mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-12-13 20:32:21 +00:00
1b6ca9d0cc
CaptureTracking becomes very expensive in large basic blocks while calling PointerMayBeCaptured. PointerMayBeCaptured scans the BB the number of times equal to the number of uses of 'BeforeHere', which is currently capped at 20 and bails out with Tracker->tooManyUses(). The bottleneck here is the number of calls to PointerMayBeCaptured * the basic block scan. In a testcase with a 82k instruction BB, PointerMayBeCaptured is called 130k times, leading to 'shouldExplore' taking 527k runs, this currently takes ~12min. To fix this we locally (within PointerMayBeCaptured) number the instructions in the basic block using a DenseMap to cache instruction positions/numbers. We build the cache incrementally every time we need to scan an unexplored part of the BB, improving compile time to only take ~2min. This triggers in the flow: DeadStoreElimination -> MepDepAnalysis -> CaptureTracking. Side note: after multiple runs in the test-suite I've seen no performance nor compile time regressions, but could note a couple of compile time improvements: Performance Improvements - Compile Time Delta Previous Current StdDev SingleSource/Benchmarks/Misc-C++/bigfib -4.48% 0.8547 0.8164 0.0022 MultiSource/Benchmarks/TSVC/LoopRerolling-dbl/LoopRerolling-dbl -1.47% 1.3912 1.3707 0.0056 Differential Revision: http://reviews.llvm.org/D7010 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@240560 91177308-0d34-0410-b5e6-96231b3b80d8
235 lines
8.3 KiB
C++
235 lines
8.3 KiB
C++
//===-- CFG.cpp - BasicBlock analysis --------------------------------------==//
|
|
//
|
|
// The LLVM Compiler Infrastructure
|
|
//
|
|
// This file is distributed under the University of Illinois Open Source
|
|
// License. See LICENSE.TXT for details.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
//
|
|
// This family of functions performs analyses on basic blocks, and instructions
|
|
// contained within basic blocks.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#include "llvm/Analysis/CFG.h"
|
|
#include "llvm/ADT/SmallSet.h"
|
|
#include "llvm/Analysis/LoopInfo.h"
|
|
#include "llvm/IR/Dominators.h"
|
|
|
|
using namespace llvm;
|
|
|
|
/// FindFunctionBackedges - Analyze the specified function to find all of the
|
|
/// loop backedges in the function and return them. This is a relatively cheap
|
|
/// (compared to computing dominators and loop info) analysis.
|
|
///
|
|
/// The output is added to Result, as pairs of <from,to> edge info.
|
|
void llvm::FindFunctionBackedges(const Function &F,
|
|
SmallVectorImpl<std::pair<const BasicBlock*,const BasicBlock*> > &Result) {
|
|
const BasicBlock *BB = &F.getEntryBlock();
|
|
if (succ_empty(BB))
|
|
return;
|
|
|
|
SmallPtrSet<const BasicBlock*, 8> Visited;
|
|
SmallVector<std::pair<const BasicBlock*, succ_const_iterator>, 8> VisitStack;
|
|
SmallPtrSet<const BasicBlock*, 8> InStack;
|
|
|
|
Visited.insert(BB);
|
|
VisitStack.push_back(std::make_pair(BB, succ_begin(BB)));
|
|
InStack.insert(BB);
|
|
do {
|
|
std::pair<const BasicBlock*, succ_const_iterator> &Top = VisitStack.back();
|
|
const BasicBlock *ParentBB = Top.first;
|
|
succ_const_iterator &I = Top.second;
|
|
|
|
bool FoundNew = false;
|
|
while (I != succ_end(ParentBB)) {
|
|
BB = *I++;
|
|
if (Visited.insert(BB).second) {
|
|
FoundNew = true;
|
|
break;
|
|
}
|
|
// Successor is in VisitStack, it's a back edge.
|
|
if (InStack.count(BB))
|
|
Result.push_back(std::make_pair(ParentBB, BB));
|
|
}
|
|
|
|
if (FoundNew) {
|
|
// Go down one level if there is a unvisited successor.
