From 398effac36a1cdae8ae2d06cd61a7ade34c154b1 Mon Sep 17 00:00:00 2001 From: Stefan Pintilie Date: Fri, 31 Mar 2023 14:42:58 -0400 Subject: [PATCH] [PowerPC] Add statistics to show the number of entries in the TOC. On Power PC some data is stored in the TOC. This pass adds statistics to show how many entries are emitted to the TOC and what types of entries those are. Reviewed By: amyk Differential Revision: https://reviews.llvm.org/D146325 --- llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp | 125 ++++++++++++-- llvm/test/CodeGen/PowerPC/ppc-TOC-stats.ll | 260 +++++++++++++++++++++++++++++ 2 files changed, 375 insertions(+), 10 deletions(-) create mode 100644 llvm/test/CodeGen/PowerPC/ppc-TOC-stats.ll diff --git a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp index 7c6fd3b..caaea40 100644 --- a/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp +++ b/llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp @@ -28,6 +28,7 @@ #include "TargetInfo/PowerPCTargetInfo.h" #include "llvm/ADT/MapVector.h" #include "llvm/ADT/SmallPtrSet.h" +#include "llvm/ADT/Statistic.h" #include "llvm/ADT/StringRef.h" #include "llvm/ADT/Twine.h" #include "llvm/BinaryFormat/ELF.h" @@ -80,6 +81,17 @@ using namespace llvm::XCOFF; #define DEBUG_TYPE "asmprinter" +STATISTIC(NumTOCEntries, "Number of Total TOC Entries Emitted."); +STATISTIC(NumTOCConstPool, "Number of Constant Pool TOC Entries."); +STATISTIC(NumTOCGlobalInternal, + "Number of Internal Linkage Global TOC Entries."); +STATISTIC(NumTOCGlobalExternal, + "Number of External Linkage Global TOC Entries."); +STATISTIC(NumTOCJumpTable, "Number of Jump Table TOC Entries."); +STATISTIC(NumTOCThreadLocal, "Number of Thread Local TOC Entries."); +STATISTIC(NumTOCBlockAddress, "Number of Block Address TOC Entries."); +STATISTIC(NumTOCEHBlock, "Number of EH Block TOC Entries."); + static cl::opt EnableSSPCanaryBitInTB( "aix-ssp-tb-bit", cl::init(false), cl::desc("Enable Passing SSP Canary info in Trackback on AIX"), cl::Hidden); @@ -148,7 +160,17 @@ public: StringRef getPassName() const override { return "PowerPC Assembly Printer"; } - MCSymbol *lookUpOrCreateTOCEntry(const MCSymbol *Sym, + enum TOCEntryType { + TOCType_ConstantPool, + TOCType_GlobalExternal, + TOCType_GlobalInternal, + TOCType_JumpTable, + TOCType_ThreadLocal, + TOCType_BlockAddress, + TOCType_EHBlock + }; + + MCSymbol *lookUpOrCreateTOCEntry(const MCSymbol *Sym, TOCEntryType Type, MCSymbolRefExpr::VariantKind Kind = MCSymbolRefExpr::VariantKind::VK_None); @@ -412,12 +434,43 @@ bool PPCAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI, unsigned OpNo, return false; } +static void collectTOCStats(PPCAsmPrinter::TOCEntryType Type) { + ++NumTOCEntries; + switch (Type) { + case PPCAsmPrinter::TOCType_ConstantPool: + ++NumTOCConstPool; + break; + case PPCAsmPrinter::TOCType_GlobalInternal: + ++NumTOCGlobalInternal; + break; + case PPCAsmPrinter::TOCType_GlobalExternal: + ++NumTOCGlobalExternal; + break; + case PPCAsmPrinter::TOCType_JumpTable: + ++NumTOCJumpTable; + break; + case PPCAsmPrinter::TOCType_ThreadLocal: + ++NumTOCThreadLocal; + break; + case PPCAsmPrinter::TOCType_BlockAddress: + ++NumTOCBlockAddress; + break; + case PPCAsmPrinter::TOCType_EHBlock: + ++NumTOCEHBlock; + break; + } +} + /// lookUpOrCreateTOCEntry -- Given a symbol, look up whether a TOC entry /// exists for it. If not, create one. Then return a symbol that references /// the TOC entry. MCSymbol * -PPCAsmPrinter::lookUpOrCreateTOCEntry(const MCSymbol *Sym, +PPCAsmPrinter::lookUpOrCreateTOCEntry(const MCSymbol *Sym, TOCEntryType Type, MCSymbolRefExpr::VariantKind Kind) { + // If this is a new TOC entry add statistics about it. + if (!TOC.contains({Sym, Kind})) + collectTOCStats(Type); + MCSymbol *&TOCEntry = TOC[{Sym, Kind}]; if (!TOCEntry) TOCEntry = createTempSymbol("C"); @@ -648,6 +701,48 @@ static MCSymbol *getMCSymbolForTOCPseudoMO(const MachineOperand &MO, } } +static bool hasTLSFlag(const MachineOperand &MO) { + unsigned Flags = MO.getTargetFlags(); + if (Flags & PPCII::MO_TLSGD_FLAG || Flags & PPCII::MO_TPREL_FLAG || + Flags & PPCII::MO_TLSLD_FLAG || Flags & PPCII::MO_TLSGDM_FLAG) + return true; + + if (Flags == PPCII::MO_TPREL_LO || Flags == PPCII::MO_TPREL_HA || + Flags == PPCII::MO_DTPREL_LO || Flags == PPCII::MO_TLSLD_LO || + Flags == PPCII::MO_TLS) + return true; + + return false; +} + +static PPCAsmPrinter::TOCEntryType +getTOCEntryTypeForMO(const MachineOperand &MO) { + // Use the target flags to determine if this MO is Thread Local. + // If we don't do this it comes out as Global. + if (hasTLSFlag(MO)) + return PPCAsmPrinter::TOCType_ThreadLocal; + + switch (MO.getType()) { + case MachineOperand::MO_GlobalAddress: { + const GlobalValue *GlobalV = MO.getGlobal(); + GlobalValue::LinkageTypes Linkage = GlobalV->getLinkage(); + if (Linkage == GlobalValue::ExternalLinkage || + Linkage == GlobalValue::AvailableExternallyLinkage || + Linkage == GlobalValue::ExternalWeakLinkage) + return PPCAsmPrinter::TOCType_GlobalExternal; + + return PPCAsmPrinter::TOCType_GlobalInternal; + } + case MachineOperand::MO_ConstantPoolIndex: + return PPCAsmPrinter::TOCType_ConstantPool; + case MachineOperand::MO_JumpTableIndex: + return PPCAsmPrinter::TOCType_JumpTable; + case MachineOperand::MO_BlockAddress: + return PPCAsmPrinter::TOCType_BlockAddress; + default: + llvm_unreachable("Unexpected operand type to get TOC type."); + } +} /// EmitInstruction -- Print out a single PowerPC MI in Darwin syntax to /// the current output stream. /// @@ -865,7 +960,8 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) { // Otherwise, use the TOC. 'TOCEntry' is a label used to reference the // storage allocated in the TOC which contains the address of // 'MOSymbol'. Said TOC entry will be synthesized later. - MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(MOSymbol, VK); + MCSymbol *TOCEntry = + lookUpOrCreateTOCEntry(MOSymbol, getTOCEntryTypeForMO(MO), VK); const MCExpr *Exp = MCSymbolRefExpr::create(TOCEntry, MCSymbolRefExpr::VK_None, OutContext); @@ -942,7 +1038,8 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) { // Map the machine operand to its corresponding MCSymbol, then map the // global address operand to be a reference to the TOC entry we will // synthesize later. - MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(MOSymbol, VK); + MCSymbol *TOCEntry = + lookUpOrCreateTOCEntry(MOSymbol, getTOCEntryTypeForMO(MO), VK); MCSymbolRefExpr::VariantKind VKExpr = IsAIX ? MCSymbolRefExpr::VK_None : MCSymbolRefExpr::VK_PPC_TOC; @@ -980,7 +1077,8 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) { // to the TOC entry we will synthesize later. 'TOCEntry' is a label used to // reference the storage allocated in the TOC which contains the address of // 'MOSymbol'. - MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(MOSymbol, VK); + MCSymbol *TOCEntry = + lookUpOrCreateTOCEntry(MOSymbol, getTOCEntryTypeForMO(MO), VK); const MCExpr *Exp = MCSymbolRefExpr::create(TOCEntry, MCSymbolRefExpr::VK_PPC_U, OutContext); @@ -1012,7 +1110,8 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) { // to the TOC entry we will synthesize later. 