Fri, 14 Mar 2014 17:28:58 -0700
8037226: compiler/7196199/Test7196199.java fails on 32-bit linux with MaxVectorSize > 16
Summary: verify YMM registers after signal processing and set limit on vector's size.
Reviewed-by: iveresov, twisti
1 /*
2 * Copyright (c) 1997, 2013, 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 clmul : 1,
65 : 1,
66 monitor : 1,
67 : 1,
68 vmx : 1,
69 : 1,
70 est : 1,
71 : 1,
72 ssse3 : 1,
73 cid : 1,
74 : 2,
75 cmpxchg16: 1,
76 : 4,
77 dca : 1,
78 sse4_1 : 1,
79 sse4_2 : 1,
80 : 2,
81 popcnt : 1,
82 : 1,
83 aes : 1,
84 : 1,
85 osxsave : 1,
86 avx : 1,
87 : 3;
88 } bits;
89 };
91 union StdCpuid1Edx {
92 uint32_t value;
93 struct {
94 uint32_t : 4,
95 tsc : 1,
96 : 3,
97 cmpxchg8 : 1,
98 : 6,
99 cmov : 1,
100 : 3,
101 clflush : 1,
102 : 3,
103 mmx : 1,
104 fxsr : 1,
105 sse : 1,
106 sse2 : 1,
107 : 1,
108 ht : 1,
109 : 3;
110 } bits;
111 };
113 union DcpCpuid4Eax {
114 uint32_t value;
115 struct {
116 uint32_t cache_type : 5,
117 : 21,
118 cores_per_cpu : 6;
119 } bits;
120 };
122 union DcpCpuid4Ebx {
123 uint32_t value;
124 struct {
125 uint32_t L1_line_size : 12,
126 partitions : 10,
127 associativity : 10;
128 } bits;
129 };
131 union TplCpuidBEbx {
132 uint32_t value;
133 struct {
134 uint32_t logical_cpus : 16,
135 : 16;
136 } bits;
137 };
139 union ExtCpuid1Ecx {
140 uint32_t value;
141 struct {
142 uint32_t LahfSahf : 1,
143 CmpLegacy : 1,
144 : 3,
145 lzcnt_intel : 1,
146 lzcnt : 1,
147 sse4a : 1,
148 misalignsse : 1,
149 prefetchw : 1,
150 : 22;
151 } bits;
152 };
154 union ExtCpuid1Edx {
155 uint32_t value;
156 struct {
157 uint32_t : 22,
158 mmx_amd : 1,
159 mmx : 1,
160 fxsr : 1,
161 : 4,
162 long_mode : 1,
163 tdnow2 : 1,
164 tdnow : 1;
165 } bits;
166 };
168 union ExtCpuid5Ex {
169 uint32_t value;
170 struct {
171 uint32_t L1_line_size : 8,
172 L1_tag_lines : 8,
173 L1_assoc : 8,
174 L1_size : 8;
175 } bits;
176 };
178 union ExtCpuid7Edx {
179 uint32_t value;
180 struct {
181 uint32_t : 8,
182 tsc_invariance : 1,
183 : 23;
184 } bits;
185 };
187 union ExtCpuid8Ecx {
188 uint32_t value;
189 struct {
190 uint32_t cores_per_cpu : 8,
191 : 24;
192 } bits;
193 };
195 union SefCpuid7Eax {
196 uint32_t value;
197 };
199 union SefCpuid7Ebx {
200 uint32_t value;
201 struct {
202 uint32_t fsgsbase : 1,
203 : 2,
204 bmi1 : 1,
205 : 1,
206 avx2 : 1,
207 : 2,
208 bmi2 : 1,
209 erms : 1,
210 : 22;
211 } bits;
212 };
214 union XemXcr0Eax {
215 uint32_t value;
216 struct {
217 uint32_t x87 : 1,
218 sse : 1,
219 ymm : 1,
220 : 29;
221 } bits;
222 };
224 protected:
225 static int _cpu;
226 static int _model;
227 static int _stepping;
228 static int _cpuFeatures; // features returned by the "cpuid" instruction
229 // 0 if this instruction is not available
230 static const char* _features_str;
232 static address _cpuinfo_segv_addr; // address of instruction which causes SEGV
233 static address _cpuinfo_cont_addr; // address of instruction after the one which causes SEGV
235 enum {
236 CPU_CX8 = (1 << 0), // next bits are from cpuid 1 (EDX)
237 CPU_CMOV = (1 << 1),
238 CPU_FXSR = (1 << 2),
239 CPU_HT = (1 << 3),
240 CPU_MMX = (1 << 4),
241 CPU_3DNOW_PREFETCH = (1 << 5), // Processor supports 3dnow prefetch and prefetchw instructions
242 // may not necessarily support other 3dnow instructions
243 CPU_SSE = (1 << 6),
244 CPU_SSE2 = (1 << 7),
245 CPU_SSE3 = (1 << 8), // SSE3 comes from cpuid 1 (ECX)
246 CPU_SSSE3 = (1 << 9),
247 CPU_SSE4A = (1 << 10),
248 CPU_SSE4_1 = (1 << 11),
249 CPU_SSE4_2 = (1 << 12),
250 CPU_POPCNT = (1 << 13),
251 CPU_LZCNT = (1 << 14),
252 CPU_TSC = (1 << 15),
253 CPU_TSCINV = (1 << 16),
254 CPU_AVX = (1 << 17),
255 CPU_AVX2 = (1 << 18),
256 CPU_AES = (1 << 19),
257 CPU_ERMS = (1 << 20), // enhanced 'rep movsb/stosb' instructions
258 CPU_CLMUL = (1 << 21), // carryless multiply for CRC
259 CPU_BMI1 = (1 << 22),
260 CPU_BMI2 = (1 << 23)
261 } cpuFeatureFlags;
263 enum {
264 // AMD
265 CPU_FAMILY_AMD_11H = 0x11,
266 // Intel
267 CPU_FAMILY_INTEL_CORE = 6,
268 CPU_MODEL_NEHALEM = 0x1e,
269 CPU_MODEL_NEHALEM_EP = 0x1a,
270 CPU_MODEL_NEHALEM_EX = 0x2e,
271 CPU_MODEL_WESTMERE = 0x25,
272 CPU_MODEL_WESTMERE_EP = 0x2c,
273 CPU_MODEL_WESTMERE_EX = 0x2f,
274 CPU_MODEL_SANDYBRIDGE = 0x2a,
275 CPU_MODEL_SANDYBRIDGE_EP = 0x2d,
276 CPU_MODEL_IVYBRIDGE_EP = 0x3a
277 } cpuExtendedFamily;
279 // cpuid information block. All info derived from executing cpuid with
280 // various function numbers is stored here. Intel and AMD info is
281 // merged in this block: accessor methods disentangle it.
282 //
283 // The info block is laid out in subblocks of 4 dwords corresponding to
284 // eax, ebx, ecx and edx, whether or not they contain anything useful.
285 struct CpuidInfo {
286 // cpuid function 0
287 uint32_t std_max_function;
288 uint32_t std_vendor_name_0;
289 uint32_t std_vendor_name_1;
290 uint32_t std_vendor_name_2;
292 // cpuid function 1
293 StdCpuid1Eax std_cpuid1_eax;
294 StdCpuid1Ebx std_cpuid1_ebx;
295 StdCpuid1Ecx std_cpuid1_ecx;
296 StdCpuid1Edx std_cpuid1_edx;
298 // cpuid function 4 (deterministic cache parameters)
299 DcpCpuid4Eax dcp_cpuid4_eax;
300 DcpCpuid4Ebx dcp_cpuid4_ebx;
301 uint32_t dcp_cpuid4_ecx; // unused currently
302 uint32_t dcp_cpuid4_edx; // unused currently
304 // cpuid function 7 (structured extended features)
305 SefCpuid7Eax sef_cpuid7_eax;
306 SefCpuid7Ebx sef_cpuid7_ebx;
307 uint32_t sef_cpuid7_ecx; // unused currently
308 uint32_t sef_cpuid7_edx; // unused currently
310 // cpuid function 0xB (processor topology)
311 // ecx = 0
312 uint32_t tpl_cpuidB0_eax;
313 TplCpuidBEbx tpl_cpuidB0_ebx;
314 uint32_t tpl_cpuidB0_ecx; // unused currently
315 uint32_t tpl_cpuidB0_edx; // unused currently
317 // ecx = 1
318 uint32_t tpl_cpuidB1_eax;
319 TplCpuidBEbx tpl_cpuidB1_ebx;
320 uint32_t tpl_cpuidB1_ecx; // unused currently
321 uint32_t tpl_cpuidB1_edx; // unused currently
323 // ecx = 2
324 uint32_t tpl_cpuidB2_eax;
325 TplCpuidBEbx tpl_cpuidB2_ebx;
326 uint32_t tpl_cpuidB2_ecx; // unused currently
327 uint32_t tpl_cpuidB2_edx; // unused currently
329 // cpuid function 0x80000000 // example, unused
330 uint32_t ext_max_function;
331 uint32_t ext_vendor_name_0;
332 uint32_t ext_vendor_name_1;
333 uint32_t ext_vendor_name_2;
335 // cpuid function 0x80000001
336 uint32_t ext_cpuid1_eax; // reserved
337 uint32_t ext_cpuid1_ebx; // reserved
338 ExtCpuid1Ecx ext_cpuid1_ecx;
339 ExtCpuid1Edx ext_cpuid1_edx;
341 // cpuid functions 0x80000002 thru 0x80000004: example, unused
342 uint32_t proc_name_0, proc_name_1, proc_name_2, proc_name_3;
343 uint32_t proc_name_4, proc_name_5, proc_name_6, proc_name_7;
344 uint32_t proc_name_8, proc_name_9, proc_name_10,proc_name_11;
346 // cpuid function 0x80000005 // AMD L1, Intel reserved
347 uint32_t ext_cpuid5_eax; // unused currently
348 uint32_t ext_cpuid5_ebx; // reserved
349 ExtCpuid5Ex ext_cpuid5_ecx; // L1 data cache info (AMD)
350 ExtCpuid5Ex ext_cpuid5_edx; // L1 instruction cache info (AMD)
