[llvm-objdump][AMDGPU] Pass ELF ABIVersion through disassembler (#78907)

Admittedly, its a bit ugly to pass the ABIVersion through onSymbolStart
but I'm not sure what a better place for it would be.

Author: epilk

llvm/include/llvm/MC/MCDisassembler/MCDisassembler.h

@@ -217,6 +217,9 @@ class MCDisassembler {
 
   const MCSubtargetInfo& getSubtargetInfo() const { return STI; }
 
+  /// ELF-specific, set the ABI version from the object header.
+  virtual void setABIVersion(unsigned Version) {}
+
   // Marked mutable because we cache it inside the disassembler, rather than
   // having to pass it around as an argument through all the autogenerated code.
   mutable raw_ostream *CommentStream = nullptr;

llvm/include/llvm/Object/ELFObjectFile.h

@@ -103,6 +103,8 @@ class ELFObjectFileBase : public ObjectFile {
 
   virtual uint16_t getEMachine() const = 0;
 
+  virtual uint8_t getEIdentABIVersion() const = 0;
+
   std::vector<ELFPltEntry> getPltEntries() const;
 
   /// Returns a vector containing a symbol version for each dynamic symbol.
@@ -251,6 +253,7 @@ ELFObjectFileBase::symbols() const {
 template <class ELFT> class ELFObjectFile : public ELFObjectFileBase {
   uint16_t getEMachine() const override;
   uint16_t getEType() const override;
+  uint8_t getEIdentABIVersion() const override;
   uint64_t getSymbolSize(DataRefImpl Sym) const override;
 
 public:
@@ -645,6 +648,10 @@ template <class ELFT> uint16_t ELFObjectFile<ELFT>::getEType() const {
   return EF.getHeader().e_type;
 }
 
+template <class ELFT> uint8_t ELFObjectFile<ELFT>::getEIdentABIVersion() const {
+  return EF.getHeader().e_ident[ELF::EI_ABIVERSION];
+}
+
 template <class ELFT>
 uint64_t ELFObjectFile<ELFT>::getSymbolSize(DataRefImpl Sym) const {
   Expected<const Elf_Sym *> SymOrErr = getSymbol(Sym);

llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp

@@ -47,12 +47,17 @@ using DecodeStatus = llvm::MCDisassembler::DecodeStatus;
 AMDGPUDisassembler::AMDGPUDisassembler(const MCSubtargetInfo &STI,
                                        MCContext &Ctx, MCInstrInfo const *MCII)
     : MCDisassembler(STI, Ctx), MCII(MCII), MRI(*Ctx.getRegisterInfo()),
-      MAI(*Ctx.getAsmInfo()), TargetMaxInstBytes(MAI.getMaxInstLength(&STI)) {
+      MAI(*Ctx.getAsmInfo()), TargetMaxInstBytes(MAI.getMaxInstLength(&STI)),
+      CodeObjectVersion(AMDGPU::getDefaultAMDHSACodeObjectVersion()) {
   // ToDo: AMDGPUDisassembler supports only VI ISA.
   if (!STI.hasFeature(AMDGPU::FeatureGCN3Encoding) && !isGFX10Plus())
     report_fatal_error("Disassembly not yet supported for subtarget");
 }
 
+void AMDGPUDisassembler::setABIVersion(unsigned Version) {
+  CodeObjectVersion = AMDGPU::getAMDHSACodeObjectVersion(Version);
+}
+
 inline static MCDisassembler::DecodeStatus
 addOperand(MCInst &Inst, const MCOperand& Opnd) {
   Inst.addOperand(Opnd);
@@ -2202,8 +2207,7 @@ AMDGPUDisassembler::decodeKernelDescriptorDirective(
                       KERNEL_CODE_PROPERTY_ENABLE_WAVEFRONT_SIZE32);
     }
 
-    // FIXME: We should be looking at the ELF header ABI version for this.
-    if (AMDGPU::getDefaultAMDHSACodeObjectVersion() >= AMDGPU::AMDHSA_COV5)
+    if (CodeObjectVersion >= AMDGPU::AMDHSA_COV5)
       PRINT_DIRECTIVE(".amdhsa_uses_dynamic_stack",
                       KERNEL_CODE_PROPERTY_USES_DYNAMIC_STACK);
 

llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h

@@ -100,12 +100,15 @@ class AMDGPUDisassembler : public MCDisassembler {
   mutable uint64_t Literal64;
   mutable bool HasLiteral;
   mutable std::optional<bool> EnableWavefrontSize32;
+  unsigned CodeObjectVersion;
 
 public:
   AMDGPUDisassembler(const MCSubtargetInfo &STI, MCContext &Ctx,
                      MCInstrInfo const *MCII);
   ~AMDGPUDisassembler() override = default;
 
+  void setABIVersion(unsigned Version) override;
+
   DecodeStatus getInstruction(MCInst &MI, uint64_t &Size,
                               ArrayRef<uint8_t> Bytes, uint64_t Address,
                               raw_ostream &CS) const override;

