2005-01-07 15:46:32 +08:00
|
|
|
//===-- SelectionDAG.cpp - Implement the SelectionDAG data structures -----===//
|
2005-04-22 06:36:52 +08:00
|
|
|
//
|
2003-10-21 03:43:21 +08:00
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
|
|
|
// This file was developed by the LLVM research group and is distributed under
|
|
|
|
// the University of Illinois Open Source License. See LICENSE.TXT for details.
|
2005-04-22 06:36:52 +08:00
|
|
|
//
|
2003-10-21 03:43:21 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
2003-08-11 22:57:33 +08:00
|
|
|
//
|
2005-01-07 15:46:32 +08:00
|
|
|
// This implements the SelectionDAG class.
|
2003-08-11 22:57:33 +08:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
#include "llvm/CodeGen/SelectionDAG.h"
|
2005-01-07 15:46:32 +08:00
|
|
|
#include "llvm/Constants.h"
|
|
|
|
#include "llvm/GlobalValue.h"
|
|
|
|
#include "llvm/Assembly/Writer.h"
|
|
|
|
#include "llvm/CodeGen/MachineBasicBlock.h"
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
#include "llvm/Target/TargetLowering.h"
|
2004-07-04 20:19:56 +08:00
|
|
|
#include <iostream>
|
2005-01-08 05:09:16 +08:00
|
|
|
#include <set>
|
2005-01-07 15:46:32 +08:00
|
|
|
#include <cmath>
|
2005-01-10 04:41:56 +08:00
|
|
|
#include <algorithm>
|
2004-06-02 12:28:06 +08:00
|
|
|
using namespace llvm;
|
2003-11-12 06:41:34 +08:00
|
|
|
|
Add some folds for == and != comparisons. This allows us to
codegen this loop in stepanov:
no_exit.i: ; preds = %entry, %no_exit.i, %then.i, %_Z5checkd.exit
%i.0.0 = phi int [ 0, %entry ], [ %i.0.0, %no_exit.i ], [ %inc.0, %_Z5checkd.exit ], [ %inc.012, %then.i ] ; <int> [#uses=3]
%indvar = phi uint [ %indvar.next, %no_exit.i ], [ 0, %entry ], [ 0, %then.i ], [ 0, %_Z5checkd.exit ] ; <uint> [#uses=3]
%result_addr.i.0 = phi double [ %tmp.4.i.i, %no_exit.i ], [ 0.000000e+00, %entry ], [ 0.000000e+00, %then.i ], [ 0.000000e+00, %_Z5checkd.exit ] ; <double> [#uses=1]
%first_addr.0.i.2.rec = cast uint %indvar to int ; <int> [#uses=1]
%first_addr.0.i.2 = getelementptr [2000 x double]* %data, int 0, uint %indvar ; <double*> [#uses=1]
%inc.i.rec = add int %first_addr.0.i.2.rec, 1 ; <int> [#uses=1]
%inc.i = getelementptr [2000 x double]* %data, int 0, int %inc.i.rec ; <double*> [#uses=1]
%tmp.3.i.i = load double* %first_addr.0.i.2 ; <double> [#uses=1]
%tmp.4.i.i = add double %result_addr.i.0, %tmp.3.i.i ; <double> [#uses=2]
%tmp.2.i = seteq double* %inc.i, getelementptr ([2000 x double]* %data, int 0, int 2000) ; <bool> [#uses=1]
%indvar.next = add uint %indvar, 1 ; <uint> [#uses=1]
br bool %tmp.2.i, label %_Z10accumulateIPddET0_T_S2_S1_.exit, label %no_exit.i
To this:
.LBB_Z4testIPddEvT_S1_T0__1: # no_exit.i
fldl data(,%eax,8)
fldl 16(%esp)
faddp %st(1)
fstpl 16(%esp)
incl %eax
movl %eax, %ecx
shll $3, %ecx
cmpl $16000, %ecx
#FP_REG_KILL
jne .LBB_Z4testIPddEvT_S1_T0__1 # no_exit.i
instead of this:
.LBB_Z4testIPddEvT_S1_T0__1: # no_exit.i
fldl data(,%eax,8)
fldl 16(%esp)
faddp %st(1)
fstpl 16(%esp)
incl %eax
leal data(,%eax,8), %ecx
leal data+16000, %edx
cmpl %edx, %ecx
#FP_REG_KILL
jne .LBB_Z4testIPddEvT_S1_T0__1 # no_exit.i
llvm-svn: 19425
2005-01-10 04:52:51 +08:00
|
|
|
static bool isCommutativeBinOp(unsigned Opcode) {
|
|
|
|
switch (Opcode) {
|
|
|
|
case ISD::ADD:
|
|
|
|
case ISD::MUL:
|
|
|
|
case ISD::AND:
|
|
|
|
case ISD::OR:
|
|
|
|
case ISD::XOR: return true;
|
|
|
|
default: return false; // FIXME: Need commutative info for user ops!
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool isAssociativeBinOp(unsigned Opcode) {
|
|
|
|
switch (Opcode) {
|
|
|
|
case ISD::ADD:
|
|
|
|
case ISD::MUL:
|
|
|
|
case ISD::AND:
|
|
|
|
case ISD::OR:
|
|
|
|
case ISD::XOR: return true;
|
|
|
|
default: return false; // FIXME: Need associative info for user ops!
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static unsigned ExactLog2(uint64_t Val) {
|
|
|
|
unsigned Count = 0;
|
|
|
|
while (Val != 1) {
|
|
|
|
Val >>= 1;
|
|
|
|
++Count;
|
|
|
|
}
|
|
|
|
return Count;
|
|
|
|
}
|
|
|
|
|
|
|
|
// isInvertibleForFree - Return true if there is no cost to emitting the logical
|
|
|
|
// inverse of this node.
|
|
|
|
static bool isInvertibleForFree(SDOperand N) {
|
|
|
|
if (isa<ConstantSDNode>(N.Val)) return true;
|
|
|
|
if (isa<SetCCSDNode>(N.Val) && N.Val->hasOneUse())
|
|
|
|
return true;
|
2005-04-22 06:36:52 +08:00
|
|
|
return false;
|
Add some folds for == and != comparisons. This allows us to
codegen this loop in stepanov:
no_exit.i: ; preds = %entry, %no_exit.i, %then.i, %_Z5checkd.exit
%i.0.0 = phi int [ 0, %entry ], [ %i.0.0, %no_exit.i ], [ %inc.0, %_Z5checkd.exit ], [ %inc.012, %then.i ] ; <int> [#uses=3]
%indvar = phi uint [ %indvar.next, %no_exit.i ], [ 0, %entry ], [ 0, %then.i ], [ 0, %_Z5checkd.exit ] ; <uint> [#uses=3]
%result_addr.i.0 = phi double [ %tmp.4.i.i, %no_exit.i ], [ 0.000000e+00, %entry ], [ 0.000000e+00, %then.i ], [ 0.000000e+00, %_Z5checkd.exit ] ; <double> [#uses=1]
%first_addr.0.i.2.rec = cast uint %indvar to int ; <int> [#uses=1]
%first_addr.0.i.2 = getelementptr [2000 x double]* %data, int 0, uint %indvar ; <double*> [#uses=1]
%inc.i.rec = add int %first_addr.0.i.2.rec, 1 ; <int> [#uses=1]
%inc.i = getelementptr [2000 x double]* %data, int 0, int %inc.i.rec ; <double*> [#uses=1]
%tmp.3.i.i = load double* %first_addr.0.i.2 ; <double> [#uses=1]
%tmp.4.i.i = add double %result_addr.i.0, %tmp.3.i.i ; <double> [#uses=2]
%tmp.2.i = seteq double* %inc.i, getelementptr ([2000 x double]* %data, int 0, int 2000) ; <bool> [#uses=1]
%indvar.next = add uint %indvar, 1 ; <uint> [#uses=1]
br bool %tmp.2.i, label %_Z10accumulateIPddET0_T_S2_S1_.exit, label %no_exit.i
To this:
.LBB_Z4testIPddEvT_S1_T0__1: # no_exit.i
fldl data(,%eax,8)
fldl 16(%esp)
faddp %st(1)
fstpl 16(%esp)
incl %eax
movl %eax, %ecx
shll $3, %ecx
cmpl $16000, %ecx
#FP_REG_KILL
jne .LBB_Z4testIPddEvT_S1_T0__1 # no_exit.i
instead of this:
.LBB_Z4testIPddEvT_S1_T0__1: # no_exit.i
fldl data(,%eax,8)
fldl 16(%esp)
faddp %st(1)
fstpl 16(%esp)
incl %eax
leal data(,%eax,8), %ecx
leal data+16000, %edx
cmpl %edx, %ecx
#FP_REG_KILL
jne .LBB_Z4testIPddEvT_S1_T0__1 # no_exit.i
llvm-svn: 19425
2005-01-10 04:52:51 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
/// getSetCCSwappedOperands - Return the operation corresponding to (Y op X)
|
|
|
|
/// when given the operation for (X op Y).
|
|
|
|
ISD::CondCode ISD::getSetCCSwappedOperands(ISD::CondCode Operation) {
|
|
|
|
// To perform this operation, we just need to swap the L and G bits of the
|
|
|
|
// operation.
|
|
|
|
unsigned OldL = (Operation >> 2) & 1;
|
|
|
|
unsigned OldG = (Operation >> 1) & 1;
|
|
|
|
return ISD::CondCode((Operation & ~6) | // Keep the N, U, E bits
|
|
|
|
(OldL << 1) | // New G bit
|
|
|
|
(OldG << 2)); // New L bit.
|
|
|
|
}
|
|
|
|
|
|
|
|
/// getSetCCInverse - Return the operation corresponding to !(X op Y), where
|
|
|
|
/// 'op' is a valid SetCC operation.
|
|
|
|
ISD::CondCode ISD::getSetCCInverse(ISD::CondCode Op, bool isInteger) {
|
|
|
|
unsigned Operation = Op;
|
|
|
|
if (isInteger)
|
|
|
|
Operation ^= 7; // Flip L, G, E bits, but not U.
|
|
|
|
else
|
|
|
|
Operation ^= 15; // Flip all of the condition bits.
|
|
|
|
if (Operation > ISD::SETTRUE2)
|
|
|
|
Operation &= ~8; // Don't let N and U bits get set.
|
|
|
|
return ISD::CondCode(Operation);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/// isSignedOp - For an integer comparison, return 1 if the comparison is a
|
|
|
|
/// signed operation and 2 if the result is an unsigned comparison. Return zero
|
|
|
|
/// if the operation does not depend on the sign of the input (setne and seteq).
|
|
|
|
static int isSignedOp(ISD::CondCode Opcode) {
|
|
|
|
switch (Opcode) {
|
|
|
|
default: assert(0 && "Illegal integer setcc operation!");
|
|
|
|
case ISD::SETEQ:
|
|
|
|
case ISD::SETNE: return 0;
|
|
|
|
case ISD::SETLT:
|
|
|
|
case ISD::SETLE:
|
|
|
|
case ISD::SETGT:
|
|
|
|
case ISD::SETGE: return 1;
|
|
|
|
case ISD::SETULT:
|
|
|
|
case ISD::SETULE:
|
|
|
|
case ISD::SETUGT:
|
|
|
|
case ISD::SETUGE: return 2;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/// getSetCCOrOperation - Return the result of a logical OR between different
|
|
|
|
/// comparisons of identical values: ((X op1 Y) | (X op2 Y)). This function
|
|
|
|
/// returns SETCC_INVALID if it is not possible to represent the resultant
|
|
|
|
/// comparison.
|
|
|
|
ISD::CondCode ISD::getSetCCOrOperation(ISD::CondCode Op1, ISD::CondCode Op2,
|
|
|
|
bool isInteger) {
|
|
|
|
if (isInteger && (isSignedOp(Op1) | isSignedOp(Op2)) == 3)
|
|
|
|
// Cannot fold a signed integer setcc with an unsigned integer setcc.
|
|
|
|
return ISD::SETCC_INVALID;
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
unsigned Op = Op1 | Op2; // Combine all of the condition bits.
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
// If the N and U bits get set then the resultant comparison DOES suddenly
|
|
|
|
// care about orderedness, and is true when ordered.
|
|
|
|
if (Op > ISD::SETTRUE2)
|
|
|
|
Op &= ~16; // Clear the N bit.
|
|
|
|
return ISD::CondCode(Op);
|
|
|
|
}
|
|
|
|
|
|
|
|
/// getSetCCAndOperation - Return the result of a logical AND between different
|
|
|
|
/// comparisons of identical values: ((X op1 Y) & (X op2 Y)). This
|
|
|
|
/// function returns zero if it is not possible to represent the resultant
|
|
|
|
/// comparison.
|
|
|
|
ISD::CondCode ISD::getSetCCAndOperation(ISD::CondCode Op1, ISD::CondCode Op2,
|
|
|
|
bool isInteger) {
|
|
|
|
if (isInteger && (isSignedOp(Op1) | isSignedOp(Op2)) == 3)
|
|
|
|
// Cannot fold a signed setcc with an unsigned setcc.
|
2005-04-22 06:36:52 +08:00
|
|
|
return ISD::SETCC_INVALID;
|
2005-01-07 15:46:32 +08:00
|
|
|
|
|
|
|
// Combine all of the condition bits.
|
|
|
|
return ISD::CondCode(Op1 & Op2);
|
|
|
|
}
|
|
|
|
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
const TargetMachine &SelectionDAG::getTarget() const {
|
|
|
|
return TLI.getTargetMachine();
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-01-08 05:09:16 +08:00
|
|
|
/// RemoveDeadNodes - This method deletes all unreachable nodes in the
|
|
|
|
/// SelectionDAG, including nodes (like loads) that have uses of their token
|
|
|
|
/// chain but no other uses and no side effect. If a node is passed in as an
|
|
|
|
/// argument, it is used as the seed for node deletion.
|
|
|
|
void SelectionDAG::RemoveDeadNodes(SDNode *N) {
|
|
|
|
std::set<SDNode*> AllNodeSet(AllNodes.begin(), AllNodes.end());
|
|
|
|
|
|
|
|
// Create a dummy node (which is not added to allnodes), that adds a reference
|
|
|
|
// to the root node, preventing it from being deleted.
|
|
|
|
SDNode *DummyNode = new SDNode(ISD::EntryToken, getRoot());
|
|
|
|
|
|
|
|
DeleteNodeIfDead(N, &AllNodeSet);
|
|
|
|
|
|
|
|
Restart:
|
|
|
|
unsigned NumNodes = AllNodeSet.size();
|
|
|
|
for (std::set<SDNode*>::iterator I = AllNodeSet.begin(), E = AllNodeSet.end();
|
|
|
|
I != E; ++I) {
|
|
|
|
// Try to delete this node.
|
|
|
|
DeleteNodeIfDead(*I, &AllNodeSet);
|
|
|
|
|
|
|
|
// If we actually deleted any nodes, do not use invalid iterators in
|
|
|
|
// AllNodeSet.
|
|
|
|
if (AllNodeSet.size() != NumNodes)
|
|
|
|
goto Restart;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Restore AllNodes.
|
|
|
|
if (AllNodes.size() != NumNodes)
|
|
|
|
AllNodes.assign(AllNodeSet.begin(), AllNodeSet.end());
|
|
|
|
|
|
|
|
// If the root changed (e.g. it was a dead load, update the root).
|
|
|
|
setRoot(DummyNode->getOperand(0));
|
|
|
|
|
|
|
|
// Now that we are done with the dummy node, delete it.
|
|
|
|
DummyNode->getOperand(0).Val->removeUser(DummyNode);
|
|
|
|
delete DummyNode;
|
|
|
|
}
|
|
|
|
|
|
|
|
void SelectionDAG::DeleteNodeIfDead(SDNode *N, void *NodeSet) {
|
|
|
|
if (!N->use_empty())
|
|
|
|
return;
|
|
|
|
|
|
|
|
// Okay, we really are going to delete this node. First take this out of the
|
|
|
|
// appropriate CSE map.
|
|
|
|
switch (N->getOpcode()) {
|
|
|
|
case ISD::Constant:
|
|
|
|
Constants.erase(std::make_pair(cast<ConstantSDNode>(N)->getValue(),
|
|
|
|
N->getValueType(0)));
|
|
|
|
break;
|
2005-02-18 04:17:32 +08:00
|
|
|
case ISD::ConstantFP: {
|
|
|
|
union {
|
|
|
|
double DV;
|
|
|
|
uint64_t IV;
|
|
|
|
};
|
|
|
|
DV = cast<ConstantFPSDNode>(N)->getValue();
|
|
|
|
ConstantFPs.erase(std::make_pair(IV, N->getValueType(0)));
|
2005-01-08 05:09:16 +08:00
|
|
|
break;
|
2005-02-18 04:17:32 +08:00
|
|
|
}
|
2005-01-08 05:09:16 +08:00
|
|
|
case ISD::GlobalAddress:
|
|
|
|
GlobalValues.erase(cast<GlobalAddressSDNode>(N)->getGlobal());
|
|
|
|
break;
|
|
|
|
case ISD::FrameIndex:
|
|
|
|
FrameIndices.erase(cast<FrameIndexSDNode>(N)->getIndex());
|
|
|
|
break;
|
|
|
|
case ISD::ConstantPool:
|
|
|
|
ConstantPoolIndices.erase(cast<ConstantPoolSDNode>(N)->getIndex());
|
|
|
|
break;
|
|
|
|
case ISD::BasicBlock:
|
|
|
|
BBNodes.erase(cast<BasicBlockSDNode>(N)->getBasicBlock());
|
|
|
|
break;
|
|
|
|
case ISD::ExternalSymbol:
|
|
|
|
ExternalSymbols.erase(cast<ExternalSymbolSDNode>(N)->getSymbol());
|
|
|
|
break;
|
|
|
|
|
|
|
|
case ISD::LOAD:
|
|
|
|
Loads.erase(std::make_pair(N->getOperand(1),
|
|
|
|
std::make_pair(N->getOperand(0),
|
|
|
|
N->getValueType(0))));
|
|
|
|
break;
|
|
|
|
case ISD::SETCC:
|
|
|
|
SetCCs.erase(std::make_pair(std::make_pair(N->getOperand(0),
|
|
|
|
N->getOperand(1)),
|
2005-01-19 03:26:36 +08:00
|
|
|
std::make_pair(
|
|
|
|
cast<SetCCSDNode>(N)->getCondition(),
|
|
|
|
N->getValueType(0))));
|
2005-01-08 05:09:16 +08:00
|
|
|
break;
|
2005-01-15 14:17:04 +08:00
|
|
|
case ISD::TRUNCSTORE:
|
|
|
|
case ISD::SIGN_EXTEND_INREG:
|
|
|
|
case ISD::FP_ROUND_INREG:
|
2005-01-15 06:38:01 +08:00
|
|
|
case ISD::EXTLOAD:
|
|
|
|
case ISD::SEXTLOAD:
|
|
|
|
case ISD::ZEXTLOAD: {
|
|
|
|
EVTStruct NN;
|
2005-04-07 08:30:13 +08:00
|
|
|
NN.Opcode = N->getOpcode();
|
2005-01-15 06:38:01 +08:00
|
|
|
NN.VT = N->getValueType(0);
|
|
|
|
NN.EVT = cast<MVTSDNode>(N)->getExtraValueType();
|
2005-01-15 14:17:04 +08:00
|
|
|
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
|
|
|
|
NN.Ops.push_back(N->getOperand(i));
|
2005-01-15 06:38:01 +08:00
|
|
|
MVTSDNodes.erase(NN);
|
|
|
|
break;
|
|
|
|
}
|
2005-01-08 05:09:16 +08:00
|
|
|
default:
|
|
|
|
if (N->getNumOperands() == 1)
|
|
|
|
UnaryOps.erase(std::make_pair(N->getOpcode(),
|
|
|
|
std::make_pair(N->getOperand(0),
|
|
|
|
N->getValueType(0))));
|
|
|
|
else if (N->getNumOperands() == 2)
|
|
|
|
BinaryOps.erase(std::make_pair(N->getOpcode(),
|
|
|
|
std::make_pair(N->getOperand(0),
|
|
|
|
N->getOperand(1))));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Next, brutally remove the operand list.
