2 * Copyright (c) 2003-2005 The Regents of The University of Michigan
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are
7 * met: redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer;
9 * redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution;
12 * neither the name of the copyright holders nor the names of its
13 * contributors may be used to endorse or promote products derived from
14 * this software without specific prior written permission.
16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 * Authors: Steve Reinhardt
31 #ifndef __CPU_STATIC_INST_HH__
32 #define __CPU_STATIC_INST_HH__
37 #include "arch/isa_traits.hh"
38 #include "arch/utility.hh"
39 #include "sim/faults.hh"
40 #include "base/bitfield.hh"
41 #include "base/hashmap.hh"
42 #include "base/misc.hh"
43 #include "base/refcnt.hh"
44 #include "cpu/op_class.hh"
45 #include "cpu/o3/dyn_inst.hh"
46 #include "sim/faults.hh"
47 #include "sim/host.hh"
49 // forward declarations
50 struct AlphaSimpleImpl;
62 class AtomicSimpleCPU;
63 class TimingSimpleCPU;
72 typedef uint32_t MicroPC;
75 * Base, ISA-independent static instruction class.
77 * The main component of this class is the vector of flags and the
78 * associated methods for reading them. Any object that can rely
79 * solely on these flags can process instructions without being
80 * recompiled for multiple ISAs.
82 class StaticInstBase : public RefCounted
86 /// Set of boolean static instruction properties.
89 /// - The IsInteger and IsFloating flags are based on the class of
90 /// registers accessed by the instruction. Although most
91 /// instructions will have exactly one of these two flags set, it
92 /// is possible for an instruction to have neither (e.g., direct
93 /// unconditional branches, memory barriers) or both (e.g., an
94 /// FP/int conversion).
95 /// - If IsMemRef is set, then exactly one of IsLoad or IsStore
97 /// - If IsControl is set, then exactly one of IsDirectControl or
98 /// IsIndirect Control will be set, and exactly one of
99 /// IsCondControl or IsUncondControl will be set.
100 /// - IsSerializing, IsMemBarrier, and IsWriteBarrier are
101 /// implemented as flags since in the current model there's no
102 /// other way for instructions to inject behavior into the
103 /// pipeline outside of fetch. Once we go to an exec-in-exec CPU
104 /// model we should be able to get rid of these flags and
105 /// implement this behavior via the execute() methods.
108 IsNop, ///< Is a no-op (no effect at all).
110 IsInteger, ///< References integer regs.
111 IsFloating, ///< References FP regs.
113 IsMemRef, ///< References memory (load, store, or prefetch).
114 IsLoad, ///< Reads from memory (load or prefetch).
115 IsStore, ///< Writes to memory.
116 IsStoreConditional, ///< Store conditional instruction.
117 IsInstPrefetch, ///< Instruction-cache prefetch.
118 IsDataPrefetch, ///< Data-cache prefetch.
119 IsCopy, ///< Fast Cache block copy
121 IsControl, ///< Control transfer instruction.
122 IsDirectControl, ///< PC relative control transfer.
123 IsIndirectControl, ///< Register indirect control transfer.
124 IsCondControl, ///< Conditional control transfer.
125 IsUncondControl, ///< Unconditional control transfer.
126 IsCall, ///< Subroutine call.
127 IsReturn, ///< Subroutine return.
129 IsCondDelaySlot,///< Conditional Delay-Slot Instruction
131 IsThreadSync, ///< Thread synchronization operation.
133 IsSerializing, ///< Serializes pipeline: won't execute until all
134 /// older instructions have committed.
137 IsMemBarrier, ///< Is a memory barrier
138 IsWriteBarrier, ///< Is a write barrier
140 IsNonSpeculative, ///< Should not be executed speculatively
141 IsQuiesce, ///< Is a quiesce instruction
143 IsIprAccess, ///< Accesses IPRs
144 IsUnverifiable, ///< Can't be verified by a checker
146 //Flags for microcode
147 IsMacroop, ///< Is a macroop containing microops
148 IsMicroop, ///< Is a microop
149 IsDelayedCommit, ///< This microop doesn't commit right away
150 IsLastMicroop, ///< This microop ends a microop sequence
151 IsFirstMicroop, ///< This microop begins a microop sequence
152 //This flag doesn't do anything yet
153 IsMicroBranch, ///< This microop branches within the microcode for a macroop
158 /// Flag values for this instruction.
