GPU llops

This commit is contained in:
George Hotz
2022-06-05 13:49:39 -07:00
parent f0fe37bd34
commit 7a3fe34db1
3 changed files with 158 additions and 142 deletions

View File

@@ -1,18 +1,18 @@
Getting the core instruction set correct is the value of tinygrad
Max size tensor is 6-D for the pool2d
Unary Ops
===
These are the simplest to reason about, and have pointwise mem access.
A and B are always the same size
Forward : A -> B
Backward (binary): (B', A) -> A'
Reduce Ops (with axis)
===