[llvm-commits] [llvm] r38478 - in /llvm/trunk/lib/Target/X86: X86ISelLowering.cpp X86ISelLowering.h X86InstrSSE.td X86RegisterInfo.cpp

[Chris Lattner]

On Jul 10, 2007, at 8:46 AM, Dan Gohman wrote:

>> Nice.  Testcase please :).  What sort of code does this improve?
>
> Ok, I added a testcase for the scalar min and max fix. That improves
> code that uses SSE min or max and an operand is spilled.

Ok, thanks!

> For vector sqrt and friends, we'll need more support in LLVM before
> they can really be useful. I'm working on a patch to add support
> for intrinsic functions with vector argument and result types that
> can be overloaded on vector length.

Nifty.  One thing to be careful of is the use of the rsqrt  
instruction.  Even with the refinement step using them should be  
limited to "fast-math" mode only.

-Chris