[all-commits] [llvm/llvm-project] d8563e: [MLIR][NVVM] Add support for converting fp4/6/8 to...

[Srinivasa Ravi via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: d8563e51b56684e7c5491647424aff55d60017fd
      https://github.com/llvm/llvm-project/commit/d8563e51b56684e7c5491647424aff55d60017fd
  Author: Srinivasa Ravi <srinivasar at nvidia.com>
  Date:   2025-10-23 (Thu, 23 Oct 2025)

  Changed paths:
    M mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td
    M mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp
    M mlir/test/Target/LLVMIR/nvvm/convert_fp4x2.mlir
    M mlir/test/Target/LLVMIR/nvvm/convert_fp6x2.mlir
    M mlir/test/Target/LLVMIR/nvvm/convert_fp8x2.mlir
    M mlir/test/Target/LLVMIR/nvvmir-invalid.mlir

  Log Message:
  -----------
  [MLIR][NVVM] Add support for converting fp4/6/8 to fp16x2 (#162439)

This change adds the following NVVM dialect Ops for converting fp4/6/8
to fp16x2:
- `convert.f4x2.to.f16x2`
- `convert.f6x2.to.f16x2`
- `convert.f8x2.to.f16x2`
- `convert.f8x2.to.bf16x2`

Tests are added in `convert_fp4x2.mlir`, `convert_fp6x2.mlir`, and
`convert_fp8x2.mlir`.

PTX Reference:
https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#data-movement-and-conversion-instructions-cvt



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications