attached file says it all, i suppose stack-variables which are never consumed, should not appear in the assembler output anymore. they are indeed removed when the variable is no array (or an array of size 1), but they're in case of this source file, even though they go out of scope without ever being used. please note that removing the external function call stops this from being triggered. Reading specs from /tmp/inst/lib/gcc-lib/i686-pc-linux-gnu/3.4/specs Configured with: Thread model: posix gcc version 3.4 20021225 (experimental) /tmp/inst/lib/gcc-lib/i686-pc-linux-gnu/3.4/cc1plus -E -D__GNUG__=3 -quiet -v -D__GNUC__=3 -D__GNUC_MINOR__=4 -D__GNUC_PATCHLEVEL__=0 -D_GNU_SOURCE test3.cpp -Wall -W -O3 test3.ii ignoring nonexistent directory "/tmp/inst/i686-pc-linux-gnu/include" #include "..." search starts here: #include <...> search starts here: /tmp/inst/include/c++/3.4 /tmp/inst/include/c++/3.4/i686-pc-linux-gnu /tmp/inst/include/c++/3.4/backward /usr/local/include /tmp/inst/include /tmp/inst/lib/gcc-lib/i686-pc-linux-gnu/3.4/include /usr/include End of search list. /tmp/inst/lib/gcc-lib/i686-pc-linux-gnu/3.4/cc1plus -fpreprocessed test3.ii -quiet -dumpbase test3.cpp -auxbase-strip test3.s -O3 -Wall -W -version -o test3.s GNU C++ version 3.4 20021225 (experimental) (i686-pc-linux-gnu) compiled by GNU C version 3.2 20020903 (Red Hat Linux 8.0 3.2-7). Release: gcc (GCC) 3.4 20021225 (experimental) Environment: intel piii, redhat 8.0 How-To-Repeat: look at the assembler output of the attached source with -O3
State-Changed-From-To: open->analyzed State-Changed-Why: Confirmed on all versions I tested.
Simpler testcase: f() { int a[1024]; } I think this is a duplicate but I cannot find the bug number.
This problem is already fixed on the tree-ssa branch so suspending untill it merges into the mainline.
Weird: The testcase: f() { int a[1024]; } is not optimized but the testcase is: f() { int a[1024]; a[0]=0; }
The problem here is that pass_remove_useless_vars is not being called for the last testcase, the one without any real code.
Ehm, pass_remove_useless_vars _is_ called, but there is no variable annotation for unused variables, so: /* Remove all unused, unaliased temporaries. Also remove unused, unaliased local variables during highly optimizing compilations. */ ann = var_ann (var); if (ann && ! ann->may_aliases && ! ann->used && ! ann->has_hidden_use && ! TREE_ADDRESSABLE (var) we already have (!ann). Something like this should work: + if (!ann) + return false; ! else if (! ann->may_aliases (etc.) I'll make a patch and try if that works.
Steven did your patch work?
Subject: Re: unused arrays not optimized away On Monday 24 May 2004 01:32, pinskia at gcc dot gnu dot org wrote: > ------- Additional Comments From pinskia at gcc dot gnu dot org 2004-05-23 > 23:32 ------- Steven did your patch work? no, but a modified one does. it's not pretty though, so i'm looking for something else now.
Fixed for 3.5.0 by: 2004-07-21 Richard Henderson <rth@redhat.com> * gimple-low.c (expand_var_p): Don't look at TREE_ADDRESSABLE, TREE_THIS_VOLATILE, may_aliases, or optimization level. (remove_useless_vars): Dump debugging info. (expand_used_vars): Move ... * cfgexpand.c (expand_used_vars): ... here. Make static. * tree-flow-inline.h (set_is_used): New. (set_default_def): Use get_var_ann. * tree-flow.h: Update decls. * tree-ssa-live.c (mark_all_vars_used_1, mark_all_vars_used): New. (create_ssa_var_map): Use it. * tree-ssa.c (set_is_used): Remove.