[llvm-commits] [llvm] r144622 - in /llvm/trunk: lib/Target/X86/X86InstrInfo.cpp lib/Target/X86/X86TargetMachine.cpp test/CodeGen/X86/avx-intrinsics-x86.ll test/CodeGen/X86/avx-logic.ll

[Jakob Stoklund Olesen]

On Nov 14, 2011, at 9:55 PM, Craig Topper wrote:

> --- llvm/trunk/lib/Target/X86/X86TargetMachine.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86TargetMachine.cpp Mon Nov 14 23:55:35 2011
> @@ -140,10 +140,7 @@
>       PM.add(createExecutionDependencyFixPass(&X86::VR128RegClass));
>       ShouldPrint = true;
>     }
> -    if (Subtarget.hasAVX2()) {
> -      // FIXME this should be turned on for just AVX, but the pass doesn't check
> -      // that instructions are valid before replacing them and there are AVX2
> -      // integer instructions in the table.
> +    if (Subtarget.hasAVX()) {
>       PM.add(createExecutionDependencyFixPass(&X86::VR256RegClass));
>       ShouldPrint = true;
>     }

Perhaps you could change this back to the simpler (Subtarget.hasSSE2() || Subtarget.hasAVX())?

It doesn't make a difference if you pass the VR128 or VR256 register class. The pass collapses aliases anyway.

/jakob

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20111114/302c6176/attachment.html>