump forward or backward.
575 return true;
576 }
577 }
578 return false;
579 }
581 // AMD features
582 static bool supports_3dnow_prefetch() { return (_cpuFeatures & CPU_3DNOW_PREFETCH) != 0; }
583 static bool supports_mmx_ext() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; }
584 static bool supports_lzcnt() { return (_cpuFeatures & CPU_LZCNT) != 0; }
585 static bool supports_sse4a() { return (_cpuFeatures & CPU_SSE4A) != 0; }
587 static bool is_amd_Barcelona() { return is_amd() &&
588 extended_cpu_family() == CPU_FAMILY_AMD_11H; }
590 // Intel and AMD newer cores support fast timestamps well
591 static bool supports_tscinv_bit() {
592 return (_cpuFeatures & CPU_TSCINV) != 0;
593 }
594 static bool supports_tscinv() {
595 return supports_tscinv_bit() &&
596 ( (is_amd() && !is_amd_Barcelona()) ||
597 is_intel_tsc_synched_at_init() );
598 }
600 // Intel Core and newer cpus have fast IDIV instruction (excluding Atom).
601 static bool has_fast_idiv() { return is_intel() && cpu_family() == 6 &&
602 supports_sse3() && _model != 0x1C; }
604 static bool supports_compare_and_exchange() { return true; }
606 static const char* cpu_features() { return _features_str; }
608 static intx allocate_prefetch_distance() {
609 // This method should be called before allocate_prefetch_style().
610 //
611 // Hardware prefetching (distance/size in bytes):
612 // Pentium 3 - 64 / 32
613 // Pentium 4 - 256 / 128
614 // Athlon - 64 / 32 ????
615 // Opteron - 128 / 64 only when 2 sequential cache lines accessed
616 // Core - 128 / 64
617 //
618 // Software prefetching (distance in bytes / instruction with best score):
619 // Pentium 3 - 128 / prefetchnta
620 // Pentium 4 - 512 / prefetchnta
621 // Athlon - 128 / prefetchnta
622 // Opteron - 256 / prefetchnta
623 // Core - 256 / prefetchnta
624 // It will be used only when AllocatePrefetchStyle > 0
626 intx count = AllocatePrefetchDistance;
627 if (count < 0) { // default ?
628 if (is_amd()) { // AMD
629 if (supports_sse2())
630 count = 256; // Opteron
631 else
632 count = 128; // Athlon
633 } else { // Intel
634 if (supports_sse2())
635 if (cpu_family() == 6) {
636 count = 256; // Pentium M, Core, Core2
637 } else {
638 count = 512; // Pentium 4
639 }
640 else
641 count = 128; // Pentium 3 (and all other old CPUs)
642 }
643 }
644 return count;
645 }
646 static intx allocate_prefetch_style() {
647 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
648 // Return 0 if AllocatePrefetchDistance was not defined.
649 return AllocatePrefetchDistance > 0 ? AllocatePrefetchStyle : 0;
650 }
652 // Prefetch interval for gc copy/scan == 9 dcache lines. Derived from
653 // 50-warehouse specjbb runs on a 2-way 1.8ghz opteron using a 4gb heap.
654 // Tested intervals from 128 to 2048 in increments of 64 == one cache line.
655 // 256 bytes (4 dcache lines) was the nearest runner-up to 576.
657 // gc copy/scan is disabled if prefetchw isn't supported, because
658 // Prefetch::write emits an inlined prefetchw on Linux.
659 // Do not use the 3dnow prefetchw instruction. It isn't supported on em64t.
660 // The used prefetcht0 instruction works for both amd64 and em64t.
661 static intx prefetch_copy_interval_in_bytes() {
662 intx interval = PrefetchCopyIntervalInBytes;
663 return interval >= 0 ? interval : 576;
664 }
665 static intx prefetch_scan_interval_in_bytes() {
666 intx interval = PrefetchScanIntervalInBytes;
667 return interval >= 0 ? interval : 576;
668 }
669 static intx prefetch_fields_ahead() {
670 intx count = PrefetchFieldsAhead;
671 return count >= 0 ? count : 1;
672 }
673 };
675 #endif // CPU_X86_VM_VM_VERSION_X86_HPP