Skip to content

Avoid amax roll for non-run modules #825

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
ksivaman merged 1 commit into from
Apr 30, 2024
Merged

Avoid amax roll for non-run modules #825

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
22 changes: 12 additions & 10 deletions transformer_engine/common/recipe/delayed_scaling.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -197,16 +197,18 @@ kernel_bulk(
const auto last_amax = ((amax_reduction_buffer != nullptr)
&& (amax_reduction_buffer[offset_in_buffer+count] != 0.0f)) ?
amax_reduction_buffer[offset_in_buffer+count] : amax_history[0];
for (size_t off = 0; off < length; off += bsize) {
const size_t i = off + tid;
float a = 0;
if (i < length) {
a = (i < length - 1) ? amax_history[(i+1)*stride] : last_amax;
amax = fmaxf(amax, a);
}
__syncthreads(); // Inplace roll
if (i < length) {
amax_history[i*stride] = (i > 0) ? a : 0;
if (last_amax != 0.0f) {
for (size_t off = 0; off < length; off += bsize) {
const size_t i = off + tid;
float a = 0;
if (i < length) {
a = (i < length - 1) ? amax_history[(i+1)*stride] : last_amax;
amax = fmaxf(amax, a);
}
__syncthreads(); // Inplace roll
if (i < length) {
amax_history[i*stride] = (i > 0) ? a : 0;
}
}
}

Expand Down