[AMDGPU] Emit MessagePack HSA Metadata for v3 code object

[scott.linder]

Summary

Continue to present HSA metadata as YAML in ASM and when output by tools (e.g. llvm-readobj), but encode it in Messagepack in the code object.

[arsenm]

include/llvm/BinaryFormat/AMDGPUMetadataVerifier.h

63

Just use msgpack::Node&?

[scott.linder]

Address feedback

[scott.linder]

lib/Target/AMDGPU/MCTargetDesc/AMDGPUHSAMetadataStreamer.cpp

800–805 ↗

(On Diff #151521)

@arsenm This calculation of the kernarg offset is analogous to what we used to have. We emitted DataLayout::getABITypeAlignment as "Align", and the runtime calculated the Offset.

Is DataLayout::getABITypeAlignment incorrect here? What is the correct way to reproduce the offsets calculated in the kernarg lowering?

[scott.linder]

Rebase

[scott.linder]

Rebase and ping

[nhaehnle]

include/llvm/Support/AMDGPUMetadata.h

455

This is admittedly a bit of bike-shedding, but what's the rationale for outlining the strings in this way?

As a general rule, I think it's better to have constant strings inline because it keeps the code base more easily discoverable. If I run into an ELF where something is wrong with .name, then inlined strings allow me to just grep for ".name" and find all the relevant pieces of code immediately. With outlining, there's a level of indirection. I can find this location here easily, but then how do I reliably find the places where it's actually used? In this particular case, I can be reasonably confident that nobody is crazy enough to do using namespace Key, so a grep for Key::Name should do the trick, but it's still an extra step of indirection.

[scott.linder]

Address feedback. There was no particular reason for not writing these inline, other than to prevent typos. With the verifier and tests we have I don't think that is an issue, and I agree these being grepable is worthwhile.

[scott.linder]

Rebase and update:

• Rename asm directive for V3 metadata
• Avoid emitting old note types for V3 code objects
• Fix LLVMBuild dependencies for libraries
• Update NT_AMDGPU_METADATA note type value

[kzhuravl]

Can you also rebase your change? I have done some changes to your patch in amd-common that you need to incorporate here.

tools/llvm-readobj/ELFDumper.cpp

3775–3796

Switch on a NoteType, then if the NoteType is ELF::NT_AMDGPU_METADATA interpret it.

[scott.linder]

Rebase tools/llvm-readobj/ELFDumper.cpp

[scott.linder]

Add missing AMDGPUNote type

[scott.linder]

Update getAMDGPUNote param name to match its new type

[kzhuravl]

what is the rationale for putting the metadata verifier in binary format instead of support?

lib/Target/AMDGPU/AMDGPUAsmPrinter.h

59

unique_ptr?

[scott.linder]

In D48179#1300366, @kzhuravl wrote:

what is the rationale for putting the metadata verifier in binary format instead of support?

It is a layering issue; BinaryFormat depends on Support, and the low-level MsgPack implementation is defined in BinaryFormat, so nothing that depends on it can live in Support.

[scott.linder]

Use unique_ptr for HSAMetadataStreamer

[scott.linder]

Rebase

[scott.linder]

Ping

[scott.linder]

Rebase, ping

[kzhuravl]

LGTM

include/llvm/BinaryFormat/AMDGPUMetadataVerifier.h

+//===- AMDGPUMetadataVerifier.h - MsgPack Types -----------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// This is a verifier for AMDGPU HSA metadata, which can verify both
+/// well-typed metadata and untyped metadata. When verifying in the non-strict
+/// mode, untyped metadata is coerced into the correct type if possible.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_BINARYFORMAT_AMDGPUMETADATAVERIFIER_H
+#define LLVM_BINARYFORMAT_AMDGPUMETADATAVERIFIER_H
+
+#include "llvm/BinaryFormat/MsgPackTypes.h"
+
+namespace llvm {
+namespace AMDGPU {
+namespace HSAMD {
+namespace V3 {
+
+/// Verifier for AMDGPU HSA metadata.
+///
+/// Operates in two modes:
+///
+/// In strict mode, metadata must already be well-typed.
+///
+/// In non-strict mode, metadata is coerced into expected types when possible.
+class MetadataVerifier {
+  bool Strict;
+
+  bool verifyScalar(msgpack::Node &Node, msgpack::ScalarNode::ScalarKind SKind,
+                    function_ref<bool(msgpack::ScalarNode &)> verifyValue = {});
+  bool verifyInteger(msgpack::Node &Node);
+  bool verifyArray(msgpack::Node &Node,
+                   function_ref<bool(msgpack::Node &)> verifyNode,
+                   Optional<size_t> Size = None);
+  bool verifyEntry(msgpack::MapNode<> &MapNode, StringRef Key, bool Required,
+                   function_ref<bool(msgpack::Node &)> verifyNode);
+  bool
+  verifyScalarEntry(msgpack::MapNode<> &MapNode, StringRef Key, bool Required,
+                    msgpack::ScalarNode::ScalarKind SKind,
+                    function_ref<bool(msgpack::ScalarNode &)> verifyValue = {});
+  bool verifyIntegerEntry(msgpack::MapNode<> &MapNode, StringRef Key,
+                          bool Required);
+  bool verifyKernelArgs(msgpack::Node &Node);
+  bool verifyKernel(msgpack::Node &Node);
+
+public:
+  /// Construct a MetadataVerifier, specifying whether it will operate in \p
+  /// Strict mode.
+  MetadataVerifier(bool Strict) : Strict(Strict) {}
+
+  /// Verify given HSA metadata.
+  ///
+  /// \returns True when successful, false when metadata is invalid.
+  bool verify(msgpack::Node &HSAMetadataRoot);
+};
+
+} // end namespace V3
+} // end namespace HSAMD
+} // end namespace AMDGPU
+} // end namespace llvm
+
+#endif // LLVM_BINARYFORMAT_AMDGPUMETADATAVERIFIER_H

include/llvm/BinaryFormat/ELF.h

 };
 
 // CET properties
 enum {
   GNU_PROPERTY_X86_FEATURE_1_IBT = 1 << 0,
   GNU_PROPERTY_X86_FEATURE_1_SHSTK = 1 << 1
 };
 
-// AMDGPU specific notes.
+// AMD specific notes. (Code Object V2)
 enum {
   // Note types with values between 0 and 9 (inclusive) are reserved.
   NT_AMD_AMDGPU_HSA_METADATA = 10,
   NT_AMD_AMDGPU_ISA = 11,
   NT_AMD_AMDGPU_PAL_METADATA = 12
 };
 
+// AMDGPU specific notes. (Code Object V3)
+enum {
+  // Note type value 0 is reserved.
+  NT_AMDGPU_METADATA = 1
+};
+
 enum {
   GNU_ABI_TAG_LINUX = 0,
   GNU_ABI_TAG_HURD = 1,
   GNU_ABI_TAG_SOLARIS = 2,
   GNU_ABI_TAG_FREEBSD = 3,
   GNU_ABI_TAG_NETBSD = 4,
   GNU_ABI_TAG_SYLLABLE = 5,
   GNU_ABI_TAG_NACL = 6,

include/llvm/Support/AMDGPUMetadata.h

 };
 
 /// Converts \p String to \p HSAMetadata.
 std::error_code fromString(std::string String, Metadata &HSAMetadata);
 
 /// Converts \p HSAMetadata to \p String.
 std::error_code toString(Metadata HSAMetadata, std::string &String);
 
+//===----------------------------------------------------------------------===//
+// HSA metadata for v3 code object.
+//===----------------------------------------------------------------------===//
+namespace V3 {
+
+/// HSA metadata major version.
+constexpr uint32_t VersionMajor = 1;
+/// HSA metadata minor version.
+constexpr uint32_t VersionMinor = 0;
+
+//===----------------------------------------------------------------------===//
+// Kernel Metadata.
+//===----------------------------------------------------------------------===//
+namespace Kernel {
+
+//===----------------------------------------------------------------------===//
+// Kernel Argument Metadata.
+//===----------------------------------------------------------------------===//
+namespace Arg {
+
+namespace Key {
+constexpr char Name[] = ".name";
+constexpr char TypeName[] = ".type_name";
+constexpr char Size[] = ".size";
+constexpr char Offset[] = ".offset";
+constexpr char ValueKind[] = ".value_kind";
+constexpr char ValueType[] = ".value_type";
+constexpr char PointeeAlign[] = ".pointee_align";
+constexpr char AddressSpace[] = ".address_space";
+constexpr char Access[] = ".access";
+constexpr char ActualAccess[] = ".actual_access";
+constexpr char IsConst[] = ".is_const";
+constexpr char IsRestrict[] = ".is_restrict";
+constexpr char IsVolatile[] = ".is_volatile";
+constexpr char IsPipe[] = ".is_pipe";
+} // end namespace Key
+
+namespace Access {
+constexpr char ReadOnly[] = "read_only";
+constexpr char WriteOnly[] = "write_only";
+constexpr char ReadWrite[] = "read_write";
+} // end namespace Access
+
+namespace AddressSpace {
+constexpr char Private[] = "private";
+constexpr char Global[] = "global";
+constexpr char Constant[] = "constant";
+constexpr char Local[] = "local";
+constexpr char Generic[] = "generic";
+constexpr char Region[] = "region";
+} // end namespace AddressSpace
+
+namespace ValueKind {
+constexpr char ByValue[] = "by_value";
+constexpr char GlobalBuffer[] = "global_buffer";
+constexpr char DynamicSharedPointer[] = "dynamic_shared_pointer";
+constexpr char Sampler[] = "sampler";
+constexpr char Image[] = "image";
+constexpr char Pipe[] = "pipe";
+constexpr char Queue[] = "queue";
+constexpr char HiddenGlobalOffsetX[] = "hidden_global_offset_x";
+constexpr char HiddenGlobalOffsetY[] = "hidden_global_offset_y";
+constexpr char HiddenGlobalOffsetZ[] = "hidden_global_offset_z";
+constexpr char HiddenNone[] = "hidden_none";
+constexpr char HiddenPrintfBuffer[] = "hidden_printf_buffer";
+constexpr char HiddenDefaultQueue[] = "hidden_default_queue";
+constexpr char HiddenCompletionAction[] = "hidden_completion_action";
+} // end namespace ValueKind
+
+namespace ValueType {
+constexpr char Struct[] = "struct";
+constexpr char I8[] = "i8";
+constexpr char U8[] = "u8";
+constexpr char I16[] = "i16";
+constexpr char U16[] = "u16";
+constexpr char F16[] = "f16";
+constexpr char I32[] = "i32";
+constexpr char U32[] = "u32";
+constexpr char F32[] = "f32";
+constexpr char I64[] = "i64";
+constexpr char U64[] = "u64";
+constexpr char F64[] = "f64";
+} // end namespace ValueType
+
+} // end namespace Arg
+
+namespace Key {
+constexpr char Name[] = ".name";
+constexpr char Symbol[] = ".symbol";
+constexpr char Language[] = ".language";
+constexpr char LanguageVersion[] = ".language_version";
+constexpr char Args[] = ".args";
+constexpr char ReqdWorkGroupSize[] = ".reqd_workgroup_size";
+constexpr char WorkGroupSizeHint[] = ".workgroup_size_hint";
+constexpr char VecTypeHint[] = ".vec_type_hint";
+constexpr char DeviceEnqueueSymbol[] = ".device_enqueue_symbol";
+constexpr char KernargSegmentSize[] = ".kernarg_segment_size";
+constexpr char GroupSegmentFixedSize[] = ".group_segment_fixed_size";
+constexpr char PrivateSegmentFixedSize[] = ".private_segment_fixed_size";
+constexpr char KernargSegmentAlign[] = ".kernarg_segment_align";
+constexpr char WavefrontSize[] = ".wavefront_size";
+constexpr char SGPRCount[] = ".sgpr_count";
+constexpr char VGPRCount[] = ".vgpr_count";
+constexpr char MaxFlatWorkGroupSize[] = ".max_flat_workgroup_size";
+constexpr char SGPRSpillCount[] = ".sgpr_spill_count";
+constexpr char VGPRSpillCount[] = ".vgpr_spill_count";
+} // end namespace Key
+
+namespace Language {
+constexpr char OpenCLC[] = "OpenCL C";
+constexpr char OpenCLCXX[] = "OpenCL C++";
+constexpr char HCC[] = "HCC";
+constexpr char HIP[] = "HIP";
+constexpr char OpenMP[] = "OpenMP";
+constexpr char Assembler[] = "Assembler";
+} // end namespace Language
+
+} // end namespace Kernel
+
+namespace Key {
+constexpr char Version[] = "amdhsa.version";
+constexpr char Printf[] = "amdhsa.printf";
+constexpr char Kernels[] = "amdhsa.kernels";
+} // end namespace Key
+
+} // end namespace V3
+
 } // end namespace HSAMD
 
 //===----------------------------------------------------------------------===//
 // PAL metadata.
 //===----------------------------------------------------------------------===//
 namespace PALMD {
 
 /// PAL metadata assembler directive.

lib/BinaryFormat/AMDGPUMetadataVerifier.cpp

+//===- AMDGPUMetadataVerifier.cpp - MsgPack Types ---------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// Implements a verifier for AMDGPU HSA metadata.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/BinaryFormat/AMDGPUMetadataVerifier.h"
+#include "llvm/Support/AMDGPUMetadata.h"
+
+namespace llvm {
+namespace AMDGPU {
+namespace HSAMD {
+namespace V3 {
+
+bool MetadataVerifier::verifyScalar(
+    msgpack::Node &Node, msgpack::ScalarNode::ScalarKind SKind,
+    function_ref<bool(msgpack::ScalarNode &)> verifyValue) {
+  auto ScalarPtr = dyn_cast<msgpack::ScalarNode>(&Node);
+  if (!ScalarPtr)
+    return false;
+  auto &Scalar = *ScalarPtr;
+  // Do not output extraneous tags for types we know from the spec.
+  Scalar.IgnoreTag = true;
+  if (Scalar.getScalarKind() != SKind) {
+    if (Strict)
+      return false;
+    // If we are not strict, we interpret string values as "implicitly typed"
+    // and attempt to coerce them to the expected type here.
+    if (Scalar.getScalarKind() != msgpack::ScalarNode::SK_String)
+      return false;
+    std::string StringValue = Scalar.getString();
+    Scalar.setScalarKind(SKind);
+    if (Scalar.inputYAML(StringValue) != StringRef())
+      return false;
+  }
+  if (verifyValue)
+    return verifyValue(Scalar);
+  return true;
+}
+
+bool MetadataVerifier::verifyInteger(msgpack::Node &Node) {
+  if (!verifyScalar(Node, msgpack::ScalarNode::SK_UInt))
+    if (!verifyScalar(Node, msgpack::ScalarNode::SK_Int))
+      return false;
+  return true;
+}
+
+bool MetadataVerifier::verifyArray(
+    msgpack::Node &Node, function_ref<bool(msgpack::Node &)> verifyNode,
+    Optional<size_t> Size) {
+  auto ArrayPtr = dyn_cast<msgpack::ArrayNode<>>(&Node);
+  if (!ArrayPtr)
+    return false;
+  auto &Array = *ArrayPtr;
+  if (Size && Array.size() != *Size)
+    return false;
+  for (auto &Item : Array)
+    if (!verifyNode(*Item.get()))
+      return false;
+
+  return true;
+}
+
+bool MetadataVerifier::verifyEntry(
+    msgpack::MapNode<> &MapNode, StringRef Key, bool Required,
+    function_ref<bool(msgpack::Node &)> verifyNode) {
+  auto Entry = MapNode.find(Key);
+  if (Entry == MapNode.end())
+    return !Required;
+  return verifyNode(*Entry->second.get());
+}
+
+bool MetadataVerifier::verifyScalarEntry(
+    msgpack::MapNode<> &MapNode, StringRef Key, bool Required,
+    msgpack::ScalarNode::ScalarKind SKind,
+    function_ref<bool(msgpack::ScalarNode &)> verifyValue) {
+  return verifyEntry(MapNode, Key, Required, [=](msgpack::Node &Node) {
+    return verifyScalar(Node, SKind, verifyValue);
+  });
+}
+
+bool MetadataVerifier::verifyIntegerEntry(msgpack::MapNode<> &MapNode,
+                                             StringRef Key, bool Required) {
+  return verifyEntry(MapNode, Key, Required, [this](msgpack::Node &Node) {
+    return verifyInteger(Node);
+  });
+}
+
+bool MetadataVerifier::verifyKernelArgs(msgpack::Node &Node) {
+  auto ArgsMapPtr = dyn_cast<msgpack::MapNode<>>(&Node);
+  if (!ArgsMapPtr)
+    return false;
+  auto &ArgsMap = *ArgsMapPtr;
+
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::Name, false,
+                         msgpack::ScalarNode::SK_String))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::TypeName, false,
+                         msgpack::ScalarNode::SK_String))
+    return false;
+  if (!verifyIntegerEntry(ArgsMap, Kernel::Arg::Key::Size, true))
+    return false;
+  if (!verifyIntegerEntry(ArgsMap, Kernel::Arg::Key::Offset, true))
+    return false;
+  if (!verifyScalarEntry(
+          ArgsMap, Kernel::Arg::Key::ValueKind, true,
+          msgpack::ScalarNode::SK_String, [this](msgpack::ScalarNode &SNode) {
+            return StringSwitch<bool>(SNode.getString())
+                .Case(Kernel::Arg::ValueKind::ByValue, true)
+                .Case(Kernel::Arg::ValueKind::GlobalBuffer, true)
+                .Case(Kernel::Arg::ValueKind::DynamicSharedPointer, true)
+                .Case(Kernel::Arg::ValueKind::Sampler, true)
+                .Case(Kernel::Arg::ValueKind::Image, true)
+                .Case(Kernel::Arg::ValueKind::Pipe, true)
+                .Case(Kernel::Arg::ValueKind::Queue, true)
+                .Case(Kernel::Arg::ValueKind::HiddenGlobalOffsetX, true)
+                .Case(Kernel::Arg::ValueKind::HiddenGlobalOffsetY, true)
+                .Case(Kernel::Arg::ValueKind::HiddenGlobalOffsetZ, true)
+                .Case(Kernel::Arg::ValueKind::HiddenNone, true)
+                .Case(Kernel::Arg::ValueKind::HiddenPrintfBuffer, true)
+                .Case(Kernel::Arg::ValueKind::HiddenDefaultQueue, true)
+                .Case(Kernel::Arg::ValueKind::HiddenCompletionAction, true)
+                .Default(false);
+          }))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::ValueType, true,
+                         msgpack::ScalarNode::SK_String,
+                         [this](msgpack::ScalarNode &SNode) {
+                           return StringSwitch<bool>(SNode.getString())
+                               .Case(Kernel::Arg::ValueType::Struct, true)
+                               .Case(Kernel::Arg::ValueType::I8, true)
+                               .Case(Kernel::Arg::ValueType::U8, true)
+                               .Case(Kernel::Arg::ValueType::I16, true)
+                               .Case(Kernel::Arg::ValueType::U16, true)
+                               .Case(Kernel::Arg::ValueType::F16, true)
+                               .Case(Kernel::Arg::ValueType::I32, true)
+                               .Case(Kernel::Arg::ValueType::U32, true)
+                               .Case(Kernel::Arg::ValueType::F32, true)
+                               .Case(Kernel::Arg::ValueType::I64, true)
+                               .Case(Kernel::Arg::ValueType::U64, true)
+                               .Case(Kernel::Arg::ValueType::F64, true)
+                               .Default(false);
+                         }))
+    return false;
+  if (!verifyIntegerEntry(ArgsMap, Kernel::Arg::Key::PointeeAlign, false))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::AddressSpace, false,
+                         msgpack::ScalarNode::SK_String,
+                         [this](msgpack::ScalarNode &SNode) {
+                           return StringSwitch<bool>(SNode.getString())
+                               .Case(Kernel::Arg::AddressSpace::Private, true)
+                               .Case(Kernel::Arg::AddressSpace::Global, true)
+                               .Case(Kernel::Arg::AddressSpace::Constant, true)
+                               .Case(Kernel::Arg::AddressSpace::Local, true)
+                               .Case(Kernel::Arg::AddressSpace::Generic, true)
+                               .Case(Kernel::Arg::AddressSpace::Region, true)
+                               .Default(false);
+                         }))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::Access, false,
+                         msgpack::ScalarNode::SK_String,
+                         [this](msgpack::ScalarNode &SNode) {
+                           return StringSwitch<bool>(SNode.getString())
+                               .Case(Kernel::Arg::Access::ReadOnly, true)
+                               .Case(Kernel::Arg::Access::WriteOnly, true)
+                               .Case(Kernel::Arg::Access::ReadWrite, true)
+                               .Default(false);
+                         }))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::ActualAccess, false,
+                         msgpack::ScalarNode::SK_String,
+                         [this](msgpack::ScalarNode &SNode) {
+                           return StringSwitch<bool>(SNode.getString())
+                               .Case(Kernel::Arg::Access::ReadOnly, true)
+                               .Case(Kernel::Arg::Access::WriteOnly, true)
+                               .Case(Kernel::Arg::Access::ReadWrite, true)
+                               .Default(false);
+                         }))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::IsConst, false,
+                         msgpack::ScalarNode::SK_Boolean))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::IsRestrict, false,
+                         msgpack::ScalarNode::SK_Boolean))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::IsVolatile, false,
+                         msgpack::ScalarNode::SK_Boolean))
+    return false;
+  if (!verifyScalarEntry(ArgsMap, Kernel::Arg::Key::IsPipe, false,
+                         msgpack::ScalarNode::SK_Boolean))
+    return false;
+
+  return true;
+}
+
+bool MetadataVerifier::verifyKernel(msgpack::Node &Node) {
+  auto KernelMapPtr = dyn_cast<msgpack::MapNode<>>(&Node);
+  if (!KernelMapPtr)
+    return false;
+  auto &KernelMap = *KernelMapPtr;
+
+  if (!verifyScalarEntry(KernelMap, Kernel::Key::Name, true,
+                         msgpack::ScalarNode::SK_String))
+    return false;
+  if (!verifyScalarEntry(KernelMap, Kernel::Key::Symbol, true,
+                         msgpack::ScalarNode::SK_String))
+    return false;
+  if (!verifyScalarEntry(KernelMap, Kernel::Key::Language, false,
+                         msgpack::ScalarNode::SK_String,
+                         [this](msgpack::ScalarNode &SNode) {
+                           return StringSwitch<bool>(SNode.getString())
+                               .Case(Kernel::Language::OpenCLC, true)
+                               .Case(Kernel::Language::OpenCLCXX, true)
+                               .Case(Kernel::Language::HCC, true)
+                               .Case(Kernel::Language::HIP, true)
+                               .Case(Kernel::Language::OpenMP, true)
+                               .Case(Kernel::Language::Assembler, true)
+                               .Default(false);
+                         }))
+    return false;
+  if (!verifyEntry(
+          KernelMap, Kernel::Key::LanguageVersion, false,
+          [this](msgpack::Node &Node) {
+            return verifyArray(
+                Node,
+                [this](msgpack::Node &Node) { return verifyInteger(Node); }, 2);
+          }))
+    return false;
+  if (!verifyEntry(KernelMap, Kernel::Key::Args, false,
+                   [this](msgpack::Node &Node) {
+                     return verifyArray(Node, [this](msgpack::Node &Node) {
+                       return verifyKernelArgs(Node);
+                     });
+                   }))
+    return false;
+  if (!verifyEntry(KernelMap, Kernel::Key::ReqdWorkGroupSize, false,
+                   [this](msgpack::Node &Node) {
+                     return verifyArray(Node,
+                                        [this](msgpack::Node &Node) {
+                                          return verifyInteger(Node);
+                                        },
+                                        3);
+                   }))
+    return false;
+  if (!verifyEntry(KernelMap, Kernel::Key::WorkGroupSizeHint, false,
+                   [this](msgpack::Node &Node) {
+                     return verifyArray(Node,
+                                        [this](msgpack::Node &Node) {
+                                          return verifyInteger(Node);
+                                        },
+                                        3);
+                   }))
+    return false;
+  if (!verifyScalarEntry(KernelMap, Kernel::Key::VecTypeHint, false,
+                         msgpack::ScalarNode::SK_String))
+    return false;
+  if (!verifyScalarEntry(KernelMap, Kernel::Key::DeviceEnqueueSymbol, false,
+                         msgpack::ScalarNode::SK_String))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::KernargSegmentSize, true))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::GroupSegmentFixedSize, true))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::PrivateSegmentFixedSize,
+                          true))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::KernargSegmentAlign, true))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::WavefrontSize, true))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::SGPRCount, true))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::VGPRCount, true))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::MaxFlatWorkGroupSize, true))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::SGPRSpillCount, false))
+    return false;
+  if (!verifyIntegerEntry(KernelMap, Kernel::Key::VGPRSpillCount, false))
+    return false;
+
+  return true;
+}
+
+bool MetadataVerifier::verify(msgpack::Node &HSAMetadataRoot) {
+  auto RootMapPtr = dyn_cast<msgpack::MapNode<>>(&HSAMetadataRoot);
+  if (!RootMapPtr)
+    return false;
+  auto &RootMap = *RootMapPtr;
+
+  if (!verifyEntry(
+          RootMap, Key::Version, true, [this](msgpack::Node &Node) {
+            return verifyArray(
+                Node,
+                [this](msgpack::Node &Node) { return verifyInteger(Node); }, 2);
+          }))
+    return false;
+  if (!verifyEntry(RootMap, Key::Printf, false, [this](msgpack::Node &Node) {
+        return verifyArray(Node, [this](msgpack::Node &Node) {
+          return verifyScalar(Node, msgpack::ScalarNode::SK_String);
+        });
+      }))
+    return false;
+  if (!verifyEntry(RootMap, Key::Kernels, true, [this](msgpack::Node &Node) {
+        return verifyArray(
+            Node, [this](msgpack::Node &Node) { return verifyKernel(Node); });
+      }))
+    return false;
+
+  return true;
+}
+
+} // end namespace V3
+} // end namespace HSAMD
+} // end namespace AMDGPU
+} // end namespace llvm

