Add bug for TODO in svdf about unrolling loop optimization.

PiperOrigin-RevId: 344377656
Change-Id: I2a06439b6bd9aa663ae5aa09766bffc7bed03b26
This commit is contained in:
Jaehong Kim 2020-11-26 00:29:28 -08:00 committed by TensorFlower Gardener
parent 62c7ae1301
commit 175187b78f

View File

@ -251,8 +251,9 @@ inline void EvalHybridSVDF(
state[i * memory_size + memory_size - 1] = scratch[i];
}
// TODO(alanchiao): can optimize hybrid case ~5% by unrolling loop in applying
// time weights so that the inner loop multiplies eight elements at a time.
// TODO(b/174275776): can optimize hybrid case ~5% by unrolling loop in
// applying time weights so that the inner loop multiplies eight elements at
// a time.
ApplyTimeWeightsBiasAndActivation(
batch_size, memory_size, num_filters, num_units, rank, weights_time_data,
bias_data, params->activation, state, scratch, output_data);