This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] require sse3 for fisttp

On Fri, 2006-09-15 at 22:51 -0700, Eric Christopher wrote:
> Let me reassure you that this instruction executed on a non-sse3 
> processor faults. That's what the #UD in the part I quoted means.

So what, the user said -march=nocona -mno-sse3, which means the user
wants the all the features the nocona has minus SSE3 vector extensions.
Note the documention is a joke here really because it just mentions MMX,
SSE, and SSE2 without mentioning SSE3.  Sounds like HJ forgot to add
that when he added PNI or changed the option to SSE3.

Really this is a mess because Intel renamed PNI directly to SSE3, SSE to
me means the vector instructions, fisttp is not a vector instruction at
all and should not be turned off by -mno-sse.

Also if an user supplies -mno-sse or even -mno-mmx, -mno-sse3 is
supplied by default which means they will not get fisttp at all which is
not what the user wanted.  He used want to disable the vector
instructions and nothing more.

-- Pinski

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]