2003-09-30 20:37:50 +02:00
|
|
|
//===-- llvm/Instruction.def - File that describes Instructions -*- C++ -*-===//
|
2015-09-22 13:14:12 +02:00
|
|
|
//
|
2003-10-20 22:19:47 +02:00
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
|
//
|
2007-12-29 23:59:10 +01:00
|
|
|
// This file is distributed under the University of Illinois Open Source
|
|
|
|
|
// License. See LICENSE.TXT for details.
|
2015-09-22 13:14:12 +02:00
|
|
|
//
|
2003-10-20 22:19:47 +02:00
|
|
|
//===----------------------------------------------------------------------===//
|
2001-10-14 19:24:33 +02:00
|
|
|
//
|
|
|
|
|
// This file contains descriptions of the various LLVM instructions. This is
|
2015-09-22 13:14:12 +02:00
|
|
|
// used as a central place for enumerating the different instructions and
|
2001-10-14 19:24:33 +02:00
|
|
|
// should eventually be the place to put comments about the instructions.
|
|
|
|
|
//
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
|
|
// NOTE: NO INCLUDE GUARD DESIRED!
|
|
|
|
|
|
2015-09-22 13:14:12 +02:00
|
|
|
// Provide definitions of macros so that users of this file do not have to
|
2006-11-08 07:47:33 +01:00
|
|
|
// define everything to use it...
|
2001-10-14 19:24:33 +02:00
|
|
|
//
|
|
|
|
|
#ifndef FIRST_TERM_INST
|
|
|
|
|
#define FIRST_TERM_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef HANDLE_TERM_INST
|
2002-03-19 06:51:18 +01:00
|
|
|
#ifndef HANDLE_INST
|
|
|
|
|
#define HANDLE_TERM_INST(num, opcode, Class)
|
|
|
|
|
#else
|
|
|
|
|
#define HANDLE_TERM_INST(num, opcode, Class) HANDLE_INST(num, opcode, Class)
|
|
|
|
|
#endif
|
2001-10-14 19:24:33 +02:00
|
|
|
#endif
|
|
|
|
|
#ifndef LAST_TERM_INST
|
|
|
|
|
#define LAST_TERM_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
#ifndef FIRST_BINARY_INST
|
|
|
|
|
#define FIRST_BINARY_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef HANDLE_BINARY_INST
|
2002-03-19 06:51:18 +01:00
|
|
|
#ifndef HANDLE_INST
|
2001-10-14 19:24:33 +02:00
|
|
|
#define HANDLE_BINARY_INST(num, opcode, instclass)
|
2002-03-19 06:51:18 +01:00
|
|
|
#else
|
|
|
|
|
#define HANDLE_BINARY_INST(num, opcode, Class) HANDLE_INST(num, opcode, Class)
|
|
|
|
|
#endif
|
2001-10-14 19:24:33 +02:00
|
|
|
#endif
|
|
|
|
|
#ifndef LAST_BINARY_INST
|
|
|
|
|
#define LAST_BINARY_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
#ifndef FIRST_MEMORY_INST
|
|
|
|
|
#define FIRST_MEMORY_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef HANDLE_MEMORY_INST
|
2002-03-19 06:51:18 +01:00
|
|
|
#ifndef HANDLE_INST
|
|
|
|
|
#define HANDLE_MEMORY_INST(num, opcode, Class)
|
|
|
|
|
#else
|
|
|
|
|
#define HANDLE_MEMORY_INST(num, opcode, Class) HANDLE_INST(num, opcode, Class)
|
|
|
|
|
#endif
|
2001-10-14 19:24:33 +02:00
|
|
|
#endif
|
|
|
|
|
#ifndef LAST_MEMORY_INST
|
|
|
|
|
#define LAST_MEMORY_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
|
2006-11-27 02:05:10 +01:00
|
|
|
#ifndef FIRST_CAST_INST
|
|
|
|
|
#define FIRST_CAST_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef HANDLE_CAST_INST