|
|
|
|
while (!N->Operands.empty()) {
|
2005-01-08 07:32:00 +08:00
|
|
|
SDNode *O = N->Operands.back().Val;
|
2005-01-08 05:09:16 +08:00
|
|
|
N->Operands.pop_back();
|
2005-01-08 07:32:00 +08:00
|
|
|
O->removeUser(N);
|
|
|
|
|
|
|
|
// Now that we removed this operand, see if there are no uses of it left.
|
|
|
|
DeleteNodeIfDead(O, NodeSet);
|
2005-01-08 05:09:16 +08:00
|
|
|
}
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-08 05:09:16 +08:00
|
|
|
// Remove the node from the nodes set and delete it.
|
|
|
|
std::set<SDNode*> &AllNodeSet = *(std::set<SDNode*>*)NodeSet;
|
|
|
|
AllNodeSet.erase(N);
|
|
|
|
|
|
|
|
// Now that the node is gone, check to see if any of the operands of this node
|
|
|
|
// are dead now.
|
2005-01-08 07:32:00 +08:00
|
|
|
delete N;
|
2005-01-08 05:09:16 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2003-08-11 22:57:33 +08:00
|
|
|
SelectionDAG::~SelectionDAG() {
|
|
|
|
for (unsigned i = 0, e = AllNodes.size(); i != e; ++i)
|
|
|
|
delete AllNodes[i];
|
|
|
|
}
|
|
|
|
|
2005-04-13 10:38:18 +08:00
|
|
|
SDOperand SelectionDAG::getZeroExtendInReg(SDOperand Op, MVT::ValueType VT) {
|
2005-04-14 03:41:05 +08:00
|
|
|
if (Op.getValueType() == VT) return Op;
|
2005-05-10 10:22:38 +08:00
|
|
|
int64_t Imm = ~0ULL >> (64-MVT::getSizeInBits(VT));
|
2005-04-13 10:38:18 +08:00
|
|
|
return getNode(ISD::AND, Op.getValueType(), Op,
|
|
|
|
getConstant(Imm, Op.getValueType()));
|
|
|
|
}
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
SDOperand SelectionDAG::getConstant(uint64_t Val, MVT::ValueType VT) {
|
|
|
|
assert(MVT::isInteger(VT) && "Cannot create FP integer constant!");
|
|
|
|
// Mask out any bits that are not valid for this constant.
|
2005-01-08 14:24:30 +08:00
|
|
|
if (VT != MVT::i64)
|
|
|
|
Val &= ((uint64_t)1 << MVT::getSizeInBits(VT)) - 1;
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
SDNode *&N = Constants[std::make_pair(Val, VT)];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new ConstantSDNode(Val, VT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
2003-08-11 22:57:33 +08:00
|
|
|
}
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
SDOperand SelectionDAG::getConstantFP(double Val, MVT::ValueType VT) {
|
|
|
|
assert(MVT::isFloatingPoint(VT) && "Cannot create integer FP constant!");
|
|
|
|
if (VT == MVT::f32)
|
|
|
|
Val = (float)Val; // Mask out extra precision.
|
|
|
|
|
2005-02-18 04:17:32 +08:00
|
|
|
// Do the map lookup using the actual bit pattern for the floating point
|
|
|
|
// value, so that we don't have problems with 0.0 comparing equal to -0.0, and
|
|
|
|
// we don't have issues with SNANs.
|
|
|
|
union {
|
|
|
|
double DV;
|
|
|
|
uint64_t IV;
|
|
|
|
};
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-02-18 04:17:32 +08:00
|
|
|
DV = Val;
|
|
|
|
|
|
|
|
SDNode *&N = ConstantFPs[std::make_pair(IV, VT)];
|
2005-01-07 15:46:32 +08:00
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new ConstantFPSDNode(Val, VT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
SDOperand SelectionDAG::getGlobalAddress(const GlobalValue *GV,
|
|
|
|
MVT::ValueType VT) {
|
|
|
|
SDNode *&N = GlobalValues[GV];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new GlobalAddressSDNode(GV,VT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand SelectionDAG::getFrameIndex(int FI, MVT::ValueType VT) {
|
|
|
|
SDNode *&N = FrameIndices[FI];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new FrameIndexSDNode(FI, VT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand SelectionDAG::getConstantPool(unsigned CPIdx, MVT::ValueType VT) {
|
|
|
|
SDNode *N = ConstantPoolIndices[CPIdx];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new ConstantPoolSDNode(CPIdx, VT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand SelectionDAG::getBasicBlock(MachineBasicBlock *MBB) {
|
|
|
|
SDNode *&N = BBNodes[MBB];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new BasicBlockSDNode(MBB);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand SelectionDAG::getExternalSymbol(const char *Sym, MVT::ValueType VT) {
|
|
|
|
SDNode *&N = ExternalSymbols[Sym];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new ExternalSymbolSDNode(Sym, VT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
2005-01-18 10:52:03 +08:00
|
|
|
SDOperand SelectionDAG::getSetCC(ISD::CondCode Cond, MVT::ValueType VT,
|
|
|
|
SDOperand N1, SDOperand N2) {
|
2005-01-07 15:46:32 +08:00
|
|
|
// These setcc operations always fold.
|
|
|
|
switch (Cond) {
|
|
|
|
default: break;
|
|
|
|
case ISD::SETFALSE:
|
2005-01-18 10:52:03 +08:00
|
|
|
case ISD::SETFALSE2: return getConstant(0, VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
case ISD::SETTRUE:
|
2005-01-18 10:52:03 +08:00
|
|
|
case ISD::SETTRUE2: return getConstant(1, VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
This patch does two things. First, it canonicalizes 'X >= C' -> 'X > C-1'
(likewise for <= >=u >=u).
Second, it implements a special case hack to turn 'X gtu SINTMAX' -> 'X lt 0'
On powerpc, for example, this changes this:
lis r2, 32767
ori r2, r2, 65535
cmplw cr0, r3, r2
bgt .LBB_test_2
into:
cmpwi cr0, r3, 0
blt .LBB_test_2
llvm-svn: 21142
2005-04-08 02:14:58 +08:00
|
|
|
if (ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.Val)) {
|
|
|
|
uint64_t C2 = N2C->getValue();
|
|
|
|
if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val)) {
|
|
|
|
uint64_t C1 = N1C->getValue();
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
// Sign extend the operands if required
|
|
|
|
if (ISD::isSignedIntSetCC(Cond)) {
|
|
|
|
C1 = N1C->getSignExtended();
|
|
|
|
C2 = N2C->getSignExtended();
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (Cond) {
|
|
|
|
default: assert(0 && "Unknown integer setcc!");
|
2005-01-18 10:52:03 +08:00
|
|
|
case ISD::SETEQ: return getConstant(C1 == C2, VT);
|
|
|
|
case ISD::SETNE: return getConstant(C1 != C2, VT);
|
|
|
|
case ISD::SETULT: return getConstant(C1 < C2, VT);
|
|
|
|
case ISD::SETUGT: return getConstant(C1 > C2, VT);
|
|
|
|
case ISD::SETULE: return getConstant(C1 <= C2, VT);
|
|
|
|
case ISD::SETUGE: return getConstant(C1 >= C2, VT);
|
|
|
|
case ISD::SETLT: return getConstant((int64_t)C1 < (int64_t)C2, VT);
|
|
|
|
case ISD::SETGT: return getConstant((int64_t)C1 > (int64_t)C2, VT);
|
|
|
|
case ISD::SETLE: return getConstant((int64_t)C1 <= (int64_t)C2, VT);
|
|
|
|
case ISD::SETGE: return getConstant((int64_t)C1 >= (int64_t)C2, VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
2005-04-08 02:58:54 +08:00
|
|
|
} else {
|
2005-04-18 12:30:45 +08:00
|
|
|
// If the LHS is a ZERO_EXTEND and if this is an ==/!= comparison, perform
|
|
|
|
// the comparison on the input.
|
|
|
|
if (N1.getOpcode() == ISD::ZERO_EXTEND) {
|
|
|
|
unsigned InSize = MVT::getSizeInBits(N1.getOperand(0).getValueType());
|
|
|
|
|
|
|
|
// If the comparison constant has bits in the upper part, the
|
|
|
|
// zero-extended value could never match.
|
|
|
|
if (C2 & (~0ULL << InSize)) {
|
|
|
|
unsigned VSize = MVT::getSizeInBits(N1.getValueType());
|
|
|
|
switch (Cond) {
|
|
|
|
case ISD::SETUGT:
|
|
|
|
case ISD::SETUGE:
|
|
|
|
case ISD::SETEQ: return getConstant(0, VT);
|
|
|
|
case ISD::SETULT:
|
|
|
|
case ISD::SETULE:
|
|
|
|
case ISD::SETNE: return getConstant(1, VT);
|
|
|
|
case ISD::SETGT:
|
|
|
|
case ISD::SETGE:
|
|
|
|
// True if the sign bit of C2 is set.
|
|
|
|
return getConstant((C2 & (1ULL << VSize)) != 0, VT);
|
|
|
|
case ISD::SETLT:
|
|
|
|
case ISD::SETLE:
|
|
|
|
// True if the sign bit of C2 isn't set.
|
|
|
|
return getConstant((C2 & (1ULL << VSize)) == 0, VT);
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Otherwise, we can perform the comparison with the low bits.
|
|
|
|
switch (Cond) {
|
|
|
|
case ISD::SETEQ:
|
|
|
|
case ISD::SETNE:
|
|
|
|
case ISD::SETUGT:
|
|
|
|
case ISD::SETUGE:
|
|
|
|
case ISD::SETULT:
|
|
|
|
case ISD::SETULE:
|
|
|
|
return getSetCC(Cond, VT, N1.getOperand(0),
|
|
|
|
getConstant(C2, N1.getOperand(0).getValueType()));
|
|
|
|
default:
|
|
|
|
break; // todo, be more careful with signed comparisons
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
This patch does two things. First, it canonicalizes 'X >= C' -> 'X > C-1'
(likewise for <= >=u >=u).
Second, it implements a special case hack to turn 'X gtu SINTMAX' -> 'X lt 0'
On powerpc, for example, this changes this:
lis r2, 32767
ori r2, r2, 65535
cmplw cr0, r3, r2
bgt .LBB_test_2
into:
cmpwi cr0, r3, 0
blt .LBB_test_2
llvm-svn: 21142
2005-04-08 02:14:58 +08:00
|
|
|
uint64_t MinVal, MaxVal;
|
|
|
|
unsigned OperandBitSize = MVT::getSizeInBits(N2C->getValueType(0));
|
|
|
|
if (ISD::isSignedIntSetCC(Cond)) {
|
|
|
|
MinVal = 1ULL << (OperandBitSize-1);
|
|
|
|
if (OperandBitSize != 1) // Avoid X >> 64, which is undefined.
|
|
|
|
MaxVal = ~0ULL >> (65-OperandBitSize);
|
|
|
|
else
|
|
|
|
MaxVal = 0;
|
|
|
|
} else {
|
|
|
|
MinVal = 0;
|
|
|
|
MaxVal = ~0ULL >> (64-OperandBitSize);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Canonicalize GE/LE comparisons to use GT/LT comparisons.
|
|
|
|
if (Cond == ISD::SETGE || Cond == ISD::SETUGE) {
|
|
|
|
if (C2 == MinVal) return getConstant(1, VT); // X >= MIN --> true
|
|
|
|
--C2; // X >= C1 --> X > (C1-1)
|
|
|
|
Cond = (Cond == ISD::SETGE) ? ISD::SETGT : ISD::SETUGT;
|
|
|
|
N2 = getConstant(C2, N2.getValueType());
|
|
|
|
N2C = cast<ConstantSDNode>(N2.Val);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (Cond == ISD::SETLE || Cond == ISD::SETULE) {
|
|
|
|
if (C2 == MaxVal) return getConstant(1, VT); // X <= MAX --> true
|
|
|
|
++C2; // X <= C1 --> X < (C1+1)
|
|
|
|
Cond = (Cond == ISD::SETLE) ? ISD::SETLT : ISD::SETULT;
|
|
|
|
N2 = getConstant(C2, N2.getValueType());
|
|
|
|
N2C = cast<ConstantSDNode>(N2.Val);
|
|
|
|
}
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-04-14 16:56:52 +08:00
|
|
|
if ((Cond == ISD::SETLT || Cond == ISD::SETULT) && C2 == MinVal)
|
|
|
|
return getConstant(0, VT); // X < MIN --> false
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-04-14 16:56:52 +08:00
|
|
|
// Canonicalize setgt X, Min --> setne X, Min
|
|
|
|
if ((Cond == ISD::SETGT || Cond == ISD::SETUGT) && C2 == MinVal)
|
|
|
|
return getSetCC(ISD::SETNE, VT, N1, N2);
|
2005-04-22 06:36:52 +08:00
|
|
|
|
canonicalize x <u 1 -> x == 0. On this testcase:
unsigned long long g;
unsigned long foo (unsigned long a) {
return (a >= g) ? 1 : 0;
}
It changes the ppc code from:
_foo:
.LBB_foo_0: ; entry
mflr r11
stw r11, 8(r1)
bl "L00000$pb"
"L00000$pb":
mflr r2
addis r2, r2, ha16(L_g$non_lazy_ptr-"L00000$pb")
lwz r2, lo16(L_g$non_lazy_ptr-"L00000$pb")(r2)
lwz r4, 0(r2)
lwz r2, 4(r2)
cmplw cr0, r3, r2
li r2, 1
li r3, 0
bge .LBB_foo_2 ; entry
.LBB_foo_1: ; entry
or r2, r3, r3
.LBB_foo_2: ; entry
cmplwi cr0, r4, 1
li r3, 1
li r5, 0
blt .LBB_foo_4 ; entry
.LBB_foo_3: ; entry
or r3, r5, r5
.LBB_foo_4: ; entry
cmpwi cr0, r4, 0
beq .LBB_foo_6 ; entry
.LBB_foo_5: ; entry
or r2, r3, r3
.LBB_foo_6: ; entry
rlwinm r3, r2, 0, 31, 31
lwz r11, 8(r1)
mtlr r11
blr
to:
_foo:
.LBB_foo_0: ; entry
mflr r11
stw r11, 8(r1)
bl "L00000$pb"
"L00000$pb":
mflr r2
addis r2, r2, ha16(L_g$non_lazy_ptr-"L00000$pb")
lwz r2, lo16(L_g$non_lazy_ptr-"L00000$pb")(r2)
lwz r4, 0(r2)
lwz r2, 4(r2)
cmplw cr0, r3, r2
li r2, 1
li r3, 0
bge .LBB_foo_2 ; entry
.LBB_foo_1: ; entry
or r2, r3, r3
.LBB_foo_2: ; entry
cntlzw r3, r4
srwi r3, r3, 5
cmpwi cr0, r4, 0
beq .LBB_foo_4 ; entry
.LBB_foo_3: ; entry
or r2, r3, r3
.LBB_foo_4: ; entry
rlwinm r3, r2, 0, 31, 31
lwz r11, 8(r1)
mtlr r11
blr
llvm-svn: 21241
2005-04-12 08:28:49 +08:00
|
|
|
// If we have setult X, 1, turn it into seteq X, 0
|
|
|
|
if ((Cond == ISD::SETLT || Cond == ISD::SETULT) && C2 == MinVal+1)
|
|
|
|
return getSetCC(ISD::SETEQ, VT, N1,
|
|
|
|
getConstant(MinVal, N1.getValueType()));
|
2005-04-14 16:56:52 +08:00
|
|
|
// If we have setugt X, Max-1, turn it into seteq X, Max
|
canonicalize x <u 1 -> x == 0. On this testcase:
unsigned long long g;
unsigned long foo (unsigned long a) {
return (a >= g) ? 1 : 0;
}
It changes the ppc code from:
_foo:
.LBB_foo_0: ; entry
mflr r11
stw r11, 8(r1)
bl "L00000$pb"
"L00000$pb":
mflr r2
addis r2, r2, ha16(L_g$non_lazy_ptr-"L00000$pb")
lwz r2, lo16(L_g$non_lazy_ptr-"L00000$pb")(r2)
lwz r4, 0(r2)
lwz r2, 4(r2)
cmplw cr0, r3, r2
li r2, 1
li r3, 0
bge .LBB_foo_2 ; entry
.LBB_foo_1: ; entry
or r2, r3, r3
.LBB_foo_2: ; entry
cmplwi cr0, r4, 1
li r3, 1
li r5, 0
blt .LBB_foo_4 ; entry
.LBB_foo_3: ; entry
or r3, r5, r5
.LBB_foo_4: ; entry
cmpwi cr0, r4, 0
beq .LBB_foo_6 ; entry
.LBB_foo_5: ; entry
or r2, r3, r3
.LBB_foo_6: ; entry
rlwinm r3, r2, 0, 31, 31
lwz r11, 8(r1)
mtlr r11
blr
to:
_foo:
.LBB_foo_0: ; entry
mflr r11
stw r11, 8(r1)
bl "L00000$pb"
"L00000$pb":
mflr r2
addis r2, r2, ha16(L_g$non_lazy_ptr-"L00000$pb")
lwz r2, lo16(L_g$non_lazy_ptr-"L00000$pb")(r2)
lwz r4, 0(r2)
lwz r2, 4(r2)
cmplw cr0, r3, r2
li r2, 1
li r3, 0
bge .LBB_foo_2 ; entry
.LBB_foo_1: ; entry
or r2, r3, r3
.LBB_foo_2: ; entry
cntlzw r3, r4
srwi r3, r3, 5
cmpwi cr0, r4, 0
beq .LBB_foo_4 ; entry
.LBB_foo_3: ; entry
or r2, r3, r3
.LBB_foo_4: ; entry
rlwinm r3, r2, 0, 31, 31
lwz r11, 8(r1)
mtlr r11
blr
llvm-svn: 21241
2005-04-12 08:28:49 +08:00
|
|
|
else if ((Cond == ISD::SETGT || Cond == ISD::SETUGT) && C2 == MaxVal-1)
|
|
|
|
return getSetCC(ISD::SETEQ, VT, N1,
|
|
|
|
getConstant(MaxVal, N1.getValueType()));
|
This patch does two things. First, it canonicalizes 'X >= C' -> 'X > C-1'
(likewise for <= >=u >=u).
Second, it implements a special case hack to turn 'X gtu SINTMAX' -> 'X lt 0'
On powerpc, for example, this changes this:
lis r2, 32767
ori r2, r2, 65535
cmplw cr0, r3, r2
bgt .LBB_test_2
into:
cmpwi cr0, r3, 0
blt .LBB_test_2
llvm-svn: 21142
2005-04-08 02:14:58 +08:00
|
|
|
|
|
|
|
// If we have "setcc X, C1", check to see if we can shrink the immediate
|
|
|
|
// by changing cc.
|
|
|
|
|
|
|
|
// SETUGT X, SINTMAX -> SETLT X, 0
|
|
|
|
if (Cond == ISD::SETUGT && OperandBitSize != 1 &&
|
|
|
|
C2 == (~0ULL >> (65-OperandBitSize)))
|
|
|
|
return getSetCC(ISD::SETLT, VT, N1, getConstant(0, N2.getValueType()));
|
|
|
|
|
|
|
|
// FIXME: Implement the rest of these.
|
|
|
|
|
Fold (x & 8) != 0 and (x & 8) == 8 into (x & 8) >> 3.
This turns this PPC code:
rlwinm r2, r3, 0, 28, 28
cmpwi cr7, r2, 8
mfcr r2
rlwinm r3, r2, 31, 31, 31
into this:
rlwinm r2, r3, 0, 28, 28
srwi r2, r2, 3
rlwinm r3, r2, 0, 31, 31
Next up, nuking the extra and.
llvm-svn: 21390
2005-04-21 14:12:41 +08:00
|
|
|
|
|
|
|
// Fold bit comparisons when we can.
|
|
|
|
if ((Cond == ISD::SETEQ || Cond == ISD::SETNE) &&
|
|
|
|
VT == N1.getValueType() && N1.getOpcode() == ISD::AND)
|
|
|
|
if (ConstantSDNode *AndRHS =
|
|
|
|
dyn_cast<ConstantSDNode>(N1.getOperand(1))) {
|
|
|
|
if (Cond == ISD::SETNE && C2 == 0) {// (X & 8) != 0 --> (X & 8) >> 3
|
|
|
|
// Perform the xform if the AND RHS is a single bit.
|
|
|
|
if ((AndRHS->getValue() & (AndRHS->getValue()-1)) == 0) {
|
|
|
|
return getNode(ISD::SRL, VT, N1,
|
|
|
|
getConstant(ExactLog2(AndRHS->getValue()),
|
|
|
|
TLI.getShiftAmountTy()));
|
|
|
|
}
|
|
|
|
} else if (Cond == ISD::SETEQ && C2 == AndRHS->getValue()) {
|
|
|
|
// (X & 8) == 8 --> (X & 8) >> 3
|
|
|
|
// Perform the xform if C2 is a single bit.
|
|
|
|
if ((C2 & (C2-1)) == 0) {
|
|
|
|
return getNode(ISD::SRL, VT, N1,
|
|
|
|
getConstant(ExactLog2(C2),TLI.getShiftAmountTy()));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
This patch does two things. First, it canonicalizes 'X >= C' -> 'X > C-1'
(likewise for <= >=u >=u).
Second, it implements a special case hack to turn 'X gtu SINTMAX' -> 'X lt 0'
On powerpc, for example, this changes this:
lis r2, 32767
ori r2, r2, 65535
cmplw cr0, r3, r2
bgt .LBB_test_2
into:
cmpwi cr0, r3, 0
blt .LBB_test_2
llvm-svn: 21142
2005-04-08 02:14:58 +08:00
|
|
|
} else if (isa<ConstantSDNode>(N1.Val)) {
|
|
|
|
// Ensure that the constant occurs on the RHS.
|
|
|
|
return getSetCC(ISD::getSetCCSwappedOperands(Cond), VT, N2, N1);
|
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
|
|
|
|
if (ConstantFPSDNode *N1C = dyn_cast<ConstantFPSDNode>(N1.Val))
|
|
|
|
if (ConstantFPSDNode *N2C = dyn_cast<ConstantFPSDNode>(N2.Val)) {
|
|
|
|
double C1 = N1C->getValue(), C2 = N2C->getValue();
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
switch (Cond) {
|
|
|
|
default: break; // FIXME: Implement the rest of these!
|
2005-01-18 10:52:03 +08:00
|
|
|
case ISD::SETEQ: return getConstant(C1 == C2, VT);
|
|
|
|
case ISD::SETNE: return getConstant(C1 != C2, VT);
|
|
|
|
case ISD::SETLT: return getConstant(C1 < C2, VT);
|
|
|
|
case ISD::SETGT: return getConstant(C1 > C2, VT);
|
|
|
|
case ISD::SETLE: return getConstant(C1 <= C2, VT);
|
|
|
|
case ISD::SETGE: return getConstant(C1 >= C2, VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// Ensure that the constant occurs on the RHS.
|
|
|
|
Cond = ISD::getSetCCSwappedOperands(Cond);
|
|
|
|
std::swap(N1, N2);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (N1 == N2) {
|
|
|
|
// We can always fold X == Y for integer setcc's.
|
|
|
|
if (MVT::isInteger(N1.getValueType()))
|
2005-01-18 10:52:03 +08:00
|
|
|
return getConstant(ISD::isTrueWhenEqual(Cond), VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
unsigned UOF = ISD::getUnorderedFlavor(Cond);
|
|
|
|
if (UOF == 2) // FP operators that are undefined on NaNs.
|
2005-01-18 10:52:03 +08:00
|
|
|
return getConstant(ISD::isTrueWhenEqual(Cond), VT);
|
2005-05-10 10:22:38 +08:00
|
|
|
if (UOF == unsigned(ISD::isTrueWhenEqual(Cond)))
|
2005-01-18 10:52:03 +08:00
|
|
|
return getConstant(UOF, VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
// Otherwise, we can't fold it. However, we can simplify it to SETUO/SETO
|
|
|
|
// if it is not already.
|
|
|
|
Cond = UOF == 0 ? ISD::SETUO : ISD::SETO;
|
|
|
|
}
|
|
|
|
|
Add some folds for == and != comparisons. This allows us to
codegen this loop in stepanov:
no_exit.i: ; preds = %entry, %no_exit.i, %then.i, %_Z5checkd.exit
%i.0.0 = phi int [ 0, %entry ], [ %i.0.0, %no_exit.i ], [ %inc.0, %_Z5checkd.exit ], [ %inc.012, %then.i ] ; <int> [#uses=3]
%indvar = phi uint [ %indvar.next, %no_exit.i ], [ 0, %entry ], [ 0, %then.i ], [ 0, %_Z5checkd.exit ] ; <uint> [#uses=3]
%result_addr.i.0 = phi double [ %tmp.4.i.i, %no_exit.i ], [ 0.000000e+00, %entry ], [ 0.000000e+00, %then.i ], [ 0.000000e+00, %_Z5checkd.exit ] ; <double> [#uses=1]
%first_addr.0.i.2.rec = cast uint %indvar to int ; <int> [#uses=1]
%first_addr.0.i.2 = getelementptr [2000 x double]* %data, int 0, uint %indvar ; <double*> [#uses=1]
%inc.i.rec = add int %first_addr.0.i.2.rec, 1 ; <int> [#uses=1]
%inc.i = getelementptr [2000 x double]* %data, int 0, int %inc.i.rec ; <double*> [#uses=1]
%tmp.3.i.i = load double* %first_addr.0.i.2 ; <double> [#uses=1]
%tmp.4.i.i = add double %result_addr.i.0, %tmp.3.i.i ; <double> [#uses=2]
%tmp.2.i = seteq double* %inc.i, getelementptr ([2000 x double]* %data, int 0, int 2000) ; <bool> [#uses=1]
%indvar.next = add uint %indvar, 1 ; <uint> [#uses=1]
br bool %tmp.2.i, label %_Z10accumulateIPddET0_T_S2_S1_.exit, label %no_exit.i
To this:
.LBB_Z4testIPddEvT_S1_T0__1: # no_exit.i
fldl data(,%eax,8)
fldl 16(%esp)
faddp %st(1)
fstpl 16(%esp)
incl %eax
movl %eax, %ecx
shll $3, %ecx
cmpl $16000, %ecx
#FP_REG_KILL
jne .LBB_Z4testIPddEvT_S1_T0__1 # no_exit.i
instead of this:
.LBB_Z4testIPddEvT_S1_T0__1: # no_exit.i
fldl data(,%eax,8)
fldl 16(%esp)
faddp %st(1)
fstpl 16(%esp)
incl %eax
leal data(,%eax,8), %ecx
leal data+16000, %edx
cmpl %edx, %ecx
#FP_REG_KILL
jne .LBB_Z4testIPddEvT_S1_T0__1 # no_exit.i
llvm-svn: 19425
2005-01-10 04:52:51 +08:00
|
|
|
if ((Cond == ISD::SETEQ || Cond == ISD::SETNE) &&
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
MVT::isInteger(N1.getValueType())) {
|
|
|
|
if (N1.getOpcode() == ISD::ADD || N1.getOpcode() == ISD::SUB ||
|
|
|
|
N1.getOpcode() == ISD::XOR) {
|
|
|
|
// Simplify (X+Y) == (X+Z) --> Y == Z
|
|
|
|
if (N1.getOpcode() == N2.getOpcode()) {
|
|
|
|
if (N1.getOperand(0) == N2.getOperand(0))
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N1.getOperand(1), N2.getOperand(1));
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
if (N1.getOperand(1) == N2.getOperand(1))
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N1.getOperand(0), N2.getOperand(0));
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
if (isCommutativeBinOp(N1.getOpcode())) {
|
|
|
|
// If X op Y == Y op X, try other combinations.
|
|
|
|
if (N1.getOperand(0) == N2.getOperand(1))
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N1.getOperand(1), N2.getOperand(0));
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
if (N1.getOperand(1) == N2.getOperand(0))
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N1.getOperand(1), N2.getOperand(1));
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
}
|
|
|
|
}
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
|
|
|
|
// FIXME: move this stuff to the DAG Combiner when it exists!
|
2005-04-22 06:36:52 +08:00
|
|
|
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
// Simplify (X+Z) == X --> Z == 0
|
|
|
|
if (N1.getOperand(0) == N2)
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N1.getOperand(1),
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
getConstant(0, N1.getValueType()));
|
|
|
|
if (N1.getOperand(1) == N2) {
|
|
|
|
if (isCommutativeBinOp(N1.getOpcode()))
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N1.getOperand(0),
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
getConstant(0, N1.getValueType()));
|
|
|
|
else {
|
|
|
|
assert(N1.getOpcode() == ISD::SUB && "Unexpected operation!");
|
|
|
|
// (Z-X) == X --> Z == X<<1
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N1.getOperand(0),
|
2005-04-22 06:36:52 +08:00
|
|
|
getNode(ISD::SHL, N2.getValueType(),
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
N2, getConstant(1, TLI.getShiftAmountTy())));
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
}
|
Add some folds for == and != comparisons. This allows us to
codegen this loop in stepanov:
no_exit.i: ; preds = %entry, %no_exit.i, %then.i, %_Z5checkd.exit
%i.0.0 = phi int [ 0, %entry ], [ %i.0.0, %no_exit.i ], [ %inc.0, %_Z5checkd.exit ], [ %inc.012, %then.i ] ; <int> [#uses=3]
%indvar = phi uint [ %indvar.next, %no_exit.i ], [ 0, %entry ], [ 0, %then.i ], [ 0, %_Z5checkd.exit ] ; <uint> [#uses=3]
%result_addr.i.0 = phi double [ %tmp.4.i.i, %no_exit.i ], [ 0.000000e+00, %entry ], [ 0.000000e+00, %then.i ], [ 0.000000e+00, %_Z5checkd.exit ] ; <double> [#uses=1]
%first_addr.0.i.2.rec = cast uint %indvar to int ; <int> [#uses=1]
%first_addr.0.i.2 = getelementptr [2000 x double]* %data, int 0, uint %indvar ; <double*> [#uses=1]
%inc.i.rec = add int %first_addr.0.i.2.rec, 1 ; <int> [#uses=1]
%inc.i = getelementptr [2000 x double]* %data, int 0, int %inc.i.rec ; <double*> [#uses=1]
%tmp.3.i.i = load double* %first_addr.0.i.2 ; <double> [#uses=1]
%tmp.4.i.i = add double %result_addr.i.0, %tmp.3.i.i ; <double> [#uses=2]
%tmp.2.i = seteq double* %inc.i, getelementptr ([2000 x double]* %data, int 0, int 2000) ; <bool> [#uses=1]
%indvar.next = add uint %indvar, 1 ; <uint> [#uses=1]
br bool %tmp.2.i, label %_Z10accumulateIPddET0_T_S2_S1_.exit, label %no_exit.i
To this:
.LBB_Z4testIPddEvT_S1_T0__1: # no_exit.i
fldl data(,%eax,8)
fldl 16(%esp)
faddp %st(1)
fstpl 16(%esp)
incl %eax
movl %eax, %ecx
shll $3, %ecx
cmpl $16000, %ecx
#FP_REG_KILL
jne .LBB_Z4testIPddEvT_S1_T0__1 # no_exit.i
instead of this:
.LBB_Z4testIPddEvT_S1_T0__1: # no_exit.i
fldl data(,%eax,8)
fldl 16(%esp)
faddp %st(1)
fstpl 16(%esp)
incl %eax
leal data(,%eax,8), %ecx
leal data+16000, %edx
cmpl %edx, %ecx
#FP_REG_KILL
jne .LBB_Z4testIPddEvT_S1_T0__1 # no_exit.i
llvm-svn: 19425
2005-01-10 04:52:51 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
if (N2.getOpcode() == ISD::ADD || N2.getOpcode() == ISD::SUB ||
|
|
|
|
N2.getOpcode() == ISD::XOR) {
|
|
|
|
// Simplify X == (X+Z) --> Z == 0
|
|
|
|
if (N2.getOperand(0) == N1)
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N2.getOperand(1),
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
getConstant(0, N2.getValueType()));
|
|
|
|
else if (N2.getOperand(1) == N1)
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Cond, VT, N2.getOperand(0),
|
Implement a couple of more simplifications. This lets us codegen:
int test2(int * P, int* Q, int A, int B) {
return P+A == P;
}
into:
test2:
movl 4(%esp), %eax
movl 12(%esp), %eax
shll $2, %eax
cmpl $0, %eax
sete %al
movzbl %al, %eax
ret
instead of:
test2:
movl 4(%esp), %eax
movl 12(%esp), %ecx
leal (%eax,%ecx,4), %ecx
cmpl %eax, %ecx
sete %al
movzbl %al, %eax
ret
ICC is producing worse code:
test2:
movl 4(%esp), %eax #8.5
movl 12(%esp), %edx #8.5
lea (%edx,%edx), %ecx #9.9
addl %ecx, %ecx #9.9
addl %eax, %ecx #9.9
cmpl %eax, %ecx #9.16
movl $0, %eax #9.16
sete %al #9.16
ret #9.16
as is GCC (looks like our old code):
test2:
movl 4(%esp), %edx
movl 12(%esp), %eax
leal (%edx,%eax,4), %ecx
cmpl %edx, %ecx
sete %al
movzbl %al, %eax
ret
llvm-svn: 19430
2005-01-10 10:03:02 +08:00
|
|
|
getConstant(0, N2.getValueType()));
|
|
|
|
}
|
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
|
2005-04-18 12:48:12 +08:00
|
|
|
// Fold away ALL boolean setcc's.
|
|
|
|
if (N1.getValueType() == MVT::i1) {
|
|
|
|
switch (Cond) {
|
|
|
|
default: assert(0 && "Unknown integer setcc!");
|
|
|
|
case ISD::SETEQ: // X == Y -> (X^Y)^1
|
|
|
|
N1 = getNode(ISD::XOR, MVT::i1,
|
|
|
|
getNode(ISD::XOR, MVT::i1, N1, N2),
|
|
|
|
getConstant(1, MVT::i1));
|
|
|
|
break;
|
|
|
|
case ISD::SETNE: // X != Y --> (X^Y)
|
|
|
|
N1 = getNode(ISD::XOR, MVT::i1, N1, N2);
|
|
|
|
break;
|
|
|
|
case ISD::SETGT: // X >s Y --> X == 0 & Y == 1 --> X^1 & Y
|
|
|
|
case ISD::SETULT: // X <u Y --> X == 0 & Y == 1 --> X^1 & Y
|
|
|
|
N1 = getNode(ISD::AND, MVT::i1, N2,
|
|
|
|
getNode(ISD::XOR, MVT::i1, N1, getConstant(1, MVT::i1)));
|
|
|
|
break;
|
|
|
|
case ISD::SETLT: // X <s Y --> X == 1 & Y == 0 --> Y^1 & X
|
|
|
|
case ISD::SETUGT: // X >u Y --> X == 1 & Y == 0 --> Y^1 & X
|
|
|
|
N1 = getNode(ISD::AND, MVT::i1, N1,
|
|
|
|
getNode(ISD::XOR, MVT::i1, N2, getConstant(1, MVT::i1)));
|
|
|
|
break;
|
|
|
|
case ISD::SETULE: // X <=u Y --> X == 0 | Y == 1 --> X^1 | Y
|
|
|
|
case ISD::SETGE: // X >=s Y --> X == 0 | Y == 1 --> X^1 | Y
|
|
|
|
N1 = getNode(ISD::OR, MVT::i1, N2,
|
|
|
|
getNode(ISD::XOR, MVT::i1, N1, getConstant(1, MVT::i1)));
|
|
|
|
break;
|
|
|
|
case ISD::SETUGE: // X >=u Y --> X == 1 | Y == 0 --> Y^1 | X
|
|
|
|
case ISD::SETLE: // X <=s Y --> X == 1 | Y == 0 --> Y^1 | X
|
|
|
|
N1 = getNode(ISD::OR, MVT::i1, N1,
|
|
|
|
getNode(ISD::XOR, MVT::i1, N2, getConstant(1, MVT::i1)));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (VT != MVT::i1)
|
|
|
|
N1 = getNode(ISD::ZERO_EXTEND, VT, N1);
|
|
|
|
return N1;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-01-19 03:26:36 +08:00
|
|
|
SetCCSDNode *&N = SetCCs[std::make_pair(std::make_pair(N1, N2),
|
|
|
|
std::make_pair(Cond, VT))];
|
2005-01-07 15:46:32 +08:00
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new SetCCSDNode(Cond, N1, N2);
|
2005-01-18 10:52:03 +08:00
|
|
|
N->setValueTypes(VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/// getNode - Gets or creates the specified node.
|
2003-08-11 22:57:33 +08:00
|
|
|
///
|
2005-01-07 15:46:32 +08:00
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT) {
|
|
|
|
SDNode *N = new SDNode(Opcode, VT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT,
|
|
|
|
SDOperand Operand) {
|
|
|
|
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Operand.Val)) {
|
|
|
|
uint64_t Val = C->getValue();
|
|
|
|
switch (Opcode) {
|
|
|
|
default: break;
|
|
|
|
case ISD::SIGN_EXTEND: return getConstant(C->getSignExtended(), VT);
|
|
|
|
case ISD::ZERO_EXTEND: return getConstant(Val, VT);
|
|
|
|
case ISD::TRUNCATE: return getConstant(Val, VT);
|
2005-01-08 16:08:56 +08:00
|
|
|
case ISD::SINT_TO_FP: return getConstantFP(C->getSignExtended(), VT);
|
|
|
|
case ISD::UINT_TO_FP: return getConstantFP(C->getValue(), VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Operand.Val))
|
|
|
|
switch (Opcode) {
|
2005-04-09 11:02:46 +08:00
|
|
|
case ISD::FNEG:
|
|
|
|
return getConstantFP(-C->getValue(), VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
case ISD::FP_ROUND:
|
|
|
|
case ISD::FP_EXTEND:
|
|
|
|
return getConstantFP(C->getValue(), VT);
|
2005-01-08 16:08:56 +08:00
|
|
|
case ISD::FP_TO_SINT:
|
|
|
|
return getConstant((int64_t)C->getValue(), VT);
|
|
|
|
case ISD::FP_TO_UINT:
|
|
|
|
return getConstant((uint64_t)C->getValue(), VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
unsigned OpOpcode = Operand.Val->getOpcode();
|
|
|
|
switch (Opcode) {
|
2005-01-22 02:01:22 +08:00
|
|
|
case ISD::TokenFactor:
|
|
|
|
return Operand; // Factor of one node? No factor.
|
2005-01-07 15:46:32 +08:00
|
|
|
case ISD::SIGN_EXTEND:
|
|
|
|
if (Operand.getValueType() == VT) return Operand; // noop extension
|
|
|
|
if (OpOpcode == ISD::SIGN_EXTEND || OpOpcode == ISD::ZERO_EXTEND)
|
|
|
|
return getNode(OpOpcode, VT, Operand.Val->getOperand(0));
|
|
|
|
break;
|
|
|
|
case ISD::ZERO_EXTEND:
|
|
|
|
if (Operand.getValueType() == VT) return Operand; // noop extension
|
If a target zero or sign extends the result of its setcc, allow folding of
this into sign/zero extension instructions later.
On PPC, for example, this testcase:
%G = external global sbyte
implementation
void %test(int %X, int %Y) {
%C = setlt int %X, %Y
%D = cast bool %C to sbyte
store sbyte %D, sbyte* %G
ret void
}
Now codegens to:
cmpw cr0, r3, r4
li r3, 1
li r4, 0
blt .LBB_test_2 ;
.LBB_test_1: ;
or r3, r4, r4
.LBB_test_2: ;
addis r2, r2, ha16(L_G$non_lazy_ptr-"L00000$pb")
lwz r2, lo16(L_G$non_lazy_ptr-"L00000$pb")(r2)
stb r3, 0(r2)
instead of:
cmpw cr0, r3, r4
li r3, 1
li r4, 0
blt .LBB_test_2 ;
.LBB_test_1: ;
or r3, r4, r4
.LBB_test_2: ;
*** rlwinm r3, r3, 0, 31, 31
addis r2, r2, ha16(L_G$non_lazy_ptr-"L00000$pb")
lwz r2, lo16(L_G$non_lazy_ptr-"L00000$pb")(r2)
stb r3, 0(r2)
llvm-svn: 21148
2005-04-08 03:43:53 +08:00
|
|
|
if (OpOpcode == ISD::ZERO_EXTEND) // (zext (zext x)) -> (zext x)
|
2005-01-13 02:51:15 +08:00
|
|
|
return getNode(ISD::ZERO_EXTEND, VT, Operand.Val->getOperand(0));
|
2005-01-07 15:46:32 +08:00
|
|
|
break;
|
|
|
|
case ISD::TRUNCATE:
|
|
|
|
if (Operand.getValueType() == VT) return Operand; // noop truncate
|
|
|
|
if (OpOpcode == ISD::TRUNCATE)
|
|
|
|
return getNode(ISD::TRUNCATE, VT, Operand.Val->getOperand(0));
|
2005-01-08 05:56:24 +08:00
|
|
|
else if (OpOpcode == ISD::ZERO_EXTEND || OpOpcode == ISD::SIGN_EXTEND) {
|
|
|
|
// If the source is smaller than the dest, we still need an extend.
|
|
|
|
if (Operand.Val->getOperand(0).getValueType() < VT)
|
|
|
|
return getNode(OpOpcode, VT, Operand.Val->getOperand(0));
|
|
|
|
else if (Operand.Val->getOperand(0).getValueType() > VT)
|
|
|
|
return getNode(ISD::TRUNCATE, VT, Operand.Val->getOperand(0));
|
|
|
|
else
|
|
|
|
return Operand.Val->getOperand(0);
|
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
break;
|
2005-04-09 11:02:46 +08:00
|
|
|
case ISD::FNEG:
|
|
|
|
if (OpOpcode == ISD::SUB) // -(X-Y) -> (Y-X)
|
|
|
|
return getNode(ISD::SUB, VT, Operand.Val->getOperand(1),
|
|
|
|
Operand.Val->getOperand(0));
|
|
|
|
if (OpOpcode == ISD::FNEG) // --X -> X
|
|
|
|
return Operand.Val->getOperand(0);
|
|
|
|
break;
|
|
|
|
case ISD::FABS:
|
|
|
|
if (OpOpcode == ISD::FNEG) // abs(-X) -> abs(X)
|
|
|
|
return getNode(ISD::FABS, VT, Operand.Val->getOperand(0));
|
|
|
|
break;
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
SDNode *&N = UnaryOps[std::make_pair(Opcode, std::make_pair(Operand, VT))];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new SDNode(Opcode, Operand);
|
|
|
|
N->setValueTypes(VT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
Make the AND elimination operation recursive and significantly more powerful,
eliminating an and for Nate's testcase:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
generating:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r2, r2, r3
rlwinm r3, r2, 0, 31, 31
blr
llvm-svn: 21315
2005-04-18 11:48:41 +08:00
|
|
|
/// MaskedValueIsZero - Return true if 'V & Mask' is known to be zero. We use
|
|
|
|
/// this predicate to simplify operations downstream. V and Mask are known to
|
|
|
|
/// be the same type.
|
|
|
|
static bool MaskedValueIsZero(const SDOperand &Op, uint64_t Mask,
|
|
|
|
const TargetLowering &TLI) {
|
|
|
|
unsigned SrcBits;
|
|
|
|
if (Mask == 0) return true;
|
2005-04-22 06:36:52 +08:00
|
|
|
|
Make the AND elimination operation recursive and significantly more powerful,
eliminating an and for Nate's testcase:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
generating:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r2, r2, r3
rlwinm r3, r2, 0, 31, 31
blr
llvm-svn: 21315
2005-04-18 11:48:41 +08:00
|
|
|
// If we know the result of a setcc has the top bits zero, use this info.
|
|
|
|
switch (Op.getOpcode()) {
|
|
|
|
case ISD::UNDEF:
|
|
|
|
return true;
|
|
|
|
case ISD::Constant:
|
|
|
|
return (cast<ConstantSDNode>(Op)->getValue() & Mask) == 0;
|
|
|
|
|
|
|
|
case ISD::SETCC:
|
2005-04-22 06:36:52 +08:00
|
|
|
return ((Mask & 1) == 0) &&
|
Make the AND elimination operation recursive and significantly more powerful,
eliminating an and for Nate's testcase:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
generating:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r2, r2, r3
rlwinm r3, r2, 0, 31, 31
blr
llvm-svn: 21315
2005-04-18 11:48:41 +08:00
|
|
|
TLI.getSetCCResultContents() == TargetLowering::ZeroOrOneSetCCResult;
|
|
|
|
|
|
|
|
case ISD::ZEXTLOAD:
|
|
|
|
SrcBits = MVT::getSizeInBits(cast<MVTSDNode>(Op)->getExtraValueType());
|
|
|
|
return (Mask & ((1ULL << SrcBits)-1)) == 0; // Returning only the zext bits.
|
|
|
|
case ISD::ZERO_EXTEND:
|
|
|
|
SrcBits = MVT::getSizeInBits(Op.getOperand(0).getValueType());
|
|
|
|
return MaskedValueIsZero(Op.getOperand(0),Mask & ((1ULL << SrcBits)-1),TLI);
|
|
|
|
|
|
|
|
case ISD::AND:
|
|
|
|
// (X & C1) & C2 == 0 iff C1 & C2 == 0.
|
Improve and elimination. On PPC, for:
bool %test(int %X) {
%Y = and int %X, 8
%Z = setne int %Y, 0
ret bool %Z
}
we now generate this:
rlwinm r2, r3, 0, 28, 28
srwi r3, r2, 3
instead of this:
rlwinm r2, r3, 0, 28, 28
srwi r2, r2, 3
rlwinm r3, r2, 0, 31, 31
I'll leave it to Nate to get it down to one instruction. :)
---------------------------------------------------------------------
llvm-svn: 21391
2005-04-21 14:28:15 +08:00
|
|
|
if (ConstantSDNode *AndRHS = dyn_cast<ConstantSDNode>(Op.getOperand(1)))
|
Make the AND elimination operation recursive and significantly more powerful,
eliminating an and for Nate's testcase:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
generating:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r2, r2, r3
rlwinm r3, r2, 0, 31, 31
blr
llvm-svn: 21315
2005-04-18 11:48:41 +08:00
|
|
|
return MaskedValueIsZero(Op.getOperand(0),AndRHS->getValue() & Mask, TLI);
|
|
|
|
|
|
|
|
// FALL THROUGH
|
|
|
|
case ISD::OR:
|
|
|
|
case ISD::XOR:
|
|
|
|
return MaskedValueIsZero(Op.getOperand(0), Mask, TLI) &&
|
|
|
|
MaskedValueIsZero(Op.getOperand(1), Mask, TLI);
|
|
|
|
case ISD::SELECT:
|
|
|
|
return MaskedValueIsZero(Op.getOperand(1), Mask, TLI) &&
|
|
|
|
MaskedValueIsZero(Op.getOperand(2), Mask, TLI);
|
Improve and elimination. On PPC, for:
bool %test(int %X) {
%Y = and int %X, 8
%Z = setne int %Y, 0
ret bool %Z
}
we now generate this:
rlwinm r2, r3, 0, 28, 28
srwi r3, r2, 3
instead of this:
rlwinm r2, r3, 0, 28, 28
srwi r2, r2, 3
rlwinm r3, r2, 0, 31, 31
I'll leave it to Nate to get it down to one instruction. :)
---------------------------------------------------------------------
llvm-svn: 21391
2005-04-21 14:28:15 +08:00
|
|
|
|
|
|
|
case ISD::SRL:
|
|
|
|
// (ushr X, C1) & C2 == 0 iff X & (C2 << C1) == 0
|
|
|
|
if (ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
|
|
|
|
uint64_t NewVal = Mask << ShAmt->getValue();
|
|
|
|
SrcBits = MVT::getSizeInBits(Op.getValueType());
|
|
|
|
if (SrcBits != 64) NewVal &= (1ULL << SrcBits)-1;
|
|
|
|
return MaskedValueIsZero(Op.getOperand(0), NewVal, TLI);
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
case ISD::SHL:
|
|
|
|
// (ushl X, C1) & C2 == 0 iff X & (C2 >> C1) == 0
|
|
|
|
if (ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
|
|
|
|
uint64_t NewVal = Mask >> ShAmt->getValue();
|
|
|
|
return MaskedValueIsZero(Op.getOperand(0), NewVal, TLI);
|
|
|
|
}
|
|
|
|
return false;
|
Codegen x < 0 | y < 0 as (x|y) < 0. This allows us to compile this to:
_foo:
or r2, r4, r3
srwi r3, r2, 31
blr
instead of:
_foo:
srwi r2, r4, 31
srwi r3, r3, 31
or r3, r2, r3
blr
llvm-svn: 21544
2005-04-26 05:03:25 +08:00
|
|
|
// TODO we could handle some SRA cases here.
|
Make the AND elimination operation recursive and significantly more powerful,
eliminating an and for Nate's testcase:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
generating:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r2, r2, r3
rlwinm r3, r2, 0, 31, 31
blr
llvm-svn: 21315
2005-04-18 11:48:41 +08:00
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT,
|
|
|
|
SDOperand N1, SDOperand N2) {
|
2005-01-16 10:23:22 +08:00
|
|
|
#ifndef NDEBUG
|
|
|
|
switch (Opcode) {
|
2005-01-20 02:01:40 +08:00
|
|
|
case ISD::TokenFactor:
|
|
|
|
assert(VT == MVT::Other && N1.getValueType() == MVT::Other &&
|
|
|
|
N2.getValueType() == MVT::Other && "Invalid token factor!");
|
|
|
|
break;
|
2005-01-16 10:23:22 +08:00
|
|
|
case ISD::AND:
|
|
|
|
case ISD::OR:
|
|
|
|
case ISD::XOR:
|
|
|
|
case ISD::UDIV:
|
|
|
|
case ISD::UREM:
|
|
|
|
assert(MVT::isInteger(VT) && "This operator does not apply to FP types!");
|
|
|
|
// fall through
|
|
|
|
case ISD::ADD:
|
|
|
|
case ISD::SUB:
|
|
|
|
case ISD::MUL:
|
|
|
|
case ISD::SDIV:
|
|
|
|
case ISD::SREM:
|
|
|
|
assert(N1.getValueType() == N2.getValueType() &&
|
|
|
|
N1.getValueType() == VT && "Binary operator types must match!");
|
|
|
|
break;
|
|
|
|
|
|
|
|
case ISD::SHL:
|
|
|
|
case ISD::SRA:
|
|
|
|
case ISD::SRL:
|
|
|
|
assert(VT == N1.getValueType() &&
|
|
|
|
"Shift operators return type must be the same as their first arg");
|
|
|
|
assert(MVT::isInteger(VT) && MVT::isInteger(N2.getValueType()) &&
|
2005-01-18 01:15:02 +08:00
|
|
|
VT != MVT::i1 && "Shifts only work on integers");
|
2005-01-16 10:23:22 +08:00
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val);
|
|
|
|
ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.Val);
|
|
|
|
if (N1C) {
|
|
|
|
if (N2C) {
|
|
|
|
uint64_t C1 = N1C->getValue(), C2 = N2C->getValue();
|
|
|
|
switch (Opcode) {
|
|
|
|
case ISD::ADD: return getConstant(C1 + C2, VT);
|
|
|
|
case ISD::SUB: return getConstant(C1 - C2, VT);
|
|
|
|
case ISD::MUL: return getConstant(C1 * C2, VT);
|
|
|
|
case ISD::UDIV:
|
|
|
|
if (C2) return getConstant(C1 / C2, VT);
|
|
|
|
break;
|
|
|
|
case ISD::UREM :
|
|
|
|
if (C2) return getConstant(C1 % C2, VT);
|
|
|
|
break;
|
|
|
|
case ISD::SDIV :
|
|
|
|
if (C2) return getConstant(N1C->getSignExtended() /
|
|
|
|
N2C->getSignExtended(), VT);
|
|
|
|
break;
|
|
|
|
case ISD::SREM :
|
|
|
|
if (C2) return getConstant(N1C->getSignExtended() %
|
|
|
|
N2C->getSignExtended(), VT);
|
|
|
|
break;
|
|
|
|
case ISD::AND : return getConstant(C1 & C2, VT);
|
|
|
|
case ISD::OR : return getConstant(C1 | C2, VT);
|
|
|
|
case ISD::XOR : return getConstant(C1 ^ C2, VT);
|
2005-01-10 08:07:15 +08:00
|
|
|
case ISD::SHL : return getConstant(C1 << (int)C2, VT);
|
|
|
|
case ISD::SRL : return getConstant(C1 >> (unsigned)C2, VT);
|
|
|
|
case ISD::SRA : return getConstant(N1C->getSignExtended() >>(int)C2, VT);
|
2005-01-07 15:46:32 +08:00
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
|
|
|
|
} else { // Cannonicalize constant to RHS if commutative
|
|
|
|
if (isCommutativeBinOp(Opcode)) {
|
|
|
|
std::swap(N1C, N2C);
|
|
|
|
std::swap(N1, N2);
|
|
|
|
}
|
|
|
|
}
|
2005-01-20 01:29:49 +08:00
|
|
|
|
|
|
|
switch (Opcode) {
|
|
|
|
default: break;
|
|
|
|
case ISD::SHL: // shl 0, X -> 0
|
|
|
|
if (N1C->isNullValue()) return N1;
|
|
|
|
break;
|
|
|
|
case ISD::SRL: // srl 0, X -> 0
|
|
|
|
if (N1C->isNullValue()) return N1;
|
|
|
|
break;
|
|
|
|
case ISD::SRA: // sra -1, X -> -1
|
|
|
|
if (N1C->isAllOnesValue()) return N1;
|
|
|
|
break;
|
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
if (N2C) {
|
|
|
|
uint64_t C2 = N2C->getValue();
|
|
|
|
|
|
|
|
switch (Opcode) {
|
|
|
|
case ISD::ADD:
|
|
|
|
if (!C2) return N1; // add X, 0 -> X
|
|
|
|
break;
|
|
|
|
case ISD::SUB:
|
|
|
|
if (!C2) return N1; // sub X, 0 -> X
|
2005-05-12 08:17:04 +08:00
|
|
|
return getNode(ISD::ADD, VT, N1, getConstant(-C2, VT));
|
2005-01-07 15:46:32 +08:00
|
|
|
case ISD::MUL:
|
|
|
|
if (!C2) return N2; // mul X, 0 -> 0
|
|
|
|
if (N2C->isAllOnesValue()) // mul X, -1 -> 0-X
|
|
|
|
return getNode(ISD::SUB, VT, getConstant(0, VT), N1);
|
|
|
|
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
// FIXME: Move this to the DAG combiner when it exists.
|
2005-01-07 15:46:32 +08:00
|
|
|
if ((C2 & C2-1) == 0) {
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
SDOperand ShAmt = getConstant(ExactLog2(C2), TLI.getShiftAmountTy());
|
|
|
|
return getNode(ISD::SHL, VT, N1, ShAmt);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
|
|
|
|
case ISD::UDIV:
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
// FIXME: Move this to the DAG combiner when it exists.
|
2005-01-07 15:46:32 +08:00
|
|
|
if ((C2 & C2-1) == 0 && C2) {
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
SDOperand ShAmt = getConstant(ExactLog2(C2), TLI.getShiftAmountTy());
|
|
|
|
return getNode(ISD::SRL, VT, N1, ShAmt);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
|
2005-01-11 12:25:13 +08:00
|
|
|
case ISD::SHL:
|
|
|
|
case ISD::SRL:
|
2005-04-13 07:12:17 +08:00
|
|
|
case ISD::SRA:
|
|
|
|
// If the shift amount is bigger than the size of the data, then all the
|
Make the AND elimination operation recursive and significantly more powerful,
eliminating an and for Nate's testcase:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
generating:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r2, r2, r3
rlwinm r3, r2, 0, 31, 31
blr
llvm-svn: 21315
2005-04-18 11:48:41 +08:00
|
|
|
// bits are shifted out. Simplify to undef.
