[llvm-commits] [PATCH] PowerPC: More support for Altivec compare operations

[David Blaikie]

On Wed, Oct 24, 2012 at 11:45 AM, Adhemerval Zanella
<azanella at linux.vnet.ibm.com> wrote:
> I thinker the switches a bit and changed to:
>
> static unsigned int getVCMPInst(MVT::SimpleValueType VecVT, ISD::CondCode CC) {
>   switch (CC) {
>     case ISD::SETEQ:
>     case ISD::SETUEQ:
>     case ISD::SETNE:
>     case ISD::SETUNE:
>       if (VecVT == MVT::v16i8)
>         return PPC::VCMPEQUB;
>       else if (VecVT == MVT::v8i16)
>         return PPC::VCMPEQUH;
>       else if (VecVT == MVT::v4i32)
>         return PPC::VCMPEQUW;
>       // v4f32 != v4f32 could be translate to unordered not equal
>       else if (VecVT == MVT::v4f32)
>         return PPC::VCMPEQFP;
>       break;
>     case ISD::SETLT:
>     case ISD::SETGT:
>     case ISD::SETLE:
>     case ISD::SETGE:
>       if (VecVT == MVT::v16i8)
>         return PPC::VCMPGTSB;
>       else if (VecVT == MVT::v8i16)
>         return PPC::VCMPGTSH;
>       else if (VecVT == MVT::v4i32)
>         return PPC::VCMPGTSW;
>       break;
>     case ISD::SETULT:
>     case ISD::SETUGT:
>     case ISD::SETUGE:
>     case ISD::SETULE:
>       if (VecVT == MVT::v16i8)
>         return PPC::VCMPGTUB;
>       else if (VecVT == MVT::v8i16)
>         return PPC::VCMPGTUH;
>       else if (VecVT == MVT::v4i32)
>         return PPC::VCMPGTUW;
>       break;
>     case ISD::SETOEQ:
>       if (VecVT == MVT::v4f32)
>         return PPC::VCMPEQFP;
>       break;
>     case ISD::SETOLT:
>     case ISD::SETOGT:
>     case ISD::SETOLE:
>       if (VecVT == MVT::v4f32)
>         return PPC::VCMPGTFP;
>       break;
>     case ISD::SETOGE:
>       if (VecVT == MVT::v4f32)
>         return PPC::VCMPGEFP;
>       break;
>     default:
>       llvm_unreachable("Invalid integer vector compare condition");
>   }
>   llvm_unreachable("Invalid integer vector compare condition");
> }
>
>
> The final 'llvm_unreachable' is needed to avoid compiler warning about unused enums.

Bikeshedding: alternatively you could just change the conditions in
the switch to asserts and then you'd unconditionally return from every
case (& unreachable in the default).

> On 10/24/2012 03:46 PM, Villmow, Micah wrote:
>> In getVCMPInst, can you put " llvm_unreachable("Invalid vector type with conditional code");" at the end of the function and just have the switch statements break out instead of duplicating this multiple times.
>> Also can you turn switches with only a single case into a if check.
>>
>>
>> In SelectSetCC, can you factor the common code out of the switch statements.
>>> -----Original Message-----
>>> From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-
>>> bounces at cs.uiuc.edu] On Behalf Of Adhemerval Zanella
>>> Sent: Wednesday, October 24, 2012 7:50 AM
>>> To: llvm-commits at cs.uiuc.edu
>>> Subject: [llvm-commits] [PATCH] PowerPC: More support for Altivec
>>> compare operations
>>>
>>> My previous patch for altivec compare support was incomplete, it didn't
>>> handle the other operators (!=, <, >, etc.). This patch add a better
>>> and more complete support for comparisons for altivec supported types
>>> (v16i8, v8i16, v4i32, and v4f32).
>>>
>>> The testcase also covers all the supported comparison operators for the
>>> altivec types.
>>>
>>> Any tips, suggestions, comments?
>>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits