relay.build fails with NHWC not supported

[Tom Gall]

Hi!

New to TVM yet, so pointers to the fine manual are certainly much appreciated.

I'm trying to use OpenCL on intel. x86_64 linux.

Failure reported is :
File "./testmobilenet-opencl.py", line 43, in
graph, lib, params = relay.build(mod, target, target_host, params=params)

File "/home/tgall/tvm/tvm/python/tvm/relay/build_module.py", line 244, in build
graph_json, mod, params = bld_mod.build(func, target, target_host, params)

File "/home/tgall/tvm/tvm/python/tvm/relay/build_module.py", line 109, in build
self._build(func, target, target_host)

File "/home/tgall/tvm/tvm/python/tvm/_ffi/_ctypes/function.py", line 207, in call
raise get_last_ffi_error()

ValueError: Traceback (most recent call last):
[bt] (8) /home/tgall/tvm/tvm/build/libtvm.so(tvm::relay::ScheduleGetter::VisitExpr(tvm::RelayExpr const&)+0x9e) [0x7f17a28aa4ae]
[bt] (7) /home/tgall/tvm/tvm/build/libtvm.so(tvm::relay::ExprFunctor<tvm::Array<tvm::Tensor, void> (tvm::RelayExpr const&)>::VisitExpr(tvm::RelayExpr const&)+0x82) [0x7f17a28a8662]
[bt] (6) /home/tgall/tvm/tvm/build/libtvm.so(tvm::relay::ExprFunctor<tvm::Array<tvm::Tensor, void> (tvm::RelayExpr const&)>::InitVTable()::{lambda(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<tvm::Array<tvm::Tensor, void> (tvm::RelayExpr const&)>)#6}::_FUN(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<tvm::Array<tvm::Tensor, void> (tvm::RelayExpr const&)>)+0x2c) [0x7f17a289acdc]
[bt] (5) /home/tgall/tvm/tvm/build/libtvm.so(tvm::relay::ScheduleGetter::VisitExpr_(tvm::relay::CallNode const*)+0x154) [0x7f17a28a5864]
[bt] (4) /home/tgall/tvm/tvm/build/libtvm.so(tvm::relay::ScheduleGetter::VisitExpr(tvm::RelayExpr const&)+0x9e) [0x7f17a28aa4ae]
[bt] (3) /home/tgall/tvm/tvm/build/libtvm.so(tvm::relay::ExprFunctor<tvm::Array<tvm::Tensor, void> (tvm::RelayExpr const&)>::VisitExpr(tvm::RelayExpr const&)+0x82) [0x7f17a28a8662]
[bt] (2) /home/tgall/tvm/tvm/build/libtvm.so(tvm::relay::ExprFunctor<tvm::Array<tvm::Tensor, void> (tvm::RelayExpr const&)>::InitVTable()::{lambda(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<tvm::Array<tvm::Tensor, void> (tvm::RelayExpr const&)>)#6}::_FUN(tvm::runtime::ObjectRef const&, tvm::relay::ExprFunctor<tvm::Array<tvm::Tensor, void> (tvm::RelayExpr const&)>)+0x2c) [0x7f17a289acdc]
[bt] (1) /home/tgall/tvm/tvm/build/libtvm.so(tvm::relay::ScheduleGetter::VisitExpr_(tvm::relay::CallNode const*)+0x6ef) [0x7f17a28a5dff]
[bt] (0) /home/tgall/tvm/tvm/build/libtvm.so(+0x38ace3) [0x7f17a20d5ce3]
File "/home/tgall/tvm/tvm/python/tvm/_ffi/_ctypes/function.py", line 72, in cfun
rv = local_pyfunc(*pyargs)
File "/home/tgall/tvm/tvm/python/tvm/relay/op/nn/_nn.py", line 216, in compute_conv2d
dilation, layout, out_dtype)
File "</home/tgall/.local/lib/python3.7/site-packages/decorator.py:decorator-gen-35>", line 2, in conv2d
File "/home/tgall/tvm/tvm/python/tvm/target.py", line 382, in dispatch_func
return dispatch_dict[k](*args, **kwargs)
File "</home/tgall/.local/lib/python3.7/site-packages/decorator.py:decorator-gen-178>", line 2, in config_dispatcher
File "/home/tgall/tvm/tvm/python/tvm/autotvm/task/dispatcher.py", line 216, in dispatch_func
return dispatch_dict['direct'](cfg, *args, **kwargs)
File "/home/tgall/tvm/tvm/python/tvm/autotvm/task/topi_integration.py", line 400, in template_call
node = f(cfg, *args, **kwargs)
File "/home/tgall/tvm/tvm/topi/python/topi/cuda/conv2d.py", line 126, in conv2d_cuda
raise ValueError("not support this layout {} yet".format(layout))
ValueError: not support this layout NHWC yet

^^^^^^^^^^^

The code that reproduces this is (to me) a pretty boring test of MobileNetv1. I have an llvm version that works FWIW.

Python:

import os
import flatbuffers

import tvm
from tvm import relay

import tflite.Model

from PIL import Image

from matplotlib import pyplot as plt

import numpy as np
from tvm.contrib import graph_runtime as runtime

model_dir = os.path.dirname(".")
tflite_model_file = os.path.join(model_dir, "mobilenet_v1_1.0_224.tflite")
tflite_model_buf = open(tflite_model_file, "rb").read()

tflite_model = tflite.Model.Model.GetRootAsModel(tflite_model_buf, 0)

resized_image = Image.open('/home/tgall/tvm/opencl-exp/mobilenet/cat.png').resize((224, 224))

image_data = np.asarray(resized_image).astype("float32")
image_data = np.expand_dims(image_data, axis=0)

image_data[:, :, :, 0] = 2.0 / 255.0 * image_data[:, :, :, 0] - 1
image_data[:, :, :, 1] = 2.0 / 255.0 * image_data[:, :, :, 1] - 1
image_data[:, :, :, 2] = 2.0 / 255.0 * image_data[:, :, :, 2] - 1
print('input', image_data.shape)

input_tensor = "input"
input_shape = (1, 224, 224, 3)
input_dtype = "float32"

mod, params = relay.frontend.from_tflite(tflite_model,
shape_dict={input_tensor: input_shape},
dtype_dict={input_tensor: input_dtype})

target = "opencl"
target_host = "llvm"

with relay.build_config(opt_level=3):
graph, lib, params = relay.build(mod, target, target_host, params=params)

module = runtime.create(graph, lib, tvm.cl(0))
module.set_input(input_tensor, tvm.nd.array(image_data))

module.set_input(**params)

module.run()

tvm_output = module.get_output(0).asnumpy()

label_file = "labels_mobilenet_quant_v1_224.txt"
label_path = os.path.join(model_dir, label_file)

with open(label_path) as f:
labels = f.readlines()

predictions = np.squeeze(tvm_output)
prediction = np.argmax(predictions)

print("The image prediction result is: id " + str(prediction) + " name: " + labels[prediction])