------- Additional Comments From pinskia at gcc dot gnu dot org 2005-08-19
11:28 -------
This has been semi fixed on the mainline:
main:
pushl %ebp
movl %esp, %ebp
pushl %esi
pushl %ebx
subl $40000000, %esp
pushl $40000000
call _Znaj
leal -40000008(%ebp), %ebx
pushl $40000000
pushl $0
pushl %ebx
movl %eax, %esi
call memset
pushl $40000000
pushl %ebx
pushl %esi
call memcpy
leal -8(%ebp), %esp
xorl %eax, %eax
popl %ebx
popl %esi
popl %ebp
ret
But in fact this is worse as the reason for the heap allocation is to avoid
the stack usage at all and now
there is 40MB allocated on the stack.
The problem is in the middle-end:
*(int[10000000] *) D.1731 = {};
Note the memcpy is an improvement over what before 4.0.0 had of writting out
element by element.
4.1's problem is a regression in terms of stack usage which should not have
happen.
--
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Component|c++ |middle-end
Ever Confirmed| |1
GCC build triplet|i386-redhat-linux |
GCC host triplet|i386-redhat-linux |
GCC target triplet|i386-redhat-linux |
Keywords| |missed-optimization
Last reconfirmed|0000-00-00 00:00:00 |2005-08-19 11:28:34
date| |
Summary|default-initializing array |[4.1 Regression] default-
|new expression uses memcpy()|initializing array new
|instead of memset(), bloats |expression uses makes stack
|executable |usage go way up
Target Milestone|--- |4.1.0
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23477
