def test_single_dim_norm():
with pm.Node(name="elem1") as graph:
m = pm.parameter("m")
x = pm.input("x", shape=m)
w = pm.state("w", shape=m)
i = pm.index(0, m - 1, name="i")
w[i] = (w[i] * x[i])
x_ = np.random.randint(0, 10, 3)
w_ = np.random.randint(0, 10, 3)
coarse_eval = graph("w", x=x_, w=w_)
np_result = x_ * w_
np.testing.assert_allclose(coarse_eval, np_result)
shape_pass = NormalizeGraph({"m": 3})
graph_shapes = shape_pass(graph)
shape_res = graph_shapes("w", x=x_, w=w_)
np.testing.assert_allclose(shape_res, np_result)
lower_pass = Lower({})
lowered_graph = lower_pass(graph_shapes)
input_info = {f"w/w({i},)": w_[i] for i in range(len(w_))}
input_info.update({f"x/x({i},)": x_[i] for i in range(len(x_))})
fine_grained_eval = lowered_graph("w/w(1,)", input_info)
assert fine_grained_eval == np_result[1]
pb_path = f"{OUTPATH}/{graph.name}.srdfg"
pm.pb_store(lowered_graph, OUTPATH)
loaded_node = pm.pb_load(pb_path)
input_info = {f"w/w({i},)": w_[i] for i in range(len(w_))}
input_info.update({f"x/x({i},)": x_[i] for i in range(len(x_))})
fine_grained_eval = loaded_node("w/w(1,)", input_info)
assert fine_grained_eval == np_result[1]
def test_conv_embedded_values(x_shape, w_shape, params):
shape_dict = {
"n": x_shape[0],
"ic": x_shape[1],
"ih": x_shape[2],
"iw": x_shape[3],
"nf": w_shape[0],
"kh": w_shape[2],
"kw": w_shape[3],
"stride": params["stride"],
"pad": params["pad"]
}
graph, input_info0, out_info, keys = conv(x_shape,
w_shape,
params,
coarse=True,
debug_matrix=True)
ngraph, input_info1, out_info, keys = conv(x_shape,
w_shape,
params,
coarse=False,
debug_matrix=True)
lower_pass = pm.Lower({})
lowered = lower_pass(ngraph)
pb_path = f"{OUTPATH}/{graph.name}.srdfg"
pm.pb_store(lowered, OUTPATH)
node = pm.pb_load(pb_path)
assert len(node.nodes) == len(lowered.nodes)
assert list(node.nodes.keys()) == list(lowered.nodes.keys())
def test_linear_deserialize():
graph_name = "linear_reg1"
with pm.Node(name=graph_name) as graph:
m = pm.placeholder("m")
x_ = pm.placeholder("x", shape=(m))
y_ = pm.placeholder("y")
w_ = pm.placeholder("w", shape=(m))
mu = pm.parameter(name="mu", default=1.0)
i = pm.index(0, (m - 1).set_name("m-1"), name="i")
h = pm.sum([i], (x_[i] * w_[i]).set_name("x*w"), name="h")
d = (h - y_).set_name("h-y")
g = (d * x_[i]).set_name("d*x")
mug = (mu * g[i]).set_name("mu*g[i]")
w_ = ((w_[i]) - mug).set_name("w_out")
x = np.random.randint(0, 10, 10)
y = np.random.randint(0, 10, 1)[0]
w = np.random.randint(0, 10, 10)
graph_res = graph("w_out", {"x": x, "y": y, "w": w})
actual_res = w - ((np.sum(x * w) - y) * x) * 1.0
np.testing.assert_allclose(graph_res, actual_res)
cwd = Path(f"{__file__}").parent
base_path = f"{cwd}/pmlang_examples"
full_path = f"{base_path}/outputs"
pb_path = f"{full_path}/{graph_name}.srdfg"
pm.pb_store(graph, full_path)
node = pm.pb_load(pb_path)
new_graph_res = node("w_out", {"x": x, "y": y, "w": w})
np.testing.assert_allclose(graph_res, new_graph_res)
np.testing.assert_allclose(actual_res, new_graph_res)
def test_linear_reg():
m_ = 3
graph, input_info, out_info, keys = linear(m=m_, coarse=True)
coarse_eval = graph(keys, input_info)
np.testing.assert_allclose(coarse_eval, out_info["w"])
fgraph, input_info, out_info, keys = linear(m=m_, coarse=False)
lower_pass = Lower({})
lowered_graph = lower_pass(fgraph, {})
all_vals = lowered_graph(keys, input_info)
out = np.asarray(all_vals).reshape(out_info["w"].shape)
np.testing.assert_allclose(out, out_info["w"])
cwd = Path(f"{__file__}").parent
base_path = f"{cwd}/pmlang_examples"
full_path = f"{base_path}/outputs"
pb_path = f"{full_path}/{graph.name}.srdfg"
pm.pb_store(lowered_graph, full_path)
loaded_node = pm.pb_load(pb_path)
_, input_info, out_info, keys = linear(m=m_, coarse=False)
loaded_res = loaded_node(keys, input_info)
out = np.asarray(loaded_res).reshape(out_info["w"].shape)
np.testing.assert_allclose(out, out_info["w"])
def test_resnet18_train():
