Mercurial > hg > graal-compiler
annotate src/cpu/sparc/vm/vm_version_sparc.cpp @ 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 | f55c4f82ab9d |
| children | f95d63e2154a |
| rev | line source |
|---|---|
| 0 | 1 /* |
|
1914
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1730
diff
changeset
|
2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved. |
| 0 | 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:
1518
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1518
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:
1518
diff
changeset
|
21 * questions. |
| 0 | 22 * |
| 23 */ | |
| 24 | |
| 25 # include "incls/_precompiled.incl" | |
| 26 # include "incls/_vm_version_sparc.cpp.incl" | |
| 27 | |
| 28 int VM_Version::_features = VM_Version::unknown_m; | |
| 29 const char* VM_Version::_features_str = ""; | |
| 30 | |
|
10
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
31 bool VM_Version::is_niagara1_plus() { |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
32 // This is a placeholder until the real test is determined. |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
33 return is_niagara1() && |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
34 (os::processor_count() > maximum_niagara1_processor_count()); |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
35 } |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
36 |
| 0 | 37 void VM_Version::initialize() { |
| 38 _features = determine_features(); | |
| 39 PrefetchCopyIntervalInBytes = prefetch_copy_interval_in_bytes(); | |
| 40 PrefetchScanIntervalInBytes = prefetch_scan_interval_in_bytes(); | |
| 41 PrefetchFieldsAhead = prefetch_fields_ahead(); | |
| 42 | |
| 43 // Allocation prefetch settings | |
| 44 intx cache_line_size = L1_data_cache_line_size(); | |
| 45 if( cache_line_size > AllocatePrefetchStepSize ) | |
| 46 AllocatePrefetchStepSize = cache_line_size; | |
| 47 if( FLAG_IS_DEFAULT(AllocatePrefetchLines) ) | |
| 48 AllocatePrefetchLines = 3; // Optimistic value | |
| 49 assert( AllocatePrefetchLines > 0, "invalid value"); | |
| 50 if( AllocatePrefetchLines < 1 ) // set valid value in product VM | |
| 51 AllocatePrefetchLines = 1; // Conservative value | |
| 52 | |
| 53 AllocatePrefetchDistance = allocate_prefetch_distance(); | |
| 54 AllocatePrefetchStyle = allocate_prefetch_style(); | |
| 55 | |
| 56 assert(AllocatePrefetchDistance % AllocatePrefetchStepSize == 0, "invalid value"); | |
| 57 | |
| 58 UseSSE = 0; // Only on x86 and x64 | |
| 59 | |
| 60 _supports_cx8 = has_v9(); | |
| 61 | |
| 62 if (is_niagara1()) { | |
| 63 // Indirect branch is the same cost as direct | |
| 64 if (FLAG_IS_DEFAULT(UseInlineCaches)) { | |
| 675 | 65 FLAG_SET_DEFAULT(UseInlineCaches, false); |
| 0 | 66 } |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
10
diff
changeset
|
67 #ifdef _LP64 |
|
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
641
diff
changeset
|
68 // 32-bit oops don't make sense for the 64-bit VM on sparc |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
641
diff
changeset
|
69 // since the 32-bit VM has the same registers and smaller objects. |
|
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
641
diff
changeset
|
70 Universe::set_narrow_oop_shift(LogMinObjAlignmentInBytes); |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
10
diff
changeset
|
71 #endif // _LP64 |
| 0 | 72 #ifdef COMPILER2 |
| 73 // Indirect branch is the same cost as direct | |
| 74 if (FLAG_IS_DEFAULT(UseJumpTables)) { | |
| 675 | 75 FLAG_SET_DEFAULT(UseJumpTables, true); |
| 0 | 76 } |
| 77 // Single-issue, so entry and loop tops are | |
| 78 // aligned on a single instruction boundary | |
| 79 if (FLAG_IS_DEFAULT(InteriorEntryAlignment)) { | |
| 675 | 80 FLAG_SET_DEFAULT(InteriorEntryAlignment, 4); |
| 0 | 81 } |
|
1367
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
82 if (is_niagara1_plus()) { |
|
1914
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1730
diff
changeset
|
83 if (has_blk_init() && AllocatePrefetchStyle > 0 && |
|
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1730
diff
changeset
|
84 FLAG_IS_DEFAULT(AllocatePrefetchStyle)) { |
|
1367
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