lib/BinaryFormat/CMakeLists.txt

 add_llvm_library(LLVMBinaryFormat
+  AMDGPUMetadataVerifier.cpp
   Dwarf.cpp
   Magic.cpp
   MsgPackReader.cpp
   MsgPackTypes.cpp
   MsgPackWriter.cpp
   Wasm.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${LLVM_MAIN_INCLUDE_DIR}/llvm/BinaryFormat
   )
   

lib/Target/AMDGPU/AMDGPUAsmPrinter.h

struct SIFunctionResourceInfo {
     bool HasRecursion = false;
 
     int32_t getTotalNumSGPRs(const GCNSubtarget &ST) const;
   };
 
   SIProgramInfo CurrentProgramInfo;
   DenseMap<const Function *, SIFunctionResourceInfo> CallGraphResourceInfo;
 
-  AMDGPU::HSAMD::MetadataStreamer HSAMetadataStream;
+  AMDGPU::HSAMD::MetadataStreamer *HSAMetadataStream;
   std::map<uint32_t, uint32_t> PALMetadataMap;
 
   uint64_t getFunctionCodeSize(const MachineFunction &MF) const;
   SIFunctionResourceInfo analyzeResourceUsage(const MachineFunction &MF) const;
 
   void readPALMetadata(Module &M);
   void getSIProgramInfo(SIProgramInfo &Out, const MachineFunction &MF);
   void getAmdKernelCode(amd_kernel_code_t &Out, const SIProgramInfo &KernelInfo,
private:
 
   amdhsa::kernel_descriptor_t getAmdhsaKernelDescriptor(
       const MachineFunction &MF,
       const SIProgramInfo &PI) const;
 
 public:
   explicit AMDGPUAsmPrinter(TargetMachine &TM,
                             std::unique_ptr<MCStreamer> Streamer);
+  ~AMDGPUAsmPrinter();
 
   StringRef getPassName() const override;
 
   const MCSubtargetInfo* getSTI() const;
 
   AMDGPUTargetStreamer* getTargetStreamer() const;
 
   bool doFinalization(Module &M) override;

lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp

 #include "llvm/MC/MCStreamer.h"
 #include "llvm/Support/AMDGPUMetadata.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Target/TargetLoweringObjectFile.h"
 
 using namespace llvm;
 using namespace llvm::AMDGPU;
+using namespace llvm::AMDGPU::HSAMD;
 
 // TODO: This should get the default rounding mode from the kernel. We just set
 // the default here, but this could change if the OpenCL rounding mode pragmas
 // are used.
 //
 // The denormal mode here should match what is reported by the OpenCL runtime
 // for the CL_FP_DENORM bit from CL_DEVICE_{HALF|SINGLE|DOUBLE}_FP_CONFIG, but
 // can also be override to flush with the -cl-denorms-are-zero compiler flag.
extern "C" void LLVMInitializeAMDGPUAsmPrinter() {
   TargetRegistry::RegisterAsmPrinter(getTheGCNTarget(),
                                      createAMDGPUAsmPrinterPass);
 }
 
 AMDGPUAsmPrinter::AMDGPUAsmPrinter(TargetMachine &TM,
                                    std::unique_ptr<MCStreamer> Streamer)
   : AsmPrinter(TM, std::move(Streamer)) {
     AMDGPUASI = static_cast<AMDGPUTargetMachine*>(&TM)->getAMDGPUAS();
-  }
+    if (IsaInfo::hasCodeObjectV3(getSTI()))
+      HSAMetadataStream = new MetadataStreamerV3();
+    else
+      HSAMetadataStream = new MetadataStreamerV2();
+  }
+
+AMDGPUAsmPrinter::~AMDGPUAsmPrinter() {
+  delete HSAMetadataStream;
 
 StringRef AMDGPUAsmPrinter::getPassName() const {
   return "AMDGPU Assembly Printer";
 }
 
 const MCSubtargetInfo* AMDGPUAsmPrinter::getSTI() const {
   return TM.getMCSubtargetInfo();
 }
 
 AMDGPUTargetStreamer* AMDGPUAsmPrinter::getTargetStreamer() const {
   if (!OutStreamer)
     return nullptr;
   return static_cast<AMDGPUTargetStreamer*>(OutStreamer->getTargetStreamer());
 }
 
 void AMDGPUAsmPrinter::EmitStartOfAsmFile(Module &M) {
-  if (IsaInfo::hasCodeObjectV3(getSTI()) &&
-      TM.getTargetTriple().getOS() == Triple::AMDHSA)
-    return;
-
   if (TM.getTargetTriple().getOS() != Triple::AMDHSA &&
       TM.getTargetTriple().getOS() != Triple::AMDPAL)
     return;
 
   if (TM.getTargetTriple().getOS() == Triple::AMDHSA)
-    HSAMetadataStream.begin(M);
+    HSAMetadataStream->begin(M);
 
   if (TM.getTargetTriple().getOS() == Triple::AMDPAL)
     readPALMetadata(M);
 
   // HSA emits NT_AMDGPU_HSA_CODE_OBJECT_VERSION for code objects v2.
   if (TM.getTargetTriple().getOS() == Triple::AMDHSA)
     getTargetStreamer()->EmitDirectiveHSACodeObjectVersion(2, 1);
 
   // HSA and PAL emit NT_AMDGPU_HSA_ISA for code objects v2.
   IsaInfo::IsaVersion ISA = IsaInfo::getIsaVersion(getSTI()->getFeatureBits());
   getTargetStreamer()->EmitDirectiveHSACodeObjectISA(
       ISA.Major, ISA.Minor, ISA.Stepping, "AMD", "AMDGPU");
 }
 
 void AMDGPUAsmPrinter::EmitEndOfAsmFile(Module &M) {
-  // TODO: Add metadata to code object v3.
-  if (IsaInfo::hasCodeObjectV3(getSTI()) &&
-      TM.getTargetTriple().getOS() == Triple::AMDHSA)
-    return;
-
   // Following code requires TargetStreamer to be present.
   if (!getTargetStreamer())
     return;
 
+  if (!IsaInfo::hasCodeObjectV3(getSTI())) {
     // Emit ISA Version (NT_AMD_AMDGPU_ISA).
     std::string ISAVersionString;
     raw_string_ostream ISAVersionStream(ISAVersionString);
     IsaInfo::streamIsaVersion(getSTI(), ISAVersionStream);
     getTargetStreamer()->EmitISAVersion(ISAVersionStream.str());
+  }
 
   // Emit HSA Metadata (NT_AMD_AMDGPU_HSA_METADATA).
   if (TM.getTargetTriple().getOS() == Triple::AMDHSA) {
-    HSAMetadataStream.end();
-    getTargetStreamer()->EmitHSAMetadata(HSAMetadataStream.getHSAMetadata());
+    HSAMetadataStream->end();
+    bool Success = HSAMetadataStream->emitTo(*getTargetStreamer());
+    (void)Success;
+    assert(Success && "Malformed HSA Metadata");
   }
 
+  if (!IsaInfo::hasCodeObjectV3(getSTI())) {
     // Emit PAL Metadata (NT_AMD_AMDGPU_PAL_METADATA).
     if (TM.getTargetTriple().getOS() == Triple::AMDPAL) {
       // Copy the PAL metadata from the map where we collected it into a vector,
       // then write it as a .note.
       PALMD::Metadata PALMetadataVector;
       for (auto i : PALMetadataMap) {
         PALMetadataVector.push_back(i.first);
         PALMetadataVector.push_back(i.second);
       }
       getTargetStreamer()->EmitPALMetadata(PALMetadataVector);
     }
   }
+}
 
 bool AMDGPUAsmPrinter::isBlockOnlyReachableByFallthrough(
   const MachineBasicBlock *MBB) const {
   if (!AsmPrinter::isBlockOnlyReachableByFallthrough(MBB))
     return false;
 
   if (MBB->empty())
     return true;
 
   // If this is a block implementing a long branch, an expression relative to
   // the start of the block is needed.  to the start of the block.
   // XXX - Is there a smarter way to check this?
   return (MBB->back().getOpcode() != AMDGPU::S_SETPC_B64);
 }
 
 void AMDGPUAsmPrinter::EmitFunctionBodyStart() {
   const SIMachineFunctionInfo &MFI = *MF->getInfo<SIMachineFunctionInfo>();
   if (!MFI.isEntryFunction())
     return;
-  if (IsaInfo::hasCodeObjectV3(getSTI()) &&
-      TM.getTargetTriple().getOS() == Triple::AMDHSA)
-    return;
 
   const GCNSubtarget &STM = MF->getSubtarget<GCNSubtarget>();
   const Function &F = MF->getFunction();
   if (STM.isAmdCodeObjectV2(F) &&
       (F.getCallingConv() == CallingConv::AMDGPU_KERNEL ||
        F.getCallingConv() == CallingConv::SPIR_KERNEL)) {
     amd_kernel_code_t KernelCode;
     getAmdKernelCode(KernelCode, CurrentProgramInfo, *MF);
     getTargetStreamer()->EmitAMDKernelCodeT(KernelCode);
   }
 
   if (TM.getTargetTriple().getOS() != Triple::AMDHSA)
     return;
 
-  HSAMetadataStream.emitKernel(*MF, CurrentProgramInfo);
+  HSAMetadataStream->emitKernel(*MF, CurrentProgramInfo);
 }
 
 void AMDGPUAsmPrinter::EmitFunctionBodyEnd() {
   const SIMachineFunctionInfo &MFI = *MF->getInfo<SIMachineFunctionInfo>();
   if (!MFI.isEntryFunction())
     return;
   if (!IsaInfo::hasCodeObjectV3(getSTI()) ||
       TM.getTargetTriple().getOS() != Triple::AMDHSA)

lib/Target/AMDGPU/AMDGPUHSAMetadataStreamer.h

 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_LIB_TARGET_AMDGPU_MCTARGETDESC_AMDGPUHSAMETADATASTREAMER_H
 #define LLVM_LIB_TARGET_AMDGPU_MCTARGETDESC_AMDGPUHSAMETADATASTREAMER_H
 
 #include "AMDGPU.h"
 #include "AMDKernelCodeT.h"
 #include "llvm/ADT/StringRef.h"
+#include "llvm/BinaryFormat/MsgPackTypes.h"
 #include "llvm/Support/AMDGPUMetadata.h"
 
 namespace llvm {
 
+class AMDGPUTargetStreamer;
 class Argument;
 class DataLayout;
 class Function;
 class MDNode;
 class Module;
 struct SIProgramInfo;
 class Type;
 
 namespace AMDGPU {
 namespace HSAMD {
 
-class MetadataStreamer final {
+class MetadataStreamer {
+public:
+  virtual ~MetadataStreamer() {};
+
+  virtual bool emitTo(AMDGPUTargetStreamer &TargetStreamer) = 0;
+
+  virtual void begin(const Module &Mod) = 0;
+
+  virtual void end() = 0;
+
+  virtual void emitKernel(const MachineFunction &MF, const SIProgramInfo &ProgramInfo) = 0;
+};
+
+class MetadataStreamerV3 final : public MetadataStreamer {
+private:
+  std::unique_ptr<msgpack::Node> HSAMetadataRoot = msgpack::makePtr<msgpack::MapNode<>>();
+  AMDGPUAS AMDGPUASI;
+
+  void dump(StringRef HSAMetadataString) const;
+
+  void verify(StringRef HSAMetadataString) const;
+
+  Optional<StringRef> getAccessQualifier(StringRef AccQual) const;
+
+  StringRef getAddressSpaceQualifer(unsigned AddressSpace) const;
+
+  StringRef getValueKind(Type *Ty, StringRef TypeQual,
+                         StringRef BaseTypeName) const;
+
+  StringRef getValueType(Type *Ty, StringRef TypeName) const;
+
+  std::string getTypeName(Type *Ty, bool Signed) const;
+
+  std::unique_ptr<msgpack::ArrayNode<>>
+  getWorkGroupDimensions(MDNode *Node) const;
+
+  std::unique_ptr<msgpack::MapNode<>>
+  getHSAKernelProps(const MachineFunction &MF, const SIProgramInfo &ProgramInfo) const;
+
+  void emitVersion();
+
+  void emitPrintf(const Module &Mod);
+
+  void emitKernelLanguage(const Function &Func, msgpack::MapNode<> &Kern);
+
+  void emitKernelAttrs(const Function &Func, msgpack::MapNode<> &Kern);
+
+  void emitKernelArgs(const Function &Func, msgpack::MapNode<> &Kern);
+
+  void emitKernelArg(const Argument &Arg, unsigned &Offset,
+                     msgpack::ArrayNode<> &Args);
+
+  void emitKernelArg(const DataLayout &DL, Type *Ty, StringRef ValueKind,
+                     unsigned &Offset, msgpack::ArrayNode<> &Args,
+                     unsigned PointeeAlign = 0,
+                     StringRef Name = "", StringRef TypeName = "",
+                     StringRef BaseTypeName = "", StringRef AccQual = "",
+                     StringRef TypeQual = "");
+
+  void emitHiddenKernelArgs(const Function &Func, unsigned &Offset,
+                            msgpack::ArrayNode<> &Args);
+
+  std::unique_ptr<msgpack::Node> &getRootMetadata(StringRef Key) {
+    return (*cast<msgpack::MapNode<>>(HSAMetadataRoot.get()))[Key];
+  }
+
+  std::unique_ptr<msgpack::Node> &getHSAMetadataRoot() {
+    return HSAMetadataRoot;
+  }
+
+public:
+  MetadataStreamerV3() = default;
+  ~MetadataStreamerV3() = default;
+
+  bool emitTo(AMDGPUTargetStreamer &TargetStreamer) override;
+
+  void begin(const Module &Mod) override;
+
+  void end() override;
+
+  void emitKernel(const MachineFunction &MF, const SIProgramInfo &ProgramInfo) override;
+};
+
+class MetadataStreamerV2 final : public MetadataStreamer {
 private:
   Metadata HSAMetadata;
   AMDGPUAS AMDGPUASI;
 
   void dump(StringRef HSAMetadataString) const;
 
   void verify(StringRef HSAMetadataString) const;
 
private:
   void emitKernelArg(const DataLayout &DL, Type *Ty, ValueKind ValueKind,
                      unsigned PointeeAlign = 0,
                      StringRef Name = "", StringRef TypeName = "",
                      StringRef BaseTypeName = "", StringRef AccQual = "",
                      StringRef TypeQual = "");
 
   void emitHiddenKernelArgs(const Function &Func);
 
-public:
-  MetadataStreamer() = default;
-  ~MetadataStreamer() = default;
-
   const Metadata &getHSAMetadata() const {
     return HSAMetadata;
   }
 
-  void begin(const Module &Mod);
+public:
+  MetadataStreamerV2() = default;
+  ~MetadataStreamerV2() = default;
+
+  bool emitTo(AMDGPUTargetStreamer &TargetStreamer) override;
+
+  void begin(const Module &Mod) override;
 
-  void end();
+  void end() override;
 
-  void emitKernel(const MachineFunction &MF, const SIProgramInfo &ProgramInfo);
+  void emitKernel(const MachineFunction &MF, const SIProgramInfo &ProgramInfo) override;
 };
 
 } // end namespace HSAMD
 } // end namespace AMDGPU
 } // end namespace llvm
 
 #endif // LLVM_LIB_TARGET_AMDGPU_MCTARGETDESC_AMDGPUHSAMETADATASTREAMER_H

lib/Target/AMDGPU/AMDGPUHSAMetadataStreamer.cpp

 /// AMDGPU HSA Metadata Streamer.
 ///
 //
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPUHSAMetadataStreamer.h"
 #include "AMDGPU.h"
 #include "AMDGPUSubtarget.h"
+#include "MCTargetDesc/AMDGPUTargetStreamer.h"
 #include "SIMachineFunctionInfo.h"
 #include "SIProgramInfo.h"
 #include "Utils/AMDGPUBaseInfo.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Support/raw_ostream.h"
 
 namespace llvm {
 
 static cl::opt<bool> DumpHSAMetadata(
     "amdgpu-dump-hsa-metadata",
     cl::desc("Dump AMDGPU HSA Metadata"));
 static cl::opt<bool> VerifyHSAMetadata(
     "amdgpu-verify-hsa-metadata",
     cl::desc("Verify AMDGPU HSA Metadata"));
 
 namespace AMDGPU {
 namespace HSAMD {
 
-void MetadataStreamer::dump(StringRef HSAMetadataString) const {
+//===----------------------------------------------------------------------===//
+// HSAMetadataStreamerV2
+//===----------------------------------------------------------------------===//
+void MetadataStreamerV2::dump(StringRef HSAMetadataString) const {
   errs() << "AMDGPU HSA Metadata:\n" << HSAMetadataString << '\n';
 }
 
-void MetadataStreamer::verify(StringRef HSAMetadataString) const {
+void MetadataStreamerV2::verify(StringRef HSAMetadataString) const {
   errs() << "AMDGPU HSA Metadata Parser Test: ";
 
   HSAMD::Metadata FromHSAMetadataString;
   if (fromString(HSAMetadataString, FromHSAMetadataString)) {
     errs() << "FAIL\n";
     return;
   }
 
   std::string ToHSAMetadataString;
   if (toString(FromHSAMetadataString, ToHSAMetadataString)) {
     errs() << "FAIL\n";
     return;
   }
 
   errs() << (HSAMetadataString == ToHSAMetadataString ? "PASS" : "FAIL")
          << '\n';
   if (HSAMetadataString != ToHSAMetadataString) {
     errs() << "Original input: " << HSAMetadataString << '\n'
            << "Produced output: " << ToHSAMetadataString << '\n';
   }
 }
 
-AccessQualifier MetadataStreamer::getAccessQualifier(StringRef AccQual) const {
+AccessQualifier MetadataStreamerV2::getAccessQualifier(StringRef AccQual) const {
   if (AccQual.empty())
     return AccessQualifier::Unknown;
 
   return StringSwitch<AccessQualifier>(AccQual)
              .Case("read_only",  AccessQualifier::ReadOnly)
              .Case("write_only", AccessQualifier::WriteOnly)
              .Case("read_write", AccessQualifier::ReadWrite)
              .Default(AccessQualifier::Default);
 }
 
-AddressSpaceQualifier MetadataStreamer::getAddressSpaceQualifer(
+AddressSpaceQualifier MetadataStreamerV2::getAddressSpaceQualifer(
     unsigned AddressSpace) const {
   if (AddressSpace == AMDGPUASI.PRIVATE_ADDRESS)
     return AddressSpaceQualifier::Private;
   if (AddressSpace == AMDGPUASI.GLOBAL_ADDRESS)
     return AddressSpaceQualifier::Global;
   if (AddressSpace == AMDGPUASI.CONSTANT_ADDRESS)
     return AddressSpaceQualifier::Constant;
   if (AddressSpace == AMDGPUASI.LOCAL_ADDRESS)
     return AddressSpaceQualifier::Local;
   if (AddressSpace == AMDGPUASI.FLAT_ADDRESS)
     return AddressSpaceQualifier::Generic;
   if (AddressSpace == AMDGPUASI.REGION_ADDRESS)
     return AddressSpaceQualifier::Region;
 
   llvm_unreachable("Unknown address space qualifier");
 }
 
-ValueKind MetadataStreamer::getValueKind(Type *Ty, StringRef TypeQual,
+ValueKind MetadataStreamerV2::getValueKind(Type *Ty, StringRef TypeQual,
                                          StringRef BaseTypeName) const {
   if (TypeQual.find("pipe") != StringRef::npos)
     return ValueKind::Pipe;
 
   return StringSwitch<ValueKind>(BaseTypeName)
              .Case("image1d_t", ValueKind::Image)
              .Case("image1d_array_t", ValueKind::Image)
              .Case("image1d_buffer_t", ValueKind::Image)
return StringSwitch<ValueKind>(BaseTypeName)
              .Default(isa<PointerType>(Ty) ?
                           (Ty->getPointerAddressSpace() ==
                            AMDGPUASI.LOCAL_ADDRESS ?
                            ValueKind::DynamicSharedPointer :
                            ValueKind::GlobalBuffer) :
                       ValueKind::ByValue);
 }
 
-ValueType MetadataStreamer::getValueType(Type *Ty, StringRef TypeName) const {
+ValueType MetadataStreamerV2::getValueType(Type *Ty, StringRef TypeName) const {
   switch (Ty->getTypeID()) {
   case Type::IntegerTyID: {
     auto Signed = !TypeName.startswith("u");
     switch (Ty->getIntegerBitWidth()) {
     case 8:
       return Signed ? ValueType::I8 : ValueType::U8;
     case 16:
       return Signed ? ValueType::I16 : ValueType::U16;
case Type::PointerTyID:
     return getValueType(Ty->getPointerElementType(), TypeName);
   case Type::VectorTyID:
     return getValueType(Ty->getVectorElementType(), TypeName);
   default:
     return ValueType::Struct;
   }
 }
 
-std::string MetadataStreamer::getTypeName(Type *Ty, bool Signed) const {
+std::string MetadataStreamerV2::getTypeName(Type *Ty, bool Signed) const {
   switch (Ty->getTypeID()) {
   case Type::IntegerTyID: {
     if (!Signed)
       return (Twine('u') + getTypeName(Ty, true)).str();
 
     auto BitWidth = Ty->getIntegerBitWidth();
     switch (BitWidth) {
     case 8:
case Type::VectorTyID: {
     auto NumElements = VecTy->getVectorNumElements();
     return (Twine(getTypeName(ElTy, Signed)) + Twine(NumElements)).str();
   }
   default:
     return "unknown";
   }
 }
 
-std::vector<uint32_t> MetadataStreamer::getWorkGroupDimensions(
+std::vector<uint32_t> MetadataStreamerV2::getWorkGroupDimensions(
     MDNode *Node) const {
   std::vector<uint32_t> Dims;
   if (Node->getNumOperands() != 3)
     return Dims;
 
   for (auto &Op : Node->operands())
     Dims.push_back(mdconst::extract<ConstantInt>(Op)->getZExtValue());
   return Dims;
 }
 
-Kernel::CodeProps::Metadata MetadataStreamer::getHSACodeProps(
+Kernel::CodeProps::Metadata MetadataStreamerV2::getHSACodeProps(
     const MachineFunction &MF,
     const SIProgramInfo &ProgramInfo) const {
   const GCNSubtarget &STM = MF.getSubtarget<GCNSubtarget>();
   const SIMachineFunctionInfo &MFI = *MF.getInfo<SIMachineFunctionInfo>();
   HSAMD::Kernel::CodeProps::Metadata HSACodeProps;
   const Function &F = MF.getFunction();
 
   assert(F.getCallingConv() == CallingConv::AMDGPU_KERNEL ||
Kernel::CodeProps::Metadata MetadataStreamerV2::getHSACodeProps(
   HSACodeProps.mIsDynamicCallStack = ProgramInfo.DynamicCallStack;
   HSACodeProps.mIsXNACKEnabled = STM.isXNACKEnabled();
   HSACodeProps.mNumSpilledSGPRs = MFI.getNumSpilledSGPRs();
   HSACodeProps.mNumSpilledVGPRs = MFI.getNumSpilledVGPRs();
 
   return HSACodeProps;
 }
 
-Kernel::DebugProps::Metadata MetadataStreamer::getHSADebugProps(
+Kernel::DebugProps::Metadata MetadataStreamerV2::getHSADebugProps(
     const MachineFunction &MF,
     const SIProgramInfo &ProgramInfo) const {
   const GCNSubtarget &STM = MF.getSubtarget<GCNSubtarget>();
   HSAMD::Kernel::DebugProps::Metadata HSADebugProps;
 
   if (!STM.debuggerSupported())
     return HSADebugProps;
 
   HSADebugProps.mDebuggerABIVersion.push_back(1);
   HSADebugProps.mDebuggerABIVersion.push_back(0);
 
   if (STM.debuggerEmitPrologue()) {
     HSADebugProps.mPrivateSegmentBufferSGPR =
         ProgramInfo.DebuggerPrivateSegmentBufferSGPR;
     HSADebugProps.mWavefrontPrivateSegmentOffsetSGPR =
         ProgramInfo.DebuggerWavefrontPrivateSegmentOffsetSGPR;
   }
 
   return HSADebugProps;
 }
 
-void MetadataStreamer::emitVersion() {
+void MetadataStreamerV2::emitVersion() {
   auto &Version = HSAMetadata.mVersion;
 
   Version.push_back(VersionMajor);
   Version.push_back(VersionMinor);
 }
 
-void MetadataStreamer::emitPrintf(const Module &Mod) {
+void MetadataStreamerV2::emitPrintf(const Module &Mod) {
   auto &Printf = HSAMetadata.mPrintf;
 
   auto Node = Mod.getNamedMetadata("llvm.printf.fmts");
   if (!Node)
     return;
 
   for (auto Op : Node->operands())
     if (Op->getNumOperands())
       Printf.push_back(cast<MDString>(Op->getOperand(0))->getString());
 }
 
-void MetadataStreamer::emitKernelLanguage(const Function &Func) {
+void MetadataStreamerV2::emitKernelLanguage(const Function &Func) {
   auto &Kernel = HSAMetadata.mKernels.back();
 
   // TODO: What about other languages?
   auto Node = Func.getParent()->getNamedMetadata("opencl.ocl.version");
   if (!Node || !Node->getNumOperands())
     return;
   auto Op0 = Node->getOperand(0);
   if (Op0->getNumOperands() <= 1)
     return;
 
   Kernel.mLanguage = "OpenCL C";
   Kernel.mLanguageVersion.push_back(
       mdconst::extract<ConstantInt>(Op0->getOperand(0))->getZExtValue());
   Kernel.mLanguageVersion.push_back(
       mdconst::extract<ConstantInt>(Op0->getOperand(1))->getZExtValue());
 }
 
-void MetadataStreamer::emitKernelAttrs(const Function &Func) {
+void MetadataStreamerV2::emitKernelAttrs(const Function &Func) {
   auto &Attrs = HSAMetadata.mKernels.back().mAttrs;
 
   if (auto Node = Func.getMetadata("reqd_work_group_size"))
     Attrs.mReqdWorkGroupSize = getWorkGroupDimensions(Node);
   if (auto Node = Func.getMetadata("work_group_size_hint"))
     Attrs.mWorkGroupSizeHint = getWorkGroupDimensions(Node);
   if (auto Node = Func.getMetadata("vec_type_hint")) {
     Attrs.mVecTypeHint = getTypeName(
         cast<ValueAsMetadata>(Node->getOperand(0))->getType(),
         mdconst::extract<ConstantInt>(Node->getOperand(1))->getZExtValue());
   }
   if (Func.hasFnAttribute("runtime-handle")) {
     Attrs.mRuntimeHandle =
         Func.getFnAttribute("runtime-handle").getValueAsString().str();
   }
 }
 
-void MetadataStreamer::emitKernelArgs(const Function &Func) {
+void MetadataStreamerV2::emitKernelArgs(const Function &Func) {
   for (auto &Arg : Func.args())
     emitKernelArg(Arg);
 
   emitHiddenKernelArgs(Func);
 }
 
-void MetadataStreamer::emitKernelArg(const Argument &Arg) {
+void MetadataStreamerV2::emitKernelArg(const Argument &Arg) {
   auto Func = Arg.getParent();
   auto ArgNo = Arg.getArgNo();
   const MDNode *Node;
 
   StringRef Name;
   Node = Func->getMetadata("kernel_arg_name");
   if (Node && ArgNo < Node->getNumOperands())
     Name = cast<MDString>(Node->getOperand(ArgNo))->getString();
if (PtrTy->getAddressSpace() == AMDGPUASI.LOCAL_ADDRESS) {
         PointeeAlign = DL.getABITypeAlignment(PtrTy->getElementType());
     }
   }
 
   emitKernelArg(DL, Ty, getValueKind(Arg.getType(), TypeQual, BaseTypeName),
                 PointeeAlign, Name, TypeName, BaseTypeName, AccQual, TypeQual);
 }
 
-void MetadataStreamer::emitKernelArg(const DataLayout &DL, Type *Ty,
-                                     ValueKind ValueKind,
-                                     unsigned PointeeAlign,
+void MetadataStreamerV2::emitKernelArg(const DataLayout &DL, Type *Ty,
+                                       ValueKind ValueKind, unsigned PointeeAlign,
                                        StringRef Name,
                                        StringRef TypeName, StringRef BaseTypeName,
                                        StringRef AccQual, StringRef TypeQual) {
   HSAMetadata.mKernels.back().mArgs.push_back(Kernel::Arg::Metadata());
   auto &Arg = HSAMetadata.mKernels.back().mArgs.back();
 
   Arg.mName = Name;
   Arg.mTypeName = TypeName;
   Arg.mSize = DL.getTypeAllocSize(Ty);
   Arg.mAlign = DL.getABITypeAlignment(Ty);
   Arg.mValueKind = ValueKind;
auto P = StringSwitch<bool*>(Key)
                  .Case("volatile", &Arg.mIsVolatile)
                  .Case("pipe",     &Arg.mIsPipe)
                  .Default(nullptr);
     if (P)
       *P = true;
   }
 }
 
-void MetadataStreamer::emitHiddenKernelArgs(const Function &Func) {
+void MetadataStreamerV2::emitHiddenKernelArgs(const Function &Func) {
   int HiddenArgNumBytes =
       getIntegerAttribute(Func, "amdgpu-implicitarg-num-bytes", 0);
 
   if (!HiddenArgNumBytes)
     return;
 
   auto &DL = Func.getParent()->getDataLayout();
   auto Int64Ty = Type::getInt64Ty(Func.getContext());
if (Func.hasFnAttribute("calls-enqueue-kernel")) {
       emitKernelArg(DL, Int8PtrTy, ValueKind::HiddenCompletionAction);
     } else {
       emitKernelArg(DL, Int8PtrTy, ValueKind::HiddenNone);
       emitKernelArg(DL, Int8PtrTy, ValueKind::HiddenNone);
     }
   }
 }
 
-void MetadataStreamer::begin(const Module &Mod) {
+bool MetadataStreamerV2::emitTo(AMDGPUTargetStreamer &TargetStreamer) {
+  return TargetStreamer.EmitHSAMetadata(getHSAMetadata());
+}
+
+void MetadataStreamerV2::begin(const Module &Mod) {
   AMDGPUASI = getAMDGPUAS(Mod);
   emitVersion();
   emitPrintf(Mod);
 }
 
-void MetadataStreamer::end() {
+void MetadataStreamerV2::end() {
   std::string HSAMetadataString;
   if (toString(HSAMetadata, HSAMetadataString))
     return;
 
   if (DumpHSAMetadata)
     dump(HSAMetadataString);
   if (VerifyHSAMetadata)
     verify(HSAMetadataString);
 }
 
-void MetadataStreamer::emitKernel(const MachineFunction &MF, const SIProgramInfo &ProgramInfo) {
+void MetadataStreamerV2::emitKernel(const MachineFunction &MF, const SIProgramInfo &ProgramInfo) {
   auto &Func = MF.getFunction();
   if (Func.getCallingConv() != CallingConv::AMDGPU_KERNEL)
     return;
 
   auto CodeProps = getHSACodeProps(MF, ProgramInfo);
   auto DebugProps = getHSADebugProps(MF, ProgramInfo);
 
   HSAMetadata.mKernels.push_back(Kernel::Metadata());
   auto &Kernel = HSAMetadata.mKernels.back();
 
   Kernel.mName = Func.getName();
   Kernel.mSymbolName = (Twine(Func.getName()) + Twine("@kd")).str();
   emitKernelLanguage(Func);
   emitKernelAttrs(Func);
   emitKernelArgs(Func);
   HSAMetadata.mKernels.back().mCodeProps = CodeProps;
   HSAMetadata.mKernels.back().mDebugProps = DebugProps;
 }
 
