[PM/Inliner] Make the new PM's inliner process call edges across an
entire SCC before iterating on newly-introduced call edges resulting
from any inlined function bodies.
This more closely matches the behavior of the old PM's inliner. While it
wasn't really clear to me initially, this behavior is actually essential
to the inliner behaving reasonably in its current design.
Because the inliner is fundamentally a bottom-up inliner and all of its
cost modeling is designed around that it often runs into trouble within
an SCC where we don't have any meaningful bottom-up ordering to use. In
addition to potentially cyclic, infinite inlining that we block with the
inline history mechanism, it can also take seemingly simple call graph
patterns within an SCC and turn them into *insanely* large functions by
accidentally working top-down across the SCC without any of the
threshold limitations that traditional top-down inliners use.
Consider this diabolical monster.cpp file that Richard Smith came up
with to help demonstrate this issue:
```
template <int N> extern const char *str;
void g(const char *);
template <bool K, int N> void f(bool *B, bool *E) {
if (K)
g(str<N>);
if (B == E)
return;
if (*B)
f<true, N + 1>(B + 1, E);
else
f<false, N + 1>(B + 1, E);
}
template <> void f<false, MAX>(bool *B, bool *E) { return f<false, 0>(B, E); }
template <> void f<true, MAX>(bool *B, bool *E) { return f<true, 0>(B, E); }
extern bool *arr, *end;
void test() { f<false, 0>(arr, end); }
```
When compiled with '-DMAX=N' for various values of N, this will create an SCC
with a reasonably large number of functions. Previously, the inliner would try
to exhaust the inlining candidates in a single function before moving on. This,
unfortunately, turns it into a top-down inliner within the SCC. Because our
thresholds were never built for that, we will incrementally decide that it is
always worth inlining and proceed to flatten the entire SCC into that one
function.
What's worse, we'll then proceed to the next function, and do the exact same
thing except we'll skip the first function, and so on. And at each step, we'll
also make some of the constant factors larger, which is awesome.
The fix in this patch is the obvious one which makes the new PM's inliner use
the same technique used by the old PM: consider all the call edges across the
entire SCC before beginning to process call edges introduced by inlining. The
result of this is essentially to distribute the inlining across the SCC so that
every function incrementally grows toward the inline thresholds rather than
allowing the inliner to grow one of the functions vastly beyond the threshold.
The code for this is a bit awkward, but it works out OK.
We could consider in the future doing something more powerful here such as
prioritized order (via lowest cost and/or profile info) and/or a code-growth
budget per SCC. However, both of those would require really substantial work
both to design the system in a way that wouldn't break really useful
abstraction decomposition properties of the current inliner and to be tuned
across a reasonably diverse set of code and workloads. It also seems really
risky in many ways. I have only found a single real-world file that triggers
the bad behavior here and it is generated code that has a pretty pathological
pattern. I'm not worried about the inliner not doing an *awesome* job here as
long as it does *ok*. On the other hand, the cases that will be tricky to get
right in a prioritized scheme with a budget will be more common and idiomatic
for at least some frontends (C++ and Rust at least). So while these approaches
are still really interesting, I'm not in a huge rush to go after them. Staying
even closer to the existing PM's behavior, especially when this easy to do,
seems like the right short to medium term approach.
I don't really have a test case that makes sense yet... I'll try to find a
variant of the IR produced by the monster template metaprogram that is both
small enough to be sane and large enough to clearly show when we get this wrong
in the future. But I'm not confident this exists. And the behavior change here
*should* be unobservable without snooping on debug logging. So there isn't
really much to test.
The test case updates come from two incidental changes:
1) We now visit functions in an SCC in the opposite order. I don't think there
really is a "right" order here, so I just update the test cases.
2) We no longer compute some analyses when an SCC has no call instructions that
we consider for inlining.
llvm-svn: 297374
diff --git a/llvm/lib/Transforms/IPO/Inliner.cpp b/llvm/lib/Transforms/IPO/Inliner.cpp
index 7049a6f..5ea26a8 100644
--- a/llvm/lib/Transforms/IPO/Inliner.cpp
+++ b/llvm/lib/Transforms/IPO/Inliner.cpp
@@ -746,20 +746,52 @@
Module &M = *InitialC.begin()->getFunction().getParent();
ProfileSummaryInfo *PSI = MAM.getCachedResult<ProfileSummaryAnalysis>(M);
- // We use a worklist of nodes to process so that we can handle if the SCC
- // structure changes and some nodes are no longer part of the current SCC. We
- // also need to use an updatable pointer for the SCC as a consequence.
- SmallVector<LazyCallGraph::Node *, 16> Nodes;
- for (auto &N : InitialC)
- Nodes.push_back(&N);
+ // We use a single common worklist for calls across the entire SCC. We
+ // process these in-order and append new calls introduced during inlining to
+ // the end.
+ //
+ // Note that this particular order of processing is actually critical to
+ // avoid very bad behaviors. Consider *highly connected* call graphs where
+ // each function contains a small amonut of code and a couple of calls to
+ // other functions. Because the LLVM inliner is fundamentally a bottom-up
+ // inliner, it can handle gracefully the fact that these all appear to be
+ // reasonable inlining candidates as it will flatten things until they become
+ // too big to inline, and then move on and flatten another batch.
