icount.H

/*
 * Copyright 2002-2019 Intel Corporation.
 * 
 * This software is provided to you as Sample Source Code as defined in the accompanying
 * End User License Agreement for the Intel(R) Software Development Products ("Agreement")
 * section 1.L.
 * 
 * This software and the related documents are provided as is, with no express or implied
 * warranties, other than those that are expressly stated in the License.
 */

#ifndef ICOUNT_H
#define ICOUNT_H

namespace INSTLIB 
{

/*! @defgroup ICOUNT
  Instrumentation for counting instruction execution
*/

/*! @ingroup ICOUNT
  The example below can be found in InstLibExamples/icount.cpp

  \include icount.cpp
*/
class ICOUNT 
{
  public:
    ICOUNT()
    {
        _mode = ModeInactive;
        
        /* Allocate 64 byte aligned data for the statistics. */
        _space = new char [(ISIMPOINT_MAX_THREADS+1)*sizeof(threadStats) -1];
        
        ADDRINT space = VoidStar2Addrint(_space);
        ADDRINT align_1 = static_cast <ADDRINT>(cacheLineSize-1);
        _stats = reinterpret_cast<threadStats *>((space+align_1) & ~align_1);
        memset (_stats, 0, ISIMPOINT_MAX_THREADS*sizeof(threadStats));
    };
    
    ~ICOUNT()
    {
        delete [] _space;
    }
    /*! @ingroup ICOUNT
      @return Total number of instructions executed. (But see @ref mode for what this means).
    */

    UINT32 MultiThreadCount() const
    {
      UINT64 multithreadCount=0;
      ASSERTX(Mode() == ModeBoth);
      for(UINT64 i=0; i<ISIMPOINT_MAX_THREADS; i++)
      {
        multithreadCount = multithreadCount + _stats[i].count;
      }
      return multithreadCount;
    }

    VOID SetMultithreadCount(UINT64 count)
    {
      ASSERTX(_mode != ModeInactive);
      for(UINT64 i=0; i<ISIMPOINT_MAX_THREADS; i++)
      {
        _stats[i].count=count;
        _stats[i].repDuplicateCount=0;
      }
    }


    UINT64 Count(THREADID tid = 0) const
    {
        ASSERTX(tid < ISIMPOINT_MAX_THREADS);
        return _stats[tid].count;
    }

    UINT64 CountWithoutRep(THREADID tid = 0) const
    {
        ASSERTX(tid < ISIMPOINT_MAX_THREADS);
        ASSERTX(Mode() == ModeBoth);
        threadStats * s = &_stats[tid];

        return s->count - s->repDuplicateCount;
    }

    /*! @ingroup ICOUNT
      Set the current count
    */
    VOID SetCount(UINT64 count, THREADID tid = 0)
    {
        ASSERTX(_mode != ModeInactive);
        ASSERTX(tid < ISIMPOINT_MAX_THREADS);
        _stats[tid].count = count;
        _stats[tid].repDuplicateCount = 0;
    }

    /*! @ingroup ICOUNT
     * The mode used for counting REP prefixed instructions.
     */
    enum mode {
        ModeInactive = -1,
        ModeNormal = 0,                         /**< Count all instructions, each REP "iteration" adds 1 */
        ModeBoth                               /**< Provide both the normal count and a count in which REP prefixed
                                                   instructions are only counted once. */
    };

    /*! @ingroup ICOUNT
     * @return the mode of the ICOUNT object.
     */
    mode Mode() const 
    {
        return _mode;
    }

    /*! @ingroup ICOUNT
      Activate the counter, must be called before PIN_StartProgram.
      @param [in] mode Determine the way in which REP prefixed operations are counted. By default (ICOUNT::ModeNormal),
                       REP prefixed instructions are counted as if REP is an implicit loop. By passing 
                       ICOUNT::ModeRepsCountedOnlyOnce you can have the counter treat each REP as only one dynamic instruction.
    */
    VOID Activate(mode m = ModeNormal)
    {
        ASSERTX(_mode == ModeInactive);
        _mode   = m;
        TRACE_AddInstrumentFunction(Trace, this);
    }

  private:
    enum {
        cacheLineSize = 64
    };

    static VOID Trace(TRACE trace, VOID * icount)
    {
#if (defined(TARGET_IA32) || defined(TARGET_IA32E))
        ICOUNT const * ic = reinterpret_cast<ICOUNT const *>(icount);
        mode m = ic->Mode();
#endif
        for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
        {
            BBL_InsertCall(bbl, IPOINT_ANYWHERE,
                           AFUNPTR(Advance),
                           IARG_FAST_ANALYSIS_CALL,
                           IARG_ADDRINT, icount, 
                           IARG_ADDRINT, ADDRINT(BBL_NumIns(bbl)), 
                           IARG_THREAD_ID, 
                           IARG_END);

            // REP prefixed instructions are an IA-32 and Intel(R) 64 feature
#if (defined(TARGET_IA32) || defined(TARGET_IA32E))
            if (m == ModeBoth)
            { // Check whether there are any REP prefixed instructions in the BBL
              // and, if so, subtract out their execution unless it is the first
              // iteration.
                for (INS ins = BBL_InsHead(bbl);
                     INS_Valid(ins);
                     ins = INS_Next(ins))
                {
                    if (INS_HasRealRep(ins))
                    {
                        INS_InsertCall(ins, IPOINT_BEFORE, 
                                       AFUNPTR(CountDuplicates),
                                       IARG_FAST_ANALYSIS_CALL,
                                       IARG_ADDRINT, icount, 
                                       IARG_FIRST_REP_ITERATION,
                                       IARG_THREAD_ID, 
                                       IARG_END);
                                       
                    }
                }
            }
#endif
        }
    }

    static VOID PIN_FAST_ANALYSIS_CALL Advance(ICOUNT * ic, ADDRINT c, THREADID tid)
    {
        // ASSERTX(tid < ISIMPOINT_MAX_THREADS);
        ic->_stats[tid].count += c;
    }

    // Accumulate the count of REP prefixed executions which aren't the first iteration. 
    //
    // We are assuming that this will be inlined, and is small, so there is no point
    // in guarding it with an InsertIf call testing IARG_FIRST_REP_ITERATION.
    static VOID PIN_FAST_ANALYSIS_CALL CountDuplicates(ICOUNT * ic, BOOL first, THREADID tid)
    {
        // ASSERTX(tid < ISIMPOINT_MAX_THREADS);
        ic->_stats[tid].repDuplicateCount += !first;
    }

    struct threadStats {
        UINT64 count;
        UINT64 repDuplicateCount;                          /* Number of REP iterations after the first */
        char padding [cacheLineSize - 2*sizeof(UINT64)];   /* Expand so we can cache align this.
                                                            * We want to avoid false sharing of the stats between threads.
                                                            */
    };

    threadStats * _stats;
    char * _space;
    mode   _mode;
};
}
#endif