Skip to content

Commit

Permalink
add missing component docs files
Browse files Browse the repository at this point in the history
  • Loading branch information
PhilippvK committed Jul 19, 2022
1 parent bd19ffa commit 1a04856
Show file tree
Hide file tree
Showing 3 changed files with 31 additions and 0 deletions.
27 changes: 27 additions & 0 deletions docs/components/AUTO_VECTORIZE.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,27 @@
# Auto-Vectorize Feature

## Usage

- Add `-f auto_vectorize` to the command line arguments

## TODOs:

- [ ] Configure loop and basic block vectorizer individually.

## Warning
- Auto vectorization is enables by default (if available) on the following optimization levels:

- GCC: `-O2`
- LLVM: `-O1`

## Configuration

- `auto_vectorize.enable`: Allows to turn off ne auto-vectorization completely (Default: `true`)
- `auto_vectorize.verbose`: Print details about auto vectorization possibilities during compililation. Need to check the MLID stdout artifact or enable `mlif.print_outputs` to be effective (Default: `false`)

## Compatibility

- Only RISC-V targets is supported at the momemt
- The supported MLIF toolchains are GCC and LLVM
- A VLEN larger equals 128 is required for this feature
- It seems like this currently needs a ELEN=64 and proper alignment (e.g. `tvmaot.alignment_bytes=8`) for the backend data. TFLMI seems to break with this.
2 changes: 2 additions & 0 deletions docs/components/auto_vectorize.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
.. include:: AUTO_VECTORIZE.md
:parser: myst_parser.sphinx_
2 changes: 2 additions & 0 deletions docs/components/muriscvnnbyoc.rst
Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
.. include:: MURISCVNNBYOC.md
:parser: myst_parser.sphinx_

0 comments on commit 1a04856

Please sign in to comment.