llvm-project/llvm/lib/Target/PowerPC/PPCSubtarget.h

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

355 lines
12 KiB
C
Raw Normal View History

//===-- PPCSubtarget.h - Define Subtarget for the PPC ----------*- C++ -*--===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// This file declares the PowerPC specific subclass of TargetSubtargetInfo.
//
//===----------------------------------------------------------------------===//
#ifndef LLVM_LIB_TARGET_POWERPC_PPCSUBTARGET_H
#define LLVM_LIB_TARGET_POWERPC_PPCSUBTARGET_H
#include "PPCFrameLowering.h"
#include "PPCISelLowering.h"
#include "PPCInstrInfo.h"
#include "llvm/ADT/Triple.h"
#include "llvm/CodeGen/SelectionDAGTargetInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/MC/MCInstrItineraries.h"
#include <string>
#define GET_SUBTARGETINFO_HEADER
#include "PPCGenSubtargetInfo.inc"
// GCC #defines PPC on Linux but we use it as our namespace name
#undef PPC
namespace llvm {
class StringRef;
namespace PPC {
// -m directive values.
enum {
DIR_NONE,
DIR_32,
DIR_440,
DIR_601,
DIR_602,
DIR_603,
DIR_7400,
DIR_750,
DIR_970,
DIR_A2,
DIR_E500,
DIR_E500mc,
DIR_E5500,
DIR_PWR3,
DIR_PWR4,
DIR_PWR5,
DIR_PWR5X,
DIR_PWR6,
DIR_PWR6X,
DIR_PWR7,
DIR_PWR8,
DIR_PWR9,
DIR_64
};
}
class GlobalValue;
class TargetMachine;
class PPCSubtarget : public PPCGenSubtargetInfo {
public:
enum POPCNTDKind {
POPCNTD_Unavailable,
POPCNTD_Slow,
POPCNTD_Fast
};
protected:
/// TargetTriple - What processor and OS we're targeting.
Triple TargetTriple;
/// stackAlignment - The minimum alignment known to hold of the stack frame on
/// entry to the function and which must be maintained by every function.
Align StackAlignment;
/// Selected instruction itineraries (one entry per itinerary class.)
InstrItineraryData InstrItins;
/// Which cpu directive was used.
unsigned DarwinDirective;
/// Used by the ISel to turn in optimizations for POWER4-derived architectures
bool HasMFOCRF;
bool Has64BitSupport;
bool Use64BitRegs;
Add CR-bit tracking to the PowerPC backend for i1 values This change enables tracking i1 values in the PowerPC backend using the condition register bits. These bits can be treated on PowerPC as separate registers; individual bit operations (and, or, xor, etc.) are supported. Tracking booleans in CR bits has several advantages: - Reduction in register pressure (because we no longer need GPRs to store boolean values). - Logical operations on booleans can be handled more efficiently; we used to have to move all results from comparisons into GPRs, perform promoted logical operations in GPRs, and then move the result back into condition register bits to be used by conditional branches. This can be very inefficient, because the throughput of these CR <-> GPR moves have high latency and low throughput (especially when other associated instructions are accounted for). - On the POWER7 and similar cores, we can increase total throughput by using the CR bits. CR bit operations have a dedicated functional unit. Most of this is more-or-less mechanical: Adjustments were needed in the calling-convention code, support was added for spilling/restoring individual condition-register bits, and conditional branch instruction definitions taking specific CR bits were added (plus patterns and code for generating bit-level operations). This is enabled by default when running at -O2 and higher. For -O0 and -O1, where the ability to debug is more important, this feature is disabled by default. Individual CR bits do not have assigned DWARF register numbers, and storing values in CR bits makes them invisible to the debugger. It is critical, however, that we don't move i1 values that have been promoted to larger values (such as those passed as function arguments) into bit registers only to quickly turn around and move the values back into GPRs (such as happens when values are returned by functions). A pair of target-specific DAG combines are added to remove the trunc/extends in: trunc(binary-ops(binary-ops(zext(x), zext(y)), ...) and: zext(binary-ops(binary-ops(trunc(x), trunc(y)), ...) In short, we only want to use CR bits where some of the i1 values come from comparisons or are used by conditional branches or selects. To put it another way, if we can do the entire i1 computation in GPRs, then we probably should (on the POWER7, the GPR-operation throughput is higher, and for all cores, the CR <-> GPR moves are expensive). POWER7 test-suite performance results (from 10 runs in each configuration): SingleSource/Benchmarks/Misc/mandel-2: 35% speedup MultiSource/Benchmarks/Prolangs-C++/city/city: 21% speedup MultiSource/Benchmarks/MiBench/automotive-susan: 23% speedup SingleSource/Benchmarks/CoyoteBench/huffbench: 13% speedup SingleSource/Benchmarks/Misc-C++/Large/sphereflake: 13% speedup SingleSource/Benchmarks/Misc-C++/mandel-text: 10% speedup SingleSource/Benchmarks/Misc-C++-EH/spirit: 10% slowdown MultiSource/Applications/lemon/lemon: 8% slowdown llvm-svn: 202451
2014-02-28 08:27:01 +08:00
bool UseCRBits;
bool HasHardFloat;
bool IsPPC64;
bool HasAltivec;
bool HasFPU;
bool HasSPE;
bool HasQPX;
bool HasVSX;
bool NeedsTwoConstNR;
bool HasP8Vector;
bool HasP8Altivec;
bool HasP8Crypto;
bool HasP9Vector;
bool HasP9Altivec;
bool HasFCPSGN;
bool HasFSQRT;
bool HasFRE, HasFRES, HasFRSQRTE, HasFRSQRTES;
bool HasRecipPrec;
bool HasSTFIWX;
bool HasLFIWAX;
bool HasFPRND;
bool HasFPCVT;
bool HasISEL;
bool HasBPERMD;
bool HasExtDiv;
bool HasCMPB;
bool HasLDBRX;
bool IsBookE;
bool HasOnlyMSYNC;
bool IsE500;
bool IsPPC4xx;
bool IsPPC6xx;
bool FeatureMFTB;
bool DeprecatedDST;
bool HasLazyResolverStubs;
bool IsLittleEndian;
bool HasICBT;
[PowerPC] Loosen ELFv1 PPC64 func descriptor loads for indirect calls Function pointers under PPC64 ELFv1 (which is used on PPC64/Linux on the POWER7, A2 and earlier cores) are really pointers to a function descriptor, a structure with three pointers: the actual pointer to the code to which to jump, the pointer to the TOC needed by the callee, and an environment pointer. We used to chain these loads, and make them opaque to the rest of the optimizer, so that they'd always occur directly before the call. This is not necessary, and in fact, highly suboptimal on embedded cores. Once the function pointer is known, the loads can be performed ahead of time; in fact, they can be hoisted out of loops. Now these function descriptors are almost always generated by the linker, and thus the contents of the descriptors are invariant. As a result, by default, we'll mark the associated loads as invariant (allowing them to be hoisted out of loops). I've added a target feature to turn this off, however, just in case someone needs that option (constructing an on-stack descriptor, casting it to a function pointer, and then calling it cannot be well-defined C/C++ code, but I can imagine some JIT-compilation system doing so). Consider this simple test: $ cat call.c typedef void (*fp)(); void bar(fp x) { for (int i = 0; i < 1600000000; ++i) x(); } $ cat main.c typedef void (*fp)(); void bar(fp x); void foo() {} int main() { bar(foo); } On the PPC A2 (the BG/Q supercomputer), marking the function-descriptor loads as invariant brings the execution time down to ~8 seconds from ~32 seconds with the loads in the loop. The difference on the POWER7 is smaller. Compiling with: gcc -std=c99 -O3 -mcpu=native call.c main.c : ~6 seconds [this is 4.8.2] clang -O3 -mcpu=native call.c main.c : ~5.3 seconds clang -O3 -mcpu=native call.c main.c -mno-invariant-function-descriptors : ~4 seconds (looks like we'd benefit from additional loop unrolling here, as a first guess, because this is faster with the extra loads) The -mno-invariant-function-descriptors will be added to Clang shortly. llvm-svn: 226207
2015-01-16 05:17:34 +08:00
bool HasInvariantFunctionDescriptors;
bool HasPartwordAtomics;
bool HasDirectMove;
bool HasHTM;
bool HasFloat128;
bool IsISA3_0;
bool UseLongCalls;
bool SecurePlt;
bool VectorsUseTwoUnits;
bool UsePPCPreRASchedStrategy;
bool UsePPCPostRASchedStrategy;
POPCNTDKind HasPOPCNTD;
2015-02-25 09:06:45 +08:00
/// When targeting QPX running a stock PPC64 Linux kernel where the stack
/// alignment has not been changed, we need to keep the 16-byte alignment
/// of the stack.
