A06 Solutions

Reference answers with reasoning. Try the ../exercises/ yourself first — the learning is in the struggle, not the answer key.

This directory has its own Makefile:

make all GPU_ARCH=gfx950      # build every solution for your GPU
make run                      # build + run all; every one must PASS
make clean

#	File	Notes
1	01_tiled_matmul_solution.cpp	Complete tiled kernel. The header comment explains each of the four pieces, why both barriers are required, and why this kernel is already bank-conflict-free at TILE = 32.
2	02_ragged_dims_solution.cpp	Zero-filled tile loads. Answers Q1–Q3: why M/N raggedness is self-contained, why a K-edge bug corrupts everything, and what an unguarded store really does (it aliases the next row before it goes out of bounds). Adds two extra shapes, including K < TILE.
3	03_intensity_math_solution.md	Full derivation: I_naive = 0.25, I_tiled = T/4, I_regblock = BM·BN / (2(BM+BN)), the ridge crossing, why one-thread-per-output cannot reach it, and how matrix cores + bf16 move the target to T ≥ 320.

A note on what "correct" means here: make run stops at the first failure, because a solution that does not PASS is a bug in the curriculum. But fast is a different bar — these are teaching kernels and they are expected to sit well below rocBLAS / hipBLASLt / cuBLAS. See ../README.md § 6.