Thu, 20 Mar 2014 11:03:06 +0100
8037915: PPC64/AIX: Several smaller fixes
Reviewed-by: kvn
1 /*
2 * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
3 * Copyright 2012, 2014 SAP AG. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
26 #include "precompiled.hpp"
27 #include "asm/assembler.inline.hpp"
28 #include "asm/macroAssembler.inline.hpp"
29 #include "compiler/disassembler.hpp"
30 #include "memory/resourceArea.hpp"
31 #include "runtime/java.hpp"
32 #include "runtime/stubCodeGenerator.hpp"
33 #include "utilities/defaultStream.hpp"
34 #include "vm_version_ppc.hpp"
35 #ifdef TARGET_OS_FAMILY_aix
36 # include "os_aix.inline.hpp"
37 #endif
38 #ifdef TARGET_OS_FAMILY_linux
39 # include "os_linux.inline.hpp"
40 #endif
42 # include <sys/sysinfo.h>
44 int VM_Version::_features = VM_Version::unknown_m;
45 int VM_Version::_measured_cache_line_size = 128; // default value
46 const char* VM_Version::_features_str = "";
47 bool VM_Version::_is_determine_features_test_running = false;
50 #define MSG(flag) \
51 if (flag && !FLAG_IS_DEFAULT(flag)) \
52 jio_fprintf(defaultStream::error_stream(), \
53 "warning: -XX:+" #flag " requires -XX:+UseSIGTRAP\n" \
54 " -XX:+" #flag " will be disabled!\n");
56 void VM_Version::initialize() {
58 // Test which instructions are supported and measure cache line size.
59 determine_features();
61 // If PowerArchitecturePPC64 hasn't been specified explicitly determine from features.
62 if (FLAG_IS_DEFAULT(PowerArchitecturePPC64)) {
63 if (VM_Version::has_popcntw()) {
64 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 7);
65 } else if (VM_Version::has_cmpb()) {
66 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 6);
67 } else if (VM_Version::has_popcntb()) {
68 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 5);
69 } else {
70 FLAG_SET_ERGO(uintx, PowerArchitecturePPC64, 0);
71 }
72 }
73 guarantee(PowerArchitecturePPC64 == 0 || PowerArchitecturePPC64 == 5 ||
74 PowerArchitecturePPC64 == 6 || PowerArchitecturePPC64 == 7,
75 "PowerArchitecturePPC64 should be 0, 5, 6 or 7");
77 if (!UseSIGTRAP) {
78 MSG(TrapBasedICMissChecks);
79 MSG(TrapBasedNotEntrantChecks);
80 MSG(TrapBasedNullChecks);
81 FLAG_SET_ERGO(bool, TrapBasedNotEntrantChecks, false);
82 FLAG_SET_ERGO(bool, TrapBasedNullChecks, false);
83 FLAG_SET_ERGO(bool, TrapBasedICMissChecks, false);
84 }
86 #ifdef COMPILER2
87 if (!UseSIGTRAP) {
88 MSG(TrapBasedRangeChecks);
89 FLAG_SET_ERGO(bool, TrapBasedRangeChecks, false);
90 }
92 // On Power6 test for section size.
93 if (PowerArchitecturePPC64 == 6) {
94 determine_section_size();
95 // TODO: PPC port } else {
96 // TODO: PPC port PdScheduling::power6SectorSize = 0x20;
97 }
99 MaxVectorSize = 8;
100 #endif
102 // Create and print feature-string.
103 char buf[(num_features+1) * 16]; // Max 16 chars per feature.
104 jio_snprintf(buf, sizeof(buf),
105 "ppc64%s%s%s%s%s%s%s%s",
106 (has_fsqrt() ? " fsqrt" : ""),
107 (has_isel() ? " isel" : ""),
108 (has_lxarxeh() ? " lxarxeh" : ""),
109 (has_cmpb() ? " cmpb" : ""),
110 //(has_mftgpr()? " mftgpr" : ""),
111 (has_popcntb() ? " popcntb" : ""),
112 (has_popcntw() ? " popcntw" : ""),
113 (has_fcfids() ? " fcfids" : ""),
114 (has_vand() ? " vand" : "")
115 // Make sure number of %s matches num_features!
116 );
117 _features_str = strdup(buf);
118 NOT_PRODUCT(if (Verbose) print_features(););
120 // PPC64 supports 8-byte compare-exchange operations (see
121 // Atomic::cmpxchg and StubGenerator::generate_atomic_cmpxchg_ptr)
122 // and 'atomic long memory ops' (see Unsafe_GetLongVolatile).
123 _supports_cx8 = true;
125 UseSSE = 0; // Only on x86 and x64
127 intx cache_line_size = _measured_cache_line_size;
129 if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) AllocatePrefetchStyle = 1;
131 if (AllocatePrefetchStyle == 4) {
132 AllocatePrefetchStepSize = cache_line_size; // Need exact value.
133 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 12; // Use larger blocks by default.
134 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 2*cache_line_size; // Default is not defined?
135 } else {
136 if (cache_line_size > AllocatePrefetchStepSize) AllocatePrefetchStepSize = cache_line_size;
137 if (FLAG_IS_DEFAULT(AllocatePrefetchLines)) AllocatePrefetchLines = 3; // Optimistic value.
138 if (AllocatePrefetchDistance < 0) AllocatePrefetchDistance = 3*cache_line_size; // Default is not defined?
139 }
141 assert(AllocatePrefetchLines > 0, "invalid value");
142 if (AllocatePrefetchLines < 1) // Set valid value in product VM.
143 AllocatePrefetchLines = 1; // Conservative value.
145 if (AllocatePrefetchStyle == 3 && AllocatePrefetchDistance < cache_line_size)
146 AllocatePrefetchStyle = 1; // Fall back if inappropriate.
148 assert(AllocatePrefetchStyle >= 0, "AllocatePrefetchStyle should be positive");
149 }
151 void VM_Version::print_features() {
152 tty->print_cr("Version: %s cache_line_size = %d", cpu_features(), get_cache_line_size());
153 }
155 #ifdef COMPILER2
156 // Determine section size on power6: If section size is 8 instructions,
157 // there should be a difference between the two testloops of ~15 %. If
158 // no difference is detected the section is assumed to be 32 instructions.
159 void VM_Version::determine_section_size() {
161 int unroll = 80;
163 const int code_size = (2* unroll * 32 + 100)*BytesPerInstWord;
165 // Allocate space for the code.
166 ResourceMark rm;
167 CodeBuffer cb("detect_section_size", code_size, 0);
168 MacroAssembler* a = new MacroAssembler(&cb);
170 uint32_t *code = (uint32_t *)a->pc();
171 // Emit code.
172 void (*test1)() = (void(*)())(void *)a->function_entry();
174 Label l1;
176 a->li(R4, 1);
177 a->sldi(R4, R4, 28);
178 a->b(l1);
179 a->align(CodeEntryAlignment);
181 a->bind(l1);
183 for (int i = 0; i < unroll; i++) {
184 // Schleife 1
185 // ------- sector 0 ------------
186 // ;; 0
187 a->nop(); // 1
188 a->fpnop0(); // 2
189 a->fpnop1(); // 3
190 a->addi(R4,R4, -1); // 4
192 // ;; 1
193 a->nop(); // 5
194 a->fmr(F6, F6); // 6
195 a->fmr(F7, F7); // 7
196 a->endgroup(); // 8
197 // ------- sector 8 ------------
199 // ;; 2
200 a->nop(); // 9
201 a->nop(); // 10
202 a->fmr(F8, F8); // 11
203 a->fmr(F9, F9); // 12
205 // ;; 3
206 a->nop(); // 13
207 a->fmr(F10, F10); // 14
208 a->fmr(F11, F11); // 15
209 a->endgroup(); // 16
210 // -------- sector 16 -------------
212 // ;; 4
213 a->nop(); // 17
214 a->nop(); // 18
215 a->fmr(F15, F15); // 19
216 a->fmr(F16, F16); // 20
218 // ;; 5
219 a->nop(); // 21
220 a->fmr(F17, F17); // 22
221 a->fmr(F18, F18); // 23
222 a->endgroup(); // 24
223 // ------- sector 24 ------------
225 // ;; 6
226 a->nop(); // 25
227 a->nop(); // 26
228 a->fmr(F19, F19); // 27
229 a->fmr(F20, F20); // 28
231 // ;; 7
232 a->nop(); // 29
233 a->fmr(F21, F21); // 30
234 a->fmr(F22, F22); // 31
235 a->brnop0(); // 32
237 // ------- sector 32 ------------
238 }
240 // ;; 8
241 a->cmpdi(CCR0, R4, unroll); // 33
242 a->bge(CCR0, l1); // 34
243 a->blr();
245 // Emit code.
