2007-12-08 20:24:38 +00:00
|
|
|
//===-- LegalizeTypes.cpp - Common code for DAG type legalizer ------------===//
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
//
|
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
2007-12-29 20:36:04 +00:00
|
|
|
// This file is distributed under the University of Illinois Open Source
|
|
|
|
// License. See LICENSE.TXT for details.
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//
|
|
|
|
// This file implements the SelectionDAG::LegalizeTypes method. It transforms
|
2007-12-08 20:24:38 +00:00
|
|
|
// an arbitrary well-formed SelectionDAG to only consist of legal types. This
|
|
|
|
// is common code shared among the LegalizeTypes*.cpp files.
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2007-12-08 20:16:06 +00:00
|
|
|
#include "LegalizeTypes.h"
|
2007-10-19 04:32:47 +00:00
|
|
|
#include "llvm/Constants.h"
|
|
|
|
#include "llvm/DerivedTypes.h"
|
2008-02-03 02:05:04 +00:00
|
|
|
#include "llvm/Support/CommandLine.h"
|
2007-11-09 13:41:39 +00:00
|
|
|
#include "llvm/Support/MathExtras.h"
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
using namespace llvm;
|
|
|
|
|
2008-02-03 02:05:04 +00:00
|
|
|
#ifndef NDEBUG
|
|
|
|
static cl::opt<bool>
|
|
|
|
ViewLegalizeTypesDAGs("view-legalize-types-dags", cl::Hidden,
|
|
|
|
cl::desc("Pop up a window to show dags before legalize types"));
|
|
|
|
#else
|
|
|
|
static const bool ViewLegalizeTypesDAGs = 0;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
|
|
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
/// run - This is the main entry point for the type legalizer. This does a
|
|
|
|
/// top-down traversal of the dag, legalizing types as it goes.
|
|
|
|
void DAGTypeLegalizer::run() {
|
|
|
|
// Create a dummy node (which is not added to allnodes), that adds a reference
|
|
|
|
// to the root node, preventing it from being deleted, and tracking any
|
|
|
|
// changes of the root.
|
|
|
|
HandleSDNode Dummy(DAG.getRoot());
|
|
|
|
|
|
|
|
// The root of the dag may dangle to deleted nodes until the type legalizer is
|
|
|
|
// done. Set it to null to avoid confusion.
|
|
|
|
DAG.setRoot(SDOperand());
|
|
|
|
|
|
|
|
// Walk all nodes in the graph, assigning them a NodeID of 'ReadyToProcess'
|
2007-10-15 13:30:18 +00:00
|
|
|
// (and remembering them) if they are leaves and assigning 'NewNode' if
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
// non-leaves.
|
|
|
|
for (SelectionDAG::allnodes_iterator I = DAG.allnodes_begin(),
|
|
|
|
E = DAG.allnodes_end(); I != E; ++I) {
|
|
|
|
if (I->getNumOperands() == 0) {
|
|
|
|
I->setNodeId(ReadyToProcess);
|
|
|
|
Worklist.push_back(I);
|
|
|
|
} else {
|
|
|
|
I->setNodeId(NewNode);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Now that we have a set of nodes to process, handle them all.
|
|
|
|
while (!Worklist.empty()) {
|
|
|
|
SDNode *N = Worklist.back();
|
|
|
|
Worklist.pop_back();
|
|
|
|
assert(N->getNodeId() == ReadyToProcess &&
|
|
|
|
"Node should be ready if on worklist!");
|
|
|
|
|
|
|
|
// Scan the values produced by the node, checking to see if any result
|
|
|
|
// types are illegal.
|
|
|
|
unsigned i = 0;
|
|
|
|
unsigned NumResults = N->getNumValues();
|
|
|
|
do {
|
2007-12-04 07:48:46 +00:00
|
|
|
MVT::ValueType ResultVT = N->getValueType(i);
|
|
|
|
LegalizeAction Action = getTypeAction(ResultVT);
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
if (Action == Promote) {
|
|
|
|
PromoteResult(N, i);
|
|
|
|
goto NodeDone;
|
|
|
|
} else if (Action == Expand) {
|
2007-12-04 07:48:46 +00:00
|
|
|
// Expand can mean 1) split integer in half 2) scalarize single-element
|
|
|
|
// vector 3) split vector in half.
|
|
|
|
if (!MVT::isVector(ResultVT))
|
|
|
|
ExpandResult(N, i);
|
|
|
|
else if (MVT::getVectorNumElements(ResultVT) == 1)
|
|
|
|
ScalarizeResult(N, i); // Scalarize the single-element vector.
|
2007-12-08 22:37:41 +00:00
|
|
|
else
|
|
|
|
SplitResult(N, i); // Split the vector in half.
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
goto NodeDone;
|
|
|
|
} else {
|
|
|
|
assert(Action == Legal && "Unknown action!");
|
|
|
|
}
|
|
|
|
} while (++i < NumResults);
|
|
|
|
|
|
|
|
// Scan the operand list for the node, handling any nodes with operands that
|
|
|
|
// are illegal.
|
|
|
|
{
|
|
|
|
unsigned NumOperands = N->getNumOperands();
|
|
|
|
bool NeedsRevisit = false;
|
|
|
|
for (i = 0; i != NumOperands; ++i) {
|
2007-12-04 07:29:51 +00:00
|
|
|
MVT::ValueType OpVT = N->getOperand(i).getValueType();
|
|
|
|
LegalizeAction Action = getTypeAction(OpVT);
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
if (Action == Promote) {
|
|
|
|
NeedsRevisit = PromoteOperand(N, i);
|
|
|
|
break;
|
|
|
|
} else if (Action == Expand) {
|
2007-12-04 07:29:51 +00:00
|
|
|
// Expand can mean 1) split integer in half 2) scalarize single-element
|
|
|
|
// vector 3) split vector in half.
|
|
|
|
if (!MVT::isVector(OpVT)) {
|
|
|
|
NeedsRevisit = ExpandOperand(N, i);
|
2007-12-04 07:48:46 +00:00
|
|
|
} else if (MVT::getVectorNumElements(OpVT) == 1) {
|
|
|
|
// Scalarize the single-element vector.
|
|
|
|
NeedsRevisit = ScalarizeOperand(N, i);
|
2007-12-04 07:29:51 +00:00
|
|
|
} else {
|
2007-12-08 22:37:41 +00:00
|
|
|
NeedsRevisit = SplitOperand(N, i); // Split the vector in half.
|
2007-12-04 07:29:51 +00:00
|
|
|
}
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
break;
|
|
|
|
} else {
|
|
|
|
assert(Action == Legal && "Unknown action!");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2007-10-15 16:46:29 +00:00
|
|
|
// If the node needs revisiting, don't add all users to the worklist etc.
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
if (NeedsRevisit)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (i == NumOperands)
|
|
|
|
DEBUG(cerr << "Legally typed node: "; N->dump(&DAG); cerr << "\n");
|
|
|
|
}
|
|
|
|
NodeDone:
|
|
|
|
|
|
|
|
// If we reach here, the node was processed, potentially creating new nodes.
|
|
|
|
// Mark it as processed and add its users to the worklist as appropriate.
|
|
|
|
N->setNodeId(Processed);
|
|
|
|
|
|
|
|
for (SDNode::use_iterator UI = N->use_begin(), E = N->use_end();
|
|
|
|
UI != E; ++UI) {
|
|
|
|
SDNode *User = *UI;
|
|
|
|
int NodeID = User->getNodeId();
|
|
|
|
assert(NodeID != ReadyToProcess && NodeID != Processed &&
|
|
|
|
"Invalid node id for user of unprocessed node!");
|
|
|
|
|
|
|
|
// This node has two options: it can either be a new node or its Node ID
|
|
|
|
// may be a count of the number of operands it has that are not ready.
|
|
|
|
if (NodeID > 0) {
|
|
|
|
User->setNodeId(NodeID-1);
|
|
|
|
|
|
|
|
// If this was the last use it was waiting on, add it to the ready list.
