"""Device-profiler zones of the addln kernel (ADDLN_EXP=prof, TT_METAL_DEVICE_PROFILER=1)."""
import sys, os, glob, csv
from pathlib import Path
from collections import defaultdict
from common import *
sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
from hamer import tt_kernels as tk
dev = open_dev()
res = ttnn.create_sharded_memory_config((192, 1280), core_grid=ttnn.CoreGrid(y=6, x=10), strategy=ttnn.ShardStrategy.BLOCK,
                                        orientation=ttnn.ShardOrientation.ROW_MAJOR)
x = ttnn.from_torch(torch.randn(1, 1, 192, 1280), dtype=ttnn.bfloat16, layout=ttnn.TILE_LAYOUT, device=dev, memory_config=res)
y = ttnn.from_torch(torch.randn(1, 1, 192, 1280), dtype=ttnn.bfloat16, layout=ttnn.TILE_LAYOUT, device=dev, memory_config=res)
C = tk.addln_constants(torch.ones(1280), torch.zeros(1280), dev)
for _ in range(5):
    s, h = tk.add_layernorm(x, y, C, 1e-6); ttnn.synchronize_device(dev); ttnn.deallocate(s); ttnn.deallocate(h)
ttnn.ReadDeviceProfiler(dev)
ttnn.close_device(dev)
f = os.environ["TT_METAL_PROFILER_DIR"] + "/.logs/profile_log_device.csv"
rows = list(csv.reader(open(f)))[2:]
# columns: PCIe slot, core_x, core_y, RISC processor type, timer_id, time[cycles], data, run host ID, zone name, type, ...
st = {}; dur = defaultdict(list)
for r in rows:
    try:
        cx, cy, risc, cyc, name, typ = int(r[1]), int(r[2]), r[3].strip(), int(r[5]), r[10].strip(), r[11].strip()
    except Exception:
        continue
    key = (cx, cy, risc, name)
    if "START" in typ: st[key] = cyc
    elif "END" in typ and key in st: dur[(risc, name)].append((cyc - st.pop(key)) / 1.35)
for k in sorted(dur):
    v = sorted(dur[k]); print(f"{k[0]:8s} {k[1]:28s} n={len(v):4d} med={v[len(v)//2]:8.0f} ns max={v[-1]:8.0f}")