Skip to content

Conversation

@diptorupd
Copy link
Contributor

The PR does the following:

  • Renames the existing DpexTarget to DpexKernelTarget and adds a new DpexTarget.
  • The DepxTarget is a stub for now and will be expanded later.
  • Splits the pipeline for kernels from the auto-offload pipeline.
  • Adds a new descriptor, dispatcher, and a new jit decorator.

Diptorup Deb added 7 commits January 23, 2023 17:41
   - Rename the existing DpexKernel*Context classes to
     DpexKernel*Context and the DpexTarget class to DpexKernelTarget.
     The changes are done to add new context and target for the dpjit
     decorator.
   - Rename target.py to kernel_target.py and move it into a
     new numba_dpex.core.targets sub-module.
   - Introduces two compiler pipelines for numba_dpex, one
     for kernel and another for njit (auo-offload).
   - Kernel compilation from FunctionIR still uses the
     OffloadCompiler. (To be fixed later)
   - Kernels from Python code are now lowered using Numba's
     NativeLowering pass.
@diptorupd diptorupd force-pushed the feature/dpex-njit-decorator branch from 20eff86 to 4ae2a27 Compare January 23, 2023 23:44
Copy link
Contributor

@DrTodd13 DrTodd13 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Lots of code here but as a starting point of the offload compiler it looks fine to me.

@diptorupd diptorupd merged commit c0e5be0 into main Jan 24, 2023
@diptorupd diptorupd deleted the feature/dpex-njit-decorator branch January 24, 2023 03:33
github-actions bot added a commit that referenced this pull request Jan 24, 2023
Initial set of changes to add a new dpjit decorator to support dpnp arrays c0e5be0
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants