ruff

_cmake/clang_format.sh

@@ -1,7 +1,7 @@
 #!/bin/bash
 clear
 echo "--ruff--"
-ruff .
+ruff check .
 echo "--cython-lint--"
 cython-lint .
 echo "--clang-format--"

_doc/examples/plot_bench_cypy_ort.py

@@ -89,7 +89,7 @@
 #############################################
 # Benchmark
 # +++++++++
-dims = list(int(i) for i in script_args.dims.split(","))
+dims = [int(i) for i in script_args.dims.split(",")]
 
 data = []
 for dim in tqdm(dims):
@@ -99,18 +99,18 @@
         number, repeat = script_args.number * 5, script_args.repeat * 5
     x = numpy.random.randn(dim, dim).astype(numpy.float32)
     t_ort = measure_time(
-        lambda: sess_ort.run(None, {"X": x})[0], number=number, repeat=50
+        lambda x=x: sess_ort.run(None, {"X": x})[0], number=number, repeat=50
     )
     t_ort["name"] = "ort"
     t_ort["dim"] = dim
     data.append(t_ort)
 
-    t_ext = measure_time(lambda: sess_ext.run([x])[0], number=number, repeat=repeat)
+    t_ext = measure_time(lambda x=x: sess_ext.run([x])[0], number=number, repeat=repeat)
     t_ext["name"] = "ext"
     t_ext["dim"] = dim
     data.append(t_ext)
 
-    t_ext2 = measure_time(lambda: sess_ext.run_1_1(x), number=number, repeat=repeat)
+    t_ext2 = measure_time(lambda x=x: sess_ext.run_1_1(x), number=number, repeat=repeat)
     t_ext2["name"] = "ext_1_1"
     t_ext2["dim"] = dim
     data.append(t_ext2)

_doc/examples/plot_bench_gemm_f8.py

@@ -111,14 +111,14 @@ def type2string(dt):
             dims.append((m, n, k))
         else:
             dims.append(int(d))
-    tests = list(int(i) for i in script_args.tests.split(","))
+    tests = [int(i) for i in script_args.tests.split(",")]
 
 pbar = tqdm(list(product(tests, dims)))
 obs = []
 for test, dim in pbar:
     pbar.set_description(f"type={test} dim={dim}")
     if test in {8, 9, 10, 12, 13}:
-        warnings.warn(f"unsupported configuration {test}.")
+        warnings.warn(f"unsupported configuration {test}.", stacklevel=0)
         continue
     mdim = dim if isinstance(dim, int) else max(dim)
     if mdim < 128:
@@ -187,7 +187,7 @@ def type2string(dt):
 # ++++++++++++++++++++++++
 
 if df.shape[0] > 0:
-    dfi = df[col_def + ["~dim", "mnk", "t-total", "t-gemm_sync"]]
+    dfi = df[[*col_def, "~dim", "mnk", "t-total", "t-gemm_sync"]]
     print(dfi)
 
 ###################################

_doc/examples/plot_bench_gemm_ort.py

@@ -237,15 +237,15 @@ def create_cast(to, cuda=False):
 #
 # The benchmark will run the following configurations.
 
-types = list(getattr(TensorProto, a) for a in script_args.types.split(","))
+types = [getattr(TensorProto, a) for a in script_args.types.split(",")]
 engine = [InferenceSession, CReferenceEvaluator]
 providers = [
     ["CUDAExecutionProvider", "CPUExecutionProvider"],
     ["CPUExecutionProvider"],
 ]
 # M, N, K
 # we use multiple of 8, otherwise, float8 does not work.
-dims = [list(int(i) for i in line.split(",")) for line in script_args.dims.split(";")]
+dims = [[int(i) for i in line.split(",")] for line in script_args.dims.split(";")]
 domains = ["onnx_extended.ortops.tutorial.cuda", "", "com.microsoft"]
 
 
@@ -420,7 +420,11 @@ def rendering_obs(obs, dim, number, repeat, domain, provider, internal_time):
         feeds = {"A": matrices[k1].numpy(), "B": matrices[k2].numpy()}
         sess = engine(onx)
         sess.run(None, feeds)
-        obs = measure_time(lambda: sess.run(None, feeds), repeat=repeat, number=number)
+        obs = measure_time(
+            lambda sess=sess, feeds=feeds: sess.run(None, feeds),
+            repeat=repeat,
+            number=number,
+        )
 
     elif engine == InferenceSession:
         if provider[0] not in get_available_providers():
@@ -441,13 +445,15 @@ def rendering_obs(obs, dim, number, repeat, domain, provider, internal_time):
         out_names = ["C"]
 
         # warmup
-        for i in range(script_args.warmup):
+        for _i in range(script_args.warmup):
             sess._sess.run_with_ort_values(the_feeds, out_names, None)[0]
 
