How do I use the RISC-V Vector (RVV) instructions in LLVM IR?

user_1404847 0
Viewed 1916

In this presentation Kruppe and Espasa give an overview of the RISC-V Vector extension (RVV) and on slide 16 they show LLVM IR samples which use the vector instructions through intrinsic functions, such as:

%vl = call i32 @llvm.riscv.vsetvl(i32 8)

At the time of the talk (April 2019) LLVM support for the V extension was developed out-of-tree at https://github.com/hanna-kruppe/rvv-llvm. However, that repository is archived now and the README file indicates that it is outdated since support for the RISC-V V extension is now developed upstream. I assume that this means that the features are now available from LLVM master at https://github.com/llvm/llvm-project.

However, when I pull the current master and build it and try to compile the sample code with llc (specifying the target with --mtriple=riscv32-unkown-none-rv32imv), I get following error:

error: ../llvm-project/build/bin/llc: test.ll:4:18: error: use of undefined value '@llvm.riscv.vsetvl'

It seems that the V extension is available, since llc -march=riscv32 -mattr=help lists it:

Available features for this target:
...
  experimental-v           - 'V' (Vector Instructions).

Do I have to explicitly enable target features that are marked as experimental? Are these vector intrinsics shown in the slides even present in the upstream version? If yes, how do I use them? If no, how do I then use vector instructions in LLVM IR?

Follow-up: in this post Eli Friedman explains that target-specific intrinsics should be defined in include/llvm/IR/IntrinsicsRISCV.td, and indeed in the archived out-of-tree repository that file contains some vector-specific intrinsics which are not present in the upstream version. If these intrinsics have not been ported upstream, what is then the correct way to use the RISC-V vector instructions?

riscv llvm-ir llvm
2 Answers

There are only two riscv triple defined in llvm/include/llvm/ADT/Triple.h: riscv32 and riscv64.

By default HasStdExtV which is the flag corresponding to experimental-v in initialized to false in llvm/lib/Target/RISCV/RISCVSubtarget.h, if you want to use this extension you need to enable this feature.

Clangs seems also supporting this feature. In clang/lib/Driver/ToolChains/Arch/RISCV.cpp you can see that march support v feature and generate experimental-v for llc.

There is nothing in include/llvm/IR/IntrinsicsRISCV.td about Vector extension , however you can find the description of the instructions from the standard 'V' Vector extension in llvm/lib/Target/RISCV/RISCVInstrInfoV.td. It still experimental. In clang also you can find that even if the v extension is supported, in the default you can see // Currently LLVM supports only "mafdc". in clang/lib/Driver/ToolChains/Arch/RISCV.cpp

user_140685970

I asked on the LLVM developper's mailing list: currently (Nov 2020) there is only MC layer (i.e. assembly level) support for the RISC-V V extension in LLVM (so no intrinsics right now).

However, there is an RFC and patch that adds initial support for vector load, store and integer add intrinsics, as well as the required infrastructure to generate vsetvl[i] instructions.

user_14048470
Related