85 // Use BIS instruction for allocation prefetch. |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
86 FLAG_SET_DEFAULT(AllocatePrefetchStyle, 3); |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
87 if (FLAG_IS_DEFAULT(AllocatePrefetchDistance)) { |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
88 // Use smaller prefetch distance on N2 with BIS |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
89 FLAG_SET_DEFAULT(AllocatePrefetchDistance, 64); |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
90 } |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
91 } |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
92 if (AllocatePrefetchStyle != 3 && FLAG_IS_DEFAULT(AllocatePrefetchDistance)) { |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
93 // Use different prefetch distance without BIS |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
94 FLAG_SET_DEFAULT(AllocatePrefetchDistance, 256); |
|
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1365
diff
changeset
|
95 } |
| 0 | 96 } |
| 97 #endif | |
| 1365 | 98 if (FLAG_IS_DEFAULT(OptoLoopAlignment)) { |
| 99 FLAG_SET_DEFAULT(OptoLoopAlignment, 4); | |
| 100 } | |
|
1518
3bfae429e2cf
6948537: CMS: BOT walkers observe out-of-thin-air zeros on sun4v sparc/CMT
ysr
parents:
1367
diff
changeset
|
101 // When using CMS, we cannot use memset() in BOT updates because |
|
3bfae429e2cf
6948537: CMS: BOT walkers observe out-of-thin-air zeros on sun4v sparc/CMT
ysr
parents:
1367
diff
changeset
|
102 // the sun4v/CMT version in libc_psr uses BIS which exposes |
|
3bfae429e2cf
6948537: CMS: BOT walkers observe out-of-thin-air zeros on sun4v sparc/CMT
ysr
parents:
1367
diff
changeset
|
103 // "phantom zeros" to concurrent readers. See 6948537. |
|
3bfae429e2cf
6948537: CMS: BOT walkers observe out-of-thin-air zeros on sun4v sparc/CMT
ysr
parents:
1367
diff
changeset
|
104 if (FLAG_IS_DEFAULT(UseMemSetInBOT) && UseConcMarkSweepGC) { |
|
3bfae429e2cf
6948537: CMS: BOT walkers observe out-of-thin-air zeros on sun4v sparc/CMT
ysr
parents:
1367
diff
changeset
|
105 FLAG_SET_DEFAULT(UseMemSetInBOT, false); |
|
3bfae429e2cf
6948537: CMS: BOT walkers observe out-of-thin-air zeros on sun4v sparc/CMT
ysr
parents:
1367
diff
changeset
|
106 } |
| 0 | 107 } |
| 108 | |
|
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
109 // Use hardware population count instruction if available. |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
110 if (has_hardware_popc()) { |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
111 if (FLAG_IS_DEFAULT(UsePopCountInstruction)) { |
| 675 | 112 FLAG_SET_DEFAULT(UsePopCountInstruction, true); |
|
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
113 } |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
114 } |
|
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
115 |
|
1730
f55c4f82ab9d
6978249: spill between cpu and fpu registers when those moves are fast
never
parents:
1579
diff
changeset
|
116 #ifdef COMPILER2 |
|
f55c4f82ab9d
6978249: spill between cpu and fpu registers when those moves are fast
never
parents:
1579
diff
changeset
|
117 // Currently not supported anywhere. |
|
f55c4f82ab9d
6978249: spill between cpu and fpu registers when those moves are fast
never
parents:
1579
diff
changeset
|
118 FLAG_SET_DEFAULT(UseFPUForSpilling, false); |
|
f55c4f82ab9d
6978249: spill between cpu and fpu registers when those moves are fast
never
parents:
1579
diff
changeset
|
119 #endif |
|
f55c4f82ab9d
6978249: spill between cpu and fpu registers when those moves are fast
never
parents:
1579
diff
changeset
|
120 |
| 0 | 121 char buf[512]; |
|
1914
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1730
diff
changeset
|
122 jio_snprintf(buf, sizeof(buf), "%s%s%s%s%s%s%s%s%s%s%s%s%s%s", |
| 0 | 123 (has_v8() ? ", has_v8" : ""), |
| 124 (has_v9() ? ", has_v9" : ""), | |
|
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
125 (has_hardware_popc() ? ", popc" : ""), |
| 0 | 126 (has_vis1() ? ", has_vis1" : ""), |
| 127 (has_vis2() ? ", has_vis2" : ""), | |
|
1914
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1730
diff
changeset
|
128 (has_blk_init() ? ", has_blk_init" : ""), |
| 0 | 129 (is_ultra3() ? ", is_ultra3" : ""), |
| 130 (is_sun4v() ? ", is_sun4v" : ""), | |
| 131 (is_niagara1() ? ", is_niagara1" : ""), | |
