maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/lib/CodeGen/GlobalISel/MachineIRBuilder.cpp

699 lines
25 KiB
C++
Raw Normal View History

[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
//===-- llvm/CodeGen/GlobalISel/MachineIRBuilder.cpp - MIBuilder--*- C++ -*-==//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
/// \file
/// This file implements the MachineIRBuidler class.
//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
#include "llvm/CodeGen/MachineRegisterInfo.h"
GlobalISel: support debug intrinsics. The translation scheme is mostly cribbed from FastISel, and it's not entirely convincing semantically. But it does seem to work in the common cases and allow variables to be printed so it can't be all wrong. llvm-svn: 293228
2017-01-27 07:39:14 +08:00
#include "llvm/IR/DebugInfo.h"
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
#include "llvm/Target/TargetInstrInfo.h"
[Target] Add a helper function to check if an opcode is invalid after isel. llvm-svn: 260590
2016-02-12 05:16:56 +08:00
#include "llvm/Target/TargetOpcodes.h"
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
#include "llvm/Target/TargetSubtargetInfo.h"
using namespace llvm;
[MachineIRBuilder] Rename the setter of MF for consistency with the getter. llvm-svn: 263262
2016-03-12 01:27:51 +08:00
void MachineIRBuilder::setMF(MachineFunction &MF) {
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
this->MF = &MF;
this->MBB = nullptr;
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
this->MRI = &MF.getRegInfo();
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
this->TII = MF.getSubtarget().getInstrInfo();
this->DL = DebugLoc();
GlobalISel: simplify MachineIRBuilder interface. MachineIRBuilder had weird before/after and beginning/end flags for the insert point. Unfortunately the non-default means that instructions will be inserted in reverse order which is almost never what anyone wants. Really, I think we just want (like IRBuilder has) the ability to insert at any C++ iterator-style point (i.e. before any instruction or before MBB.end()). So this fixes MIRBuilders to behave like IRBuilders in this respect. llvm-svn: 288980
2016-12-08 05:05:38 +08:00
this->II = MachineBasicBlock::iterator();
GlobalISel: perform multi-step legalization llvm-svn: 279758
2016-08-26 01:37:32 +08:00
this->InsertedInstr = nullptr;
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
}
GlobalISel: simplify MachineIRBuilder interface. MachineIRBuilder had weird before/after and beginning/end flags for the insert point. Unfortunately the non-default means that instructions will be inserted in reverse order which is almost never what anyone wants. Really, I think we just want (like IRBuilder has) the ability to insert at any C++ iterator-style point (i.e. before any instruction or before MBB.end()). So this fixes MIRBuilders to behave like IRBuilders in this respect. llvm-svn: 288980
2016-12-08 05:05:38 +08:00
void MachineIRBuilder::setMBB(MachineBasicBlock &MBB) {
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
this->MBB = &MBB;
GlobalISel: simplify MachineIRBuilder interface. MachineIRBuilder had weird before/after and beginning/end flags for the insert point. Unfortunately the non-default means that instructions will be inserted in reverse order which is almost never what anyone wants. Really, I think we just want (like IRBuilder has) the ability to insert at any C++ iterator-style point (i.e. before any instruction or before MBB.end()). So this fixes MIRBuilders to behave like IRBuilders in this respect. llvm-svn: 288980
2016-12-08 05:05:38 +08:00
this->II = MBB.end();
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
assert(&getMF() == MBB.getParent() &&
"Basic block is in a different function");
}
GlobalISel: simplify MachineIRBuilder interface. MachineIRBuilder had weird before/after and beginning/end flags for the insert point. Unfortunately the non-default means that instructions will be inserted in reverse order which is almost never what anyone wants. Really, I think we just want (like IRBuilder has) the ability to insert at any C++ iterator-style point (i.e. before any instruction or before MBB.end()). So this fixes MIRBuilders to behave like IRBuilders in this respect. llvm-svn: 288980
2016-12-08 05:05:38 +08:00
void MachineIRBuilder::setInstr(MachineInstr &MI) {
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
assert(MI.getParent() && "Instruction is not part of a basic block");
[MachineIRBuilder] Rename the setter for MBB for consistency with the getter. llvm-svn: 263261
2016-03-12 01:27:47 +08:00
setMBB(*MI.getParent());
GlobalISel: simplify MachineIRBuilder interface. MachineIRBuilder had weird before/after and beginning/end flags for the insert point. Unfortunately the non-default means that instructions will be inserted in reverse order which is almost never what anyone wants. Really, I think we just want (like IRBuilder has) the ability to insert at any C++ iterator-style point (i.e. before any instruction or before MBB.end()). So this fixes MIRBuilders to behave like IRBuilders in this respect. llvm-svn: 288980
2016-12-08 05:05:38 +08:00
this->II = MI.getIterator();
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
}
GlobalISel: simplify MachineIRBuilder interface. MachineIRBuilder had weird before/after and beginning/end flags for the insert point. Unfortunately the non-default means that instructions will be inserted in reverse order which is almost never what anyone wants. Really, I think we just want (like IRBuilder has) the ability to insert at any C++ iterator-style point (i.e. before any instruction or before MBB.end()). So this fixes MIRBuilders to behave like IRBuilders in this respect. llvm-svn: 288980
2016-12-08 05:05:38 +08:00
void MachineIRBuilder::setInsertPt(MachineBasicBlock &MBB,
MachineBasicBlock::iterator II) {
assert(MBB.getParent() == &getMF() &&
"Basic block is in a different function");
this->MBB = &MBB;
this->II = II;
[GlobalISel] Add a MachineIRBuilder class. Helper class to build machine instrs. This is a higher abstraction than MachineInstrBuilder. llvm-svn: 260547
2016-02-12 01:44:59 +08:00
}
GlobalISel: perform multi-step legalization llvm-svn: 279758
2016-08-26 01:37:32 +08:00
void MachineIRBuilder::recordInsertions(
std::function<void(MachineInstr *)> Inserted) {
Apply clang-tidy's performance-unnecessary-value-param to LLVM. With some minor manual fixes for using function_ref instead of std::function. No functional change intended. llvm-svn: 291904
2017-01-13 22:39:03 +08:00
InsertedInstr = std::move(Inserted);
GlobalISel: perform multi-step legalization llvm-svn: 279758
2016-08-26 01:37:32 +08:00
}
void MachineIRBuilder::stopRecordingInsertions() {
InsertedInstr = nullptr;
}
[MachineIRBuilder] Rework buildInstr API to maximize code reuse. llvm-svn: 263264
2016-03-12 01:27:58 +08:00
//------------------------------------------------------------------------------
// Build instruction variants.
