From eafb81664602a0ec514319e8a38483428eca1c13 Mon Sep 17 00:00:00 2001
From: zjgarvey <47986913+zjgarvey@users.noreply.github.com>
Date: Wed, 25 Sep 2024 16:42:45 -0700
Subject: [PATCH] [alt] adds a numerics failure reproducer (#351)

This particular Conv op signature was found to be causing numerics
failures in a few models. E.g., "maxvit_rmlp_base_rw_224.sw_in12k"
---
 alt_e2eshark/onnx_tests/operators/conv.py | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/alt_e2eshark/onnx_tests/operators/conv.py b/alt_e2eshark/onnx_tests/operators/conv.py
index 6c412bda..7f94084c 100644
--- a/alt_e2eshark/onnx_tests/operators/conv.py
+++ b/alt_e2eshark/onnx_tests/operators/conv.py
@@ -9,6 +9,21 @@
 from ..helper_classes import BuildAModel
 from e2e_testing.registry import register_with_name, register_test
 
+class ConvRepro(BuildAModel):
+    def construct_i_o_value_info(self):
+        self.input_vi = [
+            make_tensor_value_info("X", TensorProto.FLOAT, [1,256,112,112]),
+            make_tensor_value_info("W", TensorProto.FLOAT, [256,1,3,3]),
+            make_tensor_value_info("B", TensorProto.FLOAT, [256]),
+            ]
+        self.output_vi = [make_tensor_value_info("Y", TensorProto.FLOAT, [1,256,56,56])]
+    
+    def construct_nodes(self):
+        app_node = self.get_app_node()
+        app_node("Conv",["X","W","B"],["Y"],group=256,kernel_shape=[3,3],pads=[1,1,1,1],strides=[2,2])
+
+register_test(ConvRepro, "conv_depthwise_stride_2")
+
 class QConvModelBase(BuildAModel):
     def __init__(self, specs, *args, **kwargs):
         (self.N, self.Cin, self.Hin, self.Win, self.Cout, self.groups, self.Hker, self.Wker, self.pads, self.dilations, self.strides) = specs