github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-09 15:08:02 -05:00
Code Issues Packages Projects Releases Wiki Activity

remove ScheduleItem and merge it with ExecItem (#13759)

Browse Source 
* remove ExecItem and merge it with ScheduleItem

* less diff

* fix issues

* min diff

* don't change bufs in _lower

* min diff

* update

* revert

* fixes

* diff
This commit is contained in:
George Hotz
2025-12-19 17:04:24 -04:00
committed by GitHub
parent df6cde8a00
commit 744af193f0
35 changed files with 172 additions and 395 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
docs/abstractions3.py
View File

@@ -38,25 +38,19 @@ optim.schedule_step() # this will step the optimizer without running realize
# The weight Tensors have been assigned to, but not yet realized. Everything is still lazy at this point
# l1.uop and l2.uop define a computation graph
from tinygrad.engine.schedule import ScheduleItem
schedule: List[ScheduleItem] = Tensor.schedule(l1, l2)
from tinygrad.engine.schedule import ExecItem
schedule: List[ExecItem] = Tensor.schedule(l1, l2)
print(f"The schedule contains {len(schedule)} items.")
for si in schedule: print(str(si)[:80])
# *****
# 4. Lower a schedule.
# 4. Lower and run the schedule.
from tinygrad.engine.realize import lower_schedule_item, ExecItem
lowered: List[ExecItem] = [lower_schedule_item(si) for si in tqdm(schedule)]
for si in tqdm(schedule): si.run()
# *****
# 5. Run the schedule
for ei in tqdm(lowered): ei.run()
# *****
# 6. Print the weight change
# 5. Print the weight change
print("first weight change\n", l1.numpy()-l1n)
print("second weight change\n", l2.numpy()-l2n)

8
docs/developer/developer.md
View File

@@ -17,15 +17,15 @@ The `UOp` graph specifies the compute in terms of low level tinygrad ops. Not al
## Scheduling
The [scheduler](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/engine/schedule.py) converts the graph of UOps into a list of `ScheduleItem`. One `ScheduleItem` is one kernel on the GPU, and the scheduler is responsible for breaking the large compute graph into subgraphs that can fit in a kernel. `ast` specifies what compute to run, and `bufs` specifies what buffers to run it on.
The [scheduler](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/engine/schedule.py) converts the graph of UOps into a list of `ExecItem`. One `ExecItem` is one kernel on the GPU, and the scheduler is responsible for breaking the large compute graph into subgraphs that can fit in a kernel. `ast` specifies what compute to run, and `bufs` specifies what buffers to run it on.
::: tinygrad.engine.schedule.ScheduleItem
::: tinygrad.engine.schedule.ExecItem
## Lowering
The code in [realize](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/engine/realize.py) lowers `ScheduleItem` to `ExecItem` with
The code in [realize](https://github.com/tinygrad/tinygrad/tree/master/tinygrad/engine/realize.py) lowers `ExecItem` by populating its `prg` field with
::: tinygrad.engine.realize.lower_schedule
::: tinygrad.engine.realize.run_schedule
There's a ton of complexity hidden behind this, see the `codegen/` directory.