atom_application_query(5) — Macro Packages and Conventions

NAME

atom_application_query, GetProgInfo, GetObjInfo, GetObjName, GetObjOutName, GetAnalName, GetObjInstArray, GetObjInstCount, GetProcInfo, ProcName, ProcFileName, ProcPC, GetBlockInfo, BlockPC, IsBranchTarget, GetInstClass, IsInstType, GetInstInfo, InstPC, InstLineNo, GetInstBinary, GetInstRegEnum, GetInstRegUsage − Allows an Atom-tool instrumentation routine to obtain information about the parts of an application program

SYNOPSIS

#include <cmplrs/atom.inst.h>
const char ∗GetAnalName(
        void ); long GetProgInfo(
        ProgInfoType ); long GetObjInfo(
        Obj ∗,
        ObjInfoType ); const char ∗GetObjName(
        Obj ∗ ); const char ∗GetObjOutName(
        Obj ∗ ); const unsigned int ∗GetObjInstArray(
        Obj ∗ ); long GetObjInstCount(
        Obj ∗ ); long GetProcInfo(
        Proc ∗,
        ProcInfoType ); const char ∗ProcName(
        Proc ∗ ); const char ∗ProcFileName(
        Proc ∗ ); long ProcPC(
        Proc ∗ ); long GetBlockInfo(
        Block ∗,
        BlockInfoType ); long BlockPC(
        Block ∗ ); unsigned IsBranchTarget(
        Block ∗ ); IClassType GetInstClass(
        Inst ∗ ); int IsInstType(
        Inst ∗,
        ITypeType ); int GetInstInfo(
        Inst ∗,
        InstInfoType ); long InstPC(
        Inst ∗ ); long InstLineNo(
        Inst ∗ ); int GetInstBinary(
        long ); RegvType GetInstRegEnum(
        Inst ∗,
        InstInfoType ); void GetInstRegUsage(
        Inst ∗,
        InstRegUsageVec ∗ );

DESCRIPTION

Atom’s application query routines allow an instrumentation routine to obtain information about the parts of an application program.

You can use these routines only from an Atom tool’s instrumentation file. See atom(1) for a description of Atom.

GetAnalName Routine

Use the GetAnalName routine to obtain the name of the analysis file, as passed to the atom command. This routine is useful for tools that have a single instrumentation file and multiple analysis files.

GetProgInfo Routine

Use the GetProgInfo routine to obtain the number of objects in a program. The only accepted ProgInfo value is ProgNumberObjects.

Each object of a program is a self-contained program with text, unitialized, and data sections. The text section contains instructions. Initialized data includes data that must be initialized to nonzero values; unitialized data is filled with zeroes when the corresponding pages are mapped into the user’s address space.

GetObjInfo Routine

Use the GetObjInfo routine to obtain information about the specified Obj. The following ObjInfoType values return the indicated information:

ObjInfoType	Description
ObjTextStartAddress	Returns the starting compile-time address of the object’s text segment. When instrumenting a nonshared executable or the main executable of call-shared programs, the compile-time address is identical to the run-time address. For shared libraries, the compiler may place an object at a different address than its compile-time address.
ObjTextSize	Returns the size in bytes of the object’s text segment.
ObjInitDataStartAddress	Returns the starting address of the object’s data segment.
ObjInitDataSize	Returns the size in bytes of the object’s data segment.
ObjUninitDataStartAddress	Returns the starting address of the object’s bss segment.
ObjUninitDataSize	Returns the size in bytes of the object’s bss segment.
ObjNumberProcs	Returns the number of procedures in the object.
ObjNumberBlocks	Returns the number of basic blocks in the object.
ObjNumberInsts	Returns the number of instructions in the object. Note that the number of instructions returned is not usually equal to the value returned by ObjTextSize divided by the instruction size (4 bytes). The compiler aligns procedures on 16-byte boundaries by padding them on both sides with NOP instructions. Because these instructions are never executed, they are not included in the count returned by ObjNumberInsts.
ObjID	Returns a unique numeric identifier for the object. This identifier is unique to this object within the entire application program.
ObjModifyHint	Returns a Boolean value indicating whether the user requested that the object be instrumented or not. A user indicates which objects are or are not to be instrumented by specifying the −all, −incobj, or −excobj flags on the atom command line. The ObjModifyHint type returns the value OBJ_READONLY if the user did not ask for the object to be instrumented or OBJ_WRITEABLE if the user asked for the object to be instrumented. Certain tools can use this hint to decide whether to exclude particular objects from instrumentation. Although an Atom tool can disregard these flag settings and instrument objects as it sees fit, it typically honors the user’s requests.

GetObjName and GetObjOutName Routines

Use the GetObjName routine to obtain the original filename of the specified object.

Use the GetObjOutName routine to obtain the name of the instrumented object.

GetObjInstArray and GetObjInstCount Routines

Use the GetObjInstArray routine to obtain an array consisting of the 32-bit instructions included in the specified Obj.

Use GetObjInstCount to obtain the number of instructions in the array.

