[res/tfl_recipes] Add REGRESS_Issue_13863 (#14064)

This commit adds REGRESS_Issue_13863 tflite recipe.

ONE-DCO-1.0-Signed-off-by: HanJin Choi [email protected]

compiler/common-artifacts/exclude.lst

@@ -125,6 +125,7 @@ tcgenerate(ReLU6_dynamic_000) # TestDataGenerator does not support unknown dimen
 tcgenerate(ReLUN1To1_000)
 tcgenerate(ReLUN1To1_dynamic_000) # TestDataGenerator does not support unknown dimension
 tcgenerate(Reshape_003) # luci-interpreter doesn't support reshape without built-in option
+tcgenerate(Reshape_004) # has 0 in shape
 tcgenerate(ReverseSequence_000)
 tcgenerate(ReverseV2_000)
 tcgenerate(Round_000)

compiler/luci/service/src/Nodes/CircleReshape.cpp

@@ -84,15 +84,34 @@ loco::TensorShape Algorithm::visit(const luci::CircleReshape *node)
     {
       LUCI_ASSERT(const_shape_node->dtype() == S32, "Only support int32 CircleConst");
 
+      // NOTE for rank(shape_by_input before) < rank(shape_by_input after),
+      //      shape_by_input after will be filled with unknown dims
       shape_by_input.rank(const_shape_node->size<S32>());
 
       for (uint32_t axis = 0; axis < shape_by_input.rank(); ++axis)
       {
-        shape_by_input.dim(axis) = const_shape_node->at<S32>(axis);
         if (const_shape_node->at<S32>(axis) < 0)
         {
           shape_by_input.dim(axis).unset();
         }
+        else if (const_shape_node->at<S32>(axis) == 0)
+        {
+          const auto node_tensor = loco::must_cast<luci::CircleNode *>(node->tensor());
+          // set dim value to input
+          if (node_tensor->shape_status() == luci::ShapeStatus::VALID && axis < node_tensor->rank())
+            shape_by_input.dim(axis) = node_tensor->dim(axis);
+          else
+          {
+            // stop to check if this case exist for debugging
+            LUCI_ASSERT(false, "Check Reshape shape with 0");
+          }
+        }
+        else
+        {
+          shape_by_input.dim(axis).set(const_shape_node->at<S32>(axis));
+        }
+        // check valid or stop for debugging
+        assert(shape_by_input.dim(axis).value() > 0 || !shape_by_input.dim(axis).known());
       }
     }
     else
@@ -143,14 +162,26 @@ loco::TensorShape Algorithm::visit(const luci::CircleReshape *node)
   {
     for (uint32_t dim_index = 0; dim_index < output_shape.rank(); ++dim_index)
     {
-      const uint32_t dim_value = output_shape.dim(dim_index).value();
+      uint32_t dim_value = output_shape.dim(dim_index).value();
       if (not output_shape.dim(dim_index).known())
       {
         LUCI_ASSERT(unknown_dim_index == UINT32_MAX, "More than one unknown dimension");
         unknown_dim_index = dim_index;
       }
       else
       {
+        if (!dim_value)
+        {
+          // refer https://github.com/Samsung/ONE/issues/14074#issuecomment-2370795003
+          // set dim value to follow input
+          if (dim_index < input_shape.rank())
+            dim_value = input_shape.dim(dim_index).value();
+          else
+          {
+            // stop to check if this case exist for debugging
+            LUCI_ASSERT(false, "Check Reshape shape with 0");
+          }
+        }
         output_element_count *= dim_value;
       }
     }

compiler/luci/tests/test.lst

@@ -142,6 +142,7 @@ addread(Reshape_000)
 addread(Reshape_001)
 addread(Reshape_002)
 #addread(Reshape_003) # no input, no option is not supported
+addread(Reshape_004)
 addread(Reshape_U8_000)
 addread(ResizeBilinear_000)
 addread(ResizeBilinear_U8_000)
@@ -374,6 +375,7 @@ addwrite(Reshape_000)
 addwrite(Reshape_001)
 addwrite(Reshape_002)
 #addwrite(Reshape_003) # no input, no option is not supported
+addwrite(Reshape_004)
 addwrite(Reshape_U8_000)
 addwrite(ResizeBilinear_000)
 addwrite(ResizeBilinear_U8_000)

compiler/onnx-tools/onnx-extract.py

@@ -0,0 +1,38 @@
+#!/usr/bin/env python3
+
+# Copyright (c) 2024 Samsung Electronics Co., Ltd. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import onnx
+import os
+import sys
+
+
+def _help_exit(cmd_name):
+    print('Produce shape-infered ONNX file')
+    print('Usage: {0} [onnx_in_path] [onnx_out_path]'.format(cmd_name))
+    print('')
+    exit()
+
+
+def main():
+    if len(sys.argv) < 3:
+        _help_exit(os.path.basename(sys.argv[0]))
+
+    onnx.checker.check_model(sys.argv[1])
+    onnx.shape_inference.infer_shapes_path(sys.argv[1], sys.argv[2])
+
+
+if __name__ == "__main__":
+    main()

res/TensorFlowLiteRecipes/Reshape_004/test.recipe

@@ -0,0 +1,31 @@
+# NOTE test model for 0 in shape.
+# May not work in interpreter.
+operand {
+  name: "ifm"
+  type: FLOAT32
+  shape { dim: 1 dim: 3 dim: 2 dim: 3 }
+}
+operand {
+  name: "shape"
+  type: INT32
+  shape { dim: 3 }
+  filler { tag: "explicit" arg: "0" arg: "3" arg: "-1" }
+}
+operand {
+  name: "ofm"
+  type: FLOAT32
+  shape { dim: 1 dim: 3 dim: 6 }
+}
+operation {
+  type: "Reshape"
+  reshape_options {
+    new_shape: 0
+    new_shape: 3
+    new_shape: -1
+  }
+  input: "ifm"
+  input: "shape"
+  output: "ofm"
+}
+input: "ifm"
+output: "ofm"