[LPM] A targeted but somewhat horrible fix to the legacy pass manager's

querying of the pass registry.

The pass manager relies on the static registry of PassInfo objects to
perform all manner of its functionality. I don't understand why it does
much of this. My very vague understanding is that this registry is
touched both during static initialization *and* while each pass is being
constructed. As a consequence it is hard to make accessing it not
require a acquiring some lock. This lock ends up in the hot path of
setting up, tearing down, and invaliditing analyses in the legacy pass
manager.

On most systems you can observe this as a non-trivial % of the time
spent in 'ninja check-llvm'. However, I haven't really seen it be more
than 1% in extreme cases of compiling more real-world software,
including LTO.

Unfortunately, some of the GPU JITs are seeing this taking essentially
all of their time because they have very small IR running through
a small pass pipeline very many times (at least, this is the vague
understanding I have of it).

This patch tries to minimize the cost of looking up PassInfo objects by
leveraging the fact that the objects themselves are immutable and they
are allocated separately on the heap and so don't have their address
change. It also requires a change I made the last time I tried to debug
this problem which removed the ability to de-register a pass from the
registry. This patch creates a single access path to these objects
inside the PMTopLevelManager which memoizes the result of querying the
registry. This is somewhat gross as I don't really know if
PMTopLevelManager is the *right* place to put it, and I dislike using
a mutable member to memoize things, but it seems to work.

For long-lived pass managers this should completely eliminate
the cost of acquiring locks to look into the pass registry once the
memoized cache is warm. For 'ninja check' I measured about 1.5%
reduction in CPU time and in total time on a machine with 32 hardware
threads. For normal compilation, I don't know how much this will help,
sadly. We will still pay the cost while we populate the memoized cache.
I don't think it will hurt though, and for LTO or compiles with many
small functions it should still be a win. However, for tight loops
around a pass manager with many passes and small modules, this will help
tremendously. On the AArch64 backend I saw nearly 50% reductions in time
to complete 2000 cycles of spinning up and tearing down the pipeline.
Measurements from Owen of an actual long-lived pass manager show more
along the lines of 10% improvements.

Differential Revision: http://reviews.llvm.org/D7213

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@227299 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 22 insertions, 14 deletions

diff --git a/lib/IR/LegacyPassManager.cpp b/lib/IR/LegacyPassManager.cpp
index b9ab25651fa..fa8d50ec160 100644
--- a/lib/IR/LegacyPassManager.cpp
+++ b/lib/IR/LegacyPassManager.cpp
@@ -600,8 +600,7 @@ void PMTopLevelManager::schedulePass(Pass *P) {
   // If P is an analysis pass and it is available then do not
   // generate the analysis again. Stale analysis info should not be
   // available at this point.
-  const PassInfo *PI =
-    PassRegistry::getPassRegistry()->getPassInfo(P->getPassID());
+  const PassInfo *PI = findAnalysisPassInfo(P->getPassID());
   if (PI && PI->isAnalysis() && findAnalysisPass(P->getPassID())) {
     delete P;
     return;
@@ -619,7 +618,7 @@ void PMTopLevelManager::schedulePass(Pass *P) {
 
       Pass *AnalysisPass = findAnalysisPass(*I);
       if (!AnalysisPass) {
-        const PassInfo *PI = PassRegistry::getPassRegistry()->getPassInfo(*I);
+        const PassInfo *PI = findAnalysisPassInfo(*I);
 
         if (!PI) {
           // Pass P is not in the global PassRegistry
@@ -716,8 +715,7 @@ Pass *PMTopLevelManager::findAnalysisPass(AnalysisID AID) {
       return *I;
 
     // If Pass not found then check the interfaces implemented by Immutable Pass
-    const PassInfo *PassInf =
-      PassRegistry::getPassRegistry()->getPassInfo(PI);
+    const PassInfo *PassInf = findAnalysisPassInfo(PI);
     assert(PassInf && "Expected all immutable passes to be initialized");
     const std::vector<const PassInfo*> &ImmPI =
       PassInf->getInterfacesImplemented();
@@ -731,6 +729,17 @@ Pass *PMTopLevelManager::findAnalysisPass(AnalysisID AID) {
   return nullptr;
 }
 
+const PassInfo *PMTopLevelManager::findAnalysisPassInfo(AnalysisID AID) const {
+  const PassInfo *&PI = AnalysisPassInfos[AID];
+  if (!PI)
+    PI = PassRegistry::getPassRegistry()->getPassInfo(AID);
+  else
+    assert(PI == PassRegistry::getPassRegistry()->getPassInfo(AID) &&
+           "The pass info pointer changed for an analysis ID!");
+
+  return PI;
+}
+
 // Print passes managed by this top level manager.
 void PMTopLevelManager::dumpPasses() const {
 
@@ -759,8 +768,7 @@ void PMTopLevelManager::dumpArguments() const {
   dbgs() << "Pass Arguments: ";
   for (SmallVectorImpl<ImmutablePass *>::const_iterator I =
        ImmutablePasses.begin(), E = ImmutablePasses.end(); I != E; ++I)
-    if (const PassInfo *PI =
-        PassRegistry::getPassRegistry()->getPassInfo((*I)->getPassID())) {
+    if (const PassInfo *PI = findAnalysisPassInfo((*I)->getPassID())) {
       assert(PI && "Expected all immutable passes to be initialized");
       if (!PI->isAnalysisGroup())
         dbgs() << " -" << PI->getPassArgument();
@@ -824,7 +832,7 @@ void PMDataManager::recordAvailableAnalysis(Pass *P) {
 
   // This pass is the current implementation of all of the interfaces it
   // implements as well.
-  const PassInfo *PInf = PassRegistry::getPassRegistry()->getPassInfo(PI);
+  const PassInfo *PInf = TPM->findAnalysisPassInfo(PI);
   if (!PInf) return;
   const std::vector<const PassInfo*> &II = PInf->getInterfacesImplemented();
   for (unsigned i = 0, e = II.size(); i != e; ++i)
@@ -957,7 +965,7 @@ void PMDataManager::freePass(Pass *P, StringRef Msg,
   }
 
   AnalysisID PI = P->getPassID();
-  if (const PassInfo *PInf = PassRegistry::getPassRegistry()->getPassInfo(PI)) {
+  if (const PassInfo *PInf = TPM->findAnalysisPassInfo(PI)) {
     // Remove the pass itself (if it is not already removed).
     AvailableAnalysis.erase(PI);
 
@@ -1037,7 +1045,7 @@ void PMDataManager::add(Pass *P, bool ProcessAnalysis) {
   for (SmallVectorImpl<AnalysisID>::iterator
          I = ReqAnalysisNotAvailable.begin(),
          E = ReqAnalysisNotAvailable.end() ;I != E; ++I) {
-    const PassInfo *PI = PassRegistry::getPassRegistry()->getPassInfo(*I);
+    const PassInfo *PI = TPM->findAnalysisPassInfo(*I);
     Pass *AnalysisPass = PI->createPass();
     this->addLowerLevelRequiredPass(P, AnalysisPass);
   }
@@ -1142,7 +1150,7 @@ void PMDataManager::dumpPassArguments() const {
       PMD->dumpPassArguments();
     else
       if (const PassInfo *PI =
-            PassRegistry::getPassRegistry()->getPassInfo((*I)->getPassID()))
+            TPM->findAnalysisPassInfo((*I)->getPassID()))
         if (!PI->isAnalysisGroup())
           dbgs() << " -" << PI->getPassArgument();
   }
@@ -1218,7 +1226,7 @@ void PMDataManager::dumpAnalysisUsage(StringRef Msg, const Pass *P,
   dbgs() << (const void*)P << std::string(getDepth()*2+3, ' ') << Msg << " Analyses:";
   for (unsigned i = 0; i != Set.size(); ++i) {
     if (i) dbgs() << ',';
-    const PassInfo *PInf = PassRegistry::getPassRegistry()->getPassInfo(Set[i]);
+    const PassInfo *PInf = TPM->findAnalysisPassInfo(Set[i]);
     if (!PInf) {
       // Some preserved passes, such as AliasAnalysis, may not be initialized by
       // all drivers.
@@ -1658,8 +1666,8 @@ void MPPassManager::addLowerLevelRequiredPass(Pass *P, Pass *RequiredPass) {
 
     OnTheFlyManagers[P] = FPP;
   }
-  const PassInfo * RequiredPassPI =
-    PassRegistry::getPassRegistry()->getPassInfo(RequiredPass->getPassID());
+  const PassInfo *RequiredPassPI =
+      TPM->findAnalysisPassInfo(RequiredPass->getPassID());
 
   Pass *FoundPass = nullptr;
   if (RequiredPassPI && RequiredPassPI->isAnalysis()) {