Wed, 14 Dec 2011 14:54:38 -0800
7116452: Add support for AVX instructions
Summary: Added support for AVX extension to the x86 instruction set.
Reviewed-by: never
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 #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 : 3,
82 osxsave : 1,
83 avx : 1,
84 : 3;
85 } bits;
86 };
88 union StdCpuid1Edx {
89 uint32_t value;
90 struct {
91 uint32_t : 4,
92 tsc : 1,
93 : 3,
94 cmpxchg8 : 1,
95 : 6,
96 cmov : 1,
97 : 3,
98 clflush : 1,
99 : 3,
100 mmx : 1,
101 fxsr : 1,
102 sse : 1,
103 sse2 : 1,
104 : 1,
105 ht : 1,
106 : 3;
107 } bits;
108 };
110 union DcpCpuid4Eax {
111 uint32_t value;
112 struct {
113 uint32_t cache_type : 5,
114 : 21,
115 cores_per_cpu : 6;
116 } bits;
117 };
119 union DcpCpuid4Ebx {
120 uint32_t value;
121 struct {
122 uint32_t L1_line_size : 12,
123 partitions : 10,
124 associativity : 10;
125 } bits;
126 };
128 union TplCpuidBEbx {
129 uint32_t value;
130 struct {
131 uint32_t logical_cpus : 16,
132 : 16;
133 } bits;
134 };
136 union ExtCpuid1Ecx {
137 uint32_t value;
138 struct {
139 uint32_t LahfSahf : 1,
140 CmpLegacy : 1,
141 : 4,
142 lzcnt : 1,
143 sse4a : 1,
144 misalignsse : 1,
145 prefetchw : 1,
146 : 22;
147 } bits;
148 };
150 union ExtCpuid1Edx {
151 uint32_t value;
152 struct {
153 uint32_t : 22,
154 mmx_amd : 1,
155 mmx : 1,
156 fxsr : 1,
157 : 4,
158 long_mode : 1,
159 tdnow2 : 1,
160 tdnow : 1;
161 } bits;
162 };
164 union ExtCpuid5Ex {
165 uint32_t value;
166 struct {
167 uint32_t L1_line_size : 8,
168 L1_tag_lines : 8,
169 L1_assoc : 8,
170 L1_size : 8;
171 } bits;
172 };
174 union ExtCpuid8Ecx {
175 uint32_t value;
176 struct {
177 uint32_t cores_per_cpu : 8,
178 : 24;
179 } bits;
180 };
182 union SefCpuid7Eax {
183 uint32_t value;
184 };
186 union SefCpuid7Ebx {
187 uint32_t value;
188 struct {
189 uint32_t fsgsbase : 1,
190 : 2,
191 bmi1 : 1,
192 : 1,
193 avx2 : 1,
194 : 2,
195 bmi2 : 1,
196 : 23;
197 } bits;
198 };
200 union XemXcr0Eax {
201 uint32_t value;
202 struct {
203 uint32_t x87 : 1,
204 sse : 1,
205 ymm : 1,
206 : 29;
207 } bits;
208 };
210 protected:
211 static int _cpu;
212 static int _model;
213 static int _stepping;
214 static int _cpuFeatures; // features returned by the "cpuid" instruction
215 // 0 if this instruction is not available
216 static const char* _features_str;
218 enum {
219 CPU_CX8 = (1 << 0), // next bits are from cpuid 1 (EDX)
220 CPU_CMOV = (1 << 1),
221 CPU_FXSR = (1 << 2),
222 CPU_HT = (1 << 3),
223 CPU_MMX = (1 << 4),
224 CPU_3DNOW_PREFETCH = (1 << 5), // Processor supports 3dnow prefetch and prefetchw instructions
225 // may not necessarily support other 3dnow instructions
226 CPU_SSE = (1 << 6),
227 CPU_SSE2 = (1 << 7),
228 CPU_SSE3 = (1 << 8), // SSE3 comes from cpuid 1 (ECX)
229 CPU_SSSE3 = (1 << 9),
230 CPU_SSE4A = (1 << 10),
231 CPU_SSE4_1 = (1 << 11),
232 CPU_SSE4_2 = (1 << 12),
233 CPU_POPCNT = (1 << 13),
234 CPU_LZCNT = (1 << 14),
235 CPU_AVX = (1 << 15),
236 CPU_AVX2 = (1 << 16)
237 } cpuFeatureFlags;
239 // cpuid information block. All info derived from executing cpuid with
240 // various function numbers is stored here. Intel and AMD info is
241 // merged in this block: accessor methods disentangle it.
