[llvm] [Passes] Add option for LoopVersioningLICM pass. (PR #67107)

[via llvm-commits]

https://github.com/lcvon007 created https://github.com/llvm/llvm-project/pull/67107

- Revert "[DAG] getNode() - remove oneuse limit from (zext (trunc (assertzext x))) -> (assertzext x) fold"
- [Analysis] Use std::clamp (NFC)
- [SCEV] Fix -Wunused-variable in ScalarEvolutionExpander.cpp (NFC)
- [libc++abi][WebAssembly] Support Wasm EH
- [libunwind][WebAssembly] Support Wasm EH
- [SCEVExpander] Clarify absence of no-op casts (NFC)
- [llvm] Use range-based for loops (NFC)
- [clang] Enable descriptions for --print-supported-extensions (#66715)
- [Analysis] Use drop_begin (NFC)
- Revert "[SLP]Use source vector type as the original vector type instead of"
- [Passes] Add option for LoopVersioningLICM pass.


>From 2f48cf7403652799227210bdc432ebed153324be Mon Sep 17 00:00:00 2001
From: laichunfeng <laichunfeng at tencent.com>
Date: Fri, 22 Sep 2023 09:51:31 +0800
Subject: [PATCH] [Passes] Add option for LoopVersioningLICM pass.

This option was deleted before when removing cl::opts
controling optimization pass manager passes in
https://reviews.llvm.org/D137915
---
 llvm/lib/Passes/PassBuilderPipelines.cpp | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
index 5f5d92d277c1bfc..1cad0697927064e 100644
--- a/llvm/lib/Passes/PassBuilderPipelines.cpp
+++ b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -105,6 +105,7 @@
 #include "llvm/Transforms/Scalar/LoopSink.h"
 #include "llvm/Transforms/Scalar/LoopUnrollAndJamPass.h"
 #include "llvm/Transforms/Scalar/LoopUnrollPass.h"
+#include "llvm/Transforms/Scalar/LoopVersioningLICM.h"
 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h"
 #include "llvm/Transforms/Scalar/LowerExpectIntrinsic.h"
 #include "llvm/Transforms/Scalar/LowerMatrixIntrinsics.h"
@@ -271,6 +272,10 @@ static cl::opt<AttributorRunOption> AttributorRun(
                clEnumValN(AttributorRunOption::NONE, "none",
                           "disable attributor runs")));
 
+static cl::opt<bool> UseLoopVersioningLICM(
+    "enable-loop-versioning-licm", cl::init(false), cl::Hidden,
+    cl::desc("Enable the experimental Loop Versioning LICM pass"));
+
 cl::opt<bool> EnableMemProfContextDisambiguation(
     "enable-memprof-context-disambiguation", cl::init(false), cl::Hidden,
     cl::ZeroOrMore, cl::desc("Enable MemProf context disambiguation"));
@@ -1293,6 +1298,21 @@ PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level,
   if (RunPartialInlining)
     MPM.addPass(PartialInlinerPass());
 
+  // Scheduling LoopVersioningLICM when inlining is over, because after that
+  // we may see more accurate aliasing. Reason to run this late is that too
+  // early versioning may prevent further inlining due to increase of code
+  // size. By placing it just after inlining other optimizations which runs
+  // later might get benefit of no-alias assumption in clone loop.
+  if (UseLoopVersioningLICM) {
+    MPM.addPass(createModuleToFunctionPassAdaptor(
+        createFunctionToLoopPassAdaptor(LoopVersioningLICMPass())));
+    MPM.addPass(
+        createModuleToFunctionPassAdaptor(createFunctionToLoopPassAdaptor(
+            LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap,
+                     /*AllowSpeculation=*/true),
+            /*USeMemorySSA=*/true, /*UseBlockFrequencyInfo=*/false)));
+  }
+
   // Remove avail extern fns and globals definitions since we aren't compiling
   // an object file for later LTO. For LTO we want to preserve these so they
   // are eligible for inlining at link-time. Note if they are unreferenced they