Mercurial > hg > truffle
annotate src/cpu/x86/vm/vm_version_x86.hpp @ 1914:ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
Summary: Use hardware DIV instruction for long division by constant when it is faster than code with multiply.
Reviewed-by: never
author | kvn |
---|---|
date | Tue, 02 Nov 2010 09:00:37 -0700 |
parents | a83b0246bb77 |
children | f95d63e2154a |
rev | line source |
---|---|
585 | 1 /* |
1622 | 2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All Rights Reserved. |
585 | 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 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
775
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
775
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
775
diff
changeset
|
21 * questions. |
585 | 22 * |
23 */ | |
24 | |
25 class VM_Version : public Abstract_VM_Version { | |
26 public: | |
27 // cpuid result register layouts. These are all unions of a uint32_t | |
28 // (in case anyone wants access to the register as a whole) and a bitfield. | |
29 | |
30 union StdCpuid1Eax { | |
31 uint32_t value; | |
32 struct { | |
33 uint32_t stepping : 4, | |
34 model : 4, | |
35 family : 4, | |
36 proc_type : 2, | |
37 : 2, | |
38 ext_model : 4, | |
39 ext_family : 8, | |
40 : 4; | |
41 } bits; | |
42 }; | |
43 | |
44 union StdCpuid1Ebx { // example, unused | |
45 uint32_t value; | |
46 struct { | |
47 uint32_t brand_id : 8, | |
48 clflush_size : 8, | |
49 threads_per_cpu : 8, | |
50 apic_id : 8; | |
51 } bits; | |
52 }; | |
53 | |
54 union StdCpuid1Ecx { | |
55 uint32_t value; | |
56 struct { | |
57 uint32_t sse3 : 1, | |
58 : 2, | |
59 monitor : 1, | |
60 : 1, | |
61 vmx : 1, | |
62 : 1, | |
63 est : 1, | |
64 : 1, | |
65 ssse3 : 1, | |
66 cid : 1, | |
67 : 2, | |
68 cmpxchg16: 1, | |
69 : 4, | |
70 dca : 1, | |
71 sse4_1 : 1, | |
72 sse4_2 : 1, | |
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
585
diff
changeset
|
73 : 2, |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
585
diff
changeset
|
74 popcnt : 1, |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
585
diff
changeset
|
75 : 8; |
585 | 76 } bits; |
77 }; | |
78 | |
79 union StdCpuid1Edx { | |
80 uint32_t value; | |
81 struct { | |
82 uint32_t : 4, | |
83 tsc : 1, | |
84 : 3, | |
85 cmpxchg8 : 1, | |
86 : 6, | |
87 cmov : 1, | |
88 : 7, | |
89 mmx : 1, | |
90 fxsr : 1, | |
91 sse : 1, | |
92 sse2 : 1, | |
93 : 1, | |
94 ht : 1, | |
95 : 3; | |
96 } bits; | |
97 }; | |
98 | |
99 union DcpCpuid4Eax { | |
100 uint32_t value; | |
101 struct { | |
102 uint32_t cache_type : 5, | |
103 : 21, | |
104 cores_per_cpu : 6; | |
105 } bits; | |
106 }; | |
107 | |
108 union DcpCpuid4Ebx { | |
109 uint32_t value; | |
110 struct { | |
111 uint32_t L1_line_size : 12, | |
112 partitions : 10, | |
113 associativity : 10; | |
114 } bits; | |
115 }; | |
116 | |
1622 | 117 union TplCpuidBEbx { |
118 uint32_t value; | |
119 struct { | |
120 uint32_t logical_cpus : 16, | |
121 : 16; | |
122 } bits; | |
123 }; | |
124 | |
585 | 125 union ExtCpuid1Ecx { |
126 uint32_t value; | |
127 struct { | |
128 uint32_t LahfSahf : 1, | |
129 CmpLegacy : 1, | |
130 : 4, | |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
131 lzcnt : 1, |
585 | 132 sse4a : 1, |
133 misalignsse : 1, | |
134 prefetchw : 1, | |
135 : 22; | |
136 } bits; | |
137 }; | |
138 | |
139 union ExtCpuid1Edx { | |
140 uint32_t value; | |
141 struct { | |
142 uint32_t : 22, | |
143 mmx_amd : 1, | |
144 mmx : 1, | |
145 fxsr : 1, | |
