def test_const_dtype():
strides = (1, 1)
np_array = np.array(strides).astype("int32")
strides = _op.const(np_array, dtype="int64")
# strides needs to be autoconverted to int64 on Windows
assert infer_type(strides).checked_type.dtype == np.dtype(np.int64)
a = tvm.nd.array(np.random.randint(0, high=255, size=(2, 3),
dtype="uint8"))
a = _op.const(a, dtype="uint8")
aa = a.data.numpy()
assert aa.dtype == np.dtype(np.uint8)
b = _op.const(1, dtype="int8")
bb = b.data.numpy()
assert bb.dtype == np.dtype(np.int8)
kshape = (3, 10, 3, 3)
w = relay.const(np.zeros(kshape, dtype="float32"))
assert w.data.numpy().dtype == np.dtype(np.float32)
def _calculate_qparam(inp):
# reference ATen/native/quantized/cpu/qlinear_dynamic.cpp
# ChooseQuantizationParams function
mn = _op.min(inp)
mx = _op.max(inp)
# Ensure that the interval contains 0
mn = _op.minimum(mn, _op.const(0.0, dtype="float32"))
mx = _op.maximum(mx, _op.const(0.0, dtype="float32"))
qmax = 255
# reduce_range became True in v1.6
if is_version_greater_than("1.5.1"):
qmax = 127
scale = (mx - mn) / _expr.const(qmax, dtype="float32")
zero_point_from_min = -(mn / scale)
zero_point = _op.cast(_op.round(_op.clip(zero_point_from_min, 0.0, qmax)), "int32")
return scale, zero_point