|
|
InStack.insert(BB);
|
|
VisitStack.push_back(std::make_pair(BB, succ_begin(BB)));
|
|
} else {
|
|
// Go up one level.
|
|
InStack.erase(VisitStack.pop_back_val().first);
|
|
}
|
|
} while (!VisitStack.empty());
|
|
}
|
|
|
|
/// GetSuccessorNumber - Search for the specified successor of basic block BB
|
|
/// and return its position in the terminator instruction's list of
|
|
/// successors. It is an error to call this with a block that is not a
|
|
/// successor.
|
|
unsigned llvm::GetSuccessorNumber(BasicBlock *BB, BasicBlock *Succ) {
|
|
TerminatorInst *Term = BB->getTerminator();
|
|
#ifndef NDEBUG
|
|
unsigned e = Term->getNumSuccessors();
|
|
#endif
|
|
for (unsigned i = 0; ; ++i) {
|
|
assert(i != e && "Didn't find edge?");
|
|
if (Term->getSuccessor(i) == Succ)
|
|
return i;
|
|
}
|
|
}
|
|
|
|
/// isCriticalEdge - Return true if the specified edge is a critical edge.
|
|
/// Critical edges are edges from a block with multiple successors to a block
|
|
/// with multiple predecessors.
|
|
bool llvm::isCriticalEdge(const TerminatorInst *TI, unsigned SuccNum,
|
|
bool AllowIdenticalEdges) {
|
|
assert(SuccNum < TI->getNumSuccessors() && "Illegal edge specification!");
|
|
if (TI->getNumSuccessors() == 1) return false;
|
|
|
|
const BasicBlock *Dest = TI->getSuccessor(SuccNum);
|
|
const_pred_iterator I = pred_begin(Dest), E = pred_end(Dest);
|
|
|
|
// If there is more than one predecessor, this is a critical edge...
|
|
assert(I != E && "No preds, but we have an edge to the block?");
|
|
const BasicBlock *FirstPred = *I;
|
|
++I; // Skip one edge due to the incoming arc from TI.
|
|
if (!AllowIdenticalEdges)
|
|
return I != E;
|
|
|
|
// If AllowIdenticalEdges is true, then we allow this edge to be considered
|
|
// non-critical iff all preds come from TI's block.
|
|
for (; I != E; ++I)
|
|
if (*I != FirstPred)
|
|
return true;
|
|
return false;
|
|
}
|
|
|
|
// LoopInfo contains a mapping from basic block to the innermost loop. Find
|
|
// the outermost loop in the loop nest that contains BB.
|
|
static const Loop *getOutermostLoop(const LoopInfo *LI, const BasicBlock *BB) {
|
|
const Loop *L = LI->getLoopFor(BB);
|
|
if (L) {
|
|
while (const Loop *Parent = L->getParentLoop())
|
|
L = Parent;
|
|
}
|
|
return L;
|
|
}
|
|
|
|
// True if there is a loop which contains both BB1 and BB2.
|
|
static bool loopContainsBoth(const LoopInfo *LI,
|
|
const BasicBlock *BB1, const BasicBlock *BB2) {
|
|
const Loop *L1 = getOutermostLoop(LI, BB1);
|
|
const Loop *L2 = getOutermostLoop(LI, BB2);
|
|
return L1 != nullptr && L1 == L2;
|
|
}
|
|
|
|
bool llvm::isPotentiallyReachableFromMany(
|
|
SmallVectorImpl<BasicBlock *> &Worklist, BasicBlock *StopBB,
|
|
const DominatorTree *DT, const LoopInfo *LI) {
|
|
// When the stop block is unreachable, it's dominated from everywhere,
|
|
// regardless of whether there's a path between the two blocks.
|
|
if (DT && !DT->isReachableFromEntry(StopBB))
|
|
DT = nullptr;
|
|
|
|
// Limit the number of blocks we visit. The goal is to avoid run-away compile
|
|
// times on large CFGs without hampering sensible code. Arbitrarily chosen.