filename = f"resnet18_train.onnx"
filepath = f"{BENCH_DIR}/full_dnns/{filename}"
assert Path(filepath).exists()
graph = pm.from_onnx(filepath)
full_path = f"{BENCH_DIR}/full_dnns"
pb_path = f"{full_path}/resnet18_train.srdfg"
pm.pb_store(graph, full_path)
node = pm.pb_load(pb_path, verbose=True)
assert len(node.nodes) == len(graph.nodes)
def test_genesys_conv_bias():
graph = pm.pb_load(f"{LAYER_DIR}/resnet18_conv_bias.srdfg")
genesys = define_genesys(GENESYS_CFG)
program = initialize_program(graph, genesys)
program.add_compilation_step("pad_operands",
pad_operands,
preproc=True,
stage_kwargs={'shaped_nodes': []})
program.add_compilation_step("tile", tile)
program.add_compilation_step("hoist", hoist, dependencies=["tile"])
program.compile()
res = program.emit("string_final")
print(res)
def test_genesys_global_avg_pool():
from pprint import pprint
graph = pm.pb_load(f"{LAYER_DIR}/resnet18_globalaveragepool.srdfg")
genesys = define_genesys(GENESYS_CFG)
program = initialize_program(graph, genesys)
program.add_compilation_step("pad_operands",
pad_operands,
preproc=True,
stage_kwargs={'shaped_nodes': []})
program.add_compilation_step("tile", tile)
program.add_compilation_step("hoist", hoist, dependencies=["tile"])
program.compile()
res = program.emit("json_no_ops")
pprint(res)
def test_resnet_simulation():
# Load polymath program
graph = pm.pb_load(f"{BENCH_DIR}/resnet18v1.srdfg")
# Create Genesys architecture graph
genesys_cfg = parse_cfg()
genesys = generate_genesys(genesys_cfg)
# Compile the program
program = compile(graph,
genesys,
f"{BENCH_DIR}",
store_output=True,
output_type="json")
def test_genesys_gemm():
graph = pm.pb_load(f"{LAYER_DIR}/resnet18_gemm.srdfg")
batch_size_pass = pm.UpdateBatchSize(32, graph.op_name)
graph = batch_size_pass(graph)
genesys = define_genesys(GENESYS_CFG)
program = initialize_program(graph, genesys)
program.add_compilation_step("pad_operands",
pad_operands,
preproc=True,
stage_kwargs={'shaped_nodes': []})
program.add_compilation_step("tile", tile)
program.add_compilation_step("hoist", hoist, dependencies=["tile"])
program.compile()
res = program.emit("json_no_ops")
pprint(res)
def test_lenet():
filename = f"lenet.onnx"
full_path = f"{BENCH_DIR}/full_dnns"
filepath = f"{full_path}/{filename}"
pb_path = f"{full_path}/lenet.srdfg"
assert Path(filepath).exists()
graph = pm.from_onnx(filepath)
pm.pb_store(graph, full_path)
node = pm.pb_load(pb_path, verbose=True)
assert len(node.nodes) == len(graph.nodes)
for name, n in node.nodes.items():
if n.op_name == "conv":
print(n.kwargs.keys())
break
def test_tabla_linear(m_):
shape_dict = {"m": m_}
graph, input_info, out_info, keys = linear(m=m_, coarse=True)
lgraph, input_info, out_info, keys = linear(m=m_, coarse=False)
cwd = Path(f"{__file__}").parent
base_path = f"{cwd}/pmlang_examples"
full_path = f"{base_path}/outputs"
graph_name = f"{graph.name}_{m_}"
tabla_path = f"{full_path}/{graph_name}_tabla.json"
tabla_ir, tabla_graph = pm.generate_tabla(graph,
shape_dict,
tabla_path,
context_dict=input_info,
add_kwargs=True)
cwd = Path(f"{__file__}").parent
base_path = f"{cwd}/pmlang_examples"
full_path = f"{base_path}/outputs"
pb_path = f"{full_path}/{graph.name}.srdfg"
pm.pb_store(graph, full_path)
node = pm.pb_load(pb_path)
def test_lower_group_op():
with pm.Node(name="linear_reg1") as graph:
m = pm.parameter(name="m")
x = pm.input("x", shape=(m))
y = pm.input("y")
w = pm.state("w", shape=(m))
i = pm.index(0, m - 1, name="i")
h = pm.sum([i], w[i] * x[i], name="h")
m_ = 3
n_ = 3
x_ = np.random.randint(0, 10, m_)
w_ = np.random.randint(0, 10, (m_))
np_result = np.sum(x_ * w_)
np.testing.assert_allclose(graph("h", {"w": w_, "x": x_}), np_result)
np.testing.assert_allclose(graph("h", w=w_, x=x_), np_result)
shape_pass = NormalizeGraph({"m": m_, "n": n_})
graph_shapes = shape_pass(graph)
shape_res = graph_shapes("h", x=x_, w=w_)
np.testing.assert_allclose(shape_res, np_result)