352 // cpuid function 0x80000007
353 uint32_t ext_cpuid7_eax; // reserved
354 uint32_t ext_cpuid7_ebx; // reserved
355 uint32_t ext_cpuid7_ecx; // reserved
356 ExtCpuid7Edx ext_cpuid7_edx; // tscinv
358 // cpuid function 0x80000008
359 uint32_t ext_cpuid8_eax; // unused currently
360 uint32_t ext_cpuid8_ebx; // reserved
361 ExtCpuid8Ecx ext_cpuid8_ecx;
362 uint32_t ext_cpuid8_edx; // reserved
364 // extended control register XCR0 (the XFEATURE_ENABLED_MASK register)
365 XemXcr0Eax xem_xcr0_eax;
366 uint32_t xem_xcr0_edx; // reserved
368 // Space to save ymm registers after signal handle
369 int ymm_save[8*4]; // Save ymm0, ymm7, ymm8, ymm15
370 };
372 // The actual cpuid info block
373 static CpuidInfo _cpuid_info;
375 // Extractors and predicates
376 static uint32_t extended_cpu_family() {
377 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.family;
378 result += _cpuid_info.std_cpuid1_eax.bits.ext_family;
379 return result;
380 }
382 static uint32_t extended_cpu_model() {
383 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.model;
384 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4;
385 return result;
386 }
388 static uint32_t cpu_stepping() {
389 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping;
390 return result;
391 }
393 static uint logical_processor_count() {
394 uint result = threads_per_core();
395 return result;
396 }
398 static uint32_t feature_flags() {
399 uint32_t result = 0;
400 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0)
401 result |= CPU_CX8;
402 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0)
403 result |= CPU_CMOV;
404 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || (is_amd() &&
405 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0))
406 result |= CPU_FXSR;
407 // HT flag is set for multi-core processors also.
408 if (threads_per_core() > 1)
409 result |= CPU_HT;
410 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || (is_amd() &&
411 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0))
412 result |= CPU_MMX;
413 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0)
414 result |= CPU_SSE;
415 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0)
416 result |= CPU_SSE2;
417 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0)
418 result |= CPU_SSE3;
419 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0)
420 result |= CPU_SSSE3;
421 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_1 != 0)
422 result |= CPU_SSE4_1;
423 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0)
424 result |= CPU_SSE4_2;
425 if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0)
426 result |= CPU_POPCNT;
427 if (_cpuid_info.std_cpuid1_ecx.bits.avx != 0 &&
428 _cpuid_info.std_cpuid1_ecx.bits.osxsave != 0 &&
429 _cpuid_info.xem_xcr0_eax.bits.sse != 0 &&
430 _cpuid_info.xem_xcr0_eax.bits.ymm != 0) {
431 result |= CPU_AVX;
432 if (_cpuid_info.sef_cpuid7_ebx.bits.avx2 != 0)
433 result |= CPU_AVX2;
434 }
435 if(_cpuid_info.sef_cpuid7_ebx.bits.bmi1 != 0)
436 result |= CPU_BMI1;
437 if (_cpuid_info.std_cpuid1_edx.bits.tsc != 0)
438 result |= CPU_TSC;
439 if (_cpuid_info.ext_cpuid7_edx.bits.tsc_invariance != 0)
440 result |= CPU_TSCINV;
441 if (_cpuid_info.std_cpuid1_ecx.bits.aes != 0)
442 result |= CPU_AES;
443 if (_cpuid_info.sef_cpuid7_ebx.bits.erms != 0)
444 result |= CPU_ERMS;
445 if (_cpuid_info.std_cpuid1_ecx.bits.clmul != 0)
446 result |= CPU_CLMUL;
448 // AMD features.