GetProcInfo Routine

Use the GetProcInfo routine to obtain information about the specified Proc. The following ProcInfoType values return the indicated information:

ProcInfoType	Description
ProcFrameSize	Returns the size of the fixed portion of the procedure’s stack frame.
ProcIRegMask	Returns the procedure’s saved integer register mask.
ProcIRegOffset	Returns the offset to the procedure’s integer register save area in the stack frame.
ProcFRegMask	Returns the procedure’s saved floating-point register mask.
ProcFRegOffset	Returns the offset to the procedure’s floating-point register save area in the stack frame.
ProcgpPrologue	Returns the size in bytes of the global pointer (GP) prologue.
ProcgpUsed	Returns a nonzero value if the procedure uses the GP register ($gp) and zero (0) if it does not.
ProcLocalOffset	Returns the offset to the procedure’s local variables from the virtual frame pointer.
ProcFrameReg	Returns the number of the register that is being used as the procedure’s frame pointer.
ProcPcReg	Returns the number of the register that contains the procedure’s return address.
ProcNumberBlocks	Returns the number of basic blocks in the procedure.
ProcNumberInsts	Returns the number of instructions in the procedure.
ProcID	Returns a unique numeric identifier for the procedure. This identifier is unique across all shared libraries and can thus be used to create per-procedure data structures.
ProcLineLow	Returns the lowest source line in the procedure, or zero (0) if source line information is unavailable.
ProcLineHigh	Returns the highest source line in the procedure, or zero (0) if source line information is unavailable. This value and the value returned by ProcLineLow are useful for determining when the compiler has performed inline operations that have changed the line number of an instruction to a value that is not in the range of the current procedure.
ProcAddrTaken	Returns a nonzero value if the program has taken the procedure’s address and zero (0) if it has not. An Atom tool can use this value to determine if the procedure is a potential target of an indirect procedure call.
ProcIsRegFrame	Returns a nonzero value if the procedure has a register frame.

ProcName and ProcFileName Routines

Use the ProcName routine to obtain the name of the specified Proc.

Use the ProcFileName routine to obtain the name of the source file that contains the specified Proc. If local symbols are not present, the value NULL is returned. Use to ProcFileName and InstLineNo routines to identify the file name and line number of any instruction in the application program.

ProcPC Routine

Use the ProcPC routine to obtain the compile-time program counter (PC) of the first instruction in the procedure.

GetBlockInfo Routine

Use the GetBlockInfo routine to obtain information about the pecified Block. The following BlockInfoType values return the indicated information:

BlockInfoType	Description
BlockNumberInsts	Returns the number of instructions in the basic block.
BlockID	Returns a unique numeric identifier for the basic block. This identifier is unique to this basic block within its containing object.

BlockPC Routine

Use the BlockPC routine to obtain the compile-time program counter (PC) of the first instruction in the basic block.

IsBranchTarget Routine

Use the IsBranchTarget routine to determine if the specified Block is the target of a branch instruction, such as br, bsr, a conditional integer branch, or a conditional floating-point branch. IsBranchTarget returns a nonzero value if the basic block is the target of a branch and zero (0) if it is not. A tool that builds a call flow graph would find this information useful.

GetInstClass Routine

Use the GetInstClass routine to obtain the class of the specified Inst. The following IClassType values may be returned:

IClassType	Description
ClassLoad	Integer load instruction
ClassFload	Floating-point load instruction
ClassStore	Integer store data instruction
ClassFstore	Floating-point store data instruction
ClassIbranch	Integer branch instruction
ClassFbranch	Floating-point branch instruction
ClassSubroutine	Integer subroutine call instruction
ClassIarithmetic	Integer arithmetic instruction
ClassImultiplyl	Integer longword multiply instruction
ClassImultiplyq	Integer quadword multiply instruction
ClassIlogical	Logical function instruction
ClassIshift	Shift function instruction
ClassIcondmove	Conditional move instruction
ClassIcompare	Integer compare instruction
ClassFpop	Other floating-point operations
ClassFdivs	Floating-point single precision divide instruction
ClassFdivd	Floating-point double precision divide instruction
ClassNull	call pal instruction, hw_x instruction, etc

IsInstType Routine

Use the IsInstType routine to determine if the specified Inst is an instruction of the specified type. You can specify any of the following ITypeType values. IsInstType returns a nonzero value if the instruction is of the specified type and zero (0) if it is not. Note that any instruction can be of more than one ITypeType.

