Contents

• 1. Introduction
• 2. Identifiers
• 3. High Level Structure
  • 3.1. Linkage Types
  • 3.2. Calling Conventions
    • 3.2.1. Rules and Restrictions
  • 3.3. Visibility Styles
  • 3.4. DLL Storage Classes
  • 3.5. Thread Local Storage Models
  • 3.6. Runtime Preemption Specifiers
  • 3.7. Structure Types
  • 3.8. Non-Integral Pointer Type
  • 3.9. Comdats
  • 3.10. source_filename
  • 3.11. Global Variables
  • 3.12. Functions
  • 3.13. Aliases
  • 3.14. Ifuncs
  • 3.15. Named Metadata
  • 3.16. Parameter Attributes
  • 3.17. Garbage Collector Strategy Names
  • 3.18. Prefix Data
  • 3.19. Prologue Data
  • 3.20. Attribute Groups
  • 3.21. Function Attributes
  • 3.22. Global Attributes
  • 3.23. Operand Bundles
  • 3.24. Module-Level Inline Assembly
  • 3.25. Data Layout
  • 3.26. Target Triple
  • 3.27. Pointer Aliasing Rules
  • 3.28. Volatile Memory Access
  • 3.29. Memory Model for Concurrent Operations
  • 3.30. Atomic Memory Ordering Constraints
  • 3.31. Fast-Math Flags
  • 3.32. Use-list Order Directives
• 4. Type System
• 5. Constants
• 6. Other Values
  • 6.1. Inline Assembler Expressions
• 7. Metadata
  • 7.1. Metadata Nodes and Metadata Strings
• 8. ThinLTO Summary
• 9. Intrinsic Global Variables
• 10. Instructions
  • 10.1. Terminator Instructions
  • 10.2. Binary Operations
  • 10.3. Bitwise Binary Operations
  • 10.4. Vector Operations
  • 10.5. Aggregate Operations
  • 10.6. Memory Access and Addressing Operations
    • 10.6.1. alloca Instruction
    • 10.6.2. load Instruction
    • 10.6.3. store Instruction
    • 10.6.4. fence Instruction
    • 10.6.5. cmpxchg Instruction
    • 10.6.6. atomicrmw Instruction
    • 10.6.7. getelementptr Instruction
  • 10.7. Conversion Operations
  • 10.8. Other Operations
• 11. Supported Intrinsic Functions
  • 11.1. Supported Variable Argument Handling Intrinsics
  • 11.2. Supported Standard C/C++ Library Intrinsics
  • 11.3. Supported Bit Manipulations Intrinsics
  • 11.4. Supported Specialised Arithmetic Intrinsics
  • 11.5. Supported Arithmetic with Overflow Intrinsics
  • 11.6. Supported Half Precision Floating Point Intrinsics
  • 11.7. Supported Debugger Intrinsics
  • 11.8. Supported Memory Use Markers
  • 11.9. Supported General Intrinsics
• 12. Address Space
  • 12.1. Address Spaces
  • 12.2. Generic Pointers and Non-Generic Pointers
    • 12.2.1. Generic Pointers vs. Non-generic Pointers
    • 12.2.2. Conversion
    • 12.2.3. No Aliasing between Two Different Specific Address Spaces
  • 12.3. The alloca Instruction
• 13. Global Property Annotation
  • 13.1. Overview
  • 13.2. Representation of Properties
  • 13.3. Supported Properties
• 14. Texture and Surface
  • 14.1. Texture Variable and Surface Variable
  • 14.2. Accessing Texture Memory or Surface Memory
• 15. NVVM Specific Intrinsic Functions
  • 15.1. Atomic
  • 15.2. Barrier and Memory Fence
  • 15.3. Address space conversion
  • 15.4. Special Registers
  • 15.5. Texture/Surface Access
    • 15.5.1. Texture Reads
    • 15.5.2. Surface Loads
    • 15.5.3. Surface Stores
  • 15.6. Warp-level Operations
    • 15.6.1. Barrier Synchronization
    • 15.6.2. Data Movement
    • 15.6.3. Vote
    • 15.6.4. Match
    • 15.6.5. Matrix Operation
      • 15.6.5.1. Load Fragments
      • 15.6.5.2. Store Fragments
      • 15.6.5.3. Matrix Multiply-and-Accumulate
• 16. Source Level Debugging Support
• 17. NVVM ABI for PTX
  • 17.1. Linkage Types
  • 17.2. Parameter Passing and Return
• 18. Revision History
• 19. Notices
  • 19.1. Notice
  • 19.2. OpenCL
  • 19.3. Trademarks