llvm dev - Oct 2015 - backend define a register with multiple types

Hi,

I want to define a register with multiple types for a particular backend. I
checked the RegisterInfo.td for ARM backend, which covers this case. I
tried to do the similar thing but I got errors when I build the backend.

I did this as following:

In the RegisterInfo.td, I define the register class like this:

*def GRRegs : RegisterClass<"FOO", [i32, f32], 32, (add R0, R1, R2,
R3)>;*

In the InstrInfo.td, I define the instruction like this:

*def ADDINT: InstFOO<(outs GRRegs:$dst),*
*                   (ins GRRegs:$src1, GRRegs:$src2),*
*                   "add-int $dst, $src1, $src2",*
*                   [(set i32:$dst, (add i32:$src1, i32:$src2))]>;*

*def ADDFLOAT: InstFOO<(outs GRRegs:$dst),*
*                   (ins GRRegs:$src1, GRRegs:$src2),*
*                   "add-float $dst, $src1, $src2",*
*                   [(set f32:$dst, (add f32:$src1, f32:$src2))]>;*

Then when I build the backend, it gives me this error:
*error: In ADDFLOAT: Type inference contradiction found, merging 'i32'
into
'f32'*

I appreciate if if anyone can point out my errors or give me any
suggestions to make a register with multiple types works?

Regards,

Xiangyang
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20151014/8a179187/attachment.html>

Please ignore my previous email. I find the error is due to a typo in my
code. In ADDFLOAT should use 'fadd' instead of 'add' for pattern
matching.

2015-10-14 0:06 GMT-04:00 Xiangyang Guo <eceguo at gmail.com>:
> Hi,
>
> I want to define a register with multiple types for a particular backend.
> I checked the RegisterInfo.td for ARM backend, which covers this case. I
> tried to do the similar thing but I got errors when I build the backend.
>
> I did this as following:
>
> In the RegisterInfo.td, I define the register class like this:
>
> *def GRRegs : RegisterClass<"FOO", [i32, f32], 32, (add R0,
R1, R2, R3)>;*
>
> In the InstrInfo.td, I define the instruction like this:
>
> *def ADDINT: InstFOO<(outs GRRegs:$dst),*
> *                   (ins GRRegs:$src1, GRRegs:$src2),*
> *                   "add-int $dst, $src1, $src2",*
> *                   [(set i32:$dst, (add i32:$src1, i32:$src2))]>;*
>
> *def ADDFLOAT: InstFOO<(outs GRRegs:$dst),*
> *                   (ins GRRegs:$src1, GRRegs:$src2),*
> *                   "add-float $dst, $src1, $src2",*
> *                   [(set f32:$dst, (add f32:$src1, f32:$src2))]>;*
>
> Then when I build the backend, it gives me this error:
> *error: In ADDFLOAT: Type inference contradiction found, merging
'i32'
> into 'f32'*
>
> I appreciate if if anyone can point out my errors or give me any
> suggestions to make a register with multiple types works?
>
> Regards,
>
> Xiangyang
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20151014/0e2fb913/attachment.html>