|
|
|
|
if (NodeID-1 == ReadyToProcess)
|
|
|
|
Worklist.push_back(User);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Otherwise, this node is new: this is the first operand of it that
|
|
|
|
// became ready. Its new NodeID is the number of operands it has minus 1
|
|
|
|
// (as this node is now processed).
|
|
|
|
assert(NodeID == NewNode && "Unknown node ID!");
|
|
|
|
User->setNodeId(User->getNumOperands()-1);
|
|
|
|
|
|
|
|
// If the node only has a single operand, it is now ready.
|
|
|
|
if (User->getNumOperands() == 1)
|
|
|
|
Worklist.push_back(User);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// If the root changed (e.g. it was a dead load, update the root).
|
|
|
|
DAG.setRoot(Dummy.getValue());
|
|
|
|
|
|
|
|
//DAG.viewGraph();
|
|
|
|
|
|
|
|
// Remove dead nodes. This is important to do for cleanliness but also before
|
|
|
|
// the checking loop below. Implicit folding by the DAG.getNode operators can
|
|
|
|
// cause unreachable nodes to be around with their flags set to new.
|
|
|
|
DAG.RemoveDeadNodes();
|
|
|
|
|
|
|
|
// In a debug build, scan all the nodes to make sure we found them all. This
|
|
|
|
// ensures that there are no cycles and that everything got processed.
|
|
|
|
#ifndef NDEBUG
|
|
|
|
for (SelectionDAG::allnodes_iterator I = DAG.allnodes_begin(),
|
|
|
|
E = DAG.allnodes_end(); I != E; ++I) {
|
|
|
|
if (I->getNodeId() == Processed)
|
|
|
|
continue;
|
|
|
|
cerr << "Unprocessed node: ";
|
|
|
|
I->dump(&DAG); cerr << "\n";
|
|
|
|
|
|
|
|
if (I->getNodeId() == NewNode)
|
|
|
|
cerr << "New node not 'noticed'?\n";
|
|
|
|
else if (I->getNodeId() > 0)
|
|
|
|
cerr << "Operand not processed?\n";
|
|
|
|
else if (I->getNodeId() == ReadyToProcess)
|
|
|
|
cerr << "Not added to worklist?\n";
|
|
|
|
abort();
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
/// MarkNewNodes - The specified node is the root of a subtree of potentially
|
|
|
|
/// new nodes. Add the correct NodeId to mark it.
|
|
|
|
void DAGTypeLegalizer::MarkNewNodes(SDNode *N) {
|
|
|
|
// If this was an existing node that is already done, we're done.
|
|
|
|
if (N->getNodeId() != NewNode)
|
|
|
|
return;
|
|
|
|
|
|
|
|
// Okay, we know that this node is new. Recursively walk all of its operands
|
|
|
|
// to see if they are new also. The depth of this walk is bounded by the size
|
|
|
|
// of the new tree that was constructed (usually 2-3 nodes), so we don't worry
|
2007-10-15 17:16:12 +00:00
|
|
|
// about revisiting of nodes.
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
//
|
|
|
|
// As we walk the operands, keep track of the number of nodes that are
|
|
|
|
// processed. If non-zero, this will become the new nodeid of this node.
|
|
|
|
unsigned NumProcessed = 0;
|
|
|
|
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
|
|
|
|
int OpId = N->getOperand(i).Val->getNodeId();
|
|
|
|
if (OpId == NewNode)
|
|
|
|
MarkNewNodes(N->getOperand(i).Val);
|
|
|
|
else if (OpId == Processed)
|
|
|
|
++NumProcessed;
|
|
|
|
}
|
|
|
|
|
|
|
|
N->setNodeId(N->getNumOperands()-NumProcessed);
|
|
|
|
if (N->getNodeId() == ReadyToProcess)
|
|
|
|
Worklist.push_back(N);
|
|
|
|
}
|
|
|
|
|
2008-02-03 07:08:51 +00:00
|
|
|
namespace {
|
|
|
|
/// NodeUpdateListener - This class is a DAGUpdateListener that listens for
|
|
|
|
/// updates to nodes and recomputes their ready state.