lower_pass = Lower({})
lowered_graph = lower_pass(graph_shapes)
input_info = {f"w/w({i},)": w_[i] for i in range(len(w_))}
input_info.update({f"x/x({i},)": x_[i] for i in range(len(x_))})
#
fine_grained_eval = lowered_graph("h/h(4,)", input_info)
assert fine_grained_eval == np_result
pb_path = f"{OUTPATH}/linear_reg1.srdfg"
pm.pb_store(lowered_graph, OUTPATH)
loaded_node = pm.pb_load(pb_path) #
input_info = {f"w/w({i},)": w_[i] for i in range(len(w_))}
input_info.update({f"x/x({i},)": x_[i] for i in range(len(x_))})
loaded_res = loaded_node("h/h(4,)", input_info)
assert loaded_node.func_hash() == lowered_graph.func_hash()
assert loaded_res == np_result
def run_benchmark(benchmark_name, output_type="simulation"):
assert benchmark_name in ['resnet18', 'resnet50', 'maskrcnn']
graph = pm.pb_load(f"{MODEL_DIR}/{benchmark_name}.srdfg")
genesys = define_genesys(GENESYS_CFG)
program = initialize_program(graph, genesys)
program.add_compilation_step("pad_operands",
pad_operands,
preproc=True,
stage_kwargs={'shaped_nodes': []})
program.add_compilation_step("tile", tile)
program.add_compilation_step("hoist", hoist, dependencies=["tile"])
program.compile()
if output_type == 'simulation':
res = program.emit("json")
with open(f"{BENCH_DIR}/compilation_output/{benchmark_name}.json",
'w') as outfile:
json.dump(res, outfile)
else:
res = program.emit("string_final")
with open(f"{BENCH_DIR}/compilation_output/{benchmark_name}.txt",
'w') as outfile:
outfile.write(res)
def test_reco():
m_ = 3
n_ = 3
k_ = 2
shape_dict = {"m": n_, "k": k_, "n": n_}
graph, input_info, out_info, keys = reco(coarse=True, **shape_dict)
coarse_eval = graph(keys, input_info)
np.testing.assert_allclose(coarse_eval[0], out_info["w1"])
np.testing.assert_allclose(coarse_eval[1], out_info["w2"])
fgraph, input_info, out_info, keys = reco(coarse=False, **shape_dict)
lower_pass = Lower({})
lowered_graph = lower_pass(fgraph, {})
all_vals = lowered_graph(keys, input_info)
w1_elems = np.prod(out_info["w1"].shape)
w2_elems = np.prod(out_info["w2"].shape)
out1 = np.asarray(list(all_vals[0:w1_elems])).reshape(out_info["w1"].shape)
out2 = np.asarray(list(all_vals[w1_elems:])).reshape(out_info["w2"].shape)
np.testing.assert_allclose(out1, out_info["w1"])
np.testing.assert_allclose(out2, out_info["w2"])
cwd = Path(f"{__file__}").parent
base_path = f"{cwd}/pmlang_examples"
full_path = f"{base_path}/outputs"
pb_path = f"{full_path}/{graph.name}.srdfg"
pm.pb_store(lowered_graph, full_path)
loaded_node = pm.pb_load(pb_path)
_, input_info, out_info, keys = reco(coarse=False, **shape_dict)
loaded_res = loaded_node(keys, input_info)
lres1 = np.asarray(list(loaded_res[0:w1_elems])).reshape(out_info["w1"].shape)
lres2 = np.asarray(list(loaded_res[w1_elems:])).reshape(out_info["w2"].shape)
np.testing.assert_allclose(lres1, out_info["w1"])
np.testing.assert_allclose(lres2, out_info["w2"])
def test_genesys_add():
graph = pm.pb_load(f"{LAYER_DIR}/resnet18_add.srdfg")
GENESYS_DTYPES['SIMD'] = 'FXP16'
GENESYS_DTYPES['SYSTOLIC_ARRAY'] = {}
GENESYS_DTYPES['SYSTOLIC_ARRAY']['inp_weight'] = 'FXP4'
GENESYS_DTYPES['SYSTOLIC_ARRAY']['bias_out'] = 'FXP16'
update_genesys_cfg_from_dtypes()
genesys = define_genesys(GENESYS_CFG)
print(genesys.get_subgraph_node("VMEM1").size_bytes)
program = initialize_program(graph, genesys)
program.add_compilation_step("update_operand_dtypes",
update_operand_dtypes,
preproc=True,
stage_kwargs={'dtype_map': GENESYS_DTYPES})
program.add_compilation_step("pad_operands",
pad_operands,
preproc=True,
stage_kwargs={'shaped_nodes': []})
program.add_compilation_step("tile", tile)
program.add_compilation_step("hoist", hoist, dependencies=["tile"])
program.compile()
res = program.emit("json_no_ops")
pprint(res)
def get_single_conv2d_node():
resnet18 = pm.pb_load(f"{BENCH_DIR}/resnet18v1.srdfg")
for name, node in resnet18.nodes.items():
if node.name == "conv":
return node