146 : 4, | |
147 long_mode : 1, | |
148 tdnow2 : 1, | |
149 tdnow : 1; | |
150 } bits; | |
151 }; | |
152 | |
153 union ExtCpuid5Ex { | |
154 uint32_t value; | |
155 struct { | |
156 uint32_t L1_line_size : 8, | |
157 L1_tag_lines : 8, | |
158 L1_assoc : 8, | |
159 L1_size : 8; | |
160 } bits; | |
161 }; | |
162 | |
163 union ExtCpuid8Ecx { | |
164 uint32_t value; | |
165 struct { | |
166 uint32_t cores_per_cpu : 8, | |
167 : 24; | |
168 } bits; | |
169 }; | |
170 | |
171 protected: | |
172 static int _cpu; | |
173 static int _model; | |
174 static int _stepping; | |
175 static int _cpuFeatures; // features returned by the "cpuid" instruction | |
176 // 0 if this instruction is not available | |
177 static const char* _features_str; | |
178 | |
179 enum { | |
180 CPU_CX8 = (1 << 0), // next bits are from cpuid 1 (EDX) | |
181 CPU_CMOV = (1 << 1), | |
182 CPU_FXSR = (1 << 2), | |
183 CPU_HT = (1 << 3), | |
184 CPU_MMX = (1 << 4), | |
185 CPU_3DNOW = (1 << 5), // 3DNow comes from cpuid 0x80000001 (EDX) | |
186 CPU_SSE = (1 << 6), | |
187 CPU_SSE2 = (1 << 7), | |
188 CPU_SSE3 = (1 << 8), // SSE3 comes from cpuid 1 (ECX) | |
189 CPU_SSSE3 = (1 << 9), | |
190 CPU_SSE4A = (1 << 10), | |
191 CPU_SSE4_1 = (1 << 11), | |
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
585
diff
changeset
|
192 CPU_SSE4_2 = (1 << 12), |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
193 CPU_POPCNT = (1 << 13), |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
194 CPU_LZCNT = (1 << 14) |
585 | 195 } cpuFeatureFlags; |
196 | |
197 // cpuid information block. All info derived from executing cpuid with | |
198 // various function numbers is stored here. Intel and AMD info is | |
199 // merged in this block: accessor methods disentangle it. | |
200 // | |
201 // The info block is laid out in subblocks of 4 dwords corresponding to | |
202 // eax, ebx, ecx and edx, whether or not they contain anything useful. | |
203 struct CpuidInfo { | |
204 // cpuid function 0 | |
205 uint32_t std_max_function; | |
206 uint32_t std_vendor_name_0; | |
207 uint32_t std_vendor_name_1; | |
208 uint32_t std_vendor_name_2; | |
209 | |
210 // cpuid function 1 | |
211 StdCpuid1Eax std_cpuid1_eax; | |
212 StdCpuid1Ebx std_cpuid1_ebx; | |
213 StdCpuid1Ecx std_cpuid1_ecx; | |
214 StdCpuid1Edx std_cpuid1_edx; | |
215 | |
216 // cpuid function 4 (deterministic cache parameters) | |
217 DcpCpuid4Eax dcp_cpuid4_eax; | |
218 DcpCpuid4Ebx dcp_cpuid4_ebx; | |
219 uint32_t dcp_cpuid4_ecx; // unused currently | |
220 uint32_t dcp_cpuid4_edx; // unused currently | |
221 | |
1622 | 222 // cpuid function 0xB (processor topology) |
223 // ecx = 0 | |
224 uint32_t tpl_cpuidB0_eax; | |
225 TplCpuidBEbx tpl_cpuidB0_ebx; | |
226 uint32_t tpl_cpuidB0_ecx; // unused currently | |
227 uint32_t tpl_cpuidB0_edx; // unused currently | |
228 | |
229 // ecx = 1 | |
230 uint32_t tpl_cpuidB1_eax; | |
231 TplCpuidBEbx tpl_cpuidB1_ebx; | |
232 uint32_t tpl_cpuidB1_ecx; // unused currently | |
233 uint32_t tpl_cpuidB1_edx; // unused currently | |
234 | |
235 // ecx = 2 | |
236 uint32_t tpl_cpuidB2_eax; | |
237 TplCpuidBEbx tpl_cpuidB2_ebx; | |
238 uint32_t tpl_cpuidB2_ecx; // unused currently | |
239 uint32_t tpl_cpuidB2_edx; // unused currently | |
240 | |
585 | 241 // cpuid function 0x80000000 // example, unused |
242 uint32_t ext_max_function; | |
243 uint32_t ext_vendor_name_0; | |
244 uint32_t ext_vendor_name_1; | |
245 uint32_t ext_vendor_name_2; | |
246 | |
