commit	372f19e9e27c1333c0fc1e83b53d365051e81612	[log] [tgz]
author	Jake Weinstein <xboxlover360@gmail.com>	Thu Nov 17 16:01:25 2016 -0500
committer	Jake Weinstein <xboxlover360@gmail.com>	Mon Nov 28 19:35:12 2016 +0000
tree	81030a0622b9bf7af53e669bd78bca9dfbad1171
parent	a5c16983a65e2169174a0e77a54c0ce650f87b48 [diff]

libc: ARM64: update memset/strlen/memcpy/memmove to newlib/cortex-strings

* Bionic benchmarks results at the bottom

* This is a squash of the following commits:

libc: ARM64: optimize memset.

 This is an optimized memset for AArch64.  Memset is split into 4 main
 cases: small sets of up to 16 bytes, medium of 16..96 bytes which are
 fully unrolled.  Large memsets of more than 96 bytes align the
 destination and use an unrolled loop processing 64 bytes per
 iteration.  Memsets of zero of more than 256 use the dc zva
 instruction, and there are faster versions for the common ZVA sizes 64
 or 128.  STP of Q registers is used to reduce codesize without loss of
 performance.

Change-Id: I0c5b5ec5ab8a1fd0f23eee8fbacada0be08e841f

libc: ARM64: improve performance in strlen

Change-Id: Ic20f93a0052a49bd76cd6795f51e8606ccfbf11c

libc: ARM64: Optimize memcpy.

 This is an optimized memcpy for AArch64.  Copies are split into 3 main
 cases: small copies of up to 16 bytes, medium copies of 17..96 bytes
 which are fully unrolled.  Large copies of more than 96 bytes align
 the destination and use an unrolled loop processing 64 bytes per
 iteration.  In order to share code with memmove, small and medium
 copies read all data before writing, allowing any kind of overlap.  On
 a random copy test memcpy is 40.8% faster on A57 and 28.4% on A53.

Change-Id: Ibb9483e45bbc0e8ca3d5ce98a31c55dfd8a5ac28

libc: AArch64: Tune memcpy

* Further tuning for performance.

Change-Id: Id08eaab885f9743fa7575077924a947c1b88e4ff

libc: ARM64: optimize memmove for Cortex-A53

* Sadly does not work on Denver or Kryo, so can't go to generic

 This is an optimized memmove for AArch64.  All copies of up to 96
 bytes and all backward copies are done by the new memcpy.  The only
 remaining case is large forward copies which are done in the same way
 as the memcpy loop, but copying from the end rather than the start.

Tested on the Nextbit Robin with MSM8992 (Snapdragon 808):

