Skip to content

[TOPI,x86] Improve performance on int8 conv2d on x86 #9966

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
masahi merged 1 commit into from
Jan 19, 2022
Merged

[TOPI,x86] Improve performance on int8 conv2d on x86 #9966

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
10 changes: 5 additions & 5 deletions python/tvm/topi/generic/conv2d.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -228,17 +228,17 @@ def schedule_conv_NCHWc_cpu_common_int8(
batch, oc_chunk, oh, ow, oc_block = s[O].op.axis
ow_chunk, ow_block = s[O].split(ow, factor=reg_n)
s[O].reorder(oc_chunk, oh, ow_chunk, ow_block, oc_block)
s[C].compute_at(s[O], ow_block)
parallel_axis = s[O].fuse(batch, oc_chunk, oh)
s[C].compute_at(s[O], parallel_axis)
s[O].vectorize(oc_block)
s[O].parallel(parallel_axis)
elif out_ndim == 4:
batch, oc, oh, ow = s[O].op.axis
ow_chunk, ow_block = s[O].split(ow, factor=reg_n)
oc_chunk, oc_block = s[O].split(oc, factor=oc_bn)
s[O].reorder(oc_chunk, oh, ow_chunk, ow_block, oc_block)
s[C].compute_at(s[O], ow_block)
parallel_axis = s[O].fuse(batch, oc_chunk, oh)
s[C].compute_at(s[O], parallel_axis)
s[O].vectorize(oc_block)
s[O].parallel(parallel_axis)
else:
Expand Down Expand Up @@ -301,8 +301,8 @@ def schedule_conv_NCHWc_cpu_1x1_int8(
s[C].reorder(oc_chunk, oh_outer, ow_outer, oh_inner, ow_inner, oc_block)
s[C].vectorize(oc_block)

s[CC].compute_at(s[C], ow_inner)
parallel_axis = s[C].fuse(batch, oc_chunk, oh_outer)
s[CC].compute_at(s[C], parallel_axis)
if C == O:
s[C].parallel(parallel_axis)

Expand Down Expand Up @@ -346,8 +346,8 @@ def schedule_conv_NCHWc_cpu_1x1_int8(
ow_outer, ow_inner = s[O].split(ow, factor=ow_factor)
s[O].reorder(oc_chunk, oh_outer, ow_outer, oh_inner, ow_inner, oc_block)

s[C].compute_at(s[O], ow_inner)
parallel_axis = s[O].fuse(batch, oc_chunk, oh_outer)
s[C].compute_at(s[O], parallel_axis)
s[O].vectorize(oc_block)
s[O].parallel(parallel_axis)
elif out_ndim == 4:
Expand All @@ -357,8 +357,8 @@ def schedule_conv_NCHWc_cpu_1x1_int8(
ow_outer, ow_inner = s[O].split(ow, factor=ow_factor)
s[O].reorder(oc_chunk, oh_outer, ow_outer, oh_inner, ow_inner, oc_block)

s[C].compute_at(s[O], ow_inner)
parallel_axis = s[O].fuse(batch, oc_chunk, oh_outer)
s[C].compute_at(s[O], parallel_axis)
s[O].vectorize(oc_block)
s[O].parallel(parallel_axis)
else:
Expand Down