llvm dev - Sep 2014 - [LLVMdev] encoding scaled values

Hello LLVM,
I have a value that is encoded as that value shifted right by 2.  I
can see two ways to do this.
First way using the tblgen definition for this instruction, something like this:

bits<10> Value;
Inst{7-0} = Value{9-2};  // 2-bit effective downshift

Second way with custom EncoderMethod and DecoderMethods specified on
the operand.

Which way is best?

Thanks,
-steve

On further investigation, a tblgen bug prevents the td file solution
from working.  The auto-generated encoding is fine, but the decoder
properly shifts only if the value consists of more than one Inst{}
field.   An easy patch fixes this and now encode/decode work fine
without custom methods.  I'll submit the tblgen patch.

On Thu, Sep 25, 2014 at 11:16 AM, Steve King <steve at metrokings.com>
wrote:
> Hello LLVM,
> I have a value that is encoded as that value shifted right by 2.  I
> can see two ways to do this.
> First way using the tblgen definition for this instruction, something like
this:
>
> bits<10> Value;
> Inst{7-0} = Value{9-2};  // 2-bit effective downshift
>
> Second way with custom EncoderMethod and DecoderMethods specified on
> the operand.
>
> Which way is best?
>
> Thanks,
> -steve