[PM] Port ArgumentPromotion to the new pass manager.

[chandlerc]

Summary

Now that the call graph supports efficient replacement of a function and
spurious reference edges, we can port ArgumentPromotion to the new pass
manager very easily.

The old PM-specific bits are sunk into callbacks that the new PM simply
doesn't use. Unlike the old PM, the new PM simply does argument
promotion and afterward does the update to LCG reflecting the promoted
function.

Depends on D29579.

[davide]

Still I would really like these tests to use FileCheck instead of grep and check more, but I don't think it makes sense holding a review just for that. I would like to see the output to be checked for some tests (inlined), other than that I'm fine with this patch. Thanks!

test/Transforms/ArgumentPromotion/callgraph-update.ll

1–2 ↗

(On Diff #87205)

Can you check the output is the same here?

test/Transforms/ArgumentPromotion/crash.ll

1–2 ↗

(On Diff #87205)

Can you check the output is the same here?

test/Transforms/ArgumentPromotion/reserve-tbaa.ll

1–2 ↗

(On Diff #87205)

Can you check the output is the same here? This old test doesn't even check for it.

[chandlerc]

In D29580#669635, @davide wrote:

Still I would really like these tests to use FileCheck instead of grep and check more, but I don't think it makes sense holding a review just for that. I would like to see the output to be checked for some tests (inlined), other than that I'm fine with this patch. Thanks!

Sorry, I tried to do this already and just missed some cases. I've addressed all of them now. One test I just deleted because we actually didn't argument promote anything there, but the rest are now FileCheck based. =]

[chandlerc]

Rebase now that we actually don't have grep testing here.

[chandlerc]

Confirmed with Davide that this LGTM now on IRC, landing.

llvm/trunk/include/llvm/Transforms/IPO/ArgumentPromotion.h

+//===- ArgumentPromotion.h - Promote by-reference arguments -----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TRANSFORMS_IPO_ARGUMENTPROMOTION_H
+#define LLVM_TRANSFORMS_IPO_ARGUMENTPROMOTION_H
+
+#include "llvm/Analysis/CGSCCPassManager.h"
+#include "llvm/Analysis/LazyCallGraph.h"
+
+namespace llvm {
+
+/// Argument promotion pass.
+///
+/// This pass walks the functions in each SCC and for each one tries to
+/// transform it and all of its callers to replace indirect arguments with
+/// direct (by-value) arguments.
+class ArgumentPromotionPass : public PassInfoMixin<ArgumentPromotionPass> {
+public:
+  PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &AM,
+                        LazyCallGraph &CG, CGSCCUpdateResult &UR);
+};
+
+}
+
+#endif

llvm/trunk/lib/Passes/PassBuilder.cpp

 #include "llvm/IR/IRPrintingPasses.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/IR/Verifier.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/Regex.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Transforms/GCOVProfiler.h"
 #include "llvm/Transforms/IPO/AlwaysInliner.h"
+#include "llvm/Transforms/IPO/ArgumentPromotion.h"
 #include "llvm/Transforms/IPO/ConstantMerge.h"
 #include "llvm/Transforms/IPO/CrossDSOCFI.h"
 #include "llvm/Transforms/IPO/DeadArgumentElimination.h"
 #include "llvm/Transforms/IPO/ElimAvailExtern.h"
 #include "llvm/Transforms/IPO/ForceFunctionAttrs.h"
 #include "llvm/Transforms/IPO/FunctionAttrs.h"
 #include "llvm/Transforms/IPO/FunctionImport.h"
 #include "llvm/Transforms/IPO/GlobalDCE.h"

llvm/trunk/lib/Passes/PassRegistry.def

 #endif
 CGSCC_ANALYSIS("no-op-cgscc", NoOpCGSCCAnalysis())
 CGSCC_ANALYSIS("fam-proxy", FunctionAnalysisManagerCGSCCProxy())
 #undef CGSCC_ANALYSIS
 
 #ifndef CGSCC_PASS
 #define CGSCC_PASS(NAME, CREATE_PASS)
 #endif
+CGSCC_PASS("argpromotion", ArgumentPromotionPass())
 CGSCC_PASS("invalidate<all>", InvalidateAllAnalysesPass())
 CGSCC_PASS("function-attrs", PostOrderFunctionAttrsPass())
 CGSCC_PASS("inline", InlinerPass())
 CGSCC_PASS("no-op-cgscc", NoOpCGSCCPass())
 #undef CGSCC_PASS
 
 #ifndef FUNCTION_ANALYSIS
 #define FUNCTION_ANALYSIS(NAME, CREATE_PASS)

llvm/trunk/lib/Transforms/IPO/ArgumentPromotion.cpp

 //
 // Note that this transformation could also be done for arguments that are only
 // stored to (returning the value instead), but does not currently.  This case
 // would be best handled when and if LLVM begins supporting multiple return
 // values from functions.
 //
 //===----------------------------------------------------------------------===//
 
+#include "llvm/Transforms/IPO/ArgumentPromotion.h"
 #include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/ADT/Optional.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Analysis/AssumptionCache.h"
 #include "llvm/Analysis/BasicAliasAnalysis.h"
 #include "llvm/Analysis/CallGraph.h"
 #include "llvm/Analysis/CallGraphSCCPass.h"
+#include "llvm/Analysis/LazyCallGraph.h"
 #include "llvm/Analysis/Loads.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/IR/CFG.h"
 #include "llvm/IR/CallSite.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/DebugInfo.h"
 #include "llvm/IR/DerivedTypes.h"

 STATISTIC(NumArgumentsDead, "Number of dead pointer args eliminated");
 
 /// A vector used to hold the indices of a single GEP instruction
 typedef std::vector<uint64_t> IndicesVector;
 
 /// DoPromotion - This method actually performs the promotion of the specified
 /// arguments, and returns the new function.  At this point, we know that it's
 /// safe to do so.
-static CallGraphNode *
+static Function *
 doPromotion(Function *F, SmallPtrSetImpl<Argument *> &ArgsToPromote,
-            SmallPtrSetImpl<Argument *> &ByValArgsToTransform, CallGraph &CG) {
+            SmallPtrSetImpl<Argument *> &ByValArgsToTransform,
+            Optional<function_ref<void(CallSite OldCS, CallSite NewCS)>>
+                ReplaceCallSite) {
 
   // Start by computing a new prototype for the function, which is the same as
   // the old function, but has modified arguments.
   FunctionType *FTy = F->getFunctionType();
   std::vector<Type *> Params;
 
   typedef std::set<std::pair<Type *, IndicesVector>> ScalarizeTable;
 
doPromotion(Function *F, SmallPtrSetImpl<Argument *> &ArgsToPromote,
   // Recompute the parameter attributes list based on the new arguments for
   // the function.
   NF->setAttributes(AttributeSet::get(F->getContext(), AttributesVec));
   AttributesVec.clear();
 
   F->getParent()->getFunctionList().insert(F->getIterator(), NF);
   NF->takeName(F);
 
-  // Get a new callgraph node for NF.
-  CallGraphNode *NF_CGN = CG.getOrInsertFunction(NF);
-
   // Loop over all of the callers of the function, transforming the call sites
   // to pass in the loaded pointers.
   //
   SmallVector<Value *, 16> Args;
   while (!F->use_empty()) {
     CallSite CS(F->user_back());
     assert(CS.getCalledFunction() == F);
     Instruction *Call = CS.getInstruction();
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
       cast<CallInst>(New)->setTailCallKind(
           cast<CallInst>(Call)->getTailCallKind());
     }
     New->setDebugLoc(Call->getDebugLoc());
     Args.clear();
     AttributesVec.clear();
 
     // Update the callgraph to know that the callsite has been transformed.
-    CallGraphNode *CalleeNode = CG[Call->getParent()->getParent()];
-    CalleeNode->replaceCallEdge(CS, CallSite(New), NF_CGN);
+    if (ReplaceCallSite)
+      (*ReplaceCallSite)(CS, CallSite(New));
 
     if (!Call->use_empty()) {
       Call->replaceAllUsesWith(New);
       New->takeName(Call);
     }
 
     // Finally, remove the old call from the program, reducing the use-count of
     // F.
while (!I->use_empty()) {
         GEP->eraseFromParent();
       }
     }
 
     // Increment I2 past all of the arguments added for this promoted pointer.
     std::advance(I2, ArgIndices.size());
   }
 