         # benchamrk
         times = []
 
-        def fct_benchmarked():
+        def fct_benchmarked(
+            sess=sess, times=times, out_names=out_names, the_feeds=the_feeds
+        ):
             got = sess._sess.run_with_ort_values(the_feeds, out_names, None)
             if len(got) > 1:
                 times.append(got[1])

_doc/examples/plot_op_conv_denorm.py

@@ -185,9 +185,9 @@ def modify(onx, scale):
     diff0 = numpy.abs(got0 / scale - expected).max()
 
     # t1 = measure_time(lambda: sess1.run(None, feeds), repeat=2, number=5)
-    t2 = measure_time(lambda: sess2.run(None, feeds), repeat=2, number=5)
-    t3 = measure_time(lambda: sess3.run(None, feeds), repeat=2, number=5)
-    t4 = measure_time(lambda: sess4.run(None, feeds), repeat=2, number=5)
+    t2 = measure_time(lambda sess2=sess3: sess2.run(None, feeds), repeat=2, number=5)
+    t3 = measure_time(lambda sess3=sess3: sess3.run(None, feeds), repeat=2, number=5)
+    t4 = measure_time(lambda sess4=sess4: sess4.run(None, feeds), repeat=2, number=5)
     obs = dict(
         scale=scale,
         ort=t2["average"],
@@ -204,7 +204,9 @@ def modify(onx, scale):
         tb = torch.from_numpy(b)
         torch.nn.functional.conv2d(tx, tw, tb, padding=1)
         t3 = measure_time(
-            lambda: torch.nn.functional.conv2d(tx, tw, tb, padding=1),
+            lambda tx=tx, tw=tw, tb=tb: torch.nn.functional.conv2d(
+                tx, tw, tb, padding=1
+            ),
             repeat=2,
             number=5,
         )
@@ -223,8 +225,16 @@ def modify(onx, scale):
         cuda_feeds = {"X": x_ortvalue}
         sess2.run_with_ort_values(None, cuda_feeds)
         sess3.run_with_ort_values(None, cuda_feeds)
-        t2 = measure_time(lambda: sess2.run(None, cuda_feeds), repeat=2, number=5)
-        t3 = measure_time(lambda: sess3.run(None, cuda_feeds), repeat=2, number=5)
+        t2 = measure_time(
+            lambda sess2=sess2, cuda_feeds=cuda_feeds: sess2.run(None, cuda_feeds),
+            repeat=2,
+            number=5,
+        )
+        t3 = measure_time(
+            lambda sess3=sess3, cuda_feeds=cuda_feeds: sess3.run(None, cuda_feeds),
+            repeat=2,
+            number=5,
+        )
         obs["ort-cuda"] = t2["average"]
         obs["ort-cuda-opt"] = t2["average"]
 

_doc/examples/plot_op_conv_py_vs_c.py

@@ -105,8 +105,8 @@
     W = np.ones((1, 1, 3, 3), dtype=np.float32)
     B = np.array([[[[0]]]], dtype=np.float32)
     feeds = {"X": X, "W": W, "B": B}
-    t1 = measure_time(lambda: sess1.run(None, feeds))
-    t2 = measure_time(lambda: sess2.run(None, feeds))
+    t1 = measure_time(lambda feeds=feeds: sess1.run(None, feeds))
+    t2 = measure_time(lambda feeds=feeds: sess2.run(None, feeds))
     obs = dict(size=i, onnx=t1["average"], onnx_extended=t2["average"])
     data.append(obs)
     if unit_test_going() and len(data) >= 2:

_doc/examples/plot_op_gemm2_cuda.py

@@ -195,13 +195,13 @@ def benchmark(sess, sizes, label):
         bind, cuda_feeds = move_inputs(sess, feeds)
 
         begin = time.perf_counter()
-        for i in range(script_args.warmup):
+        for _i in range(script_args.warmup):
             # sess.run(None, feeds)
             sess._sess.run_with_iobinding(bind, None)
         warmup = time.perf_counter() - begin
 
         times = []
-        for i in range(script_args.repeat):
+        for _i in range(script_args.repeat):
             begin = time.perf_counter()
             # sess.run(None, feeds)
             sess._sess.run_with_iobinding(bind, None)

_doc/examples/plot_op_mul_cuda.py

@@ -210,13 +210,13 @@ def benchmark(sess, sizes, label):
         bind, cuda_feeds = move_inputs(sess, feeds)
 
         begin = time.perf_counter()
-        for i in range(script_args.warmup):
+        for _i in range(script_args.warmup):
             # sess.run(None, feeds)
             sess._sess.run_with_iobinding(bind, None)
         warmup = time.perf_counter() - begin
 
         times = []
-        for i in range(script_args.repeat):
+        for _i in range(script_args.repeat):
             begin = time.perf_counter()
             # sess.run(None, feeds)
             sess._sess.run_with_iobinding(bind, None)

