Skip to content

Shapetracker Update #32

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Bchass wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Shapetracker Update #32

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
cc03ebd
small changes
Bchass Nov 12, 2025
c70c918
Explantion for removal
Bchass Nov 13, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 10 additions & 9 deletions 20241129_viz.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,13 +20,14 @@ will allow us to see the generated code, and NOOPT tells tinygrad not to enable
see the generated metal/cuda code:

```c++
kernel void r_4(device float* data0, device float* data1, uint3 gid [[threadgroup_position_in_grid]], uint3 lid [[thread_position_in_threadgroup]]) {
float acc0 = 0.0f;
for (int ridx0 = 0; ridx0 < 4; ridx0++) {
float val0 = *(data1+ridx0);
acc0 = (acc0+val0);
kernel void r_4(device float* data0_1, device float* data1_4, uint3 gid [[threadgroup_position_in_grid]], uint3 lid [[thread_position_in_threadgroup]]) {
float acc0[1];
*(acc0+0) = 0.0f;
for (int ridx1000 = 0; ridx1000 < 4; ridx1000++) {
float val0 = (*(data1_4+ridx1000));
*(acc0+0) = ((*(acc0+0))+val0);
}
*(data0+0) = acc0;
*(data0_1+0) = (*(acc0+0));
}
```

Expand All @@ -35,9 +36,9 @@ of vectorized data type, and remove the loop. In fact, that's what the optimizat
on this time `DEBUG=5 python script.py`:

```c++
kernel void r_4(device float* data0, device float* data1, uint3 gid [[threadgroup_position_in_grid]], uint3 lid [[thread_position_in_threadgroup]]) {
float4 val0 = *((device float4*)((data1+0)));
*(data0+0) = (val0.w+val0.z+val0.x+val0.y);
kernel void r_4(device float* data0_1, device float* data1_4, uint3 gid [[threadgroup_position_in_grid]], uint3 lid [[thread_position_in_threadgroup]]) {
float4 val0 = (*((device float4*)((data1_4+0))));
*(data0_1+0) = (val0.x+val0.y+val0.z+val0.w);
}
```

Expand Down
5 changes: 5 additions & 0 deletions 20241217_st.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,10 @@
# Shapetracker

> [!NOTE]
> Shapetracker since has been removed from Tinygrad via commit 1d1e1d9d88b93db07be2244cc087f6786a5705fc. This document will stay in place for archival purposes. A replacement has been introduced - RAGNEIFY.
>
> Reasoning for the removal: https://x.com/__tinygrad__/status/1964037572503752910

> examples were tested on commit a2a4ff30dcfafc8e7763303e9d8f0955900e8617 (in case things go out of date)

Suppose you have a 2 by 2 matrix containing values 1, 2, 3, 4. These four numbers are not stored in memory as a matrix,
Expand Down