2012-02-19 10:04:03 +08:00
|
|
|
//===-- ARM.h - Top-level interface for ARM representation ------*- C++ -*-===//
|
2006-05-15 06:18:28 +08:00
|
|
|
//
|
2019-01-19 16:50:56 +08:00
|
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
|
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
|
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
2006-05-15 06:18:28 +08:00
|
|
|
//
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
//
|
|
|
|
|
// This file contains the entry points for global functions defined in the LLVM
|
|
|
|
|
// ARM back-end.
|
|
|
|
|
//
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
2014-08-14 00:26:38 +08:00
|
|
|
#ifndef LLVM_LIB_TARGET_ARM_ARM_H
|
|
|
|
|
#define LLVM_LIB_TARGET_ARM_ARM_H
|
2006-05-15 06:18:28 +08:00
|
|
|
|
2018-06-28 20:55:29 +08:00
|
|
|
#include "llvm/IR/LegacyPassManager.h"
|
2014-03-23 07:51:00 +08:00
|
|
|
#include "llvm/Support/CodeGen.h"
|
2015-06-09 02:50:43 +08:00
|
|
|
#include <functional>
|
2017-01-27 07:40:06 +08:00
|
|
|
#include <vector>
|
2006-05-15 06:18:28 +08:00
|
|
|
|
|
|
|
|
namespace llvm {
|
2007-01-19 15:51:42 +08:00
|
|
|
|
2011-07-11 11:57:24 +08:00
|
|
|
class ARMAsmPrinter;
|
2009-06-27 05:28:53 +08:00
|
|
|
class ARMBaseTargetMachine;
|
2017-04-28 17:10:38 +08:00
|
|
|
class ARMRegisterBankInfo;
|
|
|
|
|
class ARMSubtarget;
|
2017-01-27 07:40:06 +08:00
|
|
|
struct BasicBlockInfo;
|
2015-06-09 02:50:43 +08:00
|
|
|
class Function;
|
2007-01-19 15:51:42 +08:00
|
|
|
class FunctionPass;
|
2017-04-28 17:10:38 +08:00
|
|
|
class InstructionSelector;
|
2017-01-27 07:40:06 +08:00
|
|
|
class MachineBasicBlock;
|
|
|
|
|
class MachineFunction;
|
2011-07-11 11:57:24 +08:00
|
|
|
class MachineInstr;
|
|
|
|
|
class MCInst;
|
2016-07-16 10:24:10 +08:00
|
|
|
class PassRegistry;
|
2010-09-18 02:46:17 +08:00
|
|
|
|
[ARM] MVE Tail Predication
The MVE and LOB extensions of Armv8.1m can be combined to enable
'tail predication' which removes the need for a scalar remainder
loop after vectorization. Lane predication is performed implicitly
via a system register. The effects of predication is described in
Section B5.6.3 of the Armv8.1-m Arch Reference Manual, the key points
being:
- For vector operations that perform reduction across the vector and
produce a scalar result, whether the value is accumulated or not.
- For non-load instructions, the predicate flags determine if the
destination register byte is updated with the new value or if the
previous value is preserved.
- For vector store instructions, whether the store occurs or not.
- For vector load instructions, whether the value that is loaded or
whether zeros are written to that element of the destination
register.
This patch implements a pass that takes a hardware loop, containing
masked vector instructions, and converts it something that resembles
an MVE tail predicated loop. Currently, if we had code generation,
we'd generate a loop in which the VCTP would generate the predicate
and VPST would then setup the value of VPR.PO. The loads and stores
would be placed in VPT blocks so this is not tail predication, but
normal VPT predication with the predicate based upon a element
counting induction variable. Further work needs to be done to finally
produce a true tail predicated loop.
Because only the loads and stores are predicated, in both the LLVM IR
and MIR level, we will restrict support to only lane-wise operations
(no horizontal reductions). We will perform a final check on MIR
during loop finalisation too.
Another restriction, specific to MVE, is that all the vector
instructions need operate on the same number of elements. This is
because predication is performed at the byte level and this is set
on entry to the loop, or by the VCTP instead.
