2010-10-20 07:09:08 +08:00
|
|
|
//===- BasicAliasAnalysis.cpp - Stateless Alias Analysis Impl -------------===//
|
2005-04-22 05:13:18 +08:00
|
|
|
//
|
2003-10-21 03:43:21 +08:00
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
2007-12-30 04:36:04 +08:00
|
|
|
// This file is distributed under the University of Illinois Open Source
|
|
|
|
// License. See LICENSE.TXT for details.
|
2005-04-22 05:13:18 +08:00
|
|
|
//
|
2003-10-21 03:43:21 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
2003-02-27 03:41:54 +08:00
|
|
|
//
|
2010-10-20 07:09:08 +08:00
|
|
|
// This file defines the primary stateless implementation of the
|
|
|
|
// Alias Analysis interface that implements identities (two different
|
|
|
|
// globals cannot alias, etc), but does no stateful analysis.
|
2003-02-27 03:41:54 +08:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2015-08-06 15:33:15 +08:00
|
|
|
#include "llvm/Analysis/BasicAliasAnalysis.h"
|
2012-12-04 00:50:05 +08:00
|
|
|
#include "llvm/ADT/SmallVector.h"
|
2015-08-06 07:40:30 +08:00
|
|
|
#include "llvm/ADT/Statistic.h"
|
2012-12-04 00:50:05 +08:00
|
|
|
#include "llvm/Analysis/AliasAnalysis.h"
|
2017-05-14 14:18:34 +08:00
|
|
|
#include "llvm/Analysis/AssumptionCache.h"
|
2014-01-03 13:47:03 +08:00
|
|
|
#include "llvm/Analysis/CFG.h"
|
2014-01-07 19:48:04 +08:00
|
|
|
#include "llvm/Analysis/CaptureTracking.h"
|
2012-12-04 00:50:05 +08:00
|
|
|
#include "llvm/Analysis/InstructionSimplify.h"
|
2014-01-03 13:47:03 +08:00
|
|
|
#include "llvm/Analysis/LoopInfo.h"
|
2012-12-04 00:50:05 +08:00
|
|
|
#include "llvm/Analysis/MemoryBuiltins.h"
|
|
|
|
#include "llvm/Analysis/ValueTracking.h"
|
2013-01-02 19:36:10 +08:00
|
|
|
#include "llvm/IR/Constants.h"
|
|
|
|
#include "llvm/IR/DataLayout.h"
|
|
|
|
#include "llvm/IR/DerivedTypes.h"
|
2014-01-13 17:26:24 +08:00
|
|
|
#include "llvm/IR/Dominators.h"
|
2013-01-02 19:36:10 +08:00
|
|
|
#include "llvm/IR/GlobalAlias.h"
|
|
|
|
#include "llvm/IR/GlobalVariable.h"
|
|
|
|
#include "llvm/IR/Instructions.h"
|
|
|
|
#include "llvm/IR/IntrinsicInst.h"
|
|
|
|
#include "llvm/IR/LLVMContext.h"
|
|
|
|
#include "llvm/IR/Operator.h"
|
2004-03-15 11:36:49 +08:00
|
|
|
#include "llvm/Pass.h"
|
2009-07-12 04:10:48 +08:00
|
|
|
#include "llvm/Support/ErrorHandling.h"
|
2017-05-15 14:39:41 +08:00
|
|
|
#include "llvm/Support/KnownBits.h"
|
2004-09-04 02:19:51 +08:00
|
|
|
#include <algorithm>
|
2016-01-30 10:42:11 +08:00
|
|
|
|
|
|
|
#define DEBUG_TYPE "basicaa"
|
|
|
|
|
2003-11-26 02:33:40 +08:00
|
|
|
using namespace llvm;
|
2003-11-12 06:41:34 +08:00
|
|
|
|
2015-07-16 03:32:22 +08:00
|
|
|
/// Enable analysis of recursive PHI nodes.
|
2015-08-06 15:57:58 +08:00
|
|
|
static cl::opt<bool> EnableRecPhiAnalysis("basicaa-recphi", cl::Hidden,
|
|
|
|
cl::init(false));
|
2015-08-06 07:40:30 +08:00
|
|
|
/// SearchLimitReached / SearchTimes shows how often the limit of
|
|
|
|
/// to decompose GEPs is reached. It will affect the precision
|
|
|
|
/// of basic alias analysis.
|
|
|
|
STATISTIC(SearchLimitReached, "Number of times the limit to "
|
|
|
|
"decompose GEPs is reached");
|
|
|
|
STATISTIC(SearchTimes, "Number of times a GEP is decomposed");
|
|
|
|
|
2014-01-02 11:31:36 +08:00
|
|
|
/// Cutoff after which to stop analysing a set of phi nodes potentially involved
|
2016-01-18 07:13:48 +08:00
|
|
|
/// in a cycle. Because we are analysing 'through' phi nodes, we need to be
|
2014-01-03 13:47:03 +08:00
|
|
|
/// careful with value equivalence. We use reachability to make sure a value
|
|
|
|
/// cannot be involved in a cycle.
|
|
|
|
const unsigned MaxNumPhiBBsValueReachabilityCheck = 20;
|
2014-01-02 11:31:36 +08:00
|
|
|
|
2014-03-27 05:30:19 +08:00
|
|
|
// The max limit of the search depth in DecomposeGEPExpression() and
|
|
|
|
// GetUnderlyingObject(), both functions need to use the same search
|
|
|
|
// depth otherwise the algorithm in aliasGEP will assert.
|
|
|
|
static const unsigned MaxLookupSearchDepth = 6;
|
|
|
|
|
2016-12-27 18:30:45 +08:00
|
|
|
bool BasicAAResult::invalidate(Function &F, const PreservedAnalyses &PA,
|
|
|
|
FunctionAnalysisManager::Invalidator &Inv) {
|
|
|
|
// We don't care if this analysis itself is preserved, it has no state. But
|
|
|
|
// we need to check that the analyses it depends on have been. Note that we
|
|
|
|
// may be created without handles to some analyses and in that case don't
|
|
|
|
// depend on them.
|
|
|
|
if (Inv.invalidate<AssumptionAnalysis>(F, PA) ||
|
|
|
|
(DT && Inv.invalidate<DominatorTreeAnalysis>(F, PA)) ||
|
|
|
|
(LI && Inv.invalidate<LoopAnalysis>(F, PA)))
|
|
|
|
return true;
|
|
|
|
|
|
|
|
// Otherwise this analysis result remains valid.
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2008-06-16 14:30:22 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// Useful predicates
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Returns true if the pointer is to a function-local object that never
|
|
|
|
/// escapes from the function.
|
2010-07-02 04:08:40 +08:00
|
|
|
static bool isNonEscapingLocalObject(const Value *V) {
|
2008-06-16 14:30:22 +08:00
|
|
|
// If this is a local allocation, check to see if it escapes.
|
2010-07-02 04:08:40 +08:00
|
|
|
if (isa<AllocaInst>(V) || isNoAliasCall(V))
|
2009-11-20 05:57:48 +08:00
|
|
|
// Set StoreCaptures to True so that we can assume in our callers that the
|
|
|
|
// pointer is not the result of a load instruction. Currently
|
|
|
|
// PointerMayBeCaptured doesn't have any special analysis for the
|
|
|
|
// StoreCaptures=false case; if it did, our callers could be refined to be
|
|
|
|
// more precise.
|
|
|
|
return !PointerMayBeCaptured(V, false, /*StoreCaptures=*/true);
|
2009-01-06 05:19:53 +08:00
|
|
|
|
2008-06-16 14:30:22 +08:00
|
|
|
// If this is an argument that corresponds to a byval or noalias argument,
|
2009-01-06 05:19:53 +08:00
|
|
|
// then it has not escaped before entering the function. Check if it escapes
|
|
|
|
// inside the function.
|
2010-07-02 04:08:40 +08:00
|
|
|
if (const Argument *A = dyn_cast<Argument>(V))
|
2012-11-05 18:48:24 +08:00
|
|
|
if (A->hasByValAttr() || A->hasNoAliasAttr())
|
2016-01-18 07:13:48 +08:00
|
|
|
// Note even if the argument is marked nocapture, we still need to check
|
2012-11-05 18:48:24 +08:00
|
|
|
// for copies made inside the function. The nocapture attribute only
|
|
|
|
// specifies that there are no copies made that outlive the function.
|
2010-07-02 04:08:40 +08:00
|
|
|
return !PointerMayBeCaptured(V, false, /*StoreCaptures=*/true);
|
2012-11-05 18:48:24 +08:00
|
|
|
|
2008-06-16 14:30:22 +08:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Returns true if the pointer is one which would have been considered an
|
|
|
|
/// escape by isNonEscapingLocalObject.
|
2010-07-02 04:08:40 +08:00
|
|
|
static bool isEscapeSource(const Value *V) {
|
|
|
|
if (isa<CallInst>(V) || isa<InvokeInst>(V) || isa<Argument>(V))
|
|
|
|
return true;
|
2010-06-29 08:50:39 +08:00
|
|
|
|
|
|
|
// The load case works because isNonEscapingLocalObject considers all
|
|
|
|
// stores to be escapes (it passes true for the StoreCaptures argument
|
|
|
|
// to PointerMayBeCaptured).
|
|
|
|
if (isa<LoadInst>(V))
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
2008-06-16 14:30:22 +08:00
|
|
|
|
2016-01-18 07:13:48 +08:00
|
|
|
/// Returns the size of the object specified by V or UnknownSize if unknown.
|
2014-02-22 02:34:28 +08:00
|
|
|
static uint64_t getObjectSize(const Value *V, const DataLayout &DL,
|
2012-08-29 23:32:21 +08:00
|
|
|
const TargetLibraryInfo &TLI,
|
2012-02-28 04:46:07 +08:00
|
|
|
bool RoundToAlign = false) {
|
2012-06-21 23:45:28 +08:00
|
|
|
uint64_t Size;
|
2017-03-22 04:08:59 +08:00
|
|
|
ObjectSizeOpts Opts;
|
|
|
|
Opts.RoundToAlign = RoundToAlign;
|
|
|
|
if (getObjectSize(V, Size, DL, &TLI, Opts))
|
2012-06-21 23:45:28 +08:00
|
|
|
return Size;
|
2015-06-17 15:21:38 +08:00
|
|
|
return MemoryLocation::UnknownSize;
|
2011-01-19 05:16:06 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Returns true if we can prove that the object specified by V is smaller than
|
|
|
|
/// Size.
|
2011-01-19 05:16:06 +08:00
|
|
|
static bool isObjectSmallerThan(const Value *V, uint64_t Size,
|
2014-02-22 02:34:28 +08:00
|
|
|
const DataLayout &DL,
|
2012-08-29 23:32:21 +08:00
|
|
|
const TargetLibraryInfo &TLI) {
|
2013-04-10 02:16:05 +08:00
|
|
|
// Note that the meanings of the "object" are slightly different in the
|
|
|
|
// following contexts:
|
|
|
|
// c1: llvm::getObjectSize()
|
|
|
|
// c2: llvm.objectsize() intrinsic
|
|
|
|
// c3: isObjectSmallerThan()
|
|
|
|
// c1 and c2 share the same meaning; however, the meaning of "object" in c3
|
|
|
|
// refers to the "entire object".
|
|
|
|
//
|
|
|
|
// Consider this example:
|
|
|
|
// char *p = (char*)malloc(100)
|
|
|
|
// char *q = p+80;
|
|
|
|
//
|
|
|
|
// In the context of c1 and c2, the "object" pointed by q refers to the
|
|
|
|
// stretch of memory of q[0:19]. So, getObjectSize(q) should return 20.
|
|
|
|
//
|
|
|
|
// However, in the context of c3, the "object" refers to the chunk of memory
|
|
|
|
// being allocated. So, the "object" has 100 bytes, and q points to the middle
|
|
|
|
// the "object". In case q is passed to isObjectSmallerThan() as the 1st
|
|
|
|
// parameter, before the llvm::getObjectSize() is called to get the size of
|
|
|
|
// entire object, we should:
|
|
|
|
// - either rewind the pointer q to the base-address of the object in
|
|
|
|
// question (in this case rewind to p), or
|
|
|
|
// - just give up. It is up to caller to make sure the pointer is pointing
|
|
|
|
// to the base address the object.
|
2013-08-24 22:16:00 +08:00
|
|
|
//
|
2013-04-10 02:16:05 +08:00
|
|
|
// We go for 2nd option for simplicity.
|
|
|
|
if (!isIdentifiedObject(V))
|
|
|
|
return false;
|
|
|
|
|
2012-02-28 04:46:07 +08:00
|
|
|
// This function needs to use the aligned object size because we allow
|
|
|
|
// reads a bit past the end given sufficient alignment.
|
2015-08-06 15:57:58 +08:00
|
|
|
uint64_t ObjectSize = getObjectSize(V, DL, TLI, /*RoundToAlign*/ true);
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2015-06-17 15:21:38 +08:00
|
|
|
return ObjectSize != MemoryLocation::UnknownSize && ObjectSize < Size;
|
2011-01-19 05:16:06 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Returns true if we can prove that the object specified by V has size Size.
|
2015-08-06 15:57:58 +08:00
|
|
|
static bool isObjectSize(const Value *V, uint64_t Size, const DataLayout &DL,
|
|
|
|
const TargetLibraryInfo &TLI) {
|
2014-02-22 02:34:28 +08:00
|
|
|
uint64_t ObjectSize = getObjectSize(V, DL, TLI);
|
2015-06-17 15:21:38 +08:00
|
|
|
return ObjectSize != MemoryLocation::UnknownSize && ObjectSize == Size;
|
2008-06-16 14:30:22 +08:00
|
|
|
}
|
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// GetElementPtr Instruction Decomposition and Analysis
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Analyzes the specified value as a linear expression: "A*V + B", where A and
|
|
|
|
/// B are constant integers.
|
|
|
|
///
|
|
|
|
/// Returns the scale and offset values as APInts and return V as a Value*, and
|
|
|
|
/// return whether we looked through any sign or zero extends. The incoming
|
2016-01-18 07:13:48 +08:00
|
|
|
/// Value is known to have IntegerType, and it may already be sign or zero
|
2015-08-06 16:17:06 +08:00
|
|
|
/// extended.
|
2010-08-19 07:09:49 +08:00
|
|
|
///
|
|
|
|
/// Note that this looks through extends, so the high bits may not be
|
|
|
|
/// represented in the result.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
/*static*/ const Value *BasicAAResult::GetLinearExpression(
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
const Value *V, APInt &Scale, APInt &Offset, unsigned &ZExtBits,
|
|
|
|
unsigned &SExtBits, const DataLayout &DL, unsigned Depth,
|
2016-12-19 16:22:17 +08:00
|
|
|
AssumptionCache *AC, DominatorTree *DT, bool &NSW, bool &NUW) {
|
2010-08-19 06:07:29 +08:00
|
|
|
assert(V->getType()->isIntegerTy() && "Not an integer value");
|
|
|
|
|
|
|
|
// Limit our recursion depth.
|
|
|
|
if (Depth == 6) {
|
|
|
|
Scale = 1;
|
|
|
|
Offset = 0;
|
|
|
|
return V;
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
if (const ConstantInt *Const = dyn_cast<ConstantInt>(V)) {
|
2016-01-18 07:13:48 +08:00
|
|
|
// If it's a constant, just convert it to an offset and remove the variable.
|
|
|
|
// If we've been called recursively, the Offset bit width will be greater
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
// than the constant's (the Offset's always as wide as the outermost call),
|
|
|
|
// so we'll zext here and process any extension in the isa<SExtInst> &
|
|
|
|
// isa<ZExtInst> cases below.
|
|
|
|
Offset += Const->getValue().zextOrSelf(Offset.getBitWidth());
|
|
|
|
assert(Scale == 0 && "Constant values don't have a scale");
|
|
|
|
return V;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (const BinaryOperator *BOp = dyn_cast<BinaryOperator>(V)) {
|
2010-08-19 06:07:29 +08:00
|
|
|
if (ConstantInt *RHSC = dyn_cast<ConstantInt>(BOp->getOperand(1))) {
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
2016-01-18 07:13:48 +08:00
|
|
|
// If we've been called recursively, then Offset and Scale will be wider
|
|
|
|
// than the BOp operands. We'll always zext it here as we'll process sign
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
// extensions below (see the isa<SExtInst> / isa<ZExtInst> cases).
|
|
|
|
APInt RHS = RHSC->getValue().zextOrSelf(Offset.getBitWidth());
|
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
switch (BOp->getOpcode()) {
|
2015-08-06 15:57:58 +08:00
|
|
|
default:
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
// We don't understand this instruction, so we can't decompose it any
|
|
|
|
// further.
|
|
|
|
Scale = 1;
|
|
|
|
Offset = 0;
|
|
|
|
return V;
|
2010-08-19 06:07:29 +08:00
|
|
|
case Instruction::Or:
|
|
|
|
// X|C == X+C if all the bits in C are unset in X. Otherwise we can't
|
|
|
|
// analyze it.
|
2016-12-19 16:22:17 +08:00
|
|
|
if (!MaskedValueIsZero(BOp->getOperand(0), RHSC->getValue(), DL, 0, AC,
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
BOp, DT)) {
|
|
|
|
Scale = 1;
|
|
|
|
Offset = 0;
|
|
|
|
return V;
|
|
|
|
}
|
2016-08-18 04:30:52 +08:00
|
|
|
LLVM_FALLTHROUGH;
|
2010-08-19 06:07:29 +08:00
|
|
|
case Instruction::Add:
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, ZExtBits,
|
2016-12-19 16:22:17 +08:00
|
|
|
SExtBits, DL, Depth + 1, AC, DT, NSW, NUW);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
Offset += RHS;
|
|
|
|
break;
|
|
|
|
case Instruction::Sub:
|
|
|
|
V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, ZExtBits,
|
2016-12-19 16:22:17 +08:00
|
|
|
SExtBits, DL, Depth + 1, AC, DT, NSW, NUW);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
Offset -= RHS;
|
|
|
|
break;
|
2010-08-19 06:07:29 +08:00
|
|
|
case Instruction::Mul:
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, ZExtBits,
|
2016-12-19 16:22:17 +08:00
|
|
|
SExtBits, DL, Depth + 1, AC, DT, NSW, NUW);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
Offset *= RHS;
|
|
|
|
Scale *= RHS;
|
|
|
|
break;
|
2010-08-19 06:07:29 +08:00
|
|
|
case Instruction::Shl:
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, ZExtBits,
|
2016-12-19 16:22:17 +08:00
|
|
|
SExtBits, DL, Depth + 1, AC, DT, NSW, NUW);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
Offset <<= RHS.getLimitedValue();
|
|
|
|
Scale <<= RHS.getLimitedValue();
|
|
|
|
// the semantics of nsw and nuw for left shifts don't match those of
|
|
|
|
// multiplications, so we won't propagate them.
|
|
|
|
NSW = NUW = false;
|
2010-08-19 06:07:29 +08:00
|
|
|
return V;
|
|
|
|
}
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
|
|
|
if (isa<OverflowingBinaryOperator>(BOp)) {
|
|
|
|
NUW &= BOp->hasNoUnsignedWrap();
|
|
|
|
NSW &= BOp->hasNoSignedWrap();
|
|
|
|
}
|
|
|
|
return V;
|
2010-08-19 06:07:29 +08:00
|
|
|
}
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
// Since GEP indices are sign extended anyway, we don't care about the high
|
2010-08-19 07:09:49 +08:00
|
|
|
// bits of a sign or zero extended value - just scales and offsets. The
|
|
|
|
// extensions have to be consistent though.