+ //
+ // However, when processing call edges *within* an SCC we cannot rely on this
+ // bottom-up behavior. As a consequence, with heavily connected *SCCs* of
+ // functions we can end up incrementally inlining N calls into each of
+ // N functions because each incremental inlining decision looks good and we
+ // don't have a topological ordering to prevent explosions.
+ //
+ // To compensate for this, we don't process transitive edges made immediate
+ // by inlining until we've done one pass of inlining across the entire SCC.
+ // Large, highly connected SCCs still lead to some amount of code bloat in
+ // this model, but it is uniformly spread across all the functions in the SCC
+ // and eventually they all become too large to inline, rather than
+ // incrementally maknig a single function grow in a super linear fashion.
+ SmallVector<std::pair<CallSite, int>, 16> Calls;
+
+ // Populate the initial list of calls in this SCC.
+ for (auto &N : InitialC) {
+ // We want to generally process call sites top-down in order for
+ // simplifications stemming from replacing the call with the returned value
+ // after inlining to be visible to subsequent inlining decisions.
+ // FIXME: Using instructions sequence is a really bad way to do this.
+ // Instead we should do an actual RPO walk of the function body.
+ for (Instruction &I : instructions(N.getFunction()))
+ if (auto CS = CallSite(&I))
+ if (Function *Callee = CS.getCalledFunction())
+ if (!Callee->isDeclaration())
+ Calls.push_back({CS, -1});
+ }
+ if (Calls.empty())
+ return PreservedAnalyses::all();
+
+ // Capture updatable variables for the current SCC and RefSCC.
auto *C = &InitialC;
auto *RC = &C->getOuterRefSCC();
- // We also use a secondary worklist of call sites within a particular node to
- // allow quickly continuing to inline through newly inlined call sites where
- // possible.
- SmallVector<std::pair<CallSite, int>, 16> Calls;
-
// When inlining a callee produces new call sites, we want to keep track of
// the fact that they were inlined from the callee. This allows us to avoid
// infinite inlining in some obscure cases. To represent this, we use an
@@ -775,11 +807,17 @@
// defer deleting these to make it easier to handle the call graph updates.
SmallVector<Function *, 4> DeadFunctions;
- do {
- auto &N = *Nodes.pop_back_val();
+ // Loop forward over all of the calls. Note that we cannot cache the size as
+ // inlining can introduce new calls that need to be processed.
+ for (int i = 0; i < (int)Calls.size(); ++i) {
+ // We expect the calls to typically be batched with sequences of calls that
+ // have the same caller, so we first set up some shared infrastructure for
+ // this caller. We also do any pruning we can at this layer on the caller
+ // alone.
+ Function &F = *Calls[i].first.getCaller();
+ LazyCallGraph::Node &N = *CG.lookup(F);
if (CG.lookupSCC(N) != C)
continue;
- Function &F = N.getFunction();
if (F.hasFnAttribute(Attribute::OptimizeNone))
continue;
@@ -813,23 +851,14 @@
// Get the remarks emission analysis for the caller.
auto &ORE = FAM.getResult<OptimizationRemarkEmitterAnalysis>(F);
- // We want to generally process call sites top-down in order for
- // simplifications stemming from replacing the call with the returned value
- // after inlining to be visible to subsequent inlining decisions. So we
- // walk the function backwards and then process the back of the vector.
- // FIXME: Using reverse is a really bad way to do this. Instead we should
- // do an actual PO walk of the function body.
- for (Instruction &I : reverse(instructions(F)))
- if (auto CS = CallSite(&I))
- if (Function *Callee = CS.getCalledFunction())
- if (!Callee->isDeclaration())
- Calls.push_back({CS, -1});
-
+ // Now process as many calls as we have within this caller in the sequnece.
+ // We bail out as soon as the caller has to change so we can update the
+ // call graph and prepare the context of that new caller.
bool DidInline = false;
- while (!Calls.empty()) {
+ for (; i < (int)Calls.size() && Calls[i].first.getCaller() == &F; ++i) {
int InlineHistoryID;
CallSite CS;
- std::tie(CS, InlineHistoryID) = Calls.pop_back_val();
+ std::tie(CS, InlineHistoryID) = Calls[i];
Function &Callee = *CS.getCalledFunction();
if (InlineHistoryID != -1 &&
@@ -886,6 +915,10 @@
}
}
+ // Back the call index up by one to put us in a good position to go around
+ // the outer loop.
+ --i;
+
if (!DidInline)
continue;
Changed = true;
@@ -914,7 +947,7 @@
C = &updateCGAndAnalysisManagerForFunctionPass(CG, *C, N, AM, UR);
DEBUG(dbgs() << "Updated inlining SCC: " << *C << "\n");
RC = &C->getOuterRefSCC();
- } while (!Nodes.empty());
+ }
// Now that we've finished inlining all of the calls across this SCC, delete
// all of the trivially dead functions, updating the call graph and the CGSCC