[PATCH] D32871: [LV] Using VPlan to model the vectorized code and drive its transformation

[Ayal Zaks via Phabricator via llvm-commits]

Ayal created this revision.
Herald added a subscriber: mzolotukhin.

This patch contains the new VPlan model and uses it to represent the vectorized code and drive the generation of vectorized IR. It follows the breakdown of  https://reviews.llvm.org/D28975.

The patch contains

- https://reviews.llvm.org/D32200: once that patch is approved and committed, an updated and simplified version of this patch will be uploaded.
- VectorizationPlan.rst: documenting the VPlan model, covering this and follow-up patches.
- Protected methods made public to enable their reuse are annotated as such in-place to simplify the diff. They will eventually be moved and grouped together.

To recap, the introductory patch of VPlan is designed to

- capture in VPlan all current vectorization decisions,
- represent the control-flow of the vectorized loop body using VPlan's Hierarchical CFG,
- retain current vectorizer output.

Gil and Ayal.


https://reviews.llvm.org/D32871

Files:
  docs/VectorizationPlan.rst
  docs/Vectorizers.rst
  lib/Transforms/Vectorize/CMakeLists.txt
  lib/Transforms/Vectorize/LoopVectorize.cpp
  lib/Transforms/Vectorize/VPlan.cpp
  lib/Transforms/Vectorize/VPlan.h
  test/Transforms/LoopVectorize/AArch64/aarch64-predication.ll
  test/Transforms/LoopVectorize/AArch64/predication_costs.ll
  test/Transforms/LoopVectorize/SystemZ/load-store-scalarization-cost.ll
  test/Transforms/LoopVectorize/if-pred-non-void.ll
  test/Transforms/LoopVectorize/induction.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D32871.97836.patch
Type: text/x-patch
Size: 140423 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170504/a19de4b3/attachment-0001.bin>