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
if (isa<SExtInst>(V) || isa<ZExtInst>(V)) {
|
2010-08-19 06:07:29 +08:00
|
|
|
Value *CastOp = cast<CastInst>(V)->getOperand(0);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
unsigned NewWidth = V->getType()->getPrimitiveSizeInBits();
|
2010-08-19 06:07:29 +08:00
|
|
|
unsigned SmallWidth = CastOp->getType()->getPrimitiveSizeInBits();
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
unsigned OldZExtBits = ZExtBits, OldSExtBits = SExtBits;
|
|
|
|
const Value *Result =
|
|
|
|
GetLinearExpression(CastOp, Scale, Offset, ZExtBits, SExtBits, DL,
|
2016-12-19 16:22:17 +08:00
|
|
|
Depth + 1, AC, DT, NSW, NUW);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
2016-11-20 21:47:59 +08:00
|
|
|
// zext(zext(%x)) == zext(%x), and similarly for sext; we'll handle this
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
// by just incrementing the number of bits we've extended by.
|
|
|
|
unsigned ExtendedBy = NewWidth - SmallWidth;
|
|
|
|
|
|
|
|
if (isa<SExtInst>(V) && ZExtBits == 0) {
|
|
|
|
// sext(sext(%x, a), b) == sext(%x, a + b)
|
|
|
|
|
|
|
|
if (NSW) {
|
|
|
|
// We haven't sign-wrapped, so it's valid to decompose sext(%x + c)
|
|
|
|
// into sext(%x) + sext(c). We'll sext the Offset ourselves:
|
|
|
|
unsigned OldWidth = Offset.getBitWidth();
|
|
|
|
Offset = Offset.trunc(SmallWidth).sext(NewWidth).zextOrSelf(OldWidth);
|
|
|
|
} else {
|
|
|
|
// We may have signed-wrapped, so don't decompose sext(%x + c) into
|
|
|
|
// sext(%x) + sext(c)
|
|
|
|
Scale = 1;
|
|
|
|
Offset = 0;
|
|
|
|
Result = CastOp;
|
|
|
|
ZExtBits = OldZExtBits;
|
|
|
|
SExtBits = OldSExtBits;
|
|
|
|
}
|
|
|
|
SExtBits += ExtendedBy;
|
|
|
|
} else {
|
|
|
|
// sext(zext(%x, a), b) = zext(zext(%x, a), b) = zext(%x, a + b)
|
|
|
|
|
|
|
|
if (!NUW) {
|
|
|
|
// We may have unsigned-wrapped, so don't decompose zext(%x + c) into
|
|
|
|
// zext(%x) + zext(c)
|
|
|
|
Scale = 1;
|
|
|
|
Offset = 0;
|
|
|
|
Result = CastOp;
|
|
|
|
ZExtBits = OldZExtBits;
|
|
|
|
SExtBits = OldSExtBits;
|
|
|
|
}
|
|
|
|
ZExtBits += ExtendedBy;
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
return Result;
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
Scale = 1;
|
|
|
|
Offset = 0;
|
|
|
|
return V;
|
|
|
|
}
|
|
|
|
|
2016-01-30 10:42:11 +08:00
|
|
|
/// To ensure a pointer offset fits in an integer of size PointerSize
|
|
|
|
/// (in bits) when that size is smaller than 64. This is an issue in
|
|
|
|
/// particular for 32b programs with negative indices that rely on two's
|
2016-01-30 13:58:38 +08:00
|
|
|
/// complement wrap-arounds for precise alias information.
|
2016-01-30 10:42:11 +08:00
|
|
|
static int64_t adjustToPointerSize(int64_t Offset, unsigned PointerSize) {
|
|
|
|
assert(PointerSize <= 64 && "Invalid PointerSize!");
|
|
|
|
unsigned ShiftBits = 64 - PointerSize;
|
2016-01-30 13:35:09 +08:00
|
|
|
return (int64_t)((uint64_t)Offset << ShiftBits) >> ShiftBits;
|
2016-01-30 10:42:11 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// If V is a symbolic pointer expression, decompose it into a base pointer
|
|
|
|
/// with a constant offset and a number of scaled symbolic offsets.
|
2010-08-19 06:07:29 +08:00
|
|
|
///
|
2015-08-06 16:17:06 +08:00
|
|
|
/// The scaled symbolic offsets (represented by pairs of a Value* and a scale
|
|
|
|
/// in the VarIndices vector) are Value*'s that are known to be scaled by the
|
|
|
|
/// specified amount, but which may have other unrepresented high bits. As
|
|
|
|
/// such, the gep cannot necessarily be reconstructed from its decomposed form.
|
2010-08-19 06:07:29 +08:00
|
|
|
///
|
2012-10-09 00:38:25 +08:00
|
|
|
/// When DataLayout is around, this function is capable of analyzing everything
|
2014-03-27 05:30:19 +08:00
|
|
|
/// that GetUnderlyingObject can look through. To be able to do that
|
|
|
|
/// GetUnderlyingObject and DecomposeGEPExpression must use the same search
|
2015-08-06 16:17:06 +08:00
|
|
|
/// depth (MaxLookupSearchDepth). When DataLayout not is around, it just looks
|
|
|
|
/// through pointer casts.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
bool BasicAAResult::DecomposeGEPExpression(const Value *V,
|
2016-12-19 16:22:17 +08:00
|
|
|
DecomposedGEP &Decomposed, const DataLayout &DL, AssumptionCache *AC,
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
DominatorTree *DT) {
|
2010-08-19 06:07:29 +08:00
|
|
|
// Limit recursion depth to limit compile time in crazy cases.
|
2014-03-27 05:30:19 +08:00
|
|
|
unsigned MaxLookup = MaxLookupSearchDepth;
|
2015-08-06 07:40:30 +08:00
|
|
|
SearchTimes++;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.StructOffset = 0;
|
|
|
|
Decomposed.OtherOffset = 0;
|
|
|
|
Decomposed.VarIndices.clear();
|
2010-08-19 06:07:29 +08:00
|
|
|
do {
|
|
|
|
// See if this is a bitcast or GEP.
|
|
|
|
const Operator *Op = dyn_cast<Operator>(V);
|
2014-04-15 12:59:12 +08:00
|
|
|
if (!Op) {
|
2010-08-19 06:07:29 +08:00
|
|
|
// The only non-operator case we can handle are GlobalAliases.
|
|
|
|
if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
|
Don't IPO over functions that can be de-refined
Summary:
Fixes PR26774.
If you're aware of the issue, feel free to skip the "Motivation"
section and jump directly to "This patch".
Motivation:
I define "refinement" as discarding behaviors from a program that the
optimizer has license to discard. So transforming:
```
void f(unsigned x) {
unsigned t = 5 / x;
(void)t;
}
```
to
```
void f(unsigned x) { }
```
is refinement, since the behavior went from "if x == 0 then undefined
else nothing" to "nothing" (the optimizer has license to discard
undefined behavior).
Refinement is a fundamental aspect of many mid-level optimizations done
by LLVM. For instance, transforming `x == (x + 1)` to `false` also
involves refinement since the expression's value went from "if x is
`undef` then { `true` or `false` } else { `false` }" to "`false`" (by
definition, the optimizer has license to fold `undef` to any non-`undef`
value).
Unfortunately, refinement implies that the optimizer cannot assume
that the implementation of a function it can see has all of the
behavior an unoptimized or a differently optimized version of the same
function can have. This is a problem for functions with comdat
linkage, where a function can be replaced by an unoptimized or a
differently optimized version of the same source level function.
For instance, FunctionAttrs cannot assume a comdat function is
actually `readnone` even if it does not have any loads or stores in
it; since there may have been loads and stores in the "original
function" that were refined out in the currently visible variant, and
at the link step the linker may in fact choose an implementation with
a load or a store. As an example, consider a function that does two
atomic loads from the same memory location, and writes to memory only
if the two values are not equal. The optimizer is allowed to refine
this function by first CSE'ing the two loads, and the folding the
comparision to always report that the two values are equal. Such a
refined variant will look like it is `readonly`. However, the
unoptimized version of the function can still write to memory (since
the two loads //can// result in different values), and selecting the
unoptimized version at link time will retroactively invalidate
transforms we may have done under the assumption that the function
does not write to memory.
Note: this is not just a problem with atomics or with linking
differently optimized object files. See PR26774 for more realistic
examples that involved neither.
This patch:
This change introduces a new set of linkage types, predicated as
`GlobalValue::mayBeDerefined` that returns true if the linkage type
allows a function to be replaced by a differently optimized variant at
link time. It then changes a set of IPO passes to bail out if they see
such a function.
Reviewers: chandlerc, hfinkel, dexonsmith, joker.eph, rnk
Subscribers: mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D18634
llvm-svn: 265762
2016-04-08 08:48:30 +08:00
|
|
|
if (!GA->isInterposable()) {
|
2010-08-19 06:07:29 +08:00
|
|
|
V = GA->getAliasee();
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.Base = V;
|
|
|
|
return false;
|
2010-08-19 06:07:29 +08:00
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2014-07-15 08:56:40 +08:00
|
|
|
if (Op->getOpcode() == Instruction::BitCast ||
|
|
|
|
Op->getOpcode() == Instruction::AddrSpaceCast) {
|
2010-08-19 06:07:29 +08:00
|
|
|
V = Op->getOperand(0);
|
|
|
|
continue;
|
|
|
|
}
|
2010-12-16 04:49:55 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
const GEPOperator *GEPOp = dyn_cast<GEPOperator>(Op);
|
2014-04-15 12:59:12 +08:00
|
|
|
if (!GEPOp) {
|
2016-07-11 09:32:20 +08:00
|
|
|
if (auto CS = ImmutableCallSite(V))
|
|
|
|
if (const Value *RV = CS.getReturnedArgOperand()) {
|
|
|
|
V = RV;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2011-05-25 02:24:08 +08:00
|
|
|
// If it's not a GEP, hand it off to SimplifyInstruction to see if it
|
|
|
|
// can come up with something. This matches what GetUnderlyingObject does.
|
|
|
|
if (const Instruction *I = dyn_cast<Instruction>(V))
|
2016-12-19 16:22:17 +08:00
|
|
|
// TODO: Get a DominatorTree and AssumptionCache and use them here
|
|
|
|
// (these are both now available in this function, but this should be
|
|
|
|
// updated when GetUnderlyingObject is updated). TLI should be
|
|
|
|
// provided also.
|
2011-05-25 02:24:08 +08:00
|
|
|
if (const Value *Simplified =
|
2015-08-06 15:57:58 +08:00
|
|
|
SimplifyInstruction(const_cast<Instruction *>(I), DL)) {
|
2011-05-25 02:24:08 +08:00
|
|
|
V = Simplified;
|
|
|
|
continue;
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.Base = V;
|
|
|
|
return false;
|
2011-05-25 02:24:08 +08:00
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
// Don't attempt to analyze GEPs over unsized objects.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
if (!GEPOp->getSourceElementType()->isSized()) {
|
|
|
|
Decomposed.Base = V;
|
|
|
|
return false;
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2013-11-16 08:36:43 +08:00
|
|
|
unsigned AS = GEPOp->getPointerAddressSpace();
|
2010-08-19 06:07:29 +08:00
|
|
|
// Walk the indices of the GEP, accumulating them into BaseOff/VarIndices.
|
|
|
|
gep_type_iterator GTI = gep_type_begin(GEPOp);
|
2016-01-30 10:42:11 +08:00
|
|
|
unsigned PointerSize = DL.getPointerSizeInBits(AS);
|
2016-10-22 10:41:39 +08:00
|
|
|
// Assume all GEP operands are constants until proven otherwise.
|
|
|
|
bool GepHasConstantOffset = true;
|
2015-08-06 15:57:58 +08:00
|
|
|
for (User::const_op_iterator I = GEPOp->op_begin() + 1, E = GEPOp->op_end();
|
2016-12-02 10:24:42 +08:00
|
|
|
I != E; ++I, ++GTI) {
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
const Value *Index = *I;
|
2010-08-19 06:07:29 +08:00
|
|
|
// Compute the (potentially symbolic) offset in bytes for this index.
|
2016-12-02 10:24:42 +08:00
|
|
|
if (StructType *STy = GTI.getStructTypeOrNull()) {
|
2010-08-19 06:07:29 +08:00
|
|
|
// For a struct, add the member offset.
|
|
|
|
unsigned FieldNo = cast<ConstantInt>(Index)->getZExtValue();
|
2015-08-06 15:57:58 +08:00
|
|
|
if (FieldNo == 0)
|
|
|
|
continue;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.StructOffset +=
|
|
|
|
DL.getStructLayout(STy)->getElementOffset(FieldNo);
|
2010-08-19 06:07:29 +08:00
|
|
|
continue;
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
// For an array/pointer, add the element offset, explicitly scaled.
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
if (const ConstantInt *CIdx = dyn_cast<ConstantInt>(Index)) {
|
2015-08-06 15:57:58 +08:00
|
|
|
if (CIdx->isZero())
|
|
|
|
continue;
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.OtherOffset +=
|
2016-12-02 10:24:42 +08:00
|
|
|
DL.getTypeAllocSize(GTI.getIndexedType()) * CIdx->getSExtValue();
|
2010-08-19 06:07:29 +08:00
|
|
|
continue;
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2016-10-22 10:41:39 +08:00
|
|
|
GepHasConstantOffset = false;
|
|
|
|
|
2016-12-02 10:24:42 +08:00
|
|
|
uint64_t Scale = DL.getTypeAllocSize(GTI.getIndexedType());
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
unsigned ZExtBits = 0, SExtBits = 0;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 07:09:49 +08:00
|
|
|
// If the integer type is smaller than the pointer size, it is implicitly
|
|
|
|
// sign extended to pointer size.
|
2013-09-28 06:18:51 +08:00
|
|
|
unsigned Width = Index->getType()->getIntegerBitWidth();
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
if (PointerSize > Width)
|
|
|
|
SExtBits += PointerSize - Width;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 07:09:49 +08:00
|
|
|
// Use GetLinearExpression to decompose the index into a C1*V+C2 form.
|
2010-08-19 06:07:29 +08:00
|
|
|
APInt IndexScale(Width, 0), IndexOffset(Width, 0);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
bool NSW = true, NUW = true;
|
|
|
|
Index = GetLinearExpression(Index, IndexScale, IndexOffset, ZExtBits,
|
2016-12-19 16:22:17 +08:00
|
|
|
SExtBits, DL, 0, AC, DT, NSW, NUW);
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
// The GEP index scale ("Scale") scales C1*V+C2, yielding (C1*V+C2)*Scale.
|
|
|
|
// This gives us an aggregate computation of (C1*Scale)*V + C2*Scale.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.OtherOffset += IndexOffset.getSExtValue() * Scale;
|
2010-09-16 04:08:03 +08:00
|
|
|
Scale *= IndexScale.getSExtValue();
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2011-04-15 13:18:47 +08:00
|
|
|
// If we already had an occurrence of this index variable, merge this
|
2010-08-19 06:07:29 +08:00
|
|
|
// scale into it. For example, we want to handle:
|
|
|
|
// A[x][x] -> x*16 + x*4 -> x*20
|
|
|
|
// This also ensures that 'x' only appears in the index list once.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
for (unsigned i = 0, e = Decomposed.VarIndices.size(); i != e; ++i) {
|
2016-10-22 10:41:39 +08:00
|
|
|
if (Decomposed.VarIndices[i].V == Index &&
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.VarIndices[i].ZExtBits == ZExtBits &&
|
|
|
|
Decomposed.VarIndices[i].SExtBits == SExtBits) {
|
|
|
|
Scale += Decomposed.VarIndices[i].Scale;
|
|
|
|
Decomposed.VarIndices.erase(Decomposed.VarIndices.begin() + i);
|
2010-08-19 06:07:29 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
// Make sure that we have a scale that makes sense for this target's
|
|
|
|
// pointer size.
|
2016-01-30 10:42:11 +08:00
|
|
|
Scale = adjustToPointerSize(Scale, PointerSize);
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:47:56 +08:00
|
|
|
if (Scale) {
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
VariableGEPIndex Entry = {Index, ZExtBits, SExtBits,
|
2011-07-27 14:22:51 +08:00
|
|
|
static_cast<int64_t>(Scale)};
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.VarIndices.push_back(Entry);
|
2010-08-19 06:47:56 +08:00
|
|
|
}
|
2010-08-19 06:07:29 +08:00
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2016-01-30 13:52:53 +08:00
|
|
|
// Take care of wrap-arounds
|
2016-10-22 10:41:39 +08:00
|
|
|
if (GepHasConstantOffset) {
|
|
|
|
Decomposed.StructOffset =
|
|
|
|
adjustToPointerSize(Decomposed.StructOffset, PointerSize);
|
|
|
|
Decomposed.OtherOffset =
|
|
|
|
adjustToPointerSize(Decomposed.OtherOffset, PointerSize);
|
|
|
|
}
|
2016-01-30 13:52:53 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
// Analyze the base pointer next.
|
|
|
|
V = GEPOp->getOperand(0);
|
|
|
|
} while (--MaxLookup);
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-08-19 06:07:29 +08:00
|
|
|
// If the chain of expressions is too deep, just return early.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Decomposed.Base = V;
|
2015-08-06 07:40:30 +08:00
|
|
|
SearchLimitReached++;
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
return true;
|
2010-08-19 06:07:29 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Returns whether the given pointer value points to memory that is local to
|
|
|
|
/// the function, with global constants being considered local to all
|
|
|
|
/// functions.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
bool BasicAAResult::pointsToConstantMemory(const MemoryLocation &Loc,
|
|
|
|
bool OrLocal) {
|
2010-11-09 00:45:26 +08:00
|
|
|
assert(Visited.empty() && "Visited must be cleared after use!");
|
2003-12-12 07:20:16 +08:00
|
|
|
|
2010-11-09 04:26:19 +08:00
|
|
|
unsigned MaxLookup = 8;
|
2010-11-09 00:45:26 +08:00
|
|
|
SmallVector<const Value *, 16> Worklist;
|
|
|
|
Worklist.push_back(Loc.Ptr);
|
|
|
|
do {
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
const Value *V = GetUnderlyingObject(Worklist.pop_back_val(), DL);
|
2014-11-19 15:49:26 +08:00
|
|
|
if (!Visited.insert(V).second) {
|
2010-11-09 00:45:26 +08:00
|
|
|
Visited.clear();
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
return AAResultBase::pointsToConstantMemory(Loc, OrLocal);
|
2010-11-09 00:45:26 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// An alloca instruction defines local memory.
|
|
|
|
if (OrLocal && isa<AllocaInst>(V))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
// A global constant counts as local memory for our purposes.
|
|
|
|
if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
|
|
|
|
// Note: this doesn't require GV to be "ODR" because it isn't legal for a
|
|
|
|
// global to be marked constant in some modules and non-constant in
|
|
|
|
// others. GV may even be a declaration, not a definition.
|
|
|
|
if (!GV->isConstant()) {
|
|
|
|
Visited.clear();
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
return AAResultBase::pointsToConstantMemory(Loc, OrLocal);
|
2010-11-09 00:45:26 +08:00
|
|
|
}
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
// If both select values point to local memory, then so does the select.