//------------------------------------------------------------------------------
GlobalISel: give MachineInstrBuilder a uniform interface. NFC. Instead of an ad-hoc collection of "buildInstr" functions with varying numbers of registers, this uses variadic templates to provide for as many regs as needed! Also make IRtranslator use new "buildBr" function instead of some weird generic one that no-one else would really use. llvm-svn: 276762
2016-07-27 00:45:26 +08:00
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildInstr(unsigned Opcode) {
GlobalISel: handle stack-based parameters on AArch64. llvm-svn: 282153
2016-09-22 21:49:25 +08:00
return insertInstr(buildInstrNoInsert(Opcode));
}
MachineInstrBuilder MachineIRBuilder::buildInstrNoInsert(unsigned Opcode) {
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
MachineInstrBuilder MIB = BuildMI(getMF(), DL, getTII().get(Opcode));
GlobalISel: handle stack-based parameters on AArch64. llvm-svn: 282153
2016-09-22 21:49:25 +08:00
return MIB;
}
MachineInstrBuilder MachineIRBuilder::insertInstr(MachineInstrBuilder MIB) {
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
getMBB().insert(getInsertPt(), MIB);
GlobalISel: perform multi-step legalization llvm-svn: 279758
2016-08-26 01:37:32 +08:00
if (InsertedInstr)
InsertedInstr(MIB);
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
return MIB;
[GlobalISel] Teach the IRTranslator how to lower returns. llvm-svn: 260562
2016-02-12 02:53:28 +08:00
}
GlobalISel: support debug intrinsics. The translation scheme is mostly cribbed from FastISel, and it's not entirely convincing semantically. But it does seem to work in the common cases and allow variables to be printed so it can't be all wrong. llvm-svn: 293228
2017-01-27 07:39:14 +08:00
MachineInstrBuilder MachineIRBuilder::buildDirectDbgValue(
unsigned Reg, const MDNode *Variable, const MDNode *Expr) {
assert(isa<DILocalVariable>(Variable) && "not a variable");
assert(cast<DIExpression>(Expr)->isValid() && "not an expression");
assert(cast<DILocalVariable>(Variable)->isValidLocationForIntrinsic(DL) &&
"Expected inlined-at fields to agree");
return buildInstr(TargetOpcode::DBG_VALUE)
.addReg(Reg, RegState::Debug)
.addReg(0, RegState::Debug)
.addMetadata(Variable)
.addMetadata(Expr);
}
MachineInstrBuilder MachineIRBuilder::buildIndirectDbgValue(
unsigned Reg, unsigned Offset, const MDNode *Variable, const MDNode *Expr) {
assert(isa<DILocalVariable>(Variable) && "not a variable");
assert(cast<DIExpression>(Expr)->isValid() && "not an expression");
assert(cast<DILocalVariable>(Variable)->isValidLocationForIntrinsic(DL) &&
"Expected inlined-at fields to agree");
return buildInstr(TargetOpcode::DBG_VALUE)
.addReg(Reg, RegState::Debug)
.addImm(Offset)
.addMetadata(Variable)
.addMetadata(Expr);
}
MachineInstrBuilder MachineIRBuilder::buildFIDbgValue(int FI,
const MDNode *Variable,
const MDNode *Expr) {
assert(isa<DILocalVariable>(Variable) && "not a variable");
assert(cast<DIExpression>(Expr)->isValid() && "not an expression");
assert(cast<DILocalVariable>(Variable)->isValidLocationForIntrinsic(DL) &&
"Expected inlined-at fields to agree");
return buildInstr(TargetOpcode::DBG_VALUE)
.addFrameIndex(FI)
.addImm(0)
.addMetadata(Variable)
.addMetadata(Expr);
}
MachineInstrBuilder MachineIRBuilder::buildConstDbgValue(const Constant &C,
unsigned Offset,
const MDNode *Variable,
const MDNode *Expr) {
assert(isa<DILocalVariable>(Variable) && "not a variable");
assert(cast<DIExpression>(Expr)->isValid() && "not an expression");
assert(cast<DILocalVariable>(Variable)->isValidLocationForIntrinsic(DL) &&
"Expected inlined-at fields to agree");
auto MIB = buildInstr(TargetOpcode::DBG_VALUE);
if (auto *CI = dyn_cast<ConstantInt>(&C)) {
if (CI->getBitWidth() > 64)
MIB.addCImm(CI);
else
MIB.addImm(CI->getZExtValue());
[GlobalISel] Emit DBG_VALUE %noreg for non-int/fp constant values. When a dbg_value has a constant operand that isn't representable in MI, there isn't much we can do. Use %noreg (0) for those situations. This matches the SelectionDAG behavior. llvm-svn: 297200
2017-03-08 04:34:20 +08:00
} else if (auto *CFP = dyn_cast<ConstantFP>(&C)) {
[GlobalISel] Slightly clean up DBG_VALUE FP build code. I messed up my rebases leading to r297200, and ended up with stale (but working) code. Fix it. llvm-svn: 297205
2017-03-08 04:52:57 +08:00
MIB.addFPImm(CFP);
[GlobalISel] Emit DBG_VALUE %noreg for non-int/fp constant values. When a dbg_value has a constant operand that isn't representable in MI, there isn't much we can do. Use %noreg (0) for those situations. This matches the SelectionDAG behavior. llvm-svn: 297200
2017-03-08 04:34:20 +08:00
} else {
// Insert %noreg if we didn't find a usable constant and had to drop it.
MIB.addReg(0U);
}
GlobalISel: support debug intrinsics. The translation scheme is mostly cribbed from FastISel, and it's not entirely convincing semantically. But it does seem to work in the common cases and allow variables to be printed so it can't be all wrong. llvm-svn: 293228
2017-01-27 07:39:14 +08:00
return MIB.addImm(Offset).addMetadata(Variable).addMetadata(Expr);
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildFrameIndex(unsigned Res, int Idx) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Res).isPointer() && "invalid operand type");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_FRAME_INDEX)
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
.addDef(Res)
.addFrameIndex(Idx);
GlobalISel: implement alloca instruction llvm-svn: 276433
2016-07-23 00:59:52 +08:00
}
GlobalISel: implement legalization pass, with just one transformation. This adds the actual MachineLegalizeHelper to do the work and a trivial pass wrapper that legalizes all instructions in a MachineFunction. Currently the only transformation supported is splitting up a vector G_ADD into one acting on smaller vectors. llvm-svn: 276461
2016-07-23 04:03:43 +08:00
GlobalISel: support translation of global addresses. llvm-svn: 281207
2016-09-12 20:10:41 +08:00
MachineInstrBuilder MachineIRBuilder::buildGlobalValue(unsigned Res,
const GlobalValue *GV) {
assert(MRI->getType(Res).isPointer() && "invalid operand type");
assert(MRI->getType(Res).getAddressSpace() ==
GV->getType()->getAddressSpace() &&
"address space mismatch");
return buildInstr(TargetOpcode::G_GLOBAL_VALUE)
.addDef(Res)
.addGlobalAddress(GV);
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildAdd(unsigned Res, unsigned Op0,
unsigned Op1) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert((MRI->getType(Res).isScalar() || MRI->getType(Res).isVector()) &&
"invalid operand type");
assert(MRI->getType(Res) == MRI->getType(Op0) &&
MRI->getType(Res) == MRI->getType(Op1) && "type mismatch");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_ADD)
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
.addDef(Res)
.addUse(Op0)
.addUse(Op1);
GlobalISel: implement legalization pass, with just one transformation. This adds the actual MachineLegalizeHelper to do the work and a trivial pass wrapper that legalizes all instructions in a MachineFunction. Currently the only transformation supported is splitting up a vector G_ADD into one acting on smaller vectors. llvm-svn: 276461
