VOID Trace (TRACE trace, VOID *v)
{
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        for (INS ins = BBL_InsHead(bbl); INS_Valid(ins); ins = INS_Next(ins))
        {
            xed_iclass_enum_t iclass1 = static_cast<xed_iclass_enum_t>(INS_Opcode(ins));
            if (iclass1 == XED_ICLASS_FLD1 && INS_Valid(INS_Next(ins)))
            {
                xed_iclass_enum_t iclass2 = static_cast<xed_iclass_enum_t>(INS_Opcode(INS_Next(ins)));
                if (iclass2 == XED_ICLASS_FLD1 && INS_Valid(INS_Next(INS_Next(ins))))
                {
                    xed_iclass_enum_t iclass3 = static_cast<xed_iclass_enum_t>(INS_Opcode(INS_Next(INS_Next(ins))));
                    if (iclass3 == XED_ICLASS_FLD1)
                    {
                        printf ("tool: found fld1 sequence at %p\n", (void *)INS_Address(INS_Next(INS_Next(ins))));
                        fflush (stdout);
                        // Insert an analysis call that will cause the xmm scratch registers to be spilled
                        INS_InsertCall(INS_Next(INS_Next(ins)), IPOINT_AFTER, (AFUNPTR)SetXmmScratchesFun, IARG_END);
                        return;
                    }
                }
            }
        }
    }
}

VOID Trace (TRACE trace, VOID *v)
{
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        for (INS ins = BBL_InsHead(bbl); INS_Valid(ins); ins = INS_Next(ins))
        {
            xed_iclass_enum_t iclass1 = static_cast<xed_iclass_enum_t>(INS_Opcode(ins));
            if (iclass1 == XED_ICLASS_FLD1 && INS_Valid(INS_Next(ins)))
            {
                xed_iclass_enum_t iclass2 = static_cast<xed_iclass_enum_t>(INS_Opcode(INS_Next(ins)));
                if (iclass2 == XED_ICLASS_FLD1 && INS_Valid(INS_Next(INS_Next(ins))))
                {
                    xed_iclass_enum_t iclass3 = static_cast<xed_iclass_enum_t>(INS_Opcode(INS_Next(INS_Next(ins))));
                    if (iclass3 == XED_ICLASS_FLD1)
                    {
                        printf ("found fld1 sequence at %lx\n", (unsigned long)INS_Address(INS_Next(INS_Next(ins))));
                        
                        {
                            INS_InsertCall(INS_Next(INS_Next(ins)), IPOINT_AFTER, AFUNPTR(CallToFldzToTop3), IARG_END);
                            printf ("Inserted call1 to FldzToTop3 after instruction at %lx\n",
                            		(unsigned long)INS_Address(INS_Next(INS_Next(ins))));
                            
                        }
                    }
                }
            }
        }
    }
}

VOID Trace(TRACE trace, VOID *v)
{
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        BBL_InsertCall(bbl, IPOINT_ANYWHERE, (AFUNPTR)LogBBL, IARG_FAST_ANALYSIS_CALL, IARG_ADDRINT, BBL_Address(bbl), IARG_END);
    }
}

// Pin calls this function every time a new basic block is encountered
// It inserts a call to docount
VOID Trace(TRACE trace, VOID *v)
{
    // Visit every basic block in the trace
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        for (INS ins = BBL_InsHead(bbl); INS_Valid(ins); ins = INS_Next(ins))
        {
            if (INS_IsMemoryRead(ins))
            {
                INS_InsertIfCall(ins, IPOINT_BEFORE, (AFUNPTR)ReadAlways, IARG_MEMORYREAD_EA, IARG_END);
                INS_InsertThenCall(ins, IPOINT_BEFORE, (AFUNPTR)ReadRare, IARG_MEMORYREAD_EA, IARG_END);
            }
        }
            
        // Always()->Rare() are partially inlined
        BBL_InsertIfCall(bbl,   IPOINT_BEFORE, (AFUNPTR)Always, IARG_END);
        BBL_InsertThenCall(bbl, IPOINT_BEFORE, (AFUNPTR)Rare, IARG_END);

