Method and apparatus for accurate profiling of computer programs

Claim 2

A computer implemented method for profiling execution of a computer program having functions comprising instructions, the method comprising the steps of: for each function, parsing the function instructions into basic blocks, wherein a basic block consists of a contiguous sequence of at least one instruction of which the first instruction is a program control transfer destination, the last instruction is a program control transfer instruction, and any instructions between first and last instructions are neither program control transfer destinations nor program control transfer instructions; and for each basic block, performing under computer control, the steps of selecting a basic block; analyzing the selected basic block to determine a fixed number of clock cycles for the selected basic block, and to determine whether the selected basic block contains an operating system (OS) call; and adding basic block profiling code so as to be executed whenever the basic block is executed, said basic block profiling code operating to add the determined fixed number of clock cycles to a clock cycle accumulator for the function, and further including, for at least one selected basic block containing an OS call, the OS call timing profiling steps of adding timing start code to determine a start time immediately before executing the OS call; adding timing end code to determine an end time immediately after executing the OS call; adding OS call timing calculation code to determine from the start time and end time a number of clock cycles required to execute the OS call; and adding OS call timing recording code to add the determined number of clock cycles to a clock cycle accumulator for the function; wherein said OS call timing profiling steps are performed for substantially only basic blocks containing indeterminate length OS calls.

Claim 3

A computer implemented method for profiling execution of a computer program having functions, the functions having instructions, the method comprising the steps of: a) augmenting the program by adding profiling code to the program; b) executing the augmented program and collecting profiling data for functions of the program, said profiling data including self+descendants times for the functions wherein the self+descendants times are percentages of a total execution time for the program; c) constructing a call graph for the program; and d) displaying to a user a representation of the call graph including arcs connecting nodes, wherein each arc links a parent function to a child function and has a width corresponding to the self+descendants time for the child function, and wherein the step of displaying the representation of the call graph comprises displaying a call graph in which each arc has a width having a linear relationship to the percentage self+descendants time of the child function of the arc.

Claim 4

A computer implemented method for profiling execution of a computer program having functions, the functions having instructions, the method comprising the steps of: a) augmenting the program by adding profiling code to the program; b) executing the augmented program and collecting profiling data for functions of the program, said profiling data including self+descendants times for the functions wherein the self+descendants times are percentages of a total execution time for the program; c) constructing a call graph for the program; and d) displaying to a user a representation of the call graph including arcs connecting nodes, wherein each arc links a parent function to a child function and has a width corresponding to the self+descendants time for the child function, and wherein the step of displaying the representation of the call graph comprises displaying a call graph in which each arc has a width having a linear relationship to the percentage self+descendants time of the child function of the arc.

Claim 5

A computer implemented method for profiling execution of a computer program having the functions, the functions having instruction, the method comprising the steps of: a) augmenting the program by adding profiling code to the program; b) executing the augmented program and collecting profiling data for functions of the program, said profiling including self+descendants times for the functions wherein the self+descendants times are percentages of a total execution time for the program; c) constructing a call graph for the program; and d) displaying to a user a representation of the call graph including arcs connecting nodes, wherein each arc links a parent function to a child function and has a width corresponding to the self+descendants time for the child function and wherein the step of displaying the representation of the call graph comprise displaying a call graph in which each arc has a width having a logarithmic relationship to the percentage self+descendants time of the child function of the arc.

Claim 6

A computer implemented method for profiling execution of a computer program having the functions, the functions having instruction, the method comprising the steps of: a) augmenting the program by adding profiling code to the program; b) executing the augmented program and collecting profiling data for functions of the program, said profiling including self+descendants times for the functions wherein the self+descendants times are percentages of a total execution time for the program; c) constructing a call graph for the program; and d) displaying to a user a representation of the call graph including arcs connecting nodes, wherein each arc links a parent function to a child function and has a width corresponding to the self+descendants time for the child function and wherein the step of displaying the representation of the call graph comprise displaying a call graph in which each arc has a width having a logarithmic relationship to the percentage self+descendants time of the child function of the arc.

