Add challenge 76: Bilinear Image Scaling (Medium)

[claude]

Summary

• Adds challenge 76: Bilinear Image Scaling (Medium difficulty)
• Given an input image of dimensions H × W and target dimensions H_out × W_out, produce the scaled output using bilinear interpolation with the align-corners convention
• Validated locally with a correct CUDA solution passing all functional and example tests on NVIDIA Tesla T4

Why this challenge is interesting

• Each output pixel reads from 4 non-aligned input pixels — writes are coalesced but reads are not, making memory access patterns non-trivial
• Requires floating-point coordinate computation (inverse mapping: src = i * (in_size - 1) / (out_size - 1)) with correct boundary handling
• Tests understanding of 2D thread/block organization for image-shaped workloads
• Not element-wise: every output depends on a weighted blend of 4 neighbours
• Real-world operation used in every GPU vision pipeline (OpenCV CUDA, cuDNN resize, etc.)

Checklist

challenge.html

• Starts with <p> (problem description) — never <h1>
• Has <h2> sections for: Implementation Requirements, Example(s), Constraints
• First example matches generate_example_test() values (2×2 → 3×3 with [[1,3],[7,9]])
• Examples use LaTeX \begin{bmatrix} for matrices
• Constraints includes Performance is measured with H = 4,096, W = 4,096, H_out = 8,192, W_out = 8,192

challenge.py

• class Challenge inherits ChallengeBase
• __init__ calls super().__init__() with name, atol, rtol, num_gpus, access_tier
• reference_impl has assertions on shape, dtype, and device
• All 6 methods present
• generate_functional_test returns 10 cases covering edge cases (1×1, 2×2, 3×3 zeros, 4×12 width-only), powers-of-2 (16², 64², 256²), non-powers-of-2 (30×40, 100×150), realistic (1024²)
• generate_performance_test: 4096×4096 → 8192×8192 (320 MB total, well within 16 GB × 5)

Starter files

• All 6 files present: .cu, .pytorch.py, .triton.py, .jax.py, .cute.py, .mojo
• Exactly 1 parameter description comment per file, no other comments
• CUDA/Mojo use "device pointers" (medium challenge, no parenthetical)
• Python frameworks use "tensors on the GPU"; JAX also has # return output tensor directly
• Starters compile/run but do NOT produce correct output

General

• Directory follows 76_bilinear_image_scaling convention
• Linting passes: pre-commit run --all-files
• Validated with run_challenge.py --action run — status: success

🤖 Generated with Claude Code