Elle yazılan SIMD, derleyiciyi ne zaman geçer?
Aynı dot product döngüsünü dört şekilde derledim. 17 katlık farkın arkasında toplama sırası, birkaç bayrak ve derleyicinin ürettiği kod var.
9 min read
A PERSONAL SYSTEMS NOTEBOOK
Closer to
the machine.
C++, CUDA, inference & the systems underneath.
Learning by building. Understanding by measuring.
Notes by Omer Isildak
Merakla başlayan, notlarla büyüyen.CURRENT ORBITC++ / CUDAInferenceReliabilityRead. Build. Repeat.
THE NOTEBOOK
Recent notes.
THE PERSON BEHIND THE NOTES
I like the questions under the abstractions.
Why did latency climb? Where did the memory go? What actually ran on the GPU? This is where I keep those questions—and the small experiments that follow.
A little about me