[RISCV] Add OPCFG format of vector

[JojoR]

Summary

Refer to https://github.com/riscv/riscv-v-spec/blob/master/v-spec.adoc#101-vector-arithmetic-instruction-encoding

[JojoR]

@craig.topper @kito-cheng Any suggestion ?

[reames]

Reading through the spec, I only see mention of OPCFG in that table. Is this format actually used for any instruction? If so, which?

[craig.topper]

In D137694#3919832, @reames wrote:

Reading through the spec, I only see mention of OPCFG in that table. Is this format actually used for any instruction? If so, which?

It's the funct3(bits 14:12) encoding for vsetvli, vsetivli, and vsetvl.

[reames]

We should probably update RVInstSetiVLi and friends to use the new definition instead of the raw bit pattern then right?

[craig.topper]

In D137694#3919969, @reames wrote:

We should probably update RVInstSetiVLi and friends to use the new definition instead of the raw bit pattern then right?

Yes. I’ll post a patch

[reames]

LGTM since Craig said he was going to do the follow up.

[craig.topper]

Do you have commit access?

[JojoR]

@craig.topper Thanks for your help :)

llvm/lib/Target/RISCV/RISCVInstrFormatsV.td

 }
 def OPIVV : RISCVVFormat<0b000>;
 def OPFVV : RISCVVFormat<0b001>;
 def OPMVV : RISCVVFormat<0b010>;
 def OPIVI : RISCVVFormat<0b011>;
 def OPIVX : RISCVVFormat<0b100>;
 def OPFVF : RISCVVFormat<0b101>;
 def OPMVX : RISCVVFormat<0b110>;
+def OPCFG : RISCVVFormat<0b111>;
 
 class RISCVMOP<bits<2> val> {
   bits<2> Value = val;
 }
 def MOPLDUnitStride   : RISCVMOP<0b00>;
 def MOPLDIndexedUnord : RISCVMOP<0b01>;
 def MOPLDStrided      : RISCVMOP<0b10>;
 def MOPLDIndexedOrder : RISCVMOP<0b11>;