llvm-project/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded...

1246 lines
50 KiB
C++
Raw Normal View History

//===- InstCombineSimplifyDemanded.cpp ------------------------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains logic for simplifying instructions based on information
// about how they are used.
//
//===----------------------------------------------------------------------===//
#include "InstCombineInternal.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/PatternMatch.h"
using namespace llvm;
using namespace llvm::PatternMatch;
#define DEBUG_TYPE "instcombine"
/// ShrinkDemandedConstant - Check to see if the specified operand of the
/// specified instruction is a constant integer. If so, check to see if there
/// are any bits set in the constant that are not demanded. If so, shrink the
/// constant and return true.
static bool ShrinkDemandedConstant(Instruction *I, unsigned OpNo,
APInt Demanded) {
assert(I && "No instruction?");
assert(OpNo < I->getNumOperands() && "Operand index too large");
// If the operand is not a constant integer, nothing to do.
ConstantInt *OpC = dyn_cast<ConstantInt>(I->getOperand(OpNo));
if (!OpC) return false;
// If there are no bits set that aren't demanded, nothing to do.
Demanded = Demanded.zextOrTrunc(OpC->getValue().getBitWidth());
if ((~Demanded & OpC->getValue()) == 0)
return false;
// This instruction is producing bits that are not demanded. Shrink the RHS.
Demanded &= OpC->getValue();
I->setOperand(OpNo, ConstantInt::get(OpC->getType(), Demanded));
return true;
}
/// SimplifyDemandedInstructionBits - Inst is an integer instruction that
/// SimplifyDemandedBits knows about. See if the instruction has any
/// properties that allow us to simplify its operands.
bool InstCombiner::SimplifyDemandedInstructionBits(Instruction &Inst) {
unsigned BitWidth = Inst.getType()->getScalarSizeInBits();
APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
APInt DemandedMask(APInt::getAllOnesValue(BitWidth));
Value *V = SimplifyDemandedUseBits(&Inst, DemandedMask, KnownZero, KnownOne,
0, &Inst);
if (!V) return false;
if (V == &Inst) return true;
ReplaceInstUsesWith(Inst, V);
return true;
}
/// SimplifyDemandedBits - This form of SimplifyDemandedBits simplifies the
/// specified instruction operand if possible, updating it in place. It returns
/// true if it made any change and false otherwise.
bool InstCombiner::SimplifyDemandedBits(Use &U, APInt DemandedMask,
APInt &KnownZero, APInt &KnownOne,
unsigned Depth) {
auto *UserI = dyn_cast<Instruction>(U.getUser());
Value *NewVal = SimplifyDemandedUseBits(U.get(), DemandedMask, KnownZero,
KnownOne, Depth, UserI);
if (!NewVal) return false;
U = NewVal;
return true;
}
/// SimplifyDemandedUseBits - This function attempts to replace V with a simpler
/// value based on the demanded bits. When this function is called, it is known
/// that only the bits set in DemandedMask of the result of V are ever used
/// downstream. Consequently, depending on the mask and V, it may be possible
/// to replace V with a constant or one of its operands. In such cases, this
/// function does the replacement and returns true. In all other cases, it
/// returns false after analyzing the expression and setting KnownOne and known
/// to be one in the expression. KnownZero contains all the bits that are known
/// to be zero in the expression. These are provided to potentially allow the
/// caller (which might recursively be SimplifyDemandedBits itself) to simplify
/// the expression. KnownOne and KnownZero always follow the invariant that
/// KnownOne & KnownZero == 0. That is, a bit can't be both 1 and 0. Note that
/// the bits in KnownOne and KnownZero may only be accurate for those bits set
/// in DemandedMask. Note also that the bitwidth of V, DemandedMask, KnownZero
/// and KnownOne must all be the same.
///
/// This returns null if it did not change anything and it permits no
/// simplification. This returns V itself if it did some simplification of V's
/// operands based on the information about what bits are demanded. This returns
/// some other non-null value if it found out that V is equal to another value
/// in the context where the specified bits are demanded, but not for all users.
Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
APInt &KnownZero, APInt &KnownOne,
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
unsigned Depth,
Instruction *CxtI) {
2014-04-28 12:05:08 +08:00
assert(V != nullptr && "Null pointer of Value???");
assert(Depth <= 6 && "Limit Search Depth");
uint32_t BitWidth = DemandedMask.getBitWidth();
Type *VTy = V->getType();
assert(
(!VTy->isIntOrIntVectorTy() || VTy->getScalarSizeInBits() == BitWidth) &&
KnownZero.getBitWidth() == BitWidth &&
KnownOne.getBitWidth() == BitWidth &&
"Value *V, DemandedMask, KnownZero and KnownOne "
"must have same BitWidth");
if (ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
// We know all of the bits for a constant!
KnownOne = CI->getValue() & DemandedMask;
KnownZero = ~KnownOne & DemandedMask;
return nullptr;
}
if (isa<ConstantPointerNull>(V)) {
// We know all of the bits for a constant!
KnownOne.clearAllBits();
KnownZero = DemandedMask;
return nullptr;
}
KnownZero.clearAllBits();
KnownOne.clearAllBits();
if (DemandedMask == 0) { // Not demanding any bits from V.
if (isa<UndefValue>(V))
return nullptr;
return UndefValue::get(VTy);
}
if (Depth == 6) // Limit search depth.
return nullptr;
APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
APInt RHSKnownZero(BitWidth, 0), RHSKnownOne(BitWidth, 0);
Instruction *I = dyn_cast<Instruction>(V);
if (!I) {
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
computeKnownBits(V, KnownZero, KnownOne, Depth, CxtI);
return nullptr; // Only analyze instructions.
}
// If there are multiple uses of this value and we aren't at the root, then
// we can't do any simplifications of the operands, because DemandedMask
// only reflects the bits demanded by *one* of the users.
if (Depth != 0 && !I->hasOneUse()) {
// Despite the fact that we can't simplify this instruction in all User's
// context, we can at least compute the knownzero/knownone bits, and we can
// do simplifications that apply to *just* the one user if we know that
// this instruction has a simpler value in that context.
if (I->getOpcode() == Instruction::And) {
// If either the LHS or the RHS are Zero, the result is zero.
computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth + 1,
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
CxtI);
computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth + 1,
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
CxtI);
// If all of the demanded bits are known 1 on one side, return the other.
// These bits cannot contribute to the result of the 'and' in this
// context.
if ((DemandedMask & ~LHSKnownZero & RHSKnownOne) ==
(DemandedMask & ~LHSKnownZero))
return I->getOperand(0);
if ((DemandedMask & ~RHSKnownZero & LHSKnownOne) ==
(DemandedMask & ~RHSKnownZero))
return I->getOperand(1);
// If all of the demanded bits in the inputs are known zeros, return zero.
if ((DemandedMask & (RHSKnownZero|LHSKnownZero)) == DemandedMask)
return Constant::getNullValue(VTy);
} else if (I->getOpcode() == Instruction::Or) {
// We can simplify (X|Y) -> X or Y in the user's context if we know that
// only bits from X or Y are demanded.