|
|
|
|
|
#ifndef HANDLE_INST
|
|
|
|
|
#define HANDLE_CAST_INST(num, opcode, Class)
|
|
|
|
|
#else
|
|
|
|
|
#define HANDLE_CAST_INST(num, opcode, Class) HANDLE_INST(num, opcode, Class)
|
|
|
|
|
#endif
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef LAST_CAST_INST
|
|
|
|
|
#define LAST_CAST_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 06:38:55 +01:00
|
|
|
#ifndef FIRST_FUNCLETPAD_INST
|
|
|
|
|
#define FIRST_FUNCLETPAD_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef HANDLE_FUNCLETPAD_INST
|
|
|
|
|
#ifndef HANDLE_INST
|
|
|
|
|
#define HANDLE_FUNCLETPAD_INST(num, opcode, Class)
|
|
|
|
|
#else
|
|
|
|
|
#define HANDLE_FUNCLETPAD_INST(num, opcode, Class) HANDLE_INST(num, opcode, Class)
|
|
|
|
|
#endif
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef LAST_FUNCLETPAD_INST
|
|
|
|
|
#define LAST_FUNCLETPAD_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
|
2001-10-14 19:24:33 +02:00
|
|
|
#ifndef FIRST_OTHER_INST
|
|
|
|
|
#define FIRST_OTHER_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef HANDLE_OTHER_INST
|
2002-03-19 06:51:18 +01:00
|
|
|
#ifndef HANDLE_INST
|
|
|
|
|
#define HANDLE_OTHER_INST(num, opcode, Class)
|
|
|
|
|
#else
|
|
|
|
|
#define HANDLE_OTHER_INST(num, opcode, Class) HANDLE_INST(num, opcode, Class)
|
|
|
|
|
#endif
|
2001-10-14 19:24:33 +02:00
|
|
|
#endif
|
|
|
|
|
#ifndef LAST_OTHER_INST
|
|
|
|
|
#define LAST_OTHER_INST(num)
|
|
|
|
|
#endif
|
|
|
|
|
|
[IR] De-virtualize ~Value to save a vptr
Summary:
Implements PR889
Removing the virtual table pointer from Value saves 1% of RSS when doing
LTO of llc on Linux. The impact on time was positive, but too noisy to
conclusively say that performance improved. Here is a link to the
spreadsheet with the original data:
https://docs.google.com/spreadsheets/d/1F4FHir0qYnV0MEp2sYYp_BuvnJgWlWPhWOwZ6LbW7W4/edit?usp=sharing
This change makes it invalid to directly delete a Value, User, or
Instruction pointer. Instead, such code can be rewritten to a null check
and a call Value::deleteValue(). Value objects tend to have their
lifetimes managed through iplist, so for the most part, this isn't a big
deal. However, there are some places where LLVM deletes values, and
those places had to be migrated to deleteValue. I have also created
llvm::unique_value, which has a custom deleter, so it can be used in
place of std::unique_ptr<Value>.
I had to add the "DerivedUser" Deleter escape hatch for MemorySSA, which
derives from User outside of lib/IR. Code in IR cannot include MemorySSA
headers or call the MemoryAccess object destructors without introducing
a circular dependency, so we need some level of indirection.
Unfortunately, no class derived from User may have any virtual methods,
because adding a virtual method would break User::getHungOffOperands(),
which assumes that it can find the use list immediately prior to the
User object. I've added a static_assert to the appropriate OperandTraits
templates to help people avoid this trap.
Reviewers: chandlerc, mehdi_amini, pete, dberlin, george.burgess.iv
Reviewed By: chandlerc
Subscribers: krytarowski, eraman, george.burgess.iv, mzolotukhin, Prazek, nlewycky, hans, inglorion, pcc, tejohnson, dberlin, llvm-commits
Differential Revision: https://reviews.llvm.org/D31261
llvm-svn: 303362
2017-05-18 19:24:10 +02:00
|
|
|
#ifndef HANDLE_USER_INST
|
|
|
|
|
#define HANDLE_USER_INST(num, opc, Class) HANDLE_OTHER_INST(num, opc, Class)
|
|
|
|
|
#endif
|
|
|
|
|
|
2001-10-14 19:24:33 +02:00
|
|
|
// Terminator Instructions - These instructions are used to terminate a basic
|
|
|
|
|
// block of the program. Every basic block must end with one of these
|
|
|
|
|
// instructions for it to be a well formed basic block.
|
|
|
|
|
//
|
|
|
|
|
FIRST_TERM_INST ( 1)
|
2015-07-31 19:58:14 +02:00
|
|
|
HANDLE_TERM_INST ( 1, Ret , ReturnInst)
|
|
|
|
|
HANDLE_TERM_INST ( 2, Br , BranchInst)
|
|
|
|
|
HANDLE_TERM_INST ( 3, Switch , SwitchInst)
|
|
|
|
|
HANDLE_TERM_INST ( 4, IndirectBr , IndirectBrInst)
|
|
|
|
|
HANDLE_TERM_INST ( 5, Invoke , InvokeInst)
|
|
|
|
|
HANDLE_TERM_INST ( 6, Resume , ResumeInst)
|
|
|
|
|
HANDLE_TERM_INST ( 7, Unreachable , UnreachableInst)
|
|
|
|
|
HANDLE_TERM_INST ( 8, CleanupRet , CleanupReturnInst)
|
|
|
|
|
HANDLE_TERM_INST ( 9, CatchRet , CatchReturnInst)
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 06:38:55 +01:00
|
|
|
HANDLE_TERM_INST (10, CatchSwitch , CatchSwitchInst)
|
2015-12-14 19:34:23 +01:00
|
|
|
LAST_TERM_INST (10)
|
2001-10-14 19:24:33 +02:00
|
|
|
|
|
|
|
|
// Standard binary operators...
|
2015-12-14 19:34:23 +01:00
|
|
|
FIRST_BINARY_INST(11)
|
|
|
|
|
HANDLE_BINARY_INST(11, Add , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(12, FAdd , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(13, Sub , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(14, FSub , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(15, Mul , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(16, FMul , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(17, UDiv , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(18, SDiv , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(19, FDiv , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(20, URem , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(21, SRem , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(22, FRem , BinaryOperator)
|
2001-10-14 19:24:33 +02:00
|
|
|
|
2007-02-02 03:16:23 +01:00
|
|
|
// Logical operators (integer operands)
|
2015-12-14 19:34:23 +01:00
|
|
|
HANDLE_BINARY_INST(23, Shl , BinaryOperator) // Shift left (logical)
|
|
|
|
|
HANDLE_BINARY_INST(24, LShr , BinaryOperator) // Shift right (logical)
|
|
|
|
|
HANDLE_BINARY_INST(25, AShr , BinaryOperator) // Shift right (arithmetic)
|
|
|
|
|
HANDLE_BINARY_INST(26, And , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(27, Or , BinaryOperator)
|
|
|
|
|
HANDLE_BINARY_INST(28, Xor , BinaryOperator)
|
|
|
|
|
LAST_BINARY_INST(28)
|
2001-10-14 19:24:33 +02:00
|
|
|
|
|
|
|
|
// Memory operators...
|
2015-12-14 19:34:23 +01:00
|
|
|
FIRST_MEMORY_INST(29)
|
|
|
|
|
HANDLE_MEMORY_INST(29, Alloca, AllocaInst) // Stack management
|
|
|
|
|
HANDLE_MEMORY_INST(30, Load , LoadInst ) // Memory manipulation instrs
|
|
|
|
|
HANDLE_MEMORY_INST(31, Store , StoreInst )
|
|
|
|
|
HANDLE_MEMORY_INST(32, GetElementPtr, GetElementPtrInst)
|
|
|
|
|
HANDLE_MEMORY_INST(33, Fence , FenceInst )
|
|
|
|
|
HANDLE_MEMORY_INST(34, AtomicCmpXchg , AtomicCmpXchgInst )
|
|
|
|
|
HANDLE_MEMORY_INST(35, AtomicRMW , AtomicRMWInst )
|
|
|
|
|
LAST_MEMORY_INST(35)
|
2001-10-14 19:24:33 +02:00
|
|
|
|
2006-11-27 02:05:10 +01:00
|
|
|
// Cast operators ...
