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/types.hh"
39 #include "arch/registers.hh"
40 #include "config/the_isa.hh"
41 #include "base/hashmap.hh"
42 #include "base/misc.hh"
43 #include "base/refcnt.hh"
44 #include "base/types.hh"
45 #include "cpu/op_class.hh"
46 #include "sim/fault.hh"
48 // forward declarations
49 struct AlphaSimpleImpl;
57 template <class Impl> class BaseO3DynInst;
58 typedef BaseO3DynInst<O3CPUImpl> O3DynInst;
59 template <class Impl> class OzoneDynInst;
64 class AtomicSimpleCPU;
65 class TimingSimpleCPU;
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 IsIndexed, ///< Accesses memory with an indexed address computation
118 IsInstPrefetch, ///< Instruction-cache prefetch.
119 IsDataPrefetch, ///< Data-cache prefetch.
120 IsCopy, ///< Fast Cache block copy
122 IsControl, ///< Control transfer instruction.
123 IsDirectControl, ///< PC relative control transfer.
124 IsIndirectControl, ///< Register indirect control transfer.
125 IsCondControl, ///< Conditional control transfer.
126 IsUncondControl, ///< Unconditional control transfer.
127 IsCall, ///< Subroutine call.
128 IsReturn, ///< Subroutine return.
130 IsCondDelaySlot,///< Conditional Delay-Slot Instruction
132 IsThreadSync, ///< Thread synchronization operation.
134 IsSerializing, ///< Serializes pipeline: won't execute until all
135 /// older instructions have committed.
138 IsMemBarrier, ///< Is a memory barrier
139 IsWriteBarrier, ///< Is a write barrier
140 IsReadBarrier, ///< Is a read barrier
141 IsERET, /// <- Causes the IFU to stall (MIPS ISA)
143 IsNonSpeculative, ///< Should not be executed speculatively
144 IsQuiesce, ///< Is a quiesce instruction
146 IsIprAccess, ///< Accesses IPRs
147 IsUnverifiable, ///< Can't be verified by a checker
149 IsSyscall, ///< Causes a system call to be emulated in syscall
152 //Flags for microcode
153 IsMacroop, ///< Is a macroop containing microops
154 IsMicroop, ///< Is a microop
155 IsDelayedCommit, ///< This microop doesn't commit right away
156 IsLastMicroop, ///< This microop ends a microop sequence
157 IsFirstMicroop, ///< This microop begins a microop sequence
158 //This flag doesn't do anything yet
159 IsMicroBranch, ///< This microop branches within the microcode for a macroop
167 /// Flag values for this instruction.
168 std::bitset<NumFlags> flags;
173 /// See numSrcRegs().
176 /// See numDestRegs().
179 /// The following are used to track physical register usage
180 /// for machines with separate int & FP reg files.
182 int8_t _numFPDestRegs;
183 int8_t _numIntDestRegs;
187 /// It's important to initialize everything here to a sane
188 /// default, since the decoder generally only overrides
189 /// the fields that are meaningful for the particular
191 StaticInstBase(OpClass __opClass)
192 : _opClass(__opClass), _numSrcRegs(0), _numDestRegs(0),
193 _numFPDestRegs(0), _numIntDestRegs(0)
199 /// @name Register information.
200 /// The sum of numFPDestRegs() and numIntDestRegs() equals
201 /// numDestRegs(). The former two functions are used to track
202 /// physical register usage for machines with separate int & FP
205 /// Number of source registers.
206 int8_t numSrcRegs() const { return _numSrcRegs; }
207 /// Number of destination registers.
208 int8_t numDestRegs() const { return _numDestRegs; }
209 /// Number of floating-point destination regs.
210 int8_t numFPDestRegs() const { return _numFPDestRegs; }
211 /// Number of integer destination regs.
212 int8_t numIntDestRegs() const { return _numIntDestRegs; }
215 /// @name Flag accessors.
216 /// These functions are used to access the values of the various
217 /// instruction property flags. See StaticInstBase::Flags for descriptions
218 /// of the individual flags.
