Skip to content

Refactor gemm bf16 tuner #1275

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

Merged
valarLip merged 7 commits into from
Oct 29, 2025
Merged

Refactor gemm bf16 tuner #1275

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
6489094
refactor GemmTuner
yzhou103 Oct 27, 2025
ac1b9ed
update
yzhou103 Oct 28, 2025
9c48c5b
fix lint error
yzhou103 Oct 28, 2025
a04b0cf
fix lint error
yzhou103 Oct 28, 2025
bff1464
Merge branch 'main' into refactor_gemm_bf16_tuner
yzhou103 Oct 28, 2025
7f3882b
Merge branch 'main' into refactor_gemm_bf16_tuner
yzhou103 Oct 29, 2025
1b24303
Merge branch 'main' into refactor_gemm_bf16_tuner
yzhou103 Oct 29, 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
2 changes: 1 addition & 1 deletion aiter/configs/tuned_gemm.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
M,N,K,bias,dtype,outdtype,scaleAB,cu_num,libtype,solidx,splitK,soltimes,kernelName,err_ratio,tflops,bw
M,N,K,bias,dtype,outdtype,scaleAB,cu_num,libtype,solidx,splitK,us,kernelName,err_ratio,tflops,bw
64,256,5120,False,torch.bfloat16,torch.float32,False,80,asm,1,10,8.3,_ZN5aiter28bf16gemm_outf32_tn_32x64_pf3E,0.0,20.21,402.69
80,256,5120,False,torch.bfloat16,torch.float32,False,80,asm,2,10,8.8,_ZN5aiter28bf16gemm_outf32_tn_48x64_pf3E,0.0,23.83,400.29
128,256,5120,False,torch.bfloat16,torch.float32,False,80,asm,3,10,11.2,_ZN5aiter28bf16gemm_outf32_tn_64x64_pf3E,0.0,29.96,362.79
Expand Down
7 changes: 5 additions & 2 deletions aiter/utility/base_tuner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@
import torch
import pandas as pd

from abc import ABC, abstractmethod
from abc import abstractmethod
from aiter import logger
import traceback
from operator import itemgetter
Expand Down Expand Up @@ -37,6 +37,7 @@ class TunerCommon:
dtypes.fp4x2: 1,
torch.uint8: 1,
torch.uint32: 4,
dtypes.fp32: 4,
torch.int4: 1 / 2,
torch.float8_e4m3fnuz: 1,
torch.float8_e4m3fn: 1,
Expand Down Expand Up @@ -76,13 +77,15 @@ def _setup_common_arguments(self):
"-i",
"--untune_file",
default=defaults["untune_file"],
dest="untune_file",
required=False,
help="input",
)
self.parser.add_argument(
"-o",
"--tune_file",
default=defaults["tune_file"],
dest="tune_file",
required=False,
help="output: tuning result store this file",
)
Expand Down Expand Up @@ -494,7 +497,7 @@ def result_to_csv(self, resultdf, file, concat=False):
[self.success, resultdf[resultdf["us"] != self.INVALID_TIME]],
ignore_index=True,
)
update_tunedf = self.success
update_tunedf = resultdf[resultdf["us"] != self.INVALID_TIME] # self.success
if not concat:
resultdf = self.update_tunedf(old_df, update_tunedf)
else:
Expand Down
Loading