AMDGPU/SI: Use .hsatext section instead of .text for HSA

[• rafael]

lib/Target/AMDGPU/AMDGPUHSATargetObjectFile.cpp

31

This is probably wrong with comdats.

[• tstellarAMD]

Updated patch:

• Make sure assembly files start with .hsatext
• Possible fix for Comdats

lib/Target/AMDGPU/AMDGPUHSATargetObjectFile.h

+//===-- AMDGPUHSATargetObjectFile.h - AMDGPU HSA Object Info ----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// \brief This file declares the AMDGPU-specific subclass of
+/// TargetLoweringObjectFile use for targeting the HSA-runtime.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIB_TARGET_AMDGPU_AMDGPUHSATARGETOBJECTFILE_H
+#define LLVM_LIB_TARGET_AMDGPU_AMDGPUHSATARGETOBJECTFILE_H
+
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
+#include "llvm/Target/TargetMachine.h"
+
+namespace llvm {
+
+class AMDGPUHSATargetObjectFile final : public TargetLoweringObjectFileELF {
+public:
+  void Initialize(MCContext &Ctx, const TargetMachine &TM);
+
+  MCSection *SelectSectionForGlobal(const GlobalValue *GV, SectionKind Kind,
+                                    Mangler &Mang,
+                                    const TargetMachine &TM) const override;
+};
+
+} // end namespace llvm
+
+#endif

lib/Target/AMDGPU/AMDGPUHSATargetObjectFile.cpp

+//===-- AMDGPUHSATargetObjectFile.cpp - AMDGPU Object Files ---------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUHSATargetObjectFile.h"
+#include "Utils/AMDGPUBaseInfo.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCSectionELF.h"
+#include "llvm/Support/ELF.h"
+
+using namespace llvm;
+
+void AMDGPUHSATargetObjectFile::Initialize(MCContext &Ctx,
+                                           const TargetMachine &TM){
+  TargetLoweringObjectFileELF::Initialize(Ctx, TM);
+  InitializeELF(TM.Options.UseInitArray);
+
+  TextSection = AMDGPU::getHSATextSection(Ctx);
+
+}
+
+MCSection *AMDGPUHSATargetObjectFile::SelectSectionForGlobal(
+                                        const GlobalValue *GV, SectionKind Kind,
+                                        Mangler &Mang,
+                                        const TargetMachine &TM) const {
+  if (Kind.isText())
+    return getTextSection();
+
+  return TargetLoweringObjectFileELF::SelectSectionForGlobal(GV, Kind, Mang, TM);
+}

lib/Target/AMDGPU/AMDGPUTargetMachine.h

 //===----------------------------------------------------------------------===//
 // AMDGPU Target Machine (R600+)
 //===----------------------------------------------------------------------===//
 
 class AMDGPUTargetMachine : public LLVMTargetMachine {
 private:
 
 protected:
-  TargetLoweringObjectFile *TLOF;
+  std::unique_ptr<TargetLoweringObjectFile> TLOF;
   AMDGPUSubtarget Subtarget;
   AMDGPUIntrinsicInfo IntrinsicInfo;
 
 public:
   AMDGPUTargetMachine(const Target &T, const Triple &TT, StringRef FS,
                       StringRef CPU, TargetOptions Options, Reloc::Model RM,
                       CodeModel::Model CM, CodeGenOpt::Level OL);
   ~AMDGPUTargetMachine();
 
   const AMDGPUSubtarget *getSubtargetImpl() const { return &Subtarget; }
   const AMDGPUSubtarget *getSubtargetImpl(const Function &) const override {
     return &Subtarget;
   }
   const AMDGPUIntrinsicInfo *getIntrinsicInfo() const override {
     return &IntrinsicInfo;
   }
   TargetIRAnalysis getTargetIRAnalysis() override;
 
   TargetLoweringObjectFile *getObjFileLowering() const override {
-    return TLOF;
+    return TLOF.get();
   }
 };
 
 //===----------------------------------------------------------------------===//
 // R600 Target Machine (R600 -> Cayman)
 //===----------------------------------------------------------------------===//
 
 class R600TargetMachine : public AMDGPUTargetMachine {

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

 //===-- AMDGPUTargetMachine.cpp - TargetMachine for hw codegen targets-----===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file
 /// \brief The AMDGPU target machine contains all of the hardware specific
 /// information  needed to emit code for R600 and SI GPUs.
 //
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPUTargetMachine.h"
+#include "AMDGPUHSATargetObjectFile.h"
 #include "AMDGPU.h"
 #include "AMDGPUTargetTransformInfo.h"
 #include "R600ISelLowering.h"
 #include "R600InstrInfo.h"
 #include "R600MachineScheduler.h"
 #include "SIISelLowering.h"
 #include "SIInstrInfo.h"
 #include "llvm/Analysis/Passes.h"

 using namespace llvm;
 
 extern "C" void LLVMInitializeAMDGPUTarget() {
   // Register the target
   RegisterTargetMachine<R600TargetMachine> X(TheAMDGPUTarget);
   RegisterTargetMachine<GCNTargetMachine> Y(TheGCNTarget);
 }
 