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp

@@ -175,6 +175,17 @@ unsigned getDefaultAMDHSACodeObjectVersion() {
   return DefaultAMDHSACodeObjectVersion;
 }
 
+unsigned getAMDHSACodeObjectVersion(unsigned ABIVersion) {
+  switch (ABIVersion) {
+  case ELF::ELFABIVERSION_AMDGPU_HSA_V4:
+    return 4;
+  case ELF::ELFABIVERSION_AMDGPU_HSA_V5:
+    return 5;
+  default:
+    return getDefaultAMDHSACodeObjectVersion();
+  }
+}
+
 uint8_t getELFABIVersion(const Triple &T, unsigned CodeObjectVersion) {
   if (T.getOS() != Triple::AMDHSA)
     return 0;

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h

@@ -50,12 +50,15 @@ bool isHsaAbi(const MCSubtargetInfo &STI);
 /// \returns Code object version from the IR module flag.
 unsigned getAMDHSACodeObjectVersion(const Module &M);
 
+/// \returns Code object version from ELF's e_ident[EI_ABIVERSION].
+unsigned getAMDHSACodeObjectVersion(unsigned ABIVersion);
+
 /// \returns The default HSA code object version. This should only be used when
 /// we lack a more accurate CodeObjectVersion value (e.g. from the IR module
 /// flag or a .amdhsa_code_object_version directive)
 unsigned getDefaultAMDHSACodeObjectVersion();
 
-/// \returns ABIVersion suitable for use in ELF's e_ident[ABIVERSION]. \param
+/// \returns ABIVersion suitable for use in ELF's e_ident[EI_ABIVERSION]. \param
 /// CodeObjectVersion is a value returned by getAMDHSACodeObjectVersion().
 uint8_t getELFABIVersion(const Triple &OS, unsigned CodeObjectVersion);
 

llvm/test/tools/llvm-objdump/ELF/AMDGPU/kd-cov5.s

@@ -0,0 +1,21 @@
+; RUN: sed 's/CODE_OBJECT_VERSION/5/g' %s \
+; RUN:   | llvm-mc --triple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=-xnack,+wavefrontsize32,-wavefrontsize64 -filetype=obj > %t.o
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd %t.o | FileCheck %s --check-prefixes=COV5,CHECK
+
+; RUN: sed 's/CODE_OBJECT_VERSION/4/g' %s \
+; RUN:   | llvm-mc --triple=amdgcn-amd-amdhsa -mcpu=gfx1010 -mattr=-xnack,+wavefrontsize32,-wavefrontsize64 -filetype=obj > %t.o
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd %t.o | FileCheck %s --check-prefixes=COV4,CHECK
+
+;; Verify that .amdhsa_uses_dynamic_stack is only printed on COV5+.
+
+; CHECK: .amdhsa_kernel kernel
+; COV5: .amdhsa_uses_dynamic_stack 0
+; COV4-NOT: .amdhsa_uses_dynamic_stack
+; CHECK: .end_amdhsa_kernel
+
+.amdhsa_code_object_version CODE_OBJECT_VERSION
+
+.amdhsa_kernel kernel
+  .amdhsa_next_free_vgpr 32
+  .amdhsa_next_free_sgpr 32
+.end_amdhsa_kernel

llvm/test/tools/llvm-objdump/ELF/AMDGPU/kd-gfx10.s

@@ -4,7 +4,8 @@
 
 ;--- 1.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-xnack,+wavefrontsize32,-wavefrontsize64 -filetype=obj -mcpu=gfx1010 < 1.s > 1.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 1.o | tail -n +7 | tee 1-disasm.s | FileCheck 1.s
+; RUN: echo '.amdhsa_code_object_version 5' > 1-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 1.o | tail -n +7 | tee -a 1-disasm.s | FileCheck 1.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-xnack,+wavefrontsize32,-wavefrontsize64 -filetype=obj -mcpu=gfx1010 < 1-disasm.s > 1-disasm.o
 ; RUN: cmp 1.o 1-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -50,6 +51,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 1
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32
@@ -58,7 +60,8 @@
 
 ;--- 2.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-xnack,+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1010 < 2.s > 2.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 2.o | tail -n +7 | tee 2-disasm.s | FileCheck 2.s
+; RUN: echo '.amdhsa_code_object_version 5' > 2-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 2.o | tail -n +7 | tee -a 2-disasm.s | FileCheck 2.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-xnack,+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1010 < 2-disasm.s > 2-disasm.o
 ; RUN: cmp 2.o 2-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -104,6 +107,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 0
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32
@@ -112,7 +116,8 @@
 
 ;--- 3.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-xnack,+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1010 < 3.s > 3.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 3.o | tail -n +7 | tee 3-disasm.s | FileCheck 3.s
+; RUN: echo '.amdhsa_code_object_version 5' > 3-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 3.o | tail -n +7 | tee -a 3-disasm.s | FileCheck 3.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-xnack,+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1010 < 3-disasm.s > 3-disasm.o
 ; RUN: cmp 3.o 3-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -158,6 +163,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 0
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32
@@ -166,7 +172,8 @@
 
