This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [cs] Merge in tree-ssa-20020619-merge-20031207


I thought I would include some timing numbers here, since I collected them... Carbon.h is kinda like saying X11.h, but it has more in it, around 82-84k lines total. You can think of the 142x faster number as an upper bound on the speedups obtainable from the technology. In a real application, the number is closer to 2x over PCH for a cold build.

Begin forwarded message:

Below are some benchmarking numbers for:

#include <Carbon.h>

int i;

as the test case. The compilers used were checking -O2 style compilers.
The first number is the time for 5 compiles on one command line, the second number, for one compile on the command line. The below says that for normal C compiles, having 5 on a line is 1.7% faster than 5 separate compiles. For C++, it is 0.5% faster.


For the compile server numbers, the server is started fresh for a 1 unit on the command line compile, and reused for the 5 unit run. Again the times given are for these two runs. For a C server run, it is 6.3x faster when reusing 5 units. For C++, it is 77.9x faster when reusing.

And then we see times for compiling 50 units on a command line. And at the end, we see that for C++, hot compiles are around 142x faster.

	   Normal C compile:
	   9.011/1.833
	   4.91598472449536279323
	   5/4.91598472449536279323
	   1.01709022306070358450 x faster with 5 on same line
	   For a normal C++ compile:
	   13.219/2.658
	   4.97328818660647103085
	   5/4.97328818660647103085
	   1.00537105681216430894 x faster with 5 on same line
	   For a C compile server:
	   1.464/1.857
	   .78836833602584814216
	   5/0.78836833602584814216
	   6.34221311475409836069 x faster with 5 on same line (total reuse)
	   For a C++ compile server:
	   0.169/2.632
	   .06420972644376899696
	   5/0.06420972644376899696
	   77.86982248520710059230 x faster with 5 on same line (total reuse)
	   C++ compile server speedup is
	   12.27802047585817935018 x faster than C compile server speedup.
	   Time for 50 C hot compile server compiles:
	   41.680
	   Time for 50 C++ hot compile server compiles:
	   0.930
	   C++ compile server is
	   41.680/0.930
	   44.81720430107526881720 x faster than C compile server.

	   For C, the server is around
	   9.011/5/(41.680/50)
	   2.16194817658349328214 x faster than a normal compile.
	   For C++, the server is around
	   13.219/5/(0.930/50)
	   142.13978494623655913978 x faster than a normal compile.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]