1Validing libc Assembler Routines 2================================ 3This document describes how to verify incoming assembler libc routines. 4 5## Quick Start 6* First, benchmark the previous version of the routine. 7* Update the routine, run the bionic unit tests to verify the routine doesn't 8have any bugs. See the [Testing](#Testing) section for details about how to 9verify that the routine is being properly tested. 10* Rerun the benchmarks using the updated image that uses the code for 11the new routine. See the [Performance](#Performance) section for details about 12benchmarking. 13* Verify that unwind information for new routine looks correct. See 14the [Unwind Info](#unwind-info) section for details about how to verify this. 15 16When benchmarking, it's best to verify on the latest Pixel device supported. 17Make sure that you benchmark both the big and little cores to verify that 18there is no major difference in performance on each. 19 20Benchmark 64 bit memcmp: 21 22 /data/benchmarktest64/bionic-benchmarks/bionic-benchmarks --bionic_xml=string.xml --benchmark_filter=memcmp 23 24Benchmark 32 bit memcmp: 25 26 /data/benchmarktest/bionic-benchmarks/bionic-benchmarks --bionic_xml=string.xml --benchmark_filter=memcmp 27 28Locking to a specific cpu: 29 30 /data/benchmarktest/bionic-benchmarks/bionic-benchmarks --bionic_cpu=2 --bionic_xml=string.xml --benchmark_filter=memcmp 31 32## Performance 33The bionic benchmarks are used to verify the performance of changes to 34routines. For most routines, there should already be benchmarks available. 35 36Building 37-------- 38The bionic benchmarks are not built by default, they must be built separately 39and pushed on to the device. The commands below show how to do this. 40 41 mmma -j bionic/benchmarks 42 adb sync data 43 44Running 45------- 46There are two bionic benchmarks executables: 47 48 /data/benchmarktest64/bionic-benchmarks/bionic-benchmarks 49 50This is for 64 bit libc routines. 51 52 /data/benchmarktest/bionic-benchmarks/bionic-benchmarks 53 54This is for 32 bit libc routines. 55 56Here is an example of how the benchmark should be executed. For this 57command to work, you need to change directory to one of the above 58directories. 59 60 bionic-benchmarks --bionic_xml=string.xml --benchmark_filter=memcmp 61 62The last argument is the name of the one function that you want to 63benchmark. 64 65Almost all routines are already defined in the **string.xml** file in 66**bionic/benchmarks/suites**. Look at the examples in that file to see 67how to add a benchmark for a function that doesn't already exist. 68 69It can take a long time to run these tests since it attempts to test a 70large number of sizes and alignments. 71 72Results 73------- 74Bionic benchmarks is based on the [Google Benchmarks](https://github.com/google/benchmark) 75library. An example of the output looks like this: 76 77 Run on (8 X 1844 MHz CPU s) 78 CPU Caches: 79 L1 Data 32K (x8) 80 L1 Instruction 32K (x8) 81 L2 Unified 512K (x2) 82 ***WARNING*** CPU scaling is enabled, the benchmark real time measurements may be noisy and will incur extra overhead. 83 ------------------------------------------------------------------------------------------- 84 Benchmark Time CPU Iterations 85 ------------------------------------------------------------------------------------------- 86 BM_string_memcmp/1/0/0 6 ns 6 ns 120776418 164.641MB/s 87 BM_string_memcmp/1/1/1 6 ns 6 ns 120856788 164.651MB/s 88 89The smaller the time, the better the performance. 90 91Caveats 92------- 93When running the benchmarks, CPU scaling is not normally enabled. This means 94that if the device does not get up to the maximum cpu frequency, the results 95can vary wildly. It's possible to lock the cpu to the maximum frequency, but 96is beyond the scope of this document. However, most of the benchmarks max 97out the cpu very quickly on Pixel devices, and don't affect the results. 98 99Another potential issue is that the device can overheat when running the 100benchmarks. To avoid this, you can run the device in a cool environment, 101or choose a device that is less likely to overheat. To detect these kind 102of issues, you can run a subset of the tests again. At the very least, it's 103always a good idea to rerun the suite a couple of times to verify that 104there isn't a high variation in the numbers. 105 106If you want to verify a single benchmark result, you can run a single test 107using a command like this: 108 109 bionic-benchmarks --bionic_xml=string.xml --benchmark_filter=BM_string_memcmp/1/1/0 110 111Where the argument to the filter argument is the name of the benchmark from 112the output. Sometimes this filter can still match multiple benchmarks, to 113guarantee that you only run the single benchmark, you can execute the benchmark 114like so: 115 116 bionic-benchmarks --bionic_xml=string.xml --benchmark_filter=BM_string_memcmp/1/1/0$ 117 118NOTE: It is assumed that these commands are executed in adb as the shell user 119on device. If you are trying to run this using adb directly from a host 120machine, you might need to escape the special shell characters such as **$**. 121 122## Testing 123 124Run the bionic tests to verify that the new routines are valid. However, 125you should verify that there is coverage of the new routines. This is 126especially important if this is the first time a routine is assembler. 127 128Caveats 129------- 130When verifying an assembler routine that operates on buffer data (such as 131memcpy/strcpy), it's important to verify these corner cases: 132 133* Verify the routine does not read past the end of the buffers. Many 134assembler routines optimize by reading multipe bytes at a time and can 135read past the end. This kind of bug results in an infrequent and difficult to 136diagnosis crash. 137* Verify the routine handles unaligned buffers properly. Usually, a failure 138can result in an unaligned exception. 139* Verify the routine handles different sized buffers. 140 141If there are not sufficient tests for a new routine, there are a set of helper 142functions that can be used to verify the above corner cases. See the 143header **bionic/tests/buffer\_tests.h** for these routines and look at 144**bionic/tests/string\_test.cpp** for examples of how to use it. 145 146## Unwind Info 147It is also important to verify that the unwind information for these 148routines are properly set up. Here is a quick checklist of what to check: 149 150* Verify that all labels are of the format .LXXX, where XXX is any valid string 151for a label. If any other label is used, entries in the symbol table 152will be generated that include these labels. In that case, you will get 153an unwind with incorrect function information. 154* Verify that all places where pop/pushes or instructions that modify the 155sp in any way have corresponding cfi information. Along with this item, 156verify that when registers are pushed on the stack that there is cfi 157information indicating how to get the register. 158* Verify that only cfi directives are being used. This only matters for 159arm32, where it's possible to use ARM specific unwind directives. 160 161This list is not meant to be exhaustive, but a minimal set of items to verify 162before submitting a new libc assembler routine. There are difficult 163to verify unwind cases, such as around branches, where unwind information 164can be drastically different for the target of the branch and for the 165code after a branch instruction. 166