Commit Graph

15 Commits

Author SHA1 Message Date
Bill Wendling 353fbd3516 Performance improvement.
Using fwrite and fread was very *very* slow. The resulting code was multiple
times slower than GCC's implementation of gcov. Replace the fwrite/fread system
with an mmap() version.

If the `.gcda' file doesn't exist, we (re)allocate a buffer that we write
into. That gets written to the `.gcda' file in one chunk. If the `.gcda' file
already exists, we simply mmap() the file, modify the mapped data, and use
msync() to write the contents out to disk. It's much easier than implementing
our own buffering scheme, and we don't have to use fwrite's and fread's
buffering.

For those who are numbers-oriented, here are some timings:

GCC Verison
-----------

`.gcda' files don't exist:  23s
`.gcda' files do exist:     14s

LLVM Version (before this change)
---------------------------------

`.gcda' files don't exist:  28s
`.gcda' files do exist:     28s

LLVM Version (with this change)
-------------------------------

`.gcda' files don't exist:  18s
`.gcda' files do exist:      4s

It's a win-win-win-win-lose-win-win scenario!

<rdar://problem/13466086>

llvm-svn: 182563
2013-05-23 07:18:59 +00:00
Bill Wendling c054086630 Try to improve performance by using a read/write buffer instead of I/O.
The calls to fwrite/fread can be very expensive. GCC avoids this by using a
buffer to read and write from the file, thus limiting the number of fwrite/fread
calls.

<rdar://problem/13466086>

llvm-svn: 181924
2013-05-15 21:31:22 +00:00
Bill Wendling e575be4c0e I was wrong in my testing.
There isn't a speedup when using unbuffered I/O. It slows it down in fact.

llvm-svn: 181060
2013-05-03 22:06:41 +00:00
Bill Wendling 6ce98c49f4 Use unbuffered I/O. This reduces the runtime by about half. Our implementation is now only about 5 times slower than gcc's.
llvm-svn: 180980
2013-05-03 00:25:49 +00:00
Bill Wendling 7184d34f96 Add declaration for linux.
llvm-svn: 177601
2013-03-20 23:21:08 +00:00
Bill Wendling 51a6ff5799 Create a coverage initialization function.
This function replaces the call of `atexit' from being generated in the compile
units. Basically, it registers the "writeout" and "flush" functions (if
present). It will generate calls to the `atexit' function for cleanups and final
writeout functions, but only once. This is better than checking for `main',
because a library may not have a `main' function in it.
<rdar://problem/12439551>

llvm-svn: 177578
2013-03-20 21:11:47 +00:00
Richard Smith 6122d0692b Support for building the ubsan runtime when using the autoconf build system on
Darwin. Patch by Jean-Daniel Dupas, tweaked by Jonathan Sauer, simplified
somewhat by me.

llvm-svn: 167889
2012-11-13 23:55:06 +00:00
Benjamin Kramer 7aa5b663ea Update linux stdio.h to support the recent additions to GCDAProfiling.
Hopefully unbreaks compiler-rt build on linux.

llvm-svn: 164025
2012-09-17 10:23:05 +00:00
Bill Wendling 66429fe1c7 Add support for reading the GCDA file and merging the arc information.
With the advent of the __llvm_gcov_flush function, we need to be able to merge
counts into the .gcda files in an intelligent manner. This involves reading the
file if it exists, adding the counts together, and then writing the results.
<rdar://problem/12185886>

llvm-svn: 163923
2012-09-14 18:55:32 +00:00
Alexey Samsonov 80bc3b0d2b Fix atoi stub for linux introduced in r157573. This fixes compiler-rt build on linux.
llvm-svn: 157600
2012-05-29 06:55:01 +00:00
Bill Wendling 1e039681c5 Add stub for 'atoi'.
llvm-svn: 157573
2012-05-28 09:58:27 +00:00
Benjamin Kramer 185f2edd70 Fix the build on 32 bit hosts.
- Add memcpy and memcmp to the stub headers.
- __uint128_t is not available on 32 bit platforms. It's also unused so just
  comment it out for now.

llvm-svn: 153779
2012-03-30 21:37:08 +00:00
Daniel Dunbar 6e09995159 SDK/linux: Fix braindead pasto, caught by Matt Beaumont-Gay.
llvm-svn: 146188
2011-12-08 21:50:03 +00:00
Daniel Dunbar 8142a9f02b SDKs: Sketch an initial stub SDK for Linux, I believe this suffices for building
the main compiler-rt and profile modules, at least on x86.

llvm-svn: 146131
2011-12-08 02:39:23 +00:00
Daniel Dunbar e9da222f2f build/SDKs: Sketch a minimal stub SDK for Darwin.
- Motivation is explained in the README, but basically it is convenient to be
   able to build compiler-rt free standing. Since our external dependencies are
   so small, we can achieve this relatively easily by just stubbing out the
   declarations of the external dependencies.
 - This is in no way, shape, or form intended to be complete, it is just the
   minimal stubs necessary to support the stuff we use.

llvm-svn: 144843
2011-11-16 22:40:57 +00:00