        // Always()->Rare() are partially inlined
        BBL_InsertIfCall(bbl,   IPOINT_BEFORE, (AFUNPTR)AlwaysNoinline, IARG_END);
        BBL_InsertThenCall(bbl, IPOINT_BEFORE, (AFUNPTR)RareNoinline, IARG_END);

        // Noinline() is not inlined
        BBL_InsertCall(bbl, IPOINT_BEFORE, (AFUNPTR)Noinline, IARG_END);
    }
}

// We have to instrument traces in order to instrument each BBL, the API doesn't have a BBL_AddInstrumentFunction
VOID trace_instrumentation(TRACE trace, VOID *v)
{
    // We don't want to instrument the BBL contained in the Windows API
    if(is_address_in_blacklisted_modules(TRACE_Address(trace)))
        return;

    for(BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        // What's going on under the hood
        // LOG("[INSTRU] BBL Address: " + hexstr(BBL_Address(bbl)) + ", " + hexstr(BBL_NumIns(bbl)) + "\n");
        
        // Insert a call to handle_basic_block before every basic block, passing the number of instructions
        BBL_InsertCall(
            bbl,
            IPOINT_ANYWHERE,
            (AFUNPTR)handle_basic_block,
            IARG_FAST_ANALYSIS_CALL, // Use a faster linkage for calls to analysis functions. Add PIN_FAST_ANALYSIS_CALL to the declaration between the return type and the function name. You must also add IARG_FAST_ANALYSIS_CALL to the InsertCall. For example:

            IARG_UINT32,
            BBL_NumIns(bbl),

            IARG_ADDRINT,
            BBL_Address(bbl),

            IARG_END
        );
    }
}

bool TraceManager::IsNormal(RTN myrtn)
{
	BBL my_bbl=RTN_BblTail(myrtn);	
	if(BBL_Valid(my_bbl))
	{
		INS my_ins=BBL_InsTail(my_bbl);
		while(INS_Valid(my_ins))
		{
			if(INS_IsRet(my_ins))
			{
				//cerr<<"Normal Routine::"<<RTN_Name(myrtn)<<endl;
				return true;
			}
			my_ins=INS_Prev(my_ins);
		}	
		/*if(INS_IsBranch(my_ins)|| INS_IsNop(my_ins))
		{
			cerr<<"!!!Abnormal Routine::"<<RTN_Name(myrtn)<<endl;
			return false;
		}	
		my_bbl=BBL_Prev(my_bbl); */
	}
	//cerr<<"!!!!Abnormal Routine::"<<RTN_Name(myrtn)<<endl;
	return false;
}

// Pin calls this function every time a new instruction is encountered
void Trace(TRACE trace , void *v)
{
	for(BBL bbl= TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl)){
		for( INS ins = BBL_InsHead(bbl); INS_Valid(ins) ; ins =INS_Next(ins)){
			oepf.IsCurrentInOEP(ins);
		}
	}
}

VOID Trace(TRACE trace, VOID *)
{
    if (!EnableInstrumentation)
        return;

    for (BBL bbl = TRACE_BblHead(trace);  BBL_Valid(bbl);  bbl = BBL_Next(bbl))
        BBL_InsertCall(bbl, IPOINT_BEFORE, AFUNPTR(CheckPC), IARG_INST_PTR, IARG_END);
}

static VOID Trace(TRACE trace, VOID *v)
{
    RTN rtn = TRACE_Rtn(trace);
    