Before
BM_string_memcmp/8                          1000k         27    0.286 GiB/s
BM_string_memcmp/64                           50M         20    3.053 GiB/s
BM_string_memcmp/512                          20M        126    4.060 GiB/s
BM_string_memcmp/1024                         10M        234    4.372 GiB/s
BM_string_memcmp/8Ki                        1000k       1726    4.745 GiB/s
BM_string_memcmp/16Ki                        500k       3711    4.415 GiB/s
BM_string_memcmp/32Ki                        200k       8276    3.959 GiB/s
BM_string_memcmp/64Ki                        100k      16351    4.008 GiB/s
BM_string_memcpy/8                          1000k         13    0.612 GiB/s
BM_string_memcpy/64                         1000k          8    7.187 GiB/s
BM_string_memcpy/512                          50M         38   13.311 GiB/s
BM_string_memcpy/1024                         20M         86   11.858 GiB/s
BM_string_memcpy/8Ki                           5M        620   13.203 GiB/s
BM_string_memcpy/16Ki                       1000k       1265   12.950 GiB/s
BM_string_memcpy/32Ki                        500k       2977   11.004 GiB/s
BM_string_memcpy/64Ki                        500k       8003    8.188 GiB/s
BM_string_memmove/8                         1000k         11    0.684 GiB/s
BM_string_memmove/64                        1000k         16    3.855 GiB/s
BM_string_memmove/512                         50M         57    8.915 GiB/s
BM_string_memmove/1024                        20M        117    8.720 GiB/s
BM_string_memmove/8Ki                          2M        853    9.594 GiB/s
BM_string_memmove/16Ki                      1000k       1731    9.462 GiB/s
BM_string_memmove/32Ki                       500k       3566    9.189 GiB/s
BM_string_memmove/64Ki                       500k       7708    8.501 GiB/s
BM_string_memset/8                          1000k         16    0.487 GiB/s
BM_string_memset/64                         1000k         16    3.995 GiB/s
BM_string_memset/512                          50M         37   13.489 GiB/s
BM_string_memset/1024                         50M         58   17.405 GiB/s
BM_string_memset/8Ki                           5M        451   18.160 GiB/s
BM_string_memset/16Ki                          2M        883   18.554 GiB/s
BM_string_memset/32Ki                       1000k       2181   15.022 GiB/s
BM_string_memset/64Ki                        500k       4563   14.362 GiB/s
BM_string_strlen/8                          1000k          8    0.965 GiB/s
BM_string_strlen/64                         1000k         16    3.855 GiB/s
BM_string_strlen/512                          20M         92    5.540 GiB/s
BM_string_strlen/1024                         10M        167    6.111 GiB/s
BM_string_strlen/8Ki                        1000k       1237    6.620 GiB/s
BM_string_strlen/16Ki                       1000k       2765    5.923 GiB/s
BM_string_strlen/32Ki                        500k       6135    5.341 GiB/s
BM_string_strlen/64Ki                        200k      13168    4.977 GiB/s

After
BM_string_memcmp/8                          1000k         21    0.369 GiB/s
BM_string_memcmp/64                         1000k         28    2.272 GiB/s
BM_string_memcmp/512                          20M        128    3.983 GiB/s
BM_string_memcmp/1024                         10M        234    4.375 GiB/s
BM_string_memcmp/8Ki                        1000k       1732    4.728 GiB/s
BM_string_memcmp/16Ki                        500k       3485    4.701 GiB/s
BM_string_memcmp/32Ki                        500k       7031    4.660 GiB/s
BM_string_memcmp/64Ki                        200k      14296    4.584 GiB/s
BM_string_memcpy/8                          1000k          5    1.458 GiB/s
BM_string_memcpy/64                         1000k          7    8.952 GiB/s
BM_string_memcpy/512                          50M         36   13.907 GiB/s
BM_string_memcpy/1024                         20M         80   12.750 GiB/s
BM_string_memcpy/8Ki                           5M        572   14.307 GiB/s
BM_string_memcpy/16Ki                       1000k       1165   14.053 GiB/s
BM_string_memcpy/32Ki                        500k       3141   10.430 GiB/s
BM_string_memcpy/64Ki                        500k       7008    9.351 GiB/s
BM_string_memmove/8                           50M          7    1.074 GiB/s
BM_string_memmove/64                        1000k          9    6.593 GiB/s
BM_string_memmove/512                         50M         37   13.502 GiB/s
BM_string_memmove/1024                        20M         80   12.656 GiB/s
BM_string_memmove/8Ki                          5M        573   14.281 GiB/s
BM_string_memmove/16Ki                      1000k       1168   14.018 GiB/s
BM_string_memmove/32Ki                      1000k       2825   11.599 GiB/s
BM_string_memmove/64Ki                       500k       6548   10.008 GiB/s
BM_string_memset/8                          1000k          7    1.038 GiB/s
BM_string_memset/64                         1000k          8    7.151 GiB/s
BM_string_memset/512                        1000k         29   17.272 GiB/s
BM_string_memset/1024                         50M         53   18.969 GiB/s
BM_string_memset/8Ki                           5M        424   19.300 GiB/s
BM_string_memset/16Ki                          2M        846   19.350 GiB/s
BM_string_memset/32Ki                       1000k       2028   16.156 GiB/s
BM_string_memset/64Ki                        500k       4514   14.517 GiB/s
BM_string_strlen/8                          1000k          7    1.120 GiB/s
BM_string_strlen/64                         1000k         16    3.918 GiB/s
BM_string_strlen/512                          50M         64    7.894 GiB/s
BM_string_strlen/1024                         20M        104    9.815 GiB/s
BM_string_strlen/8Ki                           5M        664   12.337 GiB/s
BM_string_strlen/16Ki                       1000k       1291   12.682 GiB/s
BM_string_strlen/32Ki                       1000k       2940   11.143 GiB/s
BM_string_strlen/64Ki                        500k       6440   10.175 GiB/s