246 void (*test2)() = (void(*)())(void *)a->function_entry();
247 // uint32_t *code = (uint32_t *)a->pc();
249 Label l2;
251 a->li(R4, 1);
252 a->sldi(R4, R4, 28);
253 a->b(l2);
254 a->align(CodeEntryAlignment);
256 a->bind(l2);
258 for (int i = 0; i < unroll; i++) {
259 // Schleife 2
260 // ------- sector 0 ------------
261 // ;; 0
262 a->brnop0(); // 1
263 a->nop(); // 2
264 //a->cmpdi(CCR0, R4, unroll);
265 a->fpnop0(); // 3
266 a->fpnop1(); // 4
267 a->addi(R4,R4, -1); // 5
269 // ;; 1
271 a->nop(); // 6
272 a->fmr(F6, F6); // 7
273 a->fmr(F7, F7); // 8
274 // ------- sector 8 ---------------
276 // ;; 2
277 a->endgroup(); // 9
279 // ;; 3
280 a->nop(); // 10
281 a->nop(); // 11
282 a->fmr(F8, F8); // 12
284 // ;; 4
285 a->fmr(F9, F9); // 13
286 a->nop(); // 14
287 a->fmr(F10, F10); // 15
289 // ;; 5
290 a->fmr(F11, F11); // 16
291 // -------- sector 16 -------------
293 // ;; 6
294 a->endgroup(); // 17
296 // ;; 7
297 a->nop(); // 18
298 a->nop(); // 19
299 a->fmr(F15, F15); // 20
301 // ;; 8
302 a->fmr(F16, F16); // 21
303 a->nop(); // 22
304 a->fmr(F17, F17); // 23
306 // ;; 9
307 a->fmr(F18, F18); // 24
308 // -------- sector 24 -------------
310 // ;; 10
311 a->endgroup(); // 25
313 // ;; 11
314 a->nop(); // 26
315 a->nop(); // 27
316 a->fmr(F19, F19); // 28
318 // ;; 12
319 a->fmr(F20, F20); // 29
320 a->nop(); // 30
321 a->fmr(F21, F21); // 31
323 // ;; 13
324 a->fmr(F22, F22); // 32
325 }
327 // -------- sector 32 -------------
328 // ;; 14
329 a->cmpdi(CCR0, R4, unroll); // 33
330 a->bge(CCR0, l2); // 34
332 a->blr();
333 uint32_t *code_end = (uint32_t *)a->pc();
334 a->flush();
336 double loop1_seconds,loop2_seconds, rel_diff;
337 uint64_t start1, stop1;
339 start1 = os::current_thread_cpu_time(false);
340 (*test1)();
341 stop1 = os::current_thread_cpu_time(false);
342 loop1_seconds = (stop1- start1) / (1000 *1000 *1000.0);
345 start1 = os::current_thread_cpu_time(false);
346 (*test2)();
347 stop1 = os::current_thread_cpu_time(false);
349 loop2_seconds = (stop1 - start1) / (1000 *1000 *1000.0);
351 rel_diff = (loop2_seconds - loop1_seconds) / loop1_seconds *100;
353 if (PrintAssembly) {
354 ttyLocker ttyl;
355 tty->print_cr("Decoding section size detection stub at " INTPTR_FORMAT " before execution:", code);
356 Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
357 tty->print_cr("Time loop1 :%f", loop1_seconds);
358 tty->print_cr("Time loop2 :%f", loop2_seconds);
359 tty->print_cr("(time2 - time1) / time1 = %f %%", rel_diff);
361 if (rel_diff > 12.0) {
362 tty->print_cr("Section Size 8 Instructions");
363 } else{
364 tty->print_cr("Section Size 32 Instructions or Power5");
365 }
366 }
368 #if 0 // TODO: PPC port
369 // Set sector size (if not set explicitly).
370 if (FLAG_IS_DEFAULT(Power6SectorSize128PPC64)) {
371 if (rel_diff > 12.0) {
372 PdScheduling::power6SectorSize = 0x20;
373 } else {
374 PdScheduling::power6SectorSize = 0x80;
375 }
376 } else if (Power6SectorSize128PPC64) {
377 PdScheduling::power6SectorSize = 0x80;
378 } else {
379 PdScheduling::power6SectorSize = 0x20;
380 }
381 #endif
382 if (UsePower6SchedulerPPC64) Unimplemented();
383 }
384 #endif // COMPILER2
386 void VM_Version::determine_features() {
387 #if defined(ABI_ELFv2)
388 const int code_size = (num_features+1+2*7)*BytesPerInstWord; // TODO(asmundak): calculation is incorrect.
389 #else
390 // 7 InstWords for each call (function descriptor + blr instruction).
