1 llvm-profdata - Profile data tool
2 =================================
4 .. program:: llvm-profdata
9 :program:`llvm-profdata` *command* [*args...*]
14 The :program:`llvm-profdata` tool is a small utility for working with profile
20 * :ref:`merge <profdata-merge>`
21 * :ref:`show <profdata-show>`
22 * :ref:`overlap <profdata-overlap>`
24 .. program:: llvm-profdata merge
34 :program:`llvm-profdata merge` [*options*] [*filename...*]
39 :program:`llvm-profdata merge` takes several profile data files
40 generated by PGO instrumentation and merges them together into a single
41 indexed profile data file.
43 By default profile data is merged without modification. This means that the
44 relative importance of each input file is proportional to the number of samples
45 or counts it contains. In general, the input from a longer training run will be
46 interpreted as relatively more important than a shorter run. Depending on the
47 nature of the training runs it may be useful to adjust the weight given to each
48 input file by using the ``-weighted-input`` option.
50 Profiles passed in via ``-weighted-input``, ``-input-files``, or via positional
51 arguments are processed once for each time they are seen.
59 Print a summary of command line options.
61 .. option:: -output=output, -o=output
63 Specify the output file name. *Output* cannot be ``-`` as the resulting
64 indexed profile data can't be written to standard output.
66 .. option:: -weighted-input=weight,filename
68 Specify an input file name along with a weight. The profile counts of the
69 supplied ``filename`` will be scaled (multiplied) by the supplied
70 ``weight``, where where ``weight`` is a decimal integer >= 1.
71 Input files specified without using this option are assigned a default
72 weight of 1. Examples are shown below.
74 .. option:: -input-files=path, -f=path
76 Specify a file which contains a list of files to merge. The entries in this
77 file are newline-separated. Lines starting with '#' are skipped. Entries may
78 be of the form <filename> or <weight>,<filename>.
80 .. option:: -remapping-file=path, -r=path
82 Specify a file which contains a remapping from symbol names in the input
83 profile to the symbol names that should be used in the output profile. The
84 file should consist of lines of the form ``<input-symbol> <output-symbol>``.
85 Blank lines and lines starting with ``#`` are skipped.
87 The :doc:`llvm-cxxmap <llvm-cxxmap>` tool can be used to generate the symbol
90 .. option:: -instr (default)
92 Specify that the input profile is an instrumentation-based profile.
96 Specify that the input profile is a sample-based profile.
98 The format of the generated file can be generated in one of three ways:
100 .. option:: -binary (default)
102 Emit the profile using a binary encoding. For instrumentation-based profile
103 the output format is the indexed binary format.
107 Emit the profile in text mode. This option can also be used with both
108 sample-based and instrumentation-based profile. When this option is used
109 the profile will be dumped in the text format that is parsable by the profile
114 Emit the profile using GCC's gcov format (Not yet supported).
116 .. option:: -sparse[=true|false]
118 Do not emit function records with 0 execution count. Can only be used in
119 conjunction with -instr. Defaults to false, since it can inhibit compiler
120 optimization during PGO.
122 .. option:: -num-threads=N, -j=N
124 Use N threads to perform profile merging. When N=0, llvm-profdata auto-detects
125 an appropriate number of threads to use. This is the default.
127 .. option:: -failure-mode=[any|all]
129 Set the failure mode. There are two options: 'any' causes the merge command to
130 fail if any profiles are invalid, and 'all' causes the merge command to fail
131 only if all profiles are invalid. If 'all' is set, information from any
132 invalid profiles is excluded from the final merged product. The default
133 failure mode is 'any'.
139 Merge three profiles:
143 llvm-profdata merge foo.profdata bar.profdata baz.profdata -output merged.profdata
147 The input file `foo.profdata` is especially important, multiply its counts by 10:
151 llvm-profdata merge -weighted-input=10,foo.profdata bar.profdata baz.profdata -output merged.profdata
153 Exactly equivalent to the previous invocation (explicit form; useful for programmatic invocation):
157 llvm-profdata merge -weighted-input=10,foo.profdata -weighted-input=1,bar.profdata -weighted-input=1,baz.profdata -output merged.profdata
159 .. program:: llvm-profdata show
169 :program:`llvm-profdata show` [*options*] [*filename*]
174 :program:`llvm-profdata show` takes a profile data file and displays the
175 information about the profile counters for this file and
176 for any of the specified function(s).
178 If *filename* is omitted or is ``-``, then **llvm-profdata show** reads its
179 input from standard input.
184 .. option:: -all-functions
186 Print details for every function.
190 Print the counter values for the displayed functions.
192 .. option:: -function=string
194 Print details for a function if the function's name contains the given string.
198 Print a summary of command line options.
200 .. option:: -output=output, -o=output
202 Specify the output file name. If *output* is ``-`` or it isn't specified,
203 then the output is sent to standard output.
205 .. option:: -instr (default)
207 Specify that the input profile is an instrumentation-based profile.
211 Instruct the profile dumper to show profile counts in the text format of the
212 instrumentation-based profile data representation. By default, the profile
213 information is dumped in a more human readable form (also in text) with
218 Instruct the profile dumper to show the top ``n`` functions with the
219 hottest basic blocks in the summary section. By default, the topn functions
224 Specify that the input profile is a sample-based profile.
226 .. option:: -memop-sizes
228 Show the profiled sizes of the memory intrinsic calls for shown functions.
230 .. option:: -value-cutoff=n
232 Show only those functions whose max count values are greater or equal to ``n``.
233 By default, the value-cutoff is set to 0.
235 .. option:: -list-below-cutoff
237 Only output names of functions whose max count value are below the cutoff
242 Only show context sensitive profile counts. The default is to filter all
243 context sensitive profile counts.
245 .. program:: llvm-profdata overlap
247 .. _profdata-overlap:
255 :program:`llvm-profdata overlap` [*options*] [*base profile file*] [*test profile file*]
260 :program:`llvm-profdata overlap` takes two profile data files and displays the
261 *overlap* of counter distribution between the whole files and between any of the
264 In this command, *overlap* is defined as follows:
265 Suppose *base profile file* has the following counts:
266 {c1_1, c1_2, ..., c1_n, c1_u_1, c2_u_2, ..., c2_u_s},
267 and *test profile file* has
268 {c2_1, c2_2, ..., c2_n, c2_v_1, c2_v_2, ..., c2_v_t}.
269 Here c{1|2}_i (i = 1 .. n) are matched counters and c1_u_i (i = 1 .. s) and
270 c2_v_i (i = 1 .. v) are unmatched counters (or counters only existing in)
271 *base profile file* and *test profile file*, respectively.
272 Let sum_1 = c1_1 + c1_2 + ... + c1_n + c1_u_1 + c2_u_2 + ... + c2_u_s, and
273 sum_2 = c2_1 + c2_2 + ... + c2_n + c2_v_1 + c2_v_2 + ... + c2_v_t.
274 *overlap* = min(c1_1/sum_1, c2_1/sum_2) + min(c1_2/sum_1, c2_2/sum_2) + ...
275 + min(c1_n/sum_1, c2_n/sum_2).
277 The result overlap distribution is a percentage number, ranging from 0.0% to
278 100.0%, where 0.0% means there is no overlap and 100.0% means a perfect
281 Here is an example, if *base profile file* has counts of {400, 600}, and
282 *test profile file* has matched counts of {60000, 40000}. The *overlap* is 80%.
287 .. option:: -function=string
289 Print details for a function if the function's name contains the given string.
293 Print a summary of command line options.
295 .. option:: -o=output or -o output
297 Specify the output file name. If *output* is ``-`` or it isn't specified,
298 then the output is sent to standard output.
300 .. option:: -value-cutoff=n
302 Show only those functions whose max count values are greater or equal to ``n``.
303 By default, the value-cutoff is set to max of unsigned long long.
307 Only show overlap for the context sensitive profile counts. The default is to show
308 non-context sensitive profile counts.
313 :program:`llvm-profdata` returns 1 if the command is omitted or is invalid,
314 if it cannot read input files, or if there is a mismatch between their data.