Unverified Commit 2ad9fde4 authored by Nikita Popov's avatar Nikita Popov Committed by GitHub
Browse files

[MemDep] Use EarliestEscapeInfo (#69727) 

Use BatchAA with EarliestEscapeInfo instead of callCapturesBefore() in
MemDepAnalysis. The advantage of this is that it will also take
not-captured-before information into account for non-calls (see
test_store_before_capture for a representative example), and that this
is a cached analysis. The disadvantage is that EII is slightly less
precise than full CapturedBefore analysis.

In practice the impact is positive, with gvn.NumGVNLoad going from 22022
to 22808 on test-suite. 

The impact to compile-time is also positive, mainly in the ThinLTO
configuration.
parent ab261eb3

llvm/include/llvm/Analysis/MemoryDependenceAnalysis.h

+3 −2
Original line number Diff line number Diff line
@@ -18,6 +18,7 @@ 
#include "llvm/ADT/PointerIntPair.h"

#include "llvm/ADT/PointerSumType.h"

#include "llvm/ADT/SmallPtrSet.h"

#include "llvm/Analysis/AliasAnalysis.h"

#include "llvm/Analysis/MemoryLocation.h"

#include "llvm/IR/PassManager.h"

#include "llvm/IR/PredIteratorCache.h"
@@ -27,7 +28,6 @@ 


namespace llvm {



class AAResults;

class AssumptionCache;

class BatchAAResults;

class DominatorTree;
@@ -356,6 +356,7 @@ private: 
  const TargetLibraryInfo &TLI;

  DominatorTree &DT;

  PredIteratorCache PredCache;

  EarliestEscapeInfo EII;



  unsigned DefaultBlockScanLimit;
@@ -367,7 +368,7 @@ public: 
  MemoryDependenceResults(AAResults &AA, AssumptionCache &AC,

                          const TargetLibraryInfo &TLI, DominatorTree &DT,

                          unsigned DefaultBlockScanLimit)

      : AA(AA), AC(AC), TLI(TLI), DT(DT),

      : AA(AA), AC(AC), TLI(TLI), DT(DT), EII(DT),

        DefaultBlockScanLimit(DefaultBlockScanLimit) {}



  /// Handle invalidation in the new PM.

llvm/lib/Analysis/MemoryDependenceAnalysis.cpp

+5 −7
Original line number Diff line number Diff line
@@ -268,7 +268,7 @@ MemDepResult MemoryDependenceResults::getPointerDependencyFrom( 
MemDepResult MemoryDependenceResults::getPointerDependencyFrom(

    const MemoryLocation &MemLoc, bool isLoad, BasicBlock::iterator ScanIt,

    BasicBlock *BB, Instruction *QueryInst, unsigned *Limit) {

  BatchAAResults BatchAA(AA);

  BatchAAResults BatchAA(AA, &EII);

  return getPointerDependencyFrom(MemLoc, isLoad, ScanIt, BB, QueryInst, Limit,

                                  BatchAA);

}
@@ -645,11 +645,7 @@ MemDepResult MemoryDependenceResults::getSimplePointerDependencyFrom( 
        continue;



    // See if this instruction (e.g. a call or vaarg) mod/ref's the pointer.

    ModRefInfo MR = BatchAA.getModRefInfo(Inst, MemLoc);

    // If necessary, perform additional analysis.

    if (isModAndRefSet(MR))

      MR = BatchAA.callCapturesBefore(Inst, MemLoc, &DT);

    switch (MR) {

    switch (BatchAA.getModRefInfo(Inst, MemLoc)) {

    case ModRefInfo::NoModRef:

      // If the call has no effect on the queried pointer, just ignore it.

      continue;
@@ -1227,7 +1223,7 @@ bool MemoryDependenceResults::getNonLocalPointerDepFromBB( 
  bool GotWorklistLimit = false;

  LLVM_DEBUG(AssertSorted(*Cache));



  BatchAAResults BatchAA(AA);

  BatchAAResults BatchAA(AA, &EII);

  while (!Worklist.empty()) {

    BasicBlock *BB = Worklist.pop_back_val();
@@ -1539,6 +1535,8 @@ void MemoryDependenceResults::invalidateCachedPredecessors() { 
}



void MemoryDependenceResults::removeInstruction(Instruction *RemInst) {

  EII.removeInstruction(RemInst);



  // Walk through the Non-local dependencies, removing this one as the value

  // for any cached queries.

  NonLocalDepMapType::iterator NLDI = NonLocalDepsMap.find(RemInst);

llvm/test/CodeGen/BPF/CORE/no-narrow-load.ll

+4 −4
Original line number Diff line number Diff line
@@ -65,10 +65,10 @@ lor.end: ; preds = %lor.end.critedge, % 
  ret void, !dbg !53

}



; CHECK: r[[LOAD1:[0-9]+]] = *(u32 *)(r{{[0-9]+}} + 4)

; CHECK: r[[LOAD1]] &= 65536

; CHECK: r[[LOAD2:[0-9]+]] = *(u32 *)(r{{[0-9]+}} + 4)

; CHECK: r[[LOAD2]] &= 32768

; CHECK: r[[LOAD:[0-9]+]] = *(u32 *)(r{{[0-9]+}} + 4)

; CHECK: r[[COPY:[0-9]+]] = r[[LOAD]]

; CHECK: r[[COPY]] &= 65536

; CHECK: r[[LOAD]] &= 32768



; Function Attrs: nounwind readnone speculatable willreturn

declare void @llvm.dbg.declare(metadata, metadata, metadata) #1

llvm/test/Transforms/GVN/captured-before.ll

+1 −2
Original line number Diff line number Diff line
@@ -46,9 +46,8 @@ define i32 @test_store_before_capture(ptr %p) { 
; CHECK-NEXT:    store i32 123, ptr [[A]], align 4

; CHECK-NEXT:    [[P2:%.*]] = load ptr, ptr [[P]], align 8

; CHECK-NEXT:    store i32 42, ptr [[P2]], align 4

; CHECK-NEXT:    [[V:%.*]] = load i32, ptr [[A]], align 4

; CHECK-NEXT:    call void @capture(ptr [[A]])

; CHECK-NEXT:    ret i32 [[V]]

; CHECK-NEXT:    ret i32 123

;

  %a = alloca i32

  store i32 123, ptr %a