247 // cpuid function 0x80000001 | |
248 uint32_t ext_cpuid1_eax; // reserved | |
249 uint32_t ext_cpuid1_ebx; // reserved | |
250 ExtCpuid1Ecx ext_cpuid1_ecx; | |
251 ExtCpuid1Edx ext_cpuid1_edx; | |
252 | |
253 // cpuid functions 0x80000002 thru 0x80000004: example, unused | |
254 uint32_t proc_name_0, proc_name_1, proc_name_2, proc_name_3; | |
255 uint32_t proc_name_4, proc_name_5, proc_name_6, proc_name_7; | |
256 uint32_t proc_name_8, proc_name_9, proc_name_10,proc_name_11; | |
257 | |
258 // cpuid function 0x80000005 //AMD L1, Intel reserved | |
259 uint32_t ext_cpuid5_eax; // unused currently | |
260 uint32_t ext_cpuid5_ebx; // reserved | |
261 ExtCpuid5Ex ext_cpuid5_ecx; // L1 data cache info (AMD) | |
262 ExtCpuid5Ex ext_cpuid5_edx; // L1 instruction cache info (AMD) | |
263 | |
264 // cpuid function 0x80000008 | |
265 uint32_t ext_cpuid8_eax; // unused currently | |
266 uint32_t ext_cpuid8_ebx; // reserved | |
267 ExtCpuid8Ecx ext_cpuid8_ecx; | |
268 uint32_t ext_cpuid8_edx; // reserved | |
269 }; | |
270 | |
271 // The actual cpuid info block | |
272 static CpuidInfo _cpuid_info; | |
273 | |
274 // Extractors and predicates | |
275 static uint32_t extended_cpu_family() { | |
276 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.family; | |
277 result += _cpuid_info.std_cpuid1_eax.bits.ext_family; | |
278 return result; | |
279 } | |
280 static uint32_t extended_cpu_model() { | |
281 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.model; | |
282 result |= _cpuid_info.std_cpuid1_eax.bits.ext_model << 4; | |
283 return result; | |
284 } | |
285 static uint32_t cpu_stepping() { | |
286 uint32_t result = _cpuid_info.std_cpuid1_eax.bits.stepping; | |
287 return result; | |
288 } | |
289 static uint logical_processor_count() { | |
290 uint result = threads_per_core(); | |
291 return result; | |
292 } | |
293 static uint32_t feature_flags() { | |
294 uint32_t result = 0; | |
295 if (_cpuid_info.std_cpuid1_edx.bits.cmpxchg8 != 0) | |
296 result |= CPU_CX8; | |
297 if (_cpuid_info.std_cpuid1_edx.bits.cmov != 0) | |
298 result |= CPU_CMOV; | |
1789
a83b0246bb77
6934483: GCC 4.5 errors "suggest parentheses around something..." when compiling with -Werror and -Wall
twisti
parents:
1647
diff
changeset
|
299 if (_cpuid_info.std_cpuid1_edx.bits.fxsr != 0 || (is_amd() && |
a83b0246bb77
6934483: GCC 4.5 errors "suggest parentheses around something..." when compiling with -Werror and -Wall
twisti
parents:
1647
diff
changeset
|
300 _cpuid_info.ext_cpuid1_edx.bits.fxsr != 0)) |
585 | 301 result |= CPU_FXSR; |
302 // HT flag is set for multi-core processors also. | |
303 if (threads_per_core() > 1) | |
304 result |= CPU_HT; | |
1789
a83b0246bb77
6934483: GCC 4.5 errors "suggest parentheses around something..." when compiling with -Werror and -Wall
twisti
parents:
1647
diff
changeset
|
305 if (_cpuid_info.std_cpuid1_edx.bits.mmx != 0 || (is_amd() && |
a83b0246bb77
6934483: GCC 4.5 errors "suggest parentheses around something..." when compiling with -Werror and -Wall
twisti
parents:
1647
diff
changeset
|
306 _cpuid_info.ext_cpuid1_edx.bits.mmx != 0)) |
585 | 307 result |= CPU_MMX; |
308 if (_cpuid_info.std_cpuid1_edx.bits.sse != 0) | |
309 result |= CPU_SSE; | |
310 if (_cpuid_info.std_cpuid1_edx.bits.sse2 != 0) | |
311 result |= CPU_SSE2; | |
312 if (_cpuid_info.std_cpuid1_ecx.bits.sse3 != 0) | |
313 result |= CPU_SSE3; | |
314 if (_cpuid_info.std_cpuid1_ecx.bits.ssse3 != 0) | |
315 result |= CPU_SSSE3; | |