'TOCEntry' is a label used to // reference the storage allocated in the TOC which contains the address of // 'MOSymbol'. - MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(MOSymbol, VK); + MCSymbol *TOCEntry = + lookUpOrCreateTOCEntry(MOSymbol, getTOCEntryTypeForMO(MO), VK); const MCExpr *Exp = MCSymbolRefExpr::create(TOCEntry, MCSymbolRefExpr::VK_PPC_L, OutContext); @@ -1042,7 +1141,7 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) { MO.isGlobal() && Subtarget->isGVIndirectSymbol(MO.getGlobal()); if (GlobalToc || MO.isJTI() || MO.isBlockAddress() || (MO.isCPI() && TM.getCodeModel() == CodeModel::Large)) - MOSymbol = lookUpOrCreateTOCEntry(MOSymbol, VK); + MOSymbol = lookUpOrCreateTOCEntry(MOSymbol, getTOCEntryTypeForMO(MO), VK); VK = IsAIX ? MCSymbolRefExpr::VK_PPC_U : MCSymbolRefExpr::VK_PPC_TOC_HA; @@ -1084,7 +1183,7 @@ void PPCAsmPrinter::emitInstruction(const MachineInstr *MI) { MCSymbolRefExpr::VariantKind VK = GetVKForMO(MO); if (!MO.isCPI() || TM.getCodeModel() == CodeModel::Large) - MOSymbol = lookUpOrCreateTOCEntry(MOSymbol, VK); + MOSymbol = lookUpOrCreateTOCEntry(MOSymbol, getTOCEntryTypeForMO(MO), VK); VK = IsAIX ? MCSymbolRefExpr::VK_PPC_L : MCSymbolRefExpr::VK_PPC_TOC_LO; const MCExpr *Exp = @@ -2312,7 +2411,7 @@ void PPCAIXAsmPrinter::emitTracebackTable() { auto &Ctx = OutStreamer->getContext(); MCSymbol *EHInfoSym = TargetLoweringObjectFileXCOFF::getEHInfoTableSymbol(MF); - MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(EHInfoSym); + MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(EHInfoSym, TOCType_EHBlock); const MCSymbol *TOCBaseSym = cast(getObjFileLowering().getTOCBaseSection()) ->getQualNameSymbol(); @@ -2792,8 +2891,14 @@ void PPCAIXAsmPrinter::emitXXStructorList(const DataLayout &DL, void PPCAIXAsmPrinter::emitTTypeReference(const GlobalValue *GV, unsigned Encoding) { if (GV) { + TOCEntryType GlobalType = TOCType_GlobalInternal; + GlobalValue::LinkageTypes Linkage = GV->getLinkage(); + if (Linkage == GlobalValue::ExternalLinkage || + Linkage == GlobalValue::AvailableExternallyLinkage || + Linkage == GlobalValue::ExternalWeakLinkage) + GlobalType = TOCType_GlobalExternal; MCSymbol *TypeInfoSym = TM.getSymbol(GV); - MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(TypeInfoSym); + MCSymbol *TOCEntry = lookUpOrCreateTOCEntry(TypeInfoSym, GlobalType); const MCSymbol *TOCBaseSym = cast(getObjFileLowering().getTOCBaseSection()) ->getQualNameSymbol(); diff --git a/llvm/test/CodeGen/PowerPC/ppc-TOC-stats.ll b/llvm/test/CodeGen/PowerPC/ppc-TOC-stats.ll new file mode 100644 index 0000000..9e5fee7 --- /dev/null +++ b/llvm/test/CodeGen/PowerPC/ppc-TOC-stats.ll @@ -0,0 +1,260 @@ +;; Note: The checks for this test are manually generated. Please do not +;; run a script to update these checks. + +; RUN: llc -verify-machineinstrs -mtriple powerpc-ibm-aix-xcoff -mcpu=pwr8 \ +; RUN: --stats < %s 2>&1 | FileCheck %s --check-prefix=AIX +; RUN: llc -verify-machineinstrs -mtriple powerpc64-ibm-aix-xcoff -mcpu=pwr8 \ +; RUN: --stats < %s 2>&1 | FileCheck %s --check-prefix=AIX +; RUN: llc -verify-machineinstrs -mtriple powerpc64-unknown-linux -mcpu=pwr8 \ +; RUN: --stats < %s 2>&1 | FileCheck %s --check-prefix=LINUX +; RUN: llc -verify-machineinstrs -mtriple powerpc64le-unknown-linux -mcpu=pwr8 \ +; RUN: --stats < %s 2>&1 | FileCheck %s --check-prefix=LINUX +; RUN: llc -verify-machineinstrs -mtriple powerpc64-unknown-linux -mcpu=pwr8 \ +; RUN: -code-model=large --stats < %s 2>&1 | FileCheck %s --check-prefix=LINUXLARGE +; RUN: llc -verify-machineinstrs -mtriple powerpc64le-unknown-linux -mcpu=pwr8 \ +; RUN: -code-model=large --stats < %s 2>&1 | FileCheck %s --check-prefix=LINUXLARGE + + +; The purpose of this test is to check that the statistics about the TOC are +; are collected correctly. This