Sorry for the duplication effort on this bug, before my patch, yongchong had
also submitted a patch to the list.
I personally think my patch will be better, since patch from build system
does not solve the compability problem with gnu and -inline will do really
inline which is not a desired behavior under O0. we can add options with
-INLINE=on, -INLINE:none to work around, but this does not solve the root
cause.
Could gatekeeper on IPA or global gatekeeper help review this? it is a major
severity bug in bugzilla.
Here is his original mail:
Hi
Can a gatekeeper help review this patch, it's a bug fixed for bug#838
When building the gcc front end with opencc -g -O0, there is some
dead functions that call other undefined functions. when building with
gcc or opencc -O1 optimization above, the compiler will eliminated the
dead function. But when building with opencc -O0, dead function would
be kept, this fact will cause the undefined reference error.
This patch will add options -INLINE when build using open64 to build
the front end with DEBUG optimization. With -O0 -INLINE, the open64
can eliminated the dead functions.
Index: configure
===================================================================
--- configure (revision 3737)
+++ configure (working copy)
@@ -1418,6 +1418,9 @@
DEBUG|debug)
BUILD_OPTIMIZE=DEBUG
GCC_CONFIGURE_CFLAGS="-O0 -g -DIs_True_On"
+ if test "${BUILD_COMPILER}" = "OSP"; then
+ GCC_CONFIGURE_CFLAGS="${GCC_CONFIGURE_CFLAGS} -INLINE"
+ fi
;;
*)
{ { echo "$as_me:$LINENO: error: \"BUILD_OPTIMIZE=$BUILD_OPTIMIZE
is not supported\"" >&5
Index: configure.ac
===================================================================
--- configure.ac (revision 3737)
+++ configure.ac (working copy)
@@ -88,6 +88,9 @@
DEBUG|debug)
BUILD_OPTIMIZE=DEBUG
GCC_CONFIGURE_CFLAGS="-O0 -g -DIs_True_On"
+ if test "${BUILD_COMPILER}" = "OSP"; then
+ GCC_CONFIGURE_CFLAGS="${GCC_CONFIGURE_CFLAGS} -INLINE"
+ fi
;;
*)
AC_MSG_ERROR(["BUILD_OPTIMIZE=$BUILD_OPTIMIZE is not supported"])
--
yongchong
Regards
Gang
On Sat, Oct 8, 2011 at 9:26 AM, Gang Yu <yugang...@gmail.com> wrote:
> Hi,
>
> Could a gatekeeper please help review the fix for bug838?
> http://bugs.open64.net/show_bug.cgi?id=838
>
> symptom:
>
> opencc could not compile gcc for debug build(O0), a cutted down case as
> below:
>
> int i;
> static __inline__ void
> should_not_exist (void)
> {
> i=1;
> }
>
> yug@jupiter:~/work/bugs/bug838> $TOOLROOT/bin/opencc -c case1.i -O0
> yug@jupiter:~/work/bugs/bug838> readelf -s case1.o | grep should_not_exist
> 4: 0000000000000000 21 FUNC LOCAL DEFAULT 1 should_not_exist
> yug@jupiter:~/work/bugs/bug838> gcc -c -O0 case1.i
> yug@jupiter:~/work/bugs/bug838> readelf -s case1.o | grep should_not_exist
>
> Since opencc does not do inline on O0(non-ipa) option, so no deletion
> for should_not_exit, this incompatibility causes the break.
>
> Solution:
>
> open inline for O0, but only do dead function elimination
> for unreachable functions with inline attribute.
>
> Patch:
>
> --- a/osprey/driver/special_options.c
> +++ b/osprey/driver/special_options.c
> @@ -360,13 +360,6 @@ add_special_options (void)
> turn_down_opt_level(0, "-g changes optimization to -O0
> since no optimization level is specified");
> }
> - /* Turn off inlining when compiling -O0. We definitly want
> - * this off when compiling with -g -O0, but we don't want
> - * -g to change the generated code so we leave it off always.
> - * See bugs 1917 and 7595.
> - */
> - if (olevel == 0 && inline_t == UNDEFINED)
> - inline_t = FALSE;
> /* In the SGI world, -g3 says to emit crippled debug info for use
> * with optimized code. In the GNU/Pathscale world, -g3 says to
> emit
> Modified osprey/ipa/inline/inline.cxx
> diff --git a/osprey/ipa/inline/inline.cxx b/osprey/ipa/inline/inline.cxx
> index 7a2e8ad..a41e779 100644
> --- a/osprey/ipa/inline/inline.cxx
> +++ b/osprey/ipa/inline/inline.cxx
> @@ -635,6 +635,9 @@ Perform_inlining()
> #endif // _LIGHTWEIGHT_INLINER
> Write_inline_succ_pu();
> } else {
> +#ifdef _LIGHTWEIGHT_INLINER
> + if (Opt_Level !=0) /* O0 does not do real inline. */
> +#endif
> Inline_callees_into_caller(caller);
> #ifdef DEBUG_SYMTAB
> Scope_tab = caller->Scope();
> Modified osprey/ipa/inline/inline_driver.cxx
> diff --git a/osprey/ipa/inline/inline_driver.cxx
> b/osprey/ipa/inline/inline_driver.cxx
> index 77704bc..4aff837 100644
> --- a/osprey/ipa/inline/inline_driver.cxx
> +++ b/osprey/ipa/inline/inline_driver.cxx
> @@ -196,6 +196,10 @@ Process_Command_Line (INT argc, char **argv)
> Opt_Level = 2;
> INLINE_Enable_Split_Common = FALSE;
> break;
> + case '0': /* setup O0 for lw_inline */
> + Opt_Level = 0;
> + INLINE_Enable_Split_Common = FALSE;
> + break;
> default:
> Opt_Level = 1;
> INLINE_Enable_Split_Common = FALSE;
> Modified osprey/ipa/main/analyze/ipa_cg.cxx
> diff --git a/osprey/ipa/main/analyze/ipa_cg.cxx
> b/osprey/ipa/main/analyze/ipa_cg.cxx
> index 5b58f01..4638977 100644
> --- a/osprey/ipa/main/analyze/ipa_cg.cxx
> +++ b/osprey/ipa/main/analyze/ipa_cg.cxx
> @@ -1848,6 +1848,14 @@ Mark_Deletable_Funcs (NODE_INDEX v, DFE_ACTION
> action, mUINT8 *visited)
> node->Set_Undeletable();
> action = MARK_USED;
> visited[v] = VISITED_AND_KEEP;
> +#ifdef _LIGHTWEIGHT_INLINER
> + /* O0 does not delete functions with no inline attrib */
> + } else if (Opt_Level == 0 && ! node->Has_Inline_Attrib()) {
> + node->Clear_Deletable();
> + node->Set_Undeletable();
> + action = MARK_USED;
> + visited[v] = VISITED_AND_KEEP;
> +#endif
> } else if (visited[v] == 0)
> visited[v] = VISITED_BUT_UNDECIDED;
> break;
>
>
> Regards
> Gang
>
------------------------------------------------------------------------------
All the data continuously generated in your IT infrastructure contains a
definitive record of customers, application performance, security
threats, fraudulent activity and more. Splunk takes this data and makes
sense of it. Business sense. IT sense. Common sense.
http://p.sf.net/sfu/splunk-d2d-oct
_______________________________________________
Open64-devel mailing list
Open64-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/open64-devel
