[llvm-commits] [llvm] r154480 - in /llvm/trunk: lib/Target/ARM/ARM.td lib/Target/ARM/ARMAsmPrinter.cpp lib/Target/ARM/ARMInstrInfo.td lib/Target/ARM/ARMInstrNEON.td lib/Target/ARM/ARMInstrVFP.td lib/Target/ARM/ARMSubtarget.cpp lib/Target/ARM/ARMS

[Evan Cheng]

On Apr 11, 2012, at 4:40 AM, Anton Korobeynikov wrote:

> Hello Evan,
> 
>> Clean up ARM fused multiply + add/sub support some more: rename some isel
>> predicates.
>> Also remove NEON2 since it's not really useful and it is confusing. If
>> NEON + VFP4 implies NEON2 but NEON2 doesn't imply NEON + VFP4, what does it
>> really mean?
> There was the discussion about this in ML, consider checking it.

I only saw this:

> Here is an updated patch for the ARM fused multiply add/sub feature with the
> relevant suggestions incorporated.
> 
> In this update:
> - I assumed neon2 does not imply vfpv4, but neon and vfpv4 imply neon2.


Previously hasNEON2 is both a subtarget feature as well as a isel predicate: 'return HasNEON2 || (HasNEON && HasVFPv4);'.

This is wrong because it would break the assembler and disassembler when you pass -mattr=+vfp4 (because HasNEON2 would still be false). Furthermore, since (HasNEON && HasVFPv4) implies HasNEON2, it means NEON2 is equal or is a subset of the features implied by NEON && HasVFPv4. Since currently there are no instructions that are predicated on HasVFPv4 should not be available for a target with only NEON2, there is really no point in having this subtarget feature.

Also note:

> - Regarding Neon2 flag, Code Sourcery GCC seems to support only these flags:
> -mfpu=vfpv4 		
> -mfpu=vfpv4-d16
> -mfpu=neon-vfpv4 
> -mfpu=vfpv3-d16-fp16 
> -mfpu=vfpv3-fp16

There isn't a -mfpu=neon2 flag.

Evan


> 
> -- 
> With best regards, Anton Korobeynikov
> Faculty of Mathematics and Mechanics, Saint Petersburg State University