|
|
unsigned Limit = 32;
|
|
SmallSet<const BasicBlock*, 64> Visited;
|
|
do {
|
|
BasicBlock *BB = Worklist.pop_back_val();
|
|
if (!Visited.insert(BB).second)
|
|
continue;
|
|
if (BB == StopBB)
|
|
return true;
|
|
if (DT && DT->dominates(BB, StopBB))
|
|
return true;
|
|
if (LI && loopContainsBoth(LI, BB, StopBB))
|
|
return true;
|
|
|
|
if (!--Limit) {
|
|
// We haven't been able to prove it one way or the other. Conservatively
|
|
// answer true -- that there is potentially a path.
|
|
return true;
|
|
}
|
|
|
|
if (const Loop *Outer = LI ? getOutermostLoop(LI, BB) : nullptr) {
|
|
// All blocks in a single loop are reachable from all other blocks. From
|
|
// any of these blocks, we can skip directly to the exits of the loop,
|
|
// ignoring any other blocks inside the loop body.
|
|
Outer->getExitBlocks(Worklist);
|
|
} else {
|
|
Worklist.append(succ_begin(BB), succ_end(BB));
|
|
}
|
|
} while (!Worklist.empty());
|
|
|
|
// We have exhausted all possible paths and are certain that 'To' can not be
|
|
// reached from 'From'.
|
|
return false;
|
|
}
|
|
|
|
bool llvm::isPotentiallyReachable(const BasicBlock *A, const BasicBlock *B,
|
|
const DominatorTree *DT, const LoopInfo *LI) {
|
|
assert(A->getParent() == B->getParent() &&
|
|
"This analysis is function-local!");
|
|
|
|
SmallVector<BasicBlock*, 32> Worklist;
|
|
Worklist.push_back(const_cast<BasicBlock*>(A));
|
|
|
|
return isPotentiallyReachableFromMany(Worklist, const_cast<BasicBlock *>(B),
|
|
DT, LI);
|
|
}
|
|
|
|
bool llvm::isPotentiallyReachable(const Instruction *A, const Instruction *B,
|
|
const DominatorTree *DT, const LoopInfo *LI) {
|
|
assert(A->getParent()->getParent() == B->getParent()->getParent() &&
|
|
"This analysis is function-local!");
|
|
|
|
SmallVector<BasicBlock*, 32> Worklist;
|
|
|
|
if (A->getParent() == B->getParent()) {
|
|
// The same block case is special because it's the only time we're looking
|
|
// within a single block to see which instruction comes first. Once we
|
|
// start looking at multiple blocks, the first instruction of the block is
|
|
// reachable, so we only need to determine reachability between whole
|
|
// blocks.
|
|
BasicBlock *BB = const_cast<BasicBlock *>(A->getParent());
|
|
|
|
// If the block is in a loop then we can reach any instruction in the block
|
|
// from any other instruction in the block by going around a backedge.
|
|
if (LI && LI->getLoopFor(BB) != nullptr)
|
|
return true;
|
|
|
|
// Linear scan, start at 'A', see whether we hit 'B' or the end first.
|
|
for (BasicBlock::const_iterator I = A, E = BB->end(); I != E; ++I) {
|
|
if (&*I == B)
|
|
return true;
|
|
}
|
|
|
|
// Can't be in a loop if it's the entry block -- the entry block may not
|
|
// have predecessors.
|
|
if (BB == &BB->getParent()->getEntryBlock())
|
|
return false;
|
|
|
|
// Otherwise, continue doing the normal per-BB CFG walk.
|
|
Worklist.append(succ_begin(BB), succ_end(BB));
|
|
|
|
if (Worklist.empty()) {
|
|
// We've proven that there's no path!
|
|
return false;
|
|
}
|
|
} else {
|
|
Worklist.push_back(const_cast<BasicBlock*>(A->getParent()));
|
|
}
|
|
|
|
if (A->getParent() == &A->getParent()->getParent()->getEntryBlock())
|
|
return true;
|
|
if (B->getParent() == &A->getParent()->getParent()->getEntryBlock())
|
|
return false;
|
|
|
|
return isPotentiallyReachableFromMany(
|
|
Worklist, const_cast<BasicBlock *>(B->getParent()), DT, LI);
|
|
}
|