Advanced

Kernel Correctness Contracts

A fast GPU kernel is useful only for inputs it handles correctly. Define shapes, strides, dtypes, numerical tolerance, storage ownership and synchronization before timing a replacement.

This concept is part of Premium, which includes all concepts, course lessons and answers.

Get Premium · ₹2,000 / $256 months of access · one payment · no auto-renew

RELATED CONCEPTS

CUDA Compilation: PTX, Cubins and SASS Numerics: FP32, BF16, FP8 and FP4 Asynchronous GPU Buffer Ownership Profiling with Nsight Parallel Prefix Scan and Stream Compaction

LESSONS THAT TEACH THIS

Kernels and the MachineBalance the grid before tuning the tile18 min→Kernels and the MachineProve the kernel contract before timing18 min→

PRACTICE THIS IN REAL QUESTIONS

CUDA, Triton & Kernel EngineeringYour fused attention kernel matches the reference at 512 tokens and drifts at 8,000. Find the bug.→CUDA, Triton & Kernel EngineeringAn AI-generated kernel passes random tests and claims a 3x speedup. What would you verify?→CUDA, Triton & Kernel EngineeringMap a fused bias-GELU kernel to threads and verify the final partial block→Coding for InfraImplement symmetric INT8 quantization and prove where the reconstruction-error bound applies→CUDA, Triton & Kernel EngineeringWhy do tensor cores accumulate in fp32 when the inputs are bf16, and why do optimizers keep fp32 master weights?→Coding for InfraGiven usage records of GPU allocations, compute what each tenant owes. Write it, and say what you would test.→