Refactor Examples

[PhilippvK]

Needs rebase after #44 is merged.

Changes:

• Add more example instructions (loads for RV32/RV64, XCV, ADDW,...)
• Run examples for --riscv-xlen 32 and --riscv-xlen 64
• Run examples with/without --no-extend
• Use llvm-lit infrastructure for automated checks
• Parser: allow use of XLEN in casts and definitions
• Unify CoreDSL formatting

[PhilippvK]

@mathis-s I have the feeling that the support for immediates broke in an earlier commit ...

pattern-gen: /work/git/students/mathis/CoreDSL2LLVM/llvm/include/llvm/CodeGen/Register.h:111: llvm::MCRegister llvm::Register::asMCReg() const: Assertion `Reg == MCRegister::NoRegister || MCRegister::isPhysicalRegister(Reg)' failed.

[mathis-s]

@mathis-s I have the feeling that the support for immediates broke in an earlier commit ...

pattern-gen: /work/git/students/mathis/CoreDSL2LLVM/llvm/include/llvm/CodeGen/Register.h:111: llvm::MCRegister llvm::Register::asMCReg() const: Assertion `Reg == MCRegister::NoRegister || MCRegister::isPhysicalRegister(Reg)' failed.

Will take a look!

[PhilippvK]

@mathis-s I have the feeling that the support for immediates broke in an earlier commit ...

pattern-gen: /work/git/students/mathis/CoreDSL2LLVM/llvm/include/llvm/CodeGen/Register.h:111: llvm::MCRegister llvm::Register::asMCReg() const: Assertion `Reg == MCRegister::NoRegister || MCRegister::isPhysicalRegister(Reg)' failed.

Will take a look!

Thanks. Here is the recent ci job: https://github.com/mathis-s/CoreDSL2LLVM/actions/runs/11251833148/job/31283706173?pr=45

[mathis-s]

Should be fixed. Problem was that LLVM had a useless COPY in the MIR. I added some code to peek through that copy.

The CV_ADDN instruction is a really tough in general though because we need an intermediate result larger than XLEN. Pretty nice that it manages to generate a pattern, though I'm not sure it would match.

[PhilippvK]

Should be fixed. Problem was that LLVM had a useless COPY in the MIR. I added some code to peek through that copy.

Thanks, I just pushed a rebase.

The CV_ADDN instruction is a really tough in general though because we need an intermediate result larger than XLEN. Pretty nice that it manages to generate a pattern, though I'm not sure it would match.

That was the reason why I added the test and run lit with and without the --no-extend flag.