[LLVMdev] Error on VSELECT Dagcombiner with some architecture

Duncan Sands baldrick at free.fr
Tue May 28 08:50:11 PDT 2013

Hi JinGu Kang,

On 28/05/13 17:18, jingu kang wrote:
> Hi all,
> I met the error while compiling the code with vector type with some
> architecture. IR is as following.
>    %cmp = icmp sgt <3 x i8> %x, zeroinitializer
>    %sub = sub <3 x i8> zeroinitializer, %x
>    %cond = select <3 x i1> %cmp, <3 x i8> %x, <3 x i8> %sub
> 'select' IR is converted to 'vselect' dag and is combined to 'sra (X,
> size(X)-1); xor (add (X, Y), Y)'. There is a problem here. If the target support
> i8 type as promote integer to make size(X), size(X) has i32 type. and then type
> legalize pass do widening the vector like this.
>    v4i8 = BUILD_VECTOR 0x99d90a8: i32 = Constant<7> [ID=-3], 0x99d90a8: i32 =
> Constant<7> [ID=-3], 0x99d90a8: i32 = Constant<7> [ID=-3], 0x9a6c6c8: i8 = undef

this BUILD_VECTOR is invalid (the last element should be i32 = undef).  Please
open a bugreport with a testcase that allows people to reproduce this.

Best wishes, Duncan.

> In the end, type legalizer generates a error like this "Operands must all have
> the same type"'. I tested this code with ARM and MIPS.
> Target specific dag combiner functions runs after default dag combiner functions
> so I feel it's not possible to support target specific dag combiner function of
> vselect to fix this problem. Should I check the sra? What do you think about
> this? In order to remove this error, I just removed the vselect dag combiner
> function.
> Thanks,
> JinGu Kang
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

More information about the llvm-dev mailing list