+static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
+  if (TT.getOS() == Triple::AMDHSA)
+    return make_unique<AMDGPUHSATargetObjectFile>();
+
+  return make_unique<TargetLoweringObjectFileELF>();
+}
+
 static ScheduleDAGInstrs *createR600MachineScheduler(MachineSchedContext *C) {
   return new ScheduleDAGMILive(C, make_unique<R600SchedStrategy>());
 }
 
 static MachineSchedRegistry
 SchedCustomRegistry("r600", "Run R600's custom scheduler",
                     createR600MachineScheduler);
 

 
 AMDGPUTargetMachine::AMDGPUTargetMachine(const Target &T, const Triple &TT,
                                          StringRef CPU, StringRef FS,
                                          TargetOptions Options, Reloc::Model RM,
                                          CodeModel::Model CM,
                                          CodeGenOpt::Level OptLevel)
     : LLVMTargetMachine(T, computeDataLayout(TT), TT, CPU, FS, Options, RM, CM,
                         OptLevel),
-      TLOF(new TargetLoweringObjectFileELF()), Subtarget(TT, CPU, FS, *this),
+      TLOF(createTLOF(getTargetTriple())), Subtarget(TT, CPU, FS, *this),
       IntrinsicInfo() {
   setRequiresStructuredCFG(true);
   initAsmInfo();
 }
 
-AMDGPUTargetMachine::~AMDGPUTargetMachine() {
-  delete TLOF;
-}
+AMDGPUTargetMachine::~AMDGPUTargetMachine() { }
 
 //===----------------------------------------------------------------------===//
 // R600 Target Machine (R600 -> Cayman)
 //===----------------------------------------------------------------------===//
 
 R600TargetMachine::R600TargetMachine(const Target &T, const Triple &TT,
                                      StringRef FS, StringRef CPU,
                                      TargetOptions Options, Reloc::Model RM,

lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

#include "AMDGPUGenAsmMatcher.inc"
   /// }
 
 private:
   bool ParseDirectiveMajorMinor(uint32_t &Major, uint32_t &Minor);
   bool ParseDirectiveHSACodeObjectVersion();
   bool ParseDirectiveHSACodeObjectISA();
   bool ParseAMDKernelCodeTValue(StringRef ID, amd_kernel_code_t &Header);
   bool ParseDirectiveAMDKernelCodeT();
+  bool ParseSectionDirectiveHSAText();
 
 public:
   AMDGPUAsmParser(MCSubtargetInfo &STI, MCAsmParser &_Parser,
                const MCInstrInfo &MII,
                const MCTargetOptions &Options)
       : MCTargetAsmParser(Options), STI(STI), MII(MII), Parser(_Parser),
         ForcedEncodingSize(0){
 
if (ParseAMDKernelCodeTValue(ID, Header))
       return true;
   }
 
   getTargetStreamer().EmitAMDKernelCodeT(Header);
 
   return false;
 }
 
+bool AMDGPUAsmParser::ParseSectionDirectiveHSAText() {
+  getParser().getStreamer().SwitchSection(
+      AMDGPU::getHSATextSection(getContext()));
+  return false;
+}
+
 bool AMDGPUAsmParser::ParseDirective(AsmToken DirectiveID) {
   StringRef IDVal = DirectiveID.getString();
 
   if (IDVal == ".hsa_code_object_version")
     return ParseDirectiveHSACodeObjectVersion();
 
   if (IDVal == ".hsa_code_object_isa")
     return ParseDirectiveHSACodeObjectISA();
 
   if (IDVal == ".amd_kernel_code_t")
     return ParseDirectiveAMDKernelCodeT();
 
+  if (IDVal == ".hsatext" || IDVal == ".text")
+    return ParseSectionDirectiveHSAText();
+
   return true;
 }
 
 static bool operandsHaveModifiers(const OperandVector &Operands) {
 
   for (unsigned i = 0, e = Operands.size(); i != e; ++i) {
     const AMDGPUOperand &Op = ((AMDGPUOperand&)*Operands[i]);
     if (Op.isRegKind() && Op.hasModifiers())

lib/Target/AMDGPU/CMakeLists.txt

 tablegen(LLVM AMDGPUGenAsmMatcher.inc -gen-asm-matcher)
 add_public_tablegen_target(AMDGPUCommonTableGen)
 
 add_llvm_target(AMDGPUCodeGen
   AMDILCFGStructurizer.cpp
   AMDGPUAlwaysInlinePass.cpp
   AMDGPUAsmPrinter.cpp
   AMDGPUFrameLowering.cpp
+  AMDGPUHSATargetObjectFile.cpp
   AMDGPUIntrinsicInfo.cpp
   AMDGPUISelDAGToDAG.cpp
   AMDGPUMCInstLower.cpp
   AMDGPUMachineFunction.cpp
   AMDGPUOpenCLImageTypeLoweringPass.cpp
   AMDGPUSubtarget.cpp
   AMDGPUTargetMachine.cpp
   AMDGPUTargetTransformInfo.cpp

lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCAsmInfo.h

 // If you need to create another MCAsmInfo class, which inherits from MCAsmInfo,
 // you will need to make sure your new class sets PrivateGlobalPrefix to
 // a prefix that won't appear in a function name.  The default value
 // for PrivateGlobalPrefix is 'L', so it will consider any function starting
 // with 'L' as a local symbol.
 class AMDGPUMCAsmInfo : public MCAsmInfoELF {
 public:
   explicit AMDGPUMCAsmInfo(const Triple &TT);
+  bool shouldOmitSectionDirective(StringRef SectionName) const override;
 };
 } // namespace llvm
 #endif

lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCAsmInfo.cpp

AMDGPUMCAsmInfo::AMDGPUMCAsmInfo(const Triple &TT) : MCAsmInfoELF() {
   HasAggressiveSymbolFolding = true;
   COMMDirectiveAlignmentIsInBytes = false;
   HasDotTypeDotSizeDirective = false;
   HasNoDeadStrip = true;
   WeakRefDirective = ".weakref\t";
   //===--- Dwarf Emission Directives -----------------------------------===//
   SupportsDebugInformation = true;
 }
+
+bool AMDGPUMCAsmInfo::shouldOmitSectionDirective(StringRef SectionName) const {
+  return SectionName == ".hsatext" ||
+         MCAsmInfo::shouldOmitSectionDirective(SectionName);
+}

lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h

 #ifndef LLVM_LIB_TARGET_AMDGPU_UTILS_AMDGPUBASEINFO_H
 #define LLVM_LIB_TARGET_AMDGPU_UTILS_AMDGPUBASEINFO_H
 
 #include "AMDKernelCodeT.h"
 
 namespace llvm {
 
 class FeatureBitset;
+class MCContext;
+class MCSection;
 
 namespace AMDGPU {
 
 struct IsaVersion {
   unsigned Major;
   unsigned Minor;
   unsigned Stepping;
 };
 
 IsaVersion getIsaVersion(const FeatureBitset &Features);
 void initDefaultAMDKernelCodeT(amd_kernel_code_t &Header,
                                const FeatureBitset &Features);
+MCSection *getHSATextSection(MCContext &Ctx);
 
 } // end namespace AMDGPU
 } // end namespace llvm
 
 #endif

lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp

 //===-- AMDGPUBaseInfo.cpp - AMDGPU Base encoding information--------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 #include "AMDGPUBaseInfo.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCSectionELF.h"
 #include "llvm/MC/SubtargetFeature.h"
 
 #define GET_SUBTARGETINFO_ENUM
 #include "AMDGPUGenSubtargetInfo.inc"
 #undef GET_SUBTARGETINFO_ENUM
 
 namespace llvm {
 namespace AMDGPU {
void initDefaultAMDKernelCodeT(amd_kernel_code_t &Header,
   Header.wavefront_size = 6;
   // These alignment values are specified in powers of two, so alignment =
   // 2^n.  The minimum alignment is 2^4 = 16.
   Header.kernarg_segment_alignment = 4;
   Header.group_segment_alignment = 4;
   Header.private_segment_alignment = 4;
 }
 
+MCSection *getHSATextSection(MCContext &Ctx) {
+  return Ctx.getELFSection(".hsatext", ELF::SHT_PROGBITS,
+                           ELF::SHF_ALLOC | ELF::SHF_WRITE |
+                           ELF::SHF_EXECINSTR |
+                           ELF::SHF_AMDGPU_HSA_AGENT |
+                           ELF::SHF_AMDGPU_HSA_CODE);
+}
+
 } // End namespace AMDGPU
 } // End namespace llvm

lib/Target/AMDGPU/Utils/LLVMBuild.txt

 ;   http://llvm.org/docs/LLVMBuild.html
 ;
 ;===------------------------------------------------------------------------===;
 
 [component_0]
 type = Library
 name = AMDGPUUtils
 parent = AMDGPU
-required_libraries = Support
+required_libraries = MC Support
 add_to_library_groups = AMDGPU

test/CodeGen/AMDGPU/hsa.ll

 ; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | FileCheck --check-prefix=HSA-CI --check-prefix=HSA %s
 ; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=carrizo | FileCheck --check-prefix=HSA-VI --check-prefix=HSA %s
 ; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri -filetype=obj  | llvm-readobj -s -sd | FileCheck --check-prefix=ELF %s
 ; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | llvm-mc -filetype=obj -triple amdgcn--amdhsa -mcpu=kaveri | llvm-readobj -s -sd | FileCheck %s --check-prefix=ELF
 
 ; The SHT_NOTE section contains the output from the .hsa_code_object_*
 ; directives.
 
+; ELF: Section {
+; ELF: Name: .hsatext
+; ELF: Type: SHT_PROGBITS (0x1)
+; ELF: Flags [ (0xC00007)
+; ELF: SHF_ALLOC (0x2)
+; ELF: SHF_AMDGPU_HSA_AGENT (0x800000)
+; ELF: SHF_AMDGPU_HSA_CODE (0x400000)
+; ELF: SHF_EXECINSTR (0x4)
+; ELF: SHF_WRITE (0x1)
+; ELF: }
+
 ; ELF: SHT_NOTE
 ; ELF: 0000: 04000000 08000000 01000000 414D4400
 ; ELF: 0010: 01000000 00000000 04000000 1B000000
 ; ELF: 0020: 03000000 414D4400 04000700 07000000
 ; ELF: 0030: 00000000 00000000 414D4400 414D4447
 ; ELF: 0040: 50550000
 
 ; HSA: .hsa_code_object_version 1,0
 ; HSA-CI: .hsa_code_object_isa 7,0,0,"AMD","AMDGPU"
 ; HSA-VI: .hsa_code_object_isa 8,0,1,"AMD","AMDGPU"
 
+; HSA: .hsatext
+
 ; HSA: {{^}}simple:
 ; HSA: .amd_kernel_code_t
 ; HSA: .end_amd_kernel_code_t
 ; HSA: s_load_dwordx2 s[{{[0-9]+:[0-9]+}}], s[0:1], 0x0
 
 ; Make sure we are setting the ATC bit:
 ; HSA-CI: s_mov_b32 s[[HI:[0-9]]], 0x100f000
 ; On VI+ we also need to set MTYPE = 2
 ; HSA-VI: s_mov_b32 s[[HI:[0-9]]], 0x1100f000
 ; HSA: buffer_store_dword v{{[0-9]+}}, s[0:[[HI]]], 0
 
 define void @simple(i32 addrspace(1)* %out) {
 entry:
   store i32 0, i32 addrspace(1)* %out
   ret void
 }

test/MC/AMDGPU/hsa-text.s

+// RUN: llvm-mc -triple amdgcn--amdhsa -mcpu=kaveri -show-encoding %s | FileCheck %s --check-prefix=GCN
+
+// For compatibility reasons we treat convert .text sections to .hsatext
+
+.text
+// GCN: .hsatext

test/MC/AMDGPU/hsa.s

 // RUN: llvm-mc -triple amdgcn--amdhsa -mcpu=kaveri -show-encoding %s | FileCheck %s --check-prefix=ASM
 // RUN: llvm-mc -filetype=obj -triple amdgcn--amdhsa -mcpu=kaveri -show-encoding %s | llvm-readobj -s -sd | FileCheck %s --check-prefix=ELF
 
 // ELF: SHT_NOTE
 // ELF: 0000: 04000000 08000000 01000000 414D4400
 // ELF: 0010: 01000000 00000000 04000000 1B000000
 // ELF: 0020: 03000000 414D4400 04000700 07000000
 // ELF: 0030: 00000000 00000000 414D4400 414D4447
 // ELF: 0040: 50550000
 
+// ELF: Section {
+// ELF: Name: .hsatext
+// ELF: Type: SHT_PROGBITS (0x1)
+// ELF: Flags [ (0xC00007)
+// ELF: SHF_ALLOC (0x2)
+// ELF: SHF_AMDGPU_HSA_AGENT (0x800000)
+// ELF: SHF_AMDGPU_HSA_CODE (0x400000)
+// ELF: SHF_EXECINSTR (0x4)
+// ELF: SHF_WRITE (0x1)
+// ELF: }
+
 .hsa_code_object_version 1,0
 // ASM: .hsa_code_object_version 1,0
 
 .hsa_code_object_isa 7,0,0,"AMD","AMDGPU"
 // ASM: .hsa_code_object_isa 7,0,0,"AMD","AMDGPU"
 
-.text
+.hsatext
+// ASM: .hsatext
+
 amd_kernel_code_t_test_all:
 ; Test all amd_kernel_code_t members with non-default values.
 .amd_kernel_code_t
   kernel_code_version_major = 100
   kernel_code_version_minor = 100
   machine_kind = 0
   machine_version_major = 5
   machine_version_minor = 5