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
if (C2 >= MVT::getSizeInBits(N1.getValueType())) {
|
2005-04-13 07:12:17 +08:00
|
|
|
return getNode(ISD::UNDEF, N1.getValueType());
|
Remove the 3 HACK HACK HACKs I put in before, fixing them properly with
the new TLI that is available.
Implement support for handling out of range shifts. This allows us to
compile this code (a 64-bit rotate):
unsigned long long f3(unsigned long long x) {
return (x << 32) | (x >> (64-32));
}
into this:
f3:
mov %EDX, DWORD PTR [%ESP + 4]
mov %EAX, DWORD PTR [%ESP + 8]
ret
GCC produces this:
$ gcc t.c -masm=intel -O3 -S -o - -fomit-frame-pointer
..
f3:
push %ebx
mov %ebx, DWORD PTR [%esp+12]
mov %ecx, DWORD PTR [%esp+8]
mov %eax, %ebx
mov %edx, %ecx
pop %ebx
ret
The Simple ISEL produces (eww gross):
f3:
sub %ESP, 4
mov DWORD PTR [%ESP], %ESI
mov %EDX, DWORD PTR [%ESP + 8]
mov %ECX, DWORD PTR [%ESP + 12]
mov %EAX, 0
mov %ESI, 0
or %EAX, %ECX
or %EDX, %ESI
mov %ESI, DWORD PTR [%ESP]
add %ESP, 4
ret
llvm-svn: 19780
2005-01-23 12:39:44 +08:00
|
|
|
}
|
2005-01-11 12:25:13 +08:00
|
|
|
if (C2 == 0) return N1;
|
2005-05-10 01:06:45 +08:00
|
|
|
|
|
|
|
if (Opcode == ISD::SHL && N1.getNumOperands() == 2)
|
|
|
|
if (ConstantSDNode *OpSA = dyn_cast<ConstantSDNode>(N1.getOperand(1))) {
|
|
|
|
unsigned OpSAC = OpSA->getValue();
|
|
|
|
if (N1.getOpcode() == ISD::SHL) {
|
|
|
|
if (C2+OpSAC >= MVT::getSizeInBits(N1.getValueType()))
|
|
|
|
return getConstant(0, N1.getValueType());
|
|
|
|
return getNode(ISD::SHL, N1.getValueType(), N1.getOperand(0),
|
|
|
|
getConstant(C2+OpSAC, N2.getValueType()));
|
|
|
|
} else if (N1.getOpcode() == ISD::SRL) {
|
|
|
|
// (X >> C1) << C2: if C2 > C1, ((X & ~0<<C1) << C2-C1)
|
|
|
|
SDOperand Mask = getNode(ISD::AND, VT, N1.getOperand(0),
|
|
|
|
getConstant(~0ULL << OpSAC, VT));
|
|
|
|
if (C2 > OpSAC) {
|
|
|
|
return getNode(ISD::SHL, VT, Mask,
|
|
|
|
getConstant(C2-OpSAC, N2.getValueType()));
|
|
|
|
} else {
|
|
|
|
// (X >> C1) << C2: if C2 <= C1, ((X & ~0<<C1) >> C1-C2)
|
|
|
|
return getNode(ISD::SRL, VT, Mask,
|
|
|
|
getConstant(OpSAC-C2, N2.getValueType()));
|
|
|
|
}
|
|
|
|
} else if (N1.getOpcode() == ISD::SRA) {
|
|
|
|
// if C1 == C2, just mask out low bits.
|
|
|
|
if (C2 == OpSAC)
|
|
|
|
return getNode(ISD::AND, VT, N1.getOperand(0),
|
|
|
|
getConstant(~0ULL << C2, VT));
|
|
|
|
}
|
|
|
|
}
|
2005-01-11 12:25:13 +08:00
|
|
|
break;
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
case ISD::AND:
|
|
|
|
if (!C2) return N2; // X and 0 -> 0
|
|
|
|
if (N2C->isAllOnesValue())
|
2005-04-14 05:23:31 +08:00
|
|
|
return N1; // X and -1 -> X
|
2005-04-10 05:43:54 +08:00
|
|
|
|
Make the AND elimination operation recursive and significantly more powerful,
eliminating an and for Nate's testcase:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
generating:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r2, r2, r3
rlwinm r3, r2, 0, 31, 31
blr
llvm-svn: 21315
2005-04-18 11:48:41 +08:00
|
|
|
if (MaskedValueIsZero(N1, C2, TLI)) // X and 0 -> 0
|
|
|
|
return getConstant(0, VT);
|
|
|
|
|
Improve and elimination. On PPC, for:
bool %test(int %X) {
%Y = and int %X, 8
%Z = setne int %Y, 0
ret bool %Z
}
we now generate this:
rlwinm r2, r3, 0, 28, 28
srwi r3, r2, 3
instead of this:
rlwinm r2, r3, 0, 28, 28
srwi r2, r2, 3
rlwinm r3, r2, 0, 31, 31
I'll leave it to Nate to get it down to one instruction. :)
---------------------------------------------------------------------
llvm-svn: 21391
2005-04-21 14:28:15 +08:00
|
|
|
{
|
|
|
|
uint64_t NotC2 = ~C2;
|
|
|
|
if (VT != MVT::i64)
|
|
|
|
NotC2 &= (1ULL << MVT::getSizeInBits(VT))-1;
|
|
|
|
|
|
|
|
if (MaskedValueIsZero(N1, NotC2, TLI))
|
|
|
|
return N1; // if (X & ~C2) -> 0, the and is redundant
|
|
|
|
}
|
Make the AND elimination operation recursive and significantly more powerful,
eliminating an and for Nate's testcase:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
generating:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r2, r2, r3
rlwinm r3, r2, 0, 31, 31
blr
llvm-svn: 21315
2005-04-18 11:48:41 +08:00
|
|
|
|
2005-04-10 09:13:15 +08:00
|
|
|
// FIXME: Should add a corresponding version of this for
|
|
|
|
// ZERO_EXTEND/SIGN_EXTEND by converting them to an ANY_EXTEND node which
|
|
|
|
// we don't have yet.
|
|
|
|
|
2005-04-13 10:38:18 +08:00
|
|
|
// and (sign_extend_inreg x:16:32), 1 -> and x, 1
|
|
|
|
if (N1.getOpcode() == ISD::SIGN_EXTEND_INREG) {
|
2005-04-10 05:43:54 +08:00
|
|
|
// If we are masking out the part of our input that was extended, just
|
|
|
|
// mask the input to the extension directly.
|
|
|
|
unsigned ExtendBits =
|
|
|
|
MVT::getSizeInBits(cast<MVTSDNode>(N1)->getExtraValueType());
|
|
|
|
if ((C2 & (~0ULL << ExtendBits)) == 0)
|
|
|
|
return getNode(ISD::AND, VT, N1.getOperand(0), N2);
|
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
break;
|
|
|
|
case ISD::OR:
|
|
|
|
if (!C2)return N1; // X or 0 -> X
|
|
|
|
if (N2C->isAllOnesValue())
|
2005-04-22 12:01:18 +08:00
|
|
|
return N2; // X or -1 -> -1
|
2005-01-07 15:46:32 +08:00
|
|
|
break;
|
|
|
|
case ISD::XOR:
|
|
|
|
if (!C2) return N1; // X xor 0 -> X
|
|
|
|
if (N2C->isAllOnesValue()) {
|
|
|
|
if (SetCCSDNode *SetCC = dyn_cast<SetCCSDNode>(N1.Val)){
|
|
|
|
// !(X op Y) -> (X !op Y)
|
|
|
|
bool isInteger = MVT::isInteger(SetCC->getOperand(0).getValueType());
|
|
|
|
return getSetCC(ISD::getSetCCInverse(SetCC->getCondition(),isInteger),
|
2005-01-18 10:52:03 +08:00
|
|
|
SetCC->getValueType(0),
|
2005-01-07 15:46:32 +08:00
|
|
|
SetCC->getOperand(0), SetCC->getOperand(1));
|
|
|
|
} else if (N1.getOpcode() == ISD::AND || N1.getOpcode() == ISD::OR) {
|
|
|
|
SDNode *Op = N1.Val;
|
|
|
|
// !(X or Y) -> (!X and !Y) iff X or Y are freely invertible
|
|
|
|
// !(X and Y) -> (!X or !Y) iff X or Y are freely invertible
|
|
|
|
SDOperand LHS = Op->getOperand(0), RHS = Op->getOperand(1);
|
|
|
|
if (isInvertibleForFree(RHS) || isInvertibleForFree(LHS)) {
|
|
|
|
LHS = getNode(ISD::XOR, VT, LHS, N2); // RHS = ~LHS
|
|
|
|
RHS = getNode(ISD::XOR, VT, RHS, N2); // RHS = ~RHS
|
|
|
|
if (Op->getOpcode() == ISD::AND)
|
|
|
|
return getNode(ISD::OR, VT, LHS, RHS);
|
|
|
|
return getNode(ISD::AND, VT, LHS, RHS);
|
|
|
|
}
|
|
|
|
}
|
2005-04-22 12:01:18 +08:00
|
|
|
// X xor -1 -> not(x) ?
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Reassociate ((X op C1) op C2) if possible.
|
|
|
|
if (N1.getOpcode() == Opcode && isAssociativeBinOp(Opcode))
|
|
|
|
if (ConstantSDNode *N3C = dyn_cast<ConstantSDNode>(N1.Val->getOperand(1)))
|
2005-01-08 06:44:09 +08:00
|
|
|
return getNode(Opcode, VT, N1.Val->getOperand(0),
|
2005-01-07 15:46:32 +08:00
|
|
|
getNode(Opcode, VT, N2, N1.Val->getOperand(1)));
|
|
|
|
}
|
|
|
|
|
|
|
|
ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1.Val);
|
|
|
|
ConstantFPSDNode *N2CFP = dyn_cast<ConstantFPSDNode>(N2.Val);
|
|
|
|
if (N1CFP)
|
|
|
|
if (N2CFP) {
|
|
|
|
double C1 = N1CFP->getValue(), C2 = N2CFP->getValue();
|
|
|
|
switch (Opcode) {
|
|
|
|
case ISD::ADD: return getConstantFP(C1 + C2, VT);
|
|
|
|
case ISD::SUB: return getConstantFP(C1 - C2, VT);
|
|
|
|
case ISD::MUL: return getConstantFP(C1 * C2, VT);
|
|
|
|
case ISD::SDIV:
|
|
|
|
if (C2) return getConstantFP(C1 / C2, VT);
|
|
|
|
break;
|
|
|
|
case ISD::SREM :
|
|
|
|
if (C2) return getConstantFP(fmod(C1, C2), VT);
|
|
|
|
break;
|
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
|
|
|
|
} else { // Cannonicalize constant to RHS if commutative
|
|
|
|
if (isCommutativeBinOp(Opcode)) {
|
|
|
|
std::swap(N1CFP, N2CFP);
|
|
|
|
std::swap(N1, N2);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Finally, fold operations that do not require constants.
|
|
|
|
switch (Opcode) {
|
2005-01-20 02:01:40 +08:00
|
|
|
case ISD::TokenFactor:
|
|
|
|
if (N1.getOpcode() == ISD::EntryToken)
|
|
|
|
return N2;
|
|
|
|
if (N2.getOpcode() == ISD::EntryToken)
|
|
|
|
return N1;
|
|
|
|
break;
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
case ISD::AND:
|
|
|
|
case ISD::OR:
|
|
|
|
if (SetCCSDNode *LHS = dyn_cast<SetCCSDNode>(N1.Val))
|
|
|
|
if (SetCCSDNode *RHS = dyn_cast<SetCCSDNode>(N2.Val)) {
|
|
|
|
SDOperand LL = LHS->getOperand(0), RL = RHS->getOperand(0);
|
|
|
|
SDOperand LR = LHS->getOperand(1), RR = RHS->getOperand(1);
|
|
|
|
ISD::CondCode Op2 = RHS->getCondition();
|
|
|
|
|
Fold:
// (X != 0) | (Y != 0) -> (X|Y != 0)
// (X == 0) & (Y == 0) -> (X|Y == 0)
Compiling this:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
to this:
_bar:
or r2, r3, r4
addic r3, r2, -1
subfe r3, r3, r2
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
llvm-svn: 21316
2005-04-18 11:59:53 +08:00
|
|
|
if (LR == RR && isa<ConstantSDNode>(LR) &&
|
|
|
|
Op2 == LHS->getCondition() && MVT::isInteger(LL.getValueType())) {
|
implement some more logical compares with constants, so that:
int foo1(int x, int y) {
int t1 = x >= 0;
int t2 = y >= 0;
return t1 & t2;
}
int foo2(int x, int y) {
int t1 = x == -1;
int t2 = y == -1;
return t1 & t2;
}
produces:
_foo1:
or r2, r4, r3
srwi r2, r2, 31
xori r3, r2, 1
blr
_foo2:
and r2, r4, r3
addic r2, r2, 1
li r2, 0
addze r3, r2
blr
instead of:
_foo1:
srwi r2, r4, 31
xori r2, r2, 1
srwi r3, r3, 31
xori r3, r3, 1
and r3, r2, r3
blr
_foo2:
addic r2, r4, 1
li r2, 0
addze r2, r2
addic r3, r3, 1
li r3, 0
addze r3, r3
and r3, r2, r3
blr
llvm-svn: 21547
2005-04-26 05:20:28 +08:00
|
|
|
// (X != 0) | (Y != 0) -> (X|Y != 0)
|
|
|
|
// (X == 0) & (Y == 0) -> (X|Y == 0)
|
|
|
|
// (X < 0) | (Y < 0) -> (X|Y < 0)
|
|
|
|
if (cast<ConstantSDNode>(LR)->getValue() == 0 &&
|
|
|
|
((Op2 == ISD::SETEQ && Opcode == ISD::AND) ||
|
|
|
|
(Op2 == ISD::SETNE && Opcode == ISD::OR) ||
|
|
|
|
(Op2 == ISD::SETLT && Opcode == ISD::OR)))
|
2005-04-22 06:36:52 +08:00
|
|
|
return getSetCC(Op2, VT,
|
Fold:
// (X != 0) | (Y != 0) -> (X|Y != 0)
// (X == 0) & (Y == 0) -> (X|Y == 0)
Compiling this:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
to this:
_bar:
or r2, r3, r4
addic r3, r2, -1
subfe r3, r3, r2
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
llvm-svn: 21316
2005-04-18 11:59:53 +08:00
|
|
|
getNode(ISD::OR, LR.getValueType(), LL, RL), LR);
|
implement some more logical compares with constants, so that:
int foo1(int x, int y) {
int t1 = x >= 0;
int t2 = y >= 0;
return t1 & t2;
}
int foo2(int x, int y) {
int t1 = x == -1;
int t2 = y == -1;
return t1 & t2;
}
produces:
_foo1:
or r2, r4, r3
srwi r2, r2, 31
xori r3, r2, 1
blr
_foo2:
and r2, r4, r3
addic r2, r2, 1
li r2, 0
addze r3, r2
blr
instead of:
_foo1:
srwi r2, r4, 31
xori r2, r2, 1
srwi r3, r3, 31
xori r3, r3, 1
and r3, r2, r3
blr
_foo2:
addic r2, r4, 1
li r2, 0
addze r2, r2
addic r3, r3, 1
li r3, 0
addze r3, r3
and r3, r2, r3
blr
llvm-svn: 21547
2005-04-26 05:20:28 +08:00
|
|
|
|
|
|
|
if (cast<ConstantSDNode>(LR)->isAllOnesValue()) {
|
|
|
|
// (X == -1) & (Y == -1) -> (X&Y == -1)
|
|
|
|
// (X != -1) | (Y != -1) -> (X&Y != -1)
|
2005-04-26 09:18:33 +08:00
|
|
|
// (X > -1) | (Y > -1) -> (X&Y > -1)
|
implement some more logical compares with constants, so that:
int foo1(int x, int y) {
int t1 = x >= 0;
int t2 = y >= 0;
return t1 & t2;
}
int foo2(int x, int y) {
int t1 = x == -1;
int t2 = y == -1;
return t1 & t2;
}
produces:
_foo1:
or r2, r4, r3
srwi r2, r2, 31
xori r3, r2, 1
blr
_foo2:
and r2, r4, r3
addic r2, r2, 1
li r2, 0
addze r3, r2
blr
instead of:
_foo1:
srwi r2, r4, 31
xori r2, r2, 1
srwi r3, r3, 31
xori r3, r3, 1
and r3, r2, r3
blr
_foo2:
addic r2, r4, 1
li r2, 0
addze r2, r2
addic r3, r3, 1
li r3, 0
addze r3, r3
and r3, r2, r3
blr
llvm-svn: 21547
2005-04-26 05:20:28 +08:00
|
|
|
if ((Opcode == ISD::AND && Op2 == ISD::SETEQ) ||
|
2005-04-26 09:18:33 +08:00
|
|
|
(Opcode == ISD::OR && Op2 == ISD::SETNE) ||
|
|
|
|
(Opcode == ISD::OR && Op2 == ISD::SETGT))
|
implement some more logical compares with constants, so that:
int foo1(int x, int y) {
int t1 = x >= 0;
int t2 = y >= 0;
return t1 & t2;
}
int foo2(int x, int y) {
int t1 = x == -1;
int t2 = y == -1;
return t1 & t2;
}
produces:
_foo1:
or r2, r4, r3
srwi r2, r2, 31
xori r3, r2, 1
blr
_foo2:
and r2, r4, r3
addic r2, r2, 1
li r2, 0
addze r3, r2
blr
instead of:
_foo1:
srwi r2, r4, 31
xori r2, r2, 1
srwi r3, r3, 31
xori r3, r3, 1
and r3, r2, r3
blr
_foo2:
addic r2, r4, 1
li r2, 0
addze r2, r2
addic r3, r3, 1
li r3, 0
addze r3, r3
and r3, r2, r3
blr
llvm-svn: 21547
2005-04-26 05:20:28 +08:00
|
|