bool IsQPXStackUnaligned;
const PPCTargetMachine &TM;
PPCFrameLowering FrameLowering;
PPCInstrInfo InstrInfo;
PPCTargetLowering TLInfo;
SelectionDAGTargetInfo TSInfo;
public:
/// This constructor initializes the data members to match that
/// of the specified triple.
///
PPCSubtarget(const Triple &TT, const std::string &CPU, const std::string &FS,
const PPCTargetMachine &TM);
/// ParseSubtargetFeatures - Parses features string setting specified
/// subtarget options. Definition of function is auto generated by tblgen.
void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
/// getStackAlignment - Returns the minimum alignment known to hold of the
/// stack frame on entry to the function and which must be maintained by every
/// function for this subtarget.
Align getStackAlignment() const { return StackAlignment; }
/// getDarwinDirective - Returns the -m directive specified for the cpu.
///
unsigned getDarwinDirective() const { return DarwinDirective; }
2014-06-14 06:38:48 +08:00
/// getInstrItins - Return the instruction itineraries based on subtarget
/// selection.
const InstrItineraryData *getInstrItineraryData() const override {
return &InstrItins;
}
const PPCFrameLowering *getFrameLowering() const override {
return &FrameLowering;
}
const PPCInstrInfo *getInstrInfo() const override { return &InstrInfo; }
const PPCTargetLowering *getTargetLowering() const override {
return &TLInfo;
}
const SelectionDAGTargetInfo *getSelectionDAGInfo() const override {
return &TSInfo;
}
const PPCRegisterInfo *getRegisterInfo() const override {
return &getInstrInfo()->getRegisterInfo();
}
const PPCTargetMachine &getTargetMachine() const { return TM; }
/// initializeSubtargetDependencies - Initializes using a CPU and feature string
/// so that we can use initializer lists for subtarget initialization.
PPCSubtarget &initializeSubtargetDependencies(StringRef CPU, StringRef FS);
private:
void initializeEnvironment();
void initSubtargetFeatures(StringRef CPU, StringRef FS);
public:
/// isPPC64 - Return true if we are generating code for 64-bit pointer mode.
///
bool isPPC64() const;
/// has64BitSupport - Return true if the selected CPU supports 64-bit
/// instructions, regardless of whether we are in 32-bit or 64-bit mode.
bool has64BitSupport() const { return Has64BitSupport; }
// useSoftFloat - Return true if soft-float option is turned on.
bool useSoftFloat() const {
if (isAIXABI() && !HasHardFloat)
report_fatal_error("soft-float is not yet supported on AIX.");
return !HasHardFloat;
}
/// use64BitRegs - Return true if in 64-bit mode or if we should use 64-bit
/// registers in 32-bit mode when possible. This can only true if
/// has64BitSupport() returns true.