316 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_1 != 0) | |
317 result |= CPU_SSE4_1; | |
318 if (_cpuid_info.std_cpuid1_ecx.bits.sse4_2 != 0) | |
319 result |= CPU_SSE4_2; | |
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
585
diff
changeset
|
320 if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0) |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
585
diff
changeset
|
321 result |= CPU_POPCNT; |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
322 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
323 // AMD features. |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
324 if (is_amd()) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
325 if (_cpuid_info.ext_cpuid1_edx.bits.tdnow != 0) |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
326 result |= CPU_3DNOW; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
327 if (_cpuid_info.ext_cpuid1_ecx.bits.lzcnt != 0) |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
328 result |= CPU_LZCNT; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
329 if (_cpuid_info.ext_cpuid1_ecx.bits.sse4a != 0) |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
330 result |= CPU_SSE4A; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
331 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
332 |
585 | 333 return result; |
334 } | |
335 | |
336 static void get_processor_features(); | |
337 | |
338 public: | |
339 // Offsets for cpuid asm stub | |
340 static ByteSize std_cpuid0_offset() { return byte_offset_of(CpuidInfo, std_max_function); } | |
341 static ByteSize std_cpuid1_offset() { return byte_offset_of(CpuidInfo, std_cpuid1_eax); } | |
342 static ByteSize dcp_cpuid4_offset() { return byte_offset_of(CpuidInfo, dcp_cpuid4_eax); } | |
343 static ByteSize ext_cpuid1_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1_eax); } | |
344 static ByteSize ext_cpuid5_offset() { return byte_offset_of(CpuidInfo, ext_cpuid5_eax); } | |
345 static ByteSize ext_cpuid8_offset() { return byte_offset_of(CpuidInfo, ext_cpuid8_eax); } | |
1622 | 346 static ByteSize tpl_cpuidB0_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB0_eax); } |
347 static ByteSize tpl_cpuidB1_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB1_eax); } | |
348 static ByteSize tpl_cpuidB2_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB2_eax); } | |
585 | 349 |
350 // Initialization | |
351 static void initialize(); | |
352 | |
353 // Asserts | |
354 static void assert_is_initialized() { | |
355 assert(_cpuid_info.std_cpuid1_eax.bits.family != 0, "VM_Version not initialized"); | |
356 } | |
357 | |
358 // | |
359 // Processor family: | |
360 // 3 - 386 | |
361 // 4 - 486 | |
362 // 5 - Pentium | |
363 // 6 - PentiumPro, Pentium II, Celeron, Xeon, Pentium III, Athlon, | |
364 // Pentium M, Core Solo, Core Duo, Core2 Duo | |
365 // family 6 model: 9, 13, 14, 15 | |
366 // 0x0f - Pentium 4, Opteron | |
367 // | |
368 // Note: The cpu family should be used to select between | |
369 // instruction sequences which are valid on all Intel | |
370 // processors. Use the feature test functions below to | |
371 // determine whether a particular instruction is supported. | |
372 // | |
373 static int cpu_family() { return _cpu;} | |
374 static bool is_P6() { return cpu_family() >= 6; } | |
375 | |
376 static bool is_amd() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x68747541; } // 'htuA' | |
