commit 837dc1bc42323fa723b72fe51919bc2f013e5a26
author Michael McGeagh <michael.mcgeagh@arm.com> Tue Nov 10 12:38:25 2020 +0000
committer Michael McGeagh <michael.mcgeagh@arm.com> Tue Nov 17 14:40:26 2020 +0000
tree e38f2e5ff31a13404c252e62dd13ef877a7e6a49
parent 69b3176127ff8522903e087d56e2d2f4ec557d62

MLBEDSW-3403 Generate supported op report

A new CLI has been added that allows the generation of a report
containing a summary table of all TFLite ops that can be placed on the
NPU, and what the constraints are for that operator to be successfully
scheduled on the NPU.
This option will generate a new file, SUPPORTED_OPS.md containing this
information, in the current working directory.

Signed-off-by: Michael McGeagh <michael.mcgeagh@arm.com>
Change-Id: I6a7e2a49f251b76b2ea1168fff78e00da1910b25

ethosu/vela/supported_operators.py

diff --git a/ethosu/vela/supported_operators.py b/ethosu/vela/supported_operators.py
index 46f7a5d..ccf6104 100644
--- a/ethosu/vela/supported_operators.py
+++ b/ethosu/vela/supported_operators.py
@@ -25,6 +25,7 @@
 from .operation import get_slice_offsets
 from .operation import Op
 from .tensor import check_quantized_tens_scaling_equal
+from .tflite_mapping import BUILTIN_OPERATOR_UNKNOWN
 from .tflite_mapping import optype_to_builtintype
 
 
@@ -37,6 +38,15 @@
     return docstring
 
 
+def _optype_formatter(op_list):
+    # Convert internal op types to external names
+    output = map(optype_to_builtintype, op_list)
+    # Remove UNKNOWNs
+    output = (x for x in output if x is not BUILTIN_OPERATOR_UNKNOWN)
+    # Order alphabetically
+    return sorted(output)
+
+
 class SupportedOperators:
     # Categorised lists of supported operators
     npu_pre_ops = set((Op.SplitSliceRead,))
@@ -99,6 +109,10 @@
     filter_range = (1, 8)
     filter_height_range = (1, 256)
     filter_product_range = (1, 256 * 256)
+    # Ordered, external names of op types for the constraint reasons
+    docstring_shapeless_input_ops = _optype_formatter(shapeless_input_ops)
+    docstring_supported_int32_tensor_ops = _optype_formatter(supported_int32_tensor_ops)
+    docstring_supported_fused_activations = _optype_formatter(supported_fused_activations)
 
     def __init__(self):
         # Setup the generic constraints. Note: the order matters
@@ -279,7 +293,7 @@
         return valid, f"Output Tensor '{ofm.name}' is scalar"
 
     @classmethod
-    @docstring_format_args([shapeless_input_ops])
+    @docstring_format_args([docstring_shapeless_input_ops])
     def constraint_tens_input_scalar(cls, op):
         "Scalar Input tensors are only valid for op type: {}"
         valid = True
@@ -320,7 +334,7 @@
         return valid, ", ".join(extra)
 
     @classmethod
-    @docstring_format_args([supported_int32_tensor_ops])
+    @docstring_format_args([docstring_supported_int32_tensor_ops])
     def constraint_tens_int32_ops(cls, op):
         "Tensors which are int32 are only valid when op type is: {}"
         valid = True
@@ -377,7 +391,7 @@
         return valid, ", ".join(extra)
 
     @classmethod
-    @docstring_format_args([supported_fused_activations])
+    @docstring_format_args([docstring_supported_fused_activations])
     def constraint_faf(cls, op):
         "The fused activation function (if present) must be one of type: {}"
         if op.activation is None: