Lacking support for compressed cmv

[PeterRugg]

I think we're missing support for compressed cmv in the RISC-V backend:

llvm-project/llvm/lib/Target/RISCV/RISCVInstrInfoZCheriPurecap.td

Lines 332 to 348 in 2f07835

	//===----------------------------------------------------------------------===//
	// Compress Instruction tablegen backend.
	//===----------------------------------------------------------------------===//
	// Quadrant 0
	let Predicates = [HasZCheriPurecap, HasCheriRVC, HasStdExtC, IsCapMode] in {
	def : CompressPat<(CADDI GPCRC:$rd, CSP:$rs1, uimm10_lsb00nonzero:$imm),
	(C_CIncOffsetImm4CSPN GPCRC:$rd, CSP:$rs1, uimm10_lsb00nonzero:$imm)>;
	} // Predicates = [HasZCheriPurecap, HasCheriRVC, HasStdExtC, IsCapMode]
	// Quadrant 1
	let Predicates = [HasZCheriPurecap, HasCheriRVC, HasStdExtC, IsCapMode] in {
	def : CompressPat<(CADDI C2, C2, simm10_lsb0000nonzero:$imm),
	(C_CIncOffsetImm16CSP C2, simm10_lsb0000nonzero:$imm)>;
	} // Predicates = [HasCheri, HasCheriRVC, HasStdExtC, IsCapMode]
	//===----------------------------------------------------------------------===//

I'm seeing a lot of uncompressed cmvs in the generated code from compiling FreeBSD, which is probably hurting code size a good amount. Some unscientific sampling of random blocks in objdumps looks like it's mostly about 3%, but can be up to 10% code size growth because of it in some cases.

[PeterRugg]

Looks like Alex has a fix: CTSRD-CHERI/llvm-project#771