// If either the LHS or the RHS are One, the result is One.
computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth + 1,
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
CxtI);
computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth + 1,
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
CxtI);
// If all of the demanded bits are known zero on one side, return the
// other. These bits cannot contribute to the result of the 'or' in this
// context.
if ((DemandedMask & ~LHSKnownOne & RHSKnownZero) ==
(DemandedMask & ~LHSKnownOne))
return I->getOperand(0);
if ((DemandedMask & ~RHSKnownOne & LHSKnownZero) ==
(DemandedMask & ~RHSKnownOne))
return I->getOperand(1);
// If all of the potentially set bits on one side are known to be set on
// the other side, just use the 'other' side.
if ((DemandedMask & (~RHSKnownZero) & LHSKnownOne) ==
(DemandedMask & (~RHSKnownZero)))
return I->getOperand(0);
if ((DemandedMask & (~LHSKnownZero) & RHSKnownOne) ==
(DemandedMask & (~LHSKnownZero)))
return I->getOperand(1);
} else if (I->getOpcode() == Instruction::Xor) {
// We can simplify (X^Y) -> X or Y in the user's context if we know that
// only bits from X or Y are demanded.
computeKnownBits(I->getOperand(1), RHSKnownZero, RHSKnownOne, Depth + 1,
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
CxtI);
computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth + 1,
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
CxtI);
// If all of the demanded bits are known zero on one side, return the
// other.
if ((DemandedMask & RHSKnownZero) == DemandedMask)
return I->getOperand(0);
if ((DemandedMask & LHSKnownZero) == DemandedMask)
return I->getOperand(1);
}
// Compute the KnownZero/KnownOne bits to simplify things downstream.
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
computeKnownBits(I, KnownZero, KnownOne, Depth, CxtI);
return nullptr;
}
// If this is the root being simplified, allow it to have multiple uses,
// just set the DemandedMask to all bits so that we can try to simplify the
// operands. This allows visitTruncInst (for example) to simplify the
// operand of a trunc without duplicating all the logic below.
if (Depth == 0 && !V->hasOneUse())
DemandedMask = APInt::getAllOnesValue(BitWidth);
switch (I->getOpcode()) {
default:
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
computeKnownBits(I, KnownZero, KnownOne, Depth, CxtI);
break;
case Instruction::And:
// If either the LHS or the RHS are Zero, the result is zero.
if (SimplifyDemandedBits(I->getOperandUse(1), DemandedMask, RHSKnownZero,
RHSKnownOne, Depth + 1) ||
SimplifyDemandedBits(I->getOperandUse(0), DemandedMask & ~RHSKnownZero,
LHSKnownZero, LHSKnownOne, Depth + 1))
return I;
assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");
Add additional patterns for @llvm.assume in ValueTracking This builds on r217342, which added the infrastructure to compute known bits using assumptions (@llvm.assume calls). That original commit added only a few patterns (to catch common cases related to determining pointer alignment); this change adds several other patterns for simple cases. r217342 contained that, for assume(v & b = a), bits in the mask that are known to be one, we can propagate known bits from the a to v. It also had a known-bits transfer for assume(a = b). This patch adds: assume(~(v & b) = a) : For those bits in the mask that are known to be one, we can propagate inverted known bits from the a to v. assume(v | b = a) : For those bits in b that are known to be zero, we can propagate known bits from the a to v. assume(~(v | b) = a): For those bits in b that are known to be zero, we can propagate inverted known bits from the a to v. assume(v ^ b = a) : For those bits in b that are known to be zero, we can propagate known bits from the a to v. For those bits in b that are known to be one, we can propagate inverted known bits from the a to v. assume(~(v ^ b) = a) : For those bits in b that are known to be zero, we can propagate inverted known bits from the a to v. For those bits in b that are known to be one, we can propagate known bits from the a to v. assume(v << c = a) : For those bits in a that are known, we can propagate them to known bits in v shifted to the right by c. assume(~(v << c) = a) : For those bits in a that are known, we can propagate them inverted to known bits in v shifted to the right by c. assume(v >> c = a) : For those bits in a that are known, we can propagate them to known bits in v shifted to the right by c. assume(~(v >> c) = a) : For those bits in a that are known, we can propagate them inverted to known bits in v shifted to the right by c. assume(v >=_s c) where c is non-negative: The sign bit of v is zero assume(v >_s c) where c is at least -1: The sign bit of v is zero assume(v <=_s c) where c is negative: The sign bit of v is one assume(v <_s c) where c is non-positive: The sign bit of v is one assume(v <=_u c): Transfer the known high zero bits assume(v <_u c): Transfer the known high zero bits (if c is know to be a power of 2, transfer one more) A small addition to InstCombine was necessary for some of the test cases. The problem is that when InstCombine was simplifying and, or, etc. it would fail to check the 'do I know all of the bits' condition before checking less specific conditions and would not fully constant-fold the result. I'm not sure how to trigger this aside from using assumptions, so I've just included the change here. llvm-svn: 217343
2014-09-08 03:21:07 +08:00
// If the client is only demanding bits that we know, return the known
// constant.
if ((DemandedMask & ((RHSKnownZero | LHSKnownZero)|
(RHSKnownOne & LHSKnownOne))) == DemandedMask)
return Constant::getIntegerValue(VTy, RHSKnownOne & LHSKnownOne);
// If all of the demanded bits are known 1 on one side, return the other.
// These bits cannot contribute to the result of the 'and'.
if ((DemandedMask & ~LHSKnownZero & RHSKnownOne) ==
(DemandedMask & ~LHSKnownZero))
return I->getOperand(0);
if ((DemandedMask & ~RHSKnownZero & LHSKnownOne) ==
(DemandedMask & ~RHSKnownZero))
return I->getOperand(1);
// If all of the demanded bits in the inputs are known zeros, return zero.
if ((DemandedMask & (RHSKnownZero|LHSKnownZero)) == DemandedMask)
return Constant::getNullValue(VTy);
// If the RHS is a constant, see if we can simplify it.
if (ShrinkDemandedConstant(I, 1, DemandedMask & ~LHSKnownZero))
return I;
// Output known-1 bits are only known if set in both the LHS & RHS.
KnownOne = RHSKnownOne & LHSKnownOne;
// Output known-0 are known to be clear if zero in either the LHS | RHS.
KnownZero = RHSKnownZero | LHSKnownZero;
break;
case Instruction::Or:
// If either the LHS or the RHS are One, the result is One.
if (SimplifyDemandedBits(I->getOperandUse(1), DemandedMask, RHSKnownZero,
RHSKnownOne, Depth + 1) ||
SimplifyDemandedBits(I->getOperandUse(0), DemandedMask & ~RHSKnownOne,
LHSKnownZero, LHSKnownOne, Depth + 1))
return I;
assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");
Add additional patterns for @llvm.assume in ValueTracking This builds on r217342, which added the infrastructure to compute known bits using assumptions (@llvm.assume calls). That original commit added only a few patterns (to catch common cases related to determining pointer alignment); this change adds several other patterns for simple cases. r217342 contained that, for assume(v & b = a), bits in the mask that are known to be one, we can propagate known bits from the a to v. It also had a known-bits transfer for assume(a = b). This patch adds: assume(~(v & b) = a) : For those bits in the mask that are known to be one, we can propagate inverted known bits from the a to v. assume(v | b = a) : For those bits in b that are known to be zero, we can propagate known bits from the a to v. assume(~(v | b) = a): For those bits in b that are known to be zero, we can propagate inverted known bits from the a to v. assume(v ^ b = a) : For those bits in b that are known to be zero, we can propagate known bits from the a to v. For those bits in b that are known to be one, we can propagate inverted known bits from the a to v. assume(~(v ^ b) = a) : For those bits in b that are known to be zero, we can propagate inverted known bits from the a to v. For those bits in b that are known to be one, we can propagate known bits from the a to v. assume(v << c = a) : For those bits in a that are known, we can propagate them to known bits in v shifted to the right by c. assume(~(v << c) = a) : For those bits in a that are known, we can propagate them inverted to known bits in v shifted to the right by c. assume(v >> c = a) : For those bits in a that are known, we can propagate them to known bits in v shifted to the right by c. assume(~(v >> c) = a) : For those bits in a that are known, we can propagate them inverted to known bits in v shifted to the right by c. assume(v >=_s c) where c is non-negative: The sign bit of v is zero assume(v >_s c) where c is at least -1: The sign bit of v is zero assume(v <=_s c) where c is negative: The sign bit of v is one assume(v <_s c) where c is non-positive: The sign bit of v is one assume(v <=_u c): Transfer the known high zero bits assume(v <_u c): Transfer the known high zero bits (if c is know to be a power of 2, transfer one more) A small addition to InstCombine was necessary for some of the test cases. The problem is that when InstCombine was simplifying and, or, etc. it would fail to check the 'do I know all of the bits' condition before checking less specific conditions and would not fully constant-fold the result. I'm not sure how to trigger this aside from using assumptions, so I've just included the change here. llvm-svn: 217343
2014-09-08 03:21:07 +08:00
// If the client is only demanding bits that we know, return the known
// constant.
if ((DemandedMask & ((RHSKnownZero & LHSKnownZero)|
(RHSKnownOne | LHSKnownOne))) == DemandedMask)
return Constant::getIntegerValue(VTy, RHSKnownOne | LHSKnownOne);
// If all of the demanded bits are known zero on one side, return the other.
// These bits cannot contribute to the result of the 'or'.
if ((DemandedMask & ~LHSKnownOne & RHSKnownZero) ==
(DemandedMask & ~LHSKnownOne))
return I->getOperand(0);
if ((DemandedMask & ~RHSKnownOne & LHSKnownZero) ==
(DemandedMask & ~RHSKnownOne))
return I->getOperand(1);
// If all of the potentially set bits on one side are known to be set on
// the other side, just use the 'other' side.
if ((DemandedMask & (~RHSKnownZero) & LHSKnownOne) ==
(DemandedMask & (~RHSKnownZero)))
return I->getOperand(0);
if ((DemandedMask & (~LHSKnownZero) & RHSKnownOne) ==
(DemandedMask & (~LHSKnownZero)))
return I->getOperand(1);
// If the RHS is a constant, see if we can simplify it.
if (ShrinkDemandedConstant(I, 1, DemandedMask))
return I;
// Output known-0 bits are only known if clear in both the LHS & RHS.
KnownZero = RHSKnownZero & LHSKnownZero;
// Output known-1 are known to be set if set in either the LHS | RHS.
KnownOne = RHSKnownOne | LHSKnownOne;
break;
case Instruction::Xor: {
if (SimplifyDemandedBits(I->getOperandUse(1), DemandedMask, RHSKnownZero,
RHSKnownOne, Depth + 1) ||
SimplifyDemandedBits(I->getOperandUse(0), DemandedMask, LHSKnownZero,
LHSKnownOne, Depth + 1))
return I;
assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");
Add additional patterns for @llvm.assume in ValueTracking This builds on r217342, which added the infrastructure to compute known bits using assumptions (@llvm.assume calls). That original commit added only a few patterns (to catch common cases related to determining pointer alignment); this change adds several other patterns for simple cases. r217342 contained that, for assume(v & b = a), bits in the mask that are known to be one, we can propagate known bits from the a to v. It also had a known-bits transfer for assume(a = b). This patch adds: assume(~(v & b) = a) : For those bits in the mask that are known to be one, we can propagate inverted known bits from the a to v. assume(v | b = a) : For those bits in b that are known to be zero, we can propagate known bits from the a to v. assume(~(v | b) = a): For those bits in b that are known to be zero, we can propagate inverted known bits from the a to v. assume(v ^ b = a) : For those bits in b that are known to be zero, we can propagate known bits from the a to v. For those bits in b that are known to be one, we can propagate inverted known bits from the a to v. assume(~(v ^ b) = a) : For those bits in b that are known to be zero, we can propagate inverted known bits from the a to v. For those bits in b that are known to be one, we can propagate known bits from the a to v. assume(v << c = a) : For those bits in a that are known, we can propagate them to known bits in v shifted to the right by c. assume(~(v << c) = a) : For those bits in a that are known, we can propagate them inverted to known bits in v shifted to the right by c. assume(v >> c = a) : For those bits in a that are known, we can propagate them to known bits in v shifted to the right by c. assume(~(v >> c) = a) : For those bits in a that are known, we can propagate them inverted to known bits in v shifted to the right by c. assume(v >=_s c) where c is non-negative: The sign bit of v is zero assume(v >_s c) where c is at least -1: The sign bit of v is zero assume(v <=_s c) where c is negative: The sign bit of v is one assume(v <_s c) where c is non-positive: The sign bit of v is one assume(v <=_u c): Transfer the known high zero bits assume(v <_u c): Transfer the known high zero bits (if c is know to be a power of 2, transfer one more) A small addition to InstCombine was necessary for some of the test cases. The problem is that when InstCombine was simplifying and, or, etc. it would fail to check the 'do I know all of the bits' condition before checking less specific conditions and would not fully constant-fold the result. I'm not sure how to trigger this aside from using assumptions, so I've just included the change here. llvm-svn: 217343
2014-09-08 03:21:07 +08:00
// Output known-0 bits are known if clear or set in both the LHS & RHS.
APInt IKnownZero = (RHSKnownZero & LHSKnownZero) |
(RHSKnownOne & LHSKnownOne);
// Output known-1 are known to be set if set in only one of the LHS, RHS.
APInt IKnownOne = (RHSKnownZero & LHSKnownOne) |
(RHSKnownOne & LHSKnownZero);
// If the client is only demanding bits that we know, return the known
// constant.
if ((DemandedMask & (IKnownZero|IKnownOne)) == DemandedMask)
return Constant::getIntegerValue(VTy, IKnownOne);
// If all of the demanded bits are known zero on one side, return the other.
// These bits cannot contribute to the result of the 'xor'.
if ((DemandedMask & RHSKnownZero) == DemandedMask)
return I->getOperand(0);
if ((DemandedMask & LHSKnownZero) == DemandedMask)
return I->getOperand(1);
// If all of the demanded bits are known to be zero on one side or the
// other, turn this into an *inclusive* or.
// e.g. (A & C1)^(B & C2) -> (A & C1)|(B & C2) iff C1&C2 == 0
if ((DemandedMask & ~RHSKnownZero & ~LHSKnownZero) == 0) {
Instruction *Or =
BinaryOperator::CreateOr(I->getOperand(0), I->getOperand(1),
I->getName());
return InsertNewInstWith(Or, *I);
}
// If all of the demanded bits on one side are known, and all of the set
// bits on that side are also known to be set on the other side, turn this
// into an AND, as we know the bits will be cleared.
// e.g. (X | C1) ^ C2 --> (X | C1) & ~C2 iff (C1&C2) == C2
if ((DemandedMask & (RHSKnownZero|RHSKnownOne)) == DemandedMask) {
// all known
if ((RHSKnownOne & LHSKnownOne) == RHSKnownOne) {
Constant *AndC = Constant::getIntegerValue(VTy,
~RHSKnownOne & DemandedMask);
Instruction *And = BinaryOperator::CreateAnd(I->getOperand(0), AndC);
return InsertNewInstWith(And, *I);
}
}
// If the RHS is a constant, see if we can simplify it.