Change-Id: I635bd2798a755256f748b2af19b1a56fb85a40c6

6 files changed

tree: 81030a0622b9bf7af53e669bd78bca9dfbad1171

README.md

Working on bionic

What are the big pieces of bionic?

libc/ --- libc.so, libc.a

The C library. Stuff like fopen(3) and kill(2).

libm/ --- libm.so, libm.a

The math library. Traditionally Unix systems kept stuff like sin(3) and cos(3) in a separate library to save space in the days before shared libraries.

libdl/ --- libdl.so

The dynamic linker interface library. This is actually just a bunch of stubs that the dynamic linker replaces with pointers to its own implementation at runtime. This is where stuff like dlopen(3) lives.

libstdc++/ --- libstdc++.so

The C++ ABI support functions. The C++ compiler doesn't know how to implement thread-safe static initialization and the like, so it just calls functions that are supplied by the system. Stuff like __cxa_guard_acquire and __cxa_pure_virtual live here.

linker/ --- /system/bin/linker and /system/bin/linker64

The dynamic linker. When you run a dynamically-linked executable, its ELF file has a DT_INTERP entry that says "use the following program to start me". On Android, that's either linker or linker64 (depending on whether it's a 32-bit or 64-bit executable). It's responsible for loading the ELF executable into memory and resolving references to symbols (so that when your code tries to jump to fopen(3), say, it lands in the right place).

tests/ --- unit tests

The tests/ directory contains unit tests. Roughly arranged as one file per publicly-exported header file.

benchmarks/ --- benchmarks

The benchmarks/ directory contains benchmarks.

What's in libc/?

Adding system calls

Adding a system call usually involves:

Add entries to SYSCALLS.TXT. See SYSCALLS.TXT itself for documentation on the format.
Run the gensyscalls.py script.
Add constants (and perhaps types) to the appropriate header file. Note that you should check to see whether the constants are already in kernel uapi header files, in which case you just need to make sure that the appropriate POSIX header file in libc/include/ includes the relevant file or files.
Add function declarations to the appropriate header file.
Add the function name to the correct section in libc/libc.map.txt and run ./libc/tools/genversion-scripts.py.
Add at least basic tests. Even a test that deliberately supplies an invalid argument helps check that we're generating the right symbol and have the right declaration in the header file, and that you correctly updated the maps in step 5. (You can use strace(1) to confirm that the correct system call is being made.)

Updating kernel header files

As mentioned above, this is currently a two-step process:

Use generate_uapi_headers.sh to go from a Linux source tree to appropriate contents for external/kernel-headers/.
Run update_all.py to scrub those headers and import them into bionic.

Updating tzdata

This is fully automated (and these days handled by the libcore team, because they own icu, and that needs to be updated in sync with bionic):

Run update-tzdata.py in external/icu/tools/.

Verifying changes

If you make a change that is likely to have a wide effect on the tree (such as a libc header change), you should run make checkbuild. A regular make will not build the entire tree; just the minimum number of projects that are required for the device. Tests, additional developer tools, and various other modules will not be built. Note that make checkbuild will not be complete either, as make tests covers a few additional modules, but generally speaking make checkbuild is enough.