+//===----------------------------------------------------------------------===//
+// HSAMetadataStreamerV3
+//===----------------------------------------------------------------------===//
+
+void MetadataStreamerV3::dump(StringRef HSAMetadataString) const {
+  errs() << "AMDGPU HSA Metadata:\n" << HSAMetadataString << '\n';
+}
+
+void MetadataStreamerV3::verify(StringRef HSAMetadataString) const {
+  errs() << "AMDGPU HSA Metadata Parser Test: ";
+
+  std::unique_ptr<msgpack::Node> FromHSAMetadataString
+    = make_unique<msgpack::MapNode<>>();
+
+  yaml::Input YIn(HSAMetadataString);
+  YIn >> FromHSAMetadataString;
+  if (YIn.error()) {
+    errs() << "FAIL\n";
+    return;
+  }
+
+  std::string ToHSAMetadataString;
+  raw_string_ostream StrOS(ToHSAMetadataString);
+  yaml::Output YOut(StrOS);
+  YOut << FromHSAMetadataString;
+
+  errs() << (HSAMetadataString == StrOS.str() ? "PASS" : "FAIL")
+         << '\n';
+  if (HSAMetadataString != ToHSAMetadataString) {
+    errs() << "Original input: " << HSAMetadataString << '\n'
+           << "Produced output: " << StrOS.str() << '\n';
+  }
+}
+
+Optional<StringRef> MetadataStreamerV3::getAccessQualifier(StringRef AccQual) const {
+  return StringSwitch<Optional<StringRef>>(AccQual)
+      .Case("read_only", StringRef(V3::Kernel::Arg::Access::ReadOnly))
+      .Case("write_only", StringRef(V3::Kernel::Arg::Access::WriteOnly))
+      .Case("read_write", StringRef(V3::Kernel::Arg::Access::ReadWrite))
+      .Default(None);
+}
+
+StringRef
+MetadataStreamerV3::getAddressSpaceQualifer(unsigned AddressSpace) const {
+  if (AddressSpace == AMDGPUASI.PRIVATE_ADDRESS)
+    return V3::Kernel::Arg::AddressSpace::Private;
+  if (AddressSpace == AMDGPUASI.GLOBAL_ADDRESS)
+    return V3::Kernel::Arg::AddressSpace::Global;
+  if (AddressSpace == AMDGPUASI.CONSTANT_ADDRESS)
+    return V3::Kernel::Arg::AddressSpace::Constant;
+  if (AddressSpace == AMDGPUASI.LOCAL_ADDRESS)
+    return V3::Kernel::Arg::AddressSpace::Local;
+  if (AddressSpace == AMDGPUASI.FLAT_ADDRESS)
+    return V3::Kernel::Arg::AddressSpace::Generic;
+  if (AddressSpace == AMDGPUASI.REGION_ADDRESS)
+    return V3::Kernel::Arg::AddressSpace::Region;
+
+  llvm_unreachable("Unknown address space qualifier");
+}
+
+StringRef MetadataStreamerV3::getValueKind(Type *Ty, StringRef TypeQual,
+                                         StringRef BaseTypeName) const {
+  if (TypeQual.find("pipe") != StringRef::npos)
+    return V3::Kernel::Arg::ValueKind::Pipe;
+
+  return StringSwitch<StringRef>(BaseTypeName)
+      .Case("image1d_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image1d_array_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image1d_buffer_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image2d_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image2d_array_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image2d_array_depth_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image2d_array_msaa_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image2d_array_msaa_depth_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image2d_depth_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image2d_msaa_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image2d_msaa_depth_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("image3d_t", V3::Kernel::Arg::ValueKind::Image)
+      .Case("sampler_t", V3::Kernel::Arg::ValueKind::Sampler)
+      .Case("queue_t", V3::Kernel::Arg::ValueKind::Queue)
+      .Default(isa<PointerType>(Ty)
+                   ? (Ty->getPointerAddressSpace() == AMDGPUASI.LOCAL_ADDRESS
+                          ? V3::Kernel::Arg::ValueKind::DynamicSharedPointer
+                          : V3::Kernel::Arg::ValueKind::GlobalBuffer)
+                   : V3::Kernel::Arg::ValueKind::ByValue);
+}
+
+StringRef MetadataStreamerV3::getValueType(Type *Ty, StringRef TypeName) const {
+  switch (Ty->getTypeID()) {
+  case Type::IntegerTyID: {
+    auto Signed = !TypeName.startswith("u");
+    switch (Ty->getIntegerBitWidth()) {
+    case 8:
+      return Signed ? V3::Kernel::Arg::ValueType::I8 : V3::Kernel::Arg::ValueType::U8;
+    case 16:
+      return Signed ? V3::Kernel::Arg::ValueType::I16 : V3::Kernel::Arg::ValueType::U16;
+    case 32:
+      return Signed ? V3::Kernel::Arg::ValueType::I32 : V3::Kernel::Arg::ValueType::U32;
+    case 64:
+      return Signed ? V3::Kernel::Arg::ValueType::I64 : V3::Kernel::Arg::ValueType::U64;
+    default:
+      return V3::Kernel::Arg::ValueType::Struct;
+    }
+  }
+  case Type::HalfTyID:
+    return V3::Kernel::Arg::ValueType::F16;
+  case Type::FloatTyID:
+    return V3::Kernel::Arg::ValueType::F32;
+  case Type::DoubleTyID:
+    return V3::Kernel::Arg::ValueType::F64;
+  case Type::PointerTyID:
+    return getValueType(Ty->getPointerElementType(), TypeName);
+  case Type::VectorTyID:
+    return getValueType(Ty->getVectorElementType(), TypeName);
+  default:
+    return V3::Kernel::Arg::ValueType::Struct;
+  }
+}
+
+std::string MetadataStreamerV3::getTypeName(Type *Ty, bool Signed) const {
+  switch (Ty->getTypeID()) {
+  case Type::IntegerTyID: {
+    if (!Signed)
+      return (Twine('u') + getTypeName(Ty, true)).str();
+
+    auto BitWidth = Ty->getIntegerBitWidth();
+    switch (BitWidth) {
+    case 8:
+      return "char";
+    case 16:
+      return "short";
+    case 32:
+      return "int";
+    case 64:
+      return "long";
+    default:
+      return (Twine('i') + Twine(BitWidth)).str();
+    }
+  }
+  case Type::HalfTyID:
+    return "half";
+  case Type::FloatTyID:
+    return "float";
+  case Type::DoubleTyID:
+    return "double";
+  case Type::VectorTyID: {
+    auto VecTy = cast<VectorType>(Ty);
+    auto ElTy = VecTy->getElementType();
+    auto NumElements = VecTy->getVectorNumElements();
+    return (Twine(getTypeName(ElTy, Signed)) + Twine(NumElements)).str();
+  }
+  default:
+    return "unknown";
+  }
+}
+
+std::unique_ptr<msgpack::ArrayNode<>>
+MetadataStreamerV3::getWorkGroupDimensions(MDNode *Node) const {
+  auto Dims = make_unique<msgpack::ArrayNode<>>();
+  if (Node->getNumOperands() != 3)
+    return Dims;
+
+  for (auto &Op : Node->operands())
+    Dims->push_back(make_unique<msgpack::ScalarNode>(
+        mdconst::extract<ConstantInt>(Op)->getZExtValue()));
+  return Dims;
+}
+
+void MetadataStreamerV3::emitVersion() {
+  auto Version = make_unique<msgpack::ArrayNode<>>();
+  Version->push_back(make_unique<msgpack::ScalarNode>(V3::VersionMajor));
+  Version->push_back(make_unique<msgpack::ScalarNode>(V3::VersionMinor));
+  getRootMetadata(V3::Key::Version) = std::move(Version);
+}
+
+void MetadataStreamerV3::emitPrintf(const Module &Mod) {
+  auto Node = Mod.getNamedMetadata("llvm.printf.fmts");
+  if (!Node)
+    return;
+
+  auto Printf = make_unique<msgpack::ArrayNode<>>();
+  for (auto Op : Node->operands())
+    if (Op->getNumOperands())
+      Printf->push_back(make_unique<msgpack::ScalarNode>(cast<MDString>(Op->getOperand(0))->getString()));
+  getRootMetadata(V3::Key::Printf) = std::move(Printf);
+}
+
+void MetadataStreamerV3::emitKernelLanguage(const Function &Func,
+                                          msgpack::MapNode<> &Kern) {
+  // TODO: What about other languages?
+  auto Node = Func.getParent()->getNamedMetadata("opencl.ocl.version");
+  if (!Node || !Node->getNumOperands())
+    return;
+  auto Op0 = Node->getOperand(0);
+  if (Op0->getNumOperands() <= 1)
+    return;
+
+  Kern[V3::Kernel::Key::Language] =
+      make_unique<msgpack::ScalarNode>(V3::Kernel::Language::OpenCLC);
+  auto LanguageVersion = make_unique<msgpack::ArrayNode<>>();
+  LanguageVersion->push_back(make_unique<msgpack::ScalarNode>(
+      mdconst::extract<ConstantInt>(Op0->getOperand(0))->getZExtValue()));
+  LanguageVersion->push_back(make_unique<msgpack::ScalarNode>(
+      mdconst::extract<ConstantInt>(Op0->getOperand(1))->getZExtValue()));
+  Kern[V3::Kernel::Key::LanguageVersion] = std::move(LanguageVersion);
+}
+
+void MetadataStreamerV3::emitKernelAttrs(const Function &Func,
+                                       msgpack::MapNode<> &Kern) {
+
+  if (auto Node = Func.getMetadata("reqd_work_group_size"))
+    Kern[V3::Kernel::Key::ReqdWorkGroupSize] = getWorkGroupDimensions(Node);
+  if (auto Node = Func.getMetadata("work_group_size_hint"))
+    Kern[V3::Kernel::Key::WorkGroupSizeHint] = getWorkGroupDimensions(Node);
+  if (auto Node = Func.getMetadata("vec_type_hint")) {
+    Kern[V3::Kernel::Key::VecTypeHint] = make_unique<msgpack::ScalarNode>(
+        getTypeName(cast<ValueAsMetadata>(Node->getOperand(0))->getType(),
+                    mdconst::extract<ConstantInt>(Node->getOperand(1))
+                        ->getZExtValue()));
+  }
+  if (Func.hasFnAttribute("runtime-handle")) {
+    Kern[V3::Kernel::Key::DeviceEnqueueSymbol] = make_unique<msgpack::ScalarNode>(
+        Func.getFnAttribute("runtime-handle").getValueAsString().str());
+  }
+}
+
+void MetadataStreamerV3::emitKernelArgs(const Function &Func,
+                                      msgpack::MapNode<> &Kern) {
+  unsigned Offset = 0;
+  auto Args = make_unique<msgpack::ArrayNode<>>();
+  for (auto &Arg : Func.args())
+    emitKernelArg(Arg, Offset, *Args);
+
+  emitHiddenKernelArgs(Func, Offset, *Args);
+
+  // TODO: What about other languages?
+  if (Func.getParent()->getNamedMetadata("opencl.ocl.version")) {
+    auto &DL = Func.getParent()->getDataLayout();
+    auto Int64Ty = Type::getInt64Ty(Func.getContext());
+
+    emitKernelArg(DL, Int64Ty, V3::Kernel::Arg::ValueKind::HiddenGlobalOffsetX,
+                  Offset, *Args);
+    emitKernelArg(DL, Int64Ty, V3::Kernel::Arg::ValueKind::HiddenGlobalOffsetY,
+                  Offset, *Args);
+    emitKernelArg(DL, Int64Ty, V3::Kernel::Arg::ValueKind::HiddenGlobalOffsetZ,
+                  Offset, *Args);
+
+    auto Int8PtrTy = Type::getInt8PtrTy(Func.getContext(),
+                                        AMDGPUASI.GLOBAL_ADDRESS);
+
+    // Emit "printf buffer" argument if printf is used, otherwise emit dummy
+    // "none" argument.
+    if (Func.getParent()->getNamedMetadata("llvm.printf.fmts"))
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenPrintfBuffer,
+                    Offset, *Args);
+    else
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenNone, Offset,
+                    *Args);
+
+    // Emit "default queue" and "completion action" arguments if enqueue kernel is
+    // used, otherwise emit dummy "none" arguments.
+    if (Func.hasFnAttribute("calls-enqueue-kernel")) {
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenDefaultQueue,
+                    Offset, *Args);
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenCompletionAction,
+                    Offset, *Args);
+    } else {
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenNone, Offset,
+                    *Args);
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenNone, Offset,
+                    *Args);
+    }
+  }
+
+  Kern[V3::Kernel::Key::Args] = std::move(Args);
+}
+
+void MetadataStreamerV3::emitKernelArg(const Argument &Arg, unsigned &Offset,
+                                     msgpack::ArrayNode<> &Args) {
+  auto Func = Arg.getParent();
+  auto ArgNo = Arg.getArgNo();
+  const MDNode *Node;
+
+  StringRef Name;
+  Node = Func->getMetadata("kernel_arg_name");
+  if (Node && ArgNo < Node->getNumOperands())
+    Name = cast<MDString>(Node->getOperand(ArgNo))->getString();
+  else if (Arg.hasName())
+    Name = Arg.getName();
+
+  StringRef TypeName;
+  Node = Func->getMetadata("kernel_arg_type");
+  if (Node && ArgNo < Node->getNumOperands())
+    TypeName = cast<MDString>(Node->getOperand(ArgNo))->getString();
+
+  StringRef BaseTypeName;
+  Node = Func->getMetadata("kernel_arg_base_type");
+  if (Node && ArgNo < Node->getNumOperands())
+    BaseTypeName = cast<MDString>(Node->getOperand(ArgNo))->getString();
+
+  StringRef AccQual;
+  if (Arg.getType()->isPointerTy() && Arg.onlyReadsMemory() &&
+      Arg.hasNoAliasAttr()) {
+    AccQual = "read_only";
+  } else {
+    Node = Func->getMetadata("kernel_arg_access_qual");
+    if (Node && ArgNo < Node->getNumOperands())
+      AccQual = cast<MDString>(Node->getOperand(ArgNo))->getString();
+  }
+
+  StringRef TypeQual;
+  Node = Func->getMetadata("kernel_arg_type_qual");
+  if (Node && ArgNo < Node->getNumOperands())
+    TypeQual = cast<MDString>(Node->getOperand(ArgNo))->getString();
+
+  Type *Ty = Arg.getType();
+  const DataLayout &DL = Func->getParent()->getDataLayout();
+
+  unsigned PointeeAlign = 0;
+  if (auto PtrTy = dyn_cast<PointerType>(Ty)) {
+    if (PtrTy->getAddressSpace() == AMDGPUASI.LOCAL_ADDRESS) {
+      PointeeAlign = Arg.getParamAlignment();
+      if (PointeeAlign == 0)
+        PointeeAlign = DL.getABITypeAlignment(PtrTy->getElementType());
+    }
+  }
+
+  emitKernelArg(Func->getParent()->getDataLayout(), Arg.getType(),
+                getValueKind(Arg.getType(), TypeQual, BaseTypeName), Offset,
+                Args, PointeeAlign, Name, TypeName, BaseTypeName, AccQual, TypeQual);
+}
+
+void MetadataStreamerV3::emitKernelArg(const DataLayout &DL, Type *Ty,
+                                     StringRef ValueKind, unsigned &Offset,
+                                     msgpack::ArrayNode<> &Args,
+                                     unsigned PointeeAlign,
+                                     StringRef Name,
+                                     StringRef TypeName, StringRef BaseTypeName,
+                                     StringRef AccQual, StringRef TypeQual) {
+  auto ArgPtr = make_unique<msgpack::MapNode<>>();
+  auto &Arg = *ArgPtr;
+
+  if (!Name.empty())
+    Arg[V3::Kernel::Arg::Key::Name] = make_unique<msgpack::ScalarNode>(Name);
+  if (!TypeName.empty())
+    Arg[V3::Kernel::Arg::Key::TypeName] = make_unique<msgpack::ScalarNode>(TypeName);
+  auto Size = DL.getTypeAllocSize(Ty);
+  auto Align = DL.getABITypeAlignment(Ty);
+  Arg[V3::Kernel::Arg::Key::Size] = make_unique<msgpack::ScalarNode>(Size);
+  Offset = alignTo(Offset, Align);
+  Arg[V3::Kernel::Arg::Key::Offset] = make_unique<msgpack::ScalarNode>(Offset);
+  Offset += Size;
+  Arg[V3::Kernel::Arg::Key::ValueKind] =
+      make_unique<msgpack::ScalarNode>(ValueKind);
+  Arg[V3::Kernel::Arg::Key::ValueType] =
+      make_unique<msgpack::ScalarNode>(getValueType(Ty, BaseTypeName));
+  if (PointeeAlign)
+    Arg[V3::Kernel::Arg::Key::PointeeAlign] =
+        make_unique<msgpack::ScalarNode>(PointeeAlign);
+
+  if (auto PtrTy = dyn_cast<PointerType>(Ty))
+    Arg[V3::Kernel::Arg::Key::AddressSpace] = make_unique<msgpack::ScalarNode>(
+        getAddressSpaceQualifer(PtrTy->getAddressSpace()));
+
+  if (auto AQ = getAccessQualifier(AccQual))
+    Arg[V3::Kernel::Arg::Key::Access] = make_unique<msgpack::ScalarNode>(*AQ);
+
+  // TODO: Emit Arg[V3::Kernel::Arg::Key::ActualAccess].
+
+  SmallVector<StringRef, 1> SplitTypeQuals;
+  TypeQual.split(SplitTypeQuals, " ", -1, false);
+  for (StringRef Key : SplitTypeQuals) {
+    if (Key == "const")
+      Arg[V3::Kernel::Arg::Key::IsConst] = make_unique<msgpack::ScalarNode>(true);
+    else if (Key == "restrict")
+      Arg[V3::Kernel::Arg::Key::IsRestrict] =
+          make_unique<msgpack::ScalarNode>(true);
+    else if (Key == "volatile")
+      Arg[V3::Kernel::Arg::Key::IsVolatile] =
+          make_unique<msgpack::ScalarNode>(true);
+    else if (Key == "pipe")
+      Arg[V3::Kernel::Arg::Key::IsPipe] = make_unique<msgpack::ScalarNode>(true);
+  }
+
+  Args.push_back(std::move(ArgPtr));
+}
+
+void MetadataStreamerV3::emitHiddenKernelArgs(const Function &Func,
+                                              unsigned &Offset,
+                                              msgpack::ArrayNode<> &Args) {
+  int HiddenArgNumBytes =
+      getIntegerAttribute(Func, "amdgpu-implicitarg-num-bytes", 0);
+
+  if (!HiddenArgNumBytes)
+    return;
+
+  auto &DL = Func.getParent()->getDataLayout();
+  auto Int64Ty = Type::getInt64Ty(Func.getContext());
+
+  if (HiddenArgNumBytes >= 8)
+    emitKernelArg(DL, Int64Ty, V3::Kernel::Arg::ValueKind::HiddenGlobalOffsetX, Offset, Args);
+  if (HiddenArgNumBytes >= 16)
+    emitKernelArg(DL, Int64Ty, V3::Kernel::Arg::ValueKind::HiddenGlobalOffsetY, Offset, Args);
+  if (HiddenArgNumBytes >= 24)
+    emitKernelArg(DL, Int64Ty, V3::Kernel::Arg::ValueKind::HiddenGlobalOffsetZ, Offset, Args);
+
+  auto Int8PtrTy = Type::getInt8PtrTy(Func.getContext(),
+                                      AMDGPUASI.GLOBAL_ADDRESS);
+
+  // Emit "printf buffer" argument if printf is used, otherwise emit dummy
+  // "none" argument.
+  if (HiddenArgNumBytes >= 32) {
+    if (Func.getParent()->getNamedMetadata("llvm.printf.fmts"))
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenPrintfBuffer, Offset, Args);
+    else
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenNone, Offset, Args);
+  }
+
+  // Emit "default queue" and "completion action" arguments if enqueue kernel is
+  // used, otherwise emit dummy "none" arguments.
+  if (HiddenArgNumBytes >= 48) {
+    if (Func.hasFnAttribute("calls-enqueue-kernel")) {
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenDefaultQueue, Offset, Args);
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenCompletionAction, Offset, Args);
+    } else {
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenNone, Offset, Args);
+      emitKernelArg(DL, Int8PtrTy, V3::Kernel::Arg::ValueKind::HiddenNone, Offset, Args);
+    }
+  }
+}
+
+std::unique_ptr<msgpack::MapNode<>>
+MetadataStreamerV3::getHSAKernelProps(const MachineFunction &MF,
+                                      const SIProgramInfo &ProgramInfo) const {
+  const GCNSubtarget &STM = MF.getSubtarget<GCNSubtarget>();
+  const SIMachineFunctionInfo &MFI = *MF.getInfo<SIMachineFunctionInfo>();
+  const Function &F = MF.getFunction();
+
+  auto HSAKernelProps = make_unique<msgpack::MapNode<>>();
+  auto &Kern = *HSAKernelProps;
+
+  unsigned MaxKernArgAlign;
+  Kern[V3::Kernel::Key::KernargSegmentSize] = make_unique<msgpack::ScalarNode>(
+      STM.getKernArgSegmentSize(F, MaxKernArgAlign));
+  Kern[V3::Kernel::Key::GroupSegmentFixedSize] =
+      make_unique<msgpack::ScalarNode>(ProgramInfo.LDSSize);
+  Kern[V3::Kernel::Key::PrivateSegmentFixedSize] =
+      make_unique<msgpack::ScalarNode>(ProgramInfo.ScratchSize);
+  Kern[V3::Kernel::Key::KernargSegmentAlign] = make_unique<msgpack::ScalarNode>(
+      std::max(uint32_t(4), MaxKernArgAlign));
+  Kern[V3::Kernel::Key::WavefrontSize] =
+      make_unique<msgpack::ScalarNode>(STM.getWavefrontSize());
+  Kern[V3::Kernel::Key::SGPRCount] =
+      make_unique<msgpack::ScalarNode>(ProgramInfo.NumSGPR);
+  Kern[V3::Kernel::Key::VGPRCount] =
+      make_unique<msgpack::ScalarNode>(ProgramInfo.NumVGPR);
+  Kern[V3::Kernel::Key::MaxFlatWorkGroupSize] =
+      make_unique<msgpack::ScalarNode>(MFI.getMaxFlatWorkGroupSize());
+  Kern[V3::Kernel::Key::SGPRSpillCount] =
+      make_unique<msgpack::ScalarNode>(MFI.getNumSpilledSGPRs());
+  Kern[V3::Kernel::Key::VGPRSpillCount] =
+      make_unique<msgpack::ScalarNode>(MFI.getNumSpilledVGPRs());
+
+  return HSAKernelProps;
+}
+
+
+bool MetadataStreamerV3::emitTo(AMDGPUTargetStreamer &TargetStreamer) {
+  return TargetStreamer.EmitHSAMetadata(getHSAMetadataRoot(), true);
+}
+
+void MetadataStreamerV3::begin(const Module &Mod) {
+  AMDGPUASI = getAMDGPUAS(Mod);
+  emitVersion();
+  emitPrintf(Mod);
+  getRootMetadata(V3::Key::Kernels).reset(new msgpack::ArrayNode<>());
+}
+
+void MetadataStreamerV3::end() {
+  std::string HSAMetadataString;
+  raw_string_ostream StrOS(HSAMetadataString);
+  yaml::Output YOut(StrOS);
+  YOut << HSAMetadataRoot;
+
+  if (DumpHSAMetadata)
+    dump(StrOS.str());
+  if (VerifyHSAMetadata)
+    verify(StrOS.str());
+}
+
+void MetadataStreamerV3::emitKernel(const MachineFunction &MF, const SIProgramInfo &ProgramInfo) {
+  auto &Func = MF.getFunction();
+  auto KernelProps = getHSAKernelProps(MF, ProgramInfo);
+
+  if (Func.getCallingConv() != CallingConv::AMDGPU_KERNEL)
+    return;
+
+  auto &KernelsNode = getRootMetadata(V3::Key::Kernels);
+  auto Kernels = cast<msgpack::ArrayNode<>>(KernelsNode.get());
+
+  {
+    auto &Kern = *KernelProps;
+    Kern[V3::Kernel::Key::Name] = make_unique<msgpack::ScalarNode>(Func.getName());
+    Kern[V3::Kernel::Key::Symbol] = make_unique<msgpack::ScalarNode>(
+        (Twine(Func.getName()) + Twine(".kd")).str());
+    emitKernelLanguage(Func, Kern);
+    emitKernelAttrs(Func, Kern);
+    emitKernelArgs(Func, Kern);
+  }
+
+  Kernels->push_back(std::move(KernelProps));
+}
+
 } // end namespace HSAMD
 } // end namespace AMDGPU
 } // end namespace llvm

lib/Target/AMDGPU/AMDGPUPTNote.h

 #define LLVM_LIB_TARGET_AMDGPU_AMDGPUPTNOTE_H
 
 namespace AMDGPU {
 
 namespace ElfNote {
 
 const char SectionName[] = ".note";
 
-const char NoteName[] = "AMD";
+const char NoteNameV2[] = "AMD";
+const char NoteNameV3[] = "AMDGPU";
 
 // TODO: Remove this file once we drop code object v2.
 enum NoteType{
     NT_AMDGPU_HSA_RESERVED_0 = 0,
     NT_AMDGPU_HSA_CODE_OBJECT_VERSION = 1,
     NT_AMDGPU_HSA_HSAIL = 2,
     NT_AMDGPU_HSA_ISA = 3,
     NT_AMDGPU_HSA_PRODUCER = 4,

lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

bool AMDGPUAsmParser::ParseDirectiveHSAMetadata() {
 
   if (getLexer().is(AsmToken::Eof) && !FoundEnd) {
     return TokError(Twine("expected directive ") +
                     Twine(HSAMD::AssemblerDirectiveEnd) + Twine(" not found"));
   }
 
   YamlStream.flush();
 
-  if (!getTargetStreamer().EmitHSAMetadata(HSAMetadataString))
+  if (IsaInfo::hasCodeObjectV3(&getSTI())) {
+    if (!getTargetStreamer().EmitHSAMetadataV3(HSAMetadataString))
       return Error(getParser().getTok().getLoc(), "invalid HSA metadata");
+  } else {
+    if (!getTargetStreamer().EmitHSAMetadataV2(HSAMetadataString))
+      return Error(getParser().getTok().getLoc(), "invalid HSA metadata");
+  }
 
   return false;
 }
 
 bool AMDGPUAsmParser::ParseDirectivePALMetadata() {
   if (getSTI().getTargetTriple().getOS() != Triple::AMDPAL) {
     return Error(getParser().getTok().getLoc(),
                  (Twine(PALMD::AssemblerDirective) + Twine(" directive is "

lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.h

 //===-- AMDGPUTargetStreamer.h - AMDGPU Target Streamer --------*- C++ -*--===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_LIB_TARGET_AMDGPU_MCTARGETDESC_AMDGPUTARGETSTREAMER_H
 #define LLVM_LIB_TARGET_AMDGPU_MCTARGETDESC_AMDGPUTARGETSTREAMER_H
 
 #include "AMDKernelCodeT.h"
+#include "llvm/BinaryFormat/MsgPackTypes.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/Support/AMDGPUMetadata.h"
 #include "llvm/Support/AMDHSAKernelDescriptor.h"
 
 namespace llvm {
 #include "AMDGPUPTNote.h"
 
public:
   virtual void EmitAMDKernelCodeT(const amd_kernel_code_t &Header) = 0;
 
   virtual void EmitAMDGPUSymbolType(StringRef SymbolName, unsigned Type) = 0;
 
