fix ufmt

Author: xuzhao9

benchmarks/tagging/run.py

@@ -133,7 +133,9 @@ def merge_decorator_tags(op_name, backend_name, tags_dict):
 
     # Get decorator tags if they exist
     backend_config = REGISTERED_BENCHMARKS.get(op_name, {}).get(backend_name)
-    decorator_tags = backend_config.tags if (backend_config and backend_config.tags) else []
+    decorator_tags = (
+        backend_config.tags if (backend_config and backend_config.tags) else []
+    )
     if decorator_tags:
         # Merge decorator tags with auto-detected tags (remove duplicates)
         all_tags = list(set(decorator_tags + tags_dict["tags"]))

benchmarks/tritonparse_sweep/run.py

@@ -41,10 +41,10 @@ def setup_tritonbench_cwd():
 setup_tritonbench_cwd()
 
 import tritonparse
-from tritonparse.reproducer.orchestrator import reproduce as tritonparse_reproduce
-from tritonparse.reproducer.types import KernelImportMode
 from tritonbench.operators_collection import list_operators_by_collection
 from tritonbench.utils.run_utils import run_in_task, setup_output_dir
+from tritonparse.reproducer.orchestrator import reproduce as tritonparse_reproduce
+from tritonparse.reproducer.types import KernelImportMode
 
 NOT_WORKING_OPS = ["tritonparse_softmax_triton_softmax"]
 
@@ -93,12 +93,12 @@ def find_ndjson_files(log_dir):
 
 
 def find_reproducer_script(output: str):
-    output_line: list[str] = [ x for x in output.splitlines() if "repro_script" in x ]
+    output_line: list[str] = [x for x in output.splitlines() if "repro_script" in x]
     if len(output_line) == 0:
         return None
-    output_line = output_line[0][output_line[0].find("{"):].strip()
+    output_line = output_line[0][output_line[0].find("{") :].strip()
     output_dict = eval(output_line)
-    return output_dict['repro_script']
+    return output_dict["repro_script"]
 
 
 def run_repro_script(repro_script):

tritonbench/operators/launch_latency/cutedsl.py

@@ -1,14 +1,17 @@
 import cutlass
 import cutlass.cute as cute
 
+
 @cute.kernel
 def nop_kernel():
     pass
 
+
 @cute.jit
 def cutedsl_nop_kernel():
     nop_kernel()
 
+
 @cute.kernel
 def nop_with_args_kernel(
     t1: cute.Tensor,
@@ -33,6 +36,7 @@ def nop_with_args_kernel(
 ):
     pass
 
+
 @cute.jit
 def cutedsl_nop_with_args_kernel(
     t1: cute.Tensor,
@@ -53,5 +57,8 @@ def cutedsl_nop_with_args_kernel(
     c2: cutlass.Constexpr,
     c3: cutlass.Constexpr,
     c4: cutlass.Constexpr,
-    c5: cutlass.Constexpr,):
-    nop_with_args_kernel(t1, t2, t3, t4, t5, i1, i2, i3, i4, i5, i6, i7, i8, i9, c1, c2, c3, c4, c5)
+    c5: cutlass.Constexpr,
+):
+    nop_with_args_kernel(
+        t1, t2, t3, t4, t5, i1, i2, i3, i4, i5, i6, i7, i8, i9, c1, c2, c3, c4, c5
+    )

tritonbench/operators/launch_latency/operator.py

@@ -3,19 +3,18 @@
 
 from torch._inductor.utils import triton_version_uses_attrs_dict
 from triton.compiler import CompiledKernel
-
-from tritonbench.utils.triton_op import (
-    BenchmarkOperator,
-    register_benchmark,
-)
 from tritonbench.utils.python_utils import try_import
 
+from tritonbench.utils.triton_op import BenchmarkOperator, register_benchmark
+
 with try_import("HAