1 Using the glibc microbenchmark suite
2 ====================================
4 The glibc microbenchmark suite automatically generates code for specified
5 functions, builds and calls them repeatedly for given inputs to give some
6 basic performance properties of the function.
11 The benchmark needs python 2.7 or later in addition to the
12 dependencies required to build the GNU C Library. One may run the
13 benchmark by invoking make as follows:
17 This runs each function for 10 seconds and appends its output to
18 benchtests/bench.out. To ensure that the tests are rebuilt, one could run:
22 The duration of each test can be configured setting the BENCH_DURATION variable
23 in the call to make. One should run `make bench-clean' before changing
26 $ make BENCH_DURATION=1 bench
28 The benchmark suite does function call measurements using architecture-specific
29 high precision timing instructions whenever available. When such support is
30 not available, it uses clock_gettime (CLOCK_PROCESS_CPUTIME_ID). One can force
31 the benchmark to use clock_gettime by invoking make as follows:
33 $ make USE_CLOCK_GETTIME=1 bench
35 Again, one must run `make bench-clean' before changing the measurement method.
37 Running benchmarks on another target:
38 ====================================
40 If the target where you want to run benchmarks is not capable of building the
41 code or you're cross-building, you could build and execute the benchmark in
42 separate steps. On the build system run:
46 and then copy the source and build directories to the target and run the
47 benchmarks from the build directory as usual:
51 make sure the copy preserves timestamps by using either rsync or scp -p
52 otherwise the above command may try to build the benchmark again. Benchmarks
53 that require generated code to be executed during the build are skipped when
56 Adding a function to benchtests:
57 ===============================
59 If the name of the function is `foo', then the following procedure should allow
60 one to add `foo' to the bench tests:
62 - Append the function name to the bench variable in the Makefile.
64 - Make a file called `foo-inputs` to provide the definition and input for the
65 function. The file should have some directives telling the parser script
66 about the function and then one input per line. Directives are lines that
67 have a special meaning for the parser and they begin with two hashes '##'.
68 The following directives are recognized:
70 - args: This should be assigned a colon separated list of types of the input
71 arguments. This directive may be skipped if the function does not take any
72 inputs. One may identify output arguments by nesting them in <>. The
73 generator will create variables to get outputs from the calling function.
74 - ret: This should be assigned the type that the function returns. This
75 directive may be skipped if the function does not return a value.
76 - includes: This should be assigned a comma-separated list of headers that
77 need to be included to provide declarations for the function and types it
78 may need (specifically, this includes using "#include <header>").
79 - include-sources: This should be assigned a comma-separated list of source
80 files that need to be included to provide definitions of global variables
81 and functions (specifically, this includes using "#include "source").
82 See pthread_once-inputs and pthreads_once-source.c for an example of how
83 to use this to benchmark a function that needs state across several calls.
84 - init: Name of an initializer function to call to initialize the benchtest.
85 - name: See following section for instructions on how to use this directive.
87 Lines beginning with a single hash '#' are treated as comments. See
88 pow-inputs for an example of an input file.
90 Multiple execution units per function:
91 =====================================
93 Some functions have distinct performance characteristics for different input
94 domains and it may be necessary to measure those separately. For example, some
95 math functions perform computations at different levels of precision (64-bit vs
96 240-bit vs 768-bit) and mixing them does not give a very useful picture of the
97 performance of these functions. One could separate inputs for these domains in
98 the same file by using the `name' directive that looks something like this:
102 See the pow-inputs file for an example of what such a partitioned input file
105 It is also possible to measure throughput of a (partial) trace extracted from
106 a real workload. In this case the whole trace is iterated over multiple times
107 rather than repeating every input multiple times. This can be done via:
109 ##name: workload-<name>
114 In addition to standard benchmarking of functions, one may also generate
115 custom outputs for a set of functions. This is currently used by string
116 function benchmarks where the aim is to compare performance between
117 implementations at various alignments and for various sizes.
119 To add a benchset for `foo':
121 - Add `foo' to the benchset variable.
122 - Write your bench-foo.c that prints out the measurements to stdout.
123 - On execution, a bench-foo.out is created in $(objpfx) with the contents of
126 Reading String Benchmark Results:
127 ================================
129 Some of the string benchmark results are now in JSON to make it easier to read
130 in scripts. Use the benchtests/compare_strings.py script to show the results
131 in a tabular format, generate graphs and more. Run
133 benchtests/scripts/compare_strings.py -h
135 for usage information.