   /// \returns True on success, false on failure.
   virtual bool EmitISAVersion(StringRef IsaVersionString) = 0;
 
   /// \returns True on success, false on failure.
-  virtual bool EmitHSAMetadata(StringRef HSAMetadataString);
+  virtual bool EmitHSAMetadataV2(StringRef HSAMetadataString);
+
+  /// \returns True on success, false on failure.
+  virtual bool EmitHSAMetadataV3(StringRef HSAMetadataString);
+
+  /// Emit HSA Metadata
+  ///
+  /// When \p Strict is true, known metadata elements must already be
+  /// well-typed. When \p Strict is false, known types are inferred and
+  /// the \p HSAMetadata structure is updated with the correct types.
+  ///
+  /// \returns True on success, false on failure.
+  virtual bool EmitHSAMetadata(std::unique_ptr<msgpack::Node> &HSAMetadata,
+                               bool Strict) = 0;
 
   /// \returns True on success, false on failure.
   virtual bool EmitHSAMetadata(const AMDGPU::HSAMD::Metadata &HSAMetadata) = 0;
 
   /// \returns True on success, false on failure.
   virtual bool EmitPALMetadata(const AMDGPU::PALMD::Metadata &PALMetadata) = 0;
 
   virtual void EmitAmdhsaKernelDescriptor(
public:
   void EmitAMDKernelCodeT(const amd_kernel_code_t &Header) override;
 
   void EmitAMDGPUSymbolType(StringRef SymbolName, unsigned Type) override;
 
   /// \returns True on success, false on failure.
   bool EmitISAVersion(StringRef IsaVersionString) override;
 
   /// \returns True on success, false on failure.
+  bool EmitHSAMetadata(std::unique_ptr<msgpack::Node> &HSAMetadata,
+                       bool Strict) override;
+
+  /// \returns True on success, false on failure.
   bool EmitHSAMetadata(const AMDGPU::HSAMD::Metadata &HSAMetadata) override;
 
   /// \returns True on success, false on failure.
   bool EmitPALMetadata(const AMDGPU::PALMD::Metadata &PALMetadata) override;
 
   void EmitAmdhsaKernelDescriptor(
       const MCSubtargetInfo &STI, StringRef KernelName,
       const amdhsa::kernel_descriptor_t &KernelDescriptor, uint64_t NextVGPR,
       uint64_t NextSGPR, bool ReserveVCC, bool ReserveFlatScr,
       bool ReserveXNACK) override;
 };
 
 class AMDGPUTargetELFStreamer final : public AMDGPUTargetStreamer {
   MCStreamer &Streamer;
 
-  void EmitAMDGPUNote(const MCExpr *DescSize, unsigned NoteType,
-                      function_ref<void(MCELFStreamer &)> EmitDesc);
+  void EmitNote(StringRef Name, const MCExpr *DescSize, unsigned NoteType,
 
 public:
   AMDGPUTargetELFStreamer(MCStreamer &S, const MCSubtargetInfo &STI);
 
   MCELFStreamer &getStreamer();
 
   void EmitDirectiveAMDGCNTarget(StringRef Target) override;
 
   void EmitDirectiveHSACodeObjectVersion(uint32_t Major,
                                          uint32_t Minor) override;
 
   void EmitDirectiveHSACodeObjectISA(uint32_t Major, uint32_t Minor,
                                      uint32_t Stepping, StringRef VendorName,
                                      StringRef ArchName) override;
 
   void EmitAMDKernelCodeT(const amd_kernel_code_t &Header) override;
 
   void EmitAMDGPUSymbolType(StringRef SymbolName, unsigned Type) override;
 
   /// \returns True on success, false on failure.
   bool EmitISAVersion(StringRef IsaVersionString) override;
 
   /// \returns True on success, false on failure.
+  bool EmitHSAMetadata(std::unique_ptr<msgpack::Node> &HSAMetadata,
+                       bool Strict) override;
+
+  /// \returns True on success, false on failure.
   bool EmitHSAMetadata(const AMDGPU::HSAMD::Metadata &HSAMetadata) override;
 
   /// \returns True on success, false on failure.
   bool EmitPALMetadata(const AMDGPU::PALMD::Metadata &PALMetadata) override;
 
   void EmitAmdhsaKernelDescriptor(
       const MCSubtargetInfo &STI, StringRef KernelName,
       const amdhsa::kernel_descriptor_t &KernelDescriptor, uint64_t NextVGPR,
       uint64_t NextSGPR, bool ReserveVCC, bool ReserveFlatScr,
       bool ReserveXNACK) override;
 };
 
 }
 #endif

lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp

 //===----------------------------------------------------------------------===//
 
 #include "AMDGPUTargetStreamer.h"
 #include "AMDGPU.h"
 #include "SIDefines.h"
 #include "Utils/AMDGPUBaseInfo.h"
 #include "Utils/AMDKernelCodeTUtils.h"
 #include "llvm/ADT/Twine.h"
+#include "llvm/BinaryFormat/AMDGPUMetadataVerifier.h"
 #include "llvm/BinaryFormat/ELF.h"
+#include "llvm/BinaryFormat/MsgPackTypes.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Metadata.h"
 #include "llvm/IR/Module.h"
 #include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCELFStreamer.h"
 #include "llvm/MC/MCObjectFileInfo.h"
 #include "llvm/MC/MCSectionELF.h"
 #include "llvm/Support/FormattedStream.h"
 
 namespace llvm {
 #include "AMDGPUPTNote.h"
 }
 
 using namespace llvm;
 using namespace llvm::AMDGPU;
+using namespace llvm::AMDGPU::HSAMD;
 
 //===----------------------------------------------------------------------===//
 // AMDGPUTargetStreamer
 //===----------------------------------------------------------------------===//
 
 static const struct {
   const char *Name;
   unsigned Mach;

 
 const char *AMDGPUTargetStreamer::getMachName(unsigned Mach) {
   auto Entry = MachTable;
   for (; Entry->Name && Mach != Entry->Mach; ++Entry)
     ;
   return Entry->Name;
 }
 
-bool AMDGPUTargetStreamer::EmitHSAMetadata(StringRef HSAMetadataString) {
+bool AMDGPUTargetStreamer::EmitHSAMetadataV2(StringRef HSAMetadataString) {
   HSAMD::Metadata HSAMetadata;
   if (HSAMD::fromString(HSAMetadataString, HSAMetadata))
     return false;
 
   return EmitHSAMetadata(HSAMetadata);
 }
 
+bool AMDGPUTargetStreamer::EmitHSAMetadataV3(StringRef HSAMetadataString) {
+  std::unique_ptr<msgpack::Node> HSAMetadataRoot;
+  yaml::Input YIn(HSAMetadataString);
+  YIn >> HSAMetadataRoot;
+  if (YIn.error())
+    return false;
+  return EmitHSAMetadata(HSAMetadataRoot, false);
+}
+
 //===----------------------------------------------------------------------===//
 // AMDGPUTargetAsmStreamer
 //===----------------------------------------------------------------------===//
 
 AMDGPUTargetAsmStreamer::AMDGPUTargetAsmStreamer(MCStreamer &S,
                                                  formatted_raw_ostream &OS)
     : AMDGPUTargetStreamer(S), OS(OS) { }
 

 }
 
 bool AMDGPUTargetAsmStreamer::EmitHSAMetadata(
     const AMDGPU::HSAMD::Metadata &HSAMetadata) {
   std::string HSAMetadataString;
   if (HSAMD::toString(HSAMetadata, HSAMetadataString))
     return false;
 
-  OS << '\t' << HSAMD::AssemblerDirectiveBegin << '\n';
+  OS << '\t' << AssemblerDirectiveBegin << '\n';
   OS << HSAMetadataString << '\n';
-  OS << '\t' << HSAMD::AssemblerDirectiveEnd << '\n';
+  OS << '\t' << AssemblerDirectiveEnd << '\n';
+  return true;
+}
+
+bool AMDGPUTargetAsmStreamer::EmitHSAMetadata(
+    std::unique_ptr<msgpack::Node> &HSAMetadataRoot, bool Strict) {
+  V3::MetadataVerifier Verifier(Strict);
+  if (!Verifier.verify(*HSAMetadataRoot))
+    return false;
+
+  std::string HSAMetadataString;
+  raw_string_ostream StrOS(HSAMetadataString);
+  yaml::Output YOut(StrOS);
+  YOut << HSAMetadataRoot;
+
+  OS << '\t' << AssemblerDirectiveBegin << '\n';
+  OS << StrOS.str() << '\n';
+  OS << '\t' << AssemblerDirectiveEnd << '\n';
   return true;
 }
 
 bool AMDGPUTargetAsmStreamer::EmitPALMetadata(
     const PALMD::Metadata &PALMetadata) {
   std::string PALMetadataString;
   if (PALMD::toString(PALMetadata, PALMetadataString))
     return false;
AMDGPUTargetELFStreamer::AMDGPUTargetELFStreamer(
 
   MCA.setELFHeaderEFlags(EFlags);
 }
 
 MCELFStreamer &AMDGPUTargetELFStreamer::getStreamer() {
   return static_cast<MCELFStreamer &>(Streamer);
 }
 
-void AMDGPUTargetELFStreamer::EmitAMDGPUNote(
-    const MCExpr *DescSZ, unsigned NoteType,
+void AMDGPUTargetELFStreamer::EmitNote(
+    StringRef Name, const MCExpr *DescSZ, unsigned NoteType,
     function_ref<void(MCELFStreamer &)> EmitDesc) {
   auto &S = getStreamer();
   auto &Context = S.getContext();
 
-  auto NameSZ = sizeof(ElfNote::NoteName);
+  auto NameSZ = Name.size() + 1;
 
   S.PushSection();
   S.SwitchSection(Context.getELFSection(
     ElfNote::SectionName, ELF::SHT_NOTE, ELF::SHF_ALLOC));
   S.EmitIntValue(NameSZ, 4);                                  // namesz
   S.EmitValue(DescSZ, 4);                                     // descz
   S.EmitIntValue(NoteType, 4);                                // type
-  S.EmitBytes(StringRef(ElfNote::NoteName, NameSZ));          // name
+  S.EmitBytes(Name);                                          // name
   S.EmitValueToAlignment(4, 0, 1, 0);                         // padding 0
   EmitDesc(S);                                                // desc
   S.EmitValueToAlignment(4, 0, 1, 0);                         // padding 0
   S.PopSection();
 }
 
 void AMDGPUTargetELFStreamer::EmitDirectiveAMDGCNTarget(StringRef Target) {}
 
 void AMDGPUTargetELFStreamer::EmitDirectiveHSACodeObjectVersion(
     uint32_t Major, uint32_t Minor) {
 
-  EmitAMDGPUNote(
+  EmitNote(ElfNote::NoteNameV2,
     MCConstantExpr::create(8, getContext()),
     ElfNote::NT_AMDGPU_HSA_CODE_OBJECT_VERSION,
     [&](MCELFStreamer &OS){
       OS.EmitIntValue(Major, 4);
       OS.EmitIntValue(Minor, 4);
     }
   );
 }
 
 void
 AMDGPUTargetELFStreamer::EmitDirectiveHSACodeObjectISA(uint32_t Major,
                                                        uint32_t Minor,
                                                        uint32_t Stepping,
                                                        StringRef VendorName,
                                                        StringRef ArchName) {
   uint16_t VendorNameSize = VendorName.size() + 1;
   uint16_t ArchNameSize = ArchName.size() + 1;
 
   unsigned DescSZ = sizeof(VendorNameSize) + sizeof(ArchNameSize) +
     sizeof(Major) + sizeof(Minor) + sizeof(Stepping) +
     VendorNameSize + ArchNameSize;
 
-  EmitAMDGPUNote(
+  EmitNote(ElfNote::NoteNameV2,
     MCConstantExpr::create(DescSZ, getContext()),
     ElfNote::NT_AMDGPU_HSA_ISA,
     [&](MCELFStreamer &OS) {
       OS.EmitIntValue(VendorNameSize, 2);
       OS.EmitIntValue(ArchNameSize, 2);
       OS.EmitIntValue(Major, 4);
       OS.EmitIntValue(Minor, 4);
       OS.EmitIntValue(Stepping, 4);
bool AMDGPUTargetELFStreamer::EmitISAVersion(StringRef IsaVersionString) {
   // and a MCExpr to calculate the size of the desc field.
   auto &Context = getContext();
   auto *DescBegin = Context.createTempSymbol();
   auto *DescEnd = Context.createTempSymbol();
   auto *DescSZ = MCBinaryExpr::createSub(
     MCSymbolRefExpr::create(DescEnd, Context),
     MCSymbolRefExpr::create(DescBegin, Context), Context);
 
-  EmitAMDGPUNote(
+  EmitNote(ElfNote::NoteNameV2,
     DescSZ,
     ELF::NT_AMD_AMDGPU_ISA,
     [&](MCELFStreamer &OS) {
       OS.EmitLabel(DescBegin);
       OS.EmitBytes(IsaVersionString);
       OS.EmitLabel(DescEnd);
     }
   );
   return true;
 }
 
 bool AMDGPUTargetELFStreamer::EmitHSAMetadata(
+    std::unique_ptr<msgpack::Node> &HSAMetadataRoot, bool Strict) {
+  V3::MetadataVerifier Verifier(Strict);
+  if (!Verifier.verify(*HSAMetadataRoot))
+    return false;
+
+  std::string HSAMetadataString;
+  raw_string_ostream StrOS(HSAMetadataString);
+  msgpack::Writer MPWriter(StrOS);
+  HSAMetadataRoot->write(MPWriter);
+
+  // Create two labels to mark the beginning and end of the desc field
+  // and a MCExpr to calculate the size of the desc field.
+  auto &Context = getContext();
+  auto *DescBegin = Context.createTempSymbol();
+  auto *DescEnd = Context.createTempSymbol();
+  auto *DescSZ = MCBinaryExpr::createSub(
+    MCSymbolRefExpr::create(DescEnd, Context),
+    MCSymbolRefExpr::create(DescBegin, Context), Context);
+
+  EmitNote(ElfNote::NoteNameV3,
+    DescSZ,
+    ELF::NT_AMDGPU_METADATA,
+    [&](MCELFStreamer &OS) {
+      OS.EmitLabel(DescBegin);
+      OS.EmitBytes(StrOS.str());
+      OS.EmitLabel(DescEnd);
+    }
+  );
+  return true;
+}
+
+bool AMDGPUTargetELFStreamer::EmitHSAMetadata(
     const AMDGPU::HSAMD::Metadata &HSAMetadata) {
   std::string HSAMetadataString;
   if (HSAMD::toString(HSAMetadata, HSAMetadataString))
     return false;
 
   // Create two labels to mark the beginning and end of the desc field
   // and a MCExpr to calculate the size of the desc field.
   auto &Context = getContext();
   auto *DescBegin = Context.createTempSymbol();
   auto *DescEnd = Context.createTempSymbol();
   auto *DescSZ = MCBinaryExpr::createSub(
     MCSymbolRefExpr::create(DescEnd, Context),
     MCSymbolRefExpr::create(DescBegin, Context), Context);
 
-  EmitAMDGPUNote(
+  EmitNote(ElfNote::NoteNameV2,
     DescSZ,
     ELF::NT_AMD_AMDGPU_HSA_METADATA,
     [&](MCELFStreamer &OS) {
       OS.EmitLabel(DescBegin);
       OS.EmitBytes(HSAMetadataString);
       OS.EmitLabel(DescEnd);
     }
   );
   return true;
 }
 
 bool AMDGPUTargetELFStreamer::EmitPALMetadata(
     const PALMD::Metadata &PALMetadata) {
-  EmitAMDGPUNote(
+  EmitNote(ElfNote::NoteNameV2,
     MCConstantExpr::create(PALMetadata.size() * sizeof(uint32_t), getContext()),
     ELF::NT_AMD_AMDGPU_PAL_METADATA,
     [&](MCELFStreamer &OS){
       for (auto I : PALMetadata)
         OS.EmitIntValue(I, sizeof(uint32_t));
     }
   );
   return true;

test/CodeGen/AMDGPU/attr-amdgpu-flat-work-group-size-v3.ll

+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -mattr=+code-object-v3 < %s | FileCheck --check-prefix=CHECK %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -amdgpu-verify-hsa-metadata -filetype=obj -mattr=+code-object-v3 -o /dev/null < %s 2>&1 | FileCheck --check-prefix=PARSER %s
+
+; CHECK-LABEL: {{^}}min_64_max_64:
+; CHECK: SGPRBlocks: 0
+; CHECK: VGPRBlocks: 0
+; CHECK: NumSGPRsForWavesPerEU: 1
+; CHECK: NumVGPRsForWavesPerEU: 1
+define amdgpu_kernel void @min_64_max_64() #0 {
+entry:
+  ret void
+}
+attributes #0 = {"amdgpu-flat-work-group-size"="64,64"}
+
+; CHECK-LABEL: {{^}}min_64_max_128:
+; CHECK: SGPRBlocks: 0
+; CHECK: VGPRBlocks: 0
+; CHECK: NumSGPRsForWavesPerEU: 1
+; CHECK: NumVGPRsForWavesPerEU: 1
+define amdgpu_kernel void @min_64_max_128() #1 {
+entry:
+  ret void
+}
+attributes #1 = {"amdgpu-flat-work-group-size"="64,128"}
+
+; CHECK-LABEL: {{^}}min_128_max_128:
+; CHECK: SGPRBlocks: 0
+; CHECK: VGPRBlocks: 0
+; CHECK: NumSGPRsForWavesPerEU: 1
+; CHECK: NumVGPRsForWavesPerEU: 1
+define amdgpu_kernel void @min_128_max_128() #2 {
+entry:
+  ret void
+}
+attributes #2 = {"amdgpu-flat-work-group-size"="128,128"}
+
+; CHECK-LABEL: {{^}}min_1024_max_2048
+; CHECK: SGPRBlocks: 1
+; CHECK: VGPRBlocks: 7
+; CHECK: NumSGPRsForWavesPerEU: 12
+; CHECK: NumVGPRsForWavesPerEU: 32
+@var = addrspace(1) global float 0.0
+define amdgpu_kernel void @min_1024_max_2048() #3 {
+  %val0 = load volatile float, float addrspace(1)* @var
+  %val1 = load volatile float, float addrspace(1)* @var
+  %val2 = load volatile float, float addrspace(1)* @var
+  %val3 = load volatile float, float addrspace(1)* @var
+  %val4 = load volatile float, float addrspace(1)* @var
+  %val5 = load volatile float, float addrspace(1)* @var
+  %val6 = load volatile float, float addrspace(1)* @var
+  %val7 = load volatile float, float addrspace(1)* @var
+  %val8 = load volatile float, float addrspace(1)* @var
+  %val9 = load volatile float, float addrspace(1)* @var
+  %val10 = load volatile float, float addrspace(1)* @var
+  %val11 = load volatile float, float addrspace(1)* @var
+  %val12 = load volatile float, float addrspace(1)* @var
+  %val13 = load volatile float, float addrspace(1)* @var
+  %val14 = load volatile float, float addrspace(1)* @var
+  %val15 = load volatile float, float addrspace(1)* @var
+  %val16 = load volatile float, float addrspace(1)* @var
+  %val17 = load volatile float, float addrspace(1)* @var
+  %val18 = load volatile float, float addrspace(1)* @var
+  %val19 = load volatile float, float addrspace(1)* @var
+  %val20 = load volatile float, float addrspace(1)* @var
+  %val21 = load volatile float, float addrspace(1)* @var
+  %val22 = load volatile float, float addrspace(1)* @var
+  %val23 = load volatile float, float addrspace(1)* @var
+  %val24 = load volatile float, float addrspace(1)* @var
+  %val25 = load volatile float, float addrspace(1)* @var
+  %val26 = load volatile float, float addrspace(1)* @var
+  %val27 = load volatile float, float addrspace(1)* @var
+  %val28 = load volatile float, float addrspace(1)* @var
+  %val29 = load volatile float, float addrspace(1)* @var
+  %val30 = load volatile float, float addrspace(1)* @var
+  %val31 = load volatile float, float addrspace(1)* @var
+  %val32 = load volatile float, float addrspace(1)* @var
+  %val33 = load volatile float, float addrspace(1)* @var
+  %val34 = load volatile float, float addrspace(1)* @var
+  %val35 = load volatile float, float addrspace(1)* @var
+  %val36 = load volatile float, float addrspace(1)* @var
+  %val37 = load volatile float, float addrspace(1)* @var
+  %val38 = load volatile float, float addrspace(1)* @var
+  %val39 = load volatile float, float addrspace(1)* @var
+  %val40 = load volatile float, float addrspace(1)* @var
+
+  store volatile float %val0, float addrspace(1)* @var
+  store volatile float %val1, float addrspace(1)* @var
+  store volatile float %val2, float addrspace(1)* @var
+  store volatile float %val3, float addrspace(1)* @var
+  store volatile float %val4, float addrspace(1)* @var
+  store volatile float %val5, float addrspace(1)* @var
+  store volatile float %val6, float addrspace(1)* @var
+  store volatile float %val7, float addrspace(1)* @var
+  store volatile float %val8, float addrspace(1)* @var
+  store volatile float %val9, float addrspace(1)* @var
+  store volatile float %val10, float addrspace(1)* @var
+  store volatile float %val11, float addrspace(1)* @var
+  store volatile float %val12, float addrspace(1)* @var
+  store volatile float %val13, float addrspace(1)* @var
+  store volatile float %val14, float addrspace(1)* @var
+  store volatile float %val15, float addrspace(1)* @var
+  store volatile float %val16, float addrspace(1)* @var
+  store volatile float %val17, float addrspace(1)* @var
+  store volatile float %val18, float addrspace(1)* @var
+  store volatile float %val19, float addrspace(1)* @var
+  store volatile float %val20, float addrspace(1)* @var
+  store volatile float %val21, float addrspace(1)* @var
+  store volatile float %val22, float addrspace(1)* @var
+  store volatile float %val23, float addrspace(1)* @var
+  store volatile float %val24, float addrspace(1)* @var
+  store volatile float %val25, float addrspace(1)* @var
+  store volatile float %val26, float addrspace(1)* @var
+  store volatile float %val27, float addrspace(1)* @var
+  store volatile float %val28, float addrspace(1)* @var
+  store volatile float %val29, float addrspace(1)* @var
+  store volatile float %val30, float addrspace(1)* @var
+  store volatile float %val31, float addrspace(1)* @var
+  store volatile float %val32, float addrspace(1)* @var
+  store volatile float %val33, float addrspace(1)* @var
+  store volatile float %val34, float addrspace(1)* @var
+  store volatile float %val35, float addrspace(1)* @var
+  store volatile float %val36, float addrspace(1)* @var
+  store volatile float %val37, float addrspace(1)* @var
+  store volatile float %val38, float addrspace(1)* @var
+  store volatile float %val39, float addrspace(1)* @var
+  store volatile float %val40, float addrspace(1)* @var
+
+  ret void
+}
+attributes #3 = {"amdgpu-flat-work-group-size"="1024,2048"}
+
+; CHECK: amdhsa.kernels:
+; CHECK:   .max_flat_workgroup_size: 64
+; CHECK:   .name:                 min_64_max_64
+; CHECK:   .max_flat_workgroup_size: 128
+; CHECK:   .name:                 min_64_max_128
+; CHECK:   .max_flat_workgroup_size: 128
+; CHECK:   .name:                 min_128_max_128
+; CHECK:   .max_flat_workgroup_size: 2048
+; CHECK:   .name:                 min_1024_max_2048
+; CHECK: amdhsa.version:
+; CHECK:   - 1
+; CHECK:   - 0
+
+; PARSER: AMDGPU HSA Metadata Parser Test: PASS

test/CodeGen/AMDGPU/code-object-v3.ll

 ; OSABI-AMDHSA-ASM:     .amdhsa_reserve_vcc 0
 ; OSABI-AMDHSA-ASM:     .amdhsa_reserve_flat_scratch 0
 ; OSABI-AMDHSA-ASM: .end_amdhsa_kernel
 ; OSABI-AMDHSA-ASM: .text
 
