this problem is too simple to be wrong, makes me think something
really bad has happened. Can someone try the case for a different
target (such as IA64, SL, Mips ...?)
I'd like to know why this fails with a broader code fragment for review.
Mike,
What do you think?
Sun
On Thu, Nov 18, 2010 at 5:41 PM, 朱庆 <zqing1...@gmail.com> wrote:
> Hi, All
>
> Can gatekeeper help review this fix?
>
> Case a.c:
> struct line { char a[0];};
> static struct line line3;
> static struct line line4;
> static const char a[3]={0,1,2};
> int main()
> {
> return 0;
> }
>
> The feature of the case is: at least two global var in .bss section and the
> type size of them is 0, then a var in .rodata or .data.rel.ro.local section.
>
> $opencc -O0 a.c -keep
> a.s: Assembler messages:
> a.s:65: Error: attempt to move .org backwards
>
> $cat a.s
> 19 .text
> 20 .align 2
> 21
> 22 .section .bss, "wa",@nobits
> 23
> 24 .section .rodata, "a",@progbits
> 25 .type line3, @object
> 26 .size line3, 1
> 27 line3: # 0x0
> 28 .skip 16
> 29 .type line4, @object
> 30 .size line4, 1
> 31 line4: # 0x10
> ....
> 63
> 64 .section .rodata
> 65 .org 0x0
> 66 .align 0
> 67 .type a, @object
> 68 .size a, 3
> 69 a: # 0x0
> 70 # offset 0
> 71 .byte 0
> 72 # offset 1
> 73 .byte 1
> 74 # offset 2
> 75 .byte 2
> 76 # end of initialization for a
>
> Apperently, line25-31, should be in section .bss, so there is missing
> section name emit. To fix the problem, there are many methods(cgemit.cxx):
> 1. change all size from 0 to 1 like C++ empty calss
> 8090 size = TY_size(ST_type(sym));
> 8091 // C++ requires empty classes to have unique addresses.
> 8092 if (size == 0 && (pu == NULL || PU_cxx_lang(*pu)/*bug
> 13826*/))
> 8093 size = 1;
> 2. remove the "size!=0" from the branch
> 8122 if (!has_named_section && size != 0) {
> 8123 Change_Section_Origin (base, ofst);
> 8124 if ( !Simd_Reallocate_Objects ) {
> 8125 // If Simd_Reallocate_Objects, we already emitted
> alignment
> 8126 // inside Change_Section_Origin - may be screwed up
> but we did
> 3. add emit for section name:
> patch:
> Index: osprey/be/cg/cgemit.cxx
> ===================================================================
> --- osprey/be/cg/cgemit.cxx (revision 3403)
> +++ osprey/be/cg/cgemit.cxx (working copy)
> @@ -8167,8 +8167,13 @@
> size = TY_size(ST_type(sym));
> #ifdef KEY
> // C++ requires empty classes to have unique addresses.
> - if (size == 0)
> + if (size == 0) {
> + if (base != cur_section) {
> + fprintf ( Asm_File, "\n\t%s %s\n", AS_SECTION,
> ST_name(base));
> + cur_section = base;
> + }
> Print_Label (Asm_File, sym, 1);
> + }
> else
> #endif
> Print_Label (Asm_File, sym, size);
>
> I prefer the last one, are there any other suggestions?
>
>
> Best Regards,
> Zhu Qing
> ------------------------------------------------------------------------------
> Beautiful is writing same markup. Internet Explorer 9 supports
> standards for HTML5, CSS3, SVG 1.1, ECMAScript5, and DOM L2 & L3.
> Spend less time writing and rewriting code and more time creating great
> experiences on the web. Be a part of the beta today
> http://p.sf.net/sfu/msIE9-sfdev2dev
> _______________________________________________
> Open64-devel mailing list
> Open64-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/open64-devel
>
>
------------------------------------------------------------------------------
Beautiful is writing same markup. Internet Explorer 9 supports
standards for HTML5, CSS3, SVG 1.1, ECMAScript5, and DOM L2 & L3.
Spend less time writing and rewriting code and more time creating great
experiences on the web. Be a part of the beta today
http://p.sf.net/sfu/msIE9-sfdev2dev
_______________________________________________
Open64-devel mailing list
Open64-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/open64-devel