|
2015-09-22 13:14:12 +02:00
|
|
|
// NOTE: The order matters here because CastInst::isEliminableCastPair
|
2006-11-27 02:05:10 +01:00
|
|
|
// NOTE: (see Instructions.cpp) encodes a table based on this ordering.
|
2015-12-14 19:34:23 +01:00
|
|
|
FIRST_CAST_INST(36)
|
|
|
|
|
HANDLE_CAST_INST(36, Trunc , TruncInst ) // Truncate integers
|
|
|
|
|
HANDLE_CAST_INST(37, ZExt , ZExtInst ) // Zero extend integers
|
|
|
|
|
HANDLE_CAST_INST(38, SExt , SExtInst ) // Sign extend integers
|
|
|
|
|
HANDLE_CAST_INST(39, FPToUI , FPToUIInst ) // floating point -> UInt
|
|
|
|
|
HANDLE_CAST_INST(40, FPToSI , FPToSIInst ) // floating point -> SInt
|
|
|
|
|
HANDLE_CAST_INST(41, UIToFP , UIToFPInst ) // UInt -> floating point
|
|
|
|
|
HANDLE_CAST_INST(42, SIToFP , SIToFPInst ) // SInt -> floating point
|
|
|
|
|
HANDLE_CAST_INST(43, FPTrunc , FPTruncInst ) // Truncate floating point
|
|
|
|
|
HANDLE_CAST_INST(44, FPExt , FPExtInst ) // Extend floating point
|
|
|
|
|
HANDLE_CAST_INST(45, PtrToInt, PtrToIntInst) // Pointer -> Integer
|
|
|
|
|
HANDLE_CAST_INST(46, IntToPtr, IntToPtrInst) // Integer -> Pointer
|
|
|
|
|
HANDLE_CAST_INST(47, BitCast , BitCastInst ) // Type cast
|
|
|
|
|
HANDLE_CAST_INST(48, AddrSpaceCast, AddrSpaceCastInst) // addrspace cast
|
|
|
|
|
LAST_CAST_INST(48)
|
|
|
|
|
|
|
|
|
|
FIRST_FUNCLETPAD_INST(49)
|
|
|
|
|
HANDLE_FUNCLETPAD_INST(49, CleanupPad, CleanupPadInst)
|
|
|
|
|
HANDLE_FUNCLETPAD_INST(50, CatchPad , CatchPadInst)
|
|
|
|
|
LAST_FUNCLETPAD_INST(50)
|
2006-11-27 02:05:10 +01:00
|
|
|
|
2001-10-14 19:24:33 +02:00
|
|
|
// Other operators...
|
2015-12-14 19:34:23 +01:00
|
|
|
FIRST_OTHER_INST(51)
|
|
|
|
|
HANDLE_OTHER_INST(51, ICmp , ICmpInst ) // Integer comparison instruction
|
|
|
|
|
HANDLE_OTHER_INST(52, FCmp , FCmpInst ) // Floating point comparison instr.
|
|
|
|
|
HANDLE_OTHER_INST(53, PHI , PHINode ) // PHI node instruction
|
|
|
|
|
HANDLE_OTHER_INST(54, Call , CallInst ) // Call a function
|
|
|
|
|
HANDLE_OTHER_INST(55, Select , SelectInst ) // select instruction
|
[IR] De-virtualize ~Value to save a vptr
Summary:
Implements PR889
Removing the virtual table pointer from Value saves 1% of RSS when doing
LTO of llc on Linux. The impact on time was positive, but too noisy to
conclusively say that performance improved. Here is a link to the
spreadsheet with the original data:
https://docs.google.com/spreadsheets/d/1F4FHir0qYnV0MEp2sYYp_BuvnJgWlWPhWOwZ6LbW7W4/edit?usp=sharing
This change makes it invalid to directly delete a Value, User, or
Instruction pointer. Instead, such code can be rewritten to a null check
and a call Value::deleteValue(). Value objects tend to have their
lifetimes managed through iplist, so for the most part, this isn't a big
deal. However, there are some places where LLVM deletes values, and
those places had to be migrated to deleteValue. I have also created
llvm::unique_value, which has a custom deleter, so it can be used in
place of std::unique_ptr<Value>.