|
|
|
|
if (const SelectInst *SI = dyn_cast<SelectInst>(V)) {
|
|
|
|
Worklist.push_back(SI->getTrueValue());
|
|
|
|
Worklist.push_back(SI->getFalseValue());
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
// If all values incoming to a phi node point to local memory, then so does
|
|
|
|
// the phi.
|
|
|
|
if (const PHINode *PN = dyn_cast<PHINode>(V)) {
|
2010-11-09 04:26:19 +08:00
|
|
|
// Don't bother inspecting phi nodes with many operands.
|
|
|
|
if (PN->getNumIncomingValues() > MaxLookup) {
|
|
|
|
Visited.clear();
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
return AAResultBase::pointsToConstantMemory(Loc, OrLocal);
|
2010-11-09 04:26:19 +08:00
|
|
|
}
|
2015-05-13 04:05:31 +08:00
|
|
|
for (Value *IncValue : PN->incoming_values())
|
|
|
|
Worklist.push_back(IncValue);
|
2010-11-09 00:45:26 +08:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Otherwise be conservative.
|
|
|
|
Visited.clear();
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
return AAResultBase::pointsToConstantMemory(Loc, OrLocal);
|
2010-11-09 00:45:26 +08:00
|
|
|
|
2010-11-09 04:26:19 +08:00
|
|
|
} while (!Worklist.empty() && --MaxLookup);
|
2010-08-06 09:25:49 +08:00
|
|
|
|
2010-11-09 00:45:26 +08:00
|
|
|
Visited.clear();
|
2010-11-09 04:26:19 +08:00
|
|
|
return Worklist.empty();
|
2004-01-31 06:17:24 +08:00
|
|
|
}
|
2003-12-12 07:20:16 +08:00
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Returns the behavior when calling the given call site.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
FunctionModRefBehavior BasicAAResult::getModRefBehavior(ImmutableCallSite CS) {
|
2010-08-06 09:25:49 +08:00
|
|
|
if (CS.doesNotAccessMemory())
|
|
|
|
// Can't do better than this.
|
2015-07-23 07:15:57 +08:00
|
|
|
return FMRB_DoesNotAccessMemory;
|
2010-08-06 09:25:49 +08:00
|
|
|
|
2015-07-23 07:15:57 +08:00
|
|
|
FunctionModRefBehavior Min = FMRB_UnknownModRefBehavior;
|
2010-08-06 09:25:49 +08:00
|
|
|
|
|
|
|
// If the callsite knows it only reads memory, don't return worse
|
|
|
|
// than that.
|
|
|
|
if (CS.onlyReadsMemory())
|
2015-07-23 07:15:57 +08:00
|
|
|
Min = FMRB_OnlyReadsMemory;
|
2016-07-04 16:01:29 +08:00
|
|
|
else if (CS.doesNotReadMemory())
|
|
|
|
Min = FMRB_DoesNotReadMemory;
|
2010-08-06 09:25:49 +08:00
|
|
|
|
2015-07-11 18:30:36 +08:00
|
|
|
if (CS.onlyAccessesArgMemory())
|
2015-07-23 07:15:57 +08:00
|
|
|
Min = FunctionModRefBehavior(Min & FMRB_OnlyAccessesArgumentPointees);
|
2015-07-11 18:30:36 +08:00
|
|
|
|
[AA] Hoist the logic to reformulate various AA queries in terms of other
parts of the AA interface out of the base class of every single AA
result object.
Because this logic reformulates the query in terms of some other aspect
of the API, it would easily cause O(n^2) query patterns in alias
analysis. These could in turn be magnified further based on the number
of call arguments, and then further based on the number of AA queries
made for a particular call. This ended up causing problems for Rust that
were actually noticable enough to get a bug (PR26564) and probably other
places as well.
When originally re-working the AA infrastructure, the desire was to
regularize the pattern of refinement without losing any generality.
While I think it was successful, that is clearly proving to be too
costly. And the cost is needless: we gain no actual improvement for this
generality of making a direct query to tbaa actually be able to
re-use some other alias analysis's refinement logic for one of the other
APIs, or some such. In short, this is entirely wasted work.
To the extent possible, delegation to other API surfaces should be done
at the aggregation layer so that we can avoid re-walking the
aggregation. In fact, this significantly simplifies the logic as we no
longer need to smuggle the aggregation layer into each alias analysis
(or the TargetLibraryInfo into each alias analysis just so we can form
argument memory locations!).
However, we also have some delegation logic inside of BasicAA and some
of it even makes sense. When the delegation logic is baking in specific
knowledge of aliasing properties of the LLVM IR, as opposed to simply
reformulating the query to utilize a different alias analysis interface
entry point, it makes a lot of sense to restrict that logic to
a different layer such as BasicAA. So one aspect of the delegation that
was in every AA base class is that when we don't have operand bundles,
we re-use function AA results as a fallback for callsite alias results.
This relies on the IR properties of calls and functions w.r.t. aliasing,
and so seems a better fit to BasicAA. I've lifted the logic up to that
point where it seems to be a natural fit. This still does a bit of
redundant work (we query function attributes twice, once via the
callsite and once via the function AA query) but it is *exactly* twice
here, no more.
The end result is that all of the delegation logic is hoisted out of the
base class and into either the aggregation layer when it is a pure
retargeting to a different API surface, or into BasicAA when it relies
on the IR's aliasing properties. This should fix the quadratic query
pattern reported in PR26564, although I don't have a stand-alone test
case to reproduce it.
It also seems general goodness. Now the numerous AAs that don't need
target library info don't carry it around and depend on it. I think
I can even rip out the general access to the aggregation layer and only
expose that in BasicAA as it is the only place where we re-query in that
manner.
However, this is a non-trivial change to the AA infrastructure so I want
to get some additional eyes on this before it lands. Sadly, it can't
wait long because we should really cherry pick this into 3.8 if we're
going to go this route.
Differential Revision: http://reviews.llvm.org/D17329
llvm-svn: 262490
2016-03-02 23:56:53 +08:00
|
|
|
// If CS has operand bundles then aliasing attributes from the function it
|
|
|
|
// calls do not directly apply to the CallSite. This can be made more
|
|
|
|
// precise in the future.
|
|
|
|
if (!CS.hasOperandBundles())
|
|
|
|
if (const Function *F = CS.getCalledFunction())
|
|
|
|
Min =
|
|
|
|
FunctionModRefBehavior(Min & getBestAAResults().getModRefBehavior(F));
|
|
|
|
|
|
|
|
return Min;
|
2010-08-06 09:25:49 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Returns the behavior when calling the given function. For use when the call
|
|
|
|
/// site is not known.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
FunctionModRefBehavior BasicAAResult::getModRefBehavior(const Function *F) {
|
2010-11-09 00:08:43 +08:00
|
|
|
// If the function declares it doesn't access memory, we can't do better.
|
2010-08-06 09:25:49 +08:00
|
|
|
if (F->doesNotAccessMemory())
|
2015-07-23 07:15:57 +08:00
|
|
|
return FMRB_DoesNotAccessMemory;
|
2010-11-09 00:08:43 +08:00
|
|
|
|
2015-07-23 07:15:57 +08:00
|
|
|
FunctionModRefBehavior Min = FMRB_UnknownModRefBehavior;
|
2010-11-10 09:02:18 +08:00
|
|
|
|
2010-11-09 00:08:43 +08:00
|
|
|
// If the function declares it only reads memory, go with that.
|
2010-08-06 09:25:49 +08:00
|
|
|
if (F->onlyReadsMemory())
|
2015-07-23 07:15:57 +08:00
|
|
|
Min = FMRB_OnlyReadsMemory;
|
2016-07-04 16:01:29 +08:00
|
|
|
else if (F->doesNotReadMemory())
|
|
|
|
Min = FMRB_DoesNotReadMemory;
|
2010-08-06 09:25:49 +08:00
|
|
|
|
2015-07-11 18:30:36 +08:00
|
|
|
if (F->onlyAccessesArgMemory())
|
2015-07-23 07:15:57 +08:00
|
|
|
Min = FunctionModRefBehavior(Min & FMRB_OnlyAccessesArgumentPointees);
|
2016-11-09 05:07:42 +08:00
|
|
|
else if (F->onlyAccessesInaccessibleMemory())
|
|
|
|
Min = FunctionModRefBehavior(Min & FMRB_OnlyAccessesInaccessibleMem);
|
|
|
|
else if (F->onlyAccessesInaccessibleMemOrArgMem())
|
|
|
|
Min = FunctionModRefBehavior(Min & FMRB_OnlyAccessesInaccessibleOrArgMem);
|
2015-07-11 18:30:36 +08:00
|
|
|
|
[AA] Hoist the logic to reformulate various AA queries in terms of other
parts of the AA interface out of the base class of every single AA
result object.
Because this logic reformulates the query in terms of some other aspect
of the API, it would easily cause O(n^2) query patterns in alias
analysis. These could in turn be magnified further based on the number
of call arguments, and then further based on the number of AA queries
made for a particular call. This ended up causing problems for Rust that
were actually noticable enough to get a bug (PR26564) and probably other
places as well.
When originally re-working the AA infrastructure, the desire was to
regularize the pattern of refinement without losing any generality.
While I think it was successful, that is clearly proving to be too
costly. And the cost is needless: we gain no actual improvement for this
generality of making a direct query to tbaa actually be able to
re-use some other alias analysis's refinement logic for one of the other
APIs, or some such. In short, this is entirely wasted work.
To the extent possible, delegation to other API surfaces should be done
at the aggregation layer so that we can avoid re-walking the
aggregation. In fact, this significantly simplifies the logic as we no
longer need to smuggle the aggregation layer into each alias analysis
(or the TargetLibraryInfo into each alias analysis just so we can form
argument memory locations!).
However, we also have some delegation logic inside of BasicAA and some
of it even makes sense. When the delegation logic is baking in specific
knowledge of aliasing properties of the LLVM IR, as opposed to simply
reformulating the query to utilize a different alias analysis interface
entry point, it makes a lot of sense to restrict that logic to
a different layer such as BasicAA. So one aspect of the delegation that
was in every AA base class is that when we don't have operand bundles,
we re-use function AA results as a fallback for callsite alias results.
This relies on the IR properties of calls and functions w.r.t. aliasing,
and so seems a better fit to BasicAA. I've lifted the logic up to that
point where it seems to be a natural fit. This still does a bit of
redundant work (we query function attributes twice, once via the
callsite and once via the function AA query) but it is *exactly* twice
here, no more.
The end result is that all of the delegation logic is hoisted out of the
base class and into either the aggregation layer when it is a pure
retargeting to a different API surface, or into BasicAA when it relies
on the IR's aliasing properties. This should fix the quadratic query
pattern reported in PR26564, although I don't have a stand-alone test
case to reproduce it.
It also seems general goodness. Now the numerous AAs that don't need
target library info don't carry it around and depend on it. I think
I can even rip out the general access to the aggregation layer and only
expose that in BasicAA as it is the only place where we re-query in that
manner.
However, this is a non-trivial change to the AA infrastructure so I want
to get some additional eyes on this before it lands. Sadly, it can't
wait long because we should really cherry pick this into 3.8 if we're
going to go this route.
Differential Revision: http://reviews.llvm.org/D17329
llvm-svn: 262490
2016-03-02 23:56:53 +08:00
|
|
|
return Min;
|
2010-08-06 09:25:49 +08:00
|
|
|
}
|
2009-02-06 07:36:27 +08:00
|
|
|
|
2016-07-04 16:01:29 +08:00
|
|
|
/// Returns true if this is a writeonly (i.e Mod only) parameter.
|
2016-01-07 02:10:35 +08:00
|
|
|
static bool isWriteOnlyParam(ImmutableCallSite CS, unsigned ArgIdx,
|
|
|
|
const TargetLibraryInfo &TLI) {
|
2017-04-15 04:19:02 +08:00
|
|
|
if (CS.paramHasAttr(ArgIdx, Attribute::WriteOnly))
|
2016-07-04 16:01:29 +08:00
|
|
|
return true;
|
Improve BasicAA CS-CS queries (redux)
This reverts, "r213024 - Revert r212572 "improve BasicAA CS-CS queries", it
causes PR20303." with a fix for the bug in pr20303. As it turned out, the
relevant code was both wrong and over-conservative (because, as with the code
it replaced, it would return the overall ModRef mask even if just Ref had been
implied by the argument aliasing results). Hopefully, this correctly fixes both
problems.
Thanks to Nick Lewycky for reducing the test case for pr20303 (which I've
cleaned up a little and added in DSE's test directory). The BasicAA test has
also been updated to check for this error.
Original commit message:
BasicAA contains knowledge of certain intrinsics, such as memcpy and memset,
and uses that information to form more-accurate answers to CallSite vs. Loc
ModRef queries. Unfortunately, it did not use this information when answering
CallSite vs. CallSite queries.
Generically, when an intrinsic takes one or more pointers and the intrinsic is
marked only to read/write from its arguments, the offset/size is unknown. As a
result, the generic code that answers CallSite vs. CallSite (and CallSite vs.
Loc) queries in AA uses UnknownSize when forming Locs from an intrinsic's
arguments. While BasicAA's CallSite vs. Loc override could use more-accurate
size information for some intrinsics, it did not do the same for CallSite vs.
CallSite queries.
This change refactors the intrinsic-specific logic in BasicAA into a generic AA
query function: getArgLocation, which is overridden by BasicAA to supply the
intrinsic-specific knowledge, and used by AA's generic implementation. This
allows the intrinsic-specific knowledge to be used by both CallSite vs. Loc and
CallSite vs. CallSite queries, and simplifies the BasicAA implementation.
Currently, only one function, Mac's memset_pattern16, is handled by BasicAA
(all the rest are intrinsics). As a side-effect of this refactoring, BasicAA's
getModRefBehavior override now also returns OnlyAccessesArgumentPointees for
this function (which is an improvement).
llvm-svn: 213219
2014-07-17 09:28:25 +08:00
|
|
|
|
|
|
|
// We can bound the aliasing properties of memset_pattern16 just as we can
|
|
|
|
// for memcpy/memset. This is particularly important because the
|
|
|
|
// LoopIdiomRecognizer likes to turn loops into calls to memset_pattern16
|
2016-07-04 16:01:29 +08:00
|
|
|
// whenever possible.
|
|
|
|
// FIXME Consider handling this in InferFunctionAttr.cpp together with other
|
|
|
|
// attributes.
|
[Analysis] Add LibFunc_ prefix to enums in TargetLibraryInfo. (NFC)
Summary:
The LibFunc::Func enum holds enumerators named for libc functions.
Unfortunately, there are real situations, including libc implementations, where
function names are actually macros (musl uses "#define fopen64 fopen", for
example; any other transitively visible macro would have similar effects).
Strictly speaking, a conforming C++ Standard Library should provide any such
macros as functions instead (via <cstdio>). However, there are some "library"
functions which are not part of the standard, and thus not subject to this
rule (fopen64, for example). So, in order to be both portable and consistent,
the enum should not use the bare function names.
The old enum naming used a namespace LibFunc and an enum Func, with bare
enumerators. This patch changes LibFunc to be an enum with enumerators prefixed
with "LibFFunc_". (Unfortunately, a scoped enum is not sufficient to override
macros.)
There are additional changes required in clang.
Reviewers: rsmith
Subscribers: mehdi_amini, mzolotukhin, nemanjai, llvm-commits
Differential Revision: https://reviews.llvm.org/D28476
llvm-svn: 292848
2017-01-24 07:16:46 +08:00
|
|
|
LibFunc F;
|
2016-04-28 03:04:35 +08:00
|
|
|
if (CS.getCalledFunction() && TLI.getLibFunc(*CS.getCalledFunction(), F) &&
|
[Analysis] Add LibFunc_ prefix to enums in TargetLibraryInfo. (NFC)
Summary:
The LibFunc::Func enum holds enumerators named for libc functions.
Unfortunately, there are real situations, including libc implementations, where
function names are actually macros (musl uses "#define fopen64 fopen", for
example; any other transitively visible macro would have similar effects).
Strictly speaking, a conforming C++ Standard Library should provide any such
macros as functions instead (via <cstdio>). However, there are some "library"
functions which are not part of the standard, and thus not subject to this
rule (fopen64, for example). So, in order to be both portable and consistent,
the enum should not use the bare function names.
The old enum naming used a namespace LibFunc and an enum Func, with bare
enumerators. This patch changes LibFunc to be an enum with enumerators prefixed
with "LibFFunc_". (Unfortunately, a scoped enum is not sufficient to override
macros.)
There are additional changes required in clang.
Reviewers: rsmith
Subscribers: mehdi_amini, mzolotukhin, nemanjai, llvm-commits
Differential Revision: https://reviews.llvm.org/D28476
llvm-svn: 292848
2017-01-24 07:16:46 +08:00
|
|
|
F == LibFunc_memset_pattern16 && TLI.has(F))
|
2016-01-06 12:53:16 +08:00
|
|
|
if (ArgIdx == 0)
|
2016-01-07 02:10:35 +08:00
|
|
|
return true;
|
|
|
|
|
|
|
|
// TODO: memset_pattern4, memset_pattern8
|
|
|
|
// TODO: _chk variants
|
|
|
|
// TODO: strcmp, strcpy
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
ModRefInfo BasicAAResult::getArgModRefInfo(ImmutableCallSite CS,
|
|
|
|
unsigned ArgIdx) {
|
|
|
|
|
2016-07-04 16:01:29 +08:00
|
|
|
// Checking for known builtin intrinsics and target library functions.
|
2016-01-07 02:10:35 +08:00
|
|
|
if (isWriteOnlyParam(CS, ArgIdx, TLI))
|
|
|
|
return MRI_Mod;
|
Improve BasicAA CS-CS queries (redux)
This reverts, "r213024 - Revert r212572 "improve BasicAA CS-CS queries", it
causes PR20303." with a fix for the bug in pr20303. As it turned out, the
relevant code was both wrong and over-conservative (because, as with the code
it replaced, it would return the overall ModRef mask even if just Ref had been
implied by the argument aliasing results). Hopefully, this correctly fixes both
problems.
Thanks to Nick Lewycky for reducing the test case for pr20303 (which I've
cleaned up a little and added in DSE's test directory). The BasicAA test has
also been updated to check for this error.
Original commit message:
BasicAA contains knowledge of certain intrinsics, such as memcpy and memset,
and uses that information to form more-accurate answers to CallSite vs. Loc
ModRef queries. Unfortunately, it did not use this information when answering
CallSite vs. CallSite queries.
Generically, when an intrinsic takes one or more pointers and the intrinsic is
marked only to read/write from its arguments, the offset/size is unknown. As a
result, the generic code that answers CallSite vs. CallSite (and CallSite vs.
Loc) queries in AA uses UnknownSize when forming Locs from an intrinsic's
arguments. While BasicAA's CallSite vs. Loc override could use more-accurate
size information for some intrinsics, it did not do the same for CallSite vs.
CallSite queries.
This change refactors the intrinsic-specific logic in BasicAA into a generic AA
query function: getArgLocation, which is overridden by BasicAA to supply the
intrinsic-specific knowledge, and used by AA's generic implementation. This
allows the intrinsic-specific knowledge to be used by both CallSite vs. Loc and
CallSite vs. CallSite queries, and simplifies the BasicAA implementation.