Differential Revision: https://reviews.llvm.org/D65884
llvm-svn: 371179
2019-09-06 16:24:41 +08:00
|
|
|
Pass *createMVETailPredicationPass();
|
2019-06-25 18:45:51 +08:00
|
|
|
FunctionPass *createARMLowOverheadLoopsPass();
|
2018-06-28 20:55:29 +08:00
|
|
|
Pass *createARMParallelDSPPass();
|
2009-09-28 22:30:20 +08:00
|
|
|
FunctionPass *createARMISelDag(ARMBaseTargetMachine &TM,
|
|
|
|
|
CodeGenOpt::Level OptLevel);
|
2013-03-16 02:28:25 +08:00
|
|
|
FunctionPass *createA15SDOptimizerPass();
|
2009-06-13 17:12:55 +08:00
|
|
|
FunctionPass *createARMLoadStoreOptimizationPass(bool PreAlloc = false);
|
2009-11-07 07:52:48 +08:00
|
|
|
FunctionPass *createARMExpandPseudoPass();
|
2007-01-19 15:51:42 +08:00
|
|
|
FunctionPass *createARMConstantIslandPass();
|
2010-12-06 06:04:16 +08:00
|
|
|
FunctionPass *createMLxExpansionPass();
|
2010-07-03 05:07:09 +08:00
|
|
|
FunctionPass *createThumb2ITBlockPass();
|
2019-06-14 19:46:05 +08:00
|
|
|
FunctionPass *createMVEVPTBlockPass();
|
2014-04-02 17:03:43 +08:00
|
|
|
FunctionPass *createARMOptimizeBarriersPass();
|
2015-06-09 02:50:43 +08:00
|
|
|
FunctionPass *createThumb2SizeReductionPass(
|
|
|
|
|
std::function<bool(const Function &)> Ftor = nullptr);
|
2017-04-28 17:10:38 +08:00
|
|
|
InstructionSelector *
|
2017-05-02 17:40:49 +08:00
|
|
|
createARMInstructionSelector(const ARMBaseTargetMachine &TM, const ARMSubtarget &STI,
|
2017-04-28 17:10:38 +08:00
|
|
|
const ARMRegisterBankInfo &RBI);
|
2020-01-08 21:08:27 +08:00
|
|
|
Pass *createMVEGatherScatterLoweringPass();
|
2009-07-10 09:54:42 +08:00
|
|
|
|
2010-11-15 05:00:02 +08:00
|
|
|
void LowerARMMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI,
|
2010-12-01 11:45:07 +08:00
|
|
|
ARMAsmPrinter &AP);
|
|
|
|
|
|
2018-06-28 20:55:29 +08:00
|
|
|
void initializeARMParallelDSPPass(PassRegistry &);
|
2016-07-16 10:24:10 +08:00
|
|
|
void initializeARMLoadStoreOptPass(PassRegistry &);
|
|
|
|
|
void initializeARMPreAllocLoadStoreOptPass(PassRegistry &);
|
2017-02-13 22:07:25 +08:00
|
|
|
void initializeARMConstantIslandsPass(PassRegistry &);
|
2017-09-06 06:45:23 +08:00
|
|
|
void initializeARMExpandPseudoPass(PassRegistry &);
|
2017-12-19 20:19:08 +08:00
|
|
|
void initializeThumb2SizeReducePass(PassRegistry &);
|
2019-06-18 20:13:11 +08:00
|
|
|
void initializeThumb2ITBlockPass(PassRegistry &);
|
2019-06-14 19:46:05 +08:00
|
|
|
void initializeMVEVPTBlockPass(PassRegistry &);
|
2019-06-25 18:45:51 +08:00
|
|
|
void initializeARMLowOverheadLoopsPass(PassRegistry &);
|
[ARM] MVE Tail Predication
The MVE and LOB extensions of Armv8.1m can be combined to enable
'tail predication' which removes the need for a scalar remainder
loop after vectorization. Lane predication is performed implicitly
via a system register. The effects of predication is described in
Section B5.6.3 of the Armv8.1-m Arch Reference Manual, the key points
being:
- For vector operations that perform reduction across the vector and
produce a scalar result, whether the value is accumulated or not.
- For non-load instructions, the predicate flags determine if the
destination register byte is updated with the new value or if the
previous value is preserved.
- For vector store instructions, whether the store occurs or not.
- For vector load instructions, whether the value that is loaded or
whether zeros are written to that element of the destination
register.
This patch implements a pass that takes a hardware loop, containing
masked vector instructions, and converts it something that resembles
an MVE tail predicated loop. Currently, if we had code generation,
we'd generate a loop in which the VCTP would generate the predicate
and VPST would then setup the value of VPR.PO. The loads and stores
would be placed in VPT blocks so this is not tail predication, but
normal VPT predication with the predicate based upon a element
counting induction variable. Further work needs to be done to finally
produce a true tail predicated loop.
Because only the loads and stores are predicated, in both the LLVM IR
and MIR level, we will restrict support to only lane-wise operations
(no horizontal reductions). We will perform a final check on MIR
during loop finalisation too.
Another restriction, specific to MVE, is that all the vector
instructions need operate on the same number of elements. This is
because predication is performed at the byte level and this is set
on entry to the loop, or by the VCTP instead.
Differential Revision: https://reviews.llvm.org/D65884
llvm-svn: 371179
2019-09-06 16:24:41 +08:00
|
|
|
void initializeMVETailPredicationPass(PassRegistry &);
|
2020-01-08 21:08:27 +08:00
|
|
|
void initializeMVEGatherScatterLoweringPass(PassRegistry &);
|
2016-07-16 10:24:10 +08:00
|
|
|
|
2017-01-27 07:40:06 +08:00
|
|
|
} // end namespace llvm
|
2006-05-15 06:18:28 +08:00
|
|
|
|
2017-01-27 07:40:06 +08:00
|
|
|
#endif // LLVM_LIB_TARGET_ARM_ARM_H
|