1 llvm-exegesis - LLVM Machine Instruction Benchmark
2 ==================================================
4 .. program:: llvm-exegesis
9 :program:`llvm-exegesis` [*options*]
14 :program:`llvm-exegesis` is a benchmarking tool that uses information available
15 in LLVM to measure host machine instruction characteristics like latency,
16 throughput, or port decomposition.
18 Given an LLVM opcode name and a benchmarking mode, :program:`llvm-exegesis`
19 generates a code snippet that makes execution as serial (resp. as parallel) as
20 possible so that we can measure the latency (resp. inverse throughput/uop decomposition)
22 The code snippet is jitted and executed on the host subtarget. The time taken
23 (resp. resource usage) is measured using hardware performance counters. The
24 result is printed out as YAML to the standard output.
26 The main goal of this tool is to automatically (in)validate the LLVM's TableDef
27 scheduling models. To that end, we also provide analysis of the results.
29 :program:`llvm-exegesis` can also benchmark arbitrary user-provided code
32 EXAMPLE 1: benchmarking instructions
33 ------------------------------------
35 Assume you have an X86-64 machine. To measure the latency of a single
40 $ llvm-exegesis -mode=latency -opcode-name=ADD64rr
42 Measuring the uop decomposition or inverse throughput of an instruction works similarly:
46 $ llvm-exegesis -mode=uops -opcode-name=ADD64rr
47 $ llvm-exegesis -mode=inverse_throughput -opcode-name=ADD64rr
50 The output is a YAML document (the default is to write to stdout, but you can
51 redirect the output to a file using `-benchmarks-file`):
61 llvm_triple: x86_64-unknown-linux-gnu
62 num_repetitions: 10000
64 - { key: latency, value: 1.0058, debug_string: '' }
66 info: 'explicit self cycles, selecting one aliasing configuration.
72 To measure the latency of all instructions for the host architecture, run:
77 readonly INSTRUCTIONS=$(($(grep INSTRUCTION_LIST_END build/lib/Target/X86/X86GenInstrInfo.inc | cut -f2 -d=) - 1))
78 for INSTRUCTION in $(seq 1 ${INSTRUCTIONS});
80 ./build/bin/llvm-exegesis -mode=latency -opcode-index=${INSTRUCTION} | sed -n '/---/,$p'
83 FIXME: Provide an :program:`llvm-exegesis` option to test all instructions.
86 EXAMPLE 2: benchmarking a custom code snippet
87 ---------------------------------------------
89 To measure the latency/uops of a custom piece of code, you can specify the
90 `snippets-file` option (`-` reads from standard input).
94 $ echo "vzeroupper" | llvm-exegesis -mode=uops -snippets-file=-
96 Real-life code snippets typically depend on registers or memory.
97 :program:`llvm-exegesis` checks the liveliness of registers (i.e. any register
98 use has a corresponding def or is a "live in"). If your code depends on the
99 value of some registers, you have two options:
101 - Mark the register as requiring a definition. :program:`llvm-exegesis` will
102 automatically assign a value to the register. This can be done using the
103 directive `LLVM-EXEGESIS-DEFREG <reg name> <hex_value>`, where `<hex_value>`
104 is a bit pattern used to fill `<reg_name>`. If `<hex_value>` is smaller than
105 the register width, it will be sign-extended.
106 - Mark the register as a "live in". :program:`llvm-exegesis` will benchmark
107 using whatever value was in this registers on entry. This can be done using
108 the directive `LLVM-EXEGESIS-LIVEIN <reg name>`.
110 For example, the following code snippet depends on the values of XMM1 (which
111 will be set by the tool) and the memory buffer passed in RDI (live in).