|
return getSetCC(Op2, VT,
|
|
|
|
getNode(ISD::AND, LR.getValueType(), LL, RL), LR);
|
|
|
|
// (X > -1) & (Y > -1) -> (X|Y > -1)
|
|
|
|
if (Opcode == ISD::AND && Op2 == ISD::SETGT)
|
|
|
|
return getSetCC(Op2, VT,
|
|
|
|
getNode(ISD::OR, LR.getValueType(), LL, RL), LR);
|
|
|
|
}
|
Fold:
// (X != 0) | (Y != 0) -> (X|Y != 0)
// (X == 0) & (Y == 0) -> (X|Y == 0)
Compiling this:
int %bar(int %a, int %b) {
entry:
%tmp.1 = setne int %a, 0
%tmp.2 = setne int %b, 0
%tmp.3 = or bool %tmp.1, %tmp.2
%retval = cast bool %tmp.3 to int
ret int %retval
}
to this:
_bar:
or r2, r3, r4
addic r3, r2, -1
subfe r3, r3, r2
blr
instead of:
_bar:
addic r2, r3, -1
subfe r2, r2, r3
addic r3, r4, -1
subfe r3, r3, r4
or r3, r2, r3
blr
llvm-svn: 21316
2005-04-18 11:59:53 +08:00
|
|
|
}
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
// (X op1 Y) | (Y op2 X) -> (X op1 Y) | (X swapop2 Y)
|
|
|
|
if (LL == RR && LR == RL) {
|
|
|
|
Op2 = ISD::getSetCCSwappedOperands(Op2);
|
|
|
|
goto MatchedBackwards;
|
|
|
|
}
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
if (LL == RL && LR == RR) {
|
|
|
|
MatchedBackwards:
|
|
|
|
ISD::CondCode Result;
|
|
|
|
bool isInteger = MVT::isInteger(LL.getValueType());
|
|
|
|
if (Opcode == ISD::OR)
|
|
|
|
Result = ISD::getSetCCOrOperation(LHS->getCondition(), Op2,
|
|
|
|
isInteger);
|
|
|
|
else
|
|
|
|
Result = ISD::getSetCCAndOperation(LHS->getCondition(), Op2,
|
|
|
|
isInteger);
|
|
|
|
if (Result != ISD::SETCC_INVALID)
|
2005-01-18 10:52:03 +08:00
|
|
|
return getSetCC(Result, LHS->getValueType(0), LL, LR);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
}
|
2005-04-18 12:11:19 +08:00
|
|
|
|
|
|
|
// and/or zext(a), zext(b) -> zext(and/or a, b)
|
|
|
|
if (N1.getOpcode() == ISD::ZERO_EXTEND &&
|
|
|
|
N2.getOpcode() == ISD::ZERO_EXTEND &&
|
|
|
|
N1.getOperand(0).getValueType() == N2.getOperand(0).getValueType())
|
|
|
|
return getNode(ISD::ZERO_EXTEND, VT,
|
|
|
|
getNode(Opcode, N1.getOperand(0).getValueType(),
|
|
|
|
N1.getOperand(0), N2.getOperand(0)));
|
2005-01-07 15:46:32 +08:00
|
|
|
break;
|
|
|
|
case ISD::XOR:
|
|
|
|
if (N1 == N2) return getConstant(0, VT); // xor X, Y -> 0
|
|
|
|
break;
|
2005-04-09 11:02:46 +08:00
|
|
|
case ISD::ADD:
|
|
|
|
if (N2.getOpcode() == ISD::FNEG) // (A+ (-B) -> A-B
|
|
|
|
return getNode(ISD::SUB, VT, N1, N2.getOperand(0));
|
|
|
|
if (N1.getOpcode() == ISD::FNEG) // ((-A)+B) -> B-A
|
|
|
|
return getNode(ISD::SUB, VT, N2, N1.getOperand(0));
|
2005-04-10 12:04:49 +08:00
|
|
|
if (N1.getOpcode() == ISD::SUB && isa<ConstantSDNode>(N1.getOperand(0)) &&
|
|
|
|
cast<ConstantSDNode>(N1.getOperand(0))->getValue() == 0)
|
|
|
|
return getNode(ISD::SUB, VT, N2, N1.getOperand(1)); // (0-A)+B -> B-A
|
|
|
|
if (N2.getOpcode() == ISD::SUB && isa<ConstantSDNode>(N2.getOperand(0)) &&
|
|
|
|
cast<ConstantSDNode>(N2.getOperand(0))->getValue() == 0)
|
|
|
|
return getNode(ISD::SUB, VT, N1, N2.getOperand(1)); // A+(0-B) -> A-B
|
2005-04-09 11:02:46 +08:00
|
|
|
break;
|
Add a simple transformation. This allows us to compile one of the inner
loops in stepanov to this:
.LBB_Z5test0PdS__2: # no_exit.1
fldl data(,%eax,8)
fldl 24(%esp)
faddp %st(1)
fstl 24(%esp)
incl %eax
cmpl $2000, %eax
fstpl 16(%esp)
#FP_REG_KILL
jl .LBB_Z5test0PdS__2
instead of this:
.LBB_Z5test0PdS__2: # no_exit.1
fldl data(,%eax,8)
fldl 24(%esp)
faddp %st(1)
fstl 24(%esp)
incl %eax
movl $data, %ecx
movl %ecx, %edx
addl $16000, %edx
subl %ecx, %edx
movl %edx, %ecx
sarl $2, %ecx
shrl $29, %ecx
addl %ecx, %edx
sarl $3, %edx
cmpl %edx, %eax
fstpl 16(%esp)
#FP_REG_KILL
jl .LBB_Z5test0PdS__2
The old instruction selector produced:
.LBB_Z5test0PdS__2: # no_exit.1
fldl 24(%esp)
faddl data(,%eax,8)
fstl 24(%esp)
movl %eax, %ecx
incl %ecx
incl %eax
leal data+16000, %edx
movl $data, %edi
subl %edi, %edx
movl %edx, %edi
sarl $2, %edi
shrl $29, %edi
addl %edi, %edx
sarl $3, %edx
cmpl %edx, %ecx
fstpl 16(%esp)
#FP_REG_KILL
jl .LBB_Z5test0PdS__2 # no_exit.1
Which is even worse!
llvm-svn: 19419
2005-01-10 04:09:57 +08:00
|
|
|
case ISD::SUB:
|
|
|
|
if (N1.getOpcode() == ISD::ADD) {
|
|
|
|
if (N1.Val->getOperand(0) == N2)
|
|
|
|
return N1.Val->getOperand(1); // (A+B)-A == B
|
|
|
|
if (N1.Val->getOperand(1) == N2)
|
|
|
|
return N1.Val->getOperand(0); // (A+B)-B == A
|
|
|
|
}
|
2005-04-09 11:02:46 +08:00
|
|
|
if (N2.getOpcode() == ISD::FNEG) // (A- (-B) -> A+B
|
|
|
|
return getNode(ISD::ADD, VT, N1, N2.getOperand(0));
|
Add a simple transformation. This allows us to compile one of the inner
loops in stepanov to this:
.LBB_Z5test0PdS__2: # no_exit.1
fldl data(,%eax,8)
fldl 24(%esp)
faddp %st(1)
fstl 24(%esp)
incl %eax
cmpl $2000, %eax
fstpl 16(%esp)
#FP_REG_KILL
jl .LBB_Z5test0PdS__2
instead of this:
.LBB_Z5test0PdS__2: # no_exit.1
fldl data(,%eax,8)
fldl 24(%esp)
faddp %st(1)
fstl 24(%esp)
incl %eax
movl $data, %ecx
movl %ecx, %edx
addl $16000, %edx
subl %ecx, %edx
movl %edx, %ecx
sarl $2, %ecx
shrl $29, %ecx
addl %ecx, %edx
sarl $3, %edx
cmpl %edx, %eax
fstpl 16(%esp)
#FP_REG_KILL
jl .LBB_Z5test0PdS__2
The old instruction selector produced:
.LBB_Z5test0PdS__2: # no_exit.1
fldl 24(%esp)
faddl data(,%eax,8)
fstl 24(%esp)
movl %eax, %ecx
incl %ecx
incl %eax
leal data+16000, %edx
movl $data, %edi
subl %edi, %edx
movl %edx, %edi
sarl $2, %edi
shrl $29, %edi
addl %edi, %edx
sarl $3, %edx
cmpl %edx, %ecx
fstpl 16(%esp)
#FP_REG_KILL
jl .LBB_Z5test0PdS__2 # no_exit.1
Which is even worse!
llvm-svn: 19419
2005-01-10 04:09:57 +08:00
|
|
|
break;
|
2005-04-14 05:23:31 +08:00
|
|
|
// FIXME: figure out how to safely handle things like
|
|
|
|
// int foo(int x) { return 1 << (x & 255); }
|
|
|
|
// int bar() { return foo(256); }
|
|
|
|
#if 0
|
2005-04-13 07:32:28 +08:00
|
|
|
case ISD::SHL:
|
|
|
|
case ISD::SRL:
|
|
|
|
case ISD::SRA:
|
2005-04-13 10:58:13 +08:00
|
|
|
if (N2.getOpcode() == ISD::SIGN_EXTEND_INREG &&
|
|
|
|
cast<MVTSDNode>(N2)->getExtraValueType() != MVT::i1)
|
2005-04-13 07:32:28 +08:00
|
|
|
return getNode(Opcode, VT, N1, N2.getOperand(0));
|
2005-04-13 10:58:13 +08:00
|
|
|
else if (N2.getOpcode() == ISD::AND)
|
|
|
|
if (ConstantSDNode *AndRHS = dyn_cast<ConstantSDNode>(N2.getOperand(1))) {
|
|
|
|
// If the and is only masking out bits that cannot effect the shift,
|
|
|
|
// eliminate the and.
|
|
|
|
unsigned NumBits = MVT::getSizeInBits(VT);
|
|
|
|
if ((AndRHS->getValue() & (NumBits-1)) == NumBits-1)
|
|
|
|
return getNode(Opcode, VT, N1, N2.getOperand(0));
|
|
|
|
}
|
2005-04-13 07:32:28 +08:00
|
|
|
break;
|
2005-04-14 05:23:31 +08:00
|
|
|
#endif
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
2005-05-12 02:57:39 +08:00
|
|
|
// Memoize this node if possible.
|
|
|
|
SDNode *N;
|
2005-05-12 08:17:04 +08:00
|
|
|
if (Opcode != ISD::ADJCALLSTACKDOWN && Opcode != ISD::ADJCALLSTACKUP) {
|
2005-05-12 02:57:39 +08:00
|
|
|
SDNode *&BON = BinaryOps[std::make_pair(Opcode, std::make_pair(N1, N2))];
|
|
|
|
if (BON) return SDOperand(BON, 0);
|
|
|
|
|
|
|
|
BON = N = new SDNode(Opcode, N1, N2);
|
|
|
|
} else {
|
2005-05-12 08:17:04 +08:00
|
|
|
N = new SDNode(Opcode, N1, N2);
|
2005-05-12 02:57:39 +08:00
|
|
|
}
|
|
|
|
|
2005-05-10 04:22:17 +08:00
|
|
|
|
|
|
|
if (Opcode != ISD::READPORT && Opcode != ISD::READIO)
|
|
|
|
N->setValueTypes(VT);
|
|
|
|
else
|
|
|
|
N->setValueTypes(VT, MVT::Other);
|
2005-01-07 15:46:32 +08:00
|
|
|
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
2005-05-12 08:17:04 +08:00
|
|
|
// setAdjCallChain - This method changes the token chain of an
|
|
|
|
// ADJCALLSTACKDOWN/UP node to be the specified operand.
|
2005-05-12 02:57:39 +08:00
|
|
|
void SDNode::setAdjCallChain(SDOperand N) {
|
|
|
|
assert(N.getValueType() == MVT::Other);
|
2005-05-12 08:17:04 +08:00
|
|
|
assert((getOpcode() == ISD::ADJCALLSTACKDOWN ||
|
|
|
|
getOpcode() == ISD::ADJCALLSTACKUP) && "Cannot adjust this node!");
|
2005-05-12 02:57:39 +08:00
|
|
|
|
|
|
|
Operands[0].Val->removeUser(this);
|
|
|
|
Operands[0] = N;
|
|
|
|
N.Val->Uses.push_back(this);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
SDOperand SelectionDAG::getLoad(MVT::ValueType VT,
|
2005-04-28 04:10:01 +08:00
|
|
|
SDOperand Chain, SDOperand Ptr,
|
|
|
|
SDOperand SV) {
|
2005-01-07 15:46:32 +08:00
|
|
|
SDNode *&N = Loads[std::make_pair(Ptr, std::make_pair(Chain, VT))];
|
|
|
|
if (N) return SDOperand(N, 0);
|
2005-04-28 04:10:01 +08:00
|
|
|
N = new SDNode(ISD::LOAD, Chain, Ptr, SV);
|
2005-01-07 15:46:32 +08:00
|
|
|
|
|
|
|
// Loads have a token chain.
|
|
|
|
N->setValueTypes(VT, MVT::Other);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT,
|
|
|
|
SDOperand N1, SDOperand N2, SDOperand N3) {
|
2005-04-28 04:10:01 +08:00
|
|
|
assert(Opcode != ISD::STORE && "Store shouldn't use this anymore");
|
2005-01-07 15:46:32 +08:00
|
|
|
// Perform various simplifications.
|
|
|
|
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val);
|
|
|
|
ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.Val);
|
|
|
|
ConstantSDNode *N3C = dyn_cast<ConstantSDNode>(N3.Val);
|
|
|
|
switch (Opcode) {
|
|
|
|
case ISD::SELECT:
|
|
|
|
if (N1C)
|
|
|
|
if (N1C->getValue())
|
|
|
|
return N2; // select true, X, Y -> X
|
2005-04-22 06:36:52 +08:00
|
|
|
else
|
2005-01-07 15:46:32 +08:00
|
|
|
return N3; // select false, X, Y -> Y
|
|
|
|
|
|
|
|
if (N2 == N3) return N2; // select C, X, X -> X
|
|
|
|
|
|
|
|
if (VT == MVT::i1) { // Boolean SELECT
|
|
|
|
if (N2C) {
|
|
|
|
if (N2C->getValue()) // select C, 1, X -> C | X
|
|
|
|
return getNode(ISD::OR, VT, N1, N3);
|
|
|
|
else // select C, 0, X -> ~C & X
|
|
|
|
return getNode(ISD::AND, VT,
|
|
|
|
getNode(ISD::XOR, N1.getValueType(), N1,
|
|
|
|
getConstant(1, N1.getValueType())), N3);
|
|
|
|
} else if (N3C) {
|
|
|
|
if (N3C->getValue()) // select C, X, 1 -> ~C | X
|
|
|
|
return getNode(ISD::OR, VT,
|
|
|
|
getNode(ISD::XOR, N1.getValueType(), N1,
|
|
|
|
getConstant(1, N1.getValueType())), N2);
|
|
|
|
else // select C, X, 0 -> C & X
|
|
|
|
return getNode(ISD::AND, VT, N1, N2);
|
|
|
|
}
|
Remove some redundant checks, add a couple of new ones. This allows us to
compile this:
int foo (unsigned long a, unsigned long long g) {
return a >= g;
}
To:
foo:
movl 8(%esp), %eax
cmpl %eax, 4(%esp)
setae %al
cmpl $0, 12(%esp)
sete %cl
andb %al, %cl
movzbl %cl, %eax
ret
instead of:
foo:
movl 8(%esp), %eax
cmpl %eax, 4(%esp)
setae %al
movzbw %al, %cx
movl 12(%esp), %edx
cmpl $0, %edx
sete %al
movzbw %al, %ax
cmpl $0, %edx
cmove %cx, %ax
movzbl %al, %eax
ret
llvm-svn: 21244
2005-04-12 10:54:39 +08:00
|
|
|
|
|
|
|
if (N1 == N2) // X ? X : Y --> X ? 1 : Y --> X | Y
|
|
|
|
return getNode(ISD::OR, VT, N1, N3);
|
|
|
|
if (N1 == N3) // X ? Y : X --> X ? Y : 0 --> X & Y
|
|
|
|
return getNode(ISD::AND, VT, N1, N2);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
2005-04-09 13:15:53 +08:00
|
|
|
// If this is a selectcc, check to see if we can simplify the result.
|
|
|
|
if (SetCCSDNode *SetCC = dyn_cast<SetCCSDNode>(N1)) {
|
|
|
|
if (ConstantFPSDNode *CFP =
|
|
|
|
dyn_cast<ConstantFPSDNode>(SetCC->getOperand(1)))
|
|
|
|
if (CFP->getValue() == 0.0) { // Allow either -0.0 or 0.0
|
|
|
|
// select (setg[te] X, +/-0.0), X, fneg(X) -> fabs
|
|
|
|
if ((SetCC->getCondition() == ISD::SETGE ||
|
|
|
|
SetCC->getCondition() == ISD::SETGT) &&
|
|
|
|
N2 == SetCC->getOperand(0) && N3.getOpcode() == ISD::FNEG &&
|
|
|
|
N3.getOperand(0) == N2)
|
|
|
|
return getNode(ISD::FABS, VT, N2);
|
|
|
|
|
|
|
|
// select (setl[te] X, +/-0.0), fneg(X), X -> fabs
|
|
|
|
if ((SetCC->getCondition() == ISD::SETLT ||
|
|
|
|
SetCC->getCondition() == ISD::SETLE) &&
|
|
|
|
N3 == SetCC->getOperand(0) && N2.getOpcode() == ISD::FNEG &&
|
|
|
|
N2.getOperand(0) == N3)
|
|
|
|
return getNode(ISD::FABS, VT, N3);
|
|
|
|
}
|
2005-05-12 14:27:02 +08:00
|
|
|
// select (setlt X, 0), A, 0 -> and (sra X, size(X)-1), A
|
2005-04-14 05:23:31 +08:00
|
|
|
if (ConstantSDNode *CN =
|
|
|
|
dyn_cast<ConstantSDNode>(SetCC->getOperand(1)))
|
|
|
|
if (CN->getValue() == 0 && N3C && N3C->getValue() == 0)
|
|
|
|
if (SetCC->getCondition() == ISD::SETLT) {
|
|
|
|
MVT::ValueType XType = SetCC->getOperand(0).getValueType();
|
|
|
|
MVT::ValueType AType = N2.getValueType();
|
|
|
|
if (XType >= AType) {
|
2005-05-12 14:27:02 +08:00
|
|
|
// and (sra X, size(X)-1, A) -> "and (srl X, C2), A" iff A is a
|
|
|
|
// single-bit constant. FIXME: remove once the dag combiner
|
|
|
|
// exists.