Currently, only one function, Mac's memset_pattern16, is handled by BasicAA
(all the rest are intrinsics). As a side-effect of this refactoring, BasicAA's
getModRefBehavior override now also returns OnlyAccessesArgumentPointees for
this function (which is an improvement).
llvm-svn: 213219
2014-07-17 09:28:25 +08:00
|
|
|
|
2017-04-15 04:19:02 +08:00
|
|
|
if (CS.paramHasAttr(ArgIdx, Attribute::ReadOnly))
|
2015-10-29 00:42:00 +08:00
|
|
|
return MRI_Ref;
|
|
|
|
|
2017-04-15 04:19:02 +08:00
|
|
|
if (CS.paramHasAttr(ArgIdx, Attribute::ReadNone))
|
2015-10-29 01:54:48 +08:00
|
|
|
return MRI_NoModRef;
|
|
|
|
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
return AAResultBase::getArgModRefInfo(CS, ArgIdx);
|
Improve BasicAA CS-CS queries (redux)
This reverts, "r213024 - Revert r212572 "improve BasicAA CS-CS queries", it
causes PR20303." with a fix for the bug in pr20303. As it turned out, the
relevant code was both wrong and over-conservative (because, as with the code
it replaced, it would return the overall ModRef mask even if just Ref had been
implied by the argument aliasing results). Hopefully, this correctly fixes both
problems.
Thanks to Nick Lewycky for reducing the test case for pr20303 (which I've
cleaned up a little and added in DSE's test directory). The BasicAA test has
also been updated to check for this error.
Original commit message:
BasicAA contains knowledge of certain intrinsics, such as memcpy and memset,
and uses that information to form more-accurate answers to CallSite vs. Loc
ModRef queries. Unfortunately, it did not use this information when answering
CallSite vs. CallSite queries.
Generically, when an intrinsic takes one or more pointers and the intrinsic is
marked only to read/write from its arguments, the offset/size is unknown. As a
result, the generic code that answers CallSite vs. CallSite (and CallSite vs.
Loc) queries in AA uses UnknownSize when forming Locs from an intrinsic's
arguments. While BasicAA's CallSite vs. Loc override could use more-accurate
size information for some intrinsics, it did not do the same for CallSite vs.
CallSite queries.
This change refactors the intrinsic-specific logic in BasicAA into a generic AA
query function: getArgLocation, which is overridden by BasicAA to supply the
intrinsic-specific knowledge, and used by AA's generic implementation. This
allows the intrinsic-specific knowledge to be used by both CallSite vs. Loc and
CallSite vs. CallSite queries, and simplifies the BasicAA implementation.
Currently, only one function, Mac's memset_pattern16, is handled by BasicAA
(all the rest are intrinsics). As a side-effect of this refactoring, BasicAA's
getModRefBehavior override now also returns OnlyAccessesArgumentPointees for
this function (which is an improvement).
llvm-svn: 213219
2014-07-17 09:28:25 +08:00
|
|
|
}
|
|
|
|
|
2016-05-10 10:35:41 +08:00
|
|
|
static bool isIntrinsicCall(ImmutableCallSite CS, Intrinsic::ID IID) {
|
2014-07-26 05:13:35 +08:00
|
|
|
const IntrinsicInst *II = dyn_cast<IntrinsicInst>(CS.getInstruction());
|
2016-05-10 10:35:41 +08:00
|
|
|
return II && II->getIntrinsicID() == IID;
|
2014-07-26 05:13:35 +08:00
|
|
|
}
|
|
|
|
|
2015-09-24 13:29:31 +08:00
|
|
|
#ifndef NDEBUG
|
2015-09-24 12:59:24 +08:00
|
|
|
static const Function *getParent(const Value *V) {
|
|
|
|
if (const Instruction *inst = dyn_cast<Instruction>(V))
|
|
|
|
return inst->getParent()->getParent();
|
|
|
|
|
|
|
|
if (const Argument *arg = dyn_cast<Argument>(V))
|
|
|
|
return arg->getParent();
|
|
|
|
|
|
|
|
return nullptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool notDifferentParent(const Value *O1, const Value *O2) {
|
|
|
|
|
|
|
|
const Function *F1 = getParent(O1);
|
|
|
|
const Function *F2 = getParent(O2);
|
|
|
|
|
|
|
|
return !F1 || !F2 || F1 == F2;
|
|
|
|
}
|
2015-09-24 13:29:31 +08:00
|
|
|
#endif
|
2015-09-24 12:59:24 +08:00
|
|
|
|
|
|
|
AliasResult BasicAAResult::alias(const MemoryLocation &LocA,
|
|
|
|
const MemoryLocation &LocB) {
|
|
|
|
assert(notDifferentParent(LocA.Ptr, LocB.Ptr) &&
|
|
|
|
"BasicAliasAnalysis doesn't support interprocedural queries.");
|
|
|
|
|
|
|
|
// If we have a directly cached entry for these locations, we have recursed
|
|
|
|
// through this once, so just return the cached results. Notably, when this
|
|
|
|
// happens, we don't clear the cache.
|
|
|
|
auto CacheIt = AliasCache.find(LocPair(LocA, LocB));
|
|
|
|
if (CacheIt != AliasCache.end())
|
|
|
|
return CacheIt->second;
|
|
|
|
|
|
|
|
AliasResult Alias = aliasCheck(LocA.Ptr, LocA.Size, LocA.AATags, LocB.Ptr,
|
|
|
|
LocB.Size, LocB.AATags);
|
|
|
|
// AliasCache rarely has more than 1 or 2 elements, always use
|
|
|
|
// shrink_and_clear so it quickly returns to the inline capacity of the
|
|
|
|
// SmallDenseMap if it ever grows larger.
|
|
|
|
// FIXME: This should really be shrink_to_inline_capacity_and_clear().
|
|
|
|
AliasCache.shrink_and_clear();
|
|
|
|
VisitedPhiBBs.clear();
|
|
|
|
return Alias;
|
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Checks to see if the specified callsite can clobber the specified memory
|
|
|
|
/// object.
|
|
|
|
///
|
|
|
|
/// Since we only look at local properties of this function, we really can't
|
|
|
|
/// say much about this query. We do, however, use simple "address taken"
|
|
|
|
/// analysis on local objects.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
ModRefInfo BasicAAResult::getModRefInfo(ImmutableCallSite CS,
|
|
|
|
const MemoryLocation &Loc) {
|
2010-09-15 05:25:10 +08:00
|
|
|
assert(notDifferentParent(CS.getInstruction(), Loc.Ptr) &&
|
2010-07-07 22:27:09 +08:00
|
|
|
"AliasAnalysis query involving multiple functions!");
|
|
|
|
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
const Value *Object = GetUnderlyingObject(Loc.Ptr, DL);
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-09-15 05:25:10 +08:00
|
|
|
// If this is a tail call and Loc.Ptr points to a stack location, we know that
|
2009-11-23 00:05:05 +08:00
|
|
|
// the tail call cannot access or modify the local stack.
|
|
|
|
// We cannot exclude byval arguments here; these belong to the caller of
|
|
|
|
// the current function not to the current function, and a tail callee
|
|
|
|
// may reference them.
|
|
|
|
if (isa<AllocaInst>(Object))
|
2010-08-04 05:48:53 +08:00
|
|
|
if (const CallInst *CI = dyn_cast<CallInst>(CS.getInstruction()))
|
2009-11-23 00:05:05 +08:00
|
|
|
if (CI->isTailCall())
|
2015-07-23 07:15:57 +08:00
|
|
|
return MRI_NoModRef;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2009-11-23 00:05:05 +08:00
|
|
|
// If the pointer is to a locally allocated object that does not escape,
|
2009-11-24 00:44:43 +08:00
|
|
|
// then the call can not mod/ref the pointer unless the call takes the pointer
|
|
|
|
// as an argument, and itself doesn't capture it.
|
2009-11-24 00:46:41 +08:00
|
|
|
if (!isa<Constant>(Object) && CS.getInstruction() != Object &&
|
2010-07-02 04:08:40 +08:00
|
|
|
isNonEscapingLocalObject(Object)) {
|
2017-03-01 21:19:51 +08:00
|
|
|
|
|
|
|
// Optimistically assume that call doesn't touch Object and check this
|
|
|
|
// assumption in the following loop.
|
|
|
|
ModRefInfo Result = MRI_NoModRef;
|
|
|
|
|
2016-01-16 20:15:53 +08:00
|
|
|
unsigned OperandNo = 0;
|
|
|
|
for (auto CI = CS.data_operands_begin(), CE = CS.data_operands_end();
|
|
|
|
CI != CE; ++CI, ++OperandNo) {
|
2011-05-23 13:15:43 +08:00
|
|
|
// Only look at the no-capture or byval pointer arguments. If this
|
|
|
|
// pointer were passed to arguments that were neither of these, then it
|
|
|
|
// couldn't be no-capture.
|
2010-02-16 19:11:14 +08:00
|
|
|
if (!(*CI)->getType()->isPointerTy() ||
|
2016-12-15 13:09:15 +08:00
|
|
|
(!CS.doesNotCapture(OperandNo) &&
|
|
|
|
OperandNo < CS.getNumArgOperands() && !CS.isByValArgument(OperandNo)))
|
2009-11-24 00:44:43 +08:00
|
|
|
continue;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2017-03-01 21:19:51 +08:00
|
|
|
// Call doesn't access memory through this operand, so we don't care
|
|
|
|
// if it aliases with Object.
|
|
|
|
if (CS.doesNotAccessMemory(OperandNo))
|
|
|
|
continue;
|
|
|
|
|
2010-09-15 05:25:10 +08:00
|
|
|
// If this is a no-capture pointer argument, see if we can tell that it
|
2017-03-01 21:19:51 +08:00
|
|
|
// is impossible to alias the pointer we're checking.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
AliasResult AR =
|
|
|
|
getBestAAResults().alias(MemoryLocation(*CI), MemoryLocation(Object));
|
2017-03-01 21:19:51 +08:00
|
|
|
|
|
|
|
// Operand doesnt alias 'Object', continue looking for other aliases
|
|
|
|
if (AR == NoAlias)
|
|
|
|
continue;
|
|
|
|
// Operand aliases 'Object', but call doesn't modify it. Strengthen
|
|
|
|
// initial assumption and keep looking in case if there are more aliases.
|
|
|
|
if (CS.onlyReadsMemory(OperandNo)) {
|
|
|
|
Result = static_cast<ModRefInfo>(Result | MRI_Ref);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
// Operand aliases 'Object' but call only writes into it.
|
|
|
|
if (CS.doesNotReadMemory(OperandNo)) {
|
|
|
|
Result = static_cast<ModRefInfo>(Result | MRI_Mod);
|
|
|
|
continue;
|
2009-11-24 00:44:43 +08:00
|
|
|
}
|
2017-03-01 21:19:51 +08:00
|
|
|
// This operand aliases 'Object' and call reads and writes into it.
|
|
|
|
Result = MRI_ModRef;
|
|
|
|
break;
|
2009-11-24 00:44:43 +08:00
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2017-03-01 21:19:51 +08:00
|
|
|
// Early return if we improved mod ref information
|
|
|
|
if (Result != MRI_ModRef)
|
|
|
|
return Result;
|
2009-11-23 00:05:05 +08:00
|
|
|
}
|
|
|
|
|
2016-03-10 07:19:56 +08:00
|
|
|
// If the CallSite is to malloc or calloc, we can assume that it doesn't
|
|
|
|
// modify any IR visible value. This is only valid because we assume these
|
|
|
|
// routines do not read values visible in the IR. TODO: Consider special
|
|
|
|
// casing realloc and strdup routines which access only their arguments as
|
|
|
|
// well. Or alternatively, replace all of this with inaccessiblememonly once
|
|
|
|
// that's implemented fully.
|
|
|
|
auto *Inst = CS.getInstruction();
|
2017-04-19 05:43:46 +08:00
|
|
|
if (isMallocOrCallocLikeFn(Inst, &TLI)) {
|
2016-03-10 07:19:56 +08:00
|
|
|
// Be conservative if the accessed pointer may alias the allocation -
|
|
|
|
// fallback to the generic handling below.
|
|
|
|
if (getBestAAResults().alias(MemoryLocation(Inst), Loc) == NoAlias)
|
|
|
|
return MRI_NoModRef;
|
|
|
|
}
|
|
|
|
|
2016-12-26 06:42:27 +08:00
|
|
|
// The semantics of memcpy intrinsics forbid overlap between their respective
|
|
|
|
// operands, i.e., source and destination of any given memcpy must no-alias.
|
|
|
|
// If Loc must-aliases either one of these two locations, then it necessarily
|
|
|
|
// no-aliases the other.
|
|
|
|
if (auto *Inst = dyn_cast<MemCpyInst>(CS.getInstruction())) {
|
|
|
|
AliasResult SrcAA, DestAA;
|
|
|
|
|
|
|
|
if ((SrcAA = getBestAAResults().alias(MemoryLocation::getForSource(Inst),
|
|
|
|
Loc)) == MustAlias)
|
|
|
|
// Loc is exactly the memcpy source thus disjoint from memcpy dest.
|
|
|
|
return MRI_Ref;
|
|
|
|
if ((DestAA = getBestAAResults().alias(MemoryLocation::getForDest(Inst),
|
|
|
|
Loc)) == MustAlias)
|
|
|
|
// The converse case.
|
|
|
|
return MRI_Mod;
|
|
|
|
|
|
|
|
// It's also possible for Loc to alias both src and dest, or neither.
|
|
|
|
ModRefInfo rv = MRI_NoModRef;
|
|
|
|
if (SrcAA != NoAlias)
|
|
|
|
rv = static_cast<ModRefInfo>(rv | MRI_Ref);
|
|
|
|
if (DestAA != NoAlias)
|
|
|
|
rv = static_cast<ModRefInfo>(rv | MRI_Mod);
|
|
|
|
return rv;
|
|
|
|
}
|
|
|
|
|
2014-07-26 05:13:35 +08:00
|
|
|
// While the assume intrinsic is marked as arbitrarily writing so that
|
|
|
|
// proper control dependencies will be maintained, it never aliases any
|
|
|
|
// particular memory location.
|
2016-05-10 10:35:41 +08:00
|
|
|
if (isIntrinsicCall(CS, Intrinsic::assume))
|
2015-07-23 07:15:57 +08:00
|
|
|
return MRI_NoModRef;
|
2014-07-26 05:13:35 +08:00
|
|
|
|
2016-05-10 10:35:41 +08:00
|
|
|
// Like assumes, guard intrinsics are also marked as arbitrarily writing so
|
|
|
|
// that proper control dependencies are maintained but they never mods any
|
|
|
|
// particular memory location.
|
|
|
|
//
|
|
|
|
// *Unlike* assumes, guard intrinsics are modeled as reading memory since the
|
|
|
|
// heap state at the point the guard is issued needs to be consistent in case
|
|
|
|
// the guard invokes the "deopt" continuation.
|
|
|
|
if (isIntrinsicCall(CS, Intrinsic::experimental_guard))
|
|
|
|
return MRI_Ref;
|
|
|
|
|
2016-08-10 01:18:05 +08:00
|
|
|
// Like assumes, invariant.start intrinsics were also marked as arbitrarily
|
|
|
|
// writing so that proper control dependencies are maintained but they never
|
|
|
|
// mod any particular memory location visible to the IR.
|
|
|
|
// *Unlike* assumes (which are now modeled as NoModRef), invariant.start
|
|
|
|
// intrinsic is now modeled as reading memory. This prevents hoisting the
|
|
|
|
// invariant.start intrinsic over stores. Consider:
|
|
|
|
// *ptr = 40;
|
|
|
|
// *ptr = 50;
|
|
|
|
// invariant_start(ptr)
|
|
|
|
// int val = *ptr;
|
|
|
|
// print(val);
|
|
|
|
//
|
|
|
|
// This cannot be transformed to:
|
|
|
|
//
|
|
|
|
// *ptr = 40;
|
|
|
|
// invariant_start(ptr)
|
|
|
|
// *ptr = 50;
|
|
|
|
// int val = *ptr;
|
|
|
|
// print(val);
|
|
|
|
//
|
|
|
|
// The transformation will cause the second store to be ignored (based on
|
|
|
|
// rules of invariant.start) and print 40, while the first program always
|
|
|
|
// prints 50.
|
|
|
|
if (isIntrinsicCall(CS, Intrinsic::invariant_start))
|
|
|
|
return MRI_Ref;
|
|
|
|
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
// The AAResultBase base class has some smarts, lets use them.
|
|
|
|
return AAResultBase::getModRefInfo(CS, Loc);
|
2010-09-08 09:32:20 +08:00
|
|
|
}
|
2008-06-16 14:10:11 +08:00
|
|
|
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
ModRefInfo BasicAAResult::getModRefInfo(ImmutableCallSite CS1,
|
|
|
|
ImmutableCallSite CS2) {
|
2014-07-26 05:13:35 +08:00
|
|
|
// While the assume intrinsic is marked as arbitrarily writing so that
|
|
|
|
// proper control dependencies will be maintained, it never aliases any
|
|
|
|
// particular memory location.
|
2016-05-10 10:35:41 +08:00
|
|
|
if (isIntrinsicCall(CS1, Intrinsic::assume) ||
|
|
|
|
isIntrinsicCall(CS2, Intrinsic::assume))
|
2015-07-23 07:15:57 +08:00
|
|
|
return MRI_NoModRef;
|
2014-07-26 05:13:35 +08:00
|
|
|
|
2016-05-10 10:35:41 +08:00
|
|
|
// Like assumes, guard intrinsics are also marked as arbitrarily writing so
|
|
|
|
// that proper control dependencies are maintained but they never mod any
|
|
|
|
// particular memory location.
|
|
|
|
//
|
|
|
|
// *Unlike* assumes, guard intrinsics are modeled as reading memory since the
|
|
|
|
// heap state at the point the guard is issued needs to be consistent in case
|
|
|
|
// the guard invokes the "deopt" continuation.
|
|
|
|
|
|
|
|
// NB! This function is *not* commutative, so we specical case two
|
|
|
|
// possibilities for guard intrinsics.
|
|
|
|
|
|
|
|
if (isIntrinsicCall(CS1, Intrinsic::experimental_guard))
|
|
|
|
return getModRefBehavior(CS2) & MRI_Mod ? MRI_Ref : MRI_NoModRef;
|
|
|
|
|
|
|
|
if (isIntrinsicCall(CS2, Intrinsic::experimental_guard))
|
|
|
|
return getModRefBehavior(CS1) & MRI_Mod ? MRI_Mod : MRI_NoModRef;
|
|
|
|
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
// The AAResultBase base class has some smarts, lets use them.
|
|
|
|
return AAResultBase::getModRefInfo(CS1, CS2);
|
2014-07-26 05:13:35 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Provide ad-hoc rules to disambiguate accesses through two GEP operators,
|
|
|
|
/// both having the exact same pointer operand.
|
2015-06-22 10:16:51 +08:00
|
|
|
static AliasResult aliasSameBasePointerGEPs(const GEPOperator *GEP1,
|
|
|
|
uint64_t V1Size,
|
|
|
|
const GEPOperator *GEP2,
|
|
|
|
uint64_t V2Size,
|
|
|
|
const DataLayout &DL) {
|
2015-02-08 01:04:29 +08:00
|
|
|
|
Handle invariant.group.barrier in BasicAA
Summary:
llvm.invariant.group.barrier returns pointer that mustalias
pointer it takes. It can't be marked with `returned` attribute,
because it would be remove easily. The other reason is that
only Alias Analysis can know about this, because if any other
pass would know it, then the result would be replaced with it's
argument, which would be invalid.
