GPUFlight A thread is not an array index
An image is 11 pixels wide, yet CUDA launches 12 threads across it. Which threads own pixels, and why can a bad coordinate still give a valid-looking address?
Read the explanation →GPUFLIGHT / KNOWLEDGE LIBRARY
How GPU programs work
Understand how threads find their data, how a block shares results, and how a line of CUDA becomes machine instructions. Each reading follows one example. No GPU or setup is required.
x = 10 owns the last pixel. x = 11 has no pixel.
01 / EXPLAINERS
Three questions to start with
Read in order or follow a question.
When is shared memory ready to read?
Two threads agree on where to exchange a value. What ensures that the value is there by the time the other thread reads it?
Read the explanation →What changes between CUDA, PTX, and SASS?
Follow one increment from a line of CUDA through virtual registers to a native instruction, and see which questions the listing can answer.
Read the explanation →02 / REFERENCE
Go deeper
FIELD GUIDES
Read by subject
GLOSSARY
Find the term you need
Short definitions, with a link to an example that shows why each term matters.
After reading
Ready to profile and optimize a kernel? Visit Performance Lab ↗