2016-07-23 04:03:43 +08:00
}
GlobalISel: translate GEP instructions. Unlike SDag, we use a separate G_GEP instruction (much simplified, only taking a single byte offset) to preserve the pointer type information through selection. llvm-svn: 281205
2016-09-12 19:20:22 +08:00
MachineInstrBuilder MachineIRBuilder::buildGEP(unsigned Res, unsigned Op0,
unsigned Op1) {
assert(MRI->getType(Res).isPointer() &&
MRI->getType(Res) == MRI->getType(Op0) && "type mismatch");
assert(MRI->getType(Op1).isScalar() && "invalid offset type");
return buildInstr(TargetOpcode::G_GEP)
.addDef(Res)
.addUse(Op0)
.addUse(Op1);
}
[globalisel][legalizer] G_LOAD/G_STORE NarrowScalar should not emit G_GEP x, 0. Summary: When legalizing G_LOAD/G_STORE using NarrowScalar, we should avoid emitting %0 = G_CONSTANT ty 0 %1 = G_GEP %x, %0 since it's cheaper to not emit the redundant instructions than it is to fold them away later. Reviewers: qcolombet, t.p.northover, ab, rovka, aditya_nandakumar, kristof.beyls Reviewed By: qcolombet Subscribers: javed.absar, llvm-commits, igorb Differential Revision: https://reviews.llvm.org/D32746 llvm-svn: 305340
2017-06-14 07:42:32 +08:00
Optional<MachineInstrBuilder>
MachineIRBuilder::materializeGEP(unsigned &Res, unsigned Op0,
const LLT &ValueTy, uint64_t Value) {
assert(Res == 0 && "Res is a result argument");
assert(ValueTy.isScalar() && "invalid offset type");
if (Value == 0) {
Res = Op0;
return None;
}
Res = MRI->createGenericVirtualRegister(MRI->getType(Op0));
unsigned TmpReg = MRI->createGenericVirtualRegister(ValueTy);
buildConstant(TmpReg, Value);
return buildGEP(Res, Op0, TmpReg);
}
GlobalISel: introduce G_PTR_MASK to simplify alloca handling. This instruction clears the low bits of a pointer without requiring (possibly dodgy if pointers aren't ints) conversions to and from an integer. Since (as far as I'm aware) all masks are statically known, the instruction takes an immediate operand rather than a register to specify the mask. llvm-svn: 295103
2017-02-15 04:56:18 +08:00
MachineInstrBuilder MachineIRBuilder::buildPtrMask(unsigned Res, unsigned Op0,
uint32_t NumBits) {
assert(MRI->getType(Res).isPointer() &&
MRI->getType(Res) == MRI->getType(Op0) && "type mismatch");
return buildInstr(TargetOpcode::G_PTR_MASK)
.addDef(Res)
.addUse(Op0)
.addImm(NumBits);
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildSub(unsigned Res, unsigned Op0,
unsigned Op1) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert((MRI->getType(Res).isScalar() || MRI->getType(Res).isVector()) &&
"invalid operand type");
assert(MRI->getType(Res) == MRI->getType(Op0) &&
MRI->getType(Res) == MRI->getType(Op1) && "type mismatch");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_SUB)
GlobalISel: legalize sdiv and srem operations. llvm-svn: 279842
2016-08-27 01:46:13 +08:00
.addDef(Res)
.addUse(Op0)
.addUse(Op1);
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildMul(unsigned Res, unsigned Op0,
unsigned Op1) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert((MRI->getType(Res).isScalar() || MRI->getType(Res).isVector()) &&
"invalid operand type");
assert(MRI->getType(Res) == MRI->getType(Op0) &&
MRI->getType(Res) == MRI->getType(Op1) && "type mismatch");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_MUL)
GlobalISel: legalize sdiv and srem operations. llvm-svn: 279842
2016-08-27 01:46:13 +08:00
.addDef(Res)
.addUse(Op0)
.addUse(Op1);
}
GlobalISel: translate dynamic alloca instructions. llvm-svn: 294022
2017-02-04 02:22:45 +08:00
MachineInstrBuilder MachineIRBuilder::buildAnd(unsigned Res, unsigned Op0,
unsigned Op1) {
assert((MRI->getType(Res).isScalar() || MRI->getType(Res).isVector()) &&
"invalid operand type");
assert(MRI->getType(Res) == MRI->getType(Op0) &&
MRI->getType(Res) == MRI->getType(Op1) && "type mismatch");
return buildInstr(TargetOpcode::G_AND)
.addDef(Res)
.addUse(Op0)
.addUse(Op1);
}
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
MachineInstrBuilder MachineIRBuilder::buildBr(MachineBasicBlock &Dest) {
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_BR).addMBB(&Dest);
GlobalISel: give MachineInstrBuilder a uniform interface. NFC. Instead of an ad-hoc collection of "buildInstr" functions with varying numbers of registers, this uses variadic templates to provide for as many regs as needed! Also make IRtranslator use new "buildBr" function instead of some weird generic one that no-one else would really use. llvm-svn: 276762
2016-07-27 00:45:26 +08:00
}
[GlobalISel] Add support for indirectbr Differential Revision: https://reviews.llvm.org/D28079 llvm-svn: 293470
2017-01-30 17:13:18 +08:00
MachineInstrBuilder MachineIRBuilder::buildBrIndirect(unsigned Tgt) {
GlobalISel: add some more sanity-checking to MachineInstrBuilder. NFC. llvm-svn: 306481
2017-06-28 06:45:35 +08:00
assert(MRI->getType(Tgt).isPointer() && "invalid branch destination");
[GlobalISel] Add support for indirectbr Differential Revision: https://reviews.llvm.org/D28079 llvm-svn: 293470
2017-01-30 17:13:18 +08:00
return buildInstr(TargetOpcode::G_BRINDIRECT).addUse(Tgt);
}
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
MachineInstrBuilder MachineIRBuilder::buildCopy(unsigned Res, unsigned Op) {
GlobalISel: verify that a COPY is trivial when created. Without this check, COPY instructions can actually be one of the generic casts in disguise. That's confusing and bad. At some point during ISel this restriction has to be relaxed since the fully selected instructions will usually use COPY for those purposes. Right now I think it's possible that relaxation occurs during RegBankSelect (hence the change there). I'm not convinced that's where it belongs long-term though. llvm-svn: 306470
2017-06-28 05:41:40 +08:00
assert(MRI->getType(Res) == LLT() || MRI->getType(Op) == LLT() ||
MRI->getType(Res) == MRI->getType(Op));
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
return buildInstr(TargetOpcode::COPY).addDef(Res).addUse(Op);
GlobalISel: add specialized buildCopy function to MachineInstrBuilder. NFC. llvm-svn: 276763
2016-07-27 00:45:30 +08:00
}
GlobalISel: make G_CONSTANT take a ConstantInt rather than int64_t. This makes it more similar to the floating-point constant, and also allows for larger constants to be translated later. There's no real functional change in this patch though, just syntax updates. llvm-svn: 288712
2016-12-06 05:47:07 +08:00
MachineInstrBuilder MachineIRBuilder::buildConstant(unsigned Res,
const ConstantInt &Val) {
LLT Ty = MRI->getType(Res);
Add missing parens in assert. Summary: Add missing parens in assert, which warn in GCC. Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D27448 llvm-svn: 288792
2016-12-06 18:14:36 +08:00
assert((Ty.isScalar() || Ty.isPointer()) && "invalid operand type");
GlobalISel: make G_CONSTANT take a ConstantInt rather than int64_t. This makes it more similar to the floating-point constant, and also allows for larger constants to be translated later. There's no real functional change in this patch though, just syntax updates. llvm-svn: 288712
2016-12-06 05:47:07 +08:00
const ConstantInt *NewVal = &Val;
if (Ty.getSizeInBits() != Val.getBitWidth())
NewVal = ConstantInt::get(MF->getFunction()->getContext(),
Val.getValue().sextOrTrunc(Ty.getSizeInBits()));
return buildInstr(TargetOpcode::G_CONSTANT).addDef(Res).addCImm(NewVal);
}
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
GlobalISel: make G_CONSTANT take a ConstantInt rather than int64_t. This makes it more similar to the floating-point constant, and also allows for larger constants to be translated later. There's no real functional change in this patch though, just syntax updates. llvm-svn: 288712
2016-12-06 05:47:07 +08:00
MachineInstrBuilder MachineIRBuilder::buildConstant(unsigned Res,
int64_t Val) {
auto IntN = IntegerType::get(MF->getFunction()->getContext(),
MRI->getType(Res).getSizeInBits());
ConstantInt *CI = ConstantInt::get(IntN, Val, true);
return buildConstant(Res, *CI);
GlobalISel: implement narrowing for G_ADD. llvm-svn: 277769
2016-08-05 04:54:13 +08:00
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildFConstant(unsigned Res,
const ConstantFP &Val) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Res).isScalar() && "invalid operand type");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_FCONSTANT).addDef(Res).addFPImm(&Val);
GlobalISel: translate floating-point constants llvm-svn: 279311
2016-08-20 04:09:15 +08:00
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildBrCond(unsigned Tst,
GlobalISel: add generic conditional branch. Just the basic equivalent to DAG's condbr for now, we'll get to things like br_cc when we start doing more legalization. llvm-svn: 277184
2016-07-30 01:58:00 +08:00
MachineBasicBlock &Dest) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Tst).isScalar() && "invalid operand type");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_BRCOND).addUse(Tst).addMBB(&Dest);
GlobalISel: add generic conditional branch. Just the basic equivalent to DAG's condbr for now, we'll get to things like br_cc when we start doing more legalization. llvm-svn: 277184
2016-07-30 01:58:00 +08:00
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildLoad(unsigned Res, unsigned Addr,
MachineMemOperand &MMO) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Res).isValid() && "invalid operand type");
assert(MRI->getType(Addr).isPointer() && "invalid operand type");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_LOAD)
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
.addDef(Res)
.addUse(Addr)
.addMemOperand(&MMO);
GlobalISel: add generic load and store instructions. Pretty straightforward, the only oddity is the MachineMemOperand (which it's surprisingly difficult to share code for). llvm-svn: 276799
2016-07-27 04:23:26 +08:00
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildStore(unsigned Val, unsigned Addr,
MachineMemOperand &MMO) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Val).isValid() && "invalid operand type");
assert(MRI->getType(Addr).isPointer() && "invalid operand type");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_STORE)
CodeGen: improve MachineInstrBuilder & MachineIRBuilder interface For MachineInstrBuilder, having to manually use RegState::Define is ugly and makes register definitions clunkier than they need to be, so this adds two convenience functions: addDef and addUse. For MachineIRBuilder, we want to avoid BuildMI's first-reg-is-def rule because it's hidden away and causes bugs. So this patch switches buildInstr to returning a MachineInstrBuilder and adding *all* operands via addDef/addUse. NFC. llvm-svn: 277176
2016-07-30 01:43:52 +08:00
.addUse(Val)
.addUse(Addr)
.addMemOperand(&MMO);
GlobalISel: add generic load and store instructions. Pretty straightforward, the only oddity is the MachineMemOperand (which it's surprisingly difficult to share code for). llvm-svn: 276799
2016-07-27 04:23:26 +08:00
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildUAdde(unsigned Res,
unsigned CarryOut,
unsigned Op0, unsigned Op1,
unsigned CarryIn) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Res).isScalar() && "invalid operand type");
assert(MRI->getType(Res) == MRI->getType(Op0) &&
MRI->getType(Res) == MRI->getType(Op1) && "type mismatch");
assert(MRI->getType(CarryOut).isScalar() && "invalid operand type");
assert(MRI->getType(CarryOut) == MRI->getType(CarryIn) && "type mismatch");
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_UADDE)
GlobalISel: implement narrowing for G_ADD. llvm-svn: 277769
2016-08-05 04:54:13 +08:00
.addDef(Res)
.addDef(CarryOut)
.addUse(Op0)
.addUse(Op1)
.addUse(CarryIn);
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildAnyExt(unsigned Res, unsigned Op) {
validateTruncExt(Res, Op, true);
return buildInstr(TargetOpcode::G_ANYEXT).addDef(Res).addUse(Op);
GlobalISel: add code to widen scalar G_ADD llvm-svn: 277747
2016-08-05 02:35:11 +08:00
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildSExt(unsigned Res, unsigned Op) {
validateTruncExt(Res, Op, true);
return buildInstr(TargetOpcode::G_SEXT).addDef(Res).addUse(Op);
GlobalISel: legalize integer comparisons on AArch64. Next step is doing both legalizations at the same time! Marvel at GlobalISel's cunning. llvm-svn: 279566
2016-08-24 05:01:26 +08:00
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildZExt(unsigned Res, unsigned Op) {
validateTruncExt(Res, Op, true);
return buildInstr(TargetOpcode::G_ZEXT).addDef(Res).addUse(Op);
GlobalISel: legalize integer comparisons on AArch64. Next step is doing both legalizations at the same time! Marvel at GlobalISel's cunning. llvm-svn: 279566
2016-08-24 05:01:26 +08:00
}
GlobalISel: translate GEP instructions. Unlike SDag, we use a separate G_GEP instruction (much simplified, only taking a single byte offset) to preserve the pointer type information through selection. llvm-svn: 281205
2016-09-12 19:20:22 +08:00
MachineInstrBuilder MachineIRBuilder::buildSExtOrTrunc(unsigned Res,
unsigned Op) {
GlobalISel: add some more sanity-checking to MachineInstrBuilder. NFC. llvm-svn: 306481
2017-06-28 06:45:35 +08:00
assert(MRI->getType(Res).isScalar() || MRI->getType(Res).isVector());
assert(MRI->getType(Res).isScalar() == MRI->getType(Op).isScalar());
GlobalISel: translate GEP instructions. Unlike SDag, we use a separate G_GEP instruction (much simplified, only taking a single byte offset) to preserve the pointer type information through selection. llvm-svn: 281205
2016-09-12 19:20:22 +08:00
unsigned Opcode = TargetOpcode::COPY;
if (MRI->getType(Res).getSizeInBits() > MRI->getType(Op).getSizeInBits())
Opcode = TargetOpcode::G_SEXT;
else if (MRI->getType(Res).getSizeInBits() < MRI->getType(Op).getSizeInBits())
Opcode = TargetOpcode::G_TRUNC;
GlobalISel: add some more sanity-checking to MachineInstrBuilder. NFC. llvm-svn: 306481
2017-06-28 06:45:35 +08:00
else
assert(MRI->getType(Res) == MRI->getType(Op));
GlobalISel: translate GEP instructions. Unlike SDag, we use a separate G_GEP instruction (much simplified, only taking a single byte offset) to preserve the pointer type information through selection. llvm-svn: 281205
