add tvm fp16 yolov5s example

yolo_half.py

+from tvm import testing
+import onnx
+#import onnxruntime
+testing.utils.install_request_hook(depth=3)
+# sphinx_gallery_end_ignore                                                                                                                                                                                                  
+from PIL import Image
+import numpy as np
+from tvm.relay.transform import InferType, ToMixedPrecision
+import tvm
+from tvm import relay, auto_scheduler
+import tvm.relay.testing
+from tvm.contrib import graph_executor
+
+'''
+
+  prepare:
+    vim /tvm-0.11-dev0/python/tvm/topi/rocm/conv2d.py +79
+    set param data_type = 0
+    export PYTHONPATH=/root/tvm-0.11-dev0/python:$PYTHONPATH
+    export MIOPEN_DEBUG_CONV_IMPLICIT_GEMM=0
+  
+'''
+
+img_data = np.random.rand(1,3,640,640).astype("float16")/255                                                                                                                             
+input_name = "images"
+shape_dict = {input_name: img_data.shape}
+model_path = 'yolov5s_half.onnx'
+onnx_model = onnx.load(model_path)
+# Define the neural network and compilation target
+
+batch_size = 1
+layout = "NCHW"
+target = "rocm -libs=miopen,rocblas"                                                                                                                                                     
+#target = "rocm"     
+dtype = "float16"
+mod, params = relay.frontend.from_onnx(onnx_model, shape_dict, dtype=dtype)
+# Compile with the history best
+
+print("Compile...")
+with tvm.transform.PassContext(opt_level=3):
+    lib = relay.build(mod, target=target, params=params)
+print('Compile success!')
+
+dev = tvm.device(str(target), 0)
+module = graph_executor.GraphModule(lib["default"](dev))
+module.set_input(input_name, img_data)
+module.run()
+res = module.get_output(0)
+print("res:{}, res shape:{}".format(res, res.shape))
+
+# use onnxruntime verify tvm output
+def verify(tvm_res):
+    session = onnxruntime.InferenceSession(model_path,providers=['CPUExecutionProvider'] )
+    input_name = session.get_inputs()[0].name
+    output_name = session.get_outputs()[0].name
+    output = session.run([output_name], {input_name: img_data})
+    tvm.testing.assert_allclose(output[0], tvm_res.numpy(), rtol=1e-2, atol=1e-2)
+    print("use onnxruntime verify successfully !")  
+
+#verify(res)
+
+print("Evaluate inference time cost...")
+print(module.benchmark(dev, repeat=1, min_repeat_ms=500))
+
+
+