def validate(shape, value, dtype):
def before_left(x, elem_op, full):
return elem_op(full, x)
def after_left(x, elem_op, value):
return elem_op(relay.const(value, dtype), x)
def before_right(x, elem_op, full):
return elem_op(x, full)
def after_right(x, elem_op, value):
return elem_op(x, relay.const(value, dtype))
x = relay.var("x", shape=shape, dtype=dtype)
elem_ops = [relay.add, relay.multiply, relay.subtract, relay.divide]
full_ops = []
if value == 0:
full_ops.append(relay.zeros(shape, dtype))
full_ops.append(relay.zeros_like(x))
if value == 1:
full_ops.append(relay.ones(shape, dtype))
full_ops.append(relay.ones_like(x))
else:
full_ops.append(relay.full(relay.const(value, dtype), shape))
full_ops.append(relay.full_like(x, relay.const(value, dtype)))
for op in elem_ops:
for full in full_ops:
z = before_left(x, op, full)
zz = run_opt_pass(z, transform.SimplifyExpr())
after = run_opt_pass(after_left(x, op, value),
transform.InferType())
assert tvm.ir.structural_equal(zz, after)
z = before_right(x, op, full)
zz = run_opt_pass(z, transform.SimplifyExpr())
after = run_opt_pass(after_right(x, op, value),
transform.InferType())
assert tvm.ir.structural_equal(zz, after)
# Test the case in which x is broadcast to full's shape
full_ops = []
if value == 0:
full_ops.append(relay.zeros(shape * 2, dtype))
if value == 1:
full_ops.append(relay.ones(shape * 2, dtype))
else:
full_ops.append(relay.full(relay.const(value, dtype), shape * 2))
for op in elem_ops:
for full in full_ops:
z = before_left(x, op, full)
zz = run_opt_pass(z, transform.SimplifyExpr())
after = run_opt_pass(before_left(x, op, full),
transform.InferType())
assert tvm.ir.structural_equal(zz, after)
z = before_right(x, op, full)
zz = run_opt_pass(z, transform.SimplifyExpr())
after = run_opt_pass(before_right(x, op, full),
transform.InferType())
assert tvm.ir.structural_equal(zz, after)
def test_full_infer_type():
x = relay.var('x', relay.TensorType((), 'int8'))
y = relay.full(x, ())
yy = run_infer_type(y)
assert (yy.checked_type == relay.TensorType((), 'int8'))
x = relay.var('x', relay.TensorType((), 'float32'))
y = relay.full(x, (1, 2), 'int8')
('shape=' in y.astext())
yy = run_infer_type(y)
assert (yy.checked_type == relay.TensorType((1, 2), 'int8'))
def test_full():
# default settings: match input dtype
x = relay.var("x", relay.TensorType((), "int8"))
y = relay.full(x, ())
yy = relay.ir_pass.infer_type(y)
assert yy.checked_type == relay.TensorType((), "int8")
# change the shape and dtype
x = relay.var("x", relay.TensorType((), "float32"))
y = relay.full(x, (1, 2), "int8")
"shape=" in y.astext()
yy = relay.ir_pass.infer_type(y)
assert yy.checked_type == relay.TensorType((1, 2), "int8")
def test_full():
# default settings: match input dtype
x = relay.var("x", relay.TensorType((), "int8"))
y = relay.full(x, ())
yy = relay.ir_pass.infer_type(y)
assert yy.checked_type == relay.TensorType((), "int8")
# change the shape and dtype
x = relay.var("x", relay.TensorType((), "float32"))
y = relay.full(x, (1, 2), "int8")
"shape=" in y.astext()
yy = relay.ir_pass.infer_type(y)
assert yy.checked_type == relay.TensorType((1, 2), "int8")
def test_arange_full_and_reshape(self):
start = relay.expr.const(0.0)
stop = relay.expr.const(10.0)
step = relay.expr.const(1.0)
fill_val = relay.expr.const(1.0)
fill_shape = [10, 1]
dtype = "float32"
left = relay.arange(start, stop, step, dtype)
left = relay.reshape(left, [-1, 1])
left = relay.reshape(left, [1, -1])
right = relay.full(fill_val, fill_shape, dtype)
right = relay.reshape(right, [1, -1])
net = relay.multiply(left, right)
mod = tvm.IRModule.from_expr(net)
params = {}
xgraph = xf_relay.from_relay(mod, params)
layers = xgraph.get_layers()
assert len(layers) == 10
assert layers[0].type[0] == "Constant"
assert layers[3].type[0] == "AnyOp"
assert layers[7].type[0] == "AnyOp"
assert layers[5].shapes == [1, 10]
assert layers[8].shapes == [1, 10]
def verify_full(fill_value, src_shape, dtype):
x = relay.var("x", relay.scalar_type(dtype))
rank = len(src_shape)
dyn_src_shape = relay.var("dyn_scr_shape", relay.ty.TensorType((rank,), 'int64'))
z = relay.full(x, dyn_src_shape, dtype)
func = relay.Function([x, dyn_src_shape], z)
ref_res = np.full(src_shape, fill_value).astype(dtype)
verify_func(func, [np.array(fill_value).astype(dtype), np.array(src_shape).astype('int64')], ref_res)
def test_concretize_full_like():
dtype = "int32"
shape_like = relay.var("shape_like", shape=(3, 4, 5), dtype=dtype)
fill_value = relay.var("fill", relay.TensorType((), "float32"))
expr = relay.full_like(shape_like, fill_value)
expected = run_infer_type(relay.full(fill_value, (3, 4, 5), dtype))
actual = run_opt_pass(expr, relay.transform.SimplifyExpr())
assert tvm.ir.structural_equal(actual, expected)
def verify_full(fill_value, src_shape, dtype):
x = relay.var("x", relay.scalar_type(dtype))
z = relay.full(x, src_shape, dtype)
func = relay.Function([x], z)
ref_res = np.full(src_shape, fill_value)
for target, ctx in ctx_list():
for kind in ["graph", "debug"]:
intrp = relay.create_executor(kind, ctx=ctx, target=target)
op_res = intrp.evaluate(func)(np.array(fill_value, dtype))
tvm.testing.assert_allclose(op_res.asnumpy(), ref_res, rtol=1e-5)
def verify_full(fill_value, src_shape, dtype):
x = relay.var('x', relay.scalar_type(dtype))
z = relay.full(x, src_shape, dtype)
func = relay.Function([x], z)
ref_res = np.full(src_shape, fill_value)
for (target, ctx) in tvm.testing.enabled_targets():
for kind in ['graph', 'debug']:
intrp = relay.create_executor(kind, ctx=ctx, target=target)
op_res = intrp.evaluate(func)(np.array(fill_value, dtype))
tvm.testing.assert_allclose(op_res.asnumpy(), ref_res, rtol=1e-05)
def test_compile_full():
# Shape calculations can happen in int64. The test checks that full operator
# can handle when shapes are not int32
shape = (tvm.tir.IntImm('int32', 1), tvm.tir.IntImm('int64', 16),
tvm.tir.IntImm('int64', 16), tvm.tir.IntImm('int32', 64))
output = relay.full(relay.const(0, 'int32'), shape=shape, dtype='int32')
f = relay.Function([], output)
mod = tvm.IRModule.from_expr(f)
mod = relay.qnn.transform.CanonicalizeOps()(mod)
relay.build(mod, 'llvm')
def verify_full(fill_value, src_shape, dtype):
x = relay.var("x", relay.scalar_type(dtype))
z = relay.full(x, src_shape, dtype)
func = relay.Function([x], z)
ref_res = np.full(src_shape, fill_value)
for target, ctx in ctx_list():
for kind in ["graph", "debug"]:
intrp = relay.create_executor(kind, ctx=ctx, target=target)
op_res = intrp.evaluate(func)(np.array(fill_value, dtype))
tvm.testing.assert_allclose(op_res.asnumpy(), ref_res, rtol=1e-5)
def test_full():
# default settings: match input dtype
ib = relay.ir_builder.IRBuilder()
x = ib.param("x", relay.TensorType((), "int8"))
with ib.function(x) as func:
ib.ret(relay.full(x, ()))
ib.ret(func)
func = relay.ir_pass.infer_type(ib.env, func.to_func())
ftype = func.checked_type
assert ftype.ret_type == relay.TensorType((), "int8")
# change the shape and dtype
ib = relay.ir_builder.IRBuilder()
x = ib.param("x", relay.TensorType((), "float32"))
with ib.function(x) as func:
ib.ret(relay.full(x, (1, 2), "int8"))
ib.ret(func)
func = relay.ir_pass.infer_type(ib.env, func.to_func())
ftype = func.checked_type
assert ftype.ret_type == relay.TensorType((1, 2), "int8")
def verify_full(fill_value, fill_shape, dtype):
x = relay.var("x", relay.scalar_type(dtype))
y = relay.var("y", relay.TensorType(fill_shape, 'int64'))
z = relay.full(x, relay.shape_of(y), dtype)
func = run_infer_type(relay.Function([x, y], z))
func2 = run_opt_pass(run_opt_pass(func, transform.DynamicToStatic()), transform.InferType())
zz = func2.body
assert isinstance(zz, relay.Call)
assert zz.op == relay.op.get("full")
ref_res = np.full(fill_shape, fill_value).astype(dtype)
y_data = np.random.uniform(low=-1, high=1, size=fill_shape).astype('int64')
verify_func(func2, [fill_value, y_data], ref_res)
def test_full(self):
fill_val = relay.expr.const(1.0)
fill_shape = [10, 1]
net = relay.full(fill_val, fill_shape, "float32")
net = relay.reshape(net, [1, -1])
mod = tvm.IRModule.from_expr(net)
params = {}
xgraph = xf_relay.from_relay(mod, params)
layers = xgraph.get_layers()
assert layers[0].type[0] == "Constant"
assert layers[0].shapes == [1]
assert layers[1].type[0] == "AnyOp"
assert layers[1].shapes == [10, 1]
assert layers[2].type[0] == "Reshape"
assert layers[2].shapes == [1, 10]
def before():
dtype = 'float32'
return relay.full(relay.const(1.0, dtype), c_shape, dtype=dtype)
