llvm dev - Jan 2010 - [LLVMdev] Cygwin llvm-gcc regression

I am getting an assertion firing while building TOT llvm-gcc on Cygwin in
libgcc2.c :-

/home/ang/build/llvm-gcc/./gcc/xgcc -B/home/ang/build/llvm-gcc/./gcc/
-B/home/an
g/llvm-gcc/i686-pc-cygwin/bin/ -B/home/ang/llvm-gcc/i686-pc-cygwin/lib/
-isystem
 /home/ang/llvm-gcc/i686-pc-cygwin/include -isystem
/home/ang/llvm-gcc/i686-pc-c
ygwin/sys-include  -O2 -I/home/ang/svn/llvm-gcc/gcc/../winsup/w32api/include
-O2
 -g -O2  -DIN_GCC    -W -Wall -Wwrite-strings -Wstrict-prototypes
-Wmissing-prot
otypes -Wold-style-definition  -isystem ./include   -g  -DIN_LIBGCC2
-D__GCC_FLO
AT_NOT_NEEDED  -I. -I. -I/home/ang/svn/llvm-gcc/gcc
-I/home/ang/svn/llvm-gcc/gcc
/. -I/home/ang/svn/llvm-gcc/gcc/../include
-I/home/ang/svn/llvm-gcc/gcc/../libcp
p/include  -I/home/ang/svn/llvm-gcc/gcc/../libdecnumber -I../libdecnumber
-I/hom
e/ang/build/llvm/include -I/home/ang/svn/llvm/include -DL_powixf2 -c
/home/ang/s
vn/llvm-gcc/gcc/libgcc2.c -o libgcc/./_powixf2.o

assertion "(!TYPE_SIZE(Tr) || !Ty->isSized() || !isInt64(TYPE_SIZE(Tr),
true) ||
 getInt64(TYPE_SIZE(Tr), true) =getTargetData().getTypeAllocSizeInBits(Ty))
&&
 "LLVM type size doesn't match GCC type size!"" failed: file
"/home/ang/svn/llvm
-gcc/gcc/llvm-types.cpp", line 83

This assertion has not changed from the past and am not sure what is causing
this.

If someone more familiar with the code could have a look at this please.

Aaron
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20100108/9faffc34/attachment.html>

Hi Aaron,
> assertion "(!TYPE_SIZE(Tr) || !Ty->isSized() ||
!isInt64(TYPE_SIZE(Tr),
> true) ||
>  getInt64(TYPE_SIZE(Tr), true) == 
> getTargetData().getTypeAllocSizeInBits(Ty)) &&
>  "LLVM type size doesn't match GCC type size!"" failed:
file
> "/home/ang/svn/llvm
> -gcc/gcc/llvm-types.cpp", line 83
> 
> This assertion has not changed from the past and am not sure what is 
> causing this.
this means that gcc and llvm disagree about how big a type is.  It is
probably caused by llvm thinking long double has size 12 (or 16), while
cygwin gcc thinks it is 16 (or 12).

Ciao,

Duncan.

2010/1/10 Aaron Gray <aaronngray.lists at googlemail.com>
> 2010/1/10 Duncan Sands <baldrick at free.fr>
>
> Hi Aaron,
>>
>>
>>  Thanks, okay heres the results :-
>>>
>>> LLVM type size doesn't match GCC type size!
>>>
>>>  <real_type 0x7ff80b40 long double sizes-gimplified XF size
<integer_cst
>>> 0x7ff010e0 type <integer_type 0x7ff80060 bit_size_type>
constant invariant
>>> 96>
>>>    unit size <integer_cst 0x7ff01100 type <integer_type
0x7ff80000
>>> unsigned int> constant invariant 12>
>>>    align 32 symtab 0 alias set -1 precision 80
>>>    pointer_to_this <pointer_type 0x7ff80c60>>
>>>
>>
>> as I thought, it's a problem with long double.  GCC thinks it is 12
bytes
>> long, LLVM presumably thinks it is 16 bytes long [in reality long
double
>> is 10 bytes long, but here sizes include alignment, and different
OS's
>> choose different alignments for it].
>>
>> Does the following fix it for you?
>>
>> Index: X86Subtarget.h
>>
==================================================================>> ---
X86Subtarget.h      (revision 93111)
>> +++ X86Subtarget.h      (working copy)
>> @@ -169,7 +169,7 @@
>>       p =
"e-p:64:64-s:64-f64:64:64-i64:64:64-f80:128:128-n8:16:32:64";
>>     else if (isTargetDarwin())
>>       p =
"e-p:32:32-f64:32:64-i64:32:64-f80:128:128-n8:16:32";
>> -    else if (isTargetCygMing() || isTargetWindows())
>> +    else if (isTargetWindows())
>>       p =
"e-p:32:32-f64:64:64-i64:64:64-f80:128:128-n8:16:32";
>>     else
>>       p = "e-p:32:32-f64:32:64-i64:32:64-f80:32:32-n8:16:32";
>>
>
> Yep ! Thanks a lot Duncan, obviously not doing my job properly.
>
> http://llvm.org/viewvc/llvm-project?view=rev&revision=91745
> *
> *
> "Bump alignment requirements for windows targets to achieve
compartibility
> with vcpp. Based on patch by Michael Beck!"
>
> Looks like there is an issue to resolve here. As Cygwin will never be VC++
> compatible its probably not a good idea to change Cygwin's DataLayout
as
> well as Ming's.
>
Now, I am getting another regression in stage 2 configure :-

    "checking whether the C compiler works... configure: error: cannot run
C
compiled programs."

This one is strange, I have had it before but cannot remember what the cause
is as it was along time ago. Its odd as xgcc was used to compile libgcc, and
prev-gcc/xgcc compiles and executes a test c program, yet stage 2's
configure is failing.

Aaron
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20100110/75ee0ef9/attachment.html>

Hi Aaron,
> Now, I am getting another regression in stage 2 configure :-
> 
>     "checking whether the C compiler works... configure: error: cannot
> run C compiled programs."
check config.log to see what happened - probably the compiler crashed.

Ciao,

Duncan.