159 std::bitset<NumFlags> flags;
164 /// See numSrcRegs().
167 /// See numDestRegs().
170 /// The following are used to track physical register usage
171 /// for machines with separate int & FP reg files.
173 int8_t _numFPDestRegs;
174 int8_t _numIntDestRegs;
178 /// It's important to initialize everything here to a sane
179 /// default, since the decoder generally only overrides
180 /// the fields that are meaningful for the particular
182 StaticInstBase(OpClass __opClass)
183 : _opClass(__opClass), _numSrcRegs(0), _numDestRegs(0),
184 _numFPDestRegs(0), _numIntDestRegs(0)
190 /// @name Register information.
191 /// The sum of numFPDestRegs() and numIntDestRegs() equals
192 /// numDestRegs(). The former two functions are used to track
193 /// physical register usage for machines with separate int & FP
196 /// Number of source registers.
197 int8_t numSrcRegs() const { return _numSrcRegs; }
198 /// Number of destination registers.
199 int8_t numDestRegs() const { return _numDestRegs; }
200 /// Number of floating-point destination regs.
201 int8_t numFPDestRegs() const { return _numFPDestRegs; }
202 /// Number of integer destination regs.
203 int8_t numIntDestRegs() const { return _numIntDestRegs; }
206 /// @name Flag accessors.
207 /// These functions are used to access the values of the various
208 /// instruction property flags. See StaticInstBase::Flags for descriptions
209 /// of the individual flags.
212 bool isNop() const { return flags[IsNop]; }
214 bool isMemRef() const { return flags[IsMemRef]; }
215 bool isLoad() const { return flags[IsLoad]; }
216 bool isStore() const { return flags[IsStore]; }
217 bool isStoreConditional() const { return flags[IsStoreConditional]; }
218 bool isInstPrefetch() const { return flags[IsInstPrefetch]; }
219 bool isDataPrefetch() const { return flags[IsDataPrefetch]; }
220 bool isCopy() const { return flags[IsCopy];}
222 bool isInteger() const { return flags[IsInteger]; }
223 bool isFloating() const { return flags[IsFloating]; }
225 bool isControl() const { return flags[IsControl]; }
226 bool isCall() const { return flags[IsCall]; }
227 bool isReturn() const { return flags[IsReturn]; }
228 bool isDirectCtrl() const { return flags[IsDirectControl]; }
229 bool isIndirectCtrl() const { return flags[IsIndirectControl]; }
230 bool isCondCtrl() const { return flags[IsCondControl]; }
231 bool isUncondCtrl() const { return flags[IsUncondControl]; }
232 bool isCondDelaySlot() const { return flags[IsCondDelaySlot]; }
234 bool isThreadSync() const { return flags[IsThreadSync]; }
235 bool isSerializing() const { return flags[IsSerializing] ||
236 flags[IsSerializeBefore] ||
237 flags[IsSerializeAfter]; }
238 bool isSerializeBefore() const { return flags[IsSerializeBefore]; }
239 bool isSerializeAfter() const { return flags[IsSerializeAfter]; }
240 bool isMemBarrier() const { return flags[IsMemBarrier]; }
241 bool isWriteBarrier() const { return flags[IsWriteBarrier]; }
242 bool isNonSpeculative() const { return flags[IsNonSpeculative]; }
243 bool isQuiesce() const { return flags[IsQuiesce]; }
244 bool isIprAccess() const { return flags[IsIprAccess]; }
245 bool isUnverifiable() const { return flags[IsUnverifiable]; }
246 bool isMacroop() const { return flags[IsMacroop]; }
247 bool isMicroop() const { return flags[IsMicroop]; }
248 bool isDelayedCommit() const { return flags[IsDelayedCommit]; }
249 bool isLastMicroop() const { return flags[IsLastMicroop]; }
250 bool isFirstMicroop() const { return flags[IsFirstMicroop]; }
251 //This flag doesn't do anything yet
252 bool isMicroBranch() const { return flags[IsMicroBranch]; }
255 /// Operation class. Used to select appropriate function unit in issue.
256 OpClass opClass() const { return _opClass; }
260 // forward declaration
264 * Generic yet ISA-dependent static instruction class.
266 * This class builds on StaticInstBase, defining fields and interfaces
267 * that are generic across all ISAs but that differ in details
268 * according to the specific ISA being used.
270 class StaticInst : public StaticInstBase
274 /// Binary machine instruction type.
275 typedef TheISA::MachInst MachInst;
276 /// Binary extended machine instruction type.
277 typedef TheISA::ExtMachInst ExtMachInst;
278 /// Logical register index type.
279 typedef TheISA::RegIndex RegIndex;
282 MaxInstSrcRegs = TheISA::MaxInstSrcRegs, //< Max source regs
283 MaxInstDestRegs = TheISA::MaxInstDestRegs, //< Max dest regs
287 /// Return logical index (architectural reg num) of i'th destination reg.
288 /// Only the entries from 0 through numDestRegs()-1 are valid.
289 RegIndex destRegIdx(int i) const { return _destRegIdx[i]; }
291 /// Return logical index (architectural reg num) of i'th source reg.
292 /// Only the entries from 0 through numSrcRegs()-1 are valid.
293 RegIndex srcRegIdx(int i) const { return _srcRegIdx[i]; }
295 /// Pointer to a statically allocated "null" instruction object.
296 /// Used to give eaCompInst() and memAccInst() something to return
297 /// when called on non-memory instructions.
298 static StaticInstPtr nullStaticInstPtr;
301 * Memory references only: returns "fake" instruction representing
302 * the effective address part of the memory operation. Used to
303 * obtain the dependence info (numSrcRegs and srcRegIdx[]) for
304 * just the EA computation.
307 StaticInstPtr &eaCompInst() const { return nullStaticInstPtr; }
310 * Memory references only: returns "fake" instruction representing
311 * the memory access part of the memory operation. Used to
312 * obtain the dependence info (numSrcRegs and srcRegIdx[]) for
313 * just the memory access (not the EA computation).
316 StaticInstPtr &memAccInst() const { return nullStaticInstPtr; }
318 /// The binary machine instruction.
319 const ExtMachInst machInst;
323 /// See destRegIdx().
324 RegIndex _destRegIdx[MaxInstDestRegs];
326 RegIndex _srcRegIdx[MaxInstSrcRegs];
329 * Base mnemonic (e.g., "add"). Used by generateDisassembly()
330 * methods. Also useful to readily identify instructions from
331 * within the debugger when #cachedDisassembly has not been
334 const char *mnemonic;
337 * String representation of disassembly (lazily evaluated via
340 mutable std::string *cachedDisassembly;
343 * Internal function to generate disassembly string.
346 generateDisassembly(Addr pc, const SymbolTable *symtab) const = 0;
349 StaticInst(const char *_mnemonic, ExtMachInst _machInst, OpClass __opClass)
350 : StaticInstBase(__opClass),
351 machInst(_machInst), mnemonic(_mnemonic), cachedDisassembly(0)
353 memset(&recentDecodes, 0, 2 * sizeof(cacheElement));
358 virtual ~StaticInst()
360 if (cachedDisassembly)
361 delete cachedDisassembly;
365 * The execute() signatures are auto-generated by scons based on the
366 * set of CPU models we are compiling in today.
368 #include "cpu/static_inst_exec_sigs.hh"
371 * Return the microop that goes with a particular micropc. This should
372 * only be defined/used in macroops which will contain microops
374 virtual StaticInstPtr fetchMicroop(MicroPC micropc);
377 * Return the target address for a PC-relative branch.
378 * Invalid if not a PC-relative branch (i.e. isDirectCtrl()
381 virtual Addr branchTarget(Addr branchPC) const
383 panic("StaticInst::branchTarget() called on instruction "
384 "that is not a PC-relative branch.");
389 * Return the target address for an indirect branch (jump). The
390 * register value is read from the supplied thread context, so
391 * the result is valid only if the thread context is about to
392 * execute the branch in question. Invalid if not an indirect
393 * branch (i.e. isIndirectCtrl() should be true).