Claim 7

A computer implemented method for profiling execution of a computer program having functions, the functions having instructions, the method comprising the steps of: a) augmenting the program by adding profiling code to the program; b) executing the augmented program and collecting profiling data for functions of the program, said profiling data including self+descendants times for the functions; c) constructing a call graph showing function calls for the program; d) providing a pruning time filter value; d) displaying to a user a representation of the call graph including arcs connecting nodes, wherein each arc links a parent function to a child function, wherein a set of arcs is automatically pruned from said displayed call graph representation, wherein each of said set of said automatically pruned arcs connects to a child function having a self+descendants time less than said pruning time filter value.

Claim 8

A computer implemented method for profiling execution of a computer program having functions, the functions having instructions, the method comprising the steps of: a) augmenting the program by adding profiling code to the program; b) executing the augmented program and collecting profiling data for functions of the program, said profiling data including self+descendants times for the functions; c) constructing a call graph showing function calls for the program; d) providing a pruning time filter value; d) displaying to a user a representation of the call graph including arcs connecting nodes, wherein each arc links a parent function to a child function, wherein a set of arcs is automatically pruned from said displayed call graph representation, wherein each of said set of said automatically pruned arcs connects to a child function having a self+descendants time less than said pruning time filter value.

Claim 9

The method of claim 4, wherein said step of providing the pruning time filter value comprises automatically determining, under computer control, a time value such that a predetermined number of the functions of the program have a self+descendants time at least equal to the determined time value.

Claim 10

The method of claim 5, wherein the predetermined number of functions is less than about one hundred.

Claim 11

The method of claim 6, wherein the predetermined number of functions is about thirty.

Claim 12

A computer implemented method for profiling execution of a computer program having functions comprising instructions, the method comprising the steps of: for each function, parsing the function instructions into basic block, wherein a basic block consists of a contiguous sequence of at least one instruction of which the first instruction is a program control transfer destination, the last instruction is a program control transfer instruction, and any instructions between first and last instructions are neither program control transfer destinations nor program control transfer instructions; and for each basic block, performing under computer control, the steps of selecting a basic block; analyzing the selected basic block to determine a fixed number of clock cycles for the selected basic block, and to determine whether the selected basic block contains an operating system (OS) call; and adding basic block profiling code so as to be executed whenever the basic block is executed, said basic block profiling code operating to add the determined fixed number of clock cycles to a clock cycle accumulator for the function, and further including, for at least one selected basic block containing an OS call, the OS call timing profiling steps of adding timing start code to determine a start time immediately before executing the OS call; adding timing end code to determine an end time immediately after executing the OS call; adding OS call timing calculation code to determine from the start time and end time a number of clock cycles required to execute the OS call; and adding OS call timing recording code to add the determined number of clock cycles to a clock cycle accumulator for the function; wherein the added basic block profiling code operates to add the determined fixed number of clock cycles, and the added OS call timing recording code operates to add the determined number of cycles required to execute the OS call, to a clock cycle accumulator for the selected basic block of the function.

Claim 13

A computer implemented method for profiling execution of a computer program having functions comprising instructions, the method comprising the steps of: for each function, parsing the function instructions into basic block, wherein a basic block consists of a contiguous sequence of at least one instruction of which the first instruction is a program control transfer destination, the last instruction is a program control transfer instruction, and any instructions between first and last instructions are neither program control transfer destinations nor program control transfer instructions; and for each basic block, performing under computer control, the steps of selecting a basic block; analyzing the selected basic block to determine a fixed number of clock cycles for the selected basic block, and to determine whether the selected basic block contains an operating system (OS) call; and adding basic block profiling code so as to be executed whenever the basic block is executed, said basic block profiling code operating to add the determined fixed number of clock cycles to a clock cycle accumulator for the function, and further including, for at least one selected basic block containing an OS call, the OS call timing profiling steps of adding timing start code to determine a start time immediately before executing the OS call; adding timing end code to determine an end time immediately after executing the OS call; adding OS call timing calculation code to determine from the start time and end time a number of clock cycles required to execute the OS call; and adding OS call timing recording code to add the determined number of clock cycles to a clock cycle accumulator for the function; wherein the added basic block profiling code operates to add the determined fixed number of clock cycles, and the added OS call timing recording code operates to add the determined number of cycles required to execute the OS call, to a clock cycle accumulator for the selected basic block of the function.