|
|
|
|
class VISIBILITY_HIDDEN NodeUpdateListener :
|
|
|
|
public SelectionDAG::DAGUpdateListener {
|
|
|
|
DAGTypeLegalizer &DTL;
|
|
|
|
public:
|
|
|
|
NodeUpdateListener(DAGTypeLegalizer &dtl) : DTL(dtl) {}
|
|
|
|
|
|
|
|
virtual void NodeDeleted(SDNode *N) {
|
|
|
|
// Ignore deletes.
|
|
|
|
}
|
|
|
|
|
|
|
|
virtual void NodeUpdated(SDNode *N) {
|
|
|
|
// Node updates can mean pretty much anything. It is possible that an
|
|
|
|
// operand was set to something already processed (f.e.) in which case
|
|
|
|
// this node could become ready. Recompoute its flags.
|
2008-02-03 07:13:32 +00:00
|
|
|
if (N->getNodeId() != DAGTypeLegalizer::ReadyToProcess)
|
|
|
|
DTL.ReanalyzeNodeFlags(N);
|
2008-02-03 07:08:51 +00:00
|
|
|
}
|
|
|
|
};
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2007-11-24 07:07:01 +00:00
|
|
|
/// ReplaceValueWith - The specified value was legalized to the specified other
|
|
|
|
/// value. If they are different, update the DAG and NodeIDs replacing any uses
|
|
|
|
/// of From to use To instead.
|
|
|
|
void DAGTypeLegalizer::ReplaceValueWith(SDOperand From, SDOperand To) {
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
if (From == To) return;
|
|
|
|
|
|
|
|
// If expansion produced new nodes, make sure they are properly marked.
|
|
|
|
if (To.Val->getNodeId() == NewNode)
|
|
|
|
MarkNewNodes(To.Val);
|
|
|
|
|
|
|
|
// Anything that used the old node should now use the new one. Note that this
|
|
|
|
// can potentially cause recursive merging.
|
2008-02-03 07:08:51 +00:00
|
|
|
NodeUpdateListener NUL(*this);
|
|
|
|
DAG.ReplaceAllUsesOfValueWith(From, To, &NUL);
|
2007-11-21 16:43:19 +00:00
|
|
|
|
|
|
|
// The old node may still be present in ExpandedNodes or PromotedNodes.
|
|
|
|
// Inform them about the replacement.
|
|
|
|
ReplacedNodes[From] = To;
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
}
|
|
|
|
|
2007-11-24 07:07:01 +00:00
|
|
|
/// ReplaceNodeWith - Replace uses of the 'from' node's results with the 'to'
|
|
|
|
/// node's results. The from and to node must define identical result types.
|
|
|
|
void DAGTypeLegalizer::ReplaceNodeWith(SDNode *From, SDNode *To) {
|
|
|
|
if (From == To) return;
|
|
|
|
assert(From->getNumValues() == To->getNumValues() &&
|
|
|
|
"Node results don't match");
|
|
|
|
|
|
|
|
// If expansion produced new nodes, make sure they are properly marked.
|
|
|
|
if (To->getNodeId() == NewNode)
|
|
|
|
MarkNewNodes(To);
|
|
|
|
|
|
|
|
// Anything that used the old node should now use the new one. Note that this
|
|
|
|
// can potentially cause recursive merging.
|
2008-02-03 07:08:51 +00:00
|
|
|
NodeUpdateListener NUL(*this);
|
|
|
|
DAG.ReplaceAllUsesWith(From, To, &NUL);
|
2007-11-24 07:07:01 +00:00
|
|
|
|
|
|
|
// The old node may still be present in ExpandedNodes or PromotedNodes.
|
|
|
|
// Inform them about the replacement.
|
|
|
|
for (unsigned i = 0, e = From->getNumValues(); i != e; ++i) {
|
|
|
|
assert(From->getValueType(i) == To->getValueType(i) &&
|
|
|
|
"Node results don't match");
|
|
|
|
ReplacedNodes[SDOperand(From, i)] = SDOperand(To, i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2007-11-21 16:43:19 +00:00
|
|
|
/// RemapNode - If the specified value was already legalized to another value,
|
|
|
|
/// replace it by that value.