    ADDRINT version = TRACE_Version(trace);
    // If we are not in watch_rtn, switch back to base version
    if (!RTN_Valid(rtn) || RTN_Name(rtn) != watch_rtn)
    {
        if (version != VERSION_BASE)
            BBL_SetTargetVersion(TRACE_BblHead(trace), VERSION_BASE);
        return;
    }

    if (TRACE_Address(trace) == RTN_Address(rtn)) {
        INS ins = BBL_InsHead(TRACE_BblHead(trace));
        if (version == VERSION_BASE) 
        {
            // version_reg is used to select the version, use the first
            // argument of watch_rtn to set it
            INS_InsertCall(ins, IPOINT_BEFORE, AFUNPTR(select_version),
                           IARG_FUNCARG_ENTRYPOINT_VALUE, 0,
                           IARG_RETURN_REGS, version_reg,
                           IARG_END);
            // IF we are in the base version, decide if we should go to the
            // other versions
            // Note that the version instrumentation will occur before any
            // following instrumentation done on this ins
            INS_InsertVersionCase(ins, version_reg, 10, VERSION_1, IARG_END);
            INS_InsertVersionCase(ins, version_reg, 20, VERSION_2, IARG_END);
            printf ("Instrumentation at %p\n", reinterpret_cast<void *>(INS_Address(ins)));
        }
    }

    INS ins = BBL_InsHead(TRACE_BblHead(trace));
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl)) 
    {
        // Instrumentation depends on version
        // These instrumentations occur after the preceeding version instrumentation
        // (i.e. the instrumentation inserted by the above INS_InsertVersionCase calls
        switch(version) {
          case VERSION_BASE:
            INS_InsertCall(ins, IPOINT_BEFORE, AFUNPTR(Emit),
                             IARG_PTR, "version base", IARG_END);
            break;
          case VERSION_1:
            INS_InsertCall(ins, IPOINT_BEFORE, AFUNPTR(Emit),
                             IARG_PTR, "version 1", IARG_END);
            break;
          case VERSION_2:
            INS_InsertCall(ins, IPOINT_BEFORE, AFUNPTR(Emit),
                             IARG_PTR, "version 2", IARG_END);
            break;
          default:
            assert(0);
            break;
        }
    }
}

 static void instrument_trace(TRACE trace, VOID *v)
 {
     sse_aligner_t* pthis = static_cast<sse_aligner_t*>(v);
     bool is_read = false;
     for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
         for (INS ins = BBL_InsHead(bbl); INS_Valid(ins); ins = INS_Next(ins))
             if (check_for_sse_memop(ins, is_read, pthis))
                 rewrite_instruction(ins, is_read, pthis);
 }

/*!
 * Insert call to the CountBbl() analysis routine before every basic block 
 * of the trace.
 * This function is called every time a new trace is encountered.
 * @param[in]   trace    trace to be instrumented
 * @param[in]   v        value specified by the tool in the TRACE_AddInstrumentFunction
 *                       function call
 */
VOID Trace(TRACE trace, VOID *v)
{
    // Visit every basic block in the trace
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        // Insert a call to CountBbl() before every basic bloc, passing the number of instructions
        BBL_InsertCall(bbl, IPOINT_BEFORE, (AFUNPTR)CountBbl, IARG_UINT32, BBL_NumIns(bbl), IARG_END);
    }
}

VOID Trace(TRACE trace, VOID *v)
{
    for (BBL bbl = TRACE_BblTail(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl)) {
        if (BBL_HasFallThrough(bbl)) {
            INS ins = BBL_InsTail(bbl);
            TEST(INS_HasFallThrough(ins), "BBL_HasFallThrough or INS_HasFallThrough failed");
        }
    }
}

VOID Trace(TRACE trace, VOID *v)
{
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))

	BBL_InsertCall(bbl, IPOINT_ANYWHERE, AFUNPTR(Count),
		IARG_FAST_ANALYSIS_CALL,
		IARG_THREAD_ID,
		IARG_UINT32, BBL_NumIns(bbl),
		IARG_END);
}

VOID Trace(TRACE trace, VOID *v)
{
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        for (INS ins = BBL_InsHead(bbl); INS_Valid(ins); ins = INS_Next(ins))
        {
            RewriteIns(ins);
        }
    }
}

void Trace(TRACE trace, void *v) 
{
    // for each bbl in the application trace...
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl)) 
    {
        // args:    bbl=bbl, IPOINT_BEFORE=insert before bbl, checkBBL=ptr to function to insert, 
        //          IARG_PTR=function argument type, BBL_Address(bbl)=function argument, IARG_END=end of arguement marker
        BBL_InsertCall(bbl, IPOINT_BEFORE, (AFUNPTR)checkBBL, IARG_PTR, BBL_Address(bbl), IARG_END);
    }
} 