Running the tests

The tests are all built from the tests/ directory.

Device tests

$ mma
$ adb remount
$ adb sync
$ adb shell /data/nativetest/bionic-unit-tests/bionic-unit-tests32
$ adb shell \
    /data/nativetest/bionic-unit-tests-static/bionic-unit-tests-static32
# Only for 64-bit targets
$ adb shell /data/nativetest64/bionic-unit-tests/bionic-unit-tests64
$ adb shell \
    /data/nativetest64/bionic-unit-tests-static/bionic-unit-tests-static64

Note that we use our own custom gtest runner that offers a superset of the options documented at https://github.com/google/googletest/blob/master/googletest/docs/AdvancedGuide.md#running-test-programs-advanced-options, in particular for test isolation and parallelism (both on by default).

Host tests

The host tests require that you have lunched either an x86 or x86_64 target.

$ ./tests/run-on-host.sh 32
$ ./tests/run-on-host.sh 64   # For x86_64-bit *targets* only.

You can supply gtest flags as extra arguments to this script.

Against glibc

As a way to check that our tests do in fact test the correct behavior (and not just the behavior we think is correct), it is possible to run the tests against the host's glibc.

$ ./tests/run-on-host.sh glibc

Gathering test coverage

For either host or target coverage, you must first:

$ export NATIVE_COVERAGE=true
- Note that the build system is ignorant to this flag being toggled, i.e. if you change this flag, you will have to manually rebuild bionic.
Set bionic_coverage=true in libc/Android.mk and libm/Android.mk.

Coverage from device tests

$ mma
$ adb sync
$ adb shell \
    GCOV_PREFIX=/data/local/tmp/gcov \
    GCOV_PREFIX_STRIP=`echo $ANDROID_BUILD_TOP | grep -o / | wc -l` \
    /data/nativetest/bionic-unit-tests/bionic-unit-tests32
$ acov

acov will pull all coverage information from the device, push it to the right directories, run lcov, and open the coverage report in your browser.

Coverage from host tests

First, build and run the host tests as usual (see above).

$ croot
$ lcov -c -d $ANDROID_PRODUCT_OUT -o coverage.info
$ genhtml -o covreport coverage.info # or lcov --list coverage.info

The coverage report is now available at covreport/index.html.

Running the benchmarks

Device benchmarks

$ mma
$ adb remount
$ adb sync
$ adb shell /data/nativetest/bionic-benchmarks/bionic-benchmarks
$ adb shell /data/nativetest64/bionic-benchmarks/bionic-benchmarks

You can use --benchmark_filter=getpid to just run benchmarks with "getpid" in their name.

Host benchmarks

See the "Host tests" section of "Running the tests" above.

Attaching GDB to the tests

Bionic's test runner will run each test in its own process by default to prevent tests failures from impacting other tests. This also has the added benefit of running them in parallel, so they are much faster.

However, this also makes it difficult to run the tests under GDB. To prevent each test from being forked, run the tests with the flag --no-isolate.

32-bit ABI bugs

This probably belongs in the NDK documentation rather than here, but these are the known ABI bugs in the 32-bit ABI:

time_t is 32-bit. http://b/5819737. In the 64-bit ABI, time_t is 64-bit.
off_t is 32-bit. There is off64_t, and in newer releases there is almost-complete support for _FILE_OFFSET_BITS. Unfortunately our stdio implementation uses 32-bit offsets and -- worse -- function pointers to functions that use 32-bit offsets, so there's no good way to implement the last few pieces http://b/24807045. In the 64-bit ABI, off_t is off64_t.
sigset_t is too small on ARM and x86 (but correct on MIPS), so support for real-time signals is broken. http://b/5828899 In the 64-bit ABI, sigset_t is the correct size for every architecture.