I had to add the "DerivedUser" Deleter escape hatch for MemorySSA, which
derives from User outside of lib/IR. Code in IR cannot include MemorySSA
headers or call the MemoryAccess object destructors without introducing
a circular dependency, so we need some level of indirection.
Unfortunately, no class derived from User may have any virtual methods,
because adding a virtual method would break User::getHungOffOperands(),
which assumes that it can find the use list immediately prior to the
User object. I've added a static_assert to the appropriate OperandTraits
templates to help people avoid this trap.
Reviewers: chandlerc, mehdi_amini, pete, dberlin, george.burgess.iv
Reviewed By: chandlerc
Subscribers: krytarowski, eraman, george.burgess.iv, mzolotukhin, Prazek, nlewycky, hans, inglorion, pcc, tejohnson, dberlin, llvm-commits
Differential Revision: https://reviews.llvm.org/D31261
llvm-svn: 303362
2017-05-18 19:24:10 +02:00
|
|
|
HANDLE_USER_INST (56, UserOp1, Instruction) // May be used internally in a pass
|
|
|
|
|
HANDLE_USER_INST (57, UserOp2, Instruction) // Internal to passes only
|
2015-12-14 19:34:23 +01:00
|
|
|
HANDLE_OTHER_INST(58, VAArg , VAArgInst ) // vaarg instruction
|
|
|
|
|
HANDLE_OTHER_INST(59, ExtractElement, ExtractElementInst)// extract from vector
|
|
|
|
|
HANDLE_OTHER_INST(60, InsertElement, InsertElementInst) // insert into vector
|
|
|
|
|
HANDLE_OTHER_INST(61, ShuffleVector, ShuffleVectorInst) // shuffle two vectors.
|
|
|
|
|
HANDLE_OTHER_INST(62, ExtractValue, ExtractValueInst)// extract from aggregate
|
|
|
|
|
HANDLE_OTHER_INST(63, InsertValue, InsertValueInst) // insert into aggregate
|
|
|
|
|
HANDLE_OTHER_INST(64, LandingPad, LandingPadInst) // Landing pad instruction.
|
|
|
|
|
LAST_OTHER_INST(64)
|
2001-10-14 19:24:33 +02:00
|
|
|
|
|
|
|
|
#undef FIRST_TERM_INST
|
|
|
|
|
#undef HANDLE_TERM_INST
|
|
|
|
|
#undef LAST_TERM_INST
|
|
|
|
|
|
|
|
|
|
#undef FIRST_BINARY_INST
|
|
|
|
|
#undef HANDLE_BINARY_INST
|
|
|
|
|
#undef LAST_BINARY_INST
|
|
|
|
|
|
|
|
|
|
#undef FIRST_MEMORY_INST
|
|
|
|
|
#undef HANDLE_MEMORY_INST
|
|
|
|
|
#undef LAST_MEMORY_INST
|
|
|
|
|
|
2006-11-27 02:05:10 +01:00
|
|
|
#undef FIRST_CAST_INST
|
|
|
|
|
#undef HANDLE_CAST_INST
|
|
|
|
|
#undef LAST_CAST_INST
|
|
|
|
|
|
[IR] Reformulate LLVM's EH funclet IR
While we have successfully implemented a funclet-oriented EH scheme on
top of LLVM IR, our scheme has some notable deficiencies:
- catchendpad and cleanupendpad are necessary in the current design
but they are difficult to explain to others, even to seasoned LLVM
experts.