 ; OSABI-AMDHSA-ASM-NOT: .hsa_code_object_version
 ; OSABI-AMDHSA-ASM-NOT: .hsa_code_object_isa
 ; OSABI-AMDHSA-ASM-NOT: .amd_amdgpu_isa
-; OSABI-AMDHSA-ASM-NOT: .amd_amdgpu_hsa_metadata
 ; OSABI-AMDHSA-ASM-NOT: .amd_amdgpu_pal_metadata
 
 ; OSABI-AMDHSA-ELF: Section Headers
 ; OSABI-AMDHSA-ELF: .text   PROGBITS {{[0-9]+}} {{[0-9]+}} {{[0-9a-f]+}} {{[0-9]+}} AX {{[0-9]+}} {{[0-9]+}} 256
 ; OSABI-AMDHSA-ELF: .rodata PROGBITS {{[0-9]+}} {{[0-9]+}} {{[0-9a-f]+}} {{[0-9]+}}  A {{[0-9]+}} {{[0-9]+}} 64
 
 ; OSABI-AMDHSA-ELF: Relocation section '.rela.rodata' at offset
 ; OSABI-AMDHSA-ELF: 0000000000000010 0000000300000005 R_AMDGPU_REL64 0000000000000000 .text + 10
-; OSABI-AMDHSA-ELF: 0000000000000050 0000000300000005 R_AMDGPU_REL64 0000000000000000 .text + 110
+; OSABI-AMDHSA-ELF: 0000000000000050 0000000300000005 R_AMDGPU_REL64 0000000000000000 .text + 210
 
 ; OSABI-AMDHSA-ELF: Symbol table '.symtab' contains {{[0-9]+}} entries
 ; OSABI-AMDHSA-ELF: {{[0-9]+}}: 0000000000000000 {{[0-9]+}} FUNC   LOCAL  DEFAULT {{[0-9]+}} fadd
-; OSABI-AMDHSA-ELF: {{[0-9]+}}: 0000000000000100 {{[0-9]+}} FUNC   LOCAL  DEFAULT {{[0-9]+}} fsub
+; OSABI-AMDHSA-ELF: {{[0-9]+}}: 0000000000000200 {{[0-9]+}} FUNC   LOCAL  DEFAULT {{[0-9]+}} fsub
 ; OSABI-AMDHSA-ELF: {{[0-9]+}}: 0000000000000000 64         OBJECT GLOBAL DEFAULT {{[0-9]+}} fadd.kd
 ; OSABI-AMDHSA-ELF: {{[0-9]+}}: 0000000000000040 64         OBJECT GLOBAL DEFAULT {{[0-9]+}} fsub.kd
 
-; OSABI-AMDHSA-ELF-NOT: Displaying notes found
-
 define amdgpu_kernel void @fadd(
     float addrspace(1)* %r,
     float addrspace(1)* %a,
     float addrspace(1)* %b) {
 entry:
   %a.val = load float, float addrspace(1)* %a
   %b.val = load float, float addrspace(1)* %b
   %r.val = fadd float %a.val, %b.val

test/CodeGen/AMDGPU/hsa-metadata-deduce-ro-arg-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck %s
+
+; CHECK:        .symbol:          test_ro_arg.kd
+; CHECK:        .name:            test_ro_arg
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:       'float*'
+; CHECK-NEXT:       .value_kind:      global_buffer
+; CHECK-NEXT:       .name:            in
+; CHECK-NEXT:       .access:          read_only
+; CHECK-NEXT:       .offset:          0
+; CHECK-NEXT:       .is_const:        true
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .is_restrict:     true
+; CHECK-NEXT:       .value_type:      f32
+; CHECK-NEXT:       .address_space:   global
+; CHECK-NEXT:     - .type_name:       'float*'
+; CHECK-NEXT:       .value_kind:      global_buffer
+; CHECK-NEXT:       .name:            out
+; CHECK-NEXT:       .offset:          8
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      f32
+; CHECK-NEXT:       .address_space:   global
+
+define amdgpu_kernel void @test_ro_arg(float addrspace(1)* noalias readonly %in, float addrspace(1)* %out)
+    !kernel_arg_addr_space !0 !kernel_arg_access_qual !1 !kernel_arg_type !2
+    !kernel_arg_base_type !2 !kernel_arg_type_qual !3 {
+  ret void
+}
+
+!0 = !{i32 1, i32 1}
+!1 = !{!"none", !"none"}
+!2 = !{!"float*", !"float*"}
+!3 = !{!"const restrict", !""}

test/CodeGen/AMDGPU/hsa-metadata-enqueu-kernel-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX900 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -amdgpu-dump-hsa-metadata -amdgpu-verify-hsa-metadata -filetype=obj -o - < %s 2>&1 | FileCheck --check-prefix=PARSER %s
+
+; CHECK: ---
+; CHECK:  amdhsa.kernels:
+; CHECK:        .symbol:          test_non_enqueue_kernel_caller.kd
+; CHECK:        .name:            test_non_enqueue_kernel_caller
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:     char
+; CHECK-NEXT:       .value_kind:    by_value
+; CHECK-NEXT:       .offset:        0
+; CHECK-NEXT:       .size:          1
+; CHECK-NEXT:       .value_type:    i8
+; CHECK-NEXT:       .name:          a
+; CHECK-NEXT:     - .value_kind:    hidden_global_offset_x
+; CHECK-NEXT:       .offset:        8
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i64
+; CHECK-NEXT:     - .value_kind:    hidden_global_offset_y
+; CHECK-NEXT:       .offset:        16
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i64
+; CHECK-NEXT:     - .value_kind:    hidden_global_offset_z
+; CHECK-NEXT:       .offset:        24
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i64
+; CHECK-NOT:        .value_kind:    hidden_default_queue
+; CHECK-NOT:        .value_kind:    hidden_completion_action
+define amdgpu_kernel void @test_non_enqueue_kernel_caller(i8 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_enqueue_kernel_caller.kd
+; CHECK:        .name:            test_enqueue_kernel_caller
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:     char
+; CHECK-NEXT:       .value_kind:    by_value
+; CHECK-NEXT:       .offset:        0
+; CHECK-NEXT:       .size:          1
+; CHECK-NEXT:       .value_type:    i8
+; CHECK-NEXT:       .name:          a
+; CHECK-NEXT:     - .value_kind:    hidden_global_offset_x
+; CHECK-NEXT:       .offset:        8
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i64
+; CHECK-NEXT:     - .value_kind:    hidden_global_offset_y
+; CHECK-NEXT:       .offset:        16
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i64
+; CHECK-NEXT:     - .value_kind:    hidden_global_offset_z
+; CHECK-NEXT:       .offset:        24
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i64
+; CHECK-NEXT:     - .value_kind:    hidden_none
+; CHECK-NEXT:       .offset:        32
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i8
+; CHECK-NEXT:       .address_space: global
+; CHECK-NEXT:     - .value_kind:    hidden_default_queue
+; CHECK-NEXT:       .offset:        40
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i8
+; CHECK-NEXT:       .address_space: global
+; CHECK-NEXT:     - .value_kind:    hidden_completion_action
+; CHECK-NEXT:       .offset:        48
+; CHECK-NEXT:       .size:          8
+; CHECK-NEXT:       .value_type:    i8
+; CHECK-NEXT:       .address_space: global
+define amdgpu_kernel void @test_enqueue_kernel_caller(i8 %a) #0
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:  amdhsa.version:
+; CHECK-NEXT: - 1
+; CHECK-NEXT: - 0
+; CHECK-NOT:  amdhsa.printf:
+
+attributes #0 = { "calls-enqueue-kernel" }
+
+!1 = !{i32 0}
+!2 = !{!"none"}
+!3 = !{!"char"}
+!4 = !{!""}
+
+!opencl.ocl.version = !{!90}
+!90 = !{i32 2, i32 0}
+
+
+; PARSER: AMDGPU HSA Metadata Parser Test: PASS

test/CodeGen/AMDGPU/hsa-metadata-from-llvm-ir-full-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX700 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx802 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX802 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX900 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -amdgpu-dump-hsa-metadata -amdgpu-verify-hsa-metadata -filetype=obj -o - < %s 2>&1 | FileCheck --check-prefix=PARSER %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx802 -amdgpu-dump-hsa-metadata -amdgpu-verify-hsa-metadata -filetype=obj -o - < %s 2>&1 | FileCheck --check-prefix=PARSER %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -amdgpu-dump-hsa-metadata -amdgpu-verify-hsa-metadata -filetype=obj -o - < %s 2>&1 | FileCheck --check-prefix=PARSER %s
+
+%struct.A = type { i8, float }
+%opencl.image1d_t = type opaque
+%opencl.image2d_t = type opaque
+%opencl.image3d_t = type opaque
+%opencl.queue_t = type opaque
+%opencl.pipe_t = type opaque
+%struct.B = type { i32 addrspace(1)*}
+%opencl.clk_event_t = type opaque
+
+@__test_block_invoke_kernel_runtime_handle = external addrspace(1) externally_initialized constant i8 addrspace(1)*
+
+; CHECK: ---
+; CHECK:  amdhsa.kernels:
+; CHECK:        .symbol:          test_char.kd
+; CHECK:        .name:            test_char
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      char
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           1
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NOT:        .value_kind:     hidden_default_queue
+; CHECK-NOT:        .value_kind:     hidden_completion_action
+define amdgpu_kernel void @test_char(i8 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !9
+    !kernel_arg_base_type !9 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_ushort2.kd
+; CHECK:        .name:            test_ushort2
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      ushort2
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     u16
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_ushort2(<2 x i16> %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !10
+    !kernel_arg_base_type !10 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_int3.kd
+; CHECK:        .name:            test_int3
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int3
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           16
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_int3(<3 x i32> %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !11
+    !kernel_arg_base_type !11 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_ulong4.kd
+; CHECK:        .name:            test_ulong4
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      ulong4
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           32
+; CHECK-NEXT:       .value_type:     u64
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         48
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         56
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_ulong4(<4 x i64> %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !12
+    !kernel_arg_base_type !12 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_half8.kd
+; CHECK:        .name:            test_half8
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      half8
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           16
+; CHECK-NEXT:       .value_type:     f16
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_half8(<8 x half> %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !13
+    !kernel_arg_base_type !13 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_float16.kd
+; CHECK:        .name:            test_float16
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      float16
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           64
+; CHECK-NEXT:       .value_type:     f32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         64
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         72
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         80
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         88
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_float16(<16 x float> %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !14
+    !kernel_arg_base_type !14 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_double16.kd
+; CHECK:        .name:            test_double16
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      double16
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           128
+; CHECK-NEXT:       .value_type:     f64
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         128
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         136
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         144
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         152
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_double16(<16 x double> %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !15
+    !kernel_arg_base_type !15 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_pointer.kd
+; CHECK:        .name:            test_pointer
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      'int  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_pointer(i32 addrspace(1)* %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !16
+    !kernel_arg_base_type !16 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_image.kd
+; CHECK:        .name:            test_image
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      image2d_t
+; CHECK-NEXT:       .value_kind:     image
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_image(%opencl.image2d_t addrspace(1)* %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !17
+    !kernel_arg_base_type !17 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_sampler.kd
+; CHECK:        .name:            test_sampler
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      sampler_t
+; CHECK-NEXT:       .value_kind:     sampler
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_sampler(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !18
+    !kernel_arg_base_type !18 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_queue.kd
+; CHECK:        .name:            test_queue
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      queue_t
+; CHECK-NEXT:       .value_kind:     queue
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_queue(%opencl.queue_t addrspace(1)* %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !19
+    !kernel_arg_base_type !19 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_struct.kd
+; CHECK:        .name:            test_struct
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      struct A
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space: private
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_struct(%struct.A addrspace(5)* byval %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !20
+    !kernel_arg_base_type !20 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_i128.kd
+; CHECK:        .name:            test_i128
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      i128
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           16
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_i128(i128 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !21
+    !kernel_arg_base_type !21 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_multi_arg.kd
+; CHECK:        .name:            test_multi_arg
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .type_name:      short2
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         4
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i16
+; CHECK-NEXT:       .name:           b
+; CHECK-NEXT:     - .type_name:      char3
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .name:           c
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_multi_arg(i32 %a, <2 x i16> %b, <3 x i8> %c)
+    !kernel_arg_addr_space !22 !kernel_arg_access_qual !23 !kernel_arg_type !24
+    !kernel_arg_base_type !24 !kernel_arg_type_qual !25 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_addr_space.kd
+; CHECK:        .name:            test_addr_space
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      'int  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           g
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .type_name:      'int  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           c
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .address_space: constant
+; CHECK-NEXT:     - .type_name:      'int  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     dynamic_shared_pointer
+; CHECK-NEXT:       .name:           l
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .pointee_align:  4
+; CHECK-NEXT:       .address_space: local
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         48
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_addr_space(i32 addrspace(1)* %g,
+                                           i32 addrspace(4)* %c,
+                                           i32 addrspace(3)* %l)
+    !kernel_arg_addr_space !50 !kernel_arg_access_qual !23 !kernel_arg_type !51
+    !kernel_arg_base_type !51 !kernel_arg_type_qual !25 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_type_qual.kd
+; CHECK:        .name:            test_type_qual
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      'int  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .is_volatile:    true
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .type_name:      'int  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           b
+; CHECK-NEXT:       .is_const:       true
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .is_restrict:    true
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .type_name:      'int  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     pipe
+; CHECK-NEXT:       .name:           c
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .is_pipe:        true
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         48
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_type_qual(i32 addrspace(1)* %a,
+                                          i32 addrspace(1)* %b,
+                                          %opencl.pipe_t addrspace(1)* %c)
+    !kernel_arg_addr_space !22 !kernel_arg_access_qual !23 !kernel_arg_type !51
+    !kernel_arg_base_type !51 !kernel_arg_type_qual !70 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_access_qual.kd
+; CHECK:        .name:            test_access_qual
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      image1d_t
+; CHECK-NEXT:       .value_kind:     image
+; CHECK-NEXT:       .name:           ro
+; CHECK-NEXT:       .access:         read_only
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .type_name:      image2d_t
+; CHECK-NEXT:       .value_kind:     image
+; CHECK-NEXT:       .name:           wo
+; CHECK-NEXT:       .access:         write_only
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .type_name:      image3d_t
+; CHECK-NEXT:       .value_kind:     image
+; CHECK-NEXT:       .name:           rw
+; CHECK-NEXT:       .access:         read_write
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         48
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_access_qual(%opencl.image1d_t addrspace(1)* %ro,
+                                            %opencl.image2d_t addrspace(1)* %wo,
+                                            %opencl.image3d_t addrspace(1)* %rw)
+    !kernel_arg_addr_space !60 !kernel_arg_access_qual !61 !kernel_arg_type !62
+    !kernel_arg_base_type !62 !kernel_arg_type_qual !25 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_vec_type_hint_half.kd
+; CHECK:        .name:            test_vec_type_hint_half
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK:        .vec_type_hint:   half
+define amdgpu_kernel void @test_vec_type_hint_half(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !26 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_vec_type_hint_float.kd
+; CHECK:        .name:            test_vec_type_hint_float
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK:        .vec_type_hint:   float
+define amdgpu_kernel void @test_vec_type_hint_float(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !27 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_vec_type_hint_double.kd
+; CHECK:        .name:            test_vec_type_hint_double
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK:        .vec_type_hint:   double
+define amdgpu_kernel void @test_vec_type_hint_double(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !28 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_vec_type_hint_char.kd
+; CHECK:        .name:            test_vec_type_hint_char
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK:        .vec_type_hint:   char
+define amdgpu_kernel void @test_vec_type_hint_char(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !29 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_vec_type_hint_short.kd
+; CHECK:        .name:            test_vec_type_hint_short
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK:        .vec_type_hint:   short
+define amdgpu_kernel void @test_vec_type_hint_short(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !30 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_vec_type_hint_long.kd
+; CHECK:        .name:            test_vec_type_hint_long
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK:        .vec_type_hint:   long
+define amdgpu_kernel void @test_vec_type_hint_long(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !31 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_vec_type_hint_unknown.kd
+; CHECK:        .name:            test_vec_type_hint_unknown
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      int
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK:        .vec_type_hint:   unknown
+define amdgpu_kernel void @test_vec_type_hint_unknown(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !32 {
+  ret void
+}
+
+; CHECK:        .reqd_workgroup_size:
+; CHECK-NEXT:     - 1
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 4
+; CHECK:        .symbol:          test_reqd_wgs_vec_type_hint.kd
+; CHECK:        .name:            test_reqd_wgs_vec_type_hint
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:          int
+; CHECK-NEXT:       .value_kind:         by_value
+; CHECK-NEXT:       .offset:             0
+; CHECK-NEXT:       .size:               4
+; CHECK-NEXT:       .value_type:         i32
+; CHECK-NEXT:       .name:               a
+; CHECK-NEXT:     - .value_kind:         hidden_global_offset_x
+; CHECK-NEXT:       .offset:             8
+; CHECK-NEXT:       .size:               8
+; CHECK-NEXT:       .value_type:         i64
+; CHECK-NEXT:     - .value_kind:         hidden_global_offset_y
+; CHECK-NEXT:       .offset:             16
+; CHECK-NEXT:       .size:               8
+; CHECK-NEXT:       .value_type:         i64
+; CHECK-NEXT:     - .value_kind:         hidden_global_offset_z
+; CHECK-NEXT:       .offset:             24
+; CHECK-NEXT:       .size:               8
+; CHECK-NEXT:       .value_type:         i64
+; CHECK-NEXT:     - .value_kind:         hidden_printf_buffer
+; CHECK-NEXT:       .offset:             32
+; CHECK-NEXT:       .size:               8
+; CHECK-NEXT:       .value_type:         i8
+; CHECK-NEXT:       .address_space:      global
+; CHECK:        .vec_type_hint:       int
+define amdgpu_kernel void @test_reqd_wgs_vec_type_hint(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !5
+    !reqd_work_group_size !6 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_wgs_hint_vec_type_hint.kd
+; CHECK:        .workgroup_size_hint:
+; CHECK-NEXT:     - 8
+; CHECK-NEXT:     - 16
+; CHECK-NEXT:     - 32
+; CHECK:        .name:            test_wgs_hint_vec_type_hint
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:          int
+; CHECK-NEXT:       .value_kind:         by_value
+; CHECK-NEXT:       .offset:             0
+; CHECK-NEXT:       .size:               4
+; CHECK-NEXT:       .value_type:         i32
+; CHECK-NEXT:       .name:               a
+; CHECK-NEXT:     - .value_kind:         hidden_global_offset_x
+; CHECK-NEXT:       .offset:             8
+; CHECK-NEXT:       .size:               8
+; CHECK-NEXT:       .value_type:         i64
+; CHECK-NEXT:     - .value_kind:         hidden_global_offset_y
+; CHECK-NEXT:       .offset:             16
+; CHECK-NEXT:       .size:               8
+; CHECK-NEXT:       .value_type:         i64
+; CHECK-NEXT:     - .value_kind:         hidden_global_offset_z
+; CHECK-NEXT:       .offset:             24
+; CHECK-NEXT:       .size:               8
+; CHECK-NEXT:       .value_type:         i64
+; CHECK-NEXT:     - .value_kind:         hidden_printf_buffer
+; CHECK-NEXT:       .offset:             32
+; CHECK-NEXT:       .size:               8
+; CHECK-NEXT:       .value_type:         i8
+; CHECK-NEXT:       .address_space:      global
+; CHECK:        .vec_type_hint:       uint4
+define amdgpu_kernel void @test_wgs_hint_vec_type_hint(i32 %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !3
+    !kernel_arg_base_type !3 !kernel_arg_type_qual !4 !vec_type_hint !7
+    !work_group_size_hint !8 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_arg_ptr_to_ptr.kd
+; CHECK:        .name:            test_arg_ptr_to_ptr
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      'int  addrspace(5)* addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_arg_ptr_to_ptr(i32 addrspace(5)* addrspace(1)* %a)
+    !kernel_arg_addr_space !81 !kernel_arg_access_qual !2 !kernel_arg_type !80
+    !kernel_arg_base_type !80 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_arg_struct_contains_ptr.kd
+; CHECK:        .name:            test_arg_struct_contains_ptr
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      struct B
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space: private
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_arg_struct_contains_ptr(%struct.B addrspace(5)* byval %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !82
+    !kernel_arg_base_type !82 !kernel_arg_type_qual !4 {
+ ret void
+}
+
+; CHECK:        .symbol:          test_arg_vector_of_ptr.kd
+; CHECK:        .name:            test_arg_vector_of_ptr
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      'global int addrspace(5)* __attribute__((ext_vector_type(2)))'
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           16
+; CHECK-NEXT:       .value_type:     i32
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_arg_vector_of_ptr(<2 x i32 addrspace(1)*> %a)
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !83
+    !kernel_arg_base_type !83 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_arg_unknown_builtin_type.kd
+; CHECK:        .name:            test_arg_unknown_builtin_type
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      clk_event_t
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_arg_unknown_builtin_type(
+    %opencl.clk_event_t addrspace(1)* %a)
+    !kernel_arg_addr_space !81 !kernel_arg_access_qual !2 !kernel_arg_type !84
+    !kernel_arg_base_type !84 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_pointee_align.kd
+; CHECK:        .name:            test_pointee_align
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      'long  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     global_buffer
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .type_name:      'char  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     dynamic_shared_pointer
+; CHECK-NEXT:       .name:           b
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .pointee_align:  1
+; CHECK-NEXT:       .address_space: local
+; CHECK-NEXT:     - .type_name:      'char2  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     dynamic_shared_pointer
+; CHECK-NEXT:       .name:           c
+; CHECK-NEXT:       .offset:         12
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .pointee_align:  2
+; CHECK-NEXT:       .address_space: local
+; CHECK-NEXT:     - .type_name:      'char3  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     dynamic_shared_pointer
+; CHECK-NEXT:       .name:           d
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .pointee_align:  4
+; CHECK-NEXT:       .address_space: local
+; CHECK-NEXT:     - .type_name:      'char4  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     dynamic_shared_pointer
+; CHECK-NEXT:       .name:           e
+; CHECK-NEXT:       .offset:         20
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .pointee_align:  4
+; CHECK-NEXT:       .address_space: local
+; CHECK-NEXT:     - .type_name:      'char8  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     dynamic_shared_pointer
+; CHECK-NEXT:       .name:           f
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .pointee_align:  8
+; CHECK-NEXT:       .address_space: local
+; CHECK-NEXT:     - .type_name:      'char16  addrspace(5)*'
+; CHECK-NEXT:       .value_kind:     dynamic_shared_pointer
+; CHECK-NEXT:       .name:           g
+; CHECK-NEXT:       .offset:         28
+; CHECK-NEXT:       .size:           4
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .pointee_align:  16
+; CHECK-NEXT:       .address_space: local
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         48
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         56
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_pointee_align(i64 addrspace(1)* %a,
+                                              i8 addrspace(3)* %b,
+                                              <2 x i8> addrspace(3)* %c,
+                                              <3 x i8> addrspace(3)* %d,
+                                              <4 x i8> addrspace(3)* %e,
+                                              <8 x i8> addrspace(3)* %f,
+                                              <16 x i8> addrspace(3)* %g)
+    !kernel_arg_addr_space !91 !kernel_arg_access_qual !92 !kernel_arg_type !93
+    !kernel_arg_base_type !93 !kernel_arg_type_qual !94 {
+  ret void
+}
+
+; CHECK:        .symbol:          __test_block_invoke_kernel.kd
+; CHECK:        .device_enqueue_symbol: __test_block_invoke_kernel_runtime_handle
+; CHECK:        .name:            __test_block_invoke_kernel
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      __block_literal
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           25
+; CHECK-NEXT:       .value_type:     struct
+; CHECK-NEXT:       .name:           arg
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         48
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         56
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @__test_block_invoke_kernel(
+    <{ i32, i32, i8*, i8 addrspace(1)*, i8 }> %arg) #0
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !110
+    !kernel_arg_base_type !110 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:        .symbol:          test_enqueue_kernel_caller.kd
+; CHECK:        .name:            test_enqueue_kernel_caller
+; CHECK:        .language:        OpenCL C
+; CHECK:        .language_version:
+; CHECK-NEXT:     - 2
+; CHECK-NEXT:     - 0
+; CHECK:        .args:
+; CHECK-NEXT:     - .type_name:      char
+; CHECK-NEXT:       .value_kind:     by_value
+; CHECK-NEXT:       .offset:         0
+; CHECK-NEXT:       .size:           1
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .name:           a
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_x
+; CHECK-NEXT:       .offset:         8
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_y
+; CHECK-NEXT:       .offset:         16
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_global_offset_z
+; CHECK-NEXT:       .offset:         24
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i64
+; CHECK-NEXT:     - .value_kind:     hidden_printf_buffer
+; CHECK-NEXT:       .offset:         32
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_default_queue
+; CHECK-NEXT:       .offset:         40
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+; CHECK-NEXT:     - .value_kind:     hidden_completion_action
+; CHECK-NEXT:       .offset:         48
+; CHECK-NEXT:       .size:           8
+; CHECK-NEXT:       .value_type:     i8
+; CHECK-NEXT:       .address_space:  global
+define amdgpu_kernel void @test_enqueue_kernel_caller(i8 %a) #1
+    !kernel_arg_addr_space !1 !kernel_arg_access_qual !2 !kernel_arg_type !9
+    !kernel_arg_base_type !9 !kernel_arg_type_qual !4 {
+  ret void
+}
+
+; CHECK:  amdhsa.version:
+; CHECK-NEXT: - 1
+; CHECK-NEXT: - 0
+; CHECK:  amdhsa.printf:
+; CHECK-NEXT: - '1:1:4:%d\n'
+; CHECK-NEXT: - '2:1:8:%g\n'
+
+attributes #0 = { "runtime-handle"="__test_block_invoke_kernel_runtime_handle" }
+attributes #1 = { "calls-enqueue-kernel" }
+
+!llvm.printf.fmts = !{!100, !101}
+
+!1 = !{i32 0}
+!2 = !{!"none"}
+!3 = !{!"int"}
+!4 = !{!""}
+!5 = !{i32 undef, i32 1}
+!6 = !{i32 1, i32 2, i32 4}
+!7 = !{<4 x i32> undef, i32 0}
+!8 = !{i32 8, i32 16, i32 32}
+!9 = !{!"char"}
+!10 = !{!"ushort2"}
+!11 = !{!"int3"}
+!12 = !{!"ulong4"}
+!13 = !{!"half8"}
+!14 = !{!"float16"}
+!15 = !{!"double16"}
+!16 = !{!"int  addrspace(5)*"}
+!17 = !{!"image2d_t"}
+!18 = !{!"sampler_t"}
+!19 = !{!"queue_t"}
+!20 = !{!"struct A"}
+!21 = !{!"i128"}
+!22 = !{i32 0, i32 0, i32 0}
+!23 = !{!"none", !"none", !"none"}
+!24 = !{!"int", !"short2", !"char3"}
+!25 = !{!"", !"", !""}
+!26 = !{half undef, i32 1}
+!27 = !{float undef, i32 1}
+!28 = !{double undef, i32 1}
+!29 = !{i8 undef, i32 1}
+!30 = !{i16 undef, i32 1}
+!31 = !{i64 undef, i32 1}
+!32 = !{i32  addrspace(5)*undef, i32 1}
+!50 = !{i32 1, i32 2, i32 3}
+!51 = !{!"int  addrspace(5)*", !"int  addrspace(5)*", !"int  addrspace(5)*"}
+!60 = !{i32 1, i32 1, i32 1}
+!61 = !{!"read_only", !"write_only", !"read_write"}
+!62 = !{!"image1d_t", !"image2d_t", !"image3d_t"}
+!70 = !{!"volatile", !"const restrict", !"pipe"}
+!80 = !{!"int  addrspace(5)* addrspace(5)*"}
+!81 = !{i32 1}
+!82 = !{!"struct B"}
+!83 = !{!"global int addrspace(5)* __attribute__((ext_vector_type(2)))"}
+!84 = !{!"clk_event_t"}
+!opencl.ocl.version = !{!90}
+!90 = !{i32 2, i32 0}
+!91 = !{i32 0, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3}
+!92 = !{!"none", !"none", !"none", !"none", !"none", !"none", !"none"}
+!93 = !{!"long  addrspace(5)*", !"char  addrspace(5)*", !"char2  addrspace(5)*", !"char3  addrspace(5)*", !"char4  addrspace(5)*", !"char8  addrspace(5)*", !"char16  addrspace(5)*"}
+!94 = !{!"", !"", !"", !"", !"", !"", !""}
+!100 = !{!"1:1:4:%d\5Cn"}
+!101 = !{!"2:1:8:%g\5Cn"}
+!110 = !{!"__block_literal"}
+
+; PARSER: AMDGPU HSA Metadata Parser Test: PASS

