关于gcc:Linaro编译速度

Linaro compilation speed

我正在开发在嵌入式ARM平台上运行的软件。在更新平台的过程中,我们将从一个基于OpenEmbedded的系统切换到Linaro。

在我的机器上,使用OpenEmbedded为我们构建的32位GCC4.6.4为ARM交叉编译我们的软件目前大约需要9分钟。对于新的系统,我们现在当然正在尝试Linaro的GCC4.7二进制文件——结果令人惊讶的是,编译过程突然花费了大约两倍的时间(18分钟)。LinaroGCC4.6二进制文件有相同的问题,因此它不是特定于GCC版本的。

使用Linaro的CrosstoolNg来创建他们的编译器的调整版本(例如,尝试尽可能接近配置选项)并没有加快速度。

我们的旧gcc编译器和linaro编译器的主要区别是:

  • 旧的使用SoftFP,Linaro Hard并指定FPU
  • 旧的目标没有特定的ARM处理器/体系结构(ARM none Linux gnueabi),Linaro的gcc(ARM Linux gnueabihf)明确设置了with-target=armv7-awith-tune=cortex-a9

在gcc中更改配置选项,如启用ssp、thumb/arm模式、使用multilib、目标CPU(cortex-a8 vs a9)并不能带来改进。

对于一个简单的test.cpp来说,性能速度已经有所不同了,它只具有一个vector的主要功能,因此它与链接无关,我怀疑stl头文件造成了如此大的差异。

我没办法再做其他的调整了。有人有主意吗?

edit4:我还尝试了Ubuntu 12.04(GCC版本4.6.3(Ubuntu/Linaro 4.6.3-1ubuntu5))的ARM交叉编译器,它的编译时间与我的4.6.4版本相当。所以在Linaro的版本中似乎有一些特别的不同之处,我要么无法关闭,要么是应用了一些特殊的补丁?

edit3:Linaro GCC 4.7中的-ftime-report用于项目的实际源文件:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
Execution times (seconds)
 phase setup             :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall    2076 kB ( 2%) ggc
 phase parsing           :   1.72 (74%) usr   0.34 (85%) sys   2.04 (75%) wall   66732 kB (79%) ggc
 phase lang. deferred    :   0.28 (12%) usr   0.04 (10%) sys   0.33 (12%) wall   10215 kB (12%) ggc
 phase cgraph            :   0.32 (14%) usr   0.02 ( 5%) sys   0.33 (12%) wall    5481 kB ( 6%) ggc
 phase generate          :   0.60 (26%) usr   0.06 (15%) sys   0.66 (24%) wall   15700 kB (19%) ggc
 |name lookup            :   0.28 (12%) usr   0.02 ( 5%) sys   0.24 ( 9%) wall    8058 kB (10%) ggc
 |overload resolution    :   0.32 (14%) usr   0.06 (15%) sys   0.36 (13%) wall   10042 kB (12%) ggc
 callgraph construction  :   0.06 ( 3%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall     551 kB ( 1%) ggc
 callgraph optimization  :   0.02 ( 1%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall     224 kB ( 0%) ggc
 varpool construction    :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall      94 kB ( 0%) ggc
 df scan insns           :   0.06 ( 3%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall       5 kB ( 0%) ggc
 df reg dead/unused notes:   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall      22 kB ( 0%) ggc
 alias analysis          :   0.00 ( 0%) usr   0.02 ( 5%) sys   0.00 ( 0%) wall      11 kB ( 0%) ggc
 preprocessing           :   0.08 ( 3%) usr   0.10 (25%) sys   0.29 (11%) wall    1069 kB ( 1%) ggc
 parser (global)         :   0.58 (25%) usr   0.08 (20%) sys   0.43 (16%) wall   25145 kB (30%) ggc
 parser struct body      :   0.28 (12%) usr   0.02 ( 5%) sys   0.34 (12%) wall   12400 kB (15%) ggc
 parser enumerator list  :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall     121 kB ( 0%) ggc
 parser function body    :   0.14 ( 6%) usr   0.00 ( 0%) sys   0.15 ( 6%) wall    2435 kB ( 3%) ggc
 parser inl. func. body  :   0.10 ( 4%) usr   0.02 ( 5%) sys   0.18 ( 7%) wall    3682 kB ( 4%) ggc
 parser inl. meth. body  :   0.24 (10%) usr   0.02 ( 5%) sys   0.20 ( 7%) wall    5298 kB ( 6%) ggc
 template instantiation  :   0.58 (25%) usr   0.14 (35%) sys   0.75 (28%) wall   26588 kB (31%) ggc
 tree gimplify           :   0.02 ( 1%) usr   0.00 ( 0%) sys   0.03 ( 1%) wall     785 kB ( 1%) ggc
 tree CFG construction   :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall     543 kB ( 1%) ggc
 tree SSA other          :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall      32 kB ( 0%) ggc
 out of ssa              :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall       0 kB ( 0%) ggc
 expand                  :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall     438 kB ( 1%) ggc
 varconst                :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall       6 kB ( 0%) ggc
 integrated RA           :   0.04 ( 2%) usr   0.00 ( 0%) sys   0.09 ( 3%) wall    1313 kB ( 2%) ggc
 reload                  :   0.06 ( 3%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall      60 kB ( 0%) ggc
 thread pro- & epilogue  :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall      92 kB ( 0%) ggc
 final                   :   0.02 ( 1%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall       4 kB ( 0%) ggc
 rest of compilation     :   0.02 ( 1%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall     133 kB ( 0%) ggc
 unaccounted todo        :   0.02 ( 1%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall       0 kB ( 0%) ggc
 TOTAL                 :   2.32             0.40             2.72              84519 kB