-  NF_CGN->stealCalledFunctionsFrom(CG[F]);
-
-  // Now that the old function is dead, delete it.  If there is a dangling
-  // reference to the CallgraphNode, just leave the dead function around for
-  // someone else to nuke.
-  CallGraphNode *CGN = CG[F];
-  if (CGN->getNumReferences() == 0)
-    delete CG.removeFunctionFromModule(CGN);
-  else
-    F->setLinkage(Function::ExternalLinkage);
-
-  return NF_CGN;
+  return NF;
 }
 
 /// AllCallersPassInValidPointerForArgument - Return true if we can prove that
 /// all callees pass in a valid pointer for the specified function argument.
 static bool allCallersPassInValidPointerForArgument(Argument *Arg) {
   Function *Callee = Arg->getParent();
   const DataLayout &DL = Callee->getParent()->getDataLayout();
 
static bool canPaddingBeAccessed(Argument *arg) {
   return false;
 }
 
 /// PromoteArguments - This method checks the specified function to see if there
 /// are any promotable arguments and if it is safe to promote the function (for
 /// example, all callers are direct).  If safe to promote some arguments, it
 /// calls the DoPromotion method.
 ///
-static CallGraphNode *
-promoteArguments(CallGraphNode *CGN, CallGraph &CG,
-                 function_ref<AAResults &(Function &F)> AARGetter,
-                 unsigned MaxElements) {
-  Function *F = CGN->getFunction();
-
+static Function *
+promoteArguments(Function *F, function_ref<AAResults &(Function &F)> AARGetter,
+                 unsigned MaxElements,
+                 Optional<function_ref<void(CallSite OldCS, CallSite NewCS)>>
+                     ReplaceCallSite) {
   // Make sure that it is local to this module.
-  if (!F || !F->hasLocalLinkage())
+  if (!F->hasLocalLinkage())
     return nullptr;
 
   // Don't promote arguments for variadic functions. Adding, removing, or
   // changing non-pack parameters can change the classification of pack
   // parameters. Frontends encode that classification at the call site in the
   // IR, while in the callee the classification is determined dynamically based
   // on the number of registers consumed so far.
   if (F->isVarArg())
if (isSafeToPromoteArgument(PtrArg, PtrArg->hasByValOrInAllocaAttr(), AAR,
                                 MaxElements))
       ArgsToPromote.insert(PtrArg);
   }
 
   // No promotable pointer arguments.
   if (ArgsToPromote.empty() && ByValArgsToTransform.empty())
     return nullptr;
 
-  return doPromotion(F, ArgsToPromote, ByValArgsToTransform, CG);
+  return doPromotion(F, ArgsToPromote, ByValArgsToTransform, ReplaceCallSite);
+}
+
+PreservedAnalyses ArgumentPromotionPass::run(LazyCallGraph::SCC &C,
+                                             CGSCCAnalysisManager &AM,
+                                             LazyCallGraph &CG,
+                                             CGSCCUpdateResult &UR) {
+  bool Changed = false, LocalChange;
+
+  // Iterate until we stop promoting from this SCC.
+  do {
+    LocalChange = false;
+
+    for (LazyCallGraph::Node &N : C) {
+      Function &OldF = N.getFunction();
+
+      FunctionAnalysisManager &FAM =
+          AM.getResult<FunctionAnalysisManagerCGSCCProxy>(C, CG).getManager();
+      // FIXME: This lambda must only be used with this function. We should
+      // skip the lambda and just get the AA results directly.
+      auto AARGetter = [&](Function &F) -> AAResults & {
+        assert(&F == &OldF && "Called with an unexpected function!");
+        return FAM.getResult<AAManager>(F);
+      };
+
+      Function *NewF = promoteArguments(&OldF, AARGetter, 3u, None);
+      if (!NewF)
+        continue;
+      LocalChange = true;
+
+      // Directly substitute the functions in the call graph. Note that this
+      // requires the old function to be completely dead and completely
+      // replaced by the new function. It does no call graph updates, it merely
+      // swaps out the particular function mapped to a particular node in the
+      // graph.
+      C.getOuterRefSCC().replaceNodeFunction(N, *NewF);
+      OldF.eraseFromParent();
+    }
+
+    Changed |= LocalChange;
+  } while (LocalChange);
+
+  if (!Changed)
+    return PreservedAnalyses::all();
+
+  return PreservedAnalyses::none();
 }
 
 namespace {
 /// ArgPromotion - The 'by reference' to 'by value' argument promotion pass.
 ///
 struct ArgPromotion : public CallGraphSCCPass {
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addRequired<AssumptionCacheTracker>();
bool ArgPromotion::runOnSCC(CallGraphSCC &SCC) {
 
   bool Changed = false, LocalChange;
 