|
641
6af0a709d52b
6812587: Use auxv to determine SPARC hardware features on Solaris
twisti
parents:
196
diff
changeset
|
132 (is_niagara1_plus() ? ", is_niagara1_plus" : ""), |
|
1914
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1730
diff
changeset
|
133 (is_sparc64() ? ", is_sparc64" : ""), |
|
641
6af0a709d52b
6812587: Use auxv to determine SPARC hardware features on Solaris
twisti
parents:
196
diff
changeset
|
134 (!has_hardware_mul32() ? ", no-mul32" : ""), |
|
6af0a709d52b
6812587: Use auxv to determine SPARC hardware features on Solaris
twisti
parents:
196
diff
changeset
|
135 (!has_hardware_div32() ? ", no-div32" : ""), |
| 0 | 136 (!has_hardware_fsmuld() ? ", no-fsmuld" : "")); |
| 137 | |
| 138 // buf is started with ", " or is empty | |
| 139 _features_str = strdup(strlen(buf) > 2 ? buf + 2 : buf); | |
| 140 | |
| 141 #ifndef PRODUCT | |
| 142 if (PrintMiscellaneous && Verbose) { | |
| 143 tty->print("Allocation: "); | |
| 144 if (AllocatePrefetchStyle <= 0) { | |
| 145 tty->print_cr("no prefetching"); | |
| 146 } else { | |
| 147 if (AllocatePrefetchLines > 1) { | |
| 148 tty->print_cr("PREFETCH %d, %d lines of size %d bytes", AllocatePrefetchDistance, AllocatePrefetchLines, AllocatePrefetchStepSize); | |
| 149 } else { | |
| 150 tty->print_cr("PREFETCH %d, one line", AllocatePrefetchDistance); | |
| 151 } | |
| 152 } | |
| 153 if (PrefetchCopyIntervalInBytes > 0) { | |
| 154 tty->print_cr("PrefetchCopyIntervalInBytes %d", PrefetchCopyIntervalInBytes); | |
| 155 } | |
| 156 if (PrefetchScanIntervalInBytes > 0) { | |
| 157 tty->print_cr("PrefetchScanIntervalInBytes %d", PrefetchScanIntervalInBytes); | |
| 158 } | |
| 159 if (PrefetchFieldsAhead > 0) { | |
| 160 tty->print_cr("PrefetchFieldsAhead %d", PrefetchFieldsAhead); | |
| 161 } | |
| 162 } | |
| 163 #endif // PRODUCT | |
| 164 } | |
| 165 | |
| 166 void VM_Version::print_features() { | |
| 167 tty->print_cr("Version:%s", cpu_features()); | |
| 168 } | |
| 169 | |
| 170 int VM_Version::determine_features() { | |
| 171 if (UseV8InstrsOnly) { | |
| 172 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-V8");) | |
| 173 return generic_v8_m; | |
| 174 } | |
| 175 | |
| 176 int features = platform_features(unknown_m); // platform_features() is os_arch specific | |
| 177 | |
| 178 if (features == unknown_m) { | |
| 179 features = generic_v9_m; | |
| 180 warning("Cannot recognize SPARC version. Default to V9"); | |
| 181 } | |
| 182 | |
| 183 if (UseNiagaraInstrs) { | |
| 184 if (is_niagara1(features)) { | |
| 185 // Happy to accomodate... | |
| 186 } else { | |
| 187 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-Niagara");) | |
| 188 features = niagara1_m; | |
| 189 } | |
| 190 } else { | |
| 191 if (is_niagara1(features) && !FLAG_IS_DEFAULT(UseNiagaraInstrs)) { | |
| 192 NOT_PRODUCT(if (PrintMiscellaneous && Verbose) tty->print_cr("Version is Forced-Not-Niagara");) | |
| 193 features &= ~niagara1_unique_m; | |
| 194 } else { | |
| 195 // Happy to accomodate... | |
| 196 } | |
| 197 } | |
| 198 | |
| 199 return features; | |
| 200 } | |
| 201 | |
| 202 static int saved_features = 0; | |
| 203 | |
| 204 void VM_Version::allow_all() { | |
| 205 saved_features = _features; | |
| 206 _features = all_features_m; | |
| 207 } | |
| 208 | |
| 209 void VM_Version::revert() { | |
| 210 _features = saved_features; | |
| 211 } | |
|
10
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
212 |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
213 unsigned int VM_Version::calc_parallel_worker_threads() { |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
214 unsigned int result; |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
215 if (is_niagara1_plus()) { |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
216 result = nof_parallel_worker_threads(5, 16, 8); |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
217 } else { |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
218 result = nof_parallel_worker_threads(5, 8, 8); |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
219 } |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
220 return result; |
|
28372612af5e
6362677: Change parallel GC collector default number of parallel GC threads.
jmasa
parents:
0
diff
changeset
|
221 } |