We can think about returned pointer as something that mustalias, but
it doesn't have to be bitwise the same as the argument.
Reviewers: dberlin, chandlerc, hfinkel, sanjoy
Subscribers: reames, nlewycky, rsmith, anna, amharc
Differential Revision: https://reviews.llvm.org/D31585
llvm-svn: 301227
2017-04-25 03:37:17 +08:00
|
|
|
assert(GEP1->getPointerOperand()->stripPointerCastsAndBarriers() ==
|
|
|
|
GEP2->getPointerOperand()->stripPointerCastsAndBarriers() &&
|
2017-04-19 06:00:54 +08:00
|
|
|
GEP1->getPointerOperandType() == GEP2->getPointerOperandType() &&
|
2015-02-08 01:04:29 +08:00
|
|
|
"Expected GEPs with the same pointer operand");
|
|
|
|
|
|
|
|
// Try to determine whether GEP1 and GEP2 index through arrays, into structs,
|
|
|
|
// such that the struct field accesses provably cannot alias.
|
|
|
|
// We also need at least two indices (the pointer, and the struct field).
|
2016-06-02 02:55:32 +08:00
|
|
|
if (GEP1->getNumIndices() != GEP2->getNumIndices() ||
|
|
|
|
GEP1->getNumIndices() < 2)
|
2015-06-22 10:16:51 +08:00
|
|
|
return MayAlias;
|
2015-02-08 01:04:29 +08:00
|
|
|
|
|
|
|
// If we don't know the size of the accesses through both GEPs, we can't
|
|
|
|
// determine whether the struct fields accessed can't alias.
|
2015-06-17 15:21:38 +08:00
|
|
|
if (V1Size == MemoryLocation::UnknownSize ||
|
|
|
|
V2Size == MemoryLocation::UnknownSize)
|
2015-06-22 10:16:51 +08:00
|
|
|
return MayAlias;
|
2015-02-08 01:04:29 +08:00
|
|
|
|
|
|
|
ConstantInt *C1 =
|
|
|
|
dyn_cast<ConstantInt>(GEP1->getOperand(GEP1->getNumOperands() - 1));
|
|
|
|
ConstantInt *C2 =
|
|
|
|
dyn_cast<ConstantInt>(GEP2->getOperand(GEP2->getNumOperands() - 1));
|
|
|
|
|
2015-10-22 21:28:18 +08:00
|
|
|
// If the last (struct) indices are constants and are equal, the other indices
|
|
|
|
// might be also be dynamically equal, so the GEPs can alias.
|
2016-05-11 23:45:43 +08:00
|
|
|
if (C1 && C2 && C1->getSExtValue() == C2->getSExtValue())
|
2015-06-22 10:16:51 +08:00
|
|
|
return MayAlias;
|
2015-02-08 01:04:29 +08:00
|
|
|
|
|
|
|
// Find the last-indexed type of the GEP, i.e., the type you'd get if
|
|
|
|
// you stripped the last index.
|
|
|
|
// On the way, look at each indexed type. If there's something other
|
|
|
|
// than an array, different indices can lead to different final types.
|
|
|
|
SmallVector<Value *, 8> IntermediateIndices;
|
|
|
|
|
|
|
|
// Insert the first index; we don't need to check the type indexed
|
|
|
|
// through it as it only drops the pointer indirection.
|
|
|
|
assert(GEP1->getNumIndices() > 1 && "Not enough GEP indices to examine");
|
|
|
|
IntermediateIndices.push_back(GEP1->getOperand(1));
|
|
|
|
|
|
|
|
// Insert all the remaining indices but the last one.
|
|
|
|
// Also, check that they all index through arrays.
|
|
|
|
for (unsigned i = 1, e = GEP1->getNumIndices() - 1; i != e; ++i) {
|
|
|
|
if (!isa<ArrayType>(GetElementPtrInst::getIndexedType(
|
2015-03-31 05:41:43 +08:00
|
|
|
GEP1->getSourceElementType(), IntermediateIndices)))
|
2015-06-22 10:16:51 +08:00
|
|
|
return MayAlias;
|
2015-02-08 01:04:29 +08:00
|
|
|
IntermediateIndices.push_back(GEP1->getOperand(i + 1));
|
|
|
|
}
|
|
|
|
|
2015-10-22 21:28:18 +08:00
|
|
|
auto *Ty = GetElementPtrInst::getIndexedType(
|
|
|
|
GEP1->getSourceElementType(), IntermediateIndices);
|
|
|
|
StructType *LastIndexedStruct = dyn_cast<StructType>(Ty);
|
|
|
|
|
|
|
|
if (isa<SequentialType>(Ty)) {
|
|
|
|
// We know that:
|
|
|
|
// - both GEPs begin indexing from the exact same pointer;
|
|
|
|
// - the last indices in both GEPs are constants, indexing into a sequential
|
|
|
|
// type (array or pointer);
|
|
|
|
// - both GEPs only index through arrays prior to that.
|
|
|
|
//
|
|
|
|
// Because array indices greater than the number of elements are valid in
|
|
|
|
// GEPs, unless we know the intermediate indices are identical between
|
|
|
|
// GEP1 and GEP2 we cannot guarantee that the last indexed arrays don't
|
2015-10-23 22:17:03 +08:00
|
|
|
// partially overlap. We also need to check that the loaded size matches
|
|
|
|
// the element size, otherwise we could still have overlap.
|
|
|
|
const uint64_t ElementSize =
|
|
|
|
DL.getTypeStoreSize(cast<SequentialType>(Ty)->getElementType());
|
|
|
|
if (V1Size != ElementSize || V2Size != ElementSize)
|
|
|
|
return MayAlias;
|
|
|
|
|
2015-10-22 21:28:18 +08:00
|
|
|
for (unsigned i = 0, e = GEP1->getNumIndices() - 1; i != e; ++i)
|
|
|
|
if (GEP1->getOperand(i + 1) != GEP2->getOperand(i + 1))
|
|
|
|
return MayAlias;
|
2015-10-23 22:17:03 +08:00
|
|
|
|
2015-10-22 21:28:18 +08:00
|
|
|
// Now we know that the array/pointer that GEP1 indexes into and that
|
|
|
|
// that GEP2 indexes into must either precisely overlap or be disjoint.
|
|
|
|
// Because they cannot partially overlap and because fields in an array
|
|
|
|
// cannot overlap, if we can prove the final indices are different between
|
|
|
|
// GEP1 and GEP2, we can conclude GEP1 and GEP2 don't alias.
|
|
|
|
|
|
|
|
// If the last indices are constants, we've already checked they don't
|
|
|
|
// equal each other so we can exit early.
|
|
|
|
if (C1 && C2)
|
|
|
|
return NoAlias;
|
|
|
|
if (isKnownNonEqual(GEP1->getOperand(GEP1->getNumOperands() - 1),
|
|
|
|
GEP2->getOperand(GEP2->getNumOperands() - 1),
|
|
|
|
DL))
|
|
|
|
return NoAlias;
|
|
|
|
return MayAlias;
|
|
|
|
} else if (!LastIndexedStruct || !C1 || !C2) {
|
2015-06-22 10:16:51 +08:00
|
|
|
return MayAlias;
|
2015-10-22 21:28:18 +08:00
|
|
|
}
|
2015-02-08 01:04:29 +08:00
|
|
|
|
|
|
|
// We know that:
|
|
|
|
// - both GEPs begin indexing from the exact same pointer;
|
|
|
|
// - the last indices in both GEPs are constants, indexing into a struct;
|
|
|
|
// - said indices are different, hence, the pointed-to fields are different;
|
|
|
|
// - both GEPs only index through arrays prior to that.
|
|
|
|
//
|
|
|
|
// This lets us determine that the struct that GEP1 indexes into and the
|
|
|
|
// struct that GEP2 indexes into must either precisely overlap or be
|
|
|
|
// completely disjoint. Because they cannot partially overlap, indexing into
|
|
|
|
// different non-overlapping fields of the struct will never alias.
|
|
|
|
|
|
|
|
// Therefore, the only remaining thing needed to show that both GEPs can't
|
|
|
|
// alias is that the fields are not overlapping.
|
|
|
|
const StructLayout *SL = DL.getStructLayout(LastIndexedStruct);
|
|
|
|
const uint64_t StructSize = SL->getSizeInBytes();
|
|
|
|
const uint64_t V1Off = SL->getElementOffset(C1->getZExtValue());
|
|
|
|
const uint64_t V2Off = SL->getElementOffset(C2->getZExtValue());
|
|
|
|
|
|
|
|
auto EltsDontOverlap = [StructSize](uint64_t V1Off, uint64_t V1Size,
|
|
|
|
uint64_t V2Off, uint64_t V2Size) {
|
|
|
|
return V1Off < V2Off && V1Off + V1Size <= V2Off &&
|
|
|
|
((V2Off + V2Size <= StructSize) ||
|
|
|
|
(V2Off + V2Size - StructSize <= V1Off));
|
|
|
|
};
|
|
|
|
|
|
|
|
if (EltsDontOverlap(V1Off, V1Size, V2Off, V2Size) ||
|
|
|
|
EltsDontOverlap(V2Off, V2Size, V1Off, V1Size))
|
2015-06-22 10:16:51 +08:00
|
|
|
return NoAlias;
|
2015-02-08 01:04:29 +08:00
|
|
|
|
2015-06-22 10:16:51 +08:00
|
|
|
return MayAlias;
|
2015-02-08 01:04:29 +08:00
|
|
|
}
|
|
|
|
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
// If a we have (a) a GEP and (b) a pointer based on an alloca, and the
|
|
|
|
// beginning of the object the GEP points would have a negative offset with
|
|
|
|
// repsect to the alloca, that means the GEP can not alias pointer (b).
|
|
|
|
// Note that the pointer based on the alloca may not be a GEP. For
|
|
|
|
// example, it may be the alloca itself.
|
2016-05-27 03:30:49 +08:00
|
|
|
// The same applies if (b) is based on a GlobalVariable. Note that just being
|
|
|
|
// based on isIdentifiedObject() is not enough - we need an identified object
|
|
|
|
// that does not permit access to negative offsets. For example, a negative
|
|
|
|
// offset from a noalias argument or call can be inbounds w.r.t the actual
|
|
|
|
// underlying object.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
//
|
|
|
|
// For example, consider:
|
|
|
|
//
|
|
|
|
// struct { int f0, int f1, ...} foo;
|
|
|
|
// foo alloca;
|
|
|
|
// foo* random = bar(alloca);
|
|
|
|
// int *f0 = &alloca.f0
|
|
|
|
// int *f1 = &random->f1;
|
|
|
|
//
|
|
|
|
// Which is lowered, approximately, to:
|
|
|
|
//
|
|
|
|
// %alloca = alloca %struct.foo
|
|
|
|
// %random = call %struct.foo* @random(%struct.foo* %alloca)
|
|
|
|
// %f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
|
|
|
|
// %f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
|
|
|
|
//
|
|
|
|
// Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
|
|
|
|
// by %alloca. Since the %f1 GEP is inbounds, that means %random must also
|
|
|
|
// point into the same object. But since %f0 points to the beginning of %alloca,
|
|
|
|
// the highest %f1 can be is (%alloca + 3). This means %random can not be higher
|
|
|
|
// than (%alloca - 1), and so is not inbounds, a contradiction.
|
|
|
|
bool BasicAAResult::isGEPBaseAtNegativeOffset(const GEPOperator *GEPOp,
|
2016-05-27 03:30:49 +08:00
|
|
|
const DecomposedGEP &DecompGEP, const DecomposedGEP &DecompObject,
|
|
|
|
uint64_t ObjectAccessSize) {
|
|
|
|
// If the object access size is unknown, or the GEP isn't inbounds, bail.
|
|
|
|
if (ObjectAccessSize == MemoryLocation::UnknownSize || !GEPOp->isInBounds())
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
return false;
|
|
|
|
|
2016-05-27 03:30:49 +08:00
|
|
|
// We need the object to be an alloca or a globalvariable, and want to know
|
|
|
|
// the offset of the pointer from the object precisely, so no variable
|
|
|
|
// indices are allowed.
|
|
|
|
if (!(isa<AllocaInst>(DecompObject.Base) ||
|
|
|
|
isa<GlobalVariable>(DecompObject.Base)) ||
|
|
|
|
!DecompObject.VarIndices.empty())
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
return false;
|
|
|
|
|
2016-05-27 03:30:49 +08:00
|
|
|
int64_t ObjectBaseOffset = DecompObject.StructOffset +
|
|
|
|
DecompObject.OtherOffset;
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
|
|
|
|
// If the GEP has no variable indices, we know the precise offset
|
|
|
|
// from the base, then use it. If the GEP has variable indices, we're in
|
|
|
|
// a bit more trouble: we can't count on the constant offsets that come
|
|
|
|
// from non-struct sources, since these can be "rewound" by a negative
|
|
|
|
// variable offset. So use only offsets that came from structs.
|
|
|
|
int64_t GEPBaseOffset = DecompGEP.StructOffset;
|
|
|
|
if (DecompGEP.VarIndices.empty())
|
|
|
|
GEPBaseOffset += DecompGEP.OtherOffset;
|
|
|
|
|
2016-05-27 03:30:49 +08:00
|
|
|
return (GEPBaseOffset >= ObjectBaseOffset + (int64_t)ObjectAccessSize);
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Provides a bunch of ad-hoc rules to disambiguate a GEP instruction against
|
|
|
|
/// another pointer.
|
2009-11-26 10:11:08 +08:00
|
|
|
///
|
2015-08-06 16:17:06 +08:00
|
|
|
/// We know that V1 is a GEP, but we don't know anything about V2.
|
|
|
|
/// UnderlyingV1 is GetUnderlyingObject(GEP1, DL), UnderlyingV2 is the same for
|
|
|
|
/// V2.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
AliasResult BasicAAResult::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size,
|
|
|
|
const AAMDNodes &V1AAInfo, const Value *V2,
|
|
|
|
uint64_t V2Size, const AAMDNodes &V2AAInfo,
|
|
|
|
const Value *UnderlyingV1,
|
|
|
|
const Value *UnderlyingV2) {
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
DecomposedGEP DecompGEP1, DecompGEP2;
|
|
|
|
bool GEP1MaxLookupReached =
|
2016-12-19 16:22:17 +08:00
|
|
|
DecomposeGEPExpression(GEP1, DecompGEP1, DL, &AC, DT);
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
bool GEP2MaxLookupReached =
|
2016-12-19 16:22:17 +08:00
|
|
|
DecomposeGEPExpression(V2, DecompGEP2, DL, &AC, DT);
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
|
|
|
|
int64_t GEP1BaseOffset = DecompGEP1.StructOffset + DecompGEP1.OtherOffset;
|
|
|
|
int64_t GEP2BaseOffset = DecompGEP2.StructOffset + DecompGEP2.OtherOffset;
|
|
|
|
|
|
|
|
assert(DecompGEP1.Base == UnderlyingV1 && DecompGEP2.Base == UnderlyingV2 &&
|
|
|
|
"DecomposeGEPExpression returned a result different from "
|
|
|
|
"GetUnderlyingObject");
|
|
|
|
|
|
|
|
// If the GEP's offset relative to its base is such that the base would
|
|
|
|
// fall below the start of the object underlying V2, then the GEP and V2
|
|
|
|
// cannot alias.
|
|
|
|
if (!GEP1MaxLookupReached && !GEP2MaxLookupReached &&
|
|
|
|
isGEPBaseAtNegativeOffset(GEP1, DecompGEP1, DecompGEP2, V2Size))
|
|
|
|
return NoAlias;
|
2012-09-06 22:31:51 +08:00
|
|
|
// If we have two gep instructions with must-alias or not-alias'ing base
|
|
|
|
// pointers, figure out if the indexes to the GEP tell us anything about the
|
|
|
|
// derived pointer.
|
2009-11-26 10:11:08 +08:00
|
|
|
if (const GEPOperator *GEP2 = dyn_cast<GEPOperator>(V2)) {
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
// Check for the GEP base being at a negative offset, this time in the other
|
|
|
|
// direction.
|
|
|
|
if (!GEP1MaxLookupReached && !GEP2MaxLookupReached &&
|
|
|
|
isGEPBaseAtNegativeOffset(GEP2, DecompGEP2, DecompGEP1, V1Size))
|
|
|
|
return NoAlias;
|
2013-03-27 02:07:53 +08:00
|
|
|
// Do the base pointers alias?
|
2015-06-17 15:21:38 +08:00
|
|
|
AliasResult BaseAlias =
|
|
|
|
aliasCheck(UnderlyingV1, MemoryLocation::UnknownSize, AAMDNodes(),
|
|
|
|
UnderlyingV2, MemoryLocation::UnknownSize, AAMDNodes());
|
2013-03-27 02:07:53 +08:00
|
|
|
|
2012-09-06 22:31:51 +08:00
|
|
|
// Check for geps of non-aliasing underlying pointers where the offsets are
|
|
|
|
// identical.
|
2013-03-27 02:07:53 +08:00
|
|
|
if ((BaseAlias == MayAlias) && V1Size == V2Size) {
|
2012-09-06 22:31:51 +08:00
|
|
|
// Do the base pointers alias assuming type and size.
|
2015-08-06 15:57:58 +08:00
|
|
|
AliasResult PreciseBaseAlias = aliasCheck(UnderlyingV1, V1Size, V1AAInfo,
|
|
|
|
UnderlyingV2, V2Size, V2AAInfo);
|
2012-09-06 22:31:51 +08:00
|
|
|
if (PreciseBaseAlias == NoAlias) {
|
|
|
|
// See if the computed offset from the common pointer tells us about the
|
|
|
|
// relation of the resulting pointer.
|
2014-03-27 05:30:19 +08:00
|
|
|
// If the max search depth is reached the result is undefined
|
|
|
|
if (GEP2MaxLookupReached || GEP1MaxLookupReached)
|
|
|
|
return MayAlias;
|
|
|
|
|
2012-09-06 22:31:51 +08:00
|
|
|
// Same offsets.
|
|
|
|
if (GEP1BaseOffset == GEP2BaseOffset &&
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
DecompGEP1.VarIndices == DecompGEP2.VarIndices)
|
2012-09-06 22:31:51 +08:00
|
|
|
return NoAlias;
|
|
|
|
}
|
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2009-11-26 10:17:34 +08:00
|
|
|
// If we get a No or May, then return it immediately, no amount of analysis
|
|
|
|
// will improve this situation.
|
2015-08-06 15:57:58 +08:00
|
|
|
if (BaseAlias != MustAlias)
|
|
|
|
return BaseAlias;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2009-11-26 10:17:34 +08:00
|
|
|
// Otherwise, we have a MustAlias. Since the base pointers alias each other
|
|
|
|
// exactly, see if the computed offset from the common pointer tells us
|
|
|
|
// about the relation of the resulting pointer.
|
2015-02-08 01:04:29 +08:00
|
|
|
// If we know the two GEPs are based off of the exact same pointer (and not
|
|
|
|
// just the same underlying object), see if that tells us anything about
|
|
|
|
// the resulting pointers.
|
Handle invariant.group.barrier in BasicAA
Summary:
llvm.invariant.group.barrier returns pointer that mustalias
pointer it takes. It can't be marked with `returned` attribute,
because it would be remove easily. The other reason is that
only Alias Analysis can know about this, because if any other
pass would know it, then the result would be replaced with it's
argument, which would be invalid.
We can think about returned pointer as something that mustalias, but
it doesn't have to be bitwise the same as the argument.