_doc/examples/plot_op_scatternd_cuda.py

@@ -302,13 +302,13 @@ def benchmark(
         bind, cuda_feeds = move_inputs(sess, feeds)
 
         begin = time.perf_counter()
-        for i in range(script_args.warmup):
+        for _i in range(script_args.warmup):
             # sess.run(None, feeds)
             sess._sess.run_with_iobinding(bind, None)
         warmup = time.perf_counter() - begin
 
         times = []
-        for i in range(script_args.repeat):
+        for _i in range(script_args.repeat):
             begin = time.perf_counter()
             # sess.run(None, feeds)
             sess._sess.run_with_iobinding(bind, None)

_doc/examples/plot_op_scatternd_mask_cuda.py

@@ -333,12 +333,12 @@ def benchmark(sizes, config, itype, times_col: int = 1, times_indices: int = 1):
             sess = get_session(model)
             bind, cuda_feeds = move_inputs(sess, feeds)
             begin = time.perf_counter()
-            for i in range(script_args.warmup):
+            for _i in range(script_args.warmup):
                 sess._sess.run_with_iobinding(bind, None)
             warmup = time.perf_counter() - begin
 
             times = []
-            for i in range(script_args.repeat):
+            for _i in range(script_args.repeat):
                 begin = time.perf_counter()
                 # sess.run(None, feeds)
                 sess._sess.run_with_iobinding(bind, None)

_doc/examples/plot_op_transpose_2d_cast_cuda.py

@@ -194,13 +194,13 @@ def benchmark(
         bind, cuda_feeds = move_inputs(sess, feeds)
 
         begin = time.perf_counter()
-        for i in range(script_args.warmup):
+        for _i in range(script_args.warmup):
             # sess.run(None, feeds)
             sess._sess.run_with_iobinding(bind, None)
         warmup = time.perf_counter() - begin
 
         times = []
-        for i in range(script_args.repeat):
+        for _i in range(script_args.repeat):
             begin = time.perf_counter()
             # sess.run(None, feeds)
             sess._sess.run_with_iobinding(bind, None)
@@ -262,7 +262,10 @@ def benchmark(
     ax = pivot[["Not Fused", "Fused"]].plot(
         logx=True,
         logy=True,
-        title=f"Not Fused/Fused implementation for Transpose + Cast on CUDA\nitype={itype}",
+        title=(
+            f"Not Fused/Fused implementation for Transpose + "
+            f"Cast on CUDA\nitype={itype}"
+        ),
     )
     ax.get_figure().savefig("plot_op_transpose_2d_cast_cuda.png")
 

_doc/examples/plot_op_tree_ensemble_implementations.py

@@ -256,7 +256,7 @@ def compile_tree(
     compiler_options.SetMakeAllLeavesSameDepth(pipeline_width)
     compiler_options.SetReorderTreesByDepth(reorder_tree_by_depth)
     compiler_options.SetNumberOfFeatures(n_features)
-    assert 8 < batch_size
+    assert 8 < batch_size  # noqa: SIM300
     compiler_options.SetPipelineWidth(8)
 
     if verbose:
@@ -344,7 +344,9 @@ def make_ort_assembly_session(
 
     llc_exe = os.environ.get("TEST_LLC_EXE", "SKIP")
     if llc_exe == "SKIP":
-        warnings.warn("Unable to find environment variable 'TEST_LLC_EXE'.")
+        warnings.warn(
+            "Unable to find environment variable 'TEST_LLC_EXE'.", stacklevel=0
+        )
         return None
 
     filename = "plot_op_tree_ensemble_implementation.onnx"
@@ -383,9 +385,7 @@ def transform_model(model, use_sparse=False, **kwargs):
     onx = ModelProto()
     onx.ParseFromString(model.SerializeToString())
     att = get_node_attribute(onx.graph.node[0], "nodes_modes")
-    modes = ",".join(map(lambda s: s.decode("ascii"), att.strings)).replace(
-        "BRANCH_", ""
-    )
+    modes = ",".join([s.decode("ascii") for s in att.strings]).replace("BRANCH_", "")
     if use_sparse and "new_op_type" not in kwargs:
         kwargs["new_op_type"] = "TreeEnsembleRegressorSparse"
     if use_sparse:
@@ -538,7 +538,7 @@ def enumerate_implementations(
         disc = diff.mean()
         max_disc = diff.max()
     obs = measure_time(
-        lambda: sess.run(None, feeds),
+        lambda sess=sess, feeds=feeds: sess.run(None, feeds),
         repeat=script_args.repeat,
         number=script_args.number,
         warmup=script_args.warmup,