test/CodeGen/AMDGPU/hsa-metadata-hidden-args-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX700 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX803 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX900 --check-prefix=NOTES %s
+
+; CHECK: ---
+; CHECK:  amdhsa.kernels:
+; CHECK:        .symbol:     test.kd
+; CHECK:        .name:       test
+; CHECK:        .args:
+; CHECK-NEXT:     - .value_kind:      global_buffer
+; CHECK-NEXT:       .name:            r
+; CHECK-NEXT:       .offset:          0
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      f16
+; CHECK-NEXT:       .address_space:   global
+; CHECK-NEXT:     - .value_kind:      global_buffer
+; CHECK-NEXT:       .name:            a
+; CHECK-NEXT:       .offset:          8
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      f16
+; CHECK-NEXT:       .address_space:   global
+; CHECK-NEXT:     - .value_kind:      global_buffer
+; CHECK-NEXT:       .name:            b
+; CHECK-NEXT:       .offset:          16
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      f16
+; CHECK-NEXT:       .address_space:   global
+; CHECK-NEXT:     - .value_kind:      hidden_global_offset_x
+; CHECK-NEXT:       .offset:          24
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      i64
+; CHECK-NEXT:     - .value_kind:      hidden_global_offset_y
+; CHECK-NEXT:       .offset:          32
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      i64
+; CHECK-NEXT:     - .value_kind:      hidden_global_offset_z
+; CHECK-NEXT:       .offset:          40
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      i64
+; CHECK-NEXT:     - .value_kind:      hidden_none
+; CHECK-NEXT:       .offset:          48
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      i8
+; CHECK-NEXT:       .address_space:   global
+; CHECK-NEXT:     - .value_kind:      hidden_none
+; CHECK-NEXT:       .offset:          56
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      i8
+; CHECK-NEXT:       .address_space:   global
+; CHECK-NEXT:     - .value_kind:      hidden_none
+; CHECK-NEXT:       .offset:          64
+; CHECK-NEXT:       .size:            8
+; CHECK-NEXT:       .value_type:      i8
+; CHECK-NEXT:       .address_space:   global
+define amdgpu_kernel void @test(
+    half addrspace(1)* %r,
+    half addrspace(1)* %a,
+    half addrspace(1)* %b) {
+entry:
+  %a.val = load half, half addrspace(1)* %a
+  %b.val = load half, half addrspace(1)* %b
+  %r.val = fadd half %a.val, %b.val
+  store half %r.val, half addrspace(1)* %r
+  ret void
+}
+
+; CHECK:  amdhsa.version:
+; CHECK-NEXT: - 1
+; CHECK-NEXT: - 0
+
+!opencl.ocl.version = !{!0}
+!0 = !{i32 2, i32 0}

test/CodeGen/AMDGPU/hsa-metadata-images-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX700 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx802 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX802 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX900 --check-prefix=NOTES %s
+
+%opencl.image1d_t = type opaque
+%opencl.image1d_array_t = type opaque
+%opencl.image1d_buffer_t = type opaque
+%opencl.image2d_t = type opaque
+%opencl.image2d_array_t = type opaque
+%opencl.image2d_array_depth_t = type opaque
+%opencl.image2d_array_msaa_t = type opaque
+%opencl.image2d_array_msaa_depth_t = type opaque
+%opencl.image2d_depth_t = type opaque
+%opencl.image2d_msaa_t = type opaque
+%opencl.image2d_msaa_depth_t = type opaque
+%opencl.image3d_t = type opaque
+
+; CHECK: ---
+; CHECK:  amdhsa.kernels:
+; CHECK:      .symbol:     test.kd
+; CHECK:      .name:       test
+; CHECK:      .args:
+; CHECK:        - .type_name:  image1d_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       a
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image1d_array_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       b
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image1d_buffer_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       c
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image2d_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       d
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image2d_array_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       e
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image2d_array_depth_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       f
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image2d_array_msaa_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       g
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image2d_array_msaa_depth_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       h
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image2d_depth_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       i
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image2d_msaa_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       j
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image2d_msaa_depth_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       k
+; CHECK:          .size:       8
+; CHECK:        - .type_name:  image3d_t
+; CHECK:          .value_kind: image
+; CHECK:          .name:       l
+; CHECK:          .size:       8
+define amdgpu_kernel void @test(%opencl.image1d_t addrspace(1)* %a,
+                                %opencl.image1d_array_t addrspace(1)* %b,
+                                %opencl.image1d_buffer_t addrspace(1)* %c,
+                                %opencl.image2d_t addrspace(1)* %d,
+                                %opencl.image2d_array_t addrspace(1)* %e,
+                                %opencl.image2d_array_depth_t addrspace(1)* %f,
+                                %opencl.image2d_array_msaa_t addrspace(1)* %g,
+                                %opencl.image2d_array_msaa_depth_t addrspace(1)* %h,
+                                %opencl.image2d_depth_t addrspace(1)* %i,
+                                %opencl.image2d_msaa_t addrspace(1)* %j,
+                                %opencl.image2d_msaa_depth_t addrspace(1)* %k,
+                                %opencl.image3d_t addrspace(1)* %l)
+    !kernel_arg_type !1 !kernel_arg_base_type !1 {
+  ret void
+}
+
+; CHECK:  amdhsa.version:
+; CHECK-NEXT: - 1
+; CHECK-NEXT: - 0
+
+!1 = !{!"image1d_t", !"image1d_array_t", !"image1d_buffer_t",
+       !"image2d_t", !"image2d_array_t", !"image2d_array_depth_t",
+       !"image2d_array_msaa_t", !"image2d_array_msaa_depth_t",
+       !"image2d_depth_t", !"image2d_msaa_t", !"image2d_msaa_depth_t",
+       !"image3d_t"}

test/CodeGen/AMDGPU/hsa-metadata-invalid-ocl-version-1-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck %s
+
+; Make sure llc does not crash for invalid opencl version metadata.
+
+; CHECK: ---
+; CHECK: amdhsa.version:
+; CHECK-NEXT: - 1
+; CHECK-NEXT: - 0
+; CHECK: ...
+
+!opencl.ocl.version = !{}

test/CodeGen/AMDGPU/hsa-metadata-invalid-ocl-version-2-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck %s
+
+; Make sure llc does not crash for invalid opencl version metadata.
+
+; CHECK: ---
+; CHECK: amdhsa.version:
+; CHECK-NEXT: - 1
+; CHECK-NEXT: - 0
+; CHECK: ...
+
+!opencl.ocl.version = !{!0}
+!0 = !{}

test/CodeGen/AMDGPU/hsa-metadata-invalid-ocl-version-3-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck %s
+
+; Make sure llc does not crash for invalid opencl version metadata.
+
+; CHECK: ---
+; CHECK: amdhsa.version:
+; CHECK-NEXT: - 1
+; CHECK-NEXT: - 0
+; CHECK: ...
+
+!opencl.ocl.version = !{!0}
+!0 = !{i32 1}

test/CodeGen/AMDGPU/hsa-metadata-kernel-code-props-v3.ll

+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx700 -enable-misched=0 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX700 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -enable-misched=0 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX803 --check-prefix=NOTES %s
+; RUN: llc -mattr=+code-object-v3 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -enable-misched=0 -filetype=obj -o - < %s | llvm-readobj -elf-output-style=GNU -notes | FileCheck --check-prefix=CHECK --check-prefix=GFX900 --check-prefix=NOTES %s
+
+@var = addrspace(1) global float 0.0
+
+; CHECK: ---
+; CHECK:  amdhsa.kernels:
+
+; CHECK: - .max_flat_workgroup_size:    256
+; CHECK:   .kernarg_segment_size:       24
+; CHECK:   .private_segment_fixed_size: 0
+; CHECK:   .wavefront_size:             64
+; CHECK:   .symbol:     test.kd
+; CHECK:   .name:       test
+; CHECK:   .sgpr_count:                 8
+; CHECK:   .kernarg_segment_align:      8
+; CHECK:   .vgpr_count:                 6
+; CHECK:   .group_segment_fixed_size:   0
+define amdgpu_kernel void @test(
+    half addrspace(1)* %r,
+    half addrspace(1)* %a,
+    half addrspace(1)* %b) {
+entry:
+  %a.val = load half, half addrspace(1)* %a
+  %b.val = load half, half addrspace(1)* %b
+  %r.val = fadd half %a.val, %b.val
+  store half %r.val, half addrspace(1)* %r
+  ret void
+}
+
+; CHECK:   .symbol:     num_spilled_sgprs.kd
+; CHECK:   .name:       num_spilled_sgprs
+; GFX700:   .sgpr_spill_count: 40
+; GFX803:   .sgpr_spill_count: 24
+; GFX900:   .sgpr_spill_count: 24
+define amdgpu_kernel void @num_spilled_sgprs(
+    i32 addrspace(1)* %out0, i32 addrspace(1)* %out1, [8 x i32],
+    i32 addrspace(1)* %out2, i32 addrspace(1)* %out3, [8 x i32],
+    i32 addrspace(1)* %out4, i32 addrspace(1)* %out5, [8 x i32],
+    i32 addrspace(1)* %out6, i32 addrspace(1)* %out7, [8 x i32],
+    i32 addrspace(1)* %out8, i32 addrspace(1)* %out9, [8 x i32],
+    i32 addrspace(1)* %outa, i32 addrspace(1)* %outb, [8 x i32],
+    i32 addrspace(1)* %outc, i32 addrspace(1)* %outd, [8 x i32],
+    i32 addrspace(1)* %oute, i32 addrspace(1)* %outf, [8 x i32],
+    i32 %in0, i32 %in1, i32 %in2, i32 %in3, [8 x i32],
+    i32 %in4, i32 %in5, i32 %in6, i32 %in7, [8 x i32],
+    i32 %in8, i32 %in9, i32 %ina, i32 %inb, [8 x i32],
+    i32 %inc, i32 %ind, i32 %ine, i32 %inf) #0 {
+entry:
+  store i32 %in0, i32 addrspace(1)* %out0
+  store i32 %in1, i32 addrspace(1)* %out1
+  store i32 %in2, i32 addrspace(1)* %out2
+  store i32 %in3, i32 addrspace(1)* %out3
+  store i32 %in4, i32 addrspace(1)* %out4
+  store i32 %in5, i32 addrspace(1)* %out5
+  store i32 %in6, i32 addrspace(1)* %out6
+  store i32 %in7, i32 addrspace(1)* %out7
+  store i32 %in8, i32 addrspace(1)* %out8
+  store i32 %in9, i32 addrspace(1)* %out9
+  store i32 %ina, i32 addrspace(1)* %outa
+  store i32 %inb, i32 addrspace(1)* %outb
+  store i32 %inc, i32 addrspace(1)* %outc
+  store i32 %ind, i32 addrspace(1)* %outd
+  store i32 %ine, i32 addrspace(1)* %oute
+  store i32 %inf, i32 addrspace(1)* %outf
+  ret void
+}
+
+; CHECK:   .symbol:     num_spilled_vgprs.kd
+; CHECK:   .name:       num_spilled_vgprs
+; CHECK:   .vgpr_spill_count: 14
+define amdgpu_kernel void @num_spilled_vgprs() #1 {
+  %val0 = load volatile float, float addrspace(1)* @var
+  %val1 = load volatile float, float addrspace(1)* @var
+  %val2 = load volatile float, float addrspace(1)* @var
+  %val3 = load volatile float, float addrspace(1)* @var
+  %val4 = load volatile float, float addrspace(1)* @var
+  %val5 = load volatile float, float addrspace(1)* @var
+  %val6 = load volatile float, float addrspace(1)* @var
+  %val7 = load volatile float, float addrspace(1)* @var
+  %val8 = load volatile float, float addrspace(1)* @var
+  %val9 = load volatile float, float addrspace(1)* @var
+  %val10 = load volatile float, float addrspace(1)* @var
+  %val11 = load volatile float, float addrspace(1)* @var
+  %val12 = load volatile float, float addrspace(1)* @var
+  %val13 = load volatile float, float addrspace(1)* @var
+  %val14 = load volatile float, float addrspace(1)* @var
+  %val15 = load volatile float, float addrspace(1)* @var
+  %val16 = load volatile float, float addrspace(1)* @var
+  %val17 = load volatile float, float addrspace(1)* @var
+  %val18 = load volatile float, float addrspace(1)* @var
+  %val19 = load volatile float, float addrspace(1)* @var
+  %val20 = load volatile float, float addrspace(1)* @var
+  %val21 = load volatile float, float addrspace(1)* @var
+  %val22 = load volatile float, float addrspace(1)* @var
+  %val23 = load volatile float, float addrspace(1)* @var
+  %val24 = load volatile float, float addrspace(1)* @var
+  %val25 = load volatile float, float addrspace(1)* @var
+  %val26 = load volatile float, float addrspace(1)* @var
+  %val27 = load volatile float, float addrspace(1)* @var
+  %val28 = load volatile float, float addrspace(1)* @var
+  %val29 = load volatile float, float addrspace(1)* @var
+  %val30 = load volatile float, float addrspace(1)* @var
+
+  store volatile float %val0, float addrspace(1)* @var
+  store volatile float %val1, float addrspace(1)* @var
+  store volatile float %val2, float addrspace(1)* @var
+  store volatile float %val3, float addrspace(1)* @var
+  store volatile float %val4, float addrspace(1)* @var
+  store volatile float %val5, float addrspace(1)* @var
+  store volatile float %val6, float addrspace(1)* @var
+  store volatile float %val7, float addrspace(1)* @var
+  store volatile float %val8, float addrspace(1)* @var
+  store volatile float %val9, float addrspace(1)* @var
+  store volatile float %val10, float addrspace(1)* @var
+  store volatile float %val11, float addrspace(1)* @var
+  store volatile float %val12, float addrspace(1)* @var
+  store volatile float %val13, float addrspace(1)* @var
+  store volatile float %val14, float addrspace(1)* @var
+  store volatile float %val15, float addrspace(1)* @var
+  store volatile float %val16, float addrspace(1)* @var
+  store volatile float %val17, float addrspace(1)* @var
+  store volatile float %val18, float addrspace(1)* @var
+  store volatile float %val19, float addrspace(1)* @var
+  store volatile float %val20, float addrspace(1)* @var
+  store volatile float %val21, float addrspace(1)* @var
+  store volatile float %val22, float addrspace(1)* @var
+  store volatile float %val23, float addrspace(1)* @var
+  store volatile float %val24, float addrspace(1)* @var
+  store volatile float %val25, float addrspace(1)* @var
+  store volatile float %val26, float addrspace(1)* @var
+  store volatile float %val27, float addrspace(1)* @var
+  store volatile float %val28, float addrspace(1)* @var
+  store volatile float %val29, float addrspace(1)* @var
+  store volatile float %val30, float addrspace(1)* @var
+
+  ret void
+}
+
+; CHECK:  amdhsa.version:
+; CHECK-NEXT: - 1
+; CHECK-NEXT: - 0
+
+attributes #0 = { "amdgpu-num-sgpr"="14" }
+attributes #1 = { "amdgpu-num-vgpr"="20" }