ITypeType	Description
InstTypeLoad	Integer or floating-point load instruction
InstTypeStore	Integer or floating-point store instruction
InstTypeJump	Jump, jump to subroutine, or return instruction
InstTypeFP	Any floating-point instruction
InstTypeInt	Any integer or nonfloating-point instruction
InstTypeDiv	Single or double precision divide instruction
InstTypeMul	Integer or floating-point multiply instruction
InstTypeAdd	Integer or floating-point add instruction
InstTypeSub	Integer or floating-point subtract instruction
InstTypeCondBr	Integer or floating-point conditional branch instruction
InstTypeUncondBr	Integer or floating-point unconditional branch instruction, but not a subroutine call

GetInstInfo Routine

Use the GetInstInfo routine to parse an entire 32-bit instruction and obtain all or a portion of that instruction. You can specify any of the following InstInfoType values:

InstInfoType	Description
InstMemDisp	Returns the 16-bit memory-format displacement field, sign-extended to 32 bits (even if the instruction does not reference memory).
InstBrDisp	Returns the sign-extended, branch-format displacement field.
InstRA	Returns register field A. Supply InstA to a call to GetInstRegEnum to determine whether the instruction is an integer or floating-point instruction.
InstRB	Returns register field B. Supply InstB to a call to GetInstRegEnum to determine whether the instruction is an integer or floating-point instruction.
InstRC	Returns register field C. Supply InstC to a call to GetInstRegEnum to determine whether the instruction is an integer or floating-point instruction.
InstOpcode	Returns the instruction’s opcode.
InstBinary	Returns a 32-bit binary representation of the assembly language instruction.
InstAddrTaken	Returns a nonzero value if the instruction’s address is taken and zero (0) if it is not.
InstEntryPoint	Returns a nonzero value if the instruction is a procedure entry point and zero (0) if it is not.

InstPC Routine

Use the InstPC routine to obtain the compile-time program counter (PC) of the instruction.

InstLineNo Routine

Use the InstLineNo routine to obtain the specified instruction’s source line number. Often, an Atom tool’s instrumentation routine uses this information with the information returned by a call to the ProcFileName routine to obtain the corresponding line in the source file. A value of 0 (zero) is returned if local symbol table information is not present.

GetInstBinary Routine

Use the GetInstBinary routine to obtain a 32-bit binary representation of the assembly language instruction.

GetInstRegEnum Routine

Use the GetInstRegEnum routine to obtain the register type from an instruction field (IInstInfoType). (See the description of AddCallProto in the atom_application_instrumentation(5) reference page for a list of register types.) If the specified register is not defined for this instruction, GetInstRegEnum returns REG_NOTUSED.

For example, if you supply an argument of InstRA to GetInstRegEnum and the RA field of the instruction contains a 5, GetInstRegEnum returns REG_5 if the instruction is an integer instruction and FREG_5 if it is a floating-point instruction.

GetInstRegUsage Routine

The GetInstRegUsage routine returns a structure containing two vectors. The first vector is a bitmask with bits set for each register read by the instruction, and the second vector is a bitmask with bits set for each register stored by the instruction. Each vector is composed of two 64-bit integers: the low 32 bits correspond to the integer registers, the next 32 bits correspond to the floating-point registers, and the next two bits refer to the program counter (PC) and the cycle counter, respectively.

The format of the InstRegUsageVec structure is as follows:

typedef struct inst_reg_usage{
unsigned long ureg_bitvec[2];
unsigned long dreg_bitvec[2];
} InstRegUsageVec;

Consider the following example:

InstRegUsageVec usageVec;
Inst ∗inst = GetLastInst(GetFirstBlock(GetFirstProc()));
GetInstRegUsage(inst,&usageVec);

This small code fragment sets inst to point to the last instruction in the first basic block in the first procedure. Assume the first instruction was

ADDQ r0,r2,r7

This instruction adds of the contents of register 0 to the contents of register 2 and places the result in register 7. The value returned in usageVec.ureg_bitvec[0] is 0x5 because register 0 and register 2 are both used. The value of usageVec.dreg_bitvec[0] is set to 0x40, indicating that register 7 is set by the add instruction.

One use of this primitive would be to determine if the register loaded in one cycle is used in the next cycle. In most pipelined processors, this results in a one cycle pipeline stall (also known as a load/use conflict). The LoadUseConflict procedure in the following example uses the GetInstRegUsage routine to detect such conflicts:

   1 int LoadUseConflict(Inst ∗instA,Inst ∗instB) {
   2    int confict;
   3    InstRegUsageVec vecA,vecB;
   4    GetInstRegUsage(instA,&vecA);
   5    GetInstRegUsage(instB,&vecB);
   6    conflict = (vecA.dreg_bitvec[0] & vecB.ureg_bitvec[0]) ||
   7                vecA.dreg_bitvec[1] & vecB.ureg_bitvec[1]);
   8    return(IsInstType(instA,InstTypeLoad) && conflict);
   9 }

The arguments are two instructions, A and B. Line 3, 4, and 5 define and set the bit usage bit vectors for the two instructions. Lines 6 and 7 perform a logical-AND operation on the destination vector of instruction A with the usage vector of instruction B. Line 8 returns TRUE if instruction A is a load instruction and a load/use conflict was detected. In this example, LoadUseConflict would return TRUE because register 1 is the destination of the LDQ instruction and the source for the ADDQ instruction.

RETURN VALUES

These routines return the values described in the preceding section.

FILES

/usr/include/cmplrs/atom.inst.h
Header file containing external definitions of Atom routines

Museum

Related Articles