[Bug tree-optimization/36054] bad code generation with -ftree-vectorize

[ubizjak at gmail dot com]

------- Comment #10 from ubizjak at gmail dot com  2008-04-30 12:31 -------
(In reply to comment #7)
> > -m32?
> 
> Better, but a bunch of 45 errors like below remained.
> 
> % g++  -g -O3 -march=core2 -mfpmath=sse -msse -ftemplate-depth-4096
> -Wnon-virtual-dtor -fPIC kernel_build.ii > log 2>&1
> 
> /usr/lib/gcc/i486-linux-gnu/4.3.1/include/mmintrin.h: In function
> вЂint __vector__ _mm_add_si64(int __vector__, int
> __vector__)’:
> /usr/lib/gcc/i486-linux-gnu/4.3.1/include/mmintrin.h:312: error: cannot 
> convert
> вЂlong long int’ to вЂlong long int
> __vector__’ for argument вЂ1’ to
> вЂlong long int __vector__ __builtin_ia32_paddq(long long int
> __vector__, long long int __vector__)’

Just remove these functions (64bit MMX arithmetic and MMX shift builtins) from
the file. You are compiling with 4.4 that has changed these intrinsic functions
(and relevant builtins). The mismatch is created since this preprocessed source
includes old (4.3.1) headers.

Anyway, I can't check if 4.4 generates correct code since the test isn't
executable. For a runtime problems, executable test is needed, and since the
failure happens in Environment function, only this function should be present
in the executable.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36054