同样适用于我的GCC-4.6:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Execution times (seconds)
 callgraph construction:   0.00 ( 0%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall     527 kB ( 1%) ggc
 trivially dead code   :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 1%) wall       0 kB ( 0%) ggc
 df scan insns         :   0.02 ( 2%) usr   0.00 ( 0%) sys   0.01 ( 1%) wall       5 kB ( 0%) ggc
 df reg dead/unused notes:   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 1%) wall      22 kB ( 0%) ggc
 preprocessing         :   0.08 ( 7%) usr   0.10 (26%) sys   0.14 ( 9%) wall    1016 kB ( 1%) ggc
 parser                :   0.68 (58%) usr   0.24 (63%) sys   0.83 (52%) wall   52215 kB (76%) ggc
 name lookup           :   0.28 (24%) usr   0.02 ( 5%) sys   0.41 (26%) wall   10211 kB (15%) ggc
 inline heuristics     :   0.00 ( 0%) usr   0.02 ( 5%) sys   0.00 ( 0%) wall       0 kB ( 0%) ggc
 tree gimplify         :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 1%) wall     637 kB ( 1%) ggc
 tree CFG construction :   0.02 ( 2%) usr   0.00 ( 0%) sys   0.00 ( 0%) wall     463 kB ( 1%) ggc
 expand                :   0.02 ( 2%) usr   0.00 ( 0%) sys   0.01 ( 1%) wall     426 kB ( 1%) ggc
 varconst              :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall     132 kB ( 0%) ggc
 integrated RA         :   0.02 ( 2%) usr   0.00 ( 0%) sys   0.03 ( 2%) wall     304 kB ( 0%) ggc
 reload                :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 1%) wall      58 kB ( 0%) ggc
 machine dep reorg     :   0.02 ( 2%) usr   0.00 ( 0%) sys   0.00 ( 0%) wall       3 kB ( 0%) ggc
 final                 :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.02 ( 1%) wall       4 kB ( 0%) ggc
 rest of compilation   :   0.02 ( 2%) usr   0.00 ( 0%) sys   0.01 ( 1%) wall     133 kB ( 0%) ggc
 unaccounted todo      :   0.02 ( 2%) usr   0.00 ( 0%) sys   0.03 ( 2%) wall       0 kB ( 0%) ggc
 TOTAL                 :   1.18             0.38             1.59              68315 kB

edit2:linaro GCC 4.6.3的-ftime-report输出,用于非常简单的test.cpp(包括选项-fno-graphite-identity -fno-graphite):