242 //
243 // The info block is laid out in subblocks of 4 dwords corresponding to
244 // eax, ebx, ecx and edx, whether or not they contain anything useful.
245 struct CpuidInfo {
246 // cpuid function 0
247 uint32_t std_max_function;
248 uint32_t std_vendor_name_0;
249 uint32_t std_vendor_name_1;
250 uint32_t std_vendor_name_2;
252 // cpuid function 1
253 StdCpuid1Eax std_cpuid1_eax;
254 StdCpuid1Ebx std_cpuid1_ebx;
255 StdCpuid1Ecx std_cpuid1_ecx;
256 StdCpuid1Edx std_cpuid1_edx;
258 // cpuid function 4 (deterministic cache parameters)
259 DcpCpuid4Eax dcp_cpuid4_eax;
260 DcpCpuid4Ebx dcp_cpuid4_ebx;
261 uint32_t dcp_cpuid4_ecx; // unused currently
262 uint32_t dcp_cpuid4_edx; // unused currently
264 // cpuid function 7 (structured extended features)
265 SefCpuid7Eax sef_cpuid7_eax;
266 SefCpuid7Ebx sef_cpuid7_ebx;
267 uint32_t sef_cpuid7_ecx; // unused currently
268 uint32_t sef_cpuid7_edx; // unused currently
270 // cpuid function 0xB (processor topology)
271 // ecx = 0
272 uint32_t tpl_cpuidB0_eax;
273 TplCpuidBEbx tpl_cpuidB0_ebx;
274 uint32_t tpl_cpuidB0_ecx; // unused currently
275 uint32_t tpl_cpuidB0_edx; // unused currently
277 // ecx = 1
278 uint32_t tpl_cpuidB1_eax;
279 TplCpuidBEbx tpl_cpuidB1_ebx;
280 uint32_t tpl_cpuidB1_ecx; // unused currently
281 uint32_t tpl_cpuidB1_edx; // unused currently
283 // ecx = 2
284 uint32_t tpl_cpuidB2_eax;
285 TplCpuidBEbx tpl_cpuidB2_ebx;
286 uint32_t tpl_cpuidB2_ecx; // unused currently
287 uint32_t tpl_cpuidB2_edx; // unused currently
289 // cpuid function 0x80000000 // example, unused
290 uint32_t ext_max_function;
291 uint32_t ext_vendor_name_0;
292 uint32_t ext_vendor_name_1;
293 uint32_t ext_vendor_name_2;
295 // cpuid function 0x80000001
296 uint32_t ext_cpuid1_eax; // reserved
297 uint32_t ext_cpuid1_ebx; // reserved
298 ExtCpuid1Ecx ext_cpuid1_ecx;
299 ExtCpuid1Edx ext_cpuid1_edx;
301 // cpuid functions 0x80000002 thru 0x80000004: example, unused
302 uint32_t proc_name_0, proc_name_1, proc_name_2, proc_name_3;
303 uint32_t proc_name_4, proc_name_5, proc_name_6, proc_name_7;
304 uint32_t proc_name_8, proc_name_9, proc_name_10,proc_name_11;
306 // cpuid function 0x80000005 //AMD L1, Intel reserved
307 uint32_t ext_cpuid5_eax; // unused currently
308 uint32_t ext_cpuid5_ebx; // reserved
309 ExtCpuid5Ex ext_cpuid5_ecx; // L1 data cache info (AMD)
310 ExtCpuid5Ex ext_cpuid5_edx; // L1 instruction cache info (AMD)
312 // cpuid function 0x80000008
313 uint32_t ext_cpuid8_eax; // unused currently
314 uint32_t ext_cpuid8_ebx; // reserved
315 ExtCpuid8Ecx ext_cpuid8_ecx;
316 uint32_t ext_cpuid8_edx; // reserved
318 // extended control register XCR0 (the XFEATURE_ENABLED_MASK register)
319 XemXcr0Eax xem_xcr0_eax;
320 uint32_t xem_xcr0_edx; // reserved
321 };
323 // The actual cpuid info block
324 static CpuidInfo _cpuid_info;
326 // Extractors and predicates
327 static uint32_t extended_cpu_family() {
328 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.family;
329 result += _cpuid_info.std_cpuid1_eax.bits.ext_family;
330 return result;
331 }
332 static uint32_t extended_cpu_model() {
333 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.model;
334 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4;
335 return result;
336 }
337 static uint32_t cpu_stepping() {
338 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping;
339 return result;
340 }
341 static uint logical_processor_count() {
342 uint result = threads_per_core();
343 return result;
344 }
345 static uint32_t feature_flags() {
346 uint32_t result = 0;
347 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0)
348 result |= CPU_CX8;
349 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0)
350 result |= CPU_CMOV;
351 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || (is_amd() &&
352 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0))
353 result |= CPU_FXSR;
354 // HT flag is set for multi-core processors also.
355 if (threads_per_core() > 1)
356 result |= CPU_HT;
357 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || (is_amd() &&
358 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0))
359 result |= CPU_MMX;
360 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0)
361 result |= CPU_SSE;
362 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0)
363 result |= CPU_SSE2;
364 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0)
365 result |= CPU_SSE3;
366 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0)
367 result |= CPU_SSSE3;
368 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_1 != 0)
369 result |= CPU_SSE4_1;
370 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0)
371 result |= CPU_SSE4_2;
372 if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0)
373 result |= CPU_POPCNT;
374 if (_cpuid_info.std_cpuid1_ecx.bits.avx != 0 &&
375 _cpuid_info.std_cpuid1_ecx.bits.osxsave != 0 &&
376 _cpuid_info.xem_xcr0_eax.bits.sse != 0 &&
377 _cpuid_info.xem_xcr0_eax.bits.ymm != 0) {
378 result |= CPU_AVX;
379 if (_cpuid_info.sef_cpuid7_ebx.bits.avx2 != 0)
380 result |= CPU_AVX2;
381 }
383 // AMD features.
384 if (is_amd()) {
385 if ((_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) ||
386 (_cpuid_info.ext_cpuid1_ecx.bits.prefetchw != 0))
387 result |= CPU_3DNOW_PREFETCH;
388 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0)
389 result |= CPU_LZCNT;
390 if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0)
391 result |= CPU_SSE4A;
392 }
394 return result;
395 }
397 static void get_processor_features();
399 public:
400 // Offsets for cpuid asm stub
401 static ByteSize std_cpuid0_offset() { return byte_offset_of(CpuidInfo, std_max_function); }
402 static ByteSize std_cpuid1_offset() { return byte_offset_of(CpuidInfo, std_cpuid1_eax); }
403 static ByteSize dcp_cpuid4_offset() { return byte_offset_of(CpuidInfo, dcp_cpuid4_eax); }
404 static ByteSize sef_cpuid7_offset() { return byte_offset_of(CpuidInfo, sef_cpuid7_eax); }
405 static ByteSize ext_cpuid1_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1_eax); }
406 static ByteSize ext_cpuid5_offset() { return byte_offset_of(CpuidInfo, ext_cpuid5_eax); }
407 static ByteSize ext_cpuid8_offset() { return byte_offset_of(CpuidInfo, ext_cpuid8_eax); }
408 static ByteSize tpl_cpuidB0_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB0_eax); }
409 static ByteSize tpl_cpuidB1_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB1_eax); }
410 static ByteSize tpl_cpuidB2_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB2_eax); }
411 static ByteSize xem_xcr0_offset() { return byte_offset_of(CpuidInfo, xem_xcr0_eax); }
413 // Initialization
414 static void initialize();
416 // Asserts
417 static void assert_is_initialized() {
418 assert(_cpuid_info.std_cpuid1_eax.bits.family != 0, "VM_Version not initialized");
419 }
421 //
422 // Processor family:
423 // 3 - 386
424 // 4 - 486
425 // 5 - Pentium
426 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon,
427 // Pentium M, Core Solo, Core Duo, Core2 Duo
428 // family 6 model: 9, 13, 14, 15
429 // 0x0f - Pentium 4, Opteron
430 //
431 // Note: The cpu family should be used to select between
432 // instruction sequences which are valid on all Intel
433 // processors. Use the feature test functions below to
434 // determine whether a particular instruction is supported.
435 //
436 static int cpu_family() { return _cpu;}
437 static bool is_P6() { return cpu_family() >= 6; }
439 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA'
440 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG'
442 static bool supports_processor_topology() {
443 return (_cpuid_info.std_max_function >= 0xB) &&
444 // eax[4:0] | ebx[0:15] == 0 indicates invalid topology level.
445 // Some cpus have max cpuid >= 0xB but do not support processor topology.
446 ((_cpuid_info.tpl_cpuidB0_eax & 0x1f | _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus) != 0);
447 }
449 static uint cores_per_cpu() {
450 uint result = 1;
451 if (is_intel()) {
452 if (supports_processor_topology()) {
453 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus /
454 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
455 } else {
456 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1);
457 }
458 } else if (is_amd()) {
459 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1);
460 }
461 return result;
462 }
464 static uint threads_per_core() {
465 uint result = 1;
466 if (is_intel() && supports_processor_topology()) {
467 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus;
468 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) {
469 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu /
470 cores_per_cpu();
471 }
472 return result;
473 }
475 static intx prefetch_data_size() {
476 intx result = 0;
477 if (is_intel()) {
478 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1);
479 } else if (is_amd()) {
480 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size;
481 }
482 if (result < 32) // not defined ?
483 result = 32; // 32 bytes by default on x86 and other x64
484 return result;
485 }
487 //
488 // Feature identification
489 //
490 static bool supports_cpuid() { return _cpuFeatures != 0; }
491 static bool supports_cmpxchg8() { return (_cpuFeatures & CPU_CX8) != 0; }
492 static bool supports_cmov() { return (_cpuFeatures & CPU_CMOV) != 0; }
493 static bool supports_fxsr() { return (_cpuFeatures & CPU_FXSR) != 0; }
494 static bool supports_ht() { return (_cpuFeatures & CPU_HT) != 0; }
495 static bool supports_mmx() { return (_cpuFeatures & CPU_MMX) != 0; }
496 static bool supports_sse() { return (_cpuFeatures & CPU_SSE) != 0; }
497 static bool supports_sse2() { return (_cpuFeatures & CPU_SSE2) != 0; }
498 static bool supports_sse3() { return (_cpuFeatures & CPU_SSE3) != 0; }
499 static bool supports_ssse3() { return (_cpuFeatures & CPU_SSSE3)!= 0; }
500 static bool supports_sse4_1() { return (_cpuFeatures & CPU_SSE4_1) != 0; }
501 static bool supports_sse4_2() { return (_cpuFeatures & CPU_SSE4_2) != 0; }
502 static bool supports_popcnt() { return (_cpuFeatures & CPU_POPCNT) != 0; }
503 static bool supports_avx() { return (_cpuFeatures & CPU_AVX) != 0; }
504 static bool supports_avx2() { return (_cpuFeatures & CPU_AVX2) != 0; }
505 //
506 // AMD features
507 //
508 static bool supports_3dnow_prefetch() { return (_cpuFeatures & CPU_3DNOW_PREFETCH) != 0; }
509 static bool supports_mmx_ext() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; }
510 static bool supports_lzcnt() { return (_cpuFeatures & CPU_LZCNT) != 0; }
511 static bool supports_sse4a() { return (_cpuFeatures & CPU_SSE4A) != 0; }
513 // Intel Core and newer cpus have fast IDIV instruction (excluding Atom).