395 virtual Addr branchTarget(ThreadContext *tc) const
397 panic("StaticInst::branchTarget() called on instruction "
398 "that is not an indirect branch.");
403 * Return true if the instruction is a control transfer, and if so,
404 * return the target address as well.
406 bool hasBranchTarget(Addr pc, ThreadContext *tc, Addr &tgt) const;
409 * Return string representation of disassembled instruction.
410 * The default version of this function will call the internal
411 * virtual generateDisassembly() function to get the string,
412 * then cache it in #cachedDisassembly. If the disassembly
413 * should not be cached, this function should be overridden directly.
415 virtual const std::string &disassemble(Addr pc,
416 const SymbolTable *symtab = 0) const
418 if (!cachedDisassembly)
420 new std::string(generateDisassembly(pc, symtab));
422 return *cachedDisassembly;
425 /// Decoded instruction cache type.
426 /// For now we're using a generic hash_map; this seems to work
428 typedef m5::hash_map<ExtMachInst, StaticInstPtr> DecodeCache;
430 /// A cache of decoded instruction objects.
431 static DecodeCache decodeCache;
434 * Dump some basic stats on the decode cache hash map.
435 * Only gets called if DECODE_CACHE_HASH_STATS is defined.
437 static void dumpDecodeCacheStats();
439 /// Decode a machine instruction.
440 /// @param mach_inst The binary instruction to decode.
441 /// @retval A pointer to the corresponding StaticInst object.
442 //This is defined as inlined below.
443 static StaticInstPtr decode(ExtMachInst mach_inst, Addr addr);
445 /// Return name of machine instruction
446 std::string getName() { return mnemonic; }
448 /// Decoded instruction cache type, for address decoding.
449 /// A generic hash_map is used.
450 typedef m5::hash_map<Addr, AddrDecodePage *> AddrDecodeCache;
452 /// A cache of decoded instruction objects from addresses.
453 static AddrDecodeCache addrDecodeCache;
455 struct cacheElement {
457 AddrDecodePage *decodePage;
460 /// An array of recently decoded instructions.
461 // might not use an array if there is only two elements
462 static struct cacheElement recentDecodes[2];
464 /// Updates the recently decoded instructions entries
465 /// @param page_addr The page address recently used.
466 /// @param decodePage Pointer to decoding page containing the decoded
469 updateCache(Addr page_addr, AddrDecodePage *decodePage)
471 recentDecodes[1].page_addr = recentDecodes[0].page_addr;
472 recentDecodes[1].decodePage = recentDecodes[0].decodePage;
473 recentDecodes[0].page_addr = page_addr;
474 recentDecodes[0].decodePage = decodePage;
477 /// Searches the decoded instruction cache for instruction decoding.
478 /// If it is not found, then we decode the instruction.
479 /// Otherwise, we get the instruction from the cache and move it into
480 /// the address-to-instruction decoding page.
481 /// @param mach_inst The binary instruction to decode.
482 /// @param addr The address that contained the binary instruction.
483 /// @param decodePage Pointer to decoding page containing the instruction.
484 /// @retval A pointer to the corresponding StaticInst object.
485 //This is defined as inlined below.
486 static StaticInstPtr searchCache(ExtMachInst mach_inst, Addr addr,
487 AddrDecodePage * decodePage);
490 typedef RefCountingPtr<StaticInstBase> StaticInstBasePtr;
492 /// Reference-counted pointer to a StaticInst object.
493 /// This type should be used instead of "StaticInst *" so that
494 /// StaticInst objects can be properly reference-counted.
495 class StaticInstPtr : public RefCountingPtr<StaticInst>
500 : RefCountingPtr<StaticInst>()
504 /// Conversion from "StaticInst *".
505 StaticInstPtr(StaticInst *p)
506 : RefCountingPtr<StaticInst>(p)
510 /// Copy constructor.
511 StaticInstPtr(const StaticInstPtr &r)
512 : RefCountingPtr<StaticInst>(r)
516 /// Construct directly from machine instruction.
517 /// Calls StaticInst::decode().