1
2
3
4
5
6
7
8
Execution times (seconds)
 preprocessing         :   0.00 ( 0%) usr   0.02 (50%) sys   0.02 (10%) wall     121 kB ( 2%) ggc
 parser                :   0.10 (62%) usr   0.02 (50%) sys   0.11 (55%) wall    4022 kB (65%) ggc
 name lookup           :   0.02 (12%) usr   0.00 ( 0%) sys   0.04 (20%) wall     879 kB (14%) ggc
 tree gimplify         :   0.02 (13%) usr   0.00 ( 0%) sys   0.00 ( 0%) wall      20 kB ( 0%) ggc
 expand                :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 5%) wall      34 kB ( 1%) ggc
 integrated RA         :   0.02 (12%) usr   0.00 ( 0%) sys   0.01 ( 5%) wall      59 kB ( 1%) ggc
 TOTAL                 :   0.16             0.04             0.20               6207 kB

对于与我的旧GCC4.6.4相同的文件:

1
2
3
4
5
6
7
Execution times (seconds)
 preprocessing         :   0.02 (25%) usr   0.00 ( 0%) sys   0.02 (14%) wall     119 kB ( 2%) ggc
 parser                :   0.00 ( 0%) usr   0.04 (100%) sys   0.06 (43%) wall    4021 kB (65%) ggc
 name lookup           :   0.04 (50%) usr   0.00 ( 0%) sys   0.03 (21%) wall     879 kB (14%) ggc
 expand                :   0.02 (25%) usr   0.00 ( 0%) sys   0.01 ( 7%) wall      34 kB ( 1%) ggc
 unaccounted todo      :   0.00 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 7%) wall       0 kB ( 0%) ggc
 TOTAL                 :   0.08             0.04             0.14               6204 kB

使用这两个编译器生成预处理文件没有产生显著的差异(Linaro的gcc的输出只剩下3行)。

edit1:gcc-v旧的(路径缩短或删除(例如--sbindir)

1
2
3
4
5
6
7
8
# arm-none-linux-gnueabi-g++ -v
Using built-in specs.
COLLECT_GCC=..sysroots/i686-linux/usr/bin/arm-none-linux-gnueabi-g++
COLLECT_LTO_WRAPPER=../libexec/gcc/arm-none-linux-gnueabi/4.6.4/lto-wrapper
Target: arm-none-linux-gnueabi
Configured with: ..tmp/work/armv7a-none-linux-gnueabi/gcc-cross-4.6.3+svnr184847-r27/gcc-4_6-branch/configure --build=i686-linux --host=i686-linux --target=arm-none-linux-gnueabi  --with-gnu-ld --enable-shared --enable-languages=c,c++ --enable-threads=posix --disable-multilib --enable-c99 --enable-long-long --enable-symvers=gnu --enable-libstdcxx-pch --program-prefix=arm-none-linux-gnueabi- --without-local-prefix --enable-lto --enable-libssp --disable-bootstrap --disable-libgomp --disable-libmudflap --with-system-zlib --with-linker-hash-style=gnu --with-ppl=no --with-cloog=no --enable-cheaders=c_global --enable-languages=c,c++,fortran --disable-libunwind-exceptions --with-mpfr=..sysroots/i686-linux/usr --with-system-zlib --enable-__cxa_atexit
Thread model: posix
gcc version 4.6.4 20120303 (prerelease) (GCC)