115 # LLVM-EXEGESIS-LIVEIN RDI
116 # LLVM-EXEGESIS-DEFREG XMM1 42
117 vmulps (%rdi), %xmm1, %xmm2
118 vhaddps %xmm2, %xmm2, %xmm3
125 Assuming you have a set of benchmarked instructions (either latency or uops) as
126 YAML in file `/tmp/benchmarks.yaml`, you can analyze the results using the
131 $ llvm-exegesis -mode=analysis \
132 -benchmarks-file=/tmp/benchmarks.yaml \
133 -analysis-clusters-output-file=/tmp/clusters.csv \
134 -analysis-inconsistencies-output-file=/tmp/inconsistencies.html
136 This will group the instructions into clusters with the same performance
137 characteristics. The clusters will be written out to `/tmp/clusters.csv` in the
142 cluster_id,opcode_name,config,sched_class
144 2,ADD32ri8_DB,,WriteALU,1.00
145 2,ADD32ri_DB,,WriteALU,1.01
146 2,ADD32rr,,WriteALU,1.01
147 2,ADD32rr_DB,,WriteALU,1.00
148 2,ADD32rr_REV,,WriteALU,1.00
149 2,ADD64i32,,WriteALU,1.01
150 2,ADD64ri32,,WriteALU,1.01
151 2,MOVSX64rr32,,BSWAP32r_BSWAP64r_MOVSX64rr32,1.00
152 2,VPADDQYrr,,VPADDBYrr_VPADDDYrr_VPADDQYrr_VPADDWYrr_VPSUBBYrr_VPSUBDYrr_VPSUBQYrr_VPSUBWYrr,1.02
153 2,VPSUBQYrr,,VPADDBYrr_VPADDDYrr_VPADDQYrr_VPADDWYrr_VPSUBBYrr_VPSUBDYrr_VPSUBQYrr_VPSUBWYrr,1.01
154 2,ADD64ri8,,WriteALU,1.00
155 2,SETBr,,WriteSETCC,1.01
158 :program:`llvm-exegesis` will also analyze the clusters to point out
159 inconsistencies in the scheduling information. The output is an html file. For
160 example, `/tmp/inconsistencies.html` will contain messages like the following :
162 .. image:: llvm-exegesis-analysis.png
165 Note that the scheduling class names will be resolved only when
166 :program:`llvm-exegesis` is compiled in debug mode, else only the class id will
167 be shown. This does not invalidate any of the analysis results though.
174 Print a summary of command line options.
176 .. option:: -opcode-index=<LLVM opcode index>
178 Specify the opcode to measure, by index. See example 1 for details.
179 Either `opcode-index`, `opcode-name` or `snippets-file` must be set.
181 .. option:: -opcode-name=<opcode name 1>,<opcode name 2>,...
183 Specify the opcode to measure, by name. Several opcodes can be specified as
184 a comma-separated list. See example 1 for details.
185 Either `opcode-index`, `opcode-name` or `snippets-file` must be set.
187 .. option:: -snippets-file=<filename>
189 Specify the custom code snippet to measure. See example 2 for details.
190 Either `opcode-index`, `opcode-name` or `snippets-file` must be set.
192 .. option:: -mode=[latency|uops|inverse_throughput|analysis]
194 Specify the run mode. Note that if you pick `analysis` mode, you also need
195 to specify at least one of the `-analysis-clusters-output-file=` and
196 `-analysis-inconsistencies-output-file=`.
198 .. option:: -num-repetitions=<Number of repetition>
200 Specify the number of repetitions of the asm snippet.
201 Higher values lead to more accurate measurements but lengthen the benchmark.
203 .. option:: -benchmarks-file=</path/to/file>
205 File to read (`analysis` mode) or write (`latency`/`uops`/`inverse_throughput`
206 modes) benchmark results. "-" uses stdin/stdout.
208 .. option:: -analysis-clusters-output-file=</path/to/file>
210 If provided, write the analysis clusters as CSV to this file. "-" prints to
211 stdout. By default, this analysis is not run.
213 .. option:: -analysis-inconsistencies-output-file=</path/to/file>
215 If non-empty, write inconsistencies found during analysis to this file. `-`
216 prints to stdout. By default, this analysis is not run.
218 .. option:: -analysis-clustering=[dbscan,naive]
220 Specify the clustering algorithm to use. By default DBSCAN will be used.
221 Naive clustering algorithm is better for doing further work on the
222 `-analysis-inconsistencies-output-file=` output, it will create one cluster
223 per opcode, and check that the cluster is stable (all points are neighbours).
225 .. option:: -analysis-numpoints=<dbscan numPoints parameter>
227 Specify the numPoints parameters to be used for DBSCAN clustering
228 (`analysis` mode, DBSCAN only).
230 .. option:: -analysis-clustering-epsilon=<dbscan epsilon parameter>
232 Specify the epsilon parameter used for clustering of benchmark points
235 .. option:: -analysis-inconsistency-epsilon=<epsilon>
237 Specify the epsilon parameter used for detection of when the cluster
238 is different from the LLVM schedule profile values (`analysis` mode).
240 .. option:: -analysis-display-unstable-clusters
242 If there is more than one benchmark for an opcode, said benchmarks may end up
243 not being clustered into the same cluster if the measured performance
244 characteristics are different. by default all such opcodes are filtered out.
245 This flag will instead show only such unstable opcodes.
247 .. option:: -ignore-invalid-sched-class=false
249 If set, ignore instructions that do not have a sched class (class idx = 0).
251 .. option:: -mcpu=<cpu name>
253 If set, measure the cpu characteristics using the counters for this CPU. This
254 is useful when creating new sched models (the host CPU is unknown to LLVM).
256 .. option:: --dump-object-to-disk=true
258 By default, llvm-exegesis will dump the generated code to a temporary file to
259 enable code inspection. You may disable it to speed up the execution and save
265 :program:`llvm-exegesis` returns 0 on success. Otherwise, an error message is
266 printed to standard error, and the tool returns a non 0 value.