518 explicit StaticInstPtr(TheISA::ExtMachInst mach_inst, Addr addr)
519 : RefCountingPtr<StaticInst>(StaticInst::decode(mach_inst, addr))
523 /// Convert to pointer to StaticInstBase class.
524 operator const StaticInstBasePtr()
530 /// A page of a list of decoded instructions from an address.
533 typedef TheISA::ExtMachInst ExtMachInst;
535 StaticInstPtr instructions[TheISA::PageBytes];
536 bool valid[TheISA::PageBytes];
542 lowerMask = TheISA::PageBytes - 1;
543 memset(valid, 0, TheISA::PageBytes);
546 /// Checks if the instruction is already decoded and the machine
547 /// instruction in the cache matches the current machine instruction
548 /// related to the address
549 /// @param mach_inst The binary instruction to check
550 /// @param addr The address containing the instruction
551 inline bool decoded(ExtMachInst mach_inst, Addr addr)
553 return (valid[addr & lowerMask] &&
554 (instructions[addr & lowerMask]->machInst == mach_inst));
557 /// Returns the instruction object. decoded should be called first
558 /// to check if the instruction is valid.
559 /// @param addr The address of the instruction.
560 /// @retval A pointer to the corresponding StaticInst object.
561 inline StaticInstPtr getInst(Addr addr)
562 { return instructions[addr & lowerMask]; }
564 /// Inserts a pointer to a StaticInst object into the list of decoded
565 /// instructions on the page.
566 /// @param addr The address of the instruction.
567 /// @param si A pointer to the corresponding StaticInst object.
568 inline void insert(Addr addr, StaticInstPtr &si)
570 instructions[addr & lowerMask] = si;
571 valid[addr & lowerMask] = true;
578 StaticInst::decode(StaticInst::ExtMachInst mach_inst, Addr addr)
580 #ifdef DECODE_CACHE_HASH_STATS
581 // Simple stats on decode hash_map. Turns out the default
582 // hash function is as good as anything I could come up with.
583 const int dump_every_n = 10000000;
584 static int decodes_til_dump = dump_every_n;
586 if (--decodes_til_dump == 0) {
587 dumpDecodeCacheStats();
588 decodes_til_dump = dump_every_n;
592 Addr page_addr = addr & ~(TheISA::PageBytes - 1);
594 // checks recently decoded addresses
595 if (recentDecodes[0].decodePage &&
596 page_addr == recentDecodes[0].page_addr) {
597 if (recentDecodes[0].decodePage->decoded(mach_inst, addr))
598 return recentDecodes[0].decodePage->getInst(addr);
600 return searchCache(mach_inst, addr, recentDecodes[0].decodePage);
603 if (recentDecodes[1].decodePage &&
604 page_addr == recentDecodes[1].page_addr) {
605 if (recentDecodes[1].decodePage->decoded(mach_inst, addr))
606 return recentDecodes[1].decodePage->getInst(addr);
608 return searchCache(mach_inst, addr, recentDecodes[1].decodePage);
611 // searches the page containing the address to decode
612 AddrDecodeCache::iterator iter = addrDecodeCache.find(page_addr);
613 if (iter != addrDecodeCache.end()) {
614 updateCache(page_addr, iter->second);
615 if (iter->second->decoded(mach_inst, addr))
616 return iter->second->getInst(addr);
618 return searchCache(mach_inst, addr, iter->second);
621 // creates a new object for a page of decoded instructions
622 AddrDecodePage * decodePage = new AddrDecodePage;
623 addrDecodeCache[page_addr] = decodePage;
624 updateCache(page_addr, decodePage);
625 return searchCache(mach_inst, addr, decodePage);
629 StaticInst::searchCache(ExtMachInst mach_inst, Addr addr,
630 AddrDecodePage * decodePage)
632 DecodeCache::iterator iter = decodeCache.find(mach_inst);
633 if (iter != decodeCache.end()) {
634 decodePage->insert(addr, iter->second);
638 StaticInstPtr si = TheISA::decodeInst(mach_inst);
639 decodePage->insert(addr, si);
640 decodeCache[mach_inst] = si;
644 #endif // __CPU_STATIC_INST_HH__