Reviewers: dberlin, chandlerc, hfinkel, sanjoy
Subscribers: reames, nlewycky, rsmith, anna, amharc
Differential Revision: https://reviews.llvm.org/D31585
llvm-svn: 301227
2017-04-25 03:37:17 +08:00
|
|
|
if (GEP1->getPointerOperand()->stripPointerCastsAndBarriers() ==
|
|
|
|
GEP2->getPointerOperand()->stripPointerCastsAndBarriers() &&
|
2017-04-19 06:00:54 +08:00
|
|
|
GEP1->getPointerOperandType() == GEP2->getPointerOperandType()) {
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
AliasResult R = aliasSameBasePointerGEPs(GEP1, V1Size, GEP2, V2Size, DL);
|
2015-02-08 01:04:29 +08:00
|
|
|
// If we couldn't find anything interesting, don't abandon just yet.
|
|
|
|
if (R != MayAlias)
|
|
|
|
return R;
|
|
|
|
}
|
|
|
|
|
2016-01-18 07:13:48 +08:00
|
|
|
// If the max search depth is reached, the result is undefined
|
2014-03-27 05:30:19 +08:00
|
|
|
if (GEP2MaxLookupReached || GEP1MaxLookupReached)
|
|
|
|
return MayAlias;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2009-11-26 10:17:34 +08:00
|
|
|
// Subtract the GEP2 pointer from the GEP1 pointer to find out their
|
|
|
|
// symbolic difference.
|
|
|
|
GEP1BaseOffset -= GEP2BaseOffset;
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
GetIndexDifference(DecompGEP1.VarIndices, DecompGEP2.VarIndices);
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2009-11-26 10:17:34 +08:00
|
|
|
} else {
|
|
|
|
// Check to see if these two pointers are related by the getelementptr
|
|
|
|
// instruction. If one pointer is a GEP with a non-zero index of the other
|
|
|
|
// pointer, we know they cannot alias.
|
2009-11-27 00:52:32 +08:00
|
|
|
|
|
|
|
// If both accesses are unknown size, we can't do anything useful here.
|
2015-06-17 15:21:38 +08:00
|
|
|
if (V1Size == MemoryLocation::UnknownSize &&
|
|
|
|
V2Size == MemoryLocation::UnknownSize)
|
2009-11-26 10:17:34 +08:00
|
|
|
return MayAlias;
|
2009-10-14 14:41:49 +08:00
|
|
|
|
2015-06-17 15:21:38 +08:00
|
|
|
AliasResult R = aliasCheck(UnderlyingV1, MemoryLocation::UnknownSize,
|
2017-01-28 00:12:22 +08:00
|
|
|
AAMDNodes(), V2, MemoryLocation::UnknownSize,
|
|
|
|
V2AAInfo, nullptr, UnderlyingV2);
|
2009-11-26 10:17:34 +08:00
|
|
|
if (R != MustAlias)
|
|
|
|
// If V2 may alias GEP base pointer, conservatively returns MayAlias.
|
|
|
|
// If V2 is known not to alias GEP base pointer, then the two values
|
2017-01-28 00:12:22 +08:00
|
|
|
// cannot alias per GEP semantics: "Any memory access must be done through
|
|
|
|
// a pointer value associated with an address range of the memory access,
|
|
|
|
// otherwise the behavior is undefined.".
|
2009-11-26 10:17:34 +08:00
|
|
|
return R;
|
|
|
|
|
2014-03-27 05:30:19 +08:00
|
|
|
// If the max search depth is reached the result is undefined
|
|
|
|
if (GEP1MaxLookupReached)
|
|
|
|
return MayAlias;
|
2009-11-26 10:14:59 +08:00
|
|
|
}
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2009-11-26 10:17:34 +08:00
|
|
|
// In the two GEP Case, if there is no difference in the offsets of the
|
|
|
|
// computed pointers, the resultant pointers are a must alias. This
|
2016-01-18 07:13:48 +08:00
|
|
|
// happens when we have two lexically identical GEP's (for example).
|
2009-11-26 10:17:34 +08:00
|
|
|
//
|
|
|
|
// In the other case, if we have getelementptr <ptr>, 0, 0, 0, 0, ... and V2
|
|
|
|
// must aliases the GEP, the end result is a must alias also.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
if (GEP1BaseOffset == 0 && DecompGEP1.VarIndices.empty())
|
2009-10-14 14:41:49 +08:00
|
|
|
return MustAlias;
|
|
|
|
|
2011-09-08 10:23:31 +08:00
|
|
|
// If there is a constant difference between the pointers, but the difference
|
|
|
|
// is less than the size of the associated memory object, then we know
|
|
|
|
// that the objects are partially overlapping. If the difference is
|
|
|
|
// greater, we know they do not overlap.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
if (GEP1BaseOffset != 0 && DecompGEP1.VarIndices.empty()) {
|
2011-09-08 10:23:31 +08:00
|
|
|
if (GEP1BaseOffset >= 0) {
|
2015-06-17 15:21:38 +08:00
|
|
|
if (V2Size != MemoryLocation::UnknownSize) {
|
2011-09-08 10:23:31 +08:00
|
|
|
if ((uint64_t)GEP1BaseOffset < V2Size)
|
|
|
|
return PartialAlias;
|
|
|
|
return NoAlias;
|
|
|
|
}
|
|
|
|
} else {
|
2014-01-16 12:53:18 +08:00
|
|
|
// We have the situation where:
|
|
|
|
// + +
|
|
|
|
// | BaseOffset |
|
|
|
|
// ---------------->|
|
|
|
|
// |-->V1Size |-------> V2Size
|
|
|
|
// GEP1 V2
|
|
|
|
// We need to know that V2Size is not unknown, otherwise we might have
|
|
|
|
// stripped a gep with negative index ('gep <ptr>, -1, ...).
|
2015-06-17 15:21:38 +08:00
|
|
|
if (V1Size != MemoryLocation::UnknownSize &&
|
|
|
|
V2Size != MemoryLocation::UnknownSize) {
|
2011-09-08 10:23:31 +08:00
|
|
|
if (-(uint64_t)GEP1BaseOffset < V1Size)
|
|
|
|
return PartialAlias;
|
|
|
|
return NoAlias;
|
|
|
|
}
|
|
|
|
}
|
2010-12-14 06:50:24 +08:00
|
|
|
}
|
|
|
|
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
if (!DecompGEP1.VarIndices.empty()) {
|
2011-09-08 10:37:07 +08:00
|
|
|
uint64_t Modulo = 0;
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
bool AllPositive = true;
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
for (unsigned i = 0, e = DecompGEP1.VarIndices.size(); i != e; ++i) {
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
|
|
|
// Try to distinguish something like &A[i][1] against &A[42][0].
|
|
|
|
// Grab the least significant bit set in any of the scales. We
|
|
|
|
// don't need std::abs here (even if the scale's negative) as we'll
|
|
|
|
// be ^'ing Modulo with itself later.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
Modulo |= (uint64_t)DecompGEP1.VarIndices[i].Scale;
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
|
|
|
if (AllPositive) {
|
|
|
|
// If the Value could change between cycles, then any reasoning about
|
|
|
|
// the Value this cycle may not hold in the next cycle. We'll just
|
|
|
|
// give up if we can't determine conditions that hold for every cycle:
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
const Value *V = DecompGEP1.VarIndices[i].V;
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
2017-05-15 14:39:41 +08:00
|
|
|
KnownBits Known = computeKnownBits(V, DL, 0, &AC, nullptr, DT);
|
|
|
|
bool SignKnownZero = Known.isNonNegative();
|
|
|
|
bool SignKnownOne = Known.isNegative();
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
|
|
|
// Zero-extension widens the variable, and so forces the sign
|
|
|
|
// bit to zero.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
bool IsZExt = DecompGEP1.VarIndices[i].ZExtBits > 0 || isa<ZExtInst>(V);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
SignKnownZero |= IsZExt;
|
|
|
|
SignKnownOne &= !IsZExt;
|
|
|
|
|
|
|
|
// If the variable begins with a zero then we know it's
|
|
|
|
// positive, regardless of whether the value is signed or
|
|
|
|
// unsigned.
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
int64_t Scale = DecompGEP1.VarIndices[i].Scale;
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
AllPositive =
|
|
|
|
(SignKnownZero && Scale >= 0) || (SignKnownOne && Scale < 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-09-08 10:37:07 +08:00
|
|
|
Modulo = Modulo ^ (Modulo & (Modulo - 1));
|
|
|
|
|
|
|
|
// We can compute the difference between the two addresses
|
|
|
|
// mod Modulo. Check whether that difference guarantees that the
|
|
|
|
// two locations do not alias.
|
|
|
|
uint64_t ModOffset = (uint64_t)GEP1BaseOffset & (Modulo - 1);
|
2015-06-17 15:21:38 +08:00
|
|
|
if (V1Size != MemoryLocation::UnknownSize &&
|
|
|
|
V2Size != MemoryLocation::UnknownSize && ModOffset >= V2Size &&
|
|
|
|
V1Size <= Modulo - ModOffset)
|
2011-09-08 10:37:07 +08:00
|
|
|
return NoAlias;
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
|
|
|
// If we know all the variables are positive, then GEP1 >= GEP1BasePtr.
|
|
|
|
// If GEP1BasePtr > V2 (GEP1BaseOffset > 0) then we know the pointers
|
|
|
|
// don't alias if V2Size can fit in the gap between V2 and GEP1BasePtr.
|
|
|
|
if (AllPositive && GEP1BaseOffset > 0 && V2Size <= (uint64_t)GEP1BaseOffset)
|
|
|
|
return NoAlias;
|
|
|
|
|
[BasicAA] Improve precision of alloca vs. inbounds GEP alias queries
If a we have (a) a GEP and (b) a pointer based on an alloca, and the
beginning of the object the GEP points would have a negative offset with
repsect to the alloca, then the GEP can not alias pointer (b).
For example, consider code like:
struct { int f0, int f1, ...} foo;
...
foo alloca;
foo *random = bar(alloca);
int *f0 = &alloca.f0
int *f1 = &random->f1;
Which is lowered, approximately, to:
%alloca = alloca %struct.foo
%random = call %struct.foo* @random(%struct.foo* %alloca)
%f0 = getelementptr inbounds %struct, %struct.foo* %alloca, i32 0, i32 0
%f1 = getelementptr inbounds %struct, %struct.foo* %random, i32 0, i32 1
Assume %f1 and %f0 alias. Then %f1 would point into the object allocated
by %alloca. Since the %f1 GEP is inbounds, that means %random must also
point into the same object. But since %f0 points to the beginning of %alloca,
the highest %f1 can be is (%alloca + 3). This means %random can not be higher
than (%alloca - 1), and so is not inbounds, a contradiction.
Differential Revision: http://reviews.llvm.org/D20495
llvm-svn: 270777
2016-05-26 06:23:08 +08:00
|
|
|
if (constantOffsetHeuristic(DecompGEP1.VarIndices, V1Size, V2Size,
|
2016-12-19 16:22:17 +08:00
|
|
|
GEP1BaseOffset, &AC, DT))
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
return NoAlias;
|
2011-09-08 10:37:07 +08:00
|
|
|
}
|
2011-09-08 10:23:31 +08:00
|
|
|
|
2011-06-04 14:50:18 +08:00
|
|
|
// Statically, we can see that the base objects are the same, but the
|
|
|
|
// pointers have dynamic offsets which we can't resolve. And none of our
|
|
|
|
// little tricks above worked.
|
|
|
|
//
|
|
|
|
// TODO: Returning PartialAlias instead of MayAlias is a mild hack; the
|
|
|
|
// practical effect of this is protecting TBAA in the case of dynamic
|
2012-02-18 02:33:38 +08:00
|
|
|
// indices into arrays of unions or malloc'd memory.
|
2011-06-04 14:50:18 +08:00
|
|
|
return PartialAlias;
|
2009-10-14 02:42:04 +08:00
|
|
|
}
|
|
|
|
|
2015-06-22 10:16:51 +08:00
|
|
|
static AliasResult MergeAliasResults(AliasResult A, AliasResult B) {
|
2011-06-04 04:17:36 +08:00
|
|
|
// If the results agree, take it.
|
|
|
|
if (A == B)
|
|
|
|
return A;
|
|
|
|
// A mix of PartialAlias and MustAlias is PartialAlias.
|
2015-06-22 10:16:51 +08:00
|
|
|
if ((A == PartialAlias && B == MustAlias) ||
|
|
|
|
(B == PartialAlias && A == MustAlias))
|
|
|
|
return PartialAlias;
|
2011-06-04 04:17:36 +08:00
|
|
|
// Otherwise, we don't know anything.
|
2015-06-22 10:16:51 +08:00
|
|
|
return MayAlias;
|
2011-06-04 04:17:36 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Provides a bunch of ad-hoc rules to disambiguate a Select instruction
|
|
|
|
/// against another.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
AliasResult BasicAAResult::aliasSelect(const SelectInst *SI, uint64_t SISize,
|
|
|
|
const AAMDNodes &SIAAInfo,
|
|
|
|
const Value *V2, uint64_t V2Size,
|
2016-08-13 00:05:03 +08:00
|
|
|
const AAMDNodes &V2AAInfo,
|
|
|
|
const Value *UnderV2) {
|
2009-10-27 05:55:43 +08:00
|
|
|
// If the values are Selects with the same condition, we can do a more precise
|
|
|
|
// check: just check for aliases between the values on corresponding arms.
|
|
|
|
if (const SelectInst *SI2 = dyn_cast<SelectInst>(V2))
|
|
|
|
if (SI->getCondition() == SI2->getCondition()) {
|
2015-08-06 15:57:58 +08:00
|
|
|
AliasResult Alias = aliasCheck(SI->getTrueValue(), SISize, SIAAInfo,
|
|
|
|
SI2->getTrueValue(), V2Size, V2AAInfo);
|
2009-10-27 05:55:43 +08:00
|
|
|
if (Alias == MayAlias)
|
|
|
|
return MayAlias;
|
|
|
|
AliasResult ThisAlias =
|
2015-08-06 15:57:58 +08:00
|
|
|
aliasCheck(SI->getFalseValue(), SISize, SIAAInfo,
|
|
|
|
SI2->getFalseValue(), V2Size, V2AAInfo);
|
2011-06-04 04:17:36 +08:00
|
|
|
return MergeAliasResults(ThisAlias, Alias);
|
2009-10-27 05:55:43 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// If both arms of the Select node NoAlias or MustAlias V2, then returns
|
|
|
|
// NoAlias / MustAlias. Otherwise, returns MayAlias.
|
|
|
|
AliasResult Alias =
|
2016-08-13 00:05:03 +08:00
|
|
|
aliasCheck(V2, V2Size, V2AAInfo, SI->getTrueValue(),
|
|
|
|
SISize, SIAAInfo, UnderV2);
|
2009-10-27 05:55:43 +08:00
|
|
|
if (Alias == MayAlias)
|
|
|
|
return MayAlias;
|
2010-06-29 05:16:52 +08:00
|
|
|
|
2009-10-27 05:55:43 +08:00
|
|
|
AliasResult ThisAlias =
|
2016-08-13 00:05:03 +08:00
|
|
|
aliasCheck(V2, V2Size, V2AAInfo, SI->getFalseValue(), SISize, SIAAInfo,
|
|
|
|
UnderV2);
|
2011-06-04 04:17:36 +08:00
|
|
|
return MergeAliasResults(ThisAlias, Alias);
|
2009-10-27 05:55:43 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Provide a bunch of ad-hoc rules to disambiguate a PHI instruction against
|
|
|
|
/// another.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
AliasResult BasicAAResult::aliasPHI(const PHINode *PN, uint64_t PNSize,
|
|
|
|
const AAMDNodes &PNAAInfo, const Value *V2,
|
2016-08-13 00:05:03 +08:00
|
|
|
uint64_t V2Size, const AAMDNodes &V2AAInfo,
|
|
|
|
const Value *UnderV2) {
|
2014-01-02 11:31:36 +08:00
|
|
|
// Track phi nodes we have visited. We use this information when we determine
|
|
|
|
// value equivalence.
|
|
|
|
VisitedPhiBBs.insert(PN->getParent());
|
|
|
|
|
2009-10-27 05:55:43 +08:00
|
|
|
// If the values are PHIs in the same block, we can do a more precise
|
|
|
|
// as well as efficient check: just check for aliases between the values
|
|
|
|
// on corresponding edges.
|
|
|
|
if (const PHINode *PN2 = dyn_cast<PHINode>(V2))
|
|
|
|
if (PN2->getParent() == PN->getParent()) {
|
2015-06-17 15:18:54 +08:00
|
|
|
LocPair Locs(MemoryLocation(PN, PNSize, PNAAInfo),
|
|
|
|
MemoryLocation(V2, V2Size, V2AAInfo));
|
BasicAA: Recognize cyclic NoAlias phis
Enhances basic alias analysis to recognize phis whose first incoming values are
NoAlias and whose other incoming values are just the phi node itself through
some amount of recursion.
Example: With this change basicaa reports that ptr_phi and ptr_phi2 do not alias
each other.
bb:
ptr = ptr2 + 1
loop:
ptr_phi = phi [bb, ptr], [loop, ptr_plus_one]
ptr2_phi = phi [bb, ptr2], [loop, ptr2_plus_one]
...
ptr_plus_one = gep ptr_phi, 1
ptr2_plus_one = gep ptr2_phi, 1
This enables the elimination of one load in code like the following:
extern int foo;
int test_noalias(int *ptr, int num, int* coeff) {
int *ptr2 = ptr;
int result = (*ptr++) * (*coeff--);
while (num--) {
*ptr2++ = *ptr;
result += (*coeff--) * (*ptr++);
}
*ptr = foo;
return result;
}
Part 2/2 of fix for PR13564.
llvm-svn: 163319
2012-09-06 22:41:53 +08:00
|
|
|
if (PN > V2)
|
|
|
|
std::swap(Locs.first, Locs.second);
|
2012-12-11 07:02:41 +08:00
|
|
|
// Analyse the PHIs' inputs under the assumption that the PHIs are
|
|
|
|
// NoAlias.
|
|
|
|
// If the PHIs are May/MustAlias there must be (recursively) an input
|
|
|
|
// operand from outside the PHIs' cycle that is MayAlias/MustAlias or
|
|
|
|
// there must be an operation on the PHIs within the PHIs' value cycle
|
|
|
|
// that causes a MayAlias.
|
|
|
|
// Pretend the phis do not alias.
|
|
|
|
AliasResult Alias = NoAlias;
|
|
|
|
assert(AliasCache.count(Locs) &&
|
|
|
|
"There must exist an entry for the phi node");
|
|
|
|
AliasResult OrigAliasResult = AliasCache[Locs];
|
|
|
|
AliasCache[Locs] = NoAlias;
|
BasicAA: Recognize cyclic NoAlias phis
Enhances basic alias analysis to recognize phis whose first incoming values are
NoAlias and whose other incoming values are just the phi node itself through
some amount of recursion.
