IMPORTANT: To view this page as Markdown, append `.md` to the URL (e.g. /get-started.md). For the complete documentation index, see llms.txt.
Skip to main content
For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Mojo module

intrinsics

Provides low-level GPU intrinsic operations and memory access primitives.

Implements hardware-specific intrinsics that map directly to GPU assembly instructions, focusing on NVIDIA GPU architectures. Includes:

  • Global memory load/store operations with cache control
  • Warp-level primitives and synchronization
  • Memory fence and barrier operations
  • Atomic operations and memory ordering primitives

These low-level primitives should be used carefully as they correspond directly to hardware instructions and require understanding of the underlying GPU architecture.

Structs​

  • ​AMDBufferResource: 128-bit descriptor for a buffer resource on AMD GPUs.
  • ​CacheOperation: Represents different GPU cache operation policies.
  • ​Scope: Represents memory synchronization scope levels for GPU memory operations.

Functions​

  • ​byte_permute: Permutes bytes from two 32-bit integers based on a control mask.
  • ​cvt_pk_fp8_f32_raw: Packs 4 f32 into 4 fp8 via 2 chained v_cvt_pk_fp8_f32 ops.
  • ​ds_read_tr16_b64: Reads a 64-bit LDS transpose block using TR16 layout and returns SIMD[dtype, 4] of 16-bit types.
  • ​ds_read_tr8_b64: Reads a 64-bit LDS transpose block using TR8 layout and returns SIMD[dtype, 8] of 8-bit types.
  • ​get_ib_sts: Returns the IB status of the current thread.
  • ​implicitarg_ptr: Gets a pointer to AMD's implicit kernel arguments table.
  • ​lop: Performs an arbitrary logical operation on 3 inputs using a lookup table.
  • ​permlane_shuffle: Shuffles SIMD values across lanes using AMD permlane operations.
  • ​permlane_swap: Swaps values between lanes using AMD permlane swap instruction.
  • ​readfirstlane: Gets the value in the lowest active lane of the input operand.
  • ​sendmsg: Sends a message to fixed-function hardware.
  • ​threadfence: Enforces ordering of memory operations across threads.
  • ​warpgroup_reg_alloc: Allocates additional registers for the executing warp group.
  • ​warpgroup_reg_dealloc: Deallocates additional registers for the executing warp group.

Was this page helpful?