 ;--- 4.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-xnack,+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1010 < 4.s > 4.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 4.o | tail -n +7 | tee 4-disasm.s | FileCheck 4.s
+; RUN: echo '.amdhsa_code_object_version 5' > 4-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 4.o | tail -n +7 | tee -a 4-disasm.s | FileCheck 4.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-xnack,+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1010 < 4-disasm.s > 4-disasm.o
 ; RUN: cmp 4.o 4-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -212,6 +219,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 0
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32

llvm/test/tools/llvm-objdump/ELF/AMDGPU/kd-gfx11.s

@@ -4,7 +4,8 @@
 
 ;--- 1.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize32,-wavefrontsize64 -filetype=obj -mcpu=gfx1100 < 1.s > 1.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 1.o | tail -n +7 | tee 1-disasm.s | FileCheck 1.s
+; RUN: echo '.amdhsa_code_object_version 5' > 1-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 1.o | tail -n +7 | tee -a 1-disasm.s | FileCheck 1.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize32,-wavefrontsize64 -filetype=obj -mcpu=gfx1100 < 1-disasm.s > 1-disasm.o
 ; RUN: cmp 1.o 1-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -51,6 +52,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 1
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32
@@ -59,7 +61,8 @@
 
 ;--- 2.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1100 < 2.s > 2.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 2.o | tail -n +7 | tee 2-disasm.s | FileCheck 2.s
+; RUN: echo '.amdhsa_code_object_version 5' > 2-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 2.o | tail -n +7 | tee -a 2-disasm.s | FileCheck 2.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1100 < 2-disasm.s > 2-disasm.o
 ; RUN: cmp 2.o 2-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -106,6 +109,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 0
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32
@@ -114,7 +118,8 @@
 
 ;--- 3.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1100 < 3.s > 3.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 3.o | tail -n +7 | tee 3-disasm.s | FileCheck 3.s
+; RUN: echo '.amdhsa_code_object_version 5' > 3-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 3.o | tail -n +7 | tee -a 3-disasm.s | FileCheck 3.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1100 < 3-disasm.s > 3-disasm.o
 ; RUN: cmp 3.o 3-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -161,6 +166,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 0
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32
@@ -169,7 +175,8 @@
 
 ;--- 4.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1100 < 4.s > 4.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 4.o | tail -n +7 | tee 4-disasm.s | FileCheck 4.s
+; RUN: echo '.amdhsa_code_object_version 5' > 4-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 4.o | tail -n +7 | tee -a 4-disasm.s | FileCheck 4.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize64,-wavefrontsize32 -filetype=obj -mcpu=gfx1100 < 4-disasm.s > 4-disasm.o
 ; RUN: cmp 4.o 4-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -216,6 +223,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 0
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32

llvm/test/tools/llvm-objdump/ELF/AMDGPU/kd-gfx12.s

@@ -4,7 +4,8 @@
 
 ;--- 1.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize32,-wavefrontsize64 -filetype=obj -mcpu=gfx1200 < 1.s > 1.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 1.o | tail -n +7 | tee 1-disasm.s | FileCheck 1.s
+; RUN: echo '.amdhsa_code_object_version 5' > 1-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 1.o | tail -n +7 | tee -a 1-disasm.s | FileCheck 1.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=+wavefrontsize32,-wavefrontsize64 -filetype=obj -mcpu=gfx1200 < 1-disasm.s > 1-disasm.o
 ; RUN: cmp 1.o 1-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -48,6 +49,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 1
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32
@@ -56,7 +58,8 @@
 
 ;--- 2.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-wavefrontsize32,+wavefrontsize64 -filetype=obj -mcpu=gfx1200 < 2.s > 2.o
-; RUN: llvm-objdump --disassemble-symbols=kernel.kd 2.o | tail -n +7 | tee 2-disasm.s | FileCheck 2.s
+; RUN: echo '.amdhsa_code_object_version 5' > 2-disasm.s
+; RUN: llvm-objdump --disassemble-symbols=kernel.kd 2.o | tail -n +7 | tee -a 2-disasm.s | FileCheck 2.s
 ; RUN: llvm-mc --triple=amdgcn-amd-amdhsa -mattr=-wavefrontsize32,+wavefrontsize64 -filetype=obj -mcpu=gfx1200 < 2-disasm.s > 2-disasm.o
 ; RUN: cmp 2.o 2-disasm.o
 ; CHECK: .amdhsa_kernel kernel
@@ -100,6 +103,7 @@
 ; CHECK-NEXT: .amdhsa_wavefront_size32 0
 ; CHECK-NEXT: .amdhsa_uses_dynamic_stack 0
 ; CHECK-NEXT: .end_amdhsa_kernel
+.amdhsa_code_object_version 5
 .amdhsa_kernel kernel
   .amdhsa_next_free_vgpr 32
   .amdhsa_next_free_sgpr 32