[llvm-commits] [llvm] r168998 - in /llvm/trunk: lib/Target/PowerPC/PPCInstrAltivec.td test/CodeGen/PowerPC/buildvec_canonicalize.ll test/CodeGen/PowerPC/vec_mul.ll

NAKAMURA Takumi geek4civic at gmail.com
Fri Nov 30 05:31:52 PST 2012


Fixed in r168999.

me> Don't drop -march, or llc would assume host architecture.

s/host architecture/default target/, excuse me.

...Takumi

2012/11/30 NAKAMURA Takumi <geek4civic at gmail.com>:
> 2012/11/30 Adhemerval Zanella <azanella at linux.vnet.ibm.com>:
>> Modified: llvm/trunk/test/CodeGen/PowerPC/buildvec_canonicalize.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/buildvec_canonicalize.ll?rev=168998&r1=168997&r2=168998&view=diff
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/PowerPC/buildvec_canonicalize.ll (original)
>> +++ llvm/trunk/test/CodeGen/PowerPC/buildvec_canonicalize.ll Fri Nov 30 07:05:44 2012
>> @@ -1,10 +1,4 @@
>> -; There should be exactly one vxor here.
>> -; RUN: llc < %s -march=ppc32 -mcpu=g5 --enable-unsafe-fp-math | \
>> -; RUN:   grep vxor | count 1
>> -
>> -; There should be exactly one vsplti here.
>> -; RUN: llc < %s -march=ppc32 -mcpu=g5 --enable-unsafe-fp-math | \
>> -; RUN:   grep vsplti | count 1
>> +; RUN: llc < %s -mattr=+altivec --enable-unsafe-fp-math | FileCheck %s
>
> Don't drop -march, or llc would assume host architecture.
> -mattr=+altivec doesn't infer ppc.
>
>> Modified: llvm/trunk/test/CodeGen/PowerPC/vec_mul.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/vec_mul.ll?rev=168998&r1=168997&r2=168998&view=diff
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/PowerPC/vec_mul.ll (original)
>> +++ llvm/trunk/test/CodeGen/PowerPC/vec_mul.ll Fri Nov 30 07:05:44 2012
>> @@ -1,5 +1,4 @@
>> -; RUN: llc < %s -march=ppc32 -mcpu=g5 | not grep mullw
>> -; RUN: llc < %s -march=ppc32 -mcpu=g5 | grep vmsumuhm
>> +; RUN: llc < %s -mattr=+altivec | FileCheck %s
>
> Ditto.



More information about the llvm-commits mailing list