The output does not match the original

[jespb]

[STARTING NEW APP]
deriche
Baseline measure: 0.102523
skipped tr: [99, TrEnum.FULL_SPLIT]
skipped tr: [77, TrEnum.FULL_SPLIT]
skipped tr: [51, TrEnum.FULL_SPLIT]
skipped tr: [29, TrEnum.FULL_SPLIT]
FULL_SPLIT list validity: [True, True, True, True]
skipped tr: [111, TrEnum.TILE2D, 32, 32]
skipped tr: [85, TrEnum.TILE2D, 32, 32]
skipped tr: [56, TrEnum.TILE2D, 32, 32]
skipped tr: [30, TrEnum.TILE2D, 32, 32]
TILE 2D and 3D list validity: [True, True, True, True, True, True]
transformation_list=[
[88, TrEnum.FULL_SPLIT],
[68, TrEnum.FULL_SPLIT],
[40, TrEnum.FULL_SPLIT],
[20, TrEnum.FULL_SPLIT],
[125, TrEnum.TILE2D, 32, 32],
[70, TrEnum.TILE2D, 32, 32],
]
Tiling with size 32 ...
Is this transformation list valid: [True, True, True, True, True, True]
Transformed app: 0.110173
Thats a 0.93x speedup!
The output does not match the original
[FINISHED APP]

[jespb]

[STARTING NEW APP]
doitgen
Baseline measure: 0.187654
FULL_SPLIT list validity: [True, True]
TILE 2D and 3D list validity: [True, True, True, True]
transformation_list=[
[6, TrEnum.FULL_SPLIT],
[3, TrEnum.FULL_SPLIT],
[14, TrEnum.TILE2D, 32, 32],
[10, TrEnum.TILE2D, 32, 32],
]
Tiling with size 32 ...
Is this transformation list valid: [True, True, True, True]
Transformed app: 0.126670
Thats a 1.48x speedup!
The output does not match the original
[FINISHED APP]

[jespb]

[STARTING NEW APP]
symm
Baseline measure: 0.768797
FULL_SPLIT list validity: [True, True]
TILE 2D and 3D list validity: [True, True]
transformation_list=[
[8, TrEnum.FULL_SPLIT],
[3, TrEnum.FULL_SPLIT],
]
Tiling with size 32 ...
Is this transformation list valid: [True, True]
Transformed app: 0.887037
Thats a 0.87x speedup!
The output does not match the original
[FINISHED APP]

[jespb]

[STARTING NEW APP]
ludcmp
Baseline measure: 2.165674
skipped tr: [37, TrEnum.FULL_SPLIT]
skipped tr: [27, TrEnum.FULL_SPLIT]
skipped tr: [7, TrEnum.FULL_SPLIT]
skipped tr: [4, TrEnum.FULL_SPLIT]
FULL_SPLIT list validity: [True]
skipped tr: [21, TrEnum.TILE2D, 32, 32]
TILE 2D and 3D list validity: [True]
transformation_list=[
[17, TrEnum.FULL_SPLIT],
]
Tiling with size 32 ...
Is this transformation list valid: [True]
Transformed app: 2.200375
Thats a 0.98x speedup!
The output does not match the original
[FINISHED APP]

[vatai]

Simple example:

#!/usr/bin/env python
import difflib

import tadashi
from tadashi import TrEnum
from tadashi.apps import Polybench, Simple

# app = Polybench("deriche", compiler_options=["-DMINI_DATASET"])
app = Simple("try.c")

scop = app.scops[0]
node = scop.schedule_tree[1]
# trs = [
#     [88, TrEnum.FULL_SPLIT],
#     # [68, TrEnum.FULL_SPLIT],
#     # [40, TrEnum.FULL_SPLIT],
#     # [20, TrEnum.FULL_SPLIT],
#     # [125, TrEnum.TILE2D, 32, 32],
#     # [70, TrEnum.TILE2D, 32, 32],
# ]
print(node.yaml_str)
trs = [[2, TrEnum.FULL_SPLIT]]

valid = scop.transform_list(trs)
print(f"{valid=}")
# arrays = app.dump_arrays_and_time()["arrays"]

tapp = app.generate_code(ensure_legality=False)
print(tapp.source.read_text())
# tarrays = tapp.dump_arrays_and_time()["arrays"]

# if arrays != tarrays:
#     diff = difflib.unified_diff(arrays.splitlines(), tarrays.splitlines())
#     diff = "\n".join(diff)
#     print(diff)
#     print(app.dump_scop())
#     print("########")
#     print(tapp.dump_scop())


# node = scops.schedule_tree[18]
# print(node.yaml_str)

# tapp = app.generate_code()
# print(tapp.source.read_text())

print("DONE")

---

void
f(int _PB_H, int _PB_W, float imgOut[_PB_W][_PB_H], float y1[_PB_W][_PB_H]) {
  int j, i;
  float tm1;
#pragma scop
  for (j = 0; j < _PB_H; j++) {
    tm1 = 0.0;
    for (i = 0; i < _PB_W; i++) {
      y1[i][j] = imgOut[i][j] + tm1;
      tm1 = imgOut[i][j];
    }
  }
#pragma endscop
}