Skip to content

[Arm] parallel batch axis #3931

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
merged 5 commits into from
Sep 11, 2019
Merged

[Arm] parallel batch axis #3931

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
e4622b0
support LLVM trunk
yzhliu Sep 6, 2019
c52fa06
guard with USE_LLVM in if condition for c++14
yzhliu Sep 6, 2019
3303d6c
GREATER_EQUAL -> GREATER
yzhliu Sep 6, 2019
b2073d2
[Arm] parallel batch axis
yzhliu Sep 10, 2019
6c71b18
Merge remote-tracking branch 'upstream/master' into armfp16
yzhliu Sep 10, 2019
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
15 changes: 9 additions & 6 deletions topi/python/topi/arm_cpu/conv2d.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -280,13 +280,15 @@ def _schedule_spatial_pack(cfg, s, data_vec, kernel_vec,
s[conv].compute_at(s[last], ow)

# mark parallel
s[last].parallel(co)
p = s[last].fuse(n, co)
s[last].parallel(p)

if data_vec.op.name == 'data_vec_undilated':
_, h, _, _, _, _, _, _ = s[data_vec].op.axis
n, h, _, _, _, _, _, _ = s[data_vec].op.axis
else:
_, h, _, _, _, _ = s[data_vec].op.axis
s[data_vec].parallel(h)
n, h, _, _, _, _ = s[data_vec].op.axis
p = s[data_vec].fuse(n, h)
s[data_vec].parallel(p)

if kernel_vec.op.name == 'kernel_vec':
co, _, _, _, _ = s[kernel_vec].op.axis
Expand Down Expand Up @@ -470,8 +472,9 @@ def _schedule_winograd(cfg, s, output, last):
# output
n, co, h, w = s[last].op.axis
co, coi = cfg['tile_k'].apply(s, last, co)
s[M].compute_at(s[last], co)
s[last].parallel(co)
p = s[last].fuse(n, co)
s[M].compute_at(s[last], p)
s[last].parallel(p)

MM = s.cache_read(M, 'global', [Y])
m = get_const_int(V.shape[0]) + 1 - 3
Expand Down