377 static bool is_intel() { assert_is_initialized(); return _cpuid_info.std_vendor_name_0 == 0x756e6547; } // 'uneG' | |
378 | |
1647 | 379 static bool supports_processor_topology() { |
380 return (_cpuid_info.std_max_function >= 0xB) && | |
381 // eax[4:0] | ebx[0:15] == 0 indicates invalid topology level. | |
382 // Some cpus have max cpuid >= 0xB but do not support processor topology. | |
383 ((_cpuid_info.tpl_cpuidB0_eax & 0x1f | _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus) != 0); | |
384 } | |
385 | |
585 | 386 static uint cores_per_cpu() { |
387 uint result = 1; | |
388 if (is_intel()) { | |
1647 | 389 if (supports_processor_topology()) { |
1622 | 390 result = _cpuid_info.tpl_cpuidB1_ebx.bits.logical_cpus / |
391 _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; | |
392 } else { | |
393 result = (_cpuid_info.dcp_cpuid4_eax.bits.cores_per_cpu + 1); | |
394 } | |
585 | 395 } else if (is_amd()) { |
396 result = (_cpuid_info.ext_cpuid8_ecx.bits.cores_per_cpu + 1); | |
397 } | |
398 return result; | |
399 } | |
400 | |
401 static uint threads_per_core() { | |
402 uint result = 1; | |
1647 | 403 if (is_intel() && supports_processor_topology()) { |
1622 | 404 result = _cpuid_info.tpl_cpuidB0_ebx.bits.logical_cpus; |
405 } else if (_cpuid_info.std_cpuid1_edx.bits.ht != 0) { | |
585 | 406 result = _cpuid_info.std_cpuid1_ebx.bits.threads_per_cpu / |
407 cores_per_cpu(); | |
408 } | |
409 return result; | |
410 } | |
411 | |
412 static intx L1_data_cache_line_size() { | |
413 intx result = 0; | |
414 if (is_intel()) { | |
415 result = (_cpuid_info.dcp_cpuid4_ebx.bits.L1_line_size + 1); | |
416 } else if (is_amd()) { | |
417 result = _cpuid_info.ext_cpuid5_ecx.bits.L1_line_size; | |
418 } | |
419 if (result < 32) // not defined ? | |
420 result = 32; // 32 bytes by default on x86 and other x64 | |
421 return result; | |
422 } | |
423 | |
424 // | |
425 // Feature identification | |
426 // | |
427 static bool supports_cpuid() { return _cpuFeatures != 0; } | |
428 static bool supports_cmpxchg8() { return (_cpuFeatures & CPU_CX8) != 0; } | |
429 static bool supports_cmov() { return (_cpuFeatures & CPU_CMOV) != 0; } | |
430 static bool supports_fxsr() { return (_cpuFeatures & CPU_FXSR) != 0; } | |
431 static bool supports_ht() { return (_cpuFeatures & CPU_HT) != 0; } | |
432 static bool supports_mmx() { return (_cpuFeatures & CPU_MMX) != 0; } | |
433 static bool supports_sse() { return (_cpuFeatures & CPU_SSE) != 0; } | |
434 static bool supports_sse2() { return (_cpuFeatures & CPU_SSE2) != 0; } | |
435 static bool supports_sse3() { return (_cpuFeatures & CPU_SSE3) != 0; } | |
436 static bool supports_ssse3() { return (_cpuFeatures & CPU_SSSE3)!= 0; } | |
437 static bool supports_sse4_1() { return (_cpuFeatures & CPU_SSE4_1) != 0; } | |
438 static bool supports_sse4_2() { return (_cpuFeatures & CPU_SSE4_2) != 0; } | |
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
585
diff
changeset
|
439 static bool supports_popcnt() { return (_cpuFeatures & CPU_POPCNT) != 0; } |
585 | 440 // |
441 // AMD features | |
442 // | |
443 static bool supports_3dnow() { return (_cpuFeatures & CPU_3DNOW) != 0; } | |
444 static bool supports_mmx_ext() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.mmx_amd != 0; } | |
445 static bool supports_3dnow2() { return is_amd() && _cpuid_info.ext_cpuid1_edx.bits.tdnow2 != 0; } | |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
643
diff
changeset
|
446 static bool supports_lzcnt() { return (_cpuFeatures & CPU_LZCNT) != 0; } |