_doc/examples/plot_op_tree_ensemble_optim.py

@@ -27,7 +27,7 @@
         --max_depth=10
         --n_features=50
         --batch_size=100000
-    
+
 Another example with a full list of parameters:
 
     python plot_op_tree_ensemble_optim.py
@@ -49,7 +49,7 @@
 ::
 
     python plot_op_tree_ensemble_optim.py
-        --n_trees=100 --n_features=10 --batch_size=10000 --max_depth=8 -s SHORT        
+        --n_trees=100 --n_features=10 --batch_size=10000 --max_depth=8 -s SHORT
 """
 
 import logging
@@ -185,9 +185,7 @@ def transform_model(model, **kwargs):
     onx = ModelProto()
     onx.ParseFromString(model.SerializeToString())
     att = get_node_attribute(onx.graph.node[0], "nodes_modes")
-    modes = ",".join(map(lambda s: s.decode("ascii"), att.strings)).replace(
-        "BRANCH_", ""
-    )
+    modes = ",".join([s.decode("ascii") for s in att.strings]).replace("BRANCH_", "")
     return change_onnx_operator_domain(
         onx,
         op_type="TreeEnsembleRegressor",
@@ -307,12 +305,12 @@ def transform_model(model, **kwargs):
     )
 elif script_args.scenario == "CUSTOM":
     optim_params = dict(
-        parallel_tree=list(int(i) for i in script_args.parallel_tree.split(",")),
-        parallel_tree_N=list(int(i) for i in script_args.parallel_tree_N.split(",")),
-        parallel_N=list(int(i) for i in script_args.parallel_N.split(",")),
-        batch_size_tree=list(int(i) for i in script_args.batch_size_tree.split(",")),
-        batch_size_rows=list(int(i) for i in script_args.batch_size_rows.split(",")),
-        use_node3=list(int(i) for i in script_args.use_node3.split(",")),
+        parallel_tree=[int(i) for i in script_args.parallel_tree.split(",")],
+        parallel_tree_N=[int(i) for i in script_args.parallel_tree_N.split(",")],
+        parallel_N=[int(i) for i in script_args.parallel_N.split(",")],
+        batch_size_tree=[int(i) for i in script_args.batch_size_tree.split(",")],
+        batch_size_rows=[int(i) for i in script_args.batch_size_rows.split(",")],
+        use_node3=[int(i) for i in script_args.use_node3.split(",")],
     )
 else:
     raise ValueError(

_doc/examples/plot_op_tree_ensemble_sparse.py

@@ -22,7 +22,7 @@
         --n_features=50
         --sparsity=0.9
         --batch_size=100000
-    
+
 Another example with a full list of parameters:
 
     python plot_op_tree_ensemble_sparse.py
@@ -45,7 +45,7 @@
 ::
 
     python plot_op_tree_ensemble_sparse.py
-        --n_trees=100 --n_features=10 --batch_size=10000 --max_depth=8 -s SHORT        
+        --n_trees=100 --n_features=10 --batch_size=10000 --max_depth=8 -s SHORT
 """
 
 import logging
@@ -197,9 +197,7 @@ def transform_model(model, use_sparse=False, **kwargs):
     onx = ModelProto()
     onx.ParseFromString(model.SerializeToString())
     att = get_node_attribute(onx.graph.node[0], "nodes_modes")
-    modes = ",".join(map(lambda s: s.decode("ascii"), att.strings)).replace(
-        "BRANCH_", ""
-    )
+    modes = ",".join([s.decode("ascii") for s in att.strings]).replace("BRANCH_", "")
     if use_sparse and "new_op_type" not in kwargs:
         kwargs["new_op_type"] = "TreeEnsembleRegressorSparse"
     if use_sparse:
@@ -347,12 +345,12 @@ def transform_model(model, use_sparse=False, **kwargs):
     )
 elif script_args.scenario == "CUSTOM":
     optim_params = dict(
-        parallel_tree=list(int(i) for i in script_args.parallel_tree.split(",")),
-        parallel_tree_N=list(int(i) for i in script_args.parallel_tree_N.split(",")),
-        parallel_N=list(int(i) for i in script_args.parallel_N.split(",")),
-        batch_size_tree=list(int(i) for i in script_args.batch_size_tree.split(",")),
-        batch_size_rows=list(int(i) for i in script_args.batch_size_rows.split(",")),
-        use_node3=list(int(i) for i in script_args.use_node3.split(",")),
+        parallel_tree=[int(i) for i in script_args.parallel_tree.split(",")],
+        parallel_tree_N=[int(i) for i in script_args.parallel_tree_N.split(",")],
+        parallel_N=[int(i) for i in script_args.parallel_N.split(",")],
+        batch_size_tree=[int(i) for i in script_args.batch_size_tree.split(",")],
+        batch_size_rows=[int(i) for i in script_args.batch_size_rows.split(",")],
+        use_node3=[int(i) for i in script_args.use_node3.split(",")],
     )
 else:
     raise ValueError(