449 if (is_amd()) {
450 if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) ||
451 (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0))
452 result |= CPU_3DNOW_PREFETCH;
453 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0)
454 result |= CPU_LZCNT;
455 if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0)
456 result |= CPU_SSE4A;
457 }
458 // Intel features.
459 if(is_intel()) {
460 if(_cpuid_info.sef_cpuid7_ebx.bits.bmi2 != 0)
461 result |= CPU_BMI2;
462 if(_cpuid_info.ext_cpuid1_ecx.bits.lzcnt_intel != 0)
463 result |= CPU_LZCNT;
464 }
466 return result;
467 }
469 static bool os_supports_avx_vectors() {
470 if (!supports_avx()) {
471 return false;
472 }
473 // Verify that OS save/restore all bits of AVX registers
474 // during signal processing.
475 int nreg = 2 LP64_ONLY(+2);
476 for (int i = 0; i < 8 * nreg; i++) { // 32 bytes per ymm register
477 if (_cpuid_info.ymm_save[i] != ymm_test_value()) {
478 return false;
479 }
480 }
481 return true;
482 }
484 static void get_processor_features();
486 public:
487 // Offsets for cpuid asm stub
488 static ByteSize std_cpuid0_offset() { return byte_offset_of(CpuidInfo, std_max_function); }
489 static ByteSize std_cpuid1_offset() { return byte_offset_of(CpuidInfo, std_cpuid1_eax); }
490 static ByteSize dcp_cpuid4_offset() { return byte_offset_of(CpuidInfo, dcp_cpuid4_eax); }
491 static ByteSize sef_cpuid7_offset() { return byte_offset_of(CpuidInfo, sef_cpuid7_eax); }
492 static ByteSize ext_cpuid1_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1_eax); }
493 static ByteSize ext_cpuid5_offset() { return byte_offset_of(CpuidInfo, ext_cpuid5_eax); }
494 static ByteSize ext_cpuid7_offset() { return byte_offset_of(CpuidInfo, ext_cpuid7_eax); }
495 static ByteSize ext_cpuid8_offset() { return byte_offset_of(CpuidInfo, ext_cpuid8_eax); }
496 static ByteSize tpl_cpuidB0_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB0_eax); }
497 static ByteSize tpl_cpuidB1_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB1_eax); }
498 static ByteSize tpl_cpuidB2_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB2_eax); }
499 static ByteSize xem_xcr0_offset() { return byte_offset_of(CpuidInfo, xem_xcr0_eax); }
500 static ByteSize ymm_save_offset() { return byte_offset_of(CpuidInfo, ymm_save); }
502 // The value used to check ymm register after signal handle
503 static int ymm_test_value() { return 0xCAFEBABE; }
505 static void set_cpuinfo_segv_addr(address pc) { _cpuinfo_segv_addr = pc; }
506 static bool is_cpuinfo_segv_addr(address pc) { return _cpuinfo_segv_addr == pc; }
507 static void set_cpuinfo_cont_addr(address pc) { _cpuinfo_cont_addr = pc; }
508 static address cpuinfo_cont_addr() { return _cpuinfo_cont_addr; }
510 static void clean_cpuFeatures() { _cpuFeatures = 0; }
511 static void set_avx_cpuFeatures() { _cpuFeatures = (CPU_SSE | CPU_SSE2 | CPU_AVX); }
514 // Initialization
515 static void initialize();
517 // Asserts
518 static void assert_is_initialized() {
519 assert(_cpuid_info.std_cpuid1_eax.bits.family != 0, "VM_Version not initialized");
520 }
522 //
523 // Processor family:
524 // 3 - 386
525 // 4 - 486
526 // 5 - Pentium
527 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon,
528 // Pentium M, Core Solo, Core Duo, Core2 Duo
529 // family 6 model: 9, 13, 14, 15
530 // 0x0f - Pentium 4, Opteron
531 //
532 // Note: The cpu family should be used to select between
533 // instruction sequences which are valid on all Intel
534 // processors. Use the feature test functions below to
535 // determine whether a particular instruction is supported.
