diff options
| author | Tim Hall <tim.hall@arm.com> | 2021-05-17 22:57:46 +0100 |
|---|---|---|
| committer | tim.hall <tim.hall@arm.com> | 2021-05-21 15:42:06 +0000 |
| commit | 64556f32ff7bfca6036a6598034464b13b64a4ef (patch) | |
| tree | edf438352185ce88beddbd7f8a2ac62f459aacaa /ethosu/vela/stats_writer.py | |
| parent | 0fe2209156719ff0c17fab604dee3fef7d8b8fd7 (diff) | |
| download | ethos-u-vela-64556f32ff7bfca6036a6598034464b13b64a4ef.tar.gz | |
MLBEDSW-4219: Add tensor allocation info to summary
- Moved new tensor allocation info under --verbose-allocation flag
- Tidied up and added histogram to --verbose--allocation print
Signed-off-by: Tim Hall <tim.hall@arm.com>
Change-Id: I76fb5187319aedf86f599f57b766220cafc17326
Diffstat (limited to 'ethosu/vela/stats_writer.py')
| -rw-r--r-- | ethosu/vela/stats_writer.py | 8 |
1 files changed, 0 insertions, 8 deletions
diff --git a/ethosu/vela/stats_writer.py b/ethosu/vela/stats_writer.py index 18b8092e..fbc47f80 100644 --- a/ethosu/vela/stats_writer.py +++ b/ethosu/vela/stats_writer.py @@ -227,7 +227,6 @@ def print_performance_metrics_for_strat( bandwidths, batch_size, memory_used, - min_mem_usage, num_passes, num_cascaded_passes, n_operations=0, @@ -270,11 +269,6 @@ def print_performance_metrics_for_strat( aug_label = label + " used" print(f"Total {aug_label:25} {memory_used[mem_area] / 1024.0:12.2f} KiB", file=f) - if mem_area == MemArea.Sram and min_mem_usage: - mem_used = memory_used[[mem_area for mem_area, label in mem_area_labels if "SRAM" in label][0]] / 1024.0 - fraction = (mem_used - min_mem_usage / 1024.0) / (min_mem_usage / 1024.0) - print(f"Theoretical minimum SRAM usage{min_mem_usage/1024.0:23.2F} KiB", file=f) - print(f"Allocator overhead{100*fraction:35.2F} %", file=f) print(file=f) print(f"{num_passes:d} passes fused into {num_cascaded_passes:d}", file=f) @@ -364,7 +358,6 @@ def print_performance_metrics(nng, arch, show_cpu_operations=False, verbose_weig n_cascaded_passes = sum(len(sg.cascaded_passes) for sg in nng.subgraphs) n_operations = sum(len(ps.ops) for sg in nng.subgraphs for ps in sg.passes) cpu_operations = sum((ps.ops for sg in nng.subgraphs for ps in sg.passes if ps.placement == PassPlacement.Cpu), []) - min_mem_usage = max(sg.min_mem_usage for sg in nng.subgraphs) weights_data = ( { "original": nng.total_original_weights, @@ -382,7 +375,6 @@ def print_performance_metrics(nng, arch, show_cpu_operations=False, verbose_weig nng.bandwidths, nng.batch_size, nng.memory_used, - min_mem_usage, n_passes, n_cascaded_passes, n_operations, |