221 bool isNop() const { return flags[IsNop]; }
223 bool isMemRef() const { return flags[IsMemRef]; }
224 bool isLoad() const { return flags[IsLoad]; }
225 bool isStore() const { return flags[IsStore]; }
226 bool isStoreConditional() const { return flags[IsStoreConditional]; }
227 bool isInstPrefetch() const { return flags[IsInstPrefetch]; }
228 bool isDataPrefetch() const { return flags[IsDataPrefetch]; }
229 bool isCopy() const { return flags[IsCopy];}
231 bool isInteger() const { return flags[IsInteger]; }
232 bool isFloating() const { return flags[IsFloating]; }
234 bool isControl() const { return flags[IsControl]; }
235 bool isCall() const { return flags[IsCall]; }
236 bool isReturn() const { return flags[IsReturn]; }
237 bool isDirectCtrl() const { return flags[IsDirectControl]; }
238 bool isIndirectCtrl() const { return flags[IsIndirectControl]; }
239 bool isCondCtrl() const { return flags[IsCondControl]; }
240 bool isUncondCtrl() const { return flags[IsUncondControl]; }
241 bool isCondDelaySlot() const { return flags[IsCondDelaySlot]; }
243 bool isThreadSync() const { return flags[IsThreadSync]; }
244 bool isSerializing() const { return flags[IsSerializing] ||
245 flags[IsSerializeBefore] ||
246 flags[IsSerializeAfter]; }
247 bool isSerializeBefore() const { return flags[IsSerializeBefore]; }
248 bool isSerializeAfter() const { return flags[IsSerializeAfter]; }
249 bool isMemBarrier() const { return flags[IsMemBarrier]; }
250 bool isWriteBarrier() const { return flags[IsWriteBarrier]; }
251 bool isNonSpeculative() const { return flags[IsNonSpeculative]; }
252 bool isQuiesce() const { return flags[IsQuiesce]; }
253 bool isIprAccess() const { return flags[IsIprAccess]; }
254 bool isUnverifiable() const { return flags[IsUnverifiable]; }
255 bool isSyscall() const { return flags[IsSyscall]; }
256 bool isMacroop() const { return flags[IsMacroop]; }
257 bool isMicroop() const { return flags[IsMicroop]; }
258 bool isDelayedCommit() const { return flags[IsDelayedCommit]; }
259 bool isLastMicroop() const { return flags[IsLastMicroop]; }
260 bool isFirstMicroop() const { return flags[IsFirstMicroop]; }
261 //This flag doesn't do anything yet
262 bool isMicroBranch() const { return flags[IsMicroBranch]; }
265 void setLastMicroop() { flags[IsLastMicroop] = true; }
266 void setDelayedCommit() { flags[IsDelayedCommit] = true; }
268 /// Operation class. Used to select appropriate function unit in issue.
269 OpClass opClass() const { return _opClass; }
273 // forward declaration
277 * Generic yet ISA-dependent static instruction class.
279 * This class builds on StaticInstBase, defining fields and interfaces
280 * that are generic across all ISAs but that differ in details
281 * according to the specific ISA being used.
283 class StaticInst : public StaticInstBase
287 /// Binary machine instruction type.
288 typedef TheISA::MachInst MachInst;
289 /// Binary extended machine instruction type.
290 typedef TheISA::ExtMachInst ExtMachInst;
291 /// Logical register index type.
292 typedef TheISA::RegIndex RegIndex;
295 MaxInstSrcRegs = TheISA::MaxInstSrcRegs, //< Max source regs
296 MaxInstDestRegs = TheISA::MaxInstDestRegs, //< Max dest regs
300 /// Return logical index (architectural reg num) of i'th destination reg.
301 /// Only the entries from 0 through numDestRegs()-1 are valid.
302 RegIndex destRegIdx(int i) const { return _destRegIdx[i]; }
304 /// Return logical index (architectural reg num) of i'th source reg.
305 /// Only the entries from 0 through numSrcRegs()-1 are valid.
