Re: [cmake-developers] [CMake] CMake 2.8.12.201401116 is much slower than 2.8.12.1
On Thursday, January 16, 2014 08:41:58 AM Clinton Stimpson wrote: On Thursday, January 16, 2014 10:14:32 AM Bill Hoffman wrote: Moving this to cmake-developers list: On 1/16/2014 9:55 AM, Clinton Stimpson wrote: On one of my projects, I see ~3.0 seconds for CMake 2.8.12 and ~3.8 seconds for CMake master. I don't have an idea why, but with my test, it doesn't see like much slower. 3.8 from 3.0 is much slower in my book :) change it to minutes... 3 minutes vs 3.8 minutes is a big jump... Looking at the test timing on the bootstrap test on a slow machine, this seems to be a point that got slower: http://open.cdash.org/testDetails.php?test=229000913build=3172685 885 seconds http://open.cdash.org/testDetails.php?test=229192425build=3173776 902 seconds -Bill For this one project I got 3.0 seconds - 3.8 seconds. I narrowed it down to this commit: commit 2509c7678feb2efab79a384fd0b0b35cee8d02b0 Author: Stephen Kelly steve...@gmail.com Date: Mon Jan 13 13:04:03 2014 +0100 Qt4: Use generator expression in COMPILE_DEFINITIONS (#14692) I'm surprised how this small change added a 20% slowdown. -set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS_DEBUG QT_DEBUG) +set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS $$NOT: $CONFIG:Debug:QT_NO_DEBUG) I tried this project on Windows with CMake master (my own build) and CMake (kitware build) and CMake master was slightly faster, then modifying UseQt.cmake made it slightly faster still. Dominik, if you are seeing a massive slowdown, can you narrow it down? It may not be generator expression related at all. -Clint -- Powered by www.kitware.com Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ Follow this link to subscribe/unsubscribe: http://public.kitware.com/cgi-bin/mailman/listinfo/cmake-developers
Re: [cmake-developers] [CMake] CMake 2.8.12.201401116 is much slower than 2.8.12.1
Moving this to cmake-developers list: On 1/16/2014 9:55 AM, Clinton Stimpson wrote: On one of my projects, I see ~3.0 seconds for CMake 2.8.12 and ~3.8 seconds for CMake master. I don't have an idea why, but with my test, it doesn't see like much slower. 3.8 from 3.0 is much slower in my book :) change it to minutes... 3 minutes vs 3.8 minutes is a big jump... Looking at the test timing on the bootstrap test on a slow machine, this seems to be a point that got slower: http://open.cdash.org/testDetails.php?test=229000913build=3172685 885 seconds http://open.cdash.org/testDetails.php?test=229192425build=3173776 902 seconds -Bill -- Powered by www.kitware.com Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ Follow this link to subscribe/unsubscribe: http://public.kitware.com/cgi-bin/mailman/listinfo/cmake-developers
Re: [cmake-developers] [CMake] CMake 2.8.12.201401116 is much slower than 2.8.12.1
On Thursday, January 16, 2014 10:14:32 AM Bill Hoffman wrote: Moving this to cmake-developers list: On 1/16/2014 9:55 AM, Clinton Stimpson wrote: On one of my projects, I see ~3.0 seconds for CMake 2.8.12 and ~3.8 seconds for CMake master. I don't have an idea why, but with my test, it doesn't see like much slower. 3.8 from 3.0 is much slower in my book :) change it to minutes... 3 minutes vs 3.8 minutes is a big jump... Looking at the test timing on the bootstrap test on a slow machine, this seems to be a point that got slower: http://open.cdash.org/testDetails.php?test=229000913build=3172685 885 seconds http://open.cdash.org/testDetails.php?test=229192425build=3173776 902 seconds -Bill For this one project I got 3.0 seconds - 3.8 seconds. I narrowed it down to this commit: commit 2509c7678feb2efab79a384fd0b0b35cee8d02b0 Author: Stephen Kelly steve...@gmail.com Date: Mon Jan 13 13:04:03 2014 +0100 Qt4: Use generator expression in COMPILE_DEFINITIONS (#14692) I'm surprised how this small change added a 20% slowdown. -set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS_DEBUG QT_DEBUG) +set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS $$NOT: $CONFIG:Debug:QT_NO_DEBUG) -- Clinton Stimpson Elemental Technologies, Inc Computational Simulation Software, LLC www.csimsoft.com -- Powered by www.kitware.com Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ Follow this link to subscribe/unsubscribe: http://public.kitware.com/cgi-bin/mailman/listinfo/cmake-developers
Re: [cmake-developers] [CMake] CMake 2.8.12.201401116 is much slower than 2.8.12.1
On Thu, Jan 16, 2014 at 08:41:58 -0700, Clinton Stimpson wrote: I'm surprised how this small change added a 20% slowdown. -set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS_DEBUG QT_DEBUG) +set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS $$NOT: $CONFIG:Debug:QT_NO_DEBUG) There's a lot of places where std::string gets passed as a const char* only to be turned back into a std::string in the function. Would it be worth me trying to resurrect that branch again? It wasn't worth much last I tried, but that was before a lot of the generator expression stuff was added. I guess another question is whether the genex is evaluated once or per-target (I'd guess the latter which might be the reason especially if Qt4 is found at the top-level of a larger project). --Ben -- Powered by www.kitware.com Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ Follow this link to subscribe/unsubscribe: http://public.kitware.com/cgi-bin/mailman/listinfo/cmake-developers
Re: [cmake-developers] [CMake] CMake 2.8.12.201401116 is much slower than 2.8.12.1
On Thursday, January 16, 2014 11:49:58 AM Ben Boeckel wrote: On Thu, Jan 16, 2014 at 08:41:58 -0700, Clinton Stimpson wrote: I'm surprised how this small change added a 20% slowdown. -set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS_DEBUG QT_DEBUG) +set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS $$NOT: $CONFIG:Debug:QT_NO_DEBUG) There's a lot of places where std::string gets passed as a const char* only to be turned back into a std::string in the function. Would it be worth me trying to resurrect that branch again? It wasn't worth much last I tried, but that was before a lot of the generator expression stuff was added. I guess another question is whether the genex is evaluated once or per-target (I'd guess the latter which might be the reason especially if Qt4 is found at the top-level of a larger project). From a bit of profiling, std::string construction did take a sizable amount of time, but its not clear how much of that was redundant. However, the profile of my test case showed that the time increase is in the parser (yylex). Perhaps someone else can look further into this. Clint -- Powered by www.kitware.com Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ Follow this link to subscribe/unsubscribe: http://public.kitware.com/cgi-bin/mailman/listinfo/cmake-developers
Re: [cmake-developers] [CMake] CMake 2.8.12.201401116 is much slower than 2.8.12.1
On 01/16/2014 02:18 PM, Clinton Stimpson wrote:
On Thursday, January 16, 2014 11:49:58 AM Ben Boeckel wrote:
On Thu, Jan 16, 2014 at 08:41:58 -0700, Clinton Stimpson wrote:
I'm surprised how this small change added a 20% slowdown.
-set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS_DEBUG
QT_DEBUG)
+set_property(DIRECTORY APPEND PROPERTY COMPILE_DEFINITIONS $$NOT:
$CONFIG:Debug:QT_NO_DEBUG)
[snip]
However, the profile of my test case showed that the time increase is in the
parser (yylex). Perhaps someone else can look further into this.
The addition of '$' to the value to causes the fast-path in
cmMakefile::ExpandVariablesInString to be skipped:
http://cmake.org/gitweb?p=cmake.git;a=blob;f=Source/cmMakefile.cxx;hb=v2.8.12.1#l2535
if ( source.empty() || source.find_first_of($@\\) == source.npos)
{
return source.c_str();
}
Instead full expansion using cmCommandArgumentParserHelper occurs
which is much slower. The latter code path is ripe for optimization
by re-implementing it without flex/bison generators. Unfortunately
the current logic has so many idiosyncrasies that a re-implementation
will almost certainly introduce subtle behavior changes.
-Brad
--
Powered by www.kitware.com
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Please keep messages on-topic and check the CMake FAQ at:
http://www.cmake.org/Wiki/CMake_FAQ
Follow this link to subscribe/unsubscribe:
http://public.kitware.com/cgi-bin/mailman/listinfo/cmake-developers