|
|
|
|
if (ConstantSDNode *AC = dyn_cast<ConstantSDNode>(N2))
|
|
|
|
if ((AC->getValue() & (AC->getValue()-1)) == 0) {
|
|
|
|
unsigned ShCtV = ExactLog2(AC->getValue());
|
|
|
|
ShCtV = MVT::getSizeInBits(XType)-ShCtV-1;
|
|
|
|
SDOperand ShCt = getConstant(ShCtV, TLI.getShiftAmountTy());
|
|
|
|
SDOperand Shift = getNode(ISD::SRL, XType,
|
|
|
|
SetCC->getOperand(0), ShCt);
|
|
|
|
if (XType > AType)
|
|
|
|
Shift = getNode(ISD::TRUNCATE, AType, Shift);
|
|
|
|
return getNode(ISD::AND, AType, Shift, N2);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-04-14 05:23:31 +08:00
|
|
|
SDOperand Shift = getNode(ISD::SRA, XType, SetCC->getOperand(0),
|
|
|
|
getConstant(MVT::getSizeInBits(XType)-1,
|
|
|
|
TLI.getShiftAmountTy()));
|
|
|
|
if (XType > AType)
|
|
|
|
Shift = getNode(ISD::TRUNCATE, AType, Shift);
|
|
|
|
return getNode(ISD::AND, AType, Shift, N2);
|
|
|
|
}
|
|
|
|
}
|
2005-04-09 13:15:53 +08:00
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
break;
|
2005-01-08 06:49:57 +08:00
|
|
|
case ISD::BRCOND:
|
|
|
|
if (N2C)
|
|
|
|
if (N2C->getValue()) // Unconditional branch
|
|
|
|
return getNode(ISD::BR, MVT::Other, N1, N3);
|
|
|
|
else
|
|
|
|
return N1; // Never-taken branch
|
2005-01-08 07:32:00 +08:00
|
|
|
break;
|
2005-04-14 05:23:31 +08:00
|
|
|
// FIXME: figure out how to safely handle things like
|
|
|
|
// int foo(int x) { return 1 << (x & 255); }
|
|
|
|
// int bar() { return foo(256); }
|
|
|
|
#if 0
|
2005-04-13 07:32:28 +08:00
|
|
|
case ISD::SRA_PARTS:
|
|
|
|
case ISD::SRL_PARTS:
|
|
|
|
case ISD::SHL_PARTS:
|
2005-04-13 10:58:13 +08:00
|
|
|
if (N3.getOpcode() == ISD::SIGN_EXTEND_INREG &&
|
|
|
|
cast<MVTSDNode>(N3)->getExtraValueType() != MVT::i1)
|
2005-04-13 07:32:28 +08:00
|
|
|
return getNode(Opcode, VT, N1, N2, N3.getOperand(0));
|
2005-04-13 10:58:13 +08:00
|
|
|
else if (N3.getOpcode() == ISD::AND)
|
|
|
|
if (ConstantSDNode *AndRHS = dyn_cast<ConstantSDNode>(N3.getOperand(1))) {
|
|
|
|
// If the and is only masking out bits that cannot effect the shift,
|
|
|
|
// eliminate the and.
|
|
|
|
unsigned NumBits = MVT::getSizeInBits(VT)*2;
|
|
|
|
if ((AndRHS->getValue() & (NumBits-1)) == NumBits-1)
|
|
|
|
return getNode(Opcode, VT, N1, N2, N3.getOperand(0));
|
|
|
|
}
|
2005-04-13 07:32:28 +08:00
|
|
|
break;
|
2005-04-14 05:23:31 +08:00
|
|
|
#endif
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
SDNode *N = new SDNode(Opcode, N1, N2, N3);
|
|
|
|
switch (Opcode) {
|
2005-04-22 06:36:52 +08:00
|
|
|
default:
|
2005-01-07 15:46:32 +08:00
|
|
|
N->setValueTypes(VT);
|
|
|
|
break;
|
|
|
|
case ISD::DYNAMIC_STACKALLOC: // DYNAMIC_STACKALLOC produces pointer and chain
|
|
|
|
N->setValueTypes(VT, MVT::Other);
|
|
|
|
break;
|
2005-04-02 12:00:59 +08:00
|
|
|
|
|
|
|
case ISD::SRA_PARTS:
|
|
|
|
case ISD::SRL_PARTS:
|
|
|
|
case ISD::SHL_PARTS: {
|
|
|
|
std::vector<MVT::ValueType> V(N->getNumOperands()-1, VT);
|
|
|
|
N->setValueTypes(V);
|
|
|
|
break;
|
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// FIXME: memoize NODES
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
2005-04-28 04:10:01 +08:00
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT,
|
|
|
|
SDOperand N1, SDOperand N2, SDOperand N3,
|
|
|
|
SDOperand N4) {
|
|
|
|
assert(Opcode == ISD::STORE && "Only stores should use this");
|
|
|
|
|
|
|
|
SDNode *N = new SDNode(Opcode, N1, N2, N3, N4);
|
|
|
|
N->setValueTypes(VT);
|
|
|
|
|
|
|
|
// FIXME: memoize NODES
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
2005-05-09 12:14:13 +08:00
|
|
|
SDOperand SelectionDAG::getSrcValue(const Value *V, int Offset) {
|
2005-05-09 12:28:51 +08:00
|
|
|
assert((!V || isa<PointerType>(V->getType())) &&
|
|
|
|
"SrcValue is not a pointer?");
|
2005-05-09 12:14:13 +08:00
|
|
|
SDNode *&N = ValueNodes[std::make_pair(V, Offset)];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
|
|
|
|
N = new SrcValueSDNode(V, Offset);
|
2005-04-28 04:10:01 +08:00
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT,
|
|
|
|
std::vector<SDOperand> &Children) {
|
|
|
|
switch (Children.size()) {
|
|
|
|
case 0: return getNode(Opcode, VT);
|
|
|
|
case 1: return getNode(Opcode, VT, Children[0]);
|
|
|
|
case 2: return getNode(Opcode, VT, Children[0], Children[1]);
|
|
|
|
case 3: return getNode(Opcode, VT, Children[0], Children[1], Children[2]);
|
2005-04-09 11:27:28 +08:00
|
|
|
default: break;
|
|
|
|
}
|
|
|
|
|
|
|
|
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(Children[1].Val);
|
|
|
|
switch (Opcode) {
|
|
|
|
default: break;
|
|
|
|
case ISD::BRCONDTWOWAY:
|
|
|
|
if (N1C)
|
|
|
|
if (N1C->getValue()) // Unconditional branch to true dest.
|
|
|
|
return getNode(ISD::BR, MVT::Other, Children[0], Children[2]);
|
|
|
|
else // Unconditional branch to false dest.
|
|
|
|
return getNode(ISD::BR, MVT::Other, Children[0], Children[3]);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
// FIXME: MEMOIZE!!
|
|
|
|
SDNode *N = new SDNode(Opcode, Children);
|
|
|
|
if (Opcode != ISD::ADD_PARTS && Opcode != ISD::SUB_PARTS) {
|
|
|
|
N->setValueTypes(VT);
|
|
|
|
} else {
|
|
|
|
std::vector<MVT::ValueType> V(N->getNumOperands()/2, VT);
|
|
|
|
N->setValueTypes(V);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
2005-04-09 11:27:28 +08:00
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
|
2005-01-15 14:17:04 +08:00
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT,SDOperand N1,
|
|
|
|
MVT::ValueType EVT) {
|
|
|
|
|
|
|
|
switch (Opcode) {
|
|
|
|
default: assert(0 && "Bad opcode for this accessor!");
|
|
|
|
case ISD::FP_ROUND_INREG:
|
|
|
|
assert(VT == N1.getValueType() && "Not an inreg round!");
|
|
|
|
assert(MVT::isFloatingPoint(VT) && MVT::isFloatingPoint(EVT) &&
|
|
|
|
"Cannot FP_ROUND_INREG integer types");
|
|
|
|
if (EVT == VT) return N1; // Not actually rounding
|
|
|
|
assert(EVT < VT && "Not rounding down!");
|
constant fold FP_ROUND_INREG, ZERO_EXTEND_INREG, and SIGN_EXTEND_INREG
This allows the alpha backend to compile:
bool %test(uint %P) {
%c = seteq uint %P, 0
ret bool %c
}
into:
test:
ldgp $29, 0($27)
ZAP $16,240,$0
CMPEQ $0,0,$0
AND $0,1,$0
ret $31,($26),1
instead of:
test:
ldgp $29, 0($27)
ZAP $16,240,$0
ldiq $1,0
ZAP $1,240,$1
CMPEQ $0,$1,$0
AND $0,1,$0
ret $31,($26),1
... and fixes PR534.
llvm-svn: 20534
2005-03-10 02:37:12 +08:00
|
|
|
|
|
|
|
if (isa<ConstantFPSDNode>(N1))
|
|
|
|
return getNode(ISD::FP_EXTEND, VT, getNode(ISD::FP_ROUND, EVT, N1));
|
2005-01-15 14:17:04 +08:00
|
|
|
break;
|
|
|
|
case ISD::SIGN_EXTEND_INREG:
|
|
|
|
assert(VT == N1.getValueType() && "Not an inreg extend!");
|
|
|
|
assert(MVT::isInteger(VT) && MVT::isInteger(EVT) &&
|
|
|
|
"Cannot *_EXTEND_INREG FP types");
|
|
|
|
if (EVT == VT) return N1; // Not actually extending
|
|
|
|
assert(EVT < VT && "Not extending!");
|
2005-01-16 08:17:20 +08:00
|
|
|
|
2005-04-13 10:38:18 +08:00
|
|
|
// Extending a constant? Just return the extended constant.
|
constant fold FP_ROUND_INREG, ZERO_EXTEND_INREG, and SIGN_EXTEND_INREG
This allows the alpha backend to compile:
bool %test(uint %P) {
%c = seteq uint %P, 0
ret bool %c
}
into:
test:
ldgp $29, 0($27)
ZAP $16,240,$0
CMPEQ $0,0,$0
AND $0,1,$0
ret $31,($26),1
instead of:
test:
ldgp $29, 0($27)
ZAP $16,240,$0
ldiq $1,0
ZAP $1,240,$1
CMPEQ $0,$1,$0
AND $0,1,$0
ret $31,($26),1
... and fixes PR534.
llvm-svn: 20534
2005-03-10 02:37:12 +08:00
|
|
|
if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.Val)) {
|
|
|
|
SDOperand Tmp = getNode(ISD::TRUNCATE, EVT, N1);
|
2005-04-13 10:38:18 +08:00
|
|
|
return getNode(ISD::SIGN_EXTEND, VT, Tmp);
|
constant fold FP_ROUND_INREG, ZERO_EXTEND_INREG, and SIGN_EXTEND_INREG
This allows the alpha backend to compile:
bool %test(uint %P) {
%c = seteq uint %P, 0
ret bool %c
}
into:
test:
ldgp $29, 0($27)
ZAP $16,240,$0
CMPEQ $0,0,$0
AND $0,1,$0
ret $31,($26),1
instead of:
test:
ldgp $29, 0($27)
ZAP $16,240,$0
ldiq $1,0
ZAP $1,240,$1
CMPEQ $0,$1,$0
AND $0,1,$0
ret $31,($26),1
... and fixes PR534.
llvm-svn: 20534
2005-03-10 02:37:12 +08:00
|
|
|
}
|
|
|
|
|
2005-01-16 08:17:20 +08:00
|
|
|
// If we are sign extending an extension, use the original source.
|
2005-04-13 10:38:18 +08:00
|
|
|
if (N1.getOpcode() == ISD::SIGN_EXTEND_INREG)
|
|
|
|
if (cast<MVTSDNode>(N1)->getExtraValueType() <= EVT)
|
2005-01-16 08:17:20 +08:00
|
|
|
return N1;
|
|
|
|
|
2005-04-13 10:38:18 +08:00
|
|
|
// If we are sign extending a sextload, return just the load.
|
|
|
|
if (N1.getOpcode() == ISD::SEXTLOAD && Opcode == ISD::SIGN_EXTEND_INREG)
|
2005-04-10 12:33:08 +08:00
|
|
|
if (cast<MVTSDNode>(N1)->getExtraValueType() <= EVT)
|
|
|
|
return N1;
|
|
|
|
|
If a target zero or sign extends the result of its setcc, allow folding of
this into sign/zero extension instructions later.
On PPC, for example, this testcase:
%G = external global sbyte
implementation
void %test(int %X, int %Y) {
%C = setlt int %X, %Y
%D = cast bool %C to sbyte
store sbyte %D, sbyte* %G
ret void
}
Now codegens to:
cmpw cr0, r3, r4
li r3, 1
li r4, 0
blt .LBB_test_2 ;
.LBB_test_1: ;
or r3, r4, r4
.LBB_test_2: ;
addis r2, r2, ha16(L_G$non_lazy_ptr-"L00000$pb")
lwz r2, lo16(L_G$non_lazy_ptr-"L00000$pb")(r2)
stb r3, 0(r2)
instead of:
cmpw cr0, r3, r4
li r3, 1
li r4, 0
blt .LBB_test_2 ;
.LBB_test_1: ;
or r3, r4, r4
.LBB_test_2: ;
*** rlwinm r3, r3, 0, 31, 31
addis r2, r2, ha16(L_G$non_lazy_ptr-"L00000$pb")
lwz r2, lo16(L_G$non_lazy_ptr-"L00000$pb")(r2)
stb r3, 0(r2)
llvm-svn: 21148
2005-04-08 03:43:53 +08:00
|
|
|
// If we are extending the result of a setcc, and we already know the
|
|
|
|
// contents of the top bits, eliminate the extension.
|
2005-04-13 10:38:18 +08:00
|
|
|
if (N1.getOpcode() == ISD::SETCC &&
|
2005-04-22 06:36:52 +08:00
|
|
|
TLI.getSetCCResultContents() ==
|
2005-04-13 10:38:18 +08:00
|
|
|
TargetLowering::ZeroOrNegativeOneSetCCResult)
|
|
|
|
return N1;
|
Don't bother sign/zext_inreg'ing the result of an and operation if we know
the result does change as a result of the extend.
This improves codegen for Alpha on this testcase:
int %a(ushort* %i) {
%tmp.1 = load ushort* %i
%tmp.2 = cast ushort %tmp.1 to int
%tmp.4 = and int %tmp.2, 1
ret int %tmp.4
}
Generating:
a:
ldgp $29, 0($27)
ldwu $0,0($16)
and $0,1,$0
ret $31,($26),1
instead of:
a:
ldgp $29, 0($27)
ldwu $0,0($16)
and $0,1,$0
addl $0,0,$0
ret $31,($26),1
btw, alpha really should switch to livein/outs for args :)
llvm-svn: 21213
2005-04-11 07:37:16 +08:00
|
|
|
|
|
|
|
// If we are sign extending the result of an (and X, C) operation, and we
|
|
|
|
// know the extended bits are zeros already, don't do the extend.
|
|
|
|
if (N1.getOpcode() == ISD::AND)
|
|
|
|
if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getOperand(1))) {
|
|
|
|
uint64_t Mask = N1C->getValue();
|
|
|
|
unsigned NumBits = MVT::getSizeInBits(EVT);
|
2005-04-13 10:38:18 +08:00
|
|
|
if ((Mask & (~0ULL << (NumBits-1))) == 0)
|
|
|
|
return N1;
|
Don't bother sign/zext_inreg'ing the result of an and operation if we know
the result does change as a result of the extend.
This improves codegen for Alpha on this testcase:
int %a(ushort* %i) {
%tmp.1 = load ushort* %i
%tmp.2 = cast ushort %tmp.1 to int
%tmp.4 = and int %tmp.2, 1
ret int %tmp.4
}
Generating:
a:
ldgp $29, 0($27)
ldwu $0,0($16)
and $0,1,$0
ret $31,($26),1
instead of:
a:
ldgp $29, 0($27)
ldwu $0,0($16)
and $0,1,$0
addl $0,0,$0
ret $31,($26),1
btw, alpha really should switch to livein/outs for args :)
llvm-svn: 21213
2005-04-11 07:37:16 +08:00
|
|
|
}
|
2005-01-15 14:17:04 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
EVTStruct NN;
|
|
|
|
NN.Opcode = Opcode;
|
|
|
|
NN.VT = VT;
|
|
|
|
NN.EVT = EVT;
|
|
|
|
NN.Ops.push_back(N1);
|
|
|
|
|
|
|
|
SDNode *&N = MVTSDNodes[NN];
|
|
|
|
if (N) return SDOperand(N, 0);
|
|
|
|
N = new MVTSDNode(Opcode, VT, N1, EVT);
|
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
2005-01-15 06:08:15 +08:00
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT,SDOperand N1,
|
2005-04-28 04:10:01 +08:00
|
|
|
SDOperand N2, SDOperand N3, MVT::ValueType EVT) {
|
2005-01-15 06:08:15 +08:00
|
|
|
switch (Opcode) {
|
|
|
|
default: assert(0 && "Bad opcode for this accessor!");
|
|
|
|
case ISD::EXTLOAD:
|
|
|
|
case ISD::SEXTLOAD:
|
|
|
|
case ISD::ZEXTLOAD:
|
2005-04-13 10:38:18 +08:00
|
|
|
// If they are asking for an extending load from/to the same thing, return a
|
2005-01-15 06:08:15 +08:00
|
|
|
// normal load.
|
|
|
|
if (VT == EVT)
|
2005-04-28 04:10:01 +08:00
|
|
|
return getLoad(VT, N1, N2, N3);
|
2005-01-15 06:08:15 +08:00
|
|
|
assert(EVT < VT && "Should only be an extending load, not truncating!");
|
2005-04-22 06:36:52 +08:00
|
|
|
assert((Opcode == ISD::EXTLOAD || MVT::isInteger(VT)) &&
|
2005-01-15 06:08:15 +08:00
|
|
|
"Cannot sign/zero extend a FP load!");
|
|
|
|
assert(MVT::isInteger(VT) == MVT::isInteger(EVT) &&
|
|
|
|
"Cannot convert from FP to Int or Int -> FP!");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
EVTStruct NN;
|
|
|
|
NN.Opcode = Opcode;
|
|
|
|
NN.VT = VT;
|
|
|
|
NN.EVT = EVT;
|
|
|
|
NN.Ops.push_back(N1);
|
|
|
|
NN.Ops.push_back(N2);
|
2005-04-28 04:10:01 +08:00
|
|
|
NN.Ops.push_back(N3);
|
2005-01-15 06:08:15 +08:00
|
|
|
|
|
|
|
SDNode *&N = MVTSDNodes[NN];
|
|
|
|
if (N) return SDOperand(N, 0);
|
2005-04-28 04:10:01 +08:00
|
|
|
N = new MVTSDNode(Opcode, VT, MVT::Other, N1, N2, N3, EVT);
|
2005-01-15 06:08:15 +08:00
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand SelectionDAG::getNode(unsigned Opcode, MVT::ValueType VT,SDOperand N1,
|
2005-04-28 04:10:01 +08:00
|
|
|
SDOperand N2, SDOperand N3, SDOperand N4, MVT::ValueType EVT) {
|
2005-01-15 06:08:15 +08:00
|
|
|
switch (Opcode) {
|
|
|
|
default: assert(0 && "Bad opcode for this accessor!");
|
|
|
|
case ISD::TRUNCSTORE:
|
2005-01-15 14:17:04 +08:00
|
|
|
#if 0 // FIXME: If the target supports EVT natively, convert to a truncate/store
|
|
|
|
// If this is a truncating store of a constant, convert to the desired type
|
|
|
|
// and store it instead.