391 const int code_size = (num_features+1+2*7)*BytesPerInstWord;
392 #endif
393 int features = 0;
395 // create test area
396 enum { BUFFER_SIZE = 2*4*K }; // Needs to be >=2* max cache line size (cache line size can't exceed min page size).
397 char test_area[BUFFER_SIZE];
398 char *mid_of_test_area = &test_area[BUFFER_SIZE>>1];
400 // Allocate space for the code.
401 ResourceMark rm;
402 CodeBuffer cb("detect_cpu_features", code_size, 0);
403 MacroAssembler* a = new MacroAssembler(&cb);
405 // Must be set to true so we can generate the test code.
406 _features = VM_Version::all_features_m;
408 // Emit code.
409 void (*test)(address addr, uint64_t offset)=(void(*)(address addr, uint64_t offset))(void *)a->function_entry();
410 uint32_t *code = (uint32_t *)a->pc();
411 // Don't use R0 in ldarx.
412 // Keep R3_ARG1 unmodified, it contains &field (see below).
413 // Keep R4_ARG2 unmodified, it contains offset = 0 (see below).
414 a->fsqrt(F3, F4); // code[0] -> fsqrt_m
415 a->fsqrts(F3, F4); // code[1] -> fsqrts_m
416 a->isel(R7, R5, R6, 0); // code[2] -> isel_m
417 a->ldarx_unchecked(R7, R3_ARG1, R4_ARG2, 1); // code[3] -> lxarx_m
418 a->cmpb(R7, R5, R6); // code[4] -> bcmp
419 //a->mftgpr(R7, F3); // code[5] -> mftgpr
420 a->popcntb(R7, R5); // code[6] -> popcntb
421 a->popcntw(R7, R5); // code[7] -> popcntw
422 a->fcfids(F3, F4); // code[8] -> fcfids
423 a->vand(VR0, VR0, VR0); // code[9] -> vand
424 a->blr();
426 // Emit function to set one cache line to zero. Emit function descriptor and get pointer to it.
427 void (*zero_cacheline_func_ptr)(char*) = (void(*)(char*))(void *)a->function_entry();
428 a->dcbz(R3_ARG1); // R3_ARG1 = addr
429 a->blr();
431 uint32_t *code_end = (uint32_t *)a->pc();
432 a->flush();
433 _features = VM_Version::unknown_m;
435 // Print the detection code.
436 if (PrintAssembly) {
437 ttyLocker ttyl;
438 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " before execution:", code);
439 Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
440 }
442 // Measure cache line size.
443 memset(test_area, 0xFF, BUFFER_SIZE); // Fill test area with 0xFF.
444 (*zero_cacheline_func_ptr)(mid_of_test_area); // Call function which executes dcbz to the middle.
445 int count = 0; // count zeroed bytes
446 for (int i = 0; i < BUFFER_SIZE; i++) if (test_area[i] == 0) count++;
447 guarantee(is_power_of_2(count), "cache line size needs to be a power of 2");
448 _measured_cache_line_size = count;
450 // Execute code. Illegal instructions will be replaced by 0 in the signal handler.
451 VM_Version::_is_determine_features_test_running = true;
452 (*test)((address)mid_of_test_area, (uint64_t)0);
453 VM_Version::_is_determine_features_test_running = false;
455 // determine which instructions are legal.
456 int feature_cntr = 0;
457 if (code[feature_cntr++]) features |= fsqrt_m;
458 if (code[feature_cntr++]) features |= fsqrts_m;
459 if (code[feature_cntr++]) features |= isel_m;
460 if (code[feature_cntr++]) features |= lxarxeh_m;
461 if (code[feature_cntr++]) features |= cmpb_m;
462 //if(code[feature_cntr++])features |= mftgpr_m;
463 if (code[feature_cntr++]) features |= popcntb_m;
464 if (code[feature_cntr++]) features |= popcntw_m;
465 if (code[feature_cntr++]) features |= fcfids_m;
466 if (code[feature_cntr++]) features |= vand_m;
468 // Print the detection code.
469 if (PrintAssembly) {
470 ttyLocker ttyl;
471 tty->print_cr("Decoding cpu-feature detection stub at " INTPTR_FORMAT " after execution:", code);
472 Disassembler::decode((u_char*)code, (u_char*)code_end, tty);
473 }
475 _features = features;
476 }
479 static int saved_features = 0;
481 void VM_Version::allow_all() {
482 saved_features = _features;
483 _features = all_features_m;
484 }
486 void VM_Version::revert() {
487 _features = saved_features;
488 }