// FIXME: for XOR, we prefer to force bits to 1 if they will make a -1.
if (ShrinkDemandedConstant(I, 1, DemandedMask))
return I;
// If our LHS is an 'and' and if it has one use, and if any of the bits we
// are flipping are known to be set, then the xor is just resetting those
// bits to zero. We can just knock out bits from the 'and' and the 'xor',
// simplifying both of them.
if (Instruction *LHSInst = dyn_cast<Instruction>(I->getOperand(0)))
if (LHSInst->getOpcode() == Instruction::And && LHSInst->hasOneUse() &&
isa<ConstantInt>(I->getOperand(1)) &&
isa<ConstantInt>(LHSInst->getOperand(1)) &&
(LHSKnownOne & RHSKnownOne & DemandedMask) != 0) {
ConstantInt *AndRHS = cast<ConstantInt>(LHSInst->getOperand(1));
ConstantInt *XorRHS = cast<ConstantInt>(I->getOperand(1));
APInt NewMask = ~(LHSKnownOne & RHSKnownOne & DemandedMask);
Constant *AndC =
ConstantInt::get(I->getType(), NewMask & AndRHS->getValue());
Instruction *NewAnd = BinaryOperator::CreateAnd(I->getOperand(0), AndC);
InsertNewInstWith(NewAnd, *I);
Constant *XorC =
ConstantInt::get(I->getType(), NewMask & XorRHS->getValue());
Instruction *NewXor = BinaryOperator::CreateXor(NewAnd, XorC);
return InsertNewInstWith(NewXor, *I);
}
// Output known-0 bits are known if clear or set in both the LHS & RHS.
KnownZero= (RHSKnownZero & LHSKnownZero) | (RHSKnownOne & LHSKnownOne);
// Output known-1 are known to be set if set in only one of the LHS, RHS.
KnownOne = (RHSKnownZero & LHSKnownOne) | (RHSKnownOne & LHSKnownZero);
break;
}
case Instruction::Select:
// If this is a select as part of a min/max pattern, don't simplify any
// further in case we break the structure.
Value *LHS, *RHS;
if (matchSelectPattern(I, LHS, RHS).Flavor != SPF_UNKNOWN)
return nullptr;
if (SimplifyDemandedBits(I->getOperandUse(2), DemandedMask, RHSKnownZero,
RHSKnownOne, Depth + 1) ||
SimplifyDemandedBits(I->getOperandUse(1), DemandedMask, LHSKnownZero,
LHSKnownOne, Depth + 1))
return I;
assert(!(RHSKnownZero & RHSKnownOne) && "Bits known to be one AND zero?");
assert(!(LHSKnownZero & LHSKnownOne) && "Bits known to be one AND zero?");
// If the operands are constants, see if we can simplify them.
if (ShrinkDemandedConstant(I, 1, DemandedMask) ||
ShrinkDemandedConstant(I, 2, DemandedMask))
return I;
// Only known if known in both the LHS and RHS.
KnownOne = RHSKnownOne & LHSKnownOne;
KnownZero = RHSKnownZero & LHSKnownZero;
break;
case Instruction::Trunc: {
unsigned truncBf = I->getOperand(0)->getType()->getScalarSizeInBits();
DemandedMask = DemandedMask.zext(truncBf);
KnownZero = KnownZero.zext(truncBf);
KnownOne = KnownOne.zext(truncBf);
if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMask, KnownZero,
KnownOne, Depth + 1))
return I;
DemandedMask = DemandedMask.trunc(BitWidth);
KnownZero = KnownZero.trunc(BitWidth);
KnownOne = KnownOne.trunc(BitWidth);
assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
break;
}
case Instruction::BitCast:
if (!I->getOperand(0)->getType()->isIntOrIntVectorTy())
return nullptr; // vector->int or fp->int?
if (VectorType *DstVTy = dyn_cast<VectorType>(I->getType())) {
if (VectorType *SrcVTy =
dyn_cast<VectorType>(I->getOperand(0)->getType())) {
if (DstVTy->getNumElements() != SrcVTy->getNumElements())
// Don't touch a bitcast between vectors of different element counts.
return nullptr;
} else
// Don't touch a scalar-to-vector bitcast.
return nullptr;
} else if (I->getOperand(0)->getType()->isVectorTy())
// Don't touch a vector-to-scalar bitcast.
return nullptr;
if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMask, KnownZero,
KnownOne, Depth + 1))
return I;
assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
break;
case Instruction::ZExt: {
// Compute the bits in the result that are not present in the input.
unsigned SrcBitWidth =I->getOperand(0)->getType()->getScalarSizeInBits();
DemandedMask = DemandedMask.trunc(SrcBitWidth);
KnownZero = KnownZero.trunc(SrcBitWidth);
KnownOne = KnownOne.trunc(SrcBitWidth);
if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMask, KnownZero,
KnownOne, Depth + 1))
return I;
DemandedMask = DemandedMask.zext(BitWidth);
KnownZero = KnownZero.zext(BitWidth);
KnownOne = KnownOne.zext(BitWidth);
assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
// The top bits are known to be zero.
KnownZero |= APInt::getHighBitsSet(BitWidth, BitWidth - SrcBitWidth);
break;
}
case Instruction::SExt: {
// Compute the bits in the result that are not present in the input.
unsigned SrcBitWidth =I->getOperand(0)->getType()->getScalarSizeInBits();
APInt InputDemandedBits = DemandedMask &
APInt::getLowBitsSet(BitWidth, SrcBitWidth);
APInt NewBits(APInt::getHighBitsSet(BitWidth, BitWidth - SrcBitWidth));
// If any of the sign extended bits are demanded, we know that the sign
// bit is demanded.
if ((NewBits & DemandedMask) != 0)
InputDemandedBits.setBit(SrcBitWidth-1);
InputDemandedBits = InputDemandedBits.trunc(SrcBitWidth);
KnownZero = KnownZero.trunc(SrcBitWidth);
KnownOne = KnownOne.trunc(SrcBitWidth);
if (SimplifyDemandedBits(I->getOperandUse(0), InputDemandedBits, KnownZero,
KnownOne, Depth + 1))
return I;
InputDemandedBits = InputDemandedBits.zext(BitWidth);
KnownZero = KnownZero.zext(BitWidth);
KnownOne = KnownOne.zext(BitWidth);
assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
// If the sign bit of the input is known set or clear, then we know the
// top bits of the result.
// If the input sign bit is known zero, or if the NewBits are not demanded
// convert this into a zero extension.
if (KnownZero[SrcBitWidth-1] || (NewBits & ~DemandedMask) == NewBits) {
// Convert to ZExt cast
CastInst *NewCast = new ZExtInst(I->getOperand(0), VTy, I->getName());
return InsertNewInstWith(NewCast, *I);
} else if (KnownOne[SrcBitWidth-1]) { // Input sign bit known set
KnownOne |= NewBits;
}
break;
}
case Instruction::Add:
case Instruction::Sub: {
/// If the high-bits of an ADD/SUB are not demanded, then we do not care
/// about the high bits of the operands.
unsigned NLZ = DemandedMask.countLeadingZeros();
if (NLZ > 0) {
// Right fill the mask of bits for this ADD/SUB to demand the most
// significant bit and all those below it.
APInt DemandedFromOps(APInt::getLowBitsSet(BitWidth, BitWidth-NLZ));
if (SimplifyDemandedBits(I->getOperandUse(0), DemandedFromOps,
LHSKnownZero, LHSKnownOne, Depth + 1) ||
ShrinkDemandedConstant(I, 1, DemandedFromOps) ||
SimplifyDemandedBits(I->getOperandUse(1), DemandedFromOps,
LHSKnownZero, LHSKnownOne, Depth + 1)) {
// Disable the nsw and nuw flags here: We can no longer guarantee that
// we won't wrap after simplification. Removing the nsw/nuw flags is
// legal here because the top bit is not demanded.
BinaryOperator &BinOP = *cast<BinaryOperator>(I);
BinOP.setHasNoSignedWrap(false);
BinOP.setHasNoUnsignedWrap(false);
return I;
}
}
// Otherwise just hand the add/sub off to computeKnownBits to fill in
// the known zeros and ones.
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
computeKnownBits(V, KnownZero, KnownOne, Depth, CxtI);
break;
}
case Instruction::Shl:
if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
{
Value *VarX; ConstantInt *C1;
if (match(I->getOperand(0), m_Shr(m_Value(VarX), m_ConstantInt(C1)))) {
Instruction *Shr = cast<Instruction>(I->getOperand(0));
Value *R = SimplifyShrShlDemandedBits(Shr, I, DemandedMask,
KnownZero, KnownOne);
if (R)
return R;
}
}
uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
APInt DemandedMaskIn(DemandedMask.lshr(ShiftAmt));
// If the shift is NUW/NSW, then it does demand the high bits.
ShlOperator *IOp = cast<ShlOperator>(I);
if (IOp->hasNoSignedWrap())
DemandedMaskIn |= APInt::getHighBitsSet(BitWidth, ShiftAmt+1);
else if (IOp->hasNoUnsignedWrap())
DemandedMaskIn |= APInt::getHighBitsSet(BitWidth, ShiftAmt);
if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMaskIn, KnownZero,
KnownOne, Depth + 1))
return I;
assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
KnownZero <<= ShiftAmt;
KnownOne <<= ShiftAmt;
// low bits known zero.
if (ShiftAmt)
KnownZero |= APInt::getLowBitsSet(BitWidth, ShiftAmt);
}
break;
case Instruction::LShr:
// For a logical shift right
if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
// Unsigned shift right.
APInt DemandedMaskIn(DemandedMask.shl(ShiftAmt));
// If the shift is exact, then it does demand the low bits (and knows that
// they are zero).
if (cast<LShrOperator>(I)->isExact())
DemandedMaskIn |= APInt::getLowBitsSet(BitWidth, ShiftAmt);
if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMaskIn, KnownZero,
KnownOne, Depth + 1))
return I;
assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
KnownZero = APIntOps::lshr(KnownZero, ShiftAmt);
KnownOne = APIntOps::lshr(KnownOne, ShiftAmt);
if (ShiftAmt) {
// Compute the new bits that are at the top now.
APInt HighBits(APInt::getHighBitsSet(BitWidth, ShiftAmt));
KnownZero |= HighBits; // high bits known zero.
}
}
break;
case Instruction::AShr:
// If this is an arithmetic shift right and only the low-bit is set, we can
// always convert this into a logical shr, even if the shift amount is
// variable. The low bit of the shift cannot be an input sign bit unless
// the shift amount is >= the size of the datatype, which is undefined.
if (DemandedMask == 1) {
// Perform the logical shift right.
Instruction *NewVal = BinaryOperator::CreateLShr(
I->getOperand(0), I->getOperand(1), I->getName());
return InsertNewInstWith(NewVal, *I);
}
// If the sign bit is the only bit demanded by this ashr, then there is no
// need to do it, the shift doesn't change the high bit.
if (DemandedMask.isSignBit())
return I->getOperand(0);
if (ConstantInt *SA = dyn_cast<ConstantInt>(I->getOperand(1))) {
uint32_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
// Signed shift right.
APInt DemandedMaskIn(DemandedMask.shl(ShiftAmt));
// If any of the "high bits" are demanded, we should set the sign bit as
// demanded.
if (DemandedMask.countLeadingZeros() <= ShiftAmt)
DemandedMaskIn.setBit(BitWidth-1);
// If the shift is exact, then it does demand the low bits (and knows that
// they are zero).
if (cast<AShrOperator>(I)->isExact())
DemandedMaskIn |= APInt::getLowBitsSet(BitWidth, ShiftAmt);
if (SimplifyDemandedBits(I->getOperandUse(0), DemandedMaskIn, KnownZero,
KnownOne, Depth + 1))
return I;
assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
// Compute the new bits that are at the top now.
APInt HighBits(APInt::getHighBitsSet(BitWidth, ShiftAmt));
KnownZero = APIntOps::lshr(KnownZero, ShiftAmt);
KnownOne = APIntOps::lshr(KnownOne, ShiftAmt);
// Handle the sign bits.
APInt SignBit(APInt::getSignBit(BitWidth));
// Adjust to where it is now in the mask.
SignBit = APIntOps::lshr(SignBit, ShiftAmt);
// If the input sign bit is known to be zero, or if none of the top bits
// are demanded, turn this into an unsigned shift right.
if (BitWidth <= ShiftAmt || KnownZero[BitWidth-ShiftAmt-1] ||
(HighBits & ~DemandedMask) == HighBits) {
// Perform the logical shift right.
BinaryOperator *NewVal = BinaryOperator::CreateLShr(I->getOperand(0),
SA, I->getName());
NewVal->setIsExact(cast<BinaryOperator>(I)->isExact());
return InsertNewInstWith(NewVal, *I);
} else if ((KnownOne & SignBit) != 0) { // New bits are known one.
KnownOne |= HighBits;
}
}
break;
case Instruction::SRem:
if (ConstantInt *Rem = dyn_cast<ConstantInt>(I->getOperand(1))) {
// X % -1 demands all the bits because we don't want to introduce
// INT_MIN % -1 (== undef) by accident.
if (Rem->isAllOnesValue())
break;
APInt RA = Rem->getValue().abs();
if (RA.isPowerOf2()) {
if (DemandedMask.ult(RA)) // srem won't affect demanded bits
return I->getOperand(0);
APInt LowBits = RA - 1;
APInt Mask2 = LowBits | APInt::getSignBit(BitWidth);
if (SimplifyDemandedBits(I->getOperandUse(0), Mask2, LHSKnownZero,
LHSKnownOne, Depth + 1))
return I;
// The low bits of LHS are unchanged by the srem.
KnownZero = LHSKnownZero & LowBits;
KnownOne = LHSKnownOne & LowBits;
// If LHS is non-negative or has all low bits zero, then the upper bits
// are all zero.
if (LHSKnownZero[BitWidth-1] || ((LHSKnownZero & LowBits) == LowBits))
KnownZero |= ~LowBits;
// If LHS is negative and not all low bits are zero, then the upper bits
// are all one.
if (LHSKnownOne[BitWidth-1] && ((LHSKnownOne & LowBits) != 0))
KnownOne |= ~LowBits;
assert(!(KnownZero & KnownOne) && "Bits known to be one AND zero?");
}
}
// The sign bit is the LHS's sign bit, except when the result of the
// remainder is zero.
if (DemandedMask.isNegative() && KnownZero.isNonNegative()) {
APInt LHSKnownZero(BitWidth, 0), LHSKnownOne(BitWidth, 0);
computeKnownBits(I->getOperand(0), LHSKnownZero, LHSKnownOne, Depth + 1,
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
CxtI);
// If it's known zero, our sign bit is also zero.
if (LHSKnownZero.isNegative())
KnownZero.setBit(KnownZero.getBitWidth() - 1);
}
break;
case Instruction::URem: {
APInt KnownZero2(BitWidth, 0), KnownOne2(BitWidth, 0);
APInt AllOnes = APInt::getAllOnesValue(BitWidth);
if (SimplifyDemandedBits(I->getOperandUse(0), AllOnes, KnownZero2,
KnownOne2, Depth + 1) ||
SimplifyDemandedBits(I->getOperandUse(1), AllOnes, KnownZero2,
KnownOne2, Depth + 1))
return I;
unsigned Leaders = KnownZero2.countLeadingOnes();
Leaders = std::max(Leaders,
KnownZero2.countLeadingOnes());
KnownZero = APInt::getHighBitsSet(BitWidth, Leaders) & DemandedMask;
break;
}
case Instruction::Call:
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
switch (II->getIntrinsicID()) {
default: break;
case Intrinsic::bswap: {
// If the only bits demanded come from one byte of the bswap result,
// just shift the input byte into position to eliminate the bswap.
unsigned NLZ = DemandedMask.countLeadingZeros();
unsigned NTZ = DemandedMask.countTrailingZeros();
// Round NTZ down to the next byte. If we have 11 trailing zeros, then
// we need all the bits down to bit 8. Likewise, round NLZ. If we
// have 14 leading zeros, round to 8.
NLZ &= ~7;
NTZ &= ~7;
// If we need exactly one byte, we can do this transformation.
if (BitWidth-NLZ-NTZ == 8) {
unsigned ResultBit = NTZ;
unsigned InputBit = BitWidth-NTZ-8;
// Replace this with either a left or right shift to get the byte into
// the right place.
Instruction *NewVal;
if (InputBit > ResultBit)
2010-06-24 20:35:13 +08:00
NewVal = BinaryOperator::CreateLShr(II->getArgOperand(0),
ConstantInt::get(I->getType(), InputBit-ResultBit));
else
2010-06-24 20:35:13 +08:00
NewVal = BinaryOperator::CreateShl(II->getArgOperand(0),
ConstantInt::get(I->getType(), ResultBit-InputBit));
NewVal->takeName(I);
return InsertNewInstWith(NewVal, *I);
}
// TODO: Could compute known zero/one bits based on the input.
break;
}
case Intrinsic::x86_sse42_crc32_64_64:
KnownZero = APInt::getHighBitsSet(64, 32);
return nullptr;
}
}
Make use of @llvm.assume in ValueTracking (computeKnownBits, etc.) This change, which allows @llvm.assume to be used from within computeKnownBits (and other associated functions in ValueTracking), adds some (optional) parameters to computeKnownBits and friends. These functions now (optionally) take a "context" instruction pointer, an AssumptionTracker pointer, and also a DomTree pointer, and most of the changes are just to pass this new information when it is easily available from InstSimplify, InstCombine, etc. As explained below, the significant conceptual change is that known properties of a value might depend on the control-flow location of the use (because we care that the @llvm.assume dominates the use because assumptions have control-flow dependencies). This means that, when we ask if bits are known in a value, we might get different answers for different uses. The significant changes are all in ValueTracking. Two main changes: First, as with the rest of the code, new parameters need to be passed around. To make this easier, I grouped them into a structure, and I made internal static versions of the relevant functions that take this structure as a parameter. The new code does as you might expect, it looks for @llvm.assume calls that make use of the value we're trying to learn something about (often indirectly), attempts to pattern match that expression, and uses the result if successful. By making use of the AssumptionTracker, the process of finding @llvm.assume calls is not expensive. Part of the structure being passed around inside ValueTracking is a set of already-considered @llvm.assume calls. This is to prevent a query using, for example, the assume(a == b), to recurse on itself. The context and DT params are used to find applicable assumptions. An assumption needs to dominate the context instruction, or come after it deterministically. In this latter case we only handle the specific case where both the assumption and the context instruction are in the same block, and we need to exclude assumptions from being used to simplify their own ephemeral values (those which contribute only to the assumption) because otherwise the assumption would prove its feeding comparison trivial and would be removed. This commit adds the plumbing and the logic for a simple masked-bit propagation (just enough to write a regression test). Future commits add more patterns (and, correspondingly, more regression tests). llvm-svn: 217342
2014-09-08 02:57:58 +08:00
computeKnownBits(V, KnownZero, KnownOne, Depth, CxtI);
break;
}
// If the client is only demanding bits that we know, return the known
// constant.
if ((DemandedMask & (KnownZero|KnownOne)) == DemandedMask)
return Constant::getIntegerValue(VTy, KnownOne);
return nullptr;
}
/// Helper routine of SimplifyDemandedUseBits. It tries to simplify
/// "E1 = (X lsr C1) << C2", where the C1 and C2 are constant, into
/// "E2 = X << (C2 - C1)" or "E2 = X >> (C1 - C2)", depending on the sign
/// of "C2-C1".
///
/// Suppose E1 and E2 are generally different in bits S={bm, bm+1,
/// ..., bn}, without considering the specific value X is holding.
/// This transformation is legal iff one of following conditions is hold:
/// 1) All the bit in S are 0, in this case E1 == E2.
/// 2) We don't care those bits in S, per the input DemandedMask.
/// 3) Combination of 1) and 2). Some bits in S are 0, and we don't care the
/// rest bits.
///
/// Currently we only test condition 2).
///
/// As with SimplifyDemandedUseBits, it returns NULL if the simplification was
/// not successful.
Value *InstCombiner::SimplifyShrShlDemandedBits(Instruction *Shr,
Instruction *Shl, APInt DemandedMask, APInt &KnownZero, APInt &KnownOne) {
const APInt &ShlOp1 = cast<ConstantInt>(Shl->getOperand(1))->getValue();
const APInt &ShrOp1 = cast<ConstantInt>(Shr->getOperand(1))->getValue();
if (!ShlOp1 || !ShrOp1)
return nullptr; // Noop.
Value *VarX = Shr->getOperand(0);
Type *Ty = VarX->getType();
unsigned BitWidth = Ty->getIntegerBitWidth();
if (ShlOp1.uge(BitWidth) || ShrOp1.uge(BitWidth))
return nullptr; // Undef.
unsigned ShlAmt = ShlOp1.getZExtValue();
unsigned ShrAmt = ShrOp1.getZExtValue();
KnownOne.clearAllBits();
KnownZero = APInt::getBitsSet(KnownZero.getBitWidth(), 0, ShlAmt-1);
KnownZero &= DemandedMask;
APInt BitMask1(APInt::getAllOnesValue(BitWidth));
APInt BitMask2(APInt::getAllOnesValue(BitWidth));
bool isLshr = (Shr->getOpcode() == Instruction::LShr);
BitMask1 = isLshr ? (BitMask1.lshr(ShrAmt) << ShlAmt) :
(BitMask1.ashr(ShrAmt) << ShlAmt);
if (ShrAmt <= ShlAmt) {
BitMask2 <<= (ShlAmt - ShrAmt);
} else {
BitMask2 = isLshr ? BitMask2.lshr(ShrAmt - ShlAmt):
BitMask2.ashr(ShrAmt - ShlAmt);
}
// Check if condition-2 (see the comment to this function) is satified.