306 RegIndex srcRegIdx(int i) const { return _srcRegIdx[i]; }
308 /// Pointer to a statically allocated "null" instruction object.
309 /// Used to give eaCompInst() and memAccInst() something to return
310 /// when called on non-memory instructions.
311 static StaticInstPtr nullStaticInstPtr;
314 * Memory references only: returns "fake" instruction representing
315 * the effective address part of the memory operation. Used to
316 * obtain the dependence info (numSrcRegs and srcRegIdx[]) for
317 * just the EA computation.
320 StaticInstPtr &eaCompInst() const { return nullStaticInstPtr; }
323 * Memory references only: returns "fake" instruction representing
324 * the memory access part of the memory operation. Used to
325 * obtain the dependence info (numSrcRegs and srcRegIdx[]) for
326 * just the memory access (not the EA computation).
329 StaticInstPtr &memAccInst() const { return nullStaticInstPtr; }
331 /// The binary machine instruction.
332 const ExtMachInst machInst;
336 /// See destRegIdx().
337 RegIndex _destRegIdx[MaxInstDestRegs];
339 RegIndex _srcRegIdx[MaxInstSrcRegs];
342 * Base mnemonic (e.g., "add"). Used by generateDisassembly()
343 * methods. Also useful to readily identify instructions from
344 * within the debugger when #cachedDisassembly has not been
347 const char *mnemonic;
350 * String representation of disassembly (lazily evaluated via
353 mutable std::string *cachedDisassembly;
356 * Internal function to generate disassembly string.
359 generateDisassembly(Addr pc, const SymbolTable *symtab) const = 0;
362 StaticInst(const char *_mnemonic, ExtMachInst _machInst, OpClass __opClass)
363 : StaticInstBase(__opClass),
364 machInst(_machInst), mnemonic(_mnemonic), cachedDisassembly(0)
368 virtual ~StaticInst();
371 * The execute() signatures are auto-generated by scons based on the
372 * set of CPU models we are compiling in today.
374 #include "cpu/static_inst_exec_sigs.hh"
376 virtual void advancePC(TheISA::PCState &pcState) const = 0;
379 * Return the microop that goes with a particular micropc. This should
380 * only be defined/used in macroops which will contain microops
382 virtual StaticInstPtr fetchMicroop(MicroPC upc) const;
385 * Return the target address for a PC-relative branch.
386 * Invalid if not a PC-relative branch (i.e. isDirectCtrl()
389 virtual TheISA::PCState branchTarget(const TheISA::PCState &pc) const;
392 * Return the target address for an indirect branch (jump). The
393 * register value is read from the supplied thread context, so
394 * the result is valid only if the thread context is about to
395 * execute the branch in question. Invalid if not an indirect
396 * branch (i.e. isIndirectCtrl() should be true).
398 virtual TheISA::PCState branchTarget(ThreadContext *tc) const;
401 * Return true if the instruction is a control transfer, and if so,
402 * return the target address as well.
404 bool hasBranchTarget(const TheISA::PCState &pc, ThreadContext *tc,
405 TheISA::PCState &tgt) const;
408 * Return string representation of disassembled instruction.
409 * The default version of this function will call the internal
410 * virtual generateDisassembly() function to get the string,
411 * then cache it in #cachedDisassembly. If the disassembly
412 * should not be cached, this function should be overridden directly.
414 virtual const std::string &disassemble(Addr pc,
415 const SymbolTable *symtab = 0) const;
417 /// Decoded instruction cache type.
418 /// For now we're using a generic hash_map; this seems to work
420 typedef m5::hash_map<ExtMachInst, StaticInstPtr> DecodeCache;
422 /// A cache of decoded instruction objects.
423 static DecodeCache decodeCache;
426 * Dump some basic stats on the decode cache hash map.
427 * Only gets called if DECODE_CACHE_HASH_STATS is defined.
429 static void dumpDecodeCacheStats();
431 /// Decode a machine instruction.
432 /// @param mach_inst The binary instruction to decode.
433 /// @retval A pointer to the corresponding StaticInst object.
