llvm/lib/Transforms/Utils/AddDiscriminators.cpp
Duncan P. N. Exon Smith dad20b2ae2 IR: Split Metadata from Value
Split `Metadata` away from the `Value` class hierarchy, as part of
PR21532.  Assembly and bitcode changes are in the wings, but this is the
bulk of the change for the IR C++ API.

I have a follow-up patch prepared for `clang`.  If this breaks other
sub-projects, I apologize in advance :(.  Help me compile it on Darwin
I'll try to fix it.  FWIW, the errors should be easy to fix, so it may
be simpler to just fix it yourself.

This breaks the build for all metadata-related code that's out-of-tree.
Rest assured the transition is mechanical and the compiler should catch
almost all of the problems.

Here's a quick guide for updating your code:

  - `Metadata` is the root of a class hierarchy with three main classes:
    `MDNode`, `MDString`, and `ValueAsMetadata`.  It is distinct from
    the `Value` class hierarchy.  It is typeless -- i.e., instances do
    *not* have a `Type`.

  - `MDNode`'s operands are all `Metadata *` (instead of `Value *`).

  - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be
    replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively.

    If you're referring solely to resolved `MDNode`s -- post graph
    construction -- just use `MDNode*`.

  - `MDNode` (and the rest of `Metadata`) have only limited support for
    `replaceAllUsesWith()`.

    As long as an `MDNode` is pointing at a forward declaration -- the
    result of `MDNode::getTemporary()` -- it maintains a side map of its
    uses and can RAUW itself.  Once the forward declarations are fully
    resolved RAUW support is dropped on the ground.  This means that
    uniquing collisions on changing operands cause nodes to become
    "distinct".  (This already happened fairly commonly, whenever an
    operand went to null.)

    If you're constructing complex (non self-reference) `MDNode` cycles,
    you need to call `MDNode::resolveCycles()` on each node (or on a
    top-level node that somehow references all of the nodes).  Also,
    don't do that.  Metadata cycles (and the RAUW machinery needed to
    construct them) are expensive.

  - An `MDNode` can only refer to a `Constant` through a bridge called
    `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`).

    As a side effect, accessing an operand of an `MDNode` that is known
    to be, e.g., `ConstantInt`, takes three steps: first, cast from
    `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`;
    third, cast down to `ConstantInt`.

    The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have
    metadata schema owners transition away from using `Constant`s when
    the type isn't important (and they don't care about referring to
    `GlobalValue`s).

    In the meantime, I've added transitional API to the `mdconst`
    namespace that matches semantics with the old code, in order to
    avoid adding the error-prone three-step equivalent to every call
    site.  If your old code was:

        MDNode *N = foo();
        bar(isa             <ConstantInt>(N->getOperand(0)));
        baz(cast            <ConstantInt>(N->getOperand(1)));
        bak(cast_or_null    <ConstantInt>(N->getOperand(2)));
        bat(dyn_cast        <ConstantInt>(N->getOperand(3)));
        bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4)));

    you can trivially match its semantics with:

        MDNode *N = foo();
        bar(mdconst::hasa               <ConstantInt>(N->getOperand(0)));
        baz(mdconst::extract            <ConstantInt>(N->getOperand(1)));
        bak(mdconst::extract_or_null    <ConstantInt>(N->getOperand(2)));
        bat(mdconst::dyn_extract        <ConstantInt>(N->getOperand(3)));
        bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4)));

    and when you transition your metadata schema to `MDInt`:

        MDNode *N = foo();
        bar(isa             <MDInt>(N->getOperand(0)));
        baz(cast            <MDInt>(N->getOperand(1)));
        bak(cast_or_null    <MDInt>(N->getOperand(2)));
        bat(dyn_cast        <MDInt>(N->getOperand(3)));
        bay(dyn_cast_or_null<MDInt>(N->getOperand(4)));

  - A `CallInst` -- specifically, intrinsic instructions -- can refer to
    metadata through a bridge called `MetadataAsValue`.  This is a
    subclass of `Value` where `getType()->isMetadataTy()`.

    `MetadataAsValue` is the *only* class that can legally refer to a
    `LocalAsMetadata`, which is a bridged form of non-`Constant` values
    like `Argument` and `Instruction`.  It can also refer to any other
    `Metadata` subclass.

(I'll break all your testcases in a follow-up commit, when I propagate
this change to assembly.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@223802 91177308-0d34-0410-b5e6-96231b3b80d8
2014-12-09 18:38:53 +00:00

221 lines
8.9 KiB
C++

//===- AddDiscriminators.cpp - Insert DWARF path discriminators -----------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file adds DWARF discriminators to the IR. Path discriminators are
// used to decide what CFG path was taken inside sub-graphs whose instructions
// share the same line and column number information.
//
// The main user of this is the sample profiler. Instruction samples are
// mapped to line number information. Since a single line may be spread
// out over several basic blocks, discriminators add more precise location
// for the samples.
//
// For example,
//
// 1 #define ASSERT(P)
// 2 if (!(P))
// 3 abort()
// ...
// 100 while (true) {
// 101 ASSERT (sum < 0);
// 102 ...
// 130 }
//
// when converted to IR, this snippet looks something like:
//
// while.body: ; preds = %entry, %if.end
// %0 = load i32* %sum, align 4, !dbg !15
// %cmp = icmp slt i32 %0, 0, !dbg !15
// br i1 %cmp, label %if.end, label %if.then, !dbg !15
//
// if.then: ; preds = %while.body
// call void @abort(), !dbg !15
// br label %if.end, !dbg !15
//
// Notice that all the instructions in blocks 'while.body' and 'if.then'
// have exactly the same debug information. When this program is sampled
// at runtime, the profiler will assume that all these instructions are
// equally frequent. This, in turn, will consider the edge while.body->if.then
// to be frequently taken (which is incorrect).
//
// By adding a discriminator value to the instructions in block 'if.then',
// we can distinguish instructions at line 101 with discriminator 0 from
// the instructions at line 101 with discriminator 1.
//
// For more details about DWARF discriminators, please visit
// http://wiki.dwarfstd.org/index.php?title=Path_Discriminators
//===----------------------------------------------------------------------===//
#include "llvm/Transforms/Scalar.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DIBuilder.h"
#include "llvm/IR/DebugInfo.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
#include "llvm/Pass.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
using namespace llvm;
#define DEBUG_TYPE "add-discriminators"
namespace {
struct AddDiscriminators : public FunctionPass {
static char ID; // Pass identification, replacement for typeid
AddDiscriminators() : FunctionPass(ID) {
initializeAddDiscriminatorsPass(*PassRegistry::getPassRegistry());
}
bool runOnFunction(Function &F) override;
};
}
char AddDiscriminators::ID = 0;
INITIALIZE_PASS_BEGIN(AddDiscriminators, "add-discriminators",
"Add DWARF path discriminators", false, false)
INITIALIZE_PASS_END(AddDiscriminators, "add-discriminators",
"Add DWARF path discriminators", false, false)
// Command line option to disable discriminator generation even in the
// presence of debug information. This is only needed when debugging
// debug info generation issues.
static cl::opt<bool>
NoDiscriminators("no-discriminators", cl::init(false),
cl::desc("Disable generation of discriminator information."));
FunctionPass *llvm::createAddDiscriminatorsPass() {
return new AddDiscriminators();
}
static bool hasDebugInfo(const Function &F) {
NamedMDNode *CUNodes = F.getParent()->getNamedMetadata("llvm.dbg.cu");
return CUNodes != nullptr;
}
/// \brief Assign DWARF discriminators.
///
/// To assign discriminators, we examine the boundaries of every
/// basic block and its successors. Suppose there is a basic block B1
/// with successor B2. The last instruction I1 in B1 and the first
/// instruction I2 in B2 are located at the same file and line number.
/// This situation is illustrated in the following code snippet:
///
/// if (i < 10) x = i;
///
/// entry:
/// br i1 %cmp, label %if.then, label %if.end, !dbg !10
/// if.then:
/// %1 = load i32* %i.addr, align 4, !dbg !10
/// store i32 %1, i32* %x, align 4, !dbg !10
/// br label %if.end, !dbg !10
/// if.end:
/// ret void, !dbg !12
///
/// Notice how the branch instruction in block 'entry' and all the
/// instructions in block 'if.then' have the exact same debug location
/// information (!dbg !10).
///
/// To distinguish instructions in block 'entry' from instructions in
/// block 'if.then', we generate a new lexical block for all the
/// instruction in block 'if.then' that share the same file and line
/// location with the last instruction of block 'entry'.
///
/// This new lexical block will have the same location information as
/// the previous one, but with a new DWARF discriminator value.
///
/// One of the main uses of this discriminator value is in runtime
/// sample profilers. It allows the profiler to distinguish instructions
/// at location !dbg !10 that execute on different basic blocks. This is
/// important because while the predicate 'if (x < 10)' may have been
/// executed millions of times, the assignment 'x = i' may have only
/// executed a handful of times (meaning that the entry->if.then edge is
/// seldom taken).
///
/// If we did not have discriminator information, the profiler would
/// assign the same weight to both blocks 'entry' and 'if.then', which
/// in turn will make it conclude that the entry->if.then edge is very
/// hot.
///
/// To decide where to create new discriminator values, this function
/// traverses the CFG and examines instruction at basic block boundaries.
/// If the last instruction I1 of a block B1 is at the same file and line
/// location as instruction I2 of successor B2, then it creates a new
/// lexical block for I2 and all the instruction in B2 that share the same
/// file and line location as I2. This new lexical block will have a
/// different discriminator number than I1.
bool AddDiscriminators::runOnFunction(Function &F) {
// If the function has debug information, but the user has disabled
// discriminators, do nothing.
// Simlarly, if the function has no debug info, do nothing.
// Finally, if this module is built with dwarf versions earlier than 4,
// do nothing (discriminator support is a DWARF 4 feature).
if (NoDiscriminators ||
!hasDebugInfo(F) ||
F.getParent()->getDwarfVersion() < 4)
return false;
bool Changed = false;
Module *M = F.getParent();
LLVMContext &Ctx = M->getContext();
DIBuilder Builder(*M, /*AllowUnresolved*/ false);
// Traverse all the blocks looking for instructions in different
// blocks that are at the same file:line location.
for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I) {
BasicBlock *B = I;
TerminatorInst *Last = B->getTerminator();
DebugLoc LastLoc = Last->getDebugLoc();
if (LastLoc.isUnknown()) continue;
DILocation LastDIL(LastLoc.getAsMDNode(Ctx));
for (unsigned I = 0; I < Last->getNumSuccessors(); ++I) {
BasicBlock *Succ = Last->getSuccessor(I);
Instruction *First = Succ->getFirstNonPHIOrDbgOrLifetime();
DebugLoc FirstLoc = First->getDebugLoc();
if (FirstLoc.isUnknown()) continue;
DILocation FirstDIL(FirstLoc.getAsMDNode(Ctx));
// If the first instruction (First) of Succ is at the same file
// location as B's last instruction (Last), add a new
// discriminator for First's location and all the instructions
// in Succ that share the same location with First.
if (FirstDIL.atSameLineAs(LastDIL)) {
// Create a new lexical scope and compute a new discriminator
// number for it.
StringRef Filename = FirstDIL.getFilename();
DIScope Scope = FirstDIL.getScope();
DIFile File = Builder.createFile(Filename, Scope.getDirectory());
unsigned Discriminator = FirstDIL.computeNewDiscriminator(Ctx);
DILexicalBlockFile NewScope =
Builder.createLexicalBlockFile(Scope, File, Discriminator);
DILocation NewDIL = FirstDIL.copyWithNewScope(Ctx, NewScope);
DebugLoc newDebugLoc = DebugLoc::getFromDILocation(NewDIL);
// Attach this new debug location to First and every
// instruction following First that shares the same location.
for (BasicBlock::iterator I1(*First), E1 = Succ->end(); I1 != E1;
++I1) {
if (I1->getDebugLoc() != FirstLoc) break;
I1->setDebugLoc(newDebugLoc);
DEBUG(dbgs() << NewDIL.getFilename() << ":" << NewDIL.getLineNumber()
<< ":" << NewDIL.getColumnNumber() << ":"
<< NewDIL.getDiscriminator() << *I1 << "\n");
}
DEBUG(dbgs() << "\n");
Changed = true;
}
}
}
return Changed;
}