if ((BitMask1 & DemandedMask) == (BitMask2 & DemandedMask)) {
if (ShrAmt == ShlAmt)
return VarX;
if (!Shr->hasOneUse())
return nullptr;
BinaryOperator *New;
if (ShrAmt < ShlAmt) {
Constant *Amt = ConstantInt::get(VarX->getType(), ShlAmt - ShrAmt);
New = BinaryOperator::CreateShl(VarX, Amt);
BinaryOperator *Orig = cast<BinaryOperator>(Shl);
New->setHasNoSignedWrap(Orig->hasNoSignedWrap());
New->setHasNoUnsignedWrap(Orig->hasNoUnsignedWrap());
} else {
Constant *Amt = ConstantInt::get(VarX->getType(), ShrAmt - ShlAmt);
New = isLshr ? BinaryOperator::CreateLShr(VarX, Amt) :
BinaryOperator::CreateAShr(VarX, Amt);
if (cast<BinaryOperator>(Shr)->isExact())
New->setIsExact(true);
}
return InsertNewInstWith(New, *Shl);
}
return nullptr;
}
/// SimplifyDemandedVectorElts - The specified value produces a vector with
/// any number of elements. DemandedElts contains the set of elements that are
/// actually used by the caller. This method analyzes which elements of the
/// operand are undef and returns that information in UndefElts.
///
/// If the information about demanded elements can be used to simplify the
/// operation, the operation is simplified, then the resultant value is
/// returned. This returns null if no change was made.
Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
APInt &UndefElts,
unsigned Depth) {
unsigned VWidth = cast<VectorType>(V->getType())->getNumElements();
APInt EltMask(APInt::getAllOnesValue(VWidth));
assert((DemandedElts & ~EltMask) == 0 && "Invalid DemandedElts!");
if (isa<UndefValue>(V)) {
// If the entire vector is undefined, just return this info.
UndefElts = EltMask;
return nullptr;
}
if (DemandedElts == 0) { // If nothing is demanded, provide undef.
UndefElts = EltMask;
return UndefValue::get(V->getType());
}
UndefElts = 0;
// Handle ConstantAggregateZero, ConstantVector, ConstantDataSequential.
if (Constant *C = dyn_cast<Constant>(V)) {
// Check if this is identity. If so, return 0 since we are not simplifying
// anything.
if (DemandedElts.isAllOnesValue())
return nullptr;
Type *EltTy = cast<VectorType>(V->getType())->getElementType();
Constant *Undef = UndefValue::get(EltTy);
SmallVector<Constant*, 16> Elts;
for (unsigned i = 0; i != VWidth; ++i) {
if (!DemandedElts[i]) { // If not demanded, set to undef.
Elts.push_back(Undef);
UndefElts.setBit(i);
continue;
}
Constant *Elt = C->getAggregateElement(i);
if (!Elt) return nullptr;
if (isa<UndefValue>(Elt)) { // Already undef.
Elts.push_back(Undef);
UndefElts.setBit(i);
} else { // Otherwise, defined.
Elts.push_back(Elt);
}
}
// If we changed the constant, return it.
Constant *NewCV = ConstantVector::get(Elts);
return NewCV != C ? NewCV : nullptr;
}
// Limit search depth.
if (Depth == 10)
return nullptr;
// If multiple users are using the root value, proceed with
// simplification conservatively assuming that all elements
// are needed.
if (!V->hasOneUse()) {
// Quit if we find multiple users of a non-root value though.
// They'll be handled when it's their turn to be visited by
// the main instcombine process.
if (Depth != 0)
// TODO: Just compute the UndefElts information recursively.
return nullptr;
// Conservatively assume that all elements are needed.
DemandedElts = EltMask;
}
Instruction *I = dyn_cast<Instruction>(V);
if (!I) return nullptr; // Only analyze instructions.
bool MadeChange = false;
APInt UndefElts2(VWidth, 0);
Value *TmpV;
switch (I->getOpcode()) {
default: break;
case Instruction::InsertElement: {
// If this is a variable index, we don't know which element it overwrites.
// demand exactly the same input as we produce.
ConstantInt *Idx = dyn_cast<ConstantInt>(I->getOperand(2));
if (!Idx) {
// Note that we can't propagate undef elt info, because we don't know
// which elt is getting updated.
TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts,
UndefElts2, Depth + 1);
if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
break;
}
// If this is inserting an element that isn't demanded, remove this
// insertelement.
unsigned IdxNo = Idx->getZExtValue();
if (IdxNo >= VWidth || !DemandedElts[IdxNo]) {
Worklist.Add(I);
return I->getOperand(0);
}
// Otherwise, the element inserted overwrites whatever was there, so the
// input demanded set is simpler than the output set.
APInt DemandedElts2 = DemandedElts;
DemandedElts2.clearBit(IdxNo);
TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts2,
UndefElts, Depth + 1);
if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
// The inserted element is defined.
UndefElts.clearBit(IdxNo);
break;
}
case Instruction::ShuffleVector: {
ShuffleVectorInst *Shuffle = cast<ShuffleVectorInst>(I);
uint64_t LHSVWidth =
cast<VectorType>(Shuffle->getOperand(0)->getType())->getNumElements();
APInt LeftDemanded(LHSVWidth, 0), RightDemanded(LHSVWidth, 0);
for (unsigned i = 0; i < VWidth; i++) {
if (DemandedElts[i]) {
unsigned MaskVal = Shuffle->getMaskValue(i);
if (MaskVal != -1u) {
assert(MaskVal < LHSVWidth * 2 &&
"shufflevector mask index out of range!");
if (MaskVal < LHSVWidth)
LeftDemanded.setBit(MaskVal);
else
RightDemanded.setBit(MaskVal - LHSVWidth);
}
}
}
APInt UndefElts4(LHSVWidth, 0);
TmpV = SimplifyDemandedVectorElts(I->getOperand(0), LeftDemanded,
UndefElts4, Depth + 1);
if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
APInt UndefElts3(LHSVWidth, 0);
TmpV = SimplifyDemandedVectorElts(I->getOperand(1), RightDemanded,
UndefElts3, Depth + 1);
if (TmpV) { I->setOperand(1, TmpV); MadeChange = true; }
bool NewUndefElts = false;
for (unsigned i = 0; i < VWidth; i++) {
unsigned MaskVal = Shuffle->getMaskValue(i);
if (MaskVal == -1u) {
UndefElts.setBit(i);
} else if (!DemandedElts[i]) {
NewUndefElts = true;
UndefElts.setBit(i);
} else if (MaskVal < LHSVWidth) {
if (UndefElts4[MaskVal]) {
NewUndefElts = true;
UndefElts.setBit(i);
}
} else {
if (UndefElts3[MaskVal - LHSVWidth]) {
NewUndefElts = true;
UndefElts.setBit(i);
}
}
}
if (NewUndefElts) {
// Add additional discovered undefs.