Example: With this change basicaa reports that ptr_phi and ptr_phi2 do not alias
each other.
bb:
ptr = ptr2 + 1
loop:
ptr_phi = phi [bb, ptr], [loop, ptr_plus_one]
ptr2_phi = phi [bb, ptr2], [loop, ptr2_plus_one]
...
ptr_plus_one = gep ptr_phi, 1
ptr2_plus_one = gep ptr2_phi, 1
This enables the elimination of one load in code like the following:
extern int foo;
int test_noalias(int *ptr, int num, int* coeff) {
int *ptr2 = ptr;
int result = (*ptr++) * (*coeff--);
while (num--) {
*ptr2++ = *ptr;
result += (*coeff--) * (*ptr++);
}
*ptr = foo;
return result;
}
Part 2/2 of fix for PR13564.
llvm-svn: 163319
2012-09-06 22:41:53 +08:00
|
|
|
|
2012-11-17 10:33:15 +08:00
|
|
|
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
|
2009-10-27 05:55:43 +08:00
|
|
|
AliasResult ThisAlias =
|
2015-08-06 15:57:58 +08:00
|
|
|
aliasCheck(PN->getIncomingValue(i), PNSize, PNAAInfo,
|
|
|
|
PN2->getIncomingValueForBlock(PN->getIncomingBlock(i)),
|
|
|
|
V2Size, V2AAInfo);
|
2011-06-04 04:17:36 +08:00
|
|
|
Alias = MergeAliasResults(ThisAlias, Alias);
|
|
|
|
if (Alias == MayAlias)
|
|
|
|
break;
|
2009-10-27 05:55:43 +08:00
|
|
|
}
|
BasicAA: Recognize cyclic NoAlias phis
Enhances basic alias analysis to recognize phis whose first incoming values are
NoAlias and whose other incoming values are just the phi node itself through
some amount of recursion.
Example: With this change basicaa reports that ptr_phi and ptr_phi2 do not alias
each other.
bb:
ptr = ptr2 + 1
loop:
ptr_phi = phi [bb, ptr], [loop, ptr_plus_one]
ptr2_phi = phi [bb, ptr2], [loop, ptr2_plus_one]
...
ptr_plus_one = gep ptr_phi, 1
ptr2_plus_one = gep ptr2_phi, 1
This enables the elimination of one load in code like the following:
extern int foo;
int test_noalias(int *ptr, int num, int* coeff) {
int *ptr2 = ptr;
int result = (*ptr++) * (*coeff--);
while (num--) {
*ptr2++ = *ptr;
result += (*coeff--) * (*ptr++);
}
*ptr = foo;
return result;
}
Part 2/2 of fix for PR13564.
llvm-svn: 163319
2012-09-06 22:41:53 +08:00
|
|
|
|
|
|
|
// Reset if speculation failed.
|
2012-12-11 07:02:41 +08:00
|
|
|
if (Alias != NoAlias)
|
BasicAA: Recognize cyclic NoAlias phis
Enhances basic alias analysis to recognize phis whose first incoming values are
NoAlias and whose other incoming values are just the phi node itself through
some amount of recursion.
Example: With this change basicaa reports that ptr_phi and ptr_phi2 do not alias
each other.
bb:
ptr = ptr2 + 1
loop:
ptr_phi = phi [bb, ptr], [loop, ptr_plus_one]
ptr2_phi = phi [bb, ptr2], [loop, ptr2_plus_one]
...
ptr_plus_one = gep ptr_phi, 1
ptr2_plus_one = gep ptr2_phi, 1
This enables the elimination of one load in code like the following:
extern int foo;
int test_noalias(int *ptr, int num, int* coeff) {
int *ptr2 = ptr;
int result = (*ptr++) * (*coeff--);
while (num--) {
*ptr2++ = *ptr;
result += (*coeff--) * (*ptr++);
}
*ptr = foo;
return result;
}
Part 2/2 of fix for PR13564.
llvm-svn: 163319
2012-09-06 22:41:53 +08:00
|
|
|
AliasCache[Locs] = OrigAliasResult;
|
|
|
|
|
2009-10-27 05:55:43 +08:00
|
|
|
return Alias;
|
|
|
|
}
|
|
|
|
|
2015-08-06 15:57:58 +08:00
|
|
|
SmallPtrSet<Value *, 4> UniqueSrc;
|
|
|
|
SmallVector<Value *, 4> V1Srcs;
|
2015-07-16 03:32:22 +08:00
|
|
|
bool isRecursive = false;
|
2015-05-13 04:05:31 +08:00
|
|
|
for (Value *PV1 : PN->incoming_values()) {
|
2009-10-14 06:02:20 +08:00
|
|
|
if (isa<PHINode>(PV1))
|
|
|
|
// If any of the source itself is a PHI, return MayAlias conservatively
|
2009-10-14 14:41:49 +08:00
|
|
|
// to avoid compile time explosion. The worst possible case is if both
|
|
|
|
// sides are PHI nodes. In which case, this is O(m x n) time where 'm'
|
|
|
|
// and 'n' are the number of PHI sources.
|
2009-10-14 06:02:20 +08:00
|
|
|
return MayAlias;
|
2015-07-16 03:32:22 +08:00
|
|
|
|
|
|
|
if (EnableRecPhiAnalysis)
|
|
|
|
if (GEPOperator *PV1GEP = dyn_cast<GEPOperator>(PV1)) {
|
|
|
|
// Check whether the incoming value is a GEP that advances the pointer
|
|
|
|
// result of this PHI node (e.g. in a loop). If this is the case, we
|
|
|
|
// would recurse and always get a MayAlias. Handle this case specially
|
|
|
|
// below.
|
|
|
|
if (PV1GEP->getPointerOperand() == PN && PV1GEP->getNumIndices() == 1 &&
|
|
|
|
isa<ConstantInt>(PV1GEP->idx_begin())) {
|
|
|
|
isRecursive = true;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-11-19 15:49:26 +08:00
|
|
|
if (UniqueSrc.insert(PV1).second)
|
2009-10-14 06:02:20 +08:00
|
|
|
V1Srcs.push_back(PV1);
|
|
|
|
}
|
|
|
|
|
2015-07-16 03:32:22 +08:00
|
|
|
// If this PHI node is recursive, set the size of the accessed memory to
|
|
|
|
// unknown to represent all the possible values the GEP could advance the
|
|
|
|
// pointer to.
|
|
|
|
if (isRecursive)
|
|
|
|
PNSize = MemoryLocation::UnknownSize;
|
|
|
|
|
2015-08-06 15:57:58 +08:00
|
|
|
AliasResult Alias =
|
2016-08-13 00:05:03 +08:00
|
|
|
aliasCheck(V2, V2Size, V2AAInfo, V1Srcs[0],
|
|
|
|
PNSize, PNAAInfo, UnderV2);
|
2015-07-16 03:32:22 +08:00
|
|
|
|
2009-10-14 13:22:03 +08:00
|
|
|
// Early exit if the check of the first PHI source against V2 is MayAlias.
|
|
|
|
// Other results are not possible.
|
|
|
|
if (Alias == MayAlias)
|
|
|
|
return MayAlias;
|
|
|
|
|
2009-10-14 06:02:20 +08:00
|
|
|
// If all sources of the PHI node NoAlias or MustAlias V2, then returns
|
|
|
|
// NoAlias / MustAlias. Otherwise, returns MayAlias.
|
|
|
|
for (unsigned i = 1, e = V1Srcs.size(); i != e; ++i) {
|
|
|
|
Value *V = V1Srcs[i];
|
2009-10-27 05:55:43 +08:00
|
|
|
|
2015-08-06 15:57:58 +08:00
|
|
|
AliasResult ThisAlias =
|
2016-08-13 00:05:03 +08:00
|
|
|
aliasCheck(V2, V2Size, V2AAInfo, V, PNSize, PNAAInfo, UnderV2);
|
2011-06-04 04:17:36 +08:00
|
|
|
Alias = MergeAliasResults(ThisAlias, Alias);
|
|
|
|
if (Alias == MayAlias)
|
|
|
|
break;
|
2009-10-14 06:02:20 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
return Alias;
|
|
|
|
}
|
|
|
|
|
2015-11-17 16:15:08 +08:00
|
|
|
/// Provides a bunch of ad-hoc rules to disambiguate in common cases, such as
|
2015-08-06 16:17:06 +08:00
|
|
|
/// array references.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
AliasResult BasicAAResult::aliasCheck(const Value *V1, uint64_t V1Size,
|
|
|
|
AAMDNodes V1AAInfo, const Value *V2,
|
2016-08-13 00:05:03 +08:00
|
|
|
uint64_t V2Size, AAMDNodes V2AAInfo,
|
|
|
|
const Value *O1, const Value *O2) {
|
2010-04-09 02:11:50 +08:00
|
|
|
// If either of the memory references is empty, it doesn't matter what the
|
|
|
|
// pointer values are.
|
|
|
|
if (V1Size == 0 || V2Size == 0)
|
|
|
|
return NoAlias;
|
|
|
|
|
2009-10-14 02:42:04 +08:00
|
|
|
// Strip off any casts if they exist.
|
Handle invariant.group.barrier in BasicAA
Summary:
llvm.invariant.group.barrier returns pointer that mustalias
pointer it takes. It can't be marked with `returned` attribute,
because it would be remove easily. The other reason is that
only Alias Analysis can know about this, because if any other
pass would know it, then the result would be replaced with it's
argument, which would be invalid.
We can think about returned pointer as something that mustalias, but
it doesn't have to be bitwise the same as the argument.
Reviewers: dberlin, chandlerc, hfinkel, sanjoy
Subscribers: reames, nlewycky, rsmith, anna, amharc
Differential Revision: https://reviews.llvm.org/D31585
llvm-svn: 301227
2017-04-25 03:37:17 +08:00
|
|
|
V1 = V1->stripPointerCastsAndBarriers();
|
|
|
|
V2 = V2->stripPointerCastsAndBarriers();
|
2009-10-14 02:42:04 +08:00
|
|
|
|
2015-05-06 02:10:49 +08:00
|
|
|
// If V1 or V2 is undef, the result is NoAlias because we can always pick a
|
|
|
|
// value for undef that aliases nothing in the program.
|
|
|
|
if (isa<UndefValue>(V1) || isa<UndefValue>(V2))
|
|
|
|
return NoAlias;
|
|
|
|
|
2009-10-14 02:42:04 +08:00
|
|
|
// Are we checking for alias of the same value?
|
2016-01-18 07:13:48 +08:00
|
|
|
// Because we look 'through' phi nodes, we could look at "Value" pointers from
|
2014-01-03 13:47:03 +08:00
|
|
|
// different iterations. We must therefore make sure that this is not the
|
|
|
|
// case. The function isValueEqualInPotentialCycles ensures that this cannot
|
|
|
|
// happen by looking at the visited phi nodes and making sure they cannot
|
|
|
|
// reach the value.
|
|
|
|
if (isValueEqualInPotentialCycles(V1, V2))
|
|
|
|
return MustAlias;
|
2009-10-14 02:42:04 +08:00
|
|
|
|
2010-02-16 19:11:14 +08:00
|
|
|
if (!V1->getType()->isPointerTy() || !V2->getType()->isPointerTy())
|
2015-08-06 15:57:58 +08:00
|
|
|
return NoAlias; // Scalars cannot alias each other
|
2009-10-14 02:42:04 +08:00
|
|
|
|
|
|
|
// Figure out what objects these things are pointing to if we can.
|
2016-08-13 00:05:03 +08:00
|
|
|
if (O1 == nullptr)
|
|
|
|
O1 = GetUnderlyingObject(V1, DL, MaxLookupSearchDepth);
|
|
|
|
|
|
|
|
if (O2 == nullptr)
|
|
|
|
O2 = GetUnderlyingObject(V2, DL, MaxLookupSearchDepth);
|
2009-10-14 02:42:04 +08:00
|
|
|
|
2009-11-10 03:29:11 +08:00
|
|
|
// Null values in the default address space don't point to any object, so they
|
|
|
|
// don't alias any other pointer.
|
|
|
|
if (const ConstantPointerNull *CPN = dyn_cast<ConstantPointerNull>(O1))
|
|
|
|
if (CPN->getType()->getAddressSpace() == 0)
|
|
|
|
return NoAlias;
|
|
|
|
if (const ConstantPointerNull *CPN = dyn_cast<ConstantPointerNull>(O2))
|
|
|
|
if (CPN->getType()->getAddressSpace() == 0)
|
|
|
|
return NoAlias;
|
|
|
|
|
2009-10-14 02:42:04 +08:00
|
|
|
if (O1 != O2) {
|
2016-01-18 07:13:48 +08:00
|
|
|
// If V1/V2 point to two different objects, we know that we have no alias.
|
2010-07-07 22:27:09 +08:00
|
|
|
if (isIdentifiedObject(O1) && isIdentifiedObject(O2))
|
2009-10-14 02:42:04 +08:00
|
|
|
return NoAlias;
|
2009-11-14 14:15:14 +08:00
|
|
|
|
|
|
|
// Constant pointers can't alias with non-const isIdentifiedObject objects.
|
2010-07-07 22:27:09 +08:00
|
|
|
if ((isa<Constant>(O1) && isIdentifiedObject(O2) && !isa<Constant>(O2)) ||
|
|
|
|
(isa<Constant>(O2) && isIdentifiedObject(O1) && !isa<Constant>(O1)))
|
2009-11-14 14:15:14 +08:00
|
|
|
return NoAlias;
|
|
|
|
|
2013-05-28 16:17:48 +08:00
|
|
|
// Function arguments can't alias with things that are known to be
|
|
|
|
// unambigously identified at the function level.
|
|
|
|
if ((isa<Argument>(O1) && isIdentifiedFunctionLocal(O2)) ||
|
|
|
|
(isa<Argument>(O2) && isIdentifiedFunctionLocal(O1)))
|
2010-07-02 04:08:40 +08:00
|
|
|
return NoAlias;
|
2009-10-14 02:42:04 +08:00
|
|
|
|
|
|
|
// Most objects can't alias null.
|
2010-07-07 22:27:09 +08:00
|
|
|
if ((isa<ConstantPointerNull>(O2) && isKnownNonNull(O1)) ||
|
|
|
|
(isa<ConstantPointerNull>(O1) && isKnownNonNull(O2)))
|
2009-10-14 02:42:04 +08:00
|
|
|
return NoAlias;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2010-07-07 22:30:04 +08:00
|
|
|
// If one pointer is the result of a call/invoke or load and the other is a
|
|
|
|
// non-escaping local object within the same function, then we know the
|
|
|
|
// object couldn't escape to a point where the call could return it.
|
|
|
|
//
|
|
|
|
// Note that if the pointers are in different functions, there are a
|
|
|
|
// variety of complications. A call with a nocapture argument may still
|
|
|
|
// temporary store the nocapture argument's value in a temporary memory
|
|
|
|
// location if that memory location doesn't escape. Or it may pass a
|
|
|
|
// nocapture value to other functions as long as they don't capture it.
|
|
|
|
if (isEscapeSource(O1) && isNonEscapingLocalObject(O2))
|
|
|
|
return NoAlias;
|
|
|
|
if (isEscapeSource(O2) && isNonEscapingLocalObject(O1))
|
|
|
|
return NoAlias;
|
|
|
|
}
|
|
|
|
|
2009-10-14 02:42:04 +08:00
|
|
|
// If the size of one access is larger than the entire object on the other
|
|
|
|
// side, then we know such behavior is undefined and can assume no alias.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
if ((V1Size != MemoryLocation::UnknownSize &&
|
|
|
|
isObjectSmallerThan(O2, V1Size, DL, TLI)) ||
|
|
|
|
(V2Size != MemoryLocation::UnknownSize &&
|
|
|
|
isObjectSmallerThan(O1, V2Size, DL, TLI)))
|
|
|
|
return NoAlias;
|
2013-08-24 22:16:00 +08:00
|
|
|
|
2011-06-04 08:31:50 +08:00
|
|
|
// Check the cache before climbing up use-def chains. This also terminates
|
|
|
|
// otherwise infinitely recursive queries.
|
2015-06-17 15:18:54 +08:00
|
|
|
LocPair Locs(MemoryLocation(V1, V1Size, V1AAInfo),
|
|
|
|
MemoryLocation(V2, V2Size, V2AAInfo));
|
2011-06-04 08:31:50 +08:00
|
|
|
if (V1 > V2)
|
|
|
|
std::swap(Locs.first, Locs.second);
|
|
|
|
std::pair<AliasCacheTy::iterator, bool> Pair =
|
2015-08-06 15:57:58 +08:00
|
|
|
AliasCache.insert(std::make_pair(Locs, MayAlias));
|
2011-06-04 08:31:50 +08:00
|
|
|
if (!Pair.second)
|
|
|
|
return Pair.first->second;
|
|
|
|
|
2009-11-26 10:13:03 +08:00
|
|
|
// FIXME: This isn't aggressively handling alias(GEP, PHI) for example: if the
|
|
|
|
// GEP can't simplify, we don't even look at the PHI cases.
|
2009-10-18 07:48:54 +08:00
|
|
|
if (!isa<GEPOperator>(V1) && isa<GEPOperator>(V2)) {
|
2009-10-14 02:42:04 +08:00
|
|
|
std::swap(V1, V2);
|
|
|
|
std::swap(V1Size, V2Size);
|
2009-11-26 10:14:59 +08:00
|
|
|
std::swap(O1, O2);
|
2014-07-24 20:16:19 +08:00
|
|
|
std::swap(V1AAInfo, V2AAInfo);
|
2009-10-14 02:42:04 +08:00
|
|
|
}
|
2010-10-19 02:04:47 +08:00
|
|
|
if (const GEPOperator *GV1 = dyn_cast<GEPOperator>(V1)) {
|
2015-08-06 15:57:58 +08:00
|
|
|
AliasResult Result =
|
|
|
|
aliasGEP(GV1, V1Size, V1AAInfo, V2, V2Size, V2AAInfo, O1, O2);
|
|
|
|
if (Result != MayAlias)
|
|
|
|
return AliasCache[Locs] = Result;
|
2010-10-19 02:04:47 +08:00
|
|
|
}
|
2009-10-14 06:02:20 +08:00
|
|
|
|
|
|
|
if (isa<PHINode>(V2) && !isa<PHINode>(V1)) {
|
|
|
|
std::swap(V1, V2);
|
2016-08-13 00:05:03 +08:00
|
|
|
std::swap(O1, O2);
|
2009-10-14 06:02:20 +08:00
|
|
|
std::swap(V1Size, V2Size);
|
2014-07-24 20:16:19 +08:00
|
|
|
std::swap(V1AAInfo, V2AAInfo);
|
2009-10-14 06:02:20 +08:00
|
|
|
}
|
2010-10-19 02:04:47 +08:00
|
|
|
if (const PHINode *PN = dyn_cast<PHINode>(V1)) {
|
2016-08-13 00:05:03 +08:00
|
|
|
AliasResult Result = aliasPHI(PN, V1Size, V1AAInfo,
|
|
|
|
V2, V2Size, V2AAInfo, O2);
|
2015-08-06 15:57:58 +08:00
|
|
|
if (Result != MayAlias)
|
|
|
|
return AliasCache[Locs] = Result;
|
2010-10-19 02:04:47 +08:00
|
|
|
}
|
2005-04-22 05:13:18 +08:00
|
|
|
|
2009-10-27 05:55:43 +08:00
|
|
|
if (isa<SelectInst>(V2) && !isa<SelectInst>(V1)) {
|
|
|
|
std::swap(V1, V2);
|
2016-08-13 00:05:03 +08:00
|
|
|
std::swap(O1, O2);
|
2009-10-27 05:55:43 +08:00
|
|
|
std::swap(V1Size, V2Size);
|
2014-07-24 20:16:19 +08:00
|
|
|
std::swap(V1AAInfo, V2AAInfo);
|
2009-10-27 05:55:43 +08:00
|
|
|
}
|
2010-10-19 02:04:47 +08:00
|
|
|
if (const SelectInst *S1 = dyn_cast<SelectInst>(V1)) {
|
2015-08-06 15:57:58 +08:00
|
|
|
AliasResult Result =
|
2016-08-13 00:05:03 +08:00
|
|
|
aliasSelect(S1, V1Size, V1AAInfo, V2, V2Size, V2AAInfo, O2);
|
2015-08-06 15:57:58 +08:00
|
|
|
if (Result != MayAlias)
|
|
|
|
return AliasCache[Locs] = Result;
|
2010-10-19 02:04:47 +08:00
|
|
|
}
|
2009-10-27 05:55:43 +08:00
|
|
|
|
2011-01-19 05:16:06 +08:00
|
|
|
// If both pointers are pointing into the same object and one of them
|
2016-01-18 07:13:48 +08:00
|
|
|
// accesses the entire object, then the accesses must overlap in some way.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
if (O1 == O2)
|
2015-06-17 15:21:38 +08:00
|
|
|
if ((V1Size != MemoryLocation::UnknownSize &&
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
isObjectSize(O1, V1Size, DL, TLI)) ||
|
2015-06-17 15:21:38 +08:00
|
|
|
(V2Size != MemoryLocation::UnknownSize &&
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
isObjectSize(O2, V2Size, DL, TLI)))
|
2011-06-04 08:31:50 +08:00
|
|
|
return AliasCache[Locs] = PartialAlias;
|
2011-01-19 05:16:06 +08:00
|
|
|
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
// Recurse back into the best AA results we have, potentially with refined
|
|
|
|
// memory locations. We have already ensured that BasicAA has a MayAlias
|
|
|
|
// cache result for these, so any recursion back into BasicAA won't loop.
|
|
|
|
AliasResult Result = getBestAAResults().alias(Locs.first, Locs.second);
|
2011-06-04 08:31:50 +08:00
|
|
|
return AliasCache[Locs] = Result;
|
2003-02-27 03:41:54 +08:00
|
|
|
}
|
2014-01-02 11:31:36 +08:00
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Check whether two Values can be considered equivalent.
|
|
|
|
///
|
|
|
|
/// In addition to pointer equivalence of \p V1 and \p V2 this checks whether
|
|
|
|
/// they can not be part of a cycle in the value graph by looking at all
|
|
|
|
/// visited phi nodes an making sure that the phis cannot reach the value. We
|
|
|
|
/// have to do this because we are looking through phi nodes (That is we say
|
|
|
|
/// noalias(V, phi(VA, VB)) if noalias(V, VA) and noalias(V, VB).
