onnxruntime
Expand Qgemm UDOT kernel to 8x8 block
#8562
Merged
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Overview
Commits
19
Changes
View On
GitHub
Expand Qgemm UDOT kernel to 8x8 block
#8562
chenfucn
merged 19 commits into
master
from
cfu_dot8x8
chenfucn
requested a review
from
zhanghuanrong
5 years ago
chenfucn
requested a review
from
yufenglee
5 years ago
chenfucn
requested a review
from
tracysh
5 years ago
chenfucn
requested a review
5 years ago
chenfucn
force pushed
from
6ce8bff1
to
a10ec787
5 years ago
chenfucn
force pushed
from
a10ec787
to
bcb3868d
5 years ago
chenfucn
force pushed
from
bcb3868d
to
528fff11
5 years ago
chenfucn
force pushed
from
528fff11
to
f46cec38
5 years ago
chenfucn
force pushed
from
f46cec38
to
d8cae5ec
5 years ago
comments
0a588730
use v2 v3 for B unrolled in k dimension
c5fe1bb0
move comments before loop prefix
0cbeeb36
use same accumulators for k-dimension unroll
98d36a04
rearrange registers- prepare to process more column
c8d4523d
load A data with 128b load
7b84a138
process 8 rows
1f4ba864
dot prepack A 8 rows
cd774ffd
debug aid
2d086230
add linux/android version
e0e1d27d
optimize m4 loop prefix
9bc9ee99
revert m4 loop behavior
e291a697
optimize m1
287365e3
optimize m2 loop
6cd32657
Revert "debug aid"
6e9b82f1
copy m1 m2 optimization to aarch64
f189b722
reduce number of saved registers
0ab64607
pipelining dup instructions into the outter m8 loop
f6fd5cbd
chenfucn
force pushed
from
d8cae5ec
to
f6fd5cbd
5 years ago
tracysh
commented on 2021-08-09
adjust spaces
540efa82
chenfucn
force pushed
from
4ffa6445
to
540efa82
5 years ago
yufenglee
commented on 2021-08-17
yufenglee
commented on 2021-08-17
yufenglee
commented on 2021-08-17
yufenglee
approved these changes on 2021-08-17
chenfucn
merged
22438044
into master
5 years ago
chenfucn
deleted the cfu_dot8x8 branch
5 years ago
Login to write a write a comment.
Login via GitHub
Reviewers
yufenglee
tracysh
zhanghuanrong
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub