[llvm-dev] Sincos for X86_64's GNUX32 and ARM's GNUEABI/GNUEABIHF enviroments
Daniel Sanders via llvm-dev
llvm-dev at lists.llvm.org
Thu Jun 16 06:58:58 PDT 2016
> -----Original Message-----
> From: Renato Golin [mailto:renato.golin at linaro.org]
> Sent: 16 June 2016 09:44
> To: Daniel Sanders
> Cc: llvm-dev at lists.llvm.org
> Subject: Re: [llvm-dev] Sincos for X86_64's GNUX32 and ARM's
> GNUEABI/GNUEABIHF enviroments
> On 15 June 2016 at 11:24, Daniel Sanders via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
> > I've noticed that the GNUX32 and GNUEABI/GNUEABIHF environments
> don't make use of the sincos libcall like the GNU environment does because
> the guarding condition for this is 'TT.getEnvironment() == Triple::GNU'. The
> comment in canCombineSinCosLibcall() in LegalizeDAG.cpp suggests that it's
> intending to use sincos for all GNU environments so I'm wondering whether
> the exclusion of GNUX32/GNUEABI/GNUEABIHF is deliberate or not.
> Hi Daniel,
> I don't remember anything deliberate about that. But I haven't been
> playing around sin/cos much. To me, it looks like an omission more
> than anything.
Thanks. Seeing as you think it's an omission as well, I've posted a patch at http://reviews.llvm.org/D21431.
I've also realised I had the meaning of canCombineSinCosLibcall() backwards. GNUX32/GNUEABI/GNUEABIHF are combining sin+cos into sincos in a situation where GNU considers it to be unsafe.
More information about the llvm-dev