|
|
|
|
void DAGTypeLegalizer::RemapNode(SDOperand &N) {
|
2007-11-24 18:11:42 +00:00
|
|
|
DenseMap<SDOperand, SDOperand>::iterator I = ReplacedNodes.find(N);
|
|
|
|
if (I != ReplacedNodes.end()) {
|
|
|
|
// Use path compression to speed up future lookups if values get multiply
|
|
|
|
// replaced with other values.
|
|
|
|
RemapNode(I->second);
|
2007-11-21 16:43:19 +00:00
|
|
|
N = I->second;
|
2007-11-24 18:11:42 +00:00
|
|
|
}
|
2007-11-21 16:43:19 +00:00
|
|
|
}
|
|
|
|
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
void DAGTypeLegalizer::SetPromotedOp(SDOperand Op, SDOperand Result) {
|
|
|
|
if (Result.Val->getNodeId() == NewNode)
|
|
|
|
MarkNewNodes(Result.Val);
|
|
|
|
|
|
|
|
SDOperand &OpEntry = PromotedNodes[Op];
|
|
|
|
assert(OpEntry.Val == 0 && "Node is already promoted!");
|
|
|
|
OpEntry = Result;
|
|
|
|
}
|
|
|
|
|
2007-12-04 07:29:51 +00:00
|
|
|
void DAGTypeLegalizer::SetScalarizedOp(SDOperand Op, SDOperand Result) {
|
|
|
|
if (Result.Val->getNodeId() == NewNode)
|
|
|
|
MarkNewNodes(Result.Val);
|
|
|
|
|
|
|
|
SDOperand &OpEntry = ScalarizedNodes[Op];
|
|
|
|
assert(OpEntry.Val == 0 && "Node is already scalarized!");
|
|
|
|
OpEntry = Result;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
void DAGTypeLegalizer::GetExpandedOp(SDOperand Op, SDOperand &Lo,
|
|
|
|
SDOperand &Hi) {
|
|
|
|
std::pair<SDOperand, SDOperand> &Entry = ExpandedNodes[Op];
|
2007-11-21 16:43:19 +00:00
|
|
|
RemapNode(Entry.first);
|
|
|
|
RemapNode(Entry.second);
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
assert(Entry.first.Val && "Operand isn't expanded");
|
|
|
|
Lo = Entry.first;
|
|
|
|
Hi = Entry.second;
|
|
|
|
}
|
|
|
|
|
2007-12-08 22:40:18 +00:00
|
|
|
void DAGTypeLegalizer::SetExpandedOp(SDOperand Op, SDOperand Lo, SDOperand Hi) {
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
// Remember that this is the result of the node.
|
|
|
|
std::pair<SDOperand, SDOperand> &Entry = ExpandedNodes[Op];
|
|
|
|
assert(Entry.first.Val == 0 && "Node already expanded");
|
|
|
|
Entry.first = Lo;
|
|
|
|
Entry.second = Hi;
|
|
|
|
|
|
|
|
// Lo/Hi may have been newly allocated, if so, add nodeid's as relevant.
|
|
|
|
if (Lo.Val->getNodeId() == NewNode)
|
|
|
|
MarkNewNodes(Lo.Val);
|
|
|
|
if (Hi.Val->getNodeId() == NewNode)
|
|
|
|
MarkNewNodes(Hi.Val);
|
|
|
|
}
|
|
|
|
|
2007-12-08 22:40:18 +00:00
|
|
|
void DAGTypeLegalizer::GetSplitOp(SDOperand Op, SDOperand &Lo, SDOperand &Hi) {
|
|
|
|
std::pair<SDOperand, SDOperand> &Entry = SplitNodes[Op];
|
|
|
|
RemapNode(Entry.first);
|
|
|
|
RemapNode(Entry.second);
|
|
|
|
assert(Entry.first.Val && "Operand isn't split");
|
|
|
|
Lo = Entry.first;
|
|
|
|
Hi = Entry.second;
|
|
|
|
}
|
|
|
|
|
|
|
|
void DAGTypeLegalizer::SetSplitOp(SDOperand Op, SDOperand Lo, SDOperand Hi) {
|
|
|
|
// Remember that this is the result of the node.
