[llvm] c49b9e0 - [Doc] Proposal for vector predication

[Simon Moll via llvm-commits]

Author: Simon Moll
Date: 2020-02-10T10:35:50+01:00
New Revision: c49b9e0d3284715415cb9f473f6ab2257e778c2b

URL: https://github.com/llvm/llvm-project/commit/c49b9e0d3284715415cb9f473f6ab2257e778c2b
DIFF: https://github.com/llvm/llvm-project/commit/c49b9e0d3284715415cb9f473f6ab2257e778c2b.diff

LOG: [Doc] Proposal for vector predication

Summary:
Proposal and roadmap towards vector predication in LLVM.
This patch documents that
a) It is recognized that current LLVM is ill-equipped for vector predication.
b) The community is working on a solution.
c) A concrete prototype exists in the VP extension (D57504).

Reviewers: rkruppe, rengolin, cameron.mcinally, SjoerdMeijer, andrew.w.kaylor, craig.topper, sdesmalen, k-ishizaka, lattner, fhahn

Reviewed By: andrew.w.kaylor

Subscribers: rogfer01, merge_guards_bot, simoncook, s.egerton, llvm-commits, efocht

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D73889

Added: 
    llvm/docs/Proposals/VectorPredication.rst

Modified: 
    llvm/docs/GettingInvolved.rst

Removed: 
    


################################################################################
diff  --git a/llvm/docs/GettingInvolved.rst b/llvm/docs/GettingInvolved.rst
index 343b226231ad..64ab24f220f5 100644
--- a/llvm/docs/GettingInvolved.rst
+++ b/llvm/docs/GettingInvolved.rst
@@ -182,6 +182,7 @@ can be better.
    Proposals/TestSuite
    Proposals/VariableNames
    Proposals/VectorizationPlan
+   Proposals/VectorPredication
 
 :doc:`CodeOfConduct`
    Proposal to adopt a code of conduct on the LLVM social spaces (lists, events,
@@ -203,4 +204,7 @@ can be better.
    Proposal to change the variable names coding standard.
 
 :doc:`Proposals/VectorizationPlan`
-   Proposal to model the process and upgrade the infrastructure of LLVM's Loop Vectorizer.
\ No newline at end of file
+   Proposal to model the process and upgrade the infrastructure of LLVM's Loop Vectorizer.
+
+:doc:`Proposals/VectorPredication`
+   Proposal for predicated vector instructions in LLVM.

diff  --git a/llvm/docs/Proposals/VectorPredication.rst b/llvm/docs/Proposals/VectorPredication.rst
new file mode 100644
index 000000000000..4e93f4fea31b
--- /dev/null
+++ b/llvm/docs/Proposals/VectorPredication.rst
@@ -0,0 +1,88 @@
+==========================
+Vector Predication Roadmap
+==========================
+
+.. contents:: Table of Contents
+  :depth: 3
+  :local:
+
+Motivation
+==========
+
+This proposal defines a roadmap towards native vector predication in LLVM,
+specifically for vector instructions with a mask and/or an explicit vector
+length.  LLVM currently has no target-independent means to model predicated
+vector instructions for modern SIMD ISAs such as AVX512, ARM SVE, the RISC-V V
+extension and NEC SX-Aurora.  Only some predicated vector operations, such as
+masked loads and stores, are available through intrinsics [MaskedIR]_.
+
+The Vector Predication (VP) extensions is a concrete RFC and prototype
+implementation to achieve native vector predication in LLVM.  The VP prototype
+and all related discussions can be found in the VP patch on Phabricator
+[VPRFC]_.
+
+Roadmap
+=======
+
+1. IR-level VP intrinsics
+-------------------------
+
+- There is a consensus on the semantics/instruction set of VP.
+- VP intrinsics and attributes are available on IR level.
+- TTI has capability flags for VP (``supportsVP()``?,
+  ``haveActiveVectorLength()``?).
+
+Result: VP usable for IR-level vectorizers (LV, VPlan, RegionVectorizer),
+potential integration in Clang with builtins.
+
+2. CodeGen support
+------------------
+
+- VP intrinsics translate to first-class SDNodes
+  (eg  ``llvm.vp.fdiv.* -> vp_fdiv``).
+- VP legalization (legalize explicit vector length to mask (AVX512), legalize VP
+  SDNodes to pre-existing ones (SSE, NEON)).
+
+Result: Backend development based on VP SDNodes.
+
+3. Lift InstSimplify/InstCombine/DAGCombiner to VP
+--------------------------------------------------
+
+- Introduce PredicatedInstruction, PredicatedBinaryOperator, .. helper classes
+  that match standard vector IR and VP intrinsics.
+- Add a matcher context to PatternMatch and context-aware IR Builder APIs.
+- Incrementally lift DAGCombiner to work on VP SDNodes as well as on regular
+  vector instructions.
+- Incrementally lift InstCombine/InstSimplify to operate on VP as well as
+  regular IR instructions.
+
+Result: Optimization of VP intrinsics on par with standard vector instructions.
+
+4. Deprecate llvm.masked.* / llvm.experimental.reduce.*
+-------------------------------------------------------
+
+- Modernize llvm.masked.* / llvm.experimental.reduce* by translating to VP.
+- DCE transitional APIs.
+
+Result: VP has superseded earlier vector intrinsics.
+
+5. Predicated IR Instructions
+-----------------------------
+
+- Vector instructions have an optional mask and vector length parameter. These
+  lower to VP SDNodes (from Stage 2).
+- Phase out VP intrinsics, only keeping those that are not equivalent to
+  vectorized scalar instructions (reduce,  shuffles, ..)
+- InstCombine/InstSimplify expect predication in regular Instructions (Stage (3)
+  has laid the groundwork).
+
+Result: Native vector predication in IR.
+
+References
+==========
+
+.. [MaskedIR] `llvm.masked.*` intrinsics,
+   https://llvm.org/docs/LangRef.html#masked-vector-load-and-store-intrinsics
+
+.. [VPRFC] RFC: Prototype & Roadmap for vector predication in LLVM,
+   https://reviews.llvm.org/D57504