[llvm-commits] [llvm-gcc-4.2] r55796 - /llvm-gcc-4.2/trunk/gcc/llvm-convert.cpp

[Dale Johannesen]

On Sep 5, 2008, at 1:11 AMPDT, Duncan Sands wrote:

> Hi Dale,
>
>> Use new math intrinsics only when errno is not in use.
>> They are marked as not touching memory.
>
> this is not enough.  Some gcc math builtins may be marked
> readonly, not readnone (= does not touch memory).  It is
> then wrong to promote them to the readnone llvm intrinsics.
> Please replace the errno test with a check something like
> this:
>
> if (flags_from_decl_or_type(exp) & ECF_CONST)
>
> This will automatically handle the errno case.
>
> Thanks,
>
> Duncan.
>
> PS: gcc marks some math functions readonly ("pure") if
> their result depends on the floating point rounding mode.
> This is correct, and it would be wrong to promote to
> readnone in this case.

This doesn't do the right thing for Darwin, and is inconsistent with  
the existing handling of sqrt, which is explicitly tested in the  
testsuite.

The difficulty does seem to be the rounding mode.  I don't agree that  
reading the rounding mode should prevent marking things "readnone";  
the description can be read that way, but it's pretty clear whoever  
wrote it wasn't thinking of rounding mode.  Rounding mode is not  
memory.  There is no reason a libcall that reads rounding mode, but  
doesn't touch memory, can't be moved across loads and stores.  (It  
can't be moved across instructions that change the rounding mode, but  
we don't currently model that, and this consideration is orthogonal to  
touching memory.)