和Linaro GCC-V

1
2
3
4
5
6
7
8
# arm-linux-gnueabihf-g++ -v
Using built-in specs.
COLLECT_GCC=..compiler/bin/arm-linux-gnueabihf-g++
COLLECT_LTO_WRAPPER=../libexec/gcc/arm-linux-gnueabihf/4.7.2/lto-wrapper
Target: arm-linux-gnueabihf
Configured with: .build/src/gcc-linaro-4.7-2012.08/configure --build=i686-build_pc-linux-gnu --host=i686-build_pc-linux-gnu --target=arm-linux-gnueabihf --enable-languages=c,c++,fortran --enable-multilib --with-arch=armv7-a --with-tune=cortex-a9 --with-fpu=vfpv3-d16 --with-float=hard --with-pkgversion='crosstool-NG linaro-1.13.1-2012.08-20120827 - Linaro GCC 2012.08' --with-bugurl=https://bugs.launchpad.net/gcc-linaro --enable-__cxa_atexit --enable-libmudflap --enable-libgomp --enable-libssp --with-gmp=.. --with-mpfr=.. --with-mpc=.. --with-ppl=.. --with-cloog=.. --with-libelf=.. --with-host-libstdcxx='-L.. -lpwl' --enable-threads=posix --disable-libstdcxx-pch --enable-linker-build-id --enable-gold --with-local-prefix=.. --enable-c99 --enable-long-long --with-mode=thumb
Thread model: posix
gcc version 4.7.2 20120731 (prerelease) (crosstool-NG linaro-1.13.1-2012.08-20120827 - Linaro GCC 2012.08)

对于后者,我还做了一些调整,使--disable mullib--disable libmudflap--disable libgomp--disable mullib。

下面是Ubuntu 12.04的ARM编译器:

1
2
3
4
5
6
> arm-linux-gnueabihf-g++-4.6 -v
Using built-in specs.
COLLECT_GCC=arm-linux-gnueabihf-g++-4.6
COLLECT_LTO_WRAPPER=/usr/lib/gcc/arm-linux-gnueabihf/4.6/lto-wrapper
Target: arm-linux-gnueabihf
Configured with: ../src/configure -v --with-pkgversion='Ubuntu/Linaro 4.6.3-1ubuntu5'  --enable-languages=c,c++,fortran,objc,obj-c++ --prefix=/usr --program-suffix=-4.6 --enable-shared --enable-linker-build-id --with-system-zlib --libexecdir=/usr/lib --without-included-gettext --enable-threads=posix --with-gxx-include-dir=/usr/arm-linux-gnueabihf/include/c++/4.6.3 --libdir=/usr/lib --enable-nls --enable-clocale=gnu --enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-gnu-unique-object --enable-plugin --enable-objc-gc --enable-multilib --disable-sjlj-exceptions --with-arch=armv7-a --with-float=hard --with-fpu=vfpv3-d16 --with-mode=thumb --disable-werror --enable-checking=release --build=x86_64-linux-gnu --host=x86_64-linux-gnu --target=arm-linux-gnueabihf


您可以将-time-ftime-report传递给gcc编译器,以查明gcc占用编译时间的原因和位置。

但是为什么编辑时间对你如此重要呢?

您应该注意生成的可执行二进制文件的执行时间。

同时,向我们展示传递给您的gcc-v选项的输出。

您可以将-j选项传递给您的make命令,让它并行工作(例如,并行运行多个gcc)。您还可以降低优化级别,例如从-O3-O2-O1


好吧,你的-ftime报告测试清楚地表明"解析器"是罪魁祸首;我猜模板(一般)和STL(特别)是根本原因。

建议:

看看是否有任何方法可以在工具链中使用"预编译头"。如果可以的话,这可能会消除整个问题。

链接(很遗憾,我不确定哪一个可能适用于您,也可能不适用于您):

  • 为什么C++编译这么长时间?

  • http://gcc.gnu.org/onlinedocs/gcc-4.0.4/gcc/precompiled-headers.html

  • http://clang.llvm.org/docs/usersmanual.html预编译头

  • 在gcc中,是否可以从其他头文件中包含预编译头文件?

  • http://infocenter.arm.com/help/index.jsp?主题=/com.arm.doc.dui0472c/cihjfbhc.html