test tries to have all of the different types +; of TOC entries. + +; AIX: Statistics Collected +; AIX: 1 asmprinter - Number of Block Address TOC Entries. +; AIX: 13 asmprinter - Number of Constant Pool TOC Entries. +; AIX: 1 asmprinter - Number of EH Block TOC Entries. +; AIX: 25 asmprinter - Number of Total TOC Entries Emitted. +; AIX: 3 asmprinter - Number of External Linkage Global TOC Entries. +; AIX: 2 asmprinter - Number of Internal Linkage Global TOC Entries. +; AIX: 1 asmprinter - Number of Jump Table TOC Entries. +; AIX: 4 asmprinter - Number of Thread Local TOC Entries. + +; LINUX: Statistics Collected +; LINUX: 1 asmprinter - Number of Block Address TOC Entries. +; LINUX: 5 asmprinter - Number of Total TOC Entries Emitted. +; LINUX: 3 asmprinter - Number of External Linkage Global TOC Entries. +; LINUX: 1 asmprinter - Number of Jump Table TOC Entries. + +; LINUXLARGE: Statistics Collected +; LINUXLARGE: 1 asmprinter - Number of Block Address TOC Entries. +; LINUXLARGE: 13 asmprinter - Number of Constant Pool TOC Entries. +; LINUXLARGE: 20 asmprinter - Number of Total TOC Entries Emitted. +; LINUXLARGE: 3 asmprinter - Number of External Linkage Global TOC Entries. +; LINUXLARGE: 2 asmprinter - Number of Internal Linkage Global TOC Entries. +; LINUXLARGE: 1 asmprinter - Number of Jump Table TOC Entries. + + +@gDouble = local_unnamed_addr global double 0.000000e+00, align 8 +@TLS1 = thread_local global i32 0, align 4 +@TLS2 = external thread_local global float, align 4 +@_ZTIi = external constant ptr +@_ZL2G4 = internal unnamed_addr global i32 0, align 4 +@_ZZ9incrementvE7Element = internal unnamed_addr global i32 0, align 4 + +define noundef double @testd1() local_unnamed_addr { +entry: + ret double 3.784320e+02 +} + +define noundef float @testf1() local_unnamed_addr { +entry: + ret float 0x40039999A0000000 +} + +define noundef double @testd2() local_unnamed_addr { +entry: + ret double 6.920000e+00 +} + +define noundef <4 x i32> @testv1() local_unnamed_addr { +entry: + ret <4 x i32> +} + +define noundef double @testd3() local_unnamed_addr { +entry: + %call = tail call noundef double @calleeddf(double noundef 4.582600e+01, double noundef 0x40564F0A3D70A3D7, float noundef 0x402225E360000000) + ret double %call +} + +declare noundef double @calleeddf(double noundef, double noundef, float noundef) local_unnamed_addr + +define noundef i64 @testi() local_unnamed_addr { +entry: + ret i64 893471915 +} + +define noundef double @testld1() local_unnamed_addr { +entry: + ret double 0x417179806D5CDEBE +} + +define noundef signext i32 @testJT(i32 noundef signext %in) { +entry: + switch i32 %in, label %sw.epilog [ + i32 45, label %return + i32 86, label %sw.bb1 + i32 91, label %sw.bb2 + i32 101, label %sw.bb3 + i32 107, label %sw.bb4 + i32 76832712, label %sw.bb5 + ] + +sw.bb1: ; preds = %entry + br label %return + +sw.bb2: ; preds = %entry + br label %return + +sw.bb3: ; preds = %entry + br label %return + +sw.bb4: ; preds = %entry + br label %return + +sw.bb5: ; preds = %entry + br label %return + +sw.epilog: ; preds = %entry + br label %return + +return: ; preds = %entry, %sw.epilog, %sw.bb5, %sw.bb4, %sw.bb3, %sw.bb2, %sw.bb1 + %retval.0 = phi i32 [ 0, %sw.epilog ], [ 222, %sw.bb5 ], [ 15, %sw.bb4 ], [ 11, %sw.bb3 ], [ 17, %sw.bb2 ], [ 16, %sw.bb1 ], [ 19, %entry ] + ret i32 %retval.0 +} + +define void @setGDouble(double noundef %a) local_unnamed_addr { +entry: + store double %a, ptr @gDouble, align 8 + ret void +} + +define noundef double @getGDouble() local_unnamed_addr { +entry: + %0 = load double, ptr @gDouble, align 8 + ret double %0 +} + +define noundef signext i32 @testTLS(i32 noundef signext %a, i32 noundef signext %b, ptr nocapture noundef readonly %fIn) { +entry: + %add = add nsw i32 %b, %a + %0 = tail call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @TLS1) + store i32 %add, ptr %0, align 4 + tail call void @calleef(ptr noundef nonnull @testJT) + %1 = load i32, ptr %0, align 4 + %call = tail call noundef signext i32 %fIn(i32 noundef signext %1) + %2 = load i32, ptr %0, align 4 + %add1 = add nsw i32 %2, %call + ret i32 %add1 +} + +define float @getTLS2() local_unnamed_addr { +entry: + %0 = tail call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @TLS2) + %1 = load float, ptr %0, align 4 + ret float %1 +} + +declare nonnull ptr @llvm.threadlocal.address.p0(ptr nonnull) +declare void @calleef(ptr noundef) local_unnamed_addr + +define noundef signext i32 @testEH(i32 noundef signext %i) personality ptr @__xlcxx_personality_v1 { +entry: + %cmp = icmp slt i32 %i, 1 + br i1 %cmp, label %if.then, label %try.cont + +if.then: ; preds = %entry + %exception = tail call ptr @__cxa_allocate_exception(i64 4) + store i32 10, ptr %exception, align 16 + invoke void @__cxa_throw(ptr nonnull %exception, ptr nonnull @_ZTIi, ptr null) + to label %unreachable unwind label %lpad + +lpad: ; preds = %if.then + %0 = landingpad { ptr, i32 } + cleanup + catch ptr @_ZTIi + %1 = extractvalue { ptr, i32 } %0, 1 + %2 = tail call i32 @llvm.eh.typeid.for(ptr nonnull @_ZTIi) + %matches = icmp eq i32 %1, %2 + br i1 %matches, label %catch, label %ehcleanup + +catch: ; preds = %lpad + %3 = extractvalue { ptr, i32 } %0, 0 + %4 = tail call ptr @__cxa_begin_catch(ptr %3) + tail call void @__cxa_end_catch() + br label %try.cont + +try.cont: ; preds = %entry, %catch + %ret.0 = phi i32 [ 1, %catch ], [ 0, %entry ] + ret i32 %ret.0 + +ehcleanup: ; preds = %lpad + resume { ptr, i32 } %0 + +unreachable: ; preds = %if.then + unreachable +} + +declare ptr @__cxa_allocate_exception(i64) local_unnamed_addr +declare void @__cxa_throw(ptr, ptr, ptr) local_unnamed_addr +declare i32 @__xlcxx_personality_v1(...) +declare i32 @llvm.eh.typeid.for(ptr) +declare ptr @__cxa_begin_catch(ptr) local_unnamed_addr +declare void @__cxa_end_catch() local_unnamed_addr + +define noundef double @testISel1(i32 noundef signext %i) { +entry: + %cmp = icmp slt i32 %i, 0 + %cmp1.not = icmp eq i32 %i, 0 + %. = select i1 %cmp1.not, double 1.618030e+00, double 2.718280e+00 + %retval.0 = select i1 %cmp, double 3.141590e+00, double %. + ret double %retval.0 +} + +define noundef double @testISel2(i32 noundef signext %i) { +entry: + %cmp = icmp slt i32 %i, 0 + br i1 %cmp, label %return, label %if.else + +if.else: ; preds = %entry + %cmp1 = icmp ugt i32 %i, 30 + br i1 %cmp1, label %return, label %if.else3 + +if.else3: ; preds = %if.else + %cmp4 = icmp ugt i32 %i, 5 + %. = select i1 %cmp4, double 8.644600e+00, double 1.618030e+00 + br label %return + +return: ; preds = %if.else3, %if.else, %entry + %retval.0 = phi double [ 3.141590e+00, %entry ], [ 2.718280e+00, %if.else ], [ %., %if.else3 ] + ret double %retval.0 +} + + +define i8* @testBlockAddr() { +entry: + br label %here + +here: + ret i8* blockaddress(@testBlockAddr, %here) +} + +define noundef signext i32 @_Z5getG4v() local_unnamed_addr { +entry: + %0 = load i32, ptr @_ZL2G4, align 4 + ret i32 %0 +} + +define void @_Z5setG4i(i32 noundef signext %value) local_unnamed_addr { +entry: + store i32 %value, ptr @_ZL2G4, align 4 + ret void +} + +define noundef signext i32 @_Z9incrementv() local_unnamed_addr { +entry: + %0 = load i32, ptr @_ZZ9incrementvE7Element, align 4 + %inc = add nsw i32 %0, 1 + store i32 %inc, ptr @_ZZ9incrementvE7Element, align 4 + ret i32 %inc +} -- 2.7.4