mirror of
https://github.com/tinygrad/tinygrad.git
synced 2026-01-23 05:48:08 -05:00
f5467cfedc
* runs one metal kernel * conv2d works * ops tests are passing * const folding * all ops work * pre commit always passes * torch works * working still * fix graph test * tests passing * image almost works * image conv works * most images * fix custom * fix assignment * fix compile enet * clean up comments * fix realize return value * include shapetracker in LB repr * copy should make a copy * reenable method cache * fix lna * dtypes in graph * forward only for IMAGE=2 * simple realize * getting close * fixup new api, it's good except the kernel count * back to 197 kernels * tests should pass * go to a real float * no type_on_cpu * fix the docs * put shapetracker back in it's proper place
52 lines
2.1 KiB
Python
52 lines
2.1 KiB
Python
# TODO: move the GRAPH and DEBUG stuff to here
|
|
import gc
|
|
from tinygrad.helpers import prod
|
|
from tinygrad.tensor import Tensor
|
|
from tinygrad.lazy import LazyBuffer
|
|
from tinygrad.runtime.ops_gpu import CLBuffer
|
|
from tinygrad.ops import GlobalCounters
|
|
|
|
def print_objects():
|
|
#gc.collect()
|
|
tensors = [x for x in gc.get_objects() if isinstance(x, Tensor)]
|
|
tensor_ram_used = sum([prod(x.shape)*4 for x in tensors])
|
|
lazybuffers = [x for x in gc.get_objects() if isinstance(x, LazyBuffer)]
|
|
gpubuffers = [x for x in gc.get_objects() if isinstance(x, CLBuffer)]
|
|
realized_buffers = [x.realized for x in lazybuffers if x.realized]
|
|
gpubuffers_orphaned = [x for x in gpubuffers if x not in realized_buffers]
|
|
|
|
print(f"{len(tensors)} tensors allocated in {tensor_ram_used/1e9:.2f} GB, GPU using {GlobalCounters.mem_used/1e9:.2f} GB")
|
|
print(f"{len(lazybuffers)} lazybuffers {len(realized_buffers)} realized, {len(gpubuffers)} GPU buffers")
|
|
print(f"{len(gpubuffers_orphaned)} GPU buffers are orphaned")
|
|
|
|
cnt = 0
|
|
for tb in gpubuffers_orphaned:
|
|
bb = gc.get_referrers(tb)
|
|
for b in bb:
|
|
if b is not gpubuffers and b is not gpubuffers_orphaned:
|
|
print(tb, "\nreference", type(b), len(b), str(b)[0:150])
|
|
for x in gc.get_referrers(b):
|
|
print("double reference", str(x)[0:100])
|
|
print("\n")
|
|
if cnt == 10:
|
|
break
|
|
cnt += 1
|
|
|
|
for x in gpubuffers_orphaned:
|
|
if getattr(x, '_buf', None): del x._buf
|
|
if getattr(x, '_image', None): del x._image
|
|
|
|
return len(gpubuffers_orphaned)
|
|
|
|
"""
|
|
import gc
|
|
|
|
def print_ram():
|
|
print(GlobalCounters.mem_used/1e9, sum([prod(x.shape)*4 for x in gc.get_objects() if isinstance(x, Tensor)])/1e9)
|
|
img_count = sum([x.is_image() for x in gc.get_objects() if isinstance(x, OpenCLBuffer)])
|
|
print("img_count", img_count)
|
|
buffer_bytes = sum([x.cl.size for x in gc.get_objects() if isinstance(x, CLBuffer)])
|
|
image_bytes = sum([x.cl.row_pitch*x.cl.height for x in gc.get_objects() if isinstance(x, CLImage)])
|
|
print("buffer bytes", buffer_bytes/1e9, "image bytes", image_bytes/1e9, "sum", (buffer_bytes+image_bytes)/1e9)
|
|
"""
|