536 //
537 static int cpu_family() { return _cpu;}
538 static bool is_P6() { return cpu_family() >= 6; }
539 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA'
540 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG'
542 static bool supports_processor_topology() {
543 return (_cpuid_info.std_max_function >= 0xB) &&
544 // eax[4:0] | ebx[0:15] == 0 indicates invalid topology level.
545 // Some cpus have max cpuid >= 0xB but do not support processor topology.
546 (((_cpuid_info.tpl_cpuidB0_eax & 0x1f) | _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus) != 0);
547 }
549 static uint cores_per_cpu() {
550 uint result = 1;
551 if (is_intel()) {
552 if (supports_processor_topology()) {
553 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus /
554 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
555 } else {
556 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1);
557 }
558 } else if (is_amd()) {
559 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1);
560 }
561 return result;
562 }
564 static uint threads_per_core() {
565 uint result = 1;
566 if (is_intel() && supports_processor_topology()) {
567 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
568 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) {
569 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu /
570 cores_per_cpu();
571 }
572 return result;
573 }
575 static intx prefetch_data_size() {
576 intx result = 0;
577 if (is_intel()) {
578 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1);
579 } else if (is_amd()) {
580 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size;
581 }
582 if (result < 32) // not defined ?
583 result = 32; // 32 bytes by default on x86 and other x64
584 return result;
585 }
587 //
588 // Feature identification
589 //
590 static bool supports_cpuid() { return _cpuFeatures != 0; }
591 static bool supports_cmpxchg8() { return (_cpuFeatures & CPU_CX8) != 0; }
592 static bool supports_cmov() { return (_cpuFeatures & CPU_CMOV) != 0; }
593 static bool supports_fxsr() { return (_cpuFeatures & CPU_FXSR) != 0; }
594 static bool supports_ht() { return (_cpuFeatures & CPU_HT) != 0; }
595 static bool supports_mmx() { return (_cpuFeatures & CPU_MMX) != 0; }
596 static bool supports_sse() { return (_cpuFeatures & CPU_SSE) != 0; }
597 static bool supports_sse2() { return (_cpuFeatures & CPU_SSE2) != 0; }
598 static bool supports_sse3() { return (_cpuFeatures & CPU_SSE3) != 0; }
599 static bool supports_ssse3() { return (_cpuFeatures & CPU_SSSE3)!= 0; }
600 static bool supports_sse4_1() { return (_cpuFeatures & CPU_SSE4_1) != 0; }
601 static bool supports_sse4_2() { return (_cpuFeatures & CPU_SSE4_2) != 0; }
602 static bool supports_popcnt() { return (_cpuFeatures & CPU_POPCNT) != 0; }
603 static bool supports_avx() { return (_cpuFeatures & CPU_AVX) != 0; }
604 static bool supports_avx2() { return (_cpuFeatures & CPU_AVX2) != 0; }
605 static bool supports_tsc() { return (_cpuFeatures & CPU_TSC) != 0; }
606 static bool supports_aes() { return (_cpuFeatures & CPU_AES) != 0; }
607 static bool supports_erms() { return (_cpuFeatures & CPU_ERMS) != 0; }
608 static bool supports_clmul() { return (_cpuFeatures & CPU_CLMUL) != 0; }
609 static bool supports_bmi1() { return (_cpuFeatures & CPU_BMI1) != 0; }
610 static bool supports_bmi2() { return (_cpuFeatures & CPU_BMI2) != 0; }
611 // Intel features
612 static bool is_intel_family_core() { return is_intel() &&
613 extended_cpu_family() == CPU_FAMILY_INTEL_CORE; }
615 static bool is_intel_tsc_synched_at_init() {
616 if (is_intel_family_core()) {
617 uint32_t ext_model = extended_cpu_model();
618 if (ext_model == CPU_MODEL_NEHALEM_EP ||
619 ext_model == CPU_MODEL_WESTMERE_EP ||
620 ext_model == CPU_MODEL_SANDYBRIDGE_EP ||
621 ext_model == CPU_MODEL_IVYBRIDGE_EP) {
622 // <= 2-socket invariant tsc support. EX versions are usually used
623 // in > 2-socket systems and likely don't synchronize tscs at
624 // initialization.