bool use64BitRegs() const { return Use64BitRegs; }
Add CR-bit tracking to the PowerPC backend for i1 values This change enables tracking i1 values in the PowerPC backend using the condition register bits. These bits can be treated on PowerPC as separate registers; individual bit operations (and, or, xor, etc.) are supported. Tracking booleans in CR bits has several advantages: - Reduction in register pressure (because we no longer need GPRs to store boolean values). - Logical operations on booleans can be handled more efficiently; we used to have to move all results from comparisons into GPRs, perform promoted logical operations in GPRs, and then move the result back into condition register bits to be used by conditional branches. This can be very inefficient, because the throughput of these CR <-> GPR moves have high latency and low throughput (especially when other associated instructions are accounted for). - On the POWER7 and similar cores, we can increase total throughput by using the CR bits. CR bit operations have a dedicated functional unit. Most of this is more-or-less mechanical: Adjustments were needed in the calling-convention code, support was added for spilling/restoring individual condition-register bits, and conditional branch instruction definitions taking specific CR bits were added (plus patterns and code for generating bit-level operations). This is enabled by default when running at -O2 and higher. For -O0 and -O1, where the ability to debug is more important, this feature is disabled by default. Individual CR bits do not have assigned DWARF register numbers, and storing values in CR bits makes them invisible to the debugger. It is critical, however, that we don't move i1 values that have been promoted to larger values (such as those passed as function arguments) into bit registers only to quickly turn around and move the values back into GPRs (such as happens when values are returned by functions). A pair of target-specific DAG combines are added to remove the trunc/extends in: trunc(binary-ops(binary-ops(zext(x), zext(y)), ...) and: zext(binary-ops(binary-ops(trunc(x), trunc(y)), ...) In short, we only want to use CR bits where some of the i1 values come from comparisons or are used by conditional branches or selects. To put it another way, if we can do the entire i1 computation in GPRs, then we probably should (on the POWER7, the GPR-operation throughput is higher, and for all cores, the CR <-> GPR moves are expensive). POWER7 test-suite performance results (from 10 runs in each configuration): SingleSource/Benchmarks/Misc/mandel-2: 35% speedup MultiSource/Benchmarks/Prolangs-C++/city/city: 21% speedup MultiSource/Benchmarks/MiBench/automotive-susan: 23% speedup SingleSource/Benchmarks/CoyoteBench/huffbench: 13% speedup SingleSource/Benchmarks/Misc-C++/Large/sphereflake: 13% speedup SingleSource/Benchmarks/Misc-C++/mandel-text: 10% speedup SingleSource/Benchmarks/Misc-C++-EH/spirit: 10% slowdown MultiSource/Applications/lemon/lemon: 8% slowdown llvm-svn: 202451
2014-02-28 08:27:01 +08:00
/// useCRBits - Return true if we should store and manipulate i1 values in
/// the individual condition register bits.
bool useCRBits() const { return UseCRBits; }
/// hasLazyResolverStub - Return true if accesses to the specified global have
/// to go through a dyld lazy resolution stub. This means that an extra load
/// is required to get the address of the global.
bool hasLazyResolverStub(const GlobalValue *GV) const;
// isLittleEndian - True if generating little-endian code
bool isLittleEndian() const { return IsLittleEndian; }
// Specific obvious features.
bool hasFCPSGN() const { return HasFCPSGN; }
bool hasFSQRT() const { return HasFSQRT; }
bool hasFRE() const { return HasFRE; }
bool hasFRES() const { return HasFRES; }
bool hasFRSQRTE() const { return HasFRSQRTE; }
bool hasFRSQRTES() const { return HasFRSQRTES; }
bool hasRecipPrec() const { return HasRecipPrec; }
bool hasSTFIWX() const { return HasSTFIWX; }
bool hasLFIWAX() const { return HasLFIWAX; }
bool hasFPRND() const { return HasFPRND; }
bool hasFPCVT() const { return HasFPCVT; }
bool hasAltivec() const { return HasAltivec; }
bool hasSPE() const { return HasSPE; }