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
bool BasicAAResult::isValueEqualInPotentialCycles(const Value *V,
|
|
|
|
const Value *V2) {
|
2014-01-02 11:31:36 +08:00
|
|
|
if (V != V2)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
const Instruction *Inst = dyn_cast<Instruction>(V);
|
|
|
|
if (!Inst)
|
|
|
|
return true;
|
|
|
|
|
2015-03-21 02:05:49 +08:00
|
|
|
if (VisitedPhiBBs.empty())
|
|
|
|
return true;
|
|
|
|
|
2014-01-03 13:47:03 +08:00
|
|
|
if (VisitedPhiBBs.size() > MaxNumPhiBBsValueReachabilityCheck)
|
|
|
|
return false;
|
2014-01-02 11:31:36 +08:00
|
|
|
|
2014-01-03 13:47:03 +08:00
|
|
|
// Make sure that the visited phis cannot reach the Value. This ensures that
|
|
|
|
// the Values cannot come from different iterations of a potential cycle the
|
|
|
|
// phi nodes could be involved in.
|
2014-08-25 07:23:06 +08:00
|
|
|
for (auto *P : VisitedPhiBBs)
|
Analysis: Remove implicit ilist iterator conversions
Remove implicit ilist iterator conversions from LLVMAnalysis.
I came across something really scary in `llvm::isKnownNotFullPoison()`
which relied on `Instruction::getNextNode()` being completely broken
(not surprising, but scary nevertheless). This function is documented
(and coded to) return `nullptr` when it gets to the sentinel, but with
an `ilist_half_node` as a sentinel, the sentinel check looks into some
other memory and we don't recognize we've hit the end.
Rooting out these scary cases is the reason I'm removing the implicit
conversions before doing anything else with `ilist`; I'm not at all
surprised that clients rely on badness.
I found another scary case -- this time, not relying on badness, just
bad (but I guess getting lucky so far) -- in
`ObjectSizeOffsetEvaluator::compute_()`. Here, we save out the
insertion point, do some things, and then restore it. Previously, we
let the iterator auto-convert to `Instruction*`, and then set it back
using the `Instruction*` version:
Instruction *PrevInsertPoint = Builder.GetInsertPoint();
/* Logic that may change insert point */
if (PrevInsertPoint)
Builder.SetInsertPoint(PrevInsertPoint);
The check for `PrevInsertPoint` doesn't protect correctly against bad
accesses. If the insertion point has been set to the end of a basic
block (i.e., `SetInsertPoint(SomeBB)`), then `GetInsertPoint()` returns
an iterator pointing at the list sentinel. The version of
`SetInsertPoint()` that's getting called will then call
`PrevInsertPoint->getParent()`, which explodes horribly. The only
reason this hasn't blown up is that it's fairly unlikely the builder is
adding to the end of the block; usually, we're adding instructions
somewhere before the terminator.
llvm-svn: 249925
2015-10-10 08:53:03 +08:00
|
|
|
if (isPotentiallyReachable(&P->front(), Inst, DT, LI))
|
2014-01-03 13:47:03 +08:00
|
|
|
return false;
|
2014-01-02 11:31:36 +08:00
|
|
|
|
2014-01-03 13:47:03 +08:00
|
|
|
return true;
|
2014-01-02 11:31:36 +08:00
|
|
|
}
|
|
|
|
|
2015-08-06 16:17:06 +08:00
|
|
|
/// Computes the symbolic difference between two de-composed GEPs.
|
|
|
|
///
|
|
|
|
/// Dest and Src are the variable indices from two decomposed GetElementPtr
|
|
|
|
/// instructions GEP1 and GEP2 which have common base pointers.
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
void BasicAAResult::GetIndexDifference(
|
2014-01-02 11:31:36 +08:00
|
|
|
SmallVectorImpl<VariableGEPIndex> &Dest,
|
|
|
|
const SmallVectorImpl<VariableGEPIndex> &Src) {
|
|
|
|
if (Src.empty())
|
|
|
|
return;
|
|
|
|
|
|
|
|
for (unsigned i = 0, e = Src.size(); i != e; ++i) {
|
|
|
|
const Value *V = Src[i].V;
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
unsigned ZExtBits = Src[i].ZExtBits, SExtBits = Src[i].SExtBits;
|
2014-01-02 11:31:36 +08:00
|
|
|
int64_t Scale = Src[i].Scale;
|
|
|
|
|
|
|
|
// Find V in Dest. This is N^2, but pointer indices almost never have more
|
|
|
|
// than a few variable indexes.
|
|
|
|
for (unsigned j = 0, e = Dest.size(); j != e; ++j) {
|
2014-01-03 13:47:03 +08:00
|
|
|
if (!isValueEqualInPotentialCycles(Dest[j].V, V) ||
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
Dest[j].ZExtBits != ZExtBits || Dest[j].SExtBits != SExtBits)
|
2014-01-02 11:31:36 +08:00
|
|
|
continue;
|
|
|
|
|
|
|
|
// If we found it, subtract off Scale V's from the entry in Dest. If it
|
|
|
|
// goes to zero, remove the entry.
|
|
|
|
if (Dest[j].Scale != Scale)
|
|
|
|
Dest[j].Scale -= Scale;
|
|
|
|
else
|
|
|
|
Dest.erase(Dest.begin() + j);
|
|
|
|
Scale = 0;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
// If we didn't consume this entry, add it to the end of the Dest list.
|
|
|
|
if (Scale) {
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
VariableGEPIndex Entry = {V, ZExtBits, SExtBits, -Scale};
|
2014-01-02 11:31:36 +08:00
|
|
|
Dest.push_back(Entry);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
bool BasicAAResult::constantOffsetHeuristic(
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
const SmallVectorImpl<VariableGEPIndex> &VarIndices, uint64_t V1Size,
|
2016-12-19 16:22:17 +08:00
|
|
|
uint64_t V2Size, int64_t BaseOffset, AssumptionCache *AC,
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
DominatorTree *DT) {
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
if (VarIndices.size() != 2 || V1Size == MemoryLocation::UnknownSize ||
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
V2Size == MemoryLocation::UnknownSize)
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
return false;
|
|
|
|
|
|
|
|
const VariableGEPIndex &Var0 = VarIndices[0], &Var1 = VarIndices[1];
|
|
|
|
|
|
|
|
if (Var0.ZExtBits != Var1.ZExtBits || Var0.SExtBits != Var1.SExtBits ||
|
|
|
|
Var0.Scale != -Var1.Scale)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
unsigned Width = Var1.V->getType()->getIntegerBitWidth();
|
|
|
|
|
|
|
|
// We'll strip off the Extensions of Var0 and Var1 and do another round
|
|
|
|
// of GetLinearExpression decomposition. In the example above, if Var0
|
|
|
|
// is zext(%x + 1) we should get V1 == %x and V1Offset == 1.
|
|
|
|
|
|
|
|
APInt V0Scale(Width, 0), V0Offset(Width, 0), V1Scale(Width, 0),
|
|
|
|
V1Offset(Width, 0);
|
|
|
|
bool NSW = true, NUW = true;
|
|
|
|
unsigned V0ZExtBits = 0, V0SExtBits = 0, V1ZExtBits = 0, V1SExtBits = 0;
|
|
|
|
const Value *V0 = GetLinearExpression(Var0.V, V0Scale, V0Offset, V0ZExtBits,
|
2016-12-19 16:22:17 +08:00
|
|
|
V0SExtBits, DL, 0, AC, DT, NSW, NUW);
|
2016-02-19 06:09:30 +08:00
|
|
|
NSW = true;
|
|
|
|
NUW = true;
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
const Value *V1 = GetLinearExpression(Var1.V, V1Scale, V1Offset, V1ZExtBits,
|
2016-12-19 16:22:17 +08:00
|
|
|
V1SExtBits, DL, 0, AC, DT, NSW, NUW);
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
|
|
|
|
if (V0Scale != V1Scale || V0ZExtBits != V1ZExtBits ||
|
|
|
|
V0SExtBits != V1SExtBits || !isValueEqualInPotentialCycles(V0, V1))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
// We have a hit - Var0 and Var1 only differ by a constant offset!
|
|
|
|
|
|
|
|
// If we've been sext'ed then zext'd the maximum difference between Var0 and
|
|
|
|
// Var1 is possible to calculate, but we're just interested in the absolute
|
2015-10-24 19:38:01 +08:00
|
|
|
// minimum difference between the two. The minimum distance may occur due to
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
// wrapping; consider "add i3 %i, 5": if %i == 7 then 7 + 5 mod 8 == 4, and so
|
|
|
|
// the minimum distance between %i and %i + 5 is 3.
|
2015-10-24 19:38:01 +08:00
|
|
|
APInt MinDiff = V0Offset - V1Offset, Wrapped = -MinDiff;
|
[BasicAA] Fix the handling of sext and zext in the analysis of GEPs.
Hopefully this will end the GEPs saga!
This commit reverts r245394, i.e., it reapplies r221876 while incorporating the
fixes from D11847.
r221876 was not reapplied alone because it was not safe and D11847 was not
applied alone because it needs r221876 to produce correct results.
This should fix PR24596.
Original commit message for r221876:
Let's try this again...
This reverts r219432, plus a bug fix.
Description of the bug in r219432 (by Nick):
The bug was using AllPositive to break out of the loop; if the loop break
condition i != e is changed to i != e && AllPositive then the
test_modulo_analysis_with_global test I've added will fail as the Modulo will
be calculated incorrectly (as the last loop iteration is skipped, so Modulo
isn't updated with its Scale).
Nick also adds this comment:
ComputeSignBit is safe to use in loops as it takes into account phi nodes, and
the == EK_ZeroEx check is safe in loops as, no matter how the variable changes
between iterations, zero-extensions will always guarantee a zero sign bit. The
isValueEqualInPotentialCycles check is therefore definitely not needed as all
the variable analysis holds no matter how the variables change between loop
iterations.
And this patch also adds another enhancement to GetLinearExpression - basically
to convert ConstantInts to Offsets (see test_const_eval and
test_const_eval_scaled for the situations this improves).
Original commit message:
This reverts r218944, which reverted r218714, plus a bug fix.
Description of the bug in r218714 (by Nick):
The original patch forgot to check if the Scale in VariableGEPIndex flipped the
sign of the variable. The BasicAA pass iterates over the instructions in the
order they appear in the function, and so BasicAliasAnalysis::aliasGEP is
called with the variable it first comes across as parameter GEP1. Adding a
%reorder label puts the definition of %a after %b so aliasGEP is called with %b
as the first parameter and %a as the second. aliasGEP later calculates that %a
== %b + 1 - %idxprom where %idxprom >= 0 (if %a was passed as the first
parameter it would calculate %b == %a - 1 + %idxprom where %idxprom >= 0) -
ignoring that %idxprom is scaled by -1 here lead the patch to incorrectly
conclude that %a > %b.
Revised patch by Nick White, thanks! Thanks to Lang to isolating the bug.
Slightly modified by me to add an early exit from the loop and avoid
unnecessary, but expensive, function calls.
Original commit message:
Two related things:
1. Fixes a bug when calculating the offset in GetLinearExpression. The code
previously used zext to extend the offset, so negative offsets were converted
to large positive ones.
2. Enhance aliasGEP to deduce that, if the difference between two GEP
allocations is positive and all the variables that govern the offset are also
positive (i.e. the offset is strictly after the higher base pointer), then
locations that fit in the gap between the two base pointers are NoAlias.
Patch by Nick White!
Message from D11847:
Un-revert of r241981 and fix for PR23626. The 'Or' case of GetLinearExpression
delegates to 'Add' if possible, and if not it returns an Opaque value.
Unfortunately the Scale and Offsets weren't being set (and so defaulted to 0) -
and a scale of zero effectively removes the variable from the GEP instruction.
This meant that BasicAA would return MustAliases when it should have been
returning PartialAliases (and PR23626 was an example of the GVN pass using an
incorrect MustAlias to merge loads from what should have been different
pointers).
Differential Revision: http://reviews.llvm.org/D11847
Patch by Nick White <n.j.white@gmail.com>!
llvm-svn: 246502
2015-09-01 06:32:47 +08:00
|
|
|
MinDiff = APIntOps::umin(MinDiff, Wrapped);
|
|
|
|
uint64_t MinDiffBytes = MinDiff.getZExtValue() * std::abs(Var0.Scale);
|
|
|
|
|
|
|
|
// We can't definitely say whether GEP1 is before or after V2 due to wrapping
|
|
|
|
// arithmetic (i.e. for some values of GEP1 and V2 GEP1 < V2, and for other
|
|
|
|
// values GEP1 > V2). We'll therefore only declare NoAlias if both V1Size and
|
|
|
|
// V2Size can fit in the MinDiffBytes gap.
|
|
|
|
return V1Size + std::abs(BaseOffset) <= MinDiffBytes &&
|
|
|
|
V2Size + std::abs(BaseOffset) <= MinDiffBytes;
|
|
|
|
}
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// BasicAliasAnalysis Pass
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2016-11-24 01:53:26 +08:00
|
|
|
AnalysisKey BasicAA::Key;
|
2016-03-11 18:22:49 +08:00
|
|
|
|
2016-08-09 08:28:15 +08:00
|
|
|
BasicAAResult BasicAA::run(Function &F, FunctionAnalysisManager &AM) {
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
return BasicAAResult(F.getParent()->getDataLayout(),
|
2016-03-11 19:05:24 +08:00
|
|
|
AM.getResult<TargetLibraryAnalysis>(F),
|
2016-12-19 16:22:17 +08:00
|
|
|
AM.getResult<AssumptionAnalysis>(F),
|
2016-03-11 21:53:18 +08:00
|
|
|
&AM.getResult<DominatorTreeAnalysis>(F),
|
2016-03-11 19:05:24 +08:00
|
|
|
AM.getCachedResult<LoopAnalysis>(F));
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
}
|
|
|
|
|
2015-10-27 05:22:58 +08:00
|
|
|
BasicAAWrapperPass::BasicAAWrapperPass() : FunctionPass(ID) {
|
|
|
|
initializeBasicAAWrapperPassPass(*PassRegistry::getPassRegistry());
|
|
|
|
}
|
|
|
|
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
char BasicAAWrapperPass::ID = 0;
|
|
|
|
void BasicAAWrapperPass::anchor() {}
|
|
|
|
|
|
|
|
INITIALIZE_PASS_BEGIN(BasicAAWrapperPass, "basicaa",
|
|
|
|
"Basic Alias Analysis (stateless AA impl)", true, true)
|
2016-12-19 16:22:17 +08:00
|
|
|
INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
|
2016-03-11 21:53:18 +08:00
|
|
|
INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
|
|
|
|
INITIALIZE_PASS_END(BasicAAWrapperPass, "basicaa",
|
|
|
|
"Basic Alias Analysis (stateless AA impl)", true, true)
|
|
|
|
|
|
|
|
FunctionPass *llvm::createBasicAAWrapperPass() {
|
|
|
|
return new BasicAAWrapperPass();
|
|
|
|
}
|
|
|
|
|
|
|
|
bool BasicAAWrapperPass::runOnFunction(Function &F) {
|
2016-12-19 16:22:17 +08:00
|
|
|
auto &ACT = getAnalysis<AssumptionCacheTracker>();
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
auto &TLIWP = getAnalysis<TargetLibraryInfoWrapperPass>();
|
2016-03-11 21:53:18 +08:00
|
|
|
auto &DTWP = getAnalysis<DominatorTreeWrapperPass>();
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
auto *LIWP = getAnalysisIfAvailable<LoopInfoWrapperPass>();
|
|
|
|
|
|
|
|
Result.reset(new BasicAAResult(F.getParent()->getDataLayout(), TLIWP.getTLI(),
|
2016-12-19 16:22:17 +08:00
|
|
|
ACT.getAssumptionCache(F), &DTWP.getDomTree(),
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
LIWP ? &LIWP->getLoopInfo() : nullptr));
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
void BasicAAWrapperPass::getAnalysisUsage(AnalysisUsage &AU) const {
|
|
|
|
AU.setPreservesAll();
|
2016-12-19 16:22:17 +08:00
|
|
|
AU.addRequired<AssumptionCacheTracker>();
|
2016-03-11 21:53:18 +08:00
|
|
|
AU.addRequired<DominatorTreeWrapperPass>();
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
AU.addRequired<TargetLibraryInfoWrapperPass>();
|
|
|
|
}
|
|
|
|
|
|
|
|
BasicAAResult llvm::createLegacyPMBasicAAResult(Pass &P, Function &F) {
|
|
|
|
return BasicAAResult(
|
|
|
|
F.getParent()->getDataLayout(),
|
2016-12-19 16:22:17 +08:00
|
|
|
P.getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(),
|
|
|
|
P.getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F));
|
[PM/AA] Rebuild LLVM's alias analysis infrastructure in a way compatible
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
2015-09-10 01:55:00 +08:00
|
|
|
}
|