Skip to content
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

Jump to bottom

Small fix to mitigate overflow in Convolution.cc #2992

Merged
merged 4 commits into from
Jan 16, 2019
Merged

Small fix to mitigate overflow in Convolution.cc #2992

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
36c2f52
minor change to handle cases where <s> is appears in text
Oct 1, 2018
38e584a
Merge remote-tracking branch 'upstream/master'
Nov 8, 2018
a5f2e35
Merge remote-tracking branch 'upstream/master'
Jan 16, 2019
a3464d0
mitigating some overflow error with floats
Jan 16, 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
4 changes: 2 additions & 2 deletions src/nnet3/convolution.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -976,7 +976,7 @@ static void ComputeTempMatrixSize(const ConvolutionComputationOptions &opts,
// work out how many rows the temporary matrix should have, taking
// into account the specified memory limit.
temp_rows = computation->num_t_out * computation->num_images;
BaseFloat num_megabytes = (4 * temp_rows * temp_cols) / 1000000.0,
BaseFloat num_megabytes = (4 * (temp_rows / 1000.0) * (temp_cols / 1000.0)),
megabyte_limit = opts.max_memory_mb;
// C++ rounds down; here, we want to round up so we add one.
int32 ratio = 1.0 + num_megabytes / megabyte_limit;
Expand All @@ -986,7 +986,7 @@ static void ComputeTempMatrixSize(const ConvolutionComputationOptions &opts,
// >= temp_rows so that we don't have a small leftover piece.
int32 new_num_t_out = (computation->num_t_out + ratio - 1) / ratio;
temp_rows = new_num_t_out * computation->num_images;
BaseFloat new_num_megabytes = (4 * temp_rows * temp_cols) / 1000000.0;
BaseFloat new_num_megabytes = (4 * (temp_rows / 1000.0) * (temp_cols / 1000.0));
// make sure we're within the memory limit.
if (new_num_megabytes > 1.01 * megabyte_limit) {
KALDI_WARN << "Memory consumed in convolution is more than requested "
Expand Down