README for GPROF This is the GNU profiler. It is distributed with other "binary utilities" which should be in ../binutils. See ../binutils/README for more general notes, including where to send bug reports. This file documents the changes and new features available with this version of GNU gprof. * New Features o Long options o Supports generalized file format, without breaking backward compatibility: new file format supports basic-block execution counts and non-realtime histograms (see below) o Supports profiling at the line level: flat profiles, call-graph profiles, and execution-counts can all be displayed at a level that identifies individual lines rather than just functions o Test-coverage support (similar to Sun tcov program): source files can be annotated with the number of times a function was invoked or with the number of times each basic-block in a function was executed o Generalized histograms: not just execution-time, but arbitrary histograms are support (for example, performance counter based profiles) o Powerful mechanism to select data to be included/excluded from analysis and/or output o Support for DEC OSF/1 v3.0 o Full cross-platform profiling support: gprof uses BFD to support arbitrary, non-native object file formats and non-native byte-orders (this feature has not been tested yet) o In the call-graph function index, static function names are now printed together with the filename in which the function was defined (required bfd_find_nearest_line() support and symbolic debugging information to be present in the executable file) o Major overhaul of source code (compiles cleanly with -Wall, etc.) * Supported Platforms The current version is known to work on: o DEC OSF/1 v3.0 All features supported. o SunOS 4.1.x All features supported. o Solaris 2.3 Line-level profiling unsupported because bfd_find_nearest_line() is not fully implemented for Elf binaries. o HP-UX 9.01 Line-level profiling unsupported because bfd_find_nearest_line() is not fully implemented for SOM binaries. * Detailed Description ** User Interface Changes The command-line interface is backwards compatible with earlier versions of GNU gprof and Berkeley gprof. The only exception is the option to delete arcs from the call graph. The old syntax was: -k fromname toname while the new syntax is: -k fromname/toname This change was necessary to be compatible with long-option parsing. Also, "fromname" and "toname" can now be arbitrary symspecs rather than just function names (see below for an explanation of symspecs). For example, option "-k gprof.c/" suppresses all arcs due to calls out of file "gprof.c". *** Sym Specs It is often necessary to apply gprof only to specific parts of a program. GNU gprof has a simple but powerful mechanism to achieve this. So called {\em symspecs\/} provide the foundation for this mechanism. A symspec selects the parts of a profiled program to which an operation should be applied to. The syntax of a symspec is simple: filename_containing_a_dot | funcname_not_containing_a_dot | linenumber | ( [ any_filename ] `:' ( any_funcname | linenumber ) ) Here are some examples: main.c Selects everything in file "main.c"---the dot in the string tells gprof to interpret the string as a filename, rather than as a function name. To select a file whose name does contain a dot, a trailing colon should be specified. For example, "odd:" is interpreted as the file named "odd". main Selects all functions named "main". Notice that there may be multiple instances of the same function name because some of the definitions may be local (i.e., static). Unless a function name is unique in a program, you must use the colon notation explained below to specify a function from a specific source file. Sometimes, functionnames contain dots. In such cases, it is necessary to add a leading colon to the name. For example, ":.mul" selects function ".mul". main.c:main Selects function "main" in file "main.c". main.c:134 Selects line 134 in file "main.c". IMPLEMENTATION NOTE: The source code uses the type sym_id for symspecs. At some point, this probably ought to be changed to "sym_spec" to make reading the code easier. *** Long options GNU gprof now supports long options. The following is a list of all supported options. Options that are listed without description operate in the same manner as the corresponding option in older versions of gprof. Short Form: Long Form: ----------- ---------- -l --line Request profiling at the line-level rather than just at the function level. Source lines are identified by symbols of the form: func (file:line) where "func" is the function name, "file" is the file name and "line" is the line-number that corresponds to the line. To work properly, the binary must contain symbolic debugging information. This means that the source have to be translated with option "-g" specified. Functions for which there is no symbolic debugging information available are treated as if "--line" had not been specified. However, the line number printed with such symbols is usually incorrect and should be ignored. -a --no-static -A[symspec] --annotated-source[=symspec] Request output in the form of annotated source files. If "symspec" is specified, print output only for symbols selected by "symspec". If the option is specified multiple times, annotated output is generated for the union of all symspecs. Examples: -A Prints annotated source for all source files. -Agprof.c Prints annotated source for file gprof.c. -Afoobar Prints annotated