|
|
|
|
if (isa<Constant>(N1)) {
|
|
|
|
SDOperand Op = getNode(ISD::TRUNCATE, EVT, N1);
|
|
|
|
if (isa<Constant>(Op))
|
2005-04-22 06:36:52 +08:00
|
|
|
N1 = Op;
|
2005-01-15 14:17:04 +08:00
|
|
|
}
|
|
|
|
// Also for ConstantFP?
|
|
|
|
#endif
|
2005-01-15 06:08:15 +08:00
|
|
|
if (N1.getValueType() == EVT) // Normal store?
|
2005-04-28 04:10:01 +08:00
|
|
|
return getNode(ISD::STORE, VT, N1, N2, N3, N4);
|
2005-01-15 14:17:04 +08:00
|
|
|
assert(N2.getValueType() > EVT && "Not a truncation?");
|
|
|
|
assert(MVT::isInteger(N2.getValueType()) == MVT::isInteger(EVT) &&
|
2005-01-15 06:08:15 +08:00
|
|
|
"Can't do FP-INT conversion!");
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
EVTStruct NN;
|
|
|
|
NN.Opcode = Opcode;
|
|
|
|
NN.VT = VT;
|
|
|
|
NN.EVT = EVT;
|
|
|
|
NN.Ops.push_back(N1);
|
|
|
|
NN.Ops.push_back(N2);
|
|
|
|
NN.Ops.push_back(N3);
|
2005-04-28 04:10:01 +08:00
|
|
|
NN.Ops.push_back(N4);
|
2005-01-15 06:08:15 +08:00
|
|
|
|
|
|
|
SDNode *&N = MVTSDNodes[NN];
|
|
|
|
if (N) return SDOperand(N, 0);
|
2005-04-28 04:10:01 +08:00
|
|
|
N = new MVTSDNode(Opcode, VT, N1, N2, N3, N4, EVT);
|
2005-01-15 06:08:15 +08:00
|
|
|
AllNodes.push_back(N);
|
|
|
|
return SDOperand(N, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-01-13 02:37:47 +08:00
|
|
|
/// hasNUsesOfValue - Return true if there are exactly NUSES uses of the
|
|
|
|
/// indicated value. This method ignores uses of other values defined by this
|
|
|
|
/// operation.
|
|
|
|
bool SDNode::hasNUsesOfValue(unsigned NUses, unsigned Value) {
|
|
|
|
assert(Value < getNumValues() && "Bad value!");
|
|
|
|
|
|
|
|
// If there is only one value, this is easy.
|
|
|
|
if (getNumValues() == 1)
|
|
|
|
return use_size() == NUses;
|
|
|
|
if (Uses.size() < NUses) return false;
|
|
|
|
|
|
|
|
SDOperand TheValue(this, Value);
|
|
|
|
|
|
|
|
std::set<SDNode*> UsersHandled;
|
|
|
|
|
|
|
|
for (std::vector<SDNode*>::iterator UI = Uses.begin(), E = Uses.end();
|
|
|
|
UI != E; ++UI) {
|
|
|
|
SDNode *User = *UI;
|
|
|
|
if (User->getNumOperands() == 1 ||
|
|
|
|
UsersHandled.insert(User).second) // First time we've seen this?
|
|
|
|
for (unsigned i = 0, e = User->getNumOperands(); i != e; ++i)
|
|
|
|
if (User->getOperand(i) == TheValue) {
|
|
|
|
if (NUses == 0)
|
|
|
|
return false; // too many uses
|
|
|
|
--NUses;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Found exactly the right number of uses?
|
|
|
|
return NUses == 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-01-11 07:25:25 +08:00
|
|
|
const char *SDNode::getOperationName() const {
|
|
|
|
switch (getOpcode()) {
|
|
|
|
default: return "<<Unknown>>";
|
2005-04-01 05:24:06 +08:00
|
|
|
case ISD::PCMARKER: return "PCMarker";
|
2005-05-09 12:08:27 +08:00
|
|
|
case ISD::SRCVALUE: return "SrcValue";
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::EntryToken: return "EntryToken";
|
2005-01-14 01:59:10 +08:00
|
|
|
case ISD::TokenFactor: return "TokenFactor";
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::Constant: return "Constant";
|
|
|
|
case ISD::ConstantFP: return "ConstantFP";
|
|
|
|
case ISD::GlobalAddress: return "GlobalAddress";
|
|
|
|
case ISD::FrameIndex: return "FrameIndex";
|
|
|
|
case ISD::BasicBlock: return "BasicBlock";
|
|
|
|
case ISD::ExternalSymbol: return "ExternalSymbol";
|
|
|
|
case ISD::ConstantPool: return "ConstantPoolIndex";
|
|
|
|
case ISD::CopyToReg: return "CopyToReg";
|
|
|
|
case ISD::CopyFromReg: return "CopyFromReg";
|
2005-01-14 04:50:02 +08:00
|
|
|
case ISD::ImplicitDef: return "ImplicitDef";
|
2005-04-02 06:34:39 +08:00
|
|
|
case ISD::UNDEF: return "undef";
|
2005-01-11 07:25:25 +08:00
|
|
|
|
2005-04-02 12:58:41 +08:00
|
|
|
// Unary operators
|
|
|
|
case ISD::FABS: return "fabs";
|
|
|
|
case ISD::FNEG: return "fneg";
|
2005-04-29 05:44:03 +08:00
|
|
|
case ISD::FSQRT: return "fsqrt";
|
|
|
|
case ISD::FSIN: return "fsin";
|
|
|
|
case ISD::FCOS: return "fcos";
|
2005-04-02 12:58:41 +08:00
|
|
|
|
|
|
|
// Binary operators
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::ADD: return "add";
|
|
|
|
case ISD::SUB: return "sub";
|
|
|
|
case ISD::MUL: return "mul";
|
2005-04-06 06:36:56 +08:00
|
|
|
case ISD::MULHU: return "mulhu";
|
|
|
|
case ISD::MULHS: return "mulhs";
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::SDIV: return "sdiv";
|
|
|
|
case ISD::UDIV: return "udiv";
|
|
|
|
case ISD::SREM: return "srem";
|
|
|
|
case ISD::UREM: return "urem";
|
|
|
|
case ISD::AND: return "and";
|
|
|
|
case ISD::OR: return "or";
|
|
|
|
case ISD::XOR: return "xor";
|
|
|
|
case ISD::SHL: return "shl";
|
|
|
|
case ISD::SRA: return "sra";
|
|
|
|
case ISD::SRL: return "srl";
|
|
|
|
|
|
|
|
case ISD::SELECT: return "select";
|
2005-01-21 02:50:55 +08:00
|
|
|
case ISD::ADD_PARTS: return "add_parts";
|
|
|
|
case ISD::SUB_PARTS: return "sub_parts";
|
2005-04-02 11:30:42 +08:00
|
|
|
case ISD::SHL_PARTS: return "shl_parts";
|
|
|
|
case ISD::SRA_PARTS: return "sra_parts";
|
|
|
|
case ISD::SRL_PARTS: return "srl_parts";
|
2005-01-11 07:25:25 +08:00
|
|
|
|
2005-04-29 05:44:03 +08:00
|
|
|
// Conversion operators.
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::SIGN_EXTEND: return "sign_extend";
|
|
|
|
case ISD::ZERO_EXTEND: return "zero_extend";
|
2005-01-15 14:17:04 +08:00
|
|
|
case ISD::SIGN_EXTEND_INREG: return "sign_extend_inreg";
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::TRUNCATE: return "truncate";
|
|
|
|
case ISD::FP_ROUND: return "fp_round";
|
2005-01-15 14:17:04 +08:00
|
|
|
case ISD::FP_ROUND_INREG: return "fp_round_inreg";
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::FP_EXTEND: return "fp_extend";
|
|
|
|
|
|
|
|
case ISD::SINT_TO_FP: return "sint_to_fp";
|
|
|
|
case ISD::UINT_TO_FP: return "uint_to_fp";
|
|
|
|
case ISD::FP_TO_SINT: return "fp_to_sint";
|
|
|
|
case ISD::FP_TO_UINT: return "fp_to_uint";
|
|
|
|
|
|
|
|
// Control flow instructions
|
|
|
|
case ISD::BR: return "br";
|
|
|
|
case ISD::BRCOND: return "brcond";
|
2005-04-09 11:27:28 +08:00
|
|
|
case ISD::BRCONDTWOWAY: return "brcondtwoway";
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::RET: return "ret";
|
|
|
|
case ISD::CALL: return "call";
|
|
|
|
case ISD::ADJCALLSTACKDOWN: return "adjcallstackdown";
|
|
|
|
case ISD::ADJCALLSTACKUP: return "adjcallstackup";
|
|
|
|
|
|
|
|
// Other operators
|
|
|
|
case ISD::LOAD: return "load";
|
|
|
|
case ISD::STORE: return "store";
|
2005-01-15 06:08:15 +08:00
|
|
|
case ISD::EXTLOAD: return "extload";
|
|
|
|
case ISD::SEXTLOAD: return "sextload";
|
|
|
|
case ISD::ZEXTLOAD: return "zextload";
|
|
|
|
case ISD::TRUNCSTORE: return "truncstore";
|
|
|
|
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::DYNAMIC_STACKALLOC: return "dynamic_stackalloc";
|
|
|
|
case ISD::EXTRACT_ELEMENT: return "extract_element";
|
|
|
|
case ISD::BUILD_PAIR: return "build_pair";
|
2005-01-11 13:57:01 +08:00
|
|
|
case ISD::MEMSET: return "memset";
|
|
|
|
case ISD::MEMCPY: return "memcpy";
|
|
|
|
case ISD::MEMMOVE: return "memmove";
|
2005-01-07 15:46:32 +08:00
|
|
|
|
2005-05-11 12:50:30 +08:00
|
|
|
// Bit counting
|
|
|
|
case ISD::CTPOP: return "ctpop";
|
|
|
|
case ISD::CTTZ: return "cttz";
|
|
|
|
case ISD::CTLZ: return "ctlz";
|
|
|
|
|
|
|
|
// IO Intrinsics
|
2005-05-10 04:22:17 +08:00
|
|
|
case ISD::READPORT: return "readport";
|
|
|
|
case ISD::WRITEPORT: return "writeport";
|
|
|
|
case ISD::READIO: return "readio";
|
|
|
|
case ISD::WRITEIO: return "writeio";
|
|
|
|
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::SETCC:
|
|
|
|
const SetCCSDNode *SetCC = cast<SetCCSDNode>(this);
|
|
|
|
switch (SetCC->getCondition()) {
|
|
|
|
default: assert(0 && "Unknown setcc condition!");
|
|
|
|
case ISD::SETOEQ: return "setcc:setoeq";
|
|
|
|
case ISD::SETOGT: return "setcc:setogt";
|
|
|
|
case ISD::SETOGE: return "setcc:setoge";
|
|
|
|
case ISD::SETOLT: return "setcc:setolt";
|
|
|
|
case ISD::SETOLE: return "setcc:setole";
|
|
|
|
case ISD::SETONE: return "setcc:setone";
|
2005-04-22 06:36:52 +08:00
|
|
|
|
|
|
|
case ISD::SETO: return "setcc:seto";
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::SETUO: return "setcc:setuo";
|
|
|
|
case ISD::SETUEQ: return "setcc:setue";
|
|
|
|
case ISD::SETUGT: return "setcc:setugt";
|
|
|
|
case ISD::SETUGE: return "setcc:setuge";
|
|
|
|
case ISD::SETULT: return "setcc:setult";
|
|
|
|
case ISD::SETULE: return "setcc:setule";
|
|
|
|
case ISD::SETUNE: return "setcc:setune";
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-11 07:25:25 +08:00
|
|
|
case ISD::SETEQ: return "setcc:seteq";
|
|
|
|
case ISD::SETGT: return "setcc:setgt";
|
|
|
|
case ISD::SETGE: return "setcc:setge";
|
|
|
|
case ISD::SETLT: return "setcc:setlt";
|
|
|
|
case ISD::SETLE: return "setcc:setle";
|
|
|
|
case ISD::SETNE: return "setcc:setne";
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2005-01-07 15:46:32 +08:00
|
|
|
|
|
|
|
void SDNode::dump() const {
|
|
|
|
std::cerr << (void*)this << ": ";
|
|
|
|
|
|
|
|
for (unsigned i = 0, e = getNumValues(); i != e; ++i) {
|
|
|
|
if (i) std::cerr << ",";
|
2005-01-15 15:14:32 +08:00
|
|
|
if (getValueType(i) == MVT::Other)
|
|
|
|
std::cerr << "ch";
|
|
|
|
else
|
|
|
|
std::cerr << MVT::getValueTypeString(getValueType(i));
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
2005-01-11 07:25:25 +08:00
|
|
|
std::cerr << " = " << getOperationName();
|
2005-01-07 15:46:32 +08:00
|
|
|
|
|
|
|
std::cerr << " ";
|
|
|
|
for (unsigned i = 0, e = getNumOperands(); i != e; ++i) {
|
|
|
|
if (i) std::cerr << ", ";
|
|
|
|
std::cerr << (void*)getOperand(i).Val;
|
|
|
|
if (unsigned RN = getOperand(i).ResNo)
|
|
|
|
std::cerr << ":" << RN;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (const ConstantSDNode *CSDN = dyn_cast<ConstantSDNode>(this)) {
|
|
|
|
std::cerr << "<" << CSDN->getValue() << ">";
|
|
|
|
} else if (const ConstantFPSDNode *CSDN = dyn_cast<ConstantFPSDNode>(this)) {
|
|
|
|
std::cerr << "<" << CSDN->getValue() << ">";
|
2005-04-22 06:36:52 +08:00
|
|
|
} else if (const GlobalAddressSDNode *GADN =
|
2005-01-07 15:46:32 +08:00
|
|
|
dyn_cast<GlobalAddressSDNode>(this)) {
|
|
|
|
std::cerr << "<";
|
|
|
|
WriteAsOperand(std::cerr, GADN->getGlobal()) << ">";
|
2005-04-22 12:01:18 +08:00
|
|
|
} else if (const FrameIndexSDNode *FIDN = dyn_cast<FrameIndexSDNode>(this)) {
|
2005-01-07 15:46:32 +08:00
|
|
|
std::cerr << "<" << FIDN->getIndex() << ">";
|
|
|
|
} else if (const ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(this)){
|
|
|
|
std::cerr << "<" << CP->getIndex() << ">";
|
2005-04-22 12:01:18 +08:00
|
|
|
} else if (const BasicBlockSDNode *BBDN = dyn_cast<BasicBlockSDNode>(this)) {
|
2005-01-07 15:46:32 +08:00
|
|
|
std::cerr << "<";
|
|
|
|
const Value *LBB = (const Value*)BBDN->getBasicBlock()->getBasicBlock();
|
|
|
|
if (LBB)
|
|
|
|
std::cerr << LBB->getName() << " ";
|
|
|
|
std::cerr << (const void*)BBDN->getBasicBlock() << ">";
|
2005-01-14 04:50:02 +08:00
|
|
|
} else if (const RegSDNode *C2V = dyn_cast<RegSDNode>(this)) {
|
2005-01-07 15:46:32 +08:00
|
|
|
std::cerr << "<reg #" << C2V->getReg() << ">";
|
|
|
|
} else if (const ExternalSymbolSDNode *ES =
|
|
|
|
dyn_cast<ExternalSymbolSDNode>(this)) {
|
|
|
|
std::cerr << "'" << ES->getSymbol() << "'";
|
2005-01-16 05:11:37 +08:00
|
|
|
} else if (const MVTSDNode *M = dyn_cast<MVTSDNode>(this)) {
|
|
|
|
std::cerr << " - Ty = " << MVT::getValueTypeString(M->getExtraValueType());
|
2005-05-09 12:08:27 +08:00
|
|
|
} else if (const SrcValueSDNode *M = dyn_cast<SrcValueSDNode>(this)) {
|
|
|
|
if (M->getValue())
|
|
|
|
std::cerr << "<" << M->getValue() << ":" << M->getOffset() << ">";
|
|
|
|
else
|
|
|
|
std::cerr << "<null:" << M->getOffset() << ">";
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2005-01-10 04:38:33 +08:00
|
|
|
static void DumpNodes(SDNode *N, unsigned indent) {
|
|
|
|
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i)
|
|
|
|
if (N->getOperand(i).Val->hasOneUse())
|
|
|
|
DumpNodes(N->getOperand(i).Val, indent+2);
|
|
|
|
else
|
|
|
|
std::cerr << "\n" << std::string(indent+2, ' ')
|
|
|
|
<< (void*)N->getOperand(i).Val << ": <multiple use>";
|
2005-04-22 06:36:52 +08:00
|
|
|
|
2005-01-10 04:38:33 +08:00
|
|
|
|
|
|
|
std::cerr << "\n" << std::string(indent, ' ');
|
|
|
|
N->dump();
|
|
|
|
}
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
void SelectionDAG::dump() const {
|
|
|
|
std::cerr << "SelectionDAG has " << AllNodes.size() << " nodes:";
|
2005-01-10 04:26:36 +08:00
|
|
|
std::vector<SDNode*> Nodes(AllNodes);
|
|
|
|
std::sort(Nodes.begin(), Nodes.end());
|
|
|
|
|
|
|
|
for (unsigned i = 0, e = Nodes.size(); i != e; ++i) {
|
2005-01-10 04:38:33 +08:00
|
|
|
if (!Nodes[i]->hasOneUse() && Nodes[i] != getRoot().Val)
|
|
|
|
DumpNodes(Nodes[i], 2);
|
2005-01-07 15:46:32 +08:00
|
|
|
}
|
2005-01-10 04:38:33 +08:00
|
|
|
|
|
|
|
DumpNodes(getRoot().Val, 2);
|
|
|
|
|
2005-01-07 15:46:32 +08:00
|
|
|
std::cerr << "\n\n";
|
|
|
|
}
|
|
|
|
|