   // Iterate until we stop promoting from this SCC.
   do {
     LocalChange = false;
     // Attempt to promote arguments from all functions in this SCC.
     for (CallGraphNode *OldNode : SCC) {
-      if (CallGraphNode *NewNode =
-              promoteArguments(OldNode, CG, AARGetter, MaxElements)) {
+      Function *OldF = OldNode->getFunction();
+      if (!OldF)
+        continue;
+
+      auto ReplaceCallSite = [&](CallSite OldCS, CallSite NewCS) {
+        Function *Caller = OldCS.getInstruction()->getParent()->getParent();
+        CallGraphNode *NewCalleeNode =
+            CG.getOrInsertFunction(NewCS.getCalledFunction());
+        CallGraphNode *CallerNode = CG[Caller];
+        CallerNode->replaceCallEdge(OldCS, NewCS, NewCalleeNode);
+      };
+
+      if (Function *NewF = promoteArguments(OldF, AARGetter, MaxElements,
+                                            {ReplaceCallSite})) {
         LocalChange = true;
+
+        // Update the call graph for the newly promoted function.
+        CallGraphNode *NewNode = CG.getOrInsertFunction(NewF);
+        NewNode->stealCalledFunctionsFrom(OldNode);
+        if (OldNode->getNumReferences() == 0)
+          delete CG.removeFunctionFromModule(OldNode);
+        else
+          OldF->setLinkage(Function::ExternalLinkage);
+
+        // And updat ethe SCC we're iterating as well.
         SCC.ReplaceNode(OldNode, NewNode);
       }
     }
     // Remember that we changed something.
     Changed |= LocalChange;
   } while (LocalChange);
 
   return Changed;
 }
 
 bool ArgPromotion::doInitialization(CallGraph &CG) {
   return CallGraphSCCPass::doInitialization(CG);
 }

llvm/trunk/test/Transforms/ArgumentPromotion/aggregate-promote.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 %T = type { i32, i32, i32, i32 }
 @G = constant %T { i32 0, i32 0, i32 17, i32 25 }
 
 define internal i32 @test(%T* %p) {
 ; CHECK-LABEL: define internal i32 @test(
 ; CHECK: i32 %{{.*}}, i32 %{{.*}})
 entry:

llvm/trunk/test/Transforms/ArgumentPromotion/attrs.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 %struct.ss = type { i32, i64 }
 
 ; Don't drop 'byval' on %X here.
 define internal void @f(%struct.ss* byval %b, i32* byval %X, i32 %i) nounwind {
 ; CHECK-LABEL: define internal void @f(
 ; CHECK: i32 %[[B0:.*]], i64 %[[B1:.*]], i32* byval %X, i32 %i)
 entry:

llvm/trunk/test/Transforms/ArgumentPromotion/byval-2.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 ; Arg promotion eliminates the struct argument.
 ; FIXME: Should it eliminate the i32* argument?
 
 %struct.ss = type { i32, i64 }
 
 define internal void @f(%struct.ss* byval  %b, i32* byval %X) nounwind  {
 ; CHECK-LABEL: define internal void @f(i32 %b.0, i64 %b.1, i32* byval %X)

llvm/trunk/test/Transforms/ArgumentPromotion/byval.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
 
 %struct.ss = type { i32, i64 }
 
 define internal void @f(%struct.ss* byval  %b) nounwind  {
 ; CHECK-LABEL: define internal void @f(i32 %b.0, i64 %b.1)
 entry:

llvm/trunk/test/Transforms/ArgumentPromotion/chained.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 @G1 = constant i32 0
 @G2 = constant i32* @G1
 
 define internal i32 @test(i32** %x) {
 ; CHECK-LABEL: define internal i32 @test(
 ; CHECK: i32 %{{.*}})
 entry:

llvm/trunk/test/Transforms/ArgumentPromotion/control-flow.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 ; Don't promote around control flow.
 define internal i32 @callee(i1 %C, i32* %P) {
 ; CHECK-LABEL: define internal i32 @callee(
 ; CHECK: i1 %C, i32* %P)
 entry:
   br i1 %C, label %T, label %F
 

llvm/trunk/test/Transforms/ArgumentPromotion/control-flow2.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 ; CHECK: load i32, i32* %A
 target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
 
 define internal i32 @callee(i1 %C, i32* %P) {
         br i1 %C, label %T, label %F
 
