PTX is NVIDIA’s textual virtual instruction set for GPUs. In Triton’s compilation pipeline, LLVM IR is translated into PTX, which NVIDIA’s assembler turns into a CUBIN binary for execution. PTX offers low-level control but is difficult to work with; writing it directly is rarely necessary. section 1.2.1 section 2.3.1
Figure 2.10 The Triton compiler pipeline: from user code through AST, TTIR, and TTGIR to GPU execution via LLVM-IR, PTX, and CUBIN.
Trace PTX’s place in the Triton compiler pipeline, from user code through intermediate representations to GPU execution. section 2.3.1