VOID Trace(TRACE trace, VOID *v)
{
    // Instrument only at the head of the trace
    BBL bbl = TRACE_BblHead(trace);
    
    if (BBL_Valid(bbl))
    {
        INS_InsertCall(BBL_InsHead(bbl), IPOINT_BEFORE, (AFUNPTR)docount, IARG_UINT32, 1, IARG_END);
    }

    for (bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        for (INS ins = BBL_InsHead(bbl); INS_Valid(ins); ins = INS_Next(ins))
        {
            ASSERTX(INS_Size(ins) != 0);
        }
    }

}

VOID TraceInserted(TRACE trace, VOID *)
{
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        for (INS ins = BBL_InsHead(bbl); INS_Valid(ins); ins = INS_Next(ins))
        {
            cacheInstructions++;
        }
    }
}

// Pin calls this function every time a new basic block is encountered
// It inserts a call to docount
VOID Trace(TRACE trace, VOID *v)
{
    // Visit every basic block  in the trace
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        // Insert a call to docount for every bbl, passing the number of instructions.
        // IPOINT_ANYWHERE allows Pin to schedule the call anywhere in the bbl to obtain best performance.
        // Use a fast linkage for the call.
        BBL_InsertCall(bbl, IPOINT_ANYWHERE, AFUNPTR(docount), IARG_FAST_ANALYSIS_CALL, IARG_UINT32, BBL_NumIns(bbl), IARG_END);
    }
}

// Pin calls this function every time a new basic block is encountered.
// It inserts a call to docount.
VOID Trace(TRACE trace, VOID *v)
{
    // Visit every basic block  in the trace
    for (BBL bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl))
    {
        // Insert a call to docount for every bbl, passing the number of instructions.
        
        BBL_InsertCall(bbl, IPOINT_ANYWHERE, (AFUNPTR)docount, IARG_FAST_ANALYSIS_CALL,
                       IARG_UINT32, BBL_NumIns(bbl), IARG_THREAD_ID, IARG_END);
    }
}

/*
* trace inspection (instrumentation function)
*
* traverse the basic blocks (BBLs) on the trace and
* inspect every instruction for instrumenting it
* accordingly
*
* @trace:      instructions trace; given by PIN
* @v:		callback value
*/
static void
trace_inspect(TRACE trace, VOID *v)
{
	/* iterators */
	BBL bbl;
	INS ins;
	xed_iclass_enum_t ins_indx;

	/* versioning support */
	ADDRINT version, version_mask = (ADDRINT)v;

	if (version_mask) {
		/* 
		* ignore code cache versions that we
		* are not supposed to instrument
		*/
		version = TRACE_Version(trace);
		if ((version & version_mask) == 0)
			return;
	}

	/* traverse all the BBLs in the trace */
	for (bbl = TRACE_BblHead(trace); BBL_Valid(bbl); bbl = BBL_Next(bbl)) {
		/* traverse all the instructions in the BBL */
		for (ins = BBL_InsHead(bbl);
			INS_Valid(ins);
			ins = INS_Next(ins)) {
				/*
				* use XED to decode the instruction and
				* extract its opcode
				*/
				ins_indx = (xed_iclass_enum_t)INS_Opcode(ins);

				/* 
				* invoke the pre-ins instrumentation callback
				*/
				if (ins_desc[ins_indx].pre != NULL)
					ins_desc[ins_indx].pre(ins);

				/*
				* analyze the instruction (default handler)
				*/
				if (ins_desc[ins_indx].dflact == INSDFL_ENABLE)
					ins_inspect(ins);

				/* 
				* invoke the post-ins instrumentation callback
				*/
				if (ins_desc[ins_indx].post != NULL)
					ins_desc[ins_indx].post(ins);
		}
	}
}