commaai
/
tinygrad
mirror of https://github.com/commaai/tinygrad.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

discard children on mop shuffle, 200 -> 196 kernels

This commit is contained in:
George Hotz 2023-02-25 10:51:07 -08:00
parent 1d01842232
commit a44e8e4385
2 changed files with 7 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.github/workflows/test.yml vendored
View File

@ -161,7 +161,7 @@ jobs:
run: GPU=1 IMAGE=2 python3 test/test_ops.py
- name: Test openpilot model
run: |
ALLOWED_KERNEL_COUNT=200 FLOAT16=1 VALIDHACKS=1 DEBUGCL=1 GPU=1 IMAGE=2 python3 openpilot/compile.py
ALLOWED_KERNEL_COUNT=196 FLOAT16=1 VALIDHACKS=1 DEBUGCL=1 GPU=1 IMAGE=2 python3 openpilot/compile.py
UNSAFE_FLOAT4=1 DEBUGCL=1 GPU=1 IMAGE=2 python3 openpilot/compile.py
# disabled, this test is flaky

9
tinygrad/lazy.py
View File

@ -202,21 +202,24 @@ class LazyBuffer:
return self
# two ops in a row is one op. merge them if unresolved
if self.realized is None and self.op.op == op:
if self.realized is None and self.op.op == op and op != MovementOps.STRIDED:
# TODO: why is deleting self from children needed? shouldn't GC do it?
self.op.src[0].children.discard(self)
if op in [MovementOps.RESHAPE, MovementOps.EXPAND, MovementOps.SHRINK]:
return self.op.src[0].movement_op(op, arg)
if op == MovementOps.PERMUTE:
return self.op.src[0].movement_op(op, tuple(self.op.arg[i] for i in arg))
if op == MovementOps.PAD:
return self.op.src[0].movement_op(op, tuple((b1+b2, e1+e2) for (b1,e1),(b2,e2) in zip(self.op.arg, arg)))
# TODO: MovementOps.FLIP / MovementOps.STRIDED?
if op == MovementOps.FLIP:
return self.op.src[0].movement_op(op, tuple(i for i in arg+self.op.arg if not (i in arg and i in self.op.arg)))
# push permutes before reduce ops
if op == MovementOps.PERMUTE and PUSH_PERMUTES and self.realized is None and self.optype == ReduceOps:
# reduceops have one buffer input, permute it
narg = tuple(self.op.arg[arg[i]] for i in range(len(arg)))
src, rop = self.op.src[0], self.op.op
src.children = [y for y in src.children if self != y]
src.children.discard(self)
del self # TODO: why doesn't this delete remove it from the children
return src.movement_op(op, arg).reduce_op(rop, narg)