source for files containing a function named "foobar". The entire file will be printed, but only the function itself will be annotated with profile data. -J[symspec] --no-annotated-source[=symspec] Suppress annotated source output. If specified without argument, annotated output is suppressed completely. With an argument, annotated output is suppressed only for the symbols selected by "symspec". If the option is specified multiple times, annotated output is suppressed for the union of all symspecs. This option has lower precedence than --annotated-source -p[symspec] --flat-profile[=symspec] Request output in the form of a flat profile (unless any other output-style option is specified, this option is turned on by default). If "symspec" is specified, include only symbols selected by "symspec" in flat profile. If the option is specified multiple times, the flat profile includes symbols selected by the union of all symspecs. -P[symspec] --no-flat-profile[=symspec] Suppress output in the flat profile. If given without an argument, the flat profile is suppressed completely. If "symspec" is specified, suppress the selected symbols in the flat profile. If the option is specified multiple times, the union of the selected symbols is suppressed. This option has lower precedence than --flat-profile. -q[symspec] --graph[=symspec] Request output in the form of a call-graph (unless any other output-style option is specified, this option is turned on by default). If "symspec" is specified, include only symbols selected by "symspec" in the call-graph. If the option is specified multiple times, the call-graph includes symbols selected by the union of all symspecs. -Q[symspec] --no-graph[=symspec] Suppress output in the call-graph. If given without an argument, the call-graph is suppressed completely. With a "symspec", suppress the selected symbols from the call-graph. If the option is specified multiple times, the union of the selected symbols is suppressed. This option has lower precedence than --graph. -C[symspec] --exec-counts[=symspec] Request output in the form of execution counts. If "symspec" is present, include only symbols selected by "symspec" in the execution count listing. If the option is specified multiple times, the execution count listing includes symbols selected by the union of all symspecs. -Z[symspec] --no-exec-counts[=symspec] Suppress output in the execution count listing. If given without an argument, the listing is suppressed completely. With a "symspec", suppress the selected symbols from the call-graph. If the option is specified multiple times, the union of the selected symbols is suppressed. This option has lower precedence than --exec-counts. -i --file-info Print information about the profile files that are read. The information consists of the number and types of records present in the profile file. Currently, a profile file can contain any number and any combination of histogram, call-graph, or basic-block count records. -s --sum -x --all-lines This option affects annotated source output only. By default, only the lines at the beginning of a basic-block are annotated. If this option is specified, every line in a basic-block is annotated by repeating the annotation for the first line. This option is identical to tcov's "-a". -I dirs --directory-path=dirs This option affects annotated source output only. Specifies the list of directories to be searched for source files. The argument "dirs" is a colon separated list of directories. By default, gprof searches for source files relative to the current working directory only. -z --display-unused-functions -m num --min-count=num This option affects annotated source and execution count output only. Symbols that are executed less than "num" times are suppressed. For annotated source output, suppressed symbols are marked by five hash-marks (#####). In an execution count output, suppressed symbols do not appear at all. -L --print-path Normally, source filenames are printed with the path component suppressed. With this option, gprof can be forced to print the full pathname of source filenames. The full pathname is determined from symbolic debugging information in the image file and is relative to the directory in which the compiler was invoked. -y --separate-files This option affects annotated source output only. Normally, gprof prints annotated source files to standard-output. If this option is specified, annotated source for a file named "path/filename" is generated in the file "filename-ann". That is, annotated output is {\em always\/} generated in gprof's current working directory. Care has to be taken if a program consists of files that have identical filenames, but distinct paths. -c --static-call-graph -t num --table-length=num This option affects annotated source output only. After annotating a source file, gprof generates an execution count summary consisting of a table of lines with the top execution counts. By default, this table is ten entries long. This option can be used to change the table length or, by specifying an argument value of 0, it can be suppressed completely. -n symspec --time=symspec Only symbols selected by "symspec" are considered in total and percentage time computations. However, this option does not affect percentage time computation for the flat profile. If the option is specified multiple times, the union of all selected symbols is used in time computations. -N --no-time=symspec Exclude the symbols selected by "symspec" from total and