bool hasFPU() const { return HasFPU; }
bool hasQPX() const { return HasQPX; }
[PowerPC] Initial support for the VSX instruction set VSX is an ISA extension supported on the POWER7 and later cores that enhances floating-point vector and scalar capabilities. Among other things, this adds <2 x double> support and generally helps to reduce register pressure. The interesting part of this ISA feature is the register configuration: there are 64 new 128-bit vector registers, the 32 of which are super-registers of the existing 32 scalar floating-point registers, and the second 32 of which overlap with the 32 Altivec vector registers. This makes things like vector insertion and extraction tricky: this can be free but only if we force a restriction to the right register subclass when needed. A new "minipass" PPCVSXCopy takes care of this (although it could do a more-optimal job of it; see the comment about unnecessary copies below). Please note that, currently, VSX is not enabled by default when targeting anything because it is not yet ready for that. The assembler and disassembler are fully implemented and tested. However: - CodeGen support causes miscompiles; test-suite runtime failures: MultiSource/Benchmarks/FreeBench/distray/distray MultiSource/Benchmarks/McCat/08-main/main MultiSource/Benchmarks/Olden/voronoi/voronoi MultiSource/Benchmarks/mafft/pairlocalalign MultiSource/Benchmarks/tramp3d-v4/tramp3d-v4 SingleSource/Benchmarks/CoyoteBench/almabench SingleSource/Benchmarks/Misc/matmul_f64_4x4 - The lowering currently falls back to using Altivec instructions far more than it should. Worse, there are some things that are scalarized through the stack that shouldn't be. - A lot of unnecessary copies make it past the optimizers, and this needs to be fixed. - Many more regression tests are needed. Normally, I'd fix these things prior to committing, but there are some students and other contributors who would like to work this, and so it makes sense to move this development process upstream where it can be subject to the regular code-review procedures. llvm-svn: 203768
2014-03-13 15:58:58 +08:00
bool hasVSX() const { return HasVSX; }
bool needsTwoConstNR() const { return NeedsTwoConstNR; }
bool hasP8Vector() const { return HasP8Vector; }
bool hasP8Altivec() const { return HasP8Altivec; }
bool hasP8Crypto() const { return HasP8Crypto; }
bool hasP9Vector() const { return HasP9Vector; }
bool hasP9Altivec() const { return HasP9Altivec; }
bool hasMFOCRF() const { return HasMFOCRF; }
bool hasISEL() const { return HasISEL; }
bool hasBPERMD() const { return HasBPERMD; }
bool hasExtDiv() const { return HasExtDiv; }
bool hasCMPB() const { return HasCMPB; }
bool hasLDBRX() const { return HasLDBRX; }
bool isBookE() const { return IsBookE; }
bool hasOnlyMSYNC() const { return HasOnlyMSYNC; }
bool isPPC4xx() const { return IsPPC4xx; }
bool isPPC6xx() const { return IsPPC6xx; }
bool isSecurePlt() const {return SecurePlt; }
bool vectorsUseTwoUnits() const {return VectorsUseTwoUnits; }
bool isE500() const { return IsE500; }
bool isFeatureMFTB() const { return FeatureMFTB; }
bool isDeprecatedDST() const { return DeprecatedDST; }
bool hasICBT() const { return HasICBT; }
[PowerPC] Loosen ELFv1 PPC64 func descriptor loads for indirect calls Function pointers under PPC64 ELFv1 (which is used on PPC64/Linux on the POWER7, A2 and earlier cores) are really pointers to a function descriptor, a structure with three pointers: the actual pointer to the code to which to jump, the pointer to the TOC needed by the callee, and an environment pointer. We used to chain these loads, and make them opaque to the rest of the optimizer, so that they'd always occur directly before the call. This is not necessary, and in fact, highly suboptimal on embedded cores. Once the function pointer is known, the loads can be performed ahead of time; in fact, they can be hoisted out of loops. Now these function descriptors are almost always generated by the linker, and thus the contents of the descriptors are invariant. As a result, by default, we'll mark the associated loads as invariant (allowing them to be hoisted out of loops). I've added a target feature to turn this off, however, just in case someone needs