Forum > General
Survey: effect of the new heap.inc on the compiler.
runewalsh:
If you use trunk FPC and have a large enough project that requires several seconds to compile, try the following:
— rebuild (Shift-F9 in Lazarus) your project with FPC @ 2a30ef7acac482f88402f7ebfbc582f6edfb2393 several times, note the minimum time,
— rebuild your project with FPC @ 7e0e284fce59b7e1a0f7638db205bcc9bbb576cf several times, note the minimum time,
— post both values.
My results seem to be:
Martin_fr:
Not a compiler run. But part of some testcases I have. FPC/rtl and project compiled with -O4
From the FpDebug testcase (running both debugger code, and lots of string manipulation/comparison to check results):
Before: 16.140
After: 15.063
From the SynEdit testcase (2 tests)
Before 00:20.526 and 01:44.762
After 00:20.363 and 01:43.696
-- EDIT:
From IdeDebugger test (no significant difference)
Before 02:47.766
After 02:47.112
This test is doing a lot of mem alloc/dealloc. Mostly small objects, strings (xml) and a few arrays
jamie:
I don't worry that much about compile time if the compiler is actually doing some work to make the end product execute better and not get carried away with large in the code.
Better code analyzing normally requires a little more work to make the best choice of generated code.
It's the end results that matters.
Jamie
AlexTP:
CudaText with 'JSON lite lexer', loading 170 Mb JSON file (dummy, 79K lines).
before: ~5.3s
now: ~4.2s
Good work!
runewalsh:
I’m more interested in compilation time than everything else, because various programs can have various shares of the memory manager usage all the way down to 0%, while compiler is a known reference, and can still vary depending on the project.
My FPC compiles itself in 8.9 → 7.1 s.
Navigation
[0] Message Index
[#] Next page