commit | 9be683016e9003f474517a2060487628381fe1f5 | [log] [tgz] |
---|---|---|
author | Przemyslaw Szczepaniak <pszczepaniak@google.com> | Fri May 03 13:21:28 2019 +0100 |
committer | Przemyslaw Szczepaniak <pszczepaniak@google.com> | Tue May 07 10:45:29 2019 +0000 |
tree | b24ce615fe131baecf896e55f0dff2e4bfec44fc | |
parent | 7ddc56417fbf2e1afc1b7c219dabd1d781dddfb9 [diff] |
Make avg_pool_float_2_relaxed possible to pass with fp16 accumulator. avg_pool_float_2_relaxed is computing an average pooling with window size ~50*50. All input elements are 1, and each output is expected to be ~2500/2500 = 1.0f. However, for FP16, 2048+1 = 2048, so the actual output would be 2048/2500. Bug: 125494355 Test: avg_pool_float_2_relaxed with modifed reference AveragePool to emulate fp16 accumulator, Change-Id: I90c63fce3548e7a074c738a7fd8381ab14ab0781