percentage time computations. However, this option does not affect percentage time computation for the flat profile. This option is ignored if any --time options are specified. -w num --width=num Sets the output line width. Currently, this option affects the printing of the call-graph function index only. -e -E -f -F -k -b --brief -dnum --debug[=num] -h --help Prints a usage message. -O name --file-format=name Selects the format of the profile data files. Recognized formats are "auto", "bsd", "magic", and "prof". The last one is not yet supported. Format "auto" attempts to detect the file format automatically (this is the default behavior). It attempts to read the profile data files as "magic" files and if this fails, falls back to the "bsd" format. "bsd" forces gprof to read the data files in the BSD format. "magic" forces gprof to read the data files in the "magic" format. -T --traditional -v --version ** File Format Changes The old BSD-derived format used for profile data does not contain a magic cookie that allows to check whether a data file really is a gprof file. Furthermore, it does not provide a version number, thus rendering changes to the file format almost impossible. GNU gprof uses a new file format that provides these features. For backward compatibility, GNU gprof continues to support the old BSD-derived format, but not all features are supported with it. For example, basic-block execution counts cannot be accommodated by the old file format. The new file format is defined in header file \file{gmon_out.h}. It consists of a header containing the magic cookie and a version number, as well as some spare bytes available for future extensions. All data in a profile data file is in the native format of the host on which the profile was collected. GNU gprof adapts automatically to the byte-order in use. In the new file format, the header is followed by a sequence of records. Currently, there are three different record types: histogram records, call-graph arc records, and basic-block execution count records. Each file can contain any number of each record type. When reading a file, GNU gprof will ensure records of the same type are compatible with each other and compute the union of all records. For example, for basic-block execution counts, the union is simply the sum of all execution counts for each basic-block. *** Histogram Records Histogram records consist of a header that is followed by an array of bins. The header contains the text-segment range that the histogram spans, the size of the histogram in bytes (unlike in the old BSD format, this does not include the size of the header), the rate of the profiling clock, and the physical dimension that the bin counts represent after being scaled by the profiling clock rate. The physical dimension is specified in two parts: a long name of up to 15 characters and a single character abbreviation. For example, a histogram representing real-time would specify the long name as "seconds" and the abbreviation as "s". This feature is useful for architectures that support performance monitor hardware (which, fortunately, is becoming increasingly common). For example, under DEC OSF/1, the "uprofile" command can be used to produce a histogram of, say, instruction cache misses. In this case, the dimension in the histogram header could be set to "i-cache misses" and the abbreviation could be set to "1" (because it is simply a count, not a physical dimension). Also, the profiling rate would have to be set to 1 in this case. Histogram bins are 16-bit numbers and each bin represent an equal amount of text-space. For example, if the text-segment is one thousand bytes long and if there are ten bins in the histogram, each bin represents one hundred bytes. *** Call-Graph Records Call-graph records have a format that is identical to the one used in the BSD-derived file format. It consists of an arc in the call graph and a count indicating the number of times the arc was traversed during program execution. Arcs are specified by a pair of addresses: the first must be within caller's function and the second must be within the callee's function. When performing profiling at the function level, these addresses can point anywhere within the respective function. However, when profiling at the line-level, it is better if the addresses are as close to the call-site/entry-point as possible. This will ensure that the line-level call-graph is able to identify exactly which line of source code performed calls to a function. *** Basic-Block Execution Count Records Basic-block execution count records consist of a header followed by a sequence of address/count pairs. The header simply specifies the length of the sequence. In an address/count pair, the address identifies a basic-block and the count specifies the number of times that basic-block was executed. Any address within the basic-address can be used. IMPLEMENTATION NOTE: gcc -a can be used to instrument a program to record basic-block execution counts. However, the __bb_exit_func() that is currently present in libgcc2.c does not generate a gmon.out file in a suitable format. This should be fixed for future releases of gcc. In the meantime, contact davidm@cs.arizona.edu for a version of __bb_exit_func() to is appropriate. Copyright (C) 2012 Free Software Foundation, Inc. Copying and distribution of this file, with or without modification, are permitted in any medium without royalty provided the copyright notice and this notice are preserved.