2016-09-12 19:20:22 +08:00
return buildInstr(Opcode).addDef(Res).addUse(Op);
}
GlobalISel: translate dynamic alloca instructions. llvm-svn: 294022
2017-02-04 02:22:45 +08:00
MachineInstrBuilder MachineIRBuilder::buildZExtOrTrunc(unsigned Res,
unsigned Op) {
GlobalISel: add some more sanity-checking to MachineInstrBuilder. NFC. llvm-svn: 306481
2017-06-28 06:45:35 +08:00
assert(MRI->getType(Res).isScalar() || MRI->getType(Res).isVector());
assert(MRI->getType(Res).isScalar() == MRI->getType(Op).isScalar());
GlobalISel: translate dynamic alloca instructions. llvm-svn: 294022
2017-02-04 02:22:45 +08:00
unsigned Opcode = TargetOpcode::COPY;
if (MRI->getType(Res).getSizeInBits() > MRI->getType(Op).getSizeInBits())
Opcode = TargetOpcode::G_ZEXT;
else if (MRI->getType(Res).getSizeInBits() < MRI->getType(Op).getSizeInBits())
Opcode = TargetOpcode::G_TRUNC;
GlobalISel: add some more sanity-checking to MachineInstrBuilder. NFC. llvm-svn: 306481
2017-06-28 06:45:35 +08:00
else
assert(MRI->getType(Res) == MRI->getType(Op));
GlobalISel: translate dynamic alloca instructions. llvm-svn: 294022
2017-02-04 02:22:45 +08:00
return buildInstr(Opcode).addDef(Res).addUse(Op);
}
GlobalISel: don't emit degenerate G_INSERT instructions. Before, we were producing G_INSERT instructions that were actually closer to a cast or even a COPY when both input and output sizes are the same. This doesn't really make sense and means that everything interpreting a G_INSERT also has to handle all these kinds of casts. So now we detect these degenerate cases and emit real casts instead. llvm-svn: 297051
2017-03-07 03:04:17 +08:00
MachineInstrBuilder MachineIRBuilder::buildCast(unsigned Dst, unsigned Src) {
LLT SrcTy = MRI->getType(Src);
LLT DstTy = MRI->getType(Dst);
if (SrcTy == DstTy)
return buildCopy(Dst, Src);
unsigned Opcode;
if (SrcTy.isPointer() && DstTy.isScalar())
Opcode = TargetOpcode::G_PTRTOINT;
else if (DstTy.isPointer() && SrcTy.isScalar())
Opcode = TargetOpcode::G_INTTOPTR;
else {
assert(!SrcTy.isPointer() && !DstTy.isPointer() && "n G_ADDRCAST yet");
Opcode = TargetOpcode::G_BITCAST;
}
return buildInstr(Opcode).addDef(Dst).addUse(Src);
}
GlobalISel: restrict G_EXTRACT instruction to just one operand. A bit more painful than G_INSERT because it was more widely used, but this should simplify the handling of extract operations in most locations. llvm-svn: 297100
2017-03-07 07:50:28 +08:00
MachineInstrBuilder MachineIRBuilder::buildExtract(unsigned Res, unsigned Src,
uint64_t Index) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
#ifndef NDEBUG
assert(MRI->getType(Src).isValid() && "invalid operand type");
GlobalISel: restrict G_EXTRACT instruction to just one operand. A bit more painful than G_INSERT because it was more widely used, but this should simplify the handling of extract operations in most locations. llvm-svn: 297100
2017-03-07 07:50:28 +08:00
assert(MRI->getType(Res).isValid() && "invalid operand type");
assert(Index + MRI->getType(Res).getSizeInBits() <=
MRI->getType(Src).getSizeInBits() &&
"extracting off end of register");
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
#endif
GlobalISel: restrict G_EXTRACT instruction to just one operand. A bit more painful than G_INSERT because it was more widely used, but this should simplify the handling of extract operations in most locations. llvm-svn: 297100
2017-03-07 07:50:28 +08:00
if (MRI->getType(Res).getSizeInBits() == MRI->getType(Src).getSizeInBits()) {
assert(Index == 0 && "insertion past the end of a register");
return buildCast(Res, Src);
}
GlobalISel: improve representation of G_SEQUENCE and G_EXTRACT First, make sure all types involved are represented, rather than being implicit from the register width. Second, canonicalize all types to scalar. These operations just act in bits and don't care about vectors. Also standardize spelling of Indices in the MachineIRBuilder (NFC here). llvm-svn: 279294
2016-08-20 02:32:14 +08:00
GlobalISel: restrict G_EXTRACT instruction to just one operand. A bit more painful than G_INSERT because it was more widely used, but this should simplify the handling of extract operations in most locations. llvm-svn: 297100
2017-03-07 07:50:28 +08:00
return buildInstr(TargetOpcode::G_EXTRACT)
.addDef(Res)
.addUse(Src)
.addImm(Index);
GlobalISel: implement legalization pass, with just one transformation. This adds the actual MachineLegalizeHelper to do the work and a trivial pass wrapper that legalizes all instructions in a MachineFunction. Currently the only transformation supported is splitting up a vector G_ADD into one acting on smaller vectors. llvm-svn: 276461
2016-07-23 04:03:43 +08:00
}
GlobalISel: convert buildSequence to use non-deprecated instructions. G_SEQUENCE is going away soon so as a first step the MachineIRBuilder needs to be taught how to emulate it with alternatives. We use G_MERGE_VALUES where possible, and a sequence of G_INSERTs if not. llvm-svn: 306119
2017-06-24 00:15:37 +08:00
void MachineIRBuilder::buildSequence(unsigned Res, ArrayRef<unsigned> Ops,
ArrayRef<uint64_t> Indices) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
#ifndef NDEBUG
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
assert(Ops.size() == Indices.size() && "incompatible args");
GlobalISel: improve representation of G_SEQUENCE and G_EXTRACT First, make sure all types involved are represented, rather than being implicit from the register width. Second, canonicalize all types to scalar. These operations just act in bits and don't care about vectors. Also standardize spelling of Indices in the MachineIRBuilder (NFC here). llvm-svn: 279294
2016-08-20 02:32:14 +08:00
assert(!Ops.empty() && "invalid trivial sequence");
GlobalISel: combine extracts & sequences created for legalization Legalization ends up creating many G_SEQUENCE/G_EXTRACT pairs which leads to inefficient codegen (even for -O0), so add a quick pass over the function to remove them again. llvm-svn: 280155
2016-08-31 04:51:25 +08:00
assert(std::is_sorted(Indices.begin(), Indices.end()) &&
"sequence offsets must be in ascending order");
GlobalISel: support overflow arithmetic intrinsics. Unsigned addition and subtraction can reuse the instructions created to legalize large width operations (i.e. both produce and consume a carry flag). Signed operations and multiplies get a dedicated op-with-overflow instruction. Once this is produced the two values are combined into a struct register (which will almost always be merged with a corresponding G_EXTRACT as part of legalization). llvm-svn: 279278
2016-08-20 01:17:06 +08:00
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Res).isValid() && "invalid operand type");
for (auto Op : Ops)
assert(MRI->getType(Op).isValid() && "invalid operand type");
#endif
GlobalISel: convert buildSequence to use non-deprecated instructions. G_SEQUENCE is going away soon so as a first step the MachineIRBuilder needs to be taught how to emulate it with alternatives. We use G_MERGE_VALUES where possible, and a sequence of G_INSERTs if not. llvm-svn: 306119