514 static bool has_fast_idiv() { return is_intel() && cpu_family() == 6 &&
515 supports_sse3() && _model != 0x1C; }
517 static bool supports_compare_and_exchange() { return true; }
519 static const char* cpu_features() { return _features_str; }
521 static intx allocate_prefetch_distance() {
522 // This method should be called before allocate_prefetch_style().
523 //
524 // Hardware prefetching (distance/size in bytes):
525 // Pentium 3 - 64 / 32
526 // Pentium 4 - 256 / 128
527 // Athlon - 64 / 32 ????
528 // Opteron - 128 / 64 only when 2 sequential cache lines accessed
529 // Core - 128 / 64
530 //
531 // Software prefetching (distance in bytes / instruction with best score):
532 // Pentium 3 - 128 / prefetchnta
533 // Pentium 4 - 512 / prefetchnta
534 // Athlon - 128 / prefetchnta
535 // Opteron - 256 / prefetchnta
536 // Core - 256 / prefetchnta
537 // It will be used only when AllocatePrefetchStyle > 0
539 intx count = AllocatePrefetchDistance;
540 if (count < 0) { // default ?
541 if (is_amd()) { // AMD
542 if (supports_sse2())
543 count = 256; // Opteron
544 else
545 count = 128; // Athlon
546 } else { // Intel
547 if (supports_sse2())
548 if (cpu_family() == 6) {
549 count = 256; // Pentium M, Core, Core2
550 } else {
551 count = 512; // Pentium 4
552 }
553 else
554 count = 128; // Pentium 3 (and all other old CPUs)
555 }
556 }
557 return count;
558 }
559 static intx allocate_prefetch_style() {
560 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
561 // Return 0 if AllocatePrefetchDistance was not defined.
562 return AllocatePrefetchDistance > 0 ? AllocatePrefetchStyle : 0;
563 }
565 // Prefetch interval for gc copy/scan == 9 dcache lines. Derived from
566 // 50-warehouse specjbb runs on a 2-way 1.8ghz opteron using a 4gb heap.
567 // Tested intervals from 128 to 2048 in increments of 64 == one cache line.
568 // 256 bytes (4 dcache lines) was the nearest runner-up to 576.
570 // gc copy/scan is disabled if prefetchw isn't supported, because
571 // Prefetch::write emits an inlined prefetchw on Linux.
572 // Do not use the 3dnow prefetchw instruction. It isn't supported on em64t.
573 // The used prefetcht0 instruction works for both amd64 and em64t.
574 static intx prefetch_copy_interval_in_bytes() {
575 intx interval = PrefetchCopyIntervalInBytes;
576 return interval >= 0 ? interval : 576;
577 }
578 static intx prefetch_scan_interval_in_bytes() {
579 intx interval = PrefetchScanIntervalInBytes;
580 return interval >= 0 ? interval : 576;
581 }
582 static intx prefetch_fields_ahead() {
583 intx count = PrefetchFieldsAhead;
584 return count >= 0 ? count : 1;
585 }
586 };
588 #endif // CPU_X86_VM_VM_VERSION_X86_HPP