|
|
|
|
std::pair<SDOperand, SDOperand> &Entry = SplitNodes[Op];
|
|
|
|
assert(Entry.first.Val == 0 && "Node already split");
|
|
|
|
Entry.first = Lo;
|
|
|
|
Entry.second = Hi;
|
|
|
|
|
|
|
|
// Lo/Hi may have been newly allocated, if so, add nodeid's as relevant.
|
|
|
|
if (Lo.Val->getNodeId() == NewNode)
|
|
|
|
MarkNewNodes(Lo.Val);
|
|
|
|
if (Hi.Val->getNodeId() == NewNode)
|
|
|
|
MarkNewNodes(Hi.Val);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2007-10-19 03:33:14 +00:00
|
|
|
SDOperand DAGTypeLegalizer::CreateStackStoreLoad(SDOperand Op,
|
|
|
|
MVT::ValueType DestVT) {
|
|
|
|
// Create the stack frame object.
|
|
|
|
SDOperand FIPtr = DAG.CreateStackTemporary(DestVT);
|
|
|
|
|
|
|
|
// Emit a store to the stack slot.
|
|
|
|
SDOperand Store = DAG.getStore(DAG.getEntryNode(), Op, FIPtr, NULL, 0);
|
|
|
|
// Result is a load from the stack slot.
|
|
|
|
return DAG.getLoad(DestVT, Store, FIPtr, NULL, 0);
|
|
|
|
}
|
|
|
|
|
2007-10-20 04:07:07 +00:00
|
|
|
/// HandleMemIntrinsic - This handles memcpy/memset/memmove with invalid
|
|
|
|
/// operands. This promotes or expands the operands as required.
|
|
|
|
SDOperand DAGTypeLegalizer::HandleMemIntrinsic(SDNode *N) {
|
|
|
|
// The chain and pointer [operands #0 and #1] are always valid types.
|
|
|
|
SDOperand Chain = N->getOperand(0);
|
|
|
|
SDOperand Ptr = N->getOperand(1);
|
|
|
|
SDOperand Op2 = N->getOperand(2);
|
|
|
|
|
|
|
|
// Op #2 is either a value (memset) or a pointer. Promote it if required.
|
|
|
|
switch (getTypeAction(Op2.getValueType())) {
|
|
|
|
default: assert(0 && "Unknown action for pointer/value operand");
|
|
|
|
case Legal: break;
|
|
|
|
case Promote: Op2 = GetPromotedOp(Op2); break;
|
|
|
|
}
|
|
|
|
|
|
|
|
// The length could have any action required.
|
|
|
|
SDOperand Length = N->getOperand(3);
|
|
|
|
switch (getTypeAction(Length.getValueType())) {
|
|
|
|
default: assert(0 && "Unknown action for memop operand");
|
|
|
|
case Legal: break;
|
|
|
|
case Promote: Length = GetPromotedZExtOp(Length); break;
|
|
|
|
case Expand:
|
|
|
|
SDOperand Dummy; // discard the high part.
|
|
|
|
GetExpandedOp(Length, Length, Dummy);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand Align = N->getOperand(4);
|
|
|
|
switch (getTypeAction(Align.getValueType())) {
|
|
|
|
default: assert(0 && "Unknown action for memop operand");
|
|
|
|
case Legal: break;
|
|
|
|
case Promote: Align = GetPromotedZExtOp(Align); break;
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand AlwaysInline = N->getOperand(5);
|
|
|
|
switch (getTypeAction(AlwaysInline.getValueType())) {
|
|
|
|
default: assert(0 && "Unknown action for memop operand");
|
|
|
|
case Legal: break;
|
|
|
|
case Promote: AlwaysInline = GetPromotedZExtOp(AlwaysInline); break;
|
|
|
|
}
|
|
|
|
|
|
|
|
SDOperand Ops[] = { Chain, Ptr, Op2, Length, Align, AlwaysInline };
|
|
|
|
return DAG.UpdateNodeOperands(SDOperand(N, 0), Ops, 6);
|
|
|
|
}
|
|
|
|
|
2007-10-22 18:26:21 +00:00
|
|
|
/// SplitOp - Return the lower and upper halves of Op's bits in a value type
|
|
|
|
/// half the size of Op's.
