[llvm-commits] [llvm] r129775 - in /llvm/trunk: lib/Target/ARM/ARM.td lib/Target/ARM/ARMHazardRecognizer.cpp lib/Target/ARM/ARMISelDAGToDAG.cpp lib/Target/ARM/ARMTargetMachine.cpp lib/Target/ARM/MLxExpansionPass.cpp test/CodeGen/ARM/fmacs.ll

Renato Golin rengolin at systemcall.org
Tue Apr 19 11:56:19 PDT 2011


On 19 April 2011 19:11, Bob Wilson <bob.wilson at apple.com> wrote:
> +          // On A9, AGU and NEON/FPU are muxed.
> +          !(STI.isCortexA9() && (LastTID.mayLoad() || LastTID.mayStore())) &&

Hi Bob,

Don't take me wrong, I'm not being pedantic, but can we use less CPU
checks and more feature-support flags?

This could become something like "isFPUMuxed()" or whatever, and
setting this to true by default when building a cortex-a9 target info.

There is a number of isCortexA9(), or if (CPUString == "cortex-a8")
etc, and I'm starting a plan to eradicate them all from the ARM
back-end without polluting the target description too much.

cheers,
--renato




More information about the llvm-commits mailing list