that option (constructing an on-stack descriptor, casting it to a function pointer, and then calling it cannot be well-defined C/C++ code, but I can imagine some JIT-compilation system doing so). Consider this simple test: $ cat call.c typedef void (*fp)(); void bar(fp x) { for (int i = 0; i < 1600000000; ++i) x(); } $ cat main.c typedef void (*fp)(); void bar(fp x); void foo() {} int main() { bar(foo); } On the PPC A2 (the BG/Q supercomputer), marking the function-descriptor loads as invariant brings the execution time down to ~8 seconds from ~32 seconds with the loads in the loop. The difference on the POWER7 is smaller. Compiling with: gcc -std=c99 -O3 -mcpu=native call.c main.c : ~6 seconds [this is 4.8.2] clang -O3 -mcpu=native call.c main.c : ~5.3 seconds clang -O3 -mcpu=native call.c main.c -mno-invariant-function-descriptors : ~4 seconds (looks like we'd benefit from additional loop unrolling here, as a first guess, because this is faster with the extra loads) The -mno-invariant-function-descriptors will be added to Clang shortly. llvm-svn: 226207
2015-01-16 05:17:34 +08:00
bool hasInvariantFunctionDescriptors() const {
return HasInvariantFunctionDescriptors;
}
bool usePPCPreRASchedStrategy() const { return UsePPCPreRASchedStrategy; }
bool usePPCPostRASchedStrategy() const { return UsePPCPostRASchedStrategy; }
bool hasPartwordAtomics() const { return HasPartwordAtomics; }
bool hasDirectMove() const { return HasDirectMove; }
2015-02-25 09:06:45 +08:00
bool isQPXStackUnaligned() const { return IsQPXStackUnaligned; }
Align getPlatformStackAlignment() const {
2015-02-25 09:06:45 +08:00
if ((hasQPX() || isBGQ()) && !isQPXStackUnaligned())
return Align(32);
2015-02-25 09:06:45 +08:00
return Align(16);
2015-02-25 09:06:45 +08:00
}
// DarwinABI has a 224-byte red zone. PPC32 SVR4ABI(Non-DarwinABI) has no
// red zone and PPC64 SVR4ABI has a 288-byte red zone.
unsigned getRedZoneSize() const {
return isDarwinABI() ? 224 : (isPPC64() ? 288 : 0);
}
bool hasHTM() const { return HasHTM; }
bool hasFloat128() const { return HasFloat128; }
bool isISA3_0() const { return IsISA3_0; }
bool useLongCalls() const { return UseLongCalls; }
bool needsSwapsForVSXMemOps() const {
return hasVSX() && isLittleEndian() && !hasP9Vector();
}
2015-02-25 09:06:45 +08:00
POPCNTDKind hasPOPCNTD() const { return HasPOPCNTD; }
const Triple &getTargetTriple() const { return TargetTriple; }
/// isDarwin - True if this is any darwin platform.
bool isDarwin() const { return TargetTriple.isMacOSX(); }
/// isBGQ - True if this is a BG/Q platform.
bool isBGQ() const { return TargetTriple.getVendor() == Triple::BGQ; }
bool isTargetELF() const { return TargetTriple.isOSBinFormatELF(); }
bool isTargetMachO() const { return TargetTriple.isOSBinFormatMachO(); }
bool isTargetLinux() const { return TargetTriple.isOSLinux(); }
bool isDarwinABI() const { return isTargetMachO() || isDarwin(); }
bool isAIXABI() const { return TargetTriple.isOSAIX(); }
bool isSVR4ABI() const { return !isDarwinABI() && !isAIXABI(); }
bool isELFv2ABI() const;
bool is64BitELFABI() const { return isSVR4ABI() && isPPC64(); }
bool is32BitELFABI() const { return isSVR4ABI() && !isPPC64(); }
/// Originally, this function return hasISEL(). Now we always enable it,
/// but may expand the ISEL instruction later.
bool enableEarlyIfConversion() const override { return true; }
/// Scheduling customization.
bool enableMachineScheduler() const override;
/// Pipeliner customization.
bool enableMachinePipeliner() const override;
/// Machine Pipeliner customization
bool useDFAforSMS() const override;
/// This overrides the PostRAScheduler bit in the SchedModel for each CPU.
bool enablePostRAScheduler() const override;
AntiDepBreakMode getAntiDepBreakMode() const override;
void getCriticalPathRCs(RegClassVector &CriticalPathRCs) const override;
void overrideSchedPolicy(MachineSchedPolicy &Policy,
unsigned NumRegionInstrs) const override;
bool useAA() const override;
bool enableSubRegLiveness() const override;
/// True if the GV will be accessed via an indirect symbol.
bool isGVIndirectSymbol(const GlobalValue *GV) const;
bool isXRaySupported() const override { return IsPPC64 && IsLittleEndian; }
};
} // End llvm namespace
#endif