|
|
|
|
void DAGTypeLegalizer::SplitOp(SDOperand Op, SDOperand &Lo, SDOperand &Hi) {
|
|
|
|
unsigned NVTBits = MVT::getSizeInBits(Op.getValueType())/2;
|
|
|
|
assert(MVT::getSizeInBits(Op.getValueType()) == 2*NVTBits &&
|
|
|
|
"Cannot split odd sized integer type");
|
|
|
|
MVT::ValueType NVT = MVT::getIntegerType(NVTBits);
|
|
|
|
Lo = DAG.getNode(ISD::TRUNCATE, NVT, Op);
|
|
|
|
Hi = DAG.getNode(ISD::SRL, Op.getValueType(), Op,
|
|
|
|
DAG.getConstant(NVTBits, TLI.getShiftAmountTy()));
|
|
|
|
Hi = DAG.getNode(ISD::TRUNCATE, NVT, Hi);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// Entry Point
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
/// LegalizeTypes - This transforms the SelectionDAG into a SelectionDAG that
|
|
|
|
/// only uses types natively supported by the target.
|
|
|
|
///
|
|
|
|
/// Note that this is an involved process that may invalidate pointers into
|
|
|
|
/// the graph.
|
|
|
|
void SelectionDAG::LegalizeTypes() {
|
2008-02-03 02:05:04 +00:00
|
|
|
if (ViewLegalizeTypesDAGs) viewGraph();
|
|
|
|
|
One mundane change: Change ReplaceAllUsesOfValueWith to *optionally*
take a deleted nodes vector, instead of requiring it.
One more significant change: Implement the start of a legalizer that
just works on types. This legalizer is designed to run before the
operation legalizer and ensure just that the input dag is transformed
into an output dag whose operand and result types are all legal, even
if the operations on those types are not.
This design/impl has the following advantages:
1. When finished, this will *significantly* reduce the amount of code in
LegalizeDAG.cpp. It will remove all the code related to promotion and
expansion as well as splitting and scalarizing vectors.
2. The new code is very simple, idiomatic, and modular: unlike
LegalizeDAG.cpp, it has no 3000 line long functions. :)
3. The implementation is completely iterative instead of recursive, good
for hacking on large dags without blowing out your stack.
4. The implementation updates nodes in place when possible instead of
deallocating and reallocating the entire graph that points to some
mutated node.
5. The code nicely separates out handling of operations with invalid
results from operations with invalid operands, making some cases
simpler and easier to understand.
6. The new -debug-only=legalize-types option is very very handy :),
allowing you to easily understand what legalize types is doing.
This is not yet done. Until the ifdef added to SelectionDAGISel.cpp is
enabled, this does nothing. However, this code is sufficient to legalize
all of the code in 186.crafty, olden and freebench on an x86 machine. The
biggest issues are:
1. Vectors aren't implemented at all yet
2. SoftFP is a mess, I need to talk to Evan about it.
3. No lowering to libcalls is implemented yet.
4. Various operations are missing etc.
5. There are FIXME's for stuff I hax0r'd out, like softfp.
Hey, at least it is a step in the right direction :). If you'd like to help,
just enable the #ifdef in SelectionDAGISel.cpp and compile code with it. If
this explodes it will tell you what needs to be implemented. Help is
certainly appreciated.
Once this goes in, we can do three things:
1. Add a new pass of dag combine between the "type legalizer" and "operation
legalizer" passes. This will let us catch some long-standing isel issues
that we miss because operation legalization often obfuscates the dag with
target-specific nodes.
2. We can rip out all of the type legalization code from LegalizeDAG.cpp,
making it much smaller and simpler. When that happens we can then
reimplement the core functionality left in it in a much more efficient and
non-recursive way.
3. Once the whole legalizer is non-recursive, we can implement whole-function
selectiondags maybe...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42981 91177308-0d34-0410-b5e6-96231b3b80d8
2007-10-15 06:10:22 +00:00
|
|
|
DAGTypeLegalizer(*this).run();
|
|
|
|
}
|