 T:              ; preds = %0

llvm/trunk/test/Transforms/ArgumentPromotion/crash.ll

 ; RUN: opt -S < %s -inline -argpromotion | FileCheck %s
+; RUN: opt -S < %s -passes=inline,argpromotion | FileCheck %s
 
 %S = type { %S* }
 
 ; Inlining should nuke the invoke (and any inlined calls) here even with
 ; argument promotion running along with it.
 define void @zot() personality i32 (...)* @wibble {
 ; CHECK-LABEL: define void @zot() personality i32 (...)* @wibble
 ; CHECK-NOT: call

llvm/trunk/test/Transforms/ArgumentPromotion/dbg.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 declare void @sink(i32)
 
 ; CHECK: define internal void @test({{.*}} !dbg [[SP:![0-9]+]]
 define internal void @test(i32** %X) !dbg !2 {
   %1 = load i32*, i32** %X, align 8
   %2 = load i32, i32* %1, align 8
   call void @sink(i32 %2)

llvm/trunk/test/Transforms/ArgumentPromotion/fp80.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 %union.u = type { x86_fp80 }
 %struct.s = type { double, i16, i8, [5 x i8] }
 
 @b = internal global %struct.s { double 3.14, i16 9439, i8 25, [5 x i8] undef }, align 16

llvm/trunk/test/Transforms/ArgumentPromotion/inalloca.ll

 ; RUN: opt %s -argpromotion -sroa -S | FileCheck %s
+; RUN: opt %s -passes='argpromotion,function(sroa)' -S | FileCheck %s
 
 target datalayout = "E-p:64:64:64-a0:0:8-f32:32:32-f64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-v64:64:64-v128:128:128"
 
 %struct.ss = type { i32, i32 }
 
 ; Argpromote + sroa should change this to passing the two integers by value.
 define internal i32 @f(%struct.ss* inalloca  %s) {
 entry:

llvm/trunk/test/Transforms/ArgumentPromotion/pr27568.ll

 ; RUN: opt -S -argpromotion < %s | FileCheck %s
+; RUN: opt -S -passes=argpromotion < %s | FileCheck %s
 target triple = "x86_64-pc-windows-msvc"
 
 define internal void @callee(i8*) {
 entry:
   call void @thunk()
   ret void
 }
 

llvm/trunk/test/Transforms/ArgumentPromotion/reserve-tbaa.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 ; PR17906
 ; When we promote two arguments in a single function with different types,
 ; before the fix, we used the same tag for the newly-created two loads.
 ; This testing case makes sure that we correctly transfer the tbaa tags from the
 ; original loads to the newly-created loads when promoting pointer arguments.
 
 @a = global i32* null, align 8

llvm/trunk/test/Transforms/ArgumentPromotion/sret.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 target datalayout = "e-m:w-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-pc-windows-msvc"
 
 ; CHECK: define internal void @add(i32 %[[THIS1:.*]], i32 %[[THIS2:.*]], i32* noalias %[[SR:.*]])
 define internal void @add({i32, i32}* %this, i32* sret %r) {
   %ap = getelementptr {i32, i32}, {i32, i32}* %this, i32 0, i32 0
   %bp = getelementptr {i32, i32}, {i32, i32}* %this, i32 0, i32 1

llvm/trunk/test/Transforms/ArgumentPromotion/tail.ll

 ; RUN: opt %s -argpromotion -S -o - | FileCheck %s
+; RUN: opt %s -passes=argpromotion -S -o - | FileCheck %s
 ; PR14710
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 
 %pair = type { i32, i32 }
 
 declare i8* @foo(%pair*)
 

llvm/trunk/test/Transforms/ArgumentPromotion/variadic.ll

 ; RUN: opt < %s -argpromotion -S | FileCheck %s
+; RUN: opt < %s -passes=argpromotion -S | FileCheck %s
 
 ; Unused arguments from variadic functions cannot be eliminated as that changes
 ; their classiciation according to the SysV amd64 ABI. Clang and other frontends
 ; bake in the classification when they use things like byval, as in this test.
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"