2017-06-24 00:15:37 +08:00
LLT ResTy = MRI->getType(Res);
LLT OpTy = MRI->getType(Ops[0]);
unsigned OpSize = OpTy.getSizeInBits();
bool MaybeMerge = true;
GlobalISel: support overflow arithmetic intrinsics. Unsigned addition and subtraction can reuse the instructions created to legalize large width operations (i.e. both produce and consume a carry flag). Signed operations and multiplies get a dedicated op-with-overflow instruction. Once this is produced the two values are combined into a struct register (which will almost always be merged with a corresponding G_EXTRACT as part of legalization). llvm-svn: 279278
2016-08-20 01:17:06 +08:00
for (unsigned i = 0; i < Ops.size(); ++i) {
GlobalISel: convert buildSequence to use non-deprecated instructions. G_SEQUENCE is going away soon so as a first step the MachineIRBuilder needs to be taught how to emulate it with alternatives. We use G_MERGE_VALUES where possible, and a sequence of G_INSERTs if not. llvm-svn: 306119
2017-06-24 00:15:37 +08:00
if (MRI->getType(Ops[i]) != OpTy || Indices[i] != i * OpSize) {
MaybeMerge = false;
break;
}
}
if (MaybeMerge && Ops.size() * OpSize == ResTy.getSizeInBits()) {
buildMerge(Res, Ops);
return;
}
unsigned ResIn = MRI->createGenericVirtualRegister(ResTy);
buildUndef(ResIn);
for (unsigned i = 0; i < Ops.size(); ++i) {
unsigned ResOut =
i + 1 == Ops.size() ? Res : MRI->createGenericVirtualRegister(ResTy);
buildInsert(ResOut, ResIn, Ops[i], Indices[i]);
ResIn = ResOut;
GlobalISel: support overflow arithmetic intrinsics. Unsigned addition and subtraction can reuse the instructions created to legalize large width operations (i.e. both produce and consume a carry flag). Signed operations and multiplies get a dedicated op-with-overflow instruction. Once this is produced the two values are combined into a struct register (which will almost always be merged with a corresponding G_EXTRACT as part of legalization). llvm-svn: 279278
2016-08-20 01:17:06 +08:00
}
GlobalISel: add merge/unmerge nodes for legalization. These are simplified variants of the current G_SEQUENCE and G_EXTRACT, which assume the individual parts will be contiguous, homogeneous, and occupy the entirity of the larger register. This makes reasoning about them much easer since you only have to look at the first register being merged and the result to know what the instruction is doing. I intend to gradually replace all uses of the more complicated sequence/extract with these (or single-element insert/extracts), and then remove the older variants. For now we start with legalization. llvm-svn: 296921
2017-03-04 06:46:09 +08:00
}
GlobalISel: add buildUndef method to MachineIRBuilder. NFC. llvm-svn: 297044
2017-03-07 02:36:40 +08:00
MachineInstrBuilder MachineIRBuilder::buildUndef(unsigned Res) {
return buildInstr(TargetOpcode::IMPLICIT_DEF).addDef(Res);
}
GlobalISel: add merge/unmerge nodes for legalization. These are simplified variants of the current G_SEQUENCE and G_EXTRACT, which assume the individual parts will be contiguous, homogeneous, and occupy the entirity of the larger register. This makes reasoning about them much easer since you only have to look at the first register being merged and the result to know what the instruction is doing. I intend to gradually replace all uses of the more complicated sequence/extract with these (or single-element insert/extracts), and then remove the older variants. For now we start with legalization. llvm-svn: 296921
2017-03-04 06:46:09 +08:00
MachineInstrBuilder MachineIRBuilder::buildMerge(unsigned Res,
ArrayRef<unsigned> Ops) {
#ifndef NDEBUG
assert(!Ops.empty() && "invalid trivial sequence");
LLT Ty = MRI->getType(Ops[0]);
for (auto Reg : Ops)
assert(MRI->getType(Reg) == Ty && "type mismatch in input list");
assert(Ops.size() * MRI->getType(Ops[0]).getSizeInBits() ==
MRI->getType(Res).getSizeInBits() &&
"input operands do not cover output register");
#endif
AArch64: legalize G_EXTRACT operations. This is the dual problem to legalizing G_INSERTs so most of the code and testing was cribbed from there. llvm-svn: 306328
2017-06-27 04:34:13 +08:00
if (Ops.size() == 1)
GlobalISel: verify that a COPY is trivial when created. Without this check, COPY instructions can actually be one of the generic casts in disguise. That's confusing and bad. At some point during ISel this restriction has to be relaxed since the fully selected instructions will usually use COPY for those purposes. Right now I think it's possible that relaxation occurs during RegBankSelect (hence the change there). I'm not convinced that's where it belongs long-term though. llvm-svn: 306470
2017-06-28 05:41:40 +08:00
return buildCast(Res, Ops[0]);
AArch64: legalize G_EXTRACT operations. This is the dual problem to legalizing G_INSERTs so most of the code and testing was cribbed from there. llvm-svn: 306328
2017-06-27 04:34:13 +08:00
GlobalISel: add merge/unmerge nodes for legalization. These are simplified variants of the current G_SEQUENCE and G_EXTRACT, which assume the individual parts will be contiguous, homogeneous, and occupy the entirity of the larger register. This makes reasoning about them much easer since you only have to look at the first register being merged and the result to know what the instruction is doing. I intend to gradually replace all uses of the more complicated sequence/extract with these (or single-element insert/extracts), and then remove the older variants. For now we start with legalization. llvm-svn: 296921
2017-03-04 06:46:09 +08:00
MachineInstrBuilder MIB = buildInstr(TargetOpcode::G_MERGE_VALUES);
MIB.addDef(Res);
for (unsigned i = 0; i < Ops.size(); ++i)
MIB.addUse(Ops[i]);
return MIB;
}
MachineInstrBuilder MachineIRBuilder::buildUnmerge(ArrayRef<unsigned> Res,
unsigned Op) {
#ifndef NDEBUG
assert(!Res.empty() && "invalid trivial sequence");
LLT Ty = MRI->getType(Res[0]);
for (auto Reg : Res)
assert(MRI->getType(Reg) == Ty && "type mismatch in input list");
assert(Res.size() * MRI->getType(Res[0]).getSizeInBits() ==
MRI->getType(Op).getSizeInBits() &&
"input operands do not cover output register");
#endif
MachineInstrBuilder MIB = buildInstr(TargetOpcode::G_UNMERGE_VALUES);
for (unsigned i = 0; i < Res.size(); ++i)
MIB.addDef(Res[i]);
MIB.addUse(Op);
return MIB;
GlobalISel: implement legalization pass, with just one transformation. This adds the actual MachineLegalizeHelper to do the work and a trivial pass wrapper that legalizes all instructions in a MachineFunction. Currently the only transformation supported is splitting up a vector G_ADD into one acting on smaller vectors. llvm-svn: 276461
2016-07-23 04:03:43 +08:00
}
GlobalISel: support translation of intrinsic calls. These come in two variants for now: G_INTRINSIC and G_INTRINSIC_W_SIDE_EFFECTS. We may decide to split the latter up with finer-grained restrictions later, if necessary. llvm-svn: 277224
2016-07-30 06:32:36 +08:00
GlobalISel: constrain G_INSERT to inserting just one value per instruction. It's much easier to reason about single-value inserts and no-one was actually using the variadic variants before. llvm-svn: 296923
