Cache Prefetch Memory
Hint the memory system to bring a cache line into a specified cache level ahead of an anticipated access, without producing a register result.
Vendor-Neutral Definition
Cache[address] is fetched into the specified cache level as a hint; has no architectural effect on register state and does not block on completion.
⚠️ Cross-vendor mappings describe semantic relationships and are not guaranteed one-to-one compiler
translations.
equivalent with restrictions
Both hint the memory system to bring a cache line into cache ahead of an anticipated access, with no architectural effect on register state.
| Property | NVIDIA PTX | AMDGPU |
|---|---|---|
| Instruction | prefetch, prefetchu | global_prefetch_b8, flat_prefetch_b8 |
| ISA Layer | Virtual | Native |
| Data Types | - | b8 |
| Version / Target Introduced | PTX ISA 2.0, PTX ISA 2.0 | - |
Important Differences
- PTX's prefetch supports selecting a specific cache level (.L1/.L2) as a modifier; the AMDGPU prefetch instructions in this dataset are per-address-space (global vs. flat aperture) rather than per-cache-level, so the exact cache-level targeting behavior should be checked against AMD's documentation rather than assumed to match PTX's modifier one-for-one.
Source Evidence
- Parallel Thread Execution ISA ↗ - NVIDIA Corporation
- User Guide for AMDGPU Backend ↗ - LLVM Project
Verification method: documentation