AI Architecture & Systems

Kernel Code Generation and ISA Boundaries

SourcesNVIDIA, AMD, TPU and Neuron layer mappings.

3.9.1Instruction Selection, Instruction Scheduling, and Register Allocation#

3.9.2LLVM/NVVM/AMDGPU Backends#

3.9.3PTX, SASS, and AMD Device Instruction Sets#

3.9.4Thread-Level, Tile-Level, and Tensor-Intrinsic Lowering Paths#

3.9.5Triton IR → GPU IR → Device Code#

3.9.6Cooperation Among Vector, Matrix, and Scalar Paths#

3.9.7Distinguishing Public ISAs, Public Intrinsics, and Undisclosed Machine Instructions#

3.9.8Cross-Architecture Code Generation, Feature Detection, and Fallback#