2017-03-04 07:05:47 +08:00
MachineInstrBuilder MachineIRBuilder::buildInsert(unsigned Res, unsigned Src,
unsigned Op, unsigned Index) {
GlobalISel: add some more sanity-checking to MachineInstrBuilder. NFC. llvm-svn: 306481
2017-06-28 06:45:35 +08:00
assert(Index + MRI->getType(Op).getSizeInBits() <=
MRI->getType(Res).getSizeInBits() &&
"insertion past the end of a register");
GlobalISel: don't emit degenerate G_INSERT instructions. Before, we were producing G_INSERT instructions that were actually closer to a cast or even a COPY when both input and output sizes are the same. This doesn't really make sense and means that everything interpreting a G_INSERT also has to handle all these kinds of casts. So now we detect these degenerate cases and emit real casts instead. llvm-svn: 297051
2017-03-07 03:04:17 +08:00
if (MRI->getType(Res).getSizeInBits() == MRI->getType(Op).getSizeInBits()) {
return buildCast(Res, Op);
}
GlobalISel: constrain G_INSERT to inserting just one value per instruction. It's much easier to reason about single-value inserts and no-one was actually using the variadic variants before. llvm-svn: 296923
2017-03-04 07:05:47 +08:00
return buildInstr(TargetOpcode::G_INSERT)
.addDef(Res)
.addUse(Src)
.addUse(Op)
.addImm(Index);
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildIntrinsic(Intrinsic::ID ID,
GlobalISel: support translation of intrinsic calls. These come in two variants for now: G_INTRINSIC and G_INTRINSIC_W_SIDE_EFFECTS. We may decide to split the latter up with finer-grained restrictions later, if necessary. llvm-svn: 277224
2016-07-30 06:32:36 +08:00
unsigned Res,
bool HasSideEffects) {
auto MIB =
buildInstr(HasSideEffects ? TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
: TargetOpcode::G_INTRINSIC);
GlobalISel: support translation of intrinsic calls. These come in two variants for now: G_INTRINSIC and G_INTRINSIC_W_SIDE_EFFECTS. We may decide to split the latter up with finer-grained restrictions later, if necessary. llvm-svn: 277224
2016-07-30 06:32:36 +08:00
if (Res)
MIB.addDef(Res);
MIB.addIntrinsicID(ID);
return MIB;
}
GlobalISel: add code to widen scalar G_ADD llvm-svn: 277747
2016-08-05 02:35:11 +08:00
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildTrunc(unsigned Res, unsigned Op) {
validateTruncExt(Res, Op, false);
return buildInstr(TargetOpcode::G_TRUNC).addDef(Res).addUse(Op);
GlobalISel: add code to widen scalar G_ADD llvm-svn: 277747
2016-08-05 02:35:11 +08:00
}
GlobalISel: support irtranslation of icmp instructions. llvm-svn: 278969
2016-08-18 04:25:25 +08:00
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildFPTrunc(unsigned Res, unsigned Op) {
validateTruncExt(Res, Op, false);
return buildInstr(TargetOpcode::G_FPTRUNC).addDef(Res).addUse(Op);
GlobalISel: support legalization of G_FCONSTANTs llvm-svn: 279341
2016-08-20 06:40:08 +08:00
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildICmp(CmpInst::Predicate Pred,
GlobalISel: support irtranslation of icmp instructions. llvm-svn: 278969
2016-08-18 04:25:25 +08:00
unsigned Res, unsigned Op0,
unsigned Op1) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
#ifndef NDEBUG
assert(MRI->getType(Op0) == MRI->getType(Op0) && "type mismatch");
assert(CmpInst::isIntPredicate(Pred) && "invalid predicate");
GlobalISel: relax type constraints on G_ICMP to allow pointers. llvm-svn: 281600
2016-09-15 18:40:38 +08:00
if (MRI->getType(Op0).isScalar() || MRI->getType(Op0).isPointer())
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Res).isScalar() && "type mismatch");
else
assert(MRI->getType(Res).isVector() &&
MRI->getType(Res).getNumElements() ==
MRI->getType(Op0).getNumElements() &&
"type mismatch");
#endif
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_ICMP)
GlobalISel: support irtranslation of icmp instructions. llvm-svn: 278969
2016-08-18 04:25:25 +08:00
.addDef(Res)
.addPredicate(Pred)
.addUse(Op0)
.addUse(Op1);
}
GlobalISel: support translating select instructions. llvm-svn: 279309
2016-08-20 04:09:07 +08:00
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildFCmp(CmpInst::Predicate Pred,
GlobalISel: translate floating-point comparisons llvm-svn: 279319
2016-08-20 04:48:16 +08:00
unsigned Res, unsigned Op0,
unsigned Op1) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
#ifndef NDEBUG
assert((MRI->getType(Op0).isScalar() || MRI->getType(Op0).isVector()) &&
"invalid operand type");
assert(MRI->getType(Op0) == MRI->getType(Op1) && "type mismatch");
assert(CmpInst::isFPPredicate(Pred) && "invalid predicate");
if (MRI->getType(Op0).isScalar())
assert(MRI->getType(Res).isScalar() && "type mismatch");
else
assert(MRI->getType(Res).isVector() &&
MRI->getType(Res).getNumElements() ==
MRI->getType(Op0).getNumElements() &&
"type mismatch");
#endif
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_FCMP)
GlobalISel: translate floating-point comparisons llvm-svn: 279319
2016-08-20 04:48:16 +08:00
.addDef(Res)
.addPredicate(Pred)
.addUse(Op0)
.addUse(Op1);
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
MachineInstrBuilder MachineIRBuilder::buildSelect(unsigned Res, unsigned Tst,
GlobalISel: support translating select instructions. llvm-svn: 279309
2016-08-20 04:09:07 +08:00
unsigned Op0, unsigned Op1) {
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
#ifndef NDEBUG
GlobalISel: allow G_SELECT instructions for pointers. llvm-svn: 288835
2016-12-07 02:38:34 +08:00
LLT ResTy = MRI->getType(Res);
assert((ResTy.isScalar() || ResTy.isVector() || ResTy.isPointer()) &&
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
"invalid operand type");
GlobalISel: allow G_SELECT instructions for pointers. llvm-svn: 288835
2016-12-07 02:38:34 +08:00
assert(ResTy == MRI->getType(Op0) && ResTy == MRI->getType(Op1) &&
"type mismatch");
if (ResTy.isScalar() || ResTy.isPointer())
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
assert(MRI->getType(Tst).isScalar() && "type mismatch");
else
[GlobalISel] Relax vector G_SELECT assertion. For vector operands, the `select` instruction supports both vector and non-vector conditions. The MIR builder had an overly restrictive assertion, that only accepted vector conditions for vector selects (in effect implementing ISD::VSELECT). Make it possible to express the full range of G_SELECTs. llvm-svn: 297207
2017-03-08 04:53:03 +08:00
assert((MRI->getType(Tst).isScalar() ||
(MRI->getType(Tst).isVector() &&
MRI->getType(Tst).getNumElements() ==
MRI->getType(Op0).getNumElements())) &&
GlobalISel: fix comments and add assertions for valid instructions. llvm-svn: 281036
2016-09-09 19:46:58 +08:00
"type mismatch");
#endif
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
return buildInstr(TargetOpcode::G_SELECT)
GlobalISel: support translating select instructions. llvm-svn: 279309
2016-08-20 04:09:07 +08:00
.addDef(Res)
.addUse(Tst)
.addUse(Op0)
.addUse(Op1);
}
GlobalISel: make truncate/extend casts uniform They really should have both types represented, but early variants were created before MachineInstrs could have multiple types so they're rather ambiguous. llvm-svn: 279567