SmallVector<Constant*, 16> Elts;
for (unsigned i = 0; i < VWidth; ++i) {
if (UndefElts[i])
Elts.push_back(UndefValue::get(Type::getInt32Ty(I->getContext())));
else
Elts.push_back(ConstantInt::get(Type::getInt32Ty(I->getContext()),
Shuffle->getMaskValue(i)));
}
I->setOperand(2, ConstantVector::get(Elts));
MadeChange = true;
}
break;
}
case Instruction::Select: {
APInt LeftDemanded(DemandedElts), RightDemanded(DemandedElts);
if (ConstantVector* CV = dyn_cast<ConstantVector>(I->getOperand(0))) {
for (unsigned i = 0; i < VWidth; i++) {
if (CV->getAggregateElement(i)->isNullValue())
LeftDemanded.clearBit(i);
else
RightDemanded.clearBit(i);
}
}
TmpV = SimplifyDemandedVectorElts(I->getOperand(1), LeftDemanded, UndefElts,
Depth + 1);
if (TmpV) { I->setOperand(1, TmpV); MadeChange = true; }
TmpV = SimplifyDemandedVectorElts(I->getOperand(2), RightDemanded,
UndefElts2, Depth + 1);
if (TmpV) { I->setOperand(2, TmpV); MadeChange = true; }
// Output elements are undefined if both are undefined.
UndefElts &= UndefElts2;
break;
}
case Instruction::BitCast: {
// Vector->vector casts only.
VectorType *VTy = dyn_cast<VectorType>(I->getOperand(0)->getType());
if (!VTy) break;
unsigned InVWidth = VTy->getNumElements();
APInt InputDemandedElts(InVWidth, 0);
unsigned Ratio;
if (VWidth == InVWidth) {
// If we are converting from <4 x i32> -> <4 x f32>, we demand the same
// elements as are demanded of us.
Ratio = 1;
InputDemandedElts = DemandedElts;
} else if (VWidth > InVWidth) {
// Untested so far.
break;
// If there are more elements in the result than there are in the source,
// then an input element is live if any of the corresponding output
// elements are live.
Ratio = VWidth/InVWidth;
for (unsigned OutIdx = 0; OutIdx != VWidth; ++OutIdx) {
if (DemandedElts[OutIdx])
InputDemandedElts.setBit(OutIdx/Ratio);
}
} else {
// Untested so far.
break;
// If there are more elements in the source than there are in the result,
// then an input element is live if the corresponding output element is
// live.
Ratio = InVWidth/VWidth;
for (unsigned InIdx = 0; InIdx != InVWidth; ++InIdx)
if (DemandedElts[InIdx/Ratio])
InputDemandedElts.setBit(InIdx);
}
// div/rem demand all inputs, because they don't want divide by zero.
TmpV = SimplifyDemandedVectorElts(I->getOperand(0), InputDemandedElts,
UndefElts2, Depth + 1);
if (TmpV) {
I->setOperand(0, TmpV);
MadeChange = true;
}
UndefElts = UndefElts2;
if (VWidth > InVWidth) {
llvm_unreachable("Unimp");
// If there are more elements in the result than there are in the source,
// then an output element is undef if the corresponding input element is
// undef.
for (unsigned OutIdx = 0; OutIdx != VWidth; ++OutIdx)
if (UndefElts2[OutIdx/Ratio])
UndefElts.setBit(OutIdx);
} else if (VWidth < InVWidth) {
llvm_unreachable("Unimp");
// If there are more elements in the source than there are in the result,
// then a result element is undef if all of the corresponding input
// elements are undef.
UndefElts = ~0ULL >> (64-VWidth); // Start out all undef.
for (unsigned InIdx = 0; InIdx != InVWidth; ++InIdx)
if (!UndefElts2[InIdx]) // Not undef?
UndefElts.clearBit(InIdx/Ratio); // Clear undef bit.
}
break;
}
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
case Instruction::Add:
case Instruction::Sub:
case Instruction::Mul:
// div/rem demand all inputs, because they don't want divide by zero.
TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts, UndefElts,
Depth + 1);
if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
TmpV = SimplifyDemandedVectorElts(I->getOperand(1), DemandedElts,
UndefElts2, Depth + 1);
if (TmpV) { I->setOperand(1, TmpV); MadeChange = true; }
// Output elements are undefined if both are undefined. Consider things
// like undef&0. The result is known zero, not undef.
UndefElts &= UndefElts2;
break;
case Instruction::FPTrunc:
case Instruction::FPExt:
TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts, UndefElts,
Depth + 1);
if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
break;
case Instruction::Call: {
IntrinsicInst *II = dyn_cast<IntrinsicInst>(I);
if (!II) break;
switch (II->getIntrinsicID()) {
default: break;
// Binary vector operations that work column-wise. A dest element is a
// function of the corresponding input elements from the two inputs.
case Intrinsic::x86_sse_sub_ss:
case Intrinsic::x86_sse_mul_ss:
case Intrinsic::x86_sse_min_ss:
case Intrinsic::x86_sse_max_ss:
case Intrinsic::x86_sse2_sub_sd:
case Intrinsic::x86_sse2_mul_sd:
case Intrinsic::x86_sse2_min_sd:
case Intrinsic::x86_sse2_max_sd:
2010-06-29 00:45:00 +08:00
TmpV = SimplifyDemandedVectorElts(II->getArgOperand(0), DemandedElts,
UndefElts, Depth + 1);
2010-06-29 00:45:00 +08:00
if (TmpV) { II->setArgOperand(0, TmpV); MadeChange = true; }
TmpV = SimplifyDemandedVectorElts(II->getArgOperand(1), DemandedElts,
UndefElts2, Depth + 1);
2010-06-29 00:45:00 +08:00
if (TmpV) { II->setArgOperand(1, TmpV); MadeChange = true; }
// If only the low elt is demanded and this is a scalarizable intrinsic,
// scalarize it now.
if (DemandedElts == 1) {
switch (II->getIntrinsicID()) {
default: break;
case Intrinsic::x86_sse_sub_ss:
case Intrinsic::x86_sse_mul_ss:
case Intrinsic::x86_sse2_sub_sd:
case Intrinsic::x86_sse2_mul_sd:
// TODO: Lower MIN/MAX/ABS/etc
2010-06-24 20:35:13 +08:00
Value *LHS = II->getArgOperand(0);
Value *RHS = II->getArgOperand(1);
// Extract the element as scalars.
LHS = InsertNewInstWith(ExtractElementInst::Create(LHS,
ConstantInt::get(Type::getInt32Ty(I->getContext()), 0U)), *II);
RHS = InsertNewInstWith(ExtractElementInst::Create(RHS,
ConstantInt::get(Type::getInt32Ty(I->getContext()), 0U)), *II);
switch (II->getIntrinsicID()) {
default: llvm_unreachable("Case stmts out of sync!");
case Intrinsic::x86_sse_sub_ss:
case Intrinsic::x86_sse2_sub_sd:
TmpV = InsertNewInstWith(BinaryOperator::CreateFSub(LHS, RHS,
II->getName()), *II);
break;
case Intrinsic::x86_sse_mul_ss:
case Intrinsic::x86_sse2_mul_sd:
TmpV = InsertNewInstWith(BinaryOperator::CreateFMul(LHS, RHS,
II->getName()), *II);
break;
}
Instruction *New =
InsertElementInst::Create(
UndefValue::get(II->getType()), TmpV,
ConstantInt::get(Type::getInt32Ty(I->getContext()), 0U, false),
II->getName());
InsertNewInstWith(New, *II);
return New;
}
}
// Output elements are undefined if both are undefined. Consider things
// like undef&0. The result is known zero, not undef.
UndefElts &= UndefElts2;
break;
}
break;
}
}
return MadeChange ? I : nullptr;
}