Type Conversion Data Movement
Convert a value between integer and/or floating-point types, with a specified rounding and saturation mode.
Vendor-Neutral Definition
d = convert(a, from_type -> to_type), with an explicit rounding mode (e.g. round-to-nearest, truncate) and optional saturation on overflow.
⚠️ Cross-vendor mappings describe semantic relationships and are not guaranteed one-to-one compiler
translations.
equivalent with restrictions
Both convert a value between integer and floating-point representations.
| Property | NVIDIA PTX | AMDGPU |
|---|---|---|
| Instruction | cvt | v_cvt_f32_i32 |
| ISA Layer | Virtual | Native |
| Data Types | f16, f32, f64, s16, s32, s64, s8, u16, u32, u64, u8 | f32, i32 |
| Version / Target Introduced | PTX ISA 1.0 | gfx942/gfx1100 |
Important Differences
- Supported rounding-mode/saturation modifier combinations are not identical across every PTX cvt pair and its closest AMDGPU v_cvt_* counterpart; verify each specific type pair independently.
Source Evidence
- Parallel Thread Execution ISA ↗ — NVIDIA Corporation
- User Guide for AMDGPU Backend ↗ — LLVM Project
Verification method: documentation