2016-08-24 05:01:33 +08:00
[GlobalISel] Translate insertelement and extractelement Reviewers: qcolombet, aditya_nandakumar, dsanders, ab, t.p.northover, javed.absar Reviewed By: qcolombet Subscribers: dberris, rovka, llvm-commits, kristof.beyls Differential Revision: https://reviews.llvm.org/D30761 llvm-svn: 297495
2017-03-11 03:08:28 +08:00
MachineInstrBuilder MachineIRBuilder::buildInsertVectorElement(unsigned Res,
unsigned Val,
unsigned Elt,
unsigned Idx) {
#ifndef NDEBUG
LLT ResTy = MRI->getType(Res);
LLT ValTy = MRI->getType(Val);
LLT EltTy = MRI->getType(Elt);
LLT IdxTy = MRI->getType(Idx);
assert(ResTy.isVector() && ValTy.isVector() && "invalid operand type");
[GlobalISel] Support vector-of-pointers in LLT This fixes PR32471. As comment 10 on that bug report highlights (https://bugs.llvm.org//show_bug.cgi?id=32471#c10), there are quite a few different defendable design tradeoffs that could be made, including not representing pointers at all in LLT. I decided to go for representing vector-of-pointer as a concept in LLT, while keeping the size of the LLT type 64 bits (this is an increase from 48 bits before). My rationale for keeping pointers explicit is that on some targets probably it's very handy to have the distinction between pointer and non-pointer (e.g. 68K has a different register bank for pointers IIRC). If we keep a scalar pointer, it probably is easiest to also have a vector-of-pointers to keep LLT relatively conceptually clean and orthogonal, while we don't have a very strong reason to break that orthogonality. Once we gain more experience on the use of LLT, we can of course reconsider this direction. Rejecting vector-of-pointer types in the IRTranslator is also an option to avoid the crash reported in PR32471, but that is only a very short-term solution; also needs quite a bit of code tweaks in places, and is probably fragile. Therefore I didn't consider this the best option. llvm-svn: 300664
2017-04-19 15:23:57 +08:00
assert(IdxTy.isScalar() && "invalid operand type");
[GlobalISel] Translate insertelement and extractelement Reviewers: qcolombet, aditya_nandakumar, dsanders, ab, t.p.northover, javed.absar Reviewed By: qcolombet Subscribers: dberris, rovka, llvm-commits, kristof.beyls Differential Revision: https://reviews.llvm.org/D30761 llvm-svn: 297495
2017-03-11 03:08:28 +08:00
assert(ResTy.getNumElements() == ValTy.getNumElements() && "type mismatch");
assert(ResTy.getElementType() == EltTy && "type mismatch");
#endif
return buildInstr(TargetOpcode::G_INSERT_VECTOR_ELT)
.addDef(Res)
.addUse(Val)
.addUse(Elt)
.addUse(Idx);
}
MachineInstrBuilder MachineIRBuilder::buildExtractVectorElement(unsigned Res,
unsigned Val,
unsigned Idx) {
#ifndef NDEBUG
LLT ResTy = MRI->getType(Res);
LLT ValTy = MRI->getType(Val);
LLT IdxTy = MRI->getType(Idx);
assert(ValTy.isVector() && "invalid operand type");
[GlobalISel] Support vector-of-pointers in LLT This fixes PR32471. As comment 10 on that bug report highlights (https://bugs.llvm.org//show_bug.cgi?id=32471#c10), there are quite a few different defendable design tradeoffs that could be made, including not representing pointers at all in LLT. I decided to go for representing vector-of-pointer as a concept in LLT, while keeping the size of the LLT type 64 bits (this is an increase from 48 bits before). My rationale for keeping pointers explicit is that on some targets probably it's very handy to have the distinction between pointer and non-pointer (e.g. 68K has a different register bank for pointers IIRC). If we keep a scalar pointer, it probably is easiest to also have a vector-of-pointers to keep LLT relatively conceptually clean and orthogonal, while we don't have a very strong reason to break that orthogonality. Once we gain more experience on the use of LLT, we can of course reconsider this direction. Rejecting vector-of-pointer types in the IRTranslator is also an option to avoid the crash reported in PR32471, but that is only a very short-term solution; also needs quite a bit of code tweaks in places, and is probably fragile. Therefore I didn't consider this the best option. llvm-svn: 300664
2017-04-19 15:23:57 +08:00
assert((ResTy.isScalar() || ResTy.isPointer()) && "invalid operand type");
assert(IdxTy.isScalar() && "invalid operand type");
[GlobalISel] Translate insertelement and extractelement Reviewers: qcolombet, aditya_nandakumar, dsanders, ab, t.p.northover, javed.absar Reviewed By: qcolombet Subscribers: dberris, rovka, llvm-commits, kristof.beyls Differential Revision: https://reviews.llvm.org/D30761 llvm-svn: 297495
2017-03-11 03:08:28 +08:00
assert(ValTy.getElementType() == ResTy && "type mismatch");
#endif
return buildInstr(TargetOpcode::G_EXTRACT_VECTOR_ELT)
.addDef(Res)
.addUse(Val)
.addUse(Idx);
}
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
void MachineIRBuilder::validateTruncExt(unsigned Dst, unsigned Src,
bool IsExtend) {
#ifdef out validation code when asserts are disabled to remove unused variable warnings. llvm-svn: 279582
2016-08-24 06:14:15 +08:00
#ifndef NDEBUG
GlobalISel: move type information to MachineRegisterInfo. We want each register to have a canonical type, which means the best place to store this is in MachineRegisterInfo rather than on every MachineInstr that happens to use or define that register. Most changes following from this are pretty simple (you need an MRI anyway if you're going to be doing any transformations, so just check the type there). But legalization doesn't really want to check redundant operands (when, for example, a G_ADD only ever has one type) so I've made use of MCInstrDesc's operand type field to encode these constraints and limit legalization's work. As an added bonus, more validation is possible, both in MachineVerifier and MachineIRBuilder (coming soon). llvm-svn: 281035
2016-09-09 19:46:34 +08:00
LLT SrcTy = MRI->getType(Src);
LLT DstTy = MRI->getType(Dst);
GlobalISel: make truncate/extend casts uniform They really should have both types represented, but early variants were created before MachineInstrs could have multiple types so they're rather ambiguous. llvm-svn: 279567
2016-08-24 05:01:33 +08:00
if (DstTy.isVector()) {
assert(SrcTy.isVector() && "mismatched cast between vecot and non-vector");
assert(SrcTy.getNumElements() == DstTy.getNumElements() &&
"different number of elements in a trunc/ext");
} else
assert(DstTy.isScalar() && SrcTy.isScalar() && "invalid extend/trunc");
if (IsExtend)
assert(DstTy.getSizeInBits() > SrcTy.getSizeInBits() &&
"invalid narrowing extend");
else
assert(DstTy.getSizeInBits() < SrcTy.getSizeInBits() &&
"invalid widening trunc");
#ifdef out validation code when asserts are disabled to remove unused variable warnings. llvm-svn: 279582
2016-08-24 06:14:15 +08:00
#endif
GlobalISel: make truncate/extend casts uniform They really should have both types represented, but early variants were created before MachineInstrs could have multiple types so they're rather ambiguous. llvm-svn: 279567
2016-08-24 05:01:33 +08:00
}