- catchendpad and cleanupendpad are optimization barriers. They cannot
be split and force all potentially throwing call-sites to be invokes.
This has a noticable effect on the quality of our code generation.
- catchpad, while similar in some aspects to invoke, is fairly awkward.
It is unsplittable, starts a funclet, and has control flow to other
funclets.
- The nesting relationship between funclets is currently a property of
control flow edges. Because of this, we are forced to carefully
analyze the flow graph to see if there might potentially exist illegal
nesting among funclets. While we have logic to clone funclets when
they are illegally nested, it would be nicer if we had a
representation which forbade them upfront.
Let's clean this up a bit by doing the following:
- Instead, make catchpad more like cleanuppad and landingpad: no control
flow, just a bunch of simple operands; catchpad would be splittable.
- Introduce catchswitch, a control flow instruction designed to model
the constraints of funclet oriented EH.
- Make funclet scoping explicit by having funclet instructions consume
the token produced by the funclet which contains them.
- Remove catchendpad and cleanupendpad. Their presence can be inferred
implicitly using coloring information.
N.B. The state numbering code for the CLR has been updated but the
veracity of it's output cannot be spoken for. An expert should take a
look to make sure the results are reasonable.
Reviewers: rnk, JosephTremoulet, andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D15139
llvm-svn: 255422
2015-12-12 06:38:55 +01:00
|
|
|
#undef FIRST_FUNCLETPAD_INST
|
|
|
|
|
#undef HANDLE_FUNCLETPAD_INST
|
|
|
|
|
#undef LAST_FUNCLETPAD_INST
|
|
|
|
|
|
2001-10-14 19:24:33 +02:00
|
|
|
#undef FIRST_OTHER_INST
|
|
|
|
|
#undef HANDLE_OTHER_INST
|
|
|
|
|
#undef LAST_OTHER_INST
|
|
|
|
|
|
[IR] De-virtualize ~Value to save a vptr
Summary:
Implements PR889
Removing the virtual table pointer from Value saves 1% of RSS when doing
LTO of llc on Linux. The impact on time was positive, but too noisy to
conclusively say that performance improved. Here is a link to the
spreadsheet with the original data:
https://docs.google.com/spreadsheets/d/1F4FHir0qYnV0MEp2sYYp_BuvnJgWlWPhWOwZ6LbW7W4/edit?usp=sharing
This change makes it invalid to directly delete a Value, User, or
Instruction pointer. Instead, such code can be rewritten to a null check
and a call Value::deleteValue(). Value objects tend to have their
lifetimes managed through iplist, so for the most part, this isn't a big
deal. However, there are some places where LLVM deletes values, and
those places had to be migrated to deleteValue. I have also created
llvm::unique_value, which has a custom deleter, so it can be used in
place of std::unique_ptr<Value>.
I had to add the "DerivedUser" Deleter escape hatch for MemorySSA, which
derives from User outside of lib/IR. Code in IR cannot include MemorySSA
headers or call the MemoryAccess object destructors without introducing
a circular dependency, so we need some level of indirection.
Unfortunately, no class derived from User may have any virtual methods,
because adding a virtual method would break User::getHungOffOperands(),
which assumes that it can find the use list immediately prior to the
User object. I've added a static_assert to the appropriate OperandTraits
templates to help people avoid this trap.
Reviewers: chandlerc, mehdi_amini, pete, dberlin, george.burgess.iv
Reviewed By: chandlerc
Subscribers: krytarowski, eraman, george.burgess.iv, mzolotukhin, Prazek, nlewycky, hans, inglorion, pcc, tejohnson, dberlin, llvm-commits
Differential Revision: https://reviews.llvm.org/D31261
llvm-svn: 303362
2017-05-18 19:24:10 +02:00
|
|
|
#undef HANDLE_USER_INST
|
|
|
|
|
|
2002-03-19 06:51:18 +01:00
|
|
|
#ifdef HANDLE_INST
|
|
|
|
|
#undef HANDLE_INST
|
|
|
|
|
#endif
|
