7 lines
1.6 KiB
HTML
7 lines
1.6 KiB
HTML
<!doctype html><html lang="en"><meta charset="utf-8"><title>Coalescing and shared workgroup memory</title><h1>Coalescing and shared workgroup memory</h1><p>15-minute session: read for 3 minutes, trace for 3, experiment for 7, and complete the check for 2. Spend longer whenever you need.</p><p>GPU lanes accessing adjacent elements can allow efficient memory transactions. Workgroup-shared memory may reduce repeated global reads, but it consumes a limited resource and requires correct synchronization. Names and details vary across CUDA, Metal, Vulkan, and other APIs. Draw which lane reads which element before changing data layout; validate the output after every layout change.</p><p><strong>Practice and evidence:</strong> Predict the example or draw its state transitions. Change one assumption, test the boundary, and record the actual result. Explain what this small model does not establish about a production system.</p><pre><code class="language-python">lanes=4
|
|
contiguous=[lane for lane in range(lanes)]
|
|
strided=[lane*16 for lane in range(lanes)]
|
|
print(contiguous)
|
|
print(strided)
|
|
</code></pre><p><strong>Expected result:</strong> <code>[0, 1, 2, 3]
|
|
[0, 16, 32, 48]</code></p><h3>Practice and evidence</h3><p>Keep three notes: your prediction, the actual result, and why they match or differ. The completion check grades understanding; the practice task builds skill.</p><p><a href="https://git.agilaos.com/matt/open_agx">Language reference: git.agilaos.com</a></p><p><a href="https://docs.nvidia.com/cuda/cuda-c-best-practices-guide/">Language reference: docs.nvidia.com</a></p></html> |