diff options
| author | Jacob Bohlin <jacob.bohlin@arm.com> | 2020-09-11 10:04:15 +0200 |
|---|---|---|
| committer | patrik.gustavsson <patrik.gustavsson@arm.com> | 2020-09-17 08:18:50 +0000 |
| commit | 1a66697b80a527af6d6dd1ed235199264696767e (patch) | |
| tree | 447f19903eedb0ed163348769da28267ccf3bf47 /ethosu/vela/extract_npu_subgraphs.py | |
| parent | 1356c2ab034738bcf51822de18911cc499fa2e8e (diff) | |
| download | ethos-u-vela-1a66697b80a527af6d6dd1ed235199264696767e.tar.gz | |
MLBEDSW-2809: Redo the Tensor addressing
Added a static class TensorAddressMap that stores all Tensor addresses
based on their equivalence_id. Made the "address" field into a property
which getter and setter looks up/sets the tensor's address in
TensorAddressMap.
This makes the references to cpu_tensor/npu_tensor obsolete and they
have been removed.
Addition to scheduler: avoid SRAM spilling if an op has consumers in
other subgraphs.
Minor rework in LUTState; it will now assign a unique equivalence_id to
the SHRAM lut tensor to avoid issues with addressing. The equivalent
checks in LUTState now compares the values of the LUT instead of the the
equivalence_id.
Updated LUT unit tests accordingly.
Signed-off-by: Jacob Bohlin <jacob.bohlin@arm.com>
Change-Id: I41de5a8a4e5f07b77d6544d8d4034b754993e503
Diffstat (limited to 'ethosu/vela/extract_npu_subgraphs.py')
| -rw-r--r-- | ethosu/vela/extract_npu_subgraphs.py | 6 |
1 files changed, 0 insertions, 6 deletions
diff --git a/ethosu/vela/extract_npu_subgraphs.py b/ethosu/vela/extract_npu_subgraphs.py index 4adddc17..c0430b5d 100644 --- a/ethosu/vela/extract_npu_subgraphs.py +++ b/ethosu/vela/extract_npu_subgraphs.py @@ -70,10 +70,7 @@ def rewrite_tensor_cpu_producer_npu_consumers( orig_tens, call_ps, startup_init_ps, npu_subgraph, cpu_subgraph, subgraph_for_pass ): is_const = orig_tens.ops[0].type == "Const" - new_tens = orig_tens.clone("_npu") - orig_tens.npu_tensor = new_tens - new_tens.cpu_tensor = orig_tens op_type = "SubgraphInput" if is_const: @@ -107,9 +104,6 @@ def rewrite_tensor_npu_producer_cpu_consumers( ): new_tens = orig_tens.clone("_cpu") - new_tens.npu_tensor = orig_tens - orig_tens.cpu_tensor = new_tens - npu_subgraph.output_tensors.append(orig_tens) call_ps.outputs.append(new_tens) |