test/MC/AMDGPU/hsa-metadata-kernel-args-v3.s

+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx700 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX700 %s
+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx800 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX800 %s
+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx900 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX900 %s
+
+// CHECK:  .amd_amdgpu_hsa_metadata
+// CHECK:    amdhsa.kernels:
+// CHECK-NEXT:      - .max_flat_workgroup_size: 256
+// CHECK-NEXT:        .wavefront_size: 128
+// CHECK-NEXT:        .symbol:      'test_kernel@kd'
+// CHECK-NEXT:        .kernarg_segment_size: 8
+// CHECK-NEXT:        .private_segment_fixed_size: 32
+// CHECK-NEXT:        .name:            test_kernel
+// CHECK-NEXT:        .language:        OpenCL C
+// CHECK-NEXT:        .sgpr_count: 14
+// CHECK-NEXT:        .kernarg_segment_align: 64
+// CHECK-NEXT:        .vgpr_count: 40
+// CHECK-NEXT:        .group_segment_fixed_size: 16
+// CHECK-NEXT:        .language_version:
+// CHECK-NEXT:          - 2
+// CHECK-NEXT:          - 0
+// CHECK-NEXT:        .args:
+// CHECK-NEXT:          - .type_name:      char
+// CHECK-NEXT:            .value_kind:     by_value
+// CHECK-NEXT:            .offset:         1
+// CHECK-NEXT:            .size:          1
+// CHECK-NEXT:            .value_type:     i8
+// CHECK-NEXT:          - .value_kind:     hidden_global_offset_x
+// CHECK-NEXT:            .offset:         8
+// CHECK-NEXT:            .size:          8
+// CHECK-NEXT:            .value_type:     i64
+// CHECK-NEXT:          - .value_kind:     hidden_global_offset_y
+// CHECK-NEXT:            .offset:         8
+// CHECK-NEXT:            .size:          8
+// CHECK-NEXT:            .value_type:     i64
+// CHECK-NEXT:          - .value_kind:     hidden_global_offset_z
+// CHECK-NEXT:            .offset:         8
+// CHECK-NEXT:            .size:          8
+// CHECK-NEXT:            .value_type:     i64
+// CHECK-NEXT:          - .value_kind:     hidden_printf_buffer
+// CHECK-NEXT:            .offset:         8
+// CHECK-NEXT:            .size:          8
+// CHECK-NEXT:            .value_type:     i8
+// CHECK-NEXT:            .address_space: global
+// CHECK:    amdhsa.version:
+// CHECK-NEXT:       - 1
+// CHECK-NEXT:       - 0
+// CHECK:    amdhsa.printf:
+// CHECK-NEXT:      - '1:1:4:%d\n'
+// CHECK-NEXt:      - '2:1:8:%g\n'
+// CHECK:  .end_amd_amdgpu_hsa_metadata
+.amd_amdgpu_hsa_metadata
+  amdhsa.version:
+    - 1
+    - 0
+  amdhsa.printf:
+    - '1:1:4:%d\n'
+    - '2:1:8:%g\n'
+  amdhsa.kernels:
+    - .name:            test_kernel
+      .symbol:      test_kernel@kd
+      .language:        OpenCL C
+      .language_version:
+        - 2
+        - 0
+      .kernarg_segment_size: 8
+      .group_segment_fixed_size: 16
+      .private_segment_fixed_size: 32
+      .kernarg_segment_align: 64
+      .wavefront_size: 128
+      .sgpr_count: 14
+      .vgpr_count: 40
+      .max_flat_workgroup_size: 256
+      .args:
+        - .type_name:      char
+          .size:          1
+          .offset:         1
+          .value_kind:     by_value
+          .value_type:     i8
+        - .size:          8
+          .offset:         8
+          .value_kind:     hidden_global_offset_x
+          .value_type:     i64
+        - .size:          8
+          .offset:         8
+          .value_kind:     hidden_global_offset_y
+          .value_type:     i64
+        - .size:          8
+          .offset:         8
+          .value_kind:     hidden_global_offset_z
+          .value_type:     i64
+        - .size:          8
+          .offset:         8
+          .value_kind:     hidden_printf_buffer
+          .value_type:     i8
+          .address_space: global
+.end_amd_amdgpu_hsa_metadata

test/MC/AMDGPU/hsa-metadata-kernel-attrs-v3.s

+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx700 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX700 %s
+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx800 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX800 %s
+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx900 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX900 %s
+
+// CHECK:  .amd_amdgpu_hsa_metadata
+// CHECK:    amdhsa.kernels:
+// CHECK:      - .max_flat_workgroup_size: 256
+// CHECK:        .wavefront_size: 128
+// CHECK:        .symbol:      'test_kernel@kd'
+// CHECK:        .reqd_workgroup_size:
+// CHECK-NEXT:     - 1
+// CHECK-NEXT:     - 2
+// CHECK-NEXT:     - 4
+// CHECK:        .kernarg_segment_size: 8
+// CHECK:        .private_segment_fixed_size: 32
+// CHECK:        .workgroup_size_hint:
+// CHECK-NEXT:     - 8
+// CHECK-NEXT:     - 16
+// CHECK-NEXT:     - 32
+// CHECK:        .name:            test_kernel
+// CHECK:        .language:        OpenCL C
+// CHECK:        .sgpr_count:      14
+// CHECK:        .kernarg_segment_align: 64
+// CHECK:        .vgpr_count:      40
+// CHECK:        .language_version:
+// CHECK-NEXT:     - 2
+// CHECK-NEXT:     - 0
+// CHECK:        .vec_type_hint:       int
+// CHECK:    amdhsa.version:
+// CHECK-NEXT: - 1
+// CHECK-NEXT: - 0
+// CHECK:    amdhsa.printf:
+// CHECK:      - '1:1:4:%d\n'
+// CHECK:      - '2:1:8:%g\n'
+// CHECK: .end_amd_amdgpu_hsa_metadata
+.amd_amdgpu_hsa_metadata
+  amdhsa.version:
+    - 1
+    - 0
+  amdhsa.printf:
+    - '1:1:4:%d\n'
+    - '2:1:8:%g\n'
+  amdhsa.kernels:
+    - .name:            test_kernel
+      .symbol:      test_kernel@kd
+      .language:        OpenCL C
+      .language_version:
+        - 2
+        - 0
+      .kernarg_segment_size: 8
+      .group_segment_fixed_size: 16
+      .private_segment_fixed_size: 32
+      .kernarg_segment_align: 64
+      .wavefront_size: 128
+      .sgpr_count: 14
+      .vgpr_count: 40
+      .max_flat_workgroup_size: 256
+      .reqd_workgroup_size:
+        - 1
+        - 2
+        - 4
+      .workgroup_size_hint:
+        - 8
+        - 16
+        - 32
+      .vec_type_hint:       int
+.end_amd_amdgpu_hsa_metadata

test/MC/AMDGPU/hsa-metadata-kernel-code-props-v3.s

+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx700 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX700 %s
+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx800 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX800 %s
+// RUN: llvm-mc -mattr=+code-object-v3 -triple=amdgcn-amd-amdhsa -mcpu=gfx900 -show-encoding %s | FileCheck --check-prefix=CHECK --check-prefix=GFX900 %s
+
+// CHECK:  .amd_amdgpu_hsa_metadata
+// CHECK:    amdhsa.kernels:
+// CHECK:      - .sgpr_count: 40
+// CHECK:        .max_flat_workgroup_size:    256
+// CHECK:        .symbol: 'test_kernel@kd'
+// CHECK:        .kernarg_segment_size:      24
+// CHECK:        .group_segment_fixed_size:   24
+// CHECK:        .private_segment_fixed_size: 16
+// CHECK:        .vgpr_count: 14
+// CHECK:        .vgpr_spill_count: 1
+// CHECK:        .kernarg_segment_align:     16
+// CHECK:        .sgpr_spill_count: 1
+// CHECK:        .wavefront_size:           64
+// CHECK:        .name:       test_kernel
+// CHECK:    amdhsa.version:
+// CHECK-NEXT: - 1
+// CHECK-NEXT: - 0
+.amd_amdgpu_hsa_metadata
+  amdhsa.version:
+    - 1
+    - 0
+  amdhsa.printf:
+    - '1:1:4:%d\n'
+    - '2:1:8:%g\n'
+  amdhsa.kernels:
+    - .name:            test_kernel
+      .symbol:      test_kernel@kd
+      .kernarg_segment_size:      24
+      .group_segment_fixed_size:   24
+      .private_segment_fixed_size: 16
+      .kernarg_segment_align:     16
+      .wavefront_size:           64
+      .max_flat_workgroup_size:    256
+      .sgpr_count:               40
+      .vgpr_count:               14
+      .sgpr_spill_count:         1
+      .vgpr_spill_count:         1
+.end_amd_amdgpu_hsa_metadata

test/MC/AMDGPU/hsa-v3.s

 // ASM: .byte 0
 
 v_mov_b32_e32 v16, s3
 
 .byte .amdgcn.next_free_vgpr
 // ASM: .byte 17
 .byte .amdgcn.next_free_sgpr
 // ASM: .byte 4
+
+// Metadata
+
+.amd_amdgpu_hsa_metadata
+  amdhsa.version:
+    - 3
+    - 0
+  amdhsa.kernels:
+    - .name:       amd_kernel_code_t_test_all
+      .symbol: amd_kernel_code_t_test_all@kd
+      .kernarg_segment_size: 8
+      .group_segment_fixed_size: 16
+      .private_segment_fixed_size: 32
+      .kernarg_segment_align: 64
+      .wavefront_size: 128
+      .sgpr_count: 14
+      .vgpr_count: 40
+      .max_flat_workgroup_size: 256
+    - .name:       amd_kernel_code_t_minimal
+      .symbol: amd_kernel_code_t_minimal@kd
+      .kernarg_segment_size: 8
+      .group_segment_fixed_size: 16
+      .private_segment_fixed_size: 32
+      .kernarg_segment_align: 64
+      .wavefront_size: 128
+      .sgpr_count: 14
+      .vgpr_count: 40
+      .max_flat_workgroup_size: 256
+.end_amd_amdgpu_hsa_metadata
+
+// ASM: .amd_amdgpu_hsa_metadata
+// ASM:    amdhsa.kernels:
+// ASM:  - .sgpr_count:     14
+// ASM:    .max_flat_workgroup_size: 256
+// ASM:    .symbol:         'amd_kernel_code_t_test_all@kd'
+// ASM:    .kernarg_segment_size: 8
+// ASM:    .group_segment_fixed_size: 16
+// ASM:    .private_segment_fixed_size: 32
+// ASM:    .vgpr_count:     40
+// ASM:    .kernarg_segment_align: 64
+// ASM:    .wavefront_size: 128
+// ASM:    .name:           amd_kernel_code_t_test_all
+// ASM:  - .sgpr_count:     14
+// ASM:    .max_flat_workgroup_size: 256
+// ASM:    .symbol:         'amd_kernel_code_t_minimal@kd'
+// ASM:    .kernarg_segment_size: 8
+// ASM:    .group_segment_fixed_size: 16
+// ASM:    .private_segment_fixed_size: 32
+// ASM:    .vgpr_count:     40
+// ASM:    .kernarg_segment_align: 64
+// ASM:    .wavefront_size: 128
+// ASM:    .name:           amd_kernel_code_t_minimal
+// ASM:    amdhsa.version:
+// ASM-NEXT: - 3
+// ASM-NEXT: - 0
+// ASM: .end_amd_amdgpu_hsa_metadata

tools/llvm-readobj/ELFDumper.cpp

 #include "llvm/ADT/Optional.h"
 #include "llvm/ADT/PointerIntPair.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/Twine.h"
+#include "llvm/BinaryFormat/AMDGPUMetadataVerifier.h"
 #include "llvm/BinaryFormat/ELF.h"
+#include "llvm/BinaryFormat/MsgPackTypes.h"
 #include "llvm/Object/ELF.h"
 #include "llvm/Object/ELFObjectFile.h"
 #include "llvm/Object/ELFTypes.h"
 #include "llvm/Object/Error.h"
 #include "llvm/Object/ObjectFile.h"
 #include "llvm/Object/StackMapParser.h"
 #include "llvm/Support/AMDGPUMetadata.h"
 #include "llvm/Support/ARMAttributeParser.h"
if (Note.ID == NT)
       return std::string(Note.Name);
 
   std::string string;
   raw_string_ostream OS(string);
   OS << format("Unknown note type (0x%08x)", NT);
   return OS.str();
 }
 
-static std::string getAMDGPUNoteTypeName(const uint32_t NT) {
+static std::string getAMDNoteTypeName(const uint32_t NT) {
   static const struct {
     uint32_t ID;
     const char *Name;
   } Notes[] = {
     {ELF::NT_AMD_AMDGPU_HSA_METADATA,
      "NT_AMD_AMDGPU_HSA_METADATA (HSA Metadata)"},
     {ELF::NT_AMD_AMDGPU_ISA,
      "NT_AMD_AMDGPU_ISA (ISA Version)"},
     {ELF::NT_AMD_AMDGPU_PAL_METADATA,
      "NT_AMD_AMDGPU_PAL_METADATA (PAL Metadata)"}
   };
 
   for (const auto &Note : Notes)
     if (Note.ID == NT)
       return std::string(Note.Name);
 
   std::string string;
   raw_string_ostream OS(string);
   OS << format("Unknown note type (0x%08x)", NT);
   return OS.str();
 }
 
+static std::string getAMDGPUNoteTypeName(const uint32_t NT) {
+  if (NT == ELF::NT_AMDGPU_METADATA)
+    return std::string("NT_AMDGPU_METADATA (AMDGPU Metadata)");
+
+  std::string string;
+  raw_string_ostream OS(string);
+  OS << format("Unknown note type (0x%08x)", NT);
+  return OS.str();
+}
+
 template <typename ELFT>
 static void printGNUProperty(raw_ostream &OS, uint32_t Type, uint32_t DataSize,
                              ArrayRef<uint8_t> Data) {
   switch (Type) {
   default:
     OS << format("    <application-specific type 0x%x>\n", Type);
     return;
   case GNU_PROPERTY_STACK_SIZE: {
case ELF::NT_GNU_PROPERTY_TYPE_0:
     if (!Arr.empty())
       OS << "    <corrupted GNU_PROPERTY_TYPE_0>";
     break;
   }
   OS << '\n';
 }
 
 template <typename ELFT>
-static void printAMDGPUNote(raw_ostream &OS, uint32_t NoteType,
-                            ArrayRef<typename ELFT::Word> Words, size_t Size) {
+static void printAMDNote(raw_ostream &OS, uint32_t NoteType,
   switch (NoteType) {
   default:
     return;
     case ELF::NT_AMD_AMDGPU_HSA_METADATA:
       OS << "    HSA Metadata:\n"
          << StringRef(reinterpret_cast<const char *>(Words.data()), Size);
       break;
     case ELF::NT_AMD_AMDGPU_ISA:
case ELF::NT_AMD_AMDGPU_PAL_METADATA:
         return;
       }
       OS << PALMetadataString;
       break;
   }
   OS.flush();
 }
 
+template <typename ELFT>
+static void printAMDGPUNote(raw_ostream &OS, uint32_t NoteType,
+                            ArrayRef<typename ELFT::Word> Words, size_t Size) {
+  OS << "    AMDGPU Metadata:\n";
+  auto MsgPackString = StringRef(reinterpret_cast<const char *>(Words.data()), Size);
+  msgpack::Reader MsgPackReader(MsgPackString);
+  auto OptMsgPackNodeOrErr = msgpack::Node::read(MsgPackReader);
+  if (auto Err = OptMsgPackNodeOrErr.takeError())
+    return error(std::move(Err));
+  auto &OptMsgPackNode = *OptMsgPackNodeOrErr;
+  if (!OptMsgPackNode)
+    return;
+  auto &MsgPackNode = *OptMsgPackNode;
+
+  AMDGPU::HSAMD::V3::MetadataVerifier Verifier(true);
+  if (!Verifier.verify(*MsgPackNode))
+    reportError("invalid AMDGPU metadata");
+
+  std::string HSAMetadataString;
+  raw_string_ostream StrOS(HSAMetadataString);
+  yaml::Output YOut(StrOS);
+  YOut << MsgPackNode;
+  OS << StrOS.str();
+  OS.flush();
+}
+
 template <class ELFT>
 void GNUStyle<ELFT>::printNotes(const ELFFile<ELFT> *Obj) {
   const Elf_Ehdr *e = Obj->getHeader();
   bool IsCore = e->e_type == ELF::ET_CORE;
 
   auto PrintHeader = [&](const typename ELFT::Off Offset,
                          const typename ELFT::Addr Size) {
     OS << "Displaying notes found at file offset " << format_hex(Offset, 10)
OS << "  " << Name << std::string(22 - Name.size(), ' ')
        << format_hex(Descriptor.size(), 10) << '\t';
 
     if (Name == "GNU") {
       OS << getGNUNoteTypeName(Type) << '\n';
       printGNUNote<ELFT>(OS, Type, Descriptor, Descriptor.size());
     } else if (Name == "FreeBSD") {
       OS << getFreeBSDNoteTypeName(Type) << '\n';
     } else if (Name == "AMD") {
+      OS << getAMDNoteTypeName(Type) << '\n';
+      printAMDNote<ELFT>(OS, Type, Descriptor, Descriptor.size());
+    } else if (Name == "AMDGPU") {
       OS << getAMDGPUNoteTypeName(Type) << '\n';
       printAMDGPUNote<ELFT>(OS, Type, Descriptor, Descriptor.size());
     } else {
       OS << "Unknown note type: (" << format_hex(Type, 10) << ')';
     }
     OS << '\n';
   };