625 // Code that uses tsc values must be prepared for them to arbitrarily
626 // jump forward or backward.
627 return true;
628 }
629 }
630 return false;
631 }
633 // AMD features
634 static bool supports_3dnow_prefetch() { return (_cpuFeatures & CPU_3DNOW_PREFETCH) != 0; }
635 static bool supports_mmx_ext() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; }
636 static bool supports_lzcnt() { return (_cpuFeatures & CPU_LZCNT) != 0; }
637 static bool supports_sse4a() { return (_cpuFeatures & CPU_SSE4A) != 0; }
639 static bool is_amd_Barcelona() { return is_amd() &&
640 extended_cpu_family() == CPU_FAMILY_AMD_11H; }
642 // Intel and AMD newer cores support fast timestamps well
643 static bool supports_tscinv_bit() {
644 return (_cpuFeatures & CPU_TSCINV) != 0;
645 }
646 static bool supports_tscinv() {
647 return supports_tscinv_bit() &&
648 ( (is_amd() && !is_amd_Barcelona()) ||
649 is_intel_tsc_synched_at_init() );
650 }
652 // Intel Core and newer cpus have fast IDIV instruction (excluding Atom).
653 static bool has_fast_idiv() { return is_intel() && cpu_family() == 6 &&
654 supports_sse3() && _model != 0x1C; }
656 static bool supports_compare_and_exchange() { return true; }
658 static const char* cpu_features() { return _features_str; }
660 static intx allocate_prefetch_distance() {
661 // This method should be called before allocate_prefetch_style().
662 //
663 // Hardware prefetching (distance/size in bytes):
664 // Pentium 3 - 64 / 32
665 // Pentium 4 - 256 / 128
666 // Athlon - 64 / 32 ????
667 // Opteron - 128 / 64 only when 2 sequential cache lines accessed
668 // Core - 128 / 64
669 //
670 // Software prefetching (distance in bytes / instruction with best score):
671 // Pentium 3 - 128 / prefetchnta
672 // Pentium 4 - 512 / prefetchnta
673 // Athlon - 128 / prefetchnta
674 // Opteron - 256 / prefetchnta
675 // Core - 256 / prefetchnta
676 // It will be used only when AllocatePrefetchStyle > 0
678 intx count = AllocatePrefetchDistance;
679 if (count < 0) { // default ?
680 if (is_amd()) { // AMD
681 if (supports_sse2())
682 count = 256; // Opteron
683 else
684 count = 128; // Athlon
685 } else { // Intel
686 if (supports_sse2())
687 if (cpu_family() == 6) {
688 count = 256; // Pentium M, Core, Core2
689 } else {
690 count = 512; // Pentium 4
691 }
692 else
693 count = 128; // Pentium 3 (and all other old CPUs)
694 }
695 }
696 return count;
697 }
698 static intx allocate_prefetch_style() {
699 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
700 // Return 0 if AllocatePrefetchDistance was not defined.
701 return AllocatePrefetchDistance > 0 ? AllocatePrefetchStyle : 0;
702 }
704 // Prefetch interval for gc copy/scan == 9 dcache lines. Derived from
705 // 50-warehouse specjbb runs on a 2-way 1.8ghz opteron using a 4gb heap.
706 // Tested intervals from 128 to 2048 in increments of 64 == one cache line.
707 // 256 bytes (4 dcache lines) was the nearest runner-up to 576.
709 // gc copy/scan is disabled if prefetchw isn't supported, because
710 // Prefetch::write emits an inlined prefetchw on Linux.
711 // Do not use the 3dnow prefetchw instruction. It isn't supported on em64t.
712 // The used prefetcht0 instruction works for both amd64 and em64t.
713 static intx prefetch_copy_interval_in_bytes() {
714 intx interval = PrefetchCopyIntervalInBytes;
715 return interval >= 0 ? interval : 576;
716 }
717 static intx prefetch_scan_interval_in_bytes() {
718 intx interval = PrefetchScanIntervalInBytes;
719 return interval >= 0 ? interval : 576;
720 }
721 static intx prefetch_fields_ahead() {
722 intx count = PrefetchFieldsAhead;
723 return count >= 0 ? count : 1;
724 }
725 };
727 #endif // CPU_X86_VM_VM_VERSION_X86_HPP