585 | 447 static bool supports_sse4a() { return (_cpuFeatures & CPU_SSE4A) != 0; } |
448 | |
1914
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1789
diff
changeset
|
449 // Intel Core and newer cpus have fast IDIV instruction (excluding Atom). |
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1789
diff
changeset
|
450 static bool has_fast_idiv() { return is_intel() && cpu_family() == 6 && |
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1789
diff
changeset
|
451 supports_sse3() && _model != 0x1C; } |
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1789
diff
changeset
|
452 |
585 | 453 static bool supports_compare_and_exchange() { return true; } |
454 | |
455 static const char* cpu_features() { return _features_str; } | |
456 | |
457 static intx allocate_prefetch_distance() { | |
458 // This method should be called before allocate_prefetch_style(). | |
459 // | |
460 // Hardware prefetching (distance/size in bytes): | |
461 // Pentium 3 - 64 / 32 | |
462 // Pentium 4 - 256 / 128 | |
463 // Athlon - 64 / 32 ???? | |
464 // Opteron - 128 / 64 only when 2 sequential cache lines accessed | |
465 // Core - 128 / 64 | |
466 // | |
467 // Software prefetching (distance in bytes / instruction with best score): | |
468 // Pentium 3 - 128 / prefetchnta | |
469 // Pentium 4 - 512 / prefetchnta | |
470 // Athlon - 128 / prefetchnta | |
471 // Opteron - 256 / prefetchnta | |
472 // Core - 256 / prefetchnta | |
473 // It will be used only when AllocatePrefetchStyle > 0 | |
474 | |
475 intx count = AllocatePrefetchDistance; | |
476 if (count < 0) { // default ? | |
477 if (is_amd()) { // AMD | |
478 if (supports_sse2()) | |
479 count = 256; // Opteron | |
480 else | |
481 count = 128; // Athlon | |
482 } else { // Intel | |
483 if (supports_sse2()) | |
484 if (cpu_family() == 6) { | |
485 count = 256; // Pentium M, Core, Core2 | |
486 } else { | |
487 count = 512; // Pentium 4 | |
488 } | |
489 else | |
490 count = 128; // Pentium 3 (and all other old CPUs) | |
491 } | |
492 } | |
493 return count; | |
494 } | |
495 static intx allocate_prefetch_style() { | |
496 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive"); | |
497 // Return 0 if AllocatePrefetchDistance was not defined. | |
498 return AllocatePrefetchDistance > 0 ? AllocatePrefetchStyle : 0; | |
499 } | |
500 | |
501 // Prefetch interval for gc copy/scan == 9 dcache lines. Derived from | |
502 // 50-warehouse specjbb runs on a 2-way 1.8ghz opteron using a 4gb heap. | |
503 // Tested intervals from 128 to 2048 in increments of 64 == one cache line. | |
504 // 256 bytes (4 dcache lines) was the nearest runner-up to 576. | |
505 | |
506 // gc copy/scan is disabled if prefetchw isn't supported, because | |
507 // Prefetch::write emits an inlined prefetchw on Linux. | |
508 // Do not use the 3dnow prefetchw instruction. It isn't supported on em64t. | |
509 // The used prefetcht0 instruction works for both amd64 and em64t. | |
510 static intx prefetch_copy_interval_in_bytes() { | |
511 intx interval = PrefetchCopyIntervalInBytes; | |
512 return interval >= 0 ? interval : 576; | |
513 } | |
514 static intx prefetch_scan_interval_in_bytes() { | |
515 intx interval = PrefetchScanIntervalInBytes; | |
516 return interval >= 0 ? interval : 576; | |
517 } | |
518 static intx prefetch_fields_ahead() { | |
519 intx count = PrefetchFieldsAhead; | |
520 return count >= 0 ? count : 1; | |
521 } | |
522 }; |