434 //This is defined as inlined below.
435 static StaticInstPtr decode(ExtMachInst mach_inst, Addr addr);
437 /// Return name of machine instruction
438 std::string getName() { return mnemonic; }
440 /// Decoded instruction cache type, for address decoding.
441 /// A generic hash_map is used.
442 typedef m5::hash_map<Addr, AddrDecodePage *> AddrDecodeCache;
444 /// A cache of decoded instruction objects from addresses.
445 static AddrDecodeCache addrDecodeCache;
450 AddrDecodePage *decodePage;
452 cacheElement() : decodePage(NULL) { }
455 /// An array of recently decoded instructions.
456 // might not use an array if there is only two elements
457 static struct cacheElement recentDecodes[2];
459 /// Updates the recently decoded instructions entries
460 /// @param page_addr The page address recently used.
461 /// @param decodePage Pointer to decoding page containing the decoded
464 updateCache(Addr page_addr, AddrDecodePage *decodePage)
466 recentDecodes[1].page_addr = recentDecodes[0].page_addr;
467 recentDecodes[1].decodePage = recentDecodes[0].decodePage;
468 recentDecodes[0].page_addr = page_addr;
469 recentDecodes[0].decodePage = decodePage;
472 /// Searches the decoded instruction cache for instruction decoding.
473 /// If it is not found, then we decode the instruction.
474 /// Otherwise, we get the instruction from the cache and move it into
475 /// the address-to-instruction decoding page.
476 /// @param mach_inst The binary instruction to decode.
477 /// @param addr The address that contained the binary instruction.
478 /// @param decodePage Pointer to decoding page containing the instruction.
479 /// @retval A pointer to the corresponding StaticInst object.
480 //This is defined as inlined below.
481 static StaticInstPtr searchCache(ExtMachInst mach_inst, Addr addr,
482 AddrDecodePage *decodePage);
485 typedef RefCountingPtr<StaticInstBase> StaticInstBasePtr;
487 /// Reference-counted pointer to a StaticInst object.
488 /// This type should be used instead of "StaticInst *" so that
489 /// StaticInst objects can be properly reference-counted.
490 class StaticInstPtr : public RefCountingPtr<StaticInst>
495 : RefCountingPtr<StaticInst>()
499 /// Conversion from "StaticInst *".
500 StaticInstPtr(StaticInst *p)
501 : RefCountingPtr<StaticInst>(p)
505 /// Copy constructor.
506 StaticInstPtr(const StaticInstPtr &r)
507 : RefCountingPtr<StaticInst>(r)
511 /// Construct directly from machine instruction.
512 /// Calls StaticInst::decode().
513 explicit StaticInstPtr(TheISA::ExtMachInst mach_inst, Addr addr)
514 : RefCountingPtr<StaticInst>(StaticInst::decode(mach_inst, addr))
518 /// Convert to pointer to StaticInstBase class.
519 operator const StaticInstBasePtr()
525 /// A page of a list of decoded instructions from an address.
528 typedef TheISA::ExtMachInst ExtMachInst;
530 StaticInstPtr instructions[TheISA::PageBytes];
531 bool valid[TheISA::PageBytes];
538 lowerMask = TheISA::PageBytes - 1;
539 memset(valid, 0, TheISA::PageBytes);
542 /// Checks if the instruction is already decoded and the machine
543 /// instruction in the cache matches the current machine instruction
544 /// related to the address
545 /// @param mach_inst The binary instruction to check
546 /// @param addr The address containing the instruction
548 decoded(ExtMachInst mach_inst, Addr addr)
550 return (valid[addr & lowerMask] &&
551 (instructions[addr & lowerMask]->machInst == mach_inst));
554 /// Returns the instruction object. decoded should be called first
555 /// to check if the instruction is valid.
556 /// @param addr The address of the instruction.
557 /// @retval A pointer to the corresponding StaticInst object.
561 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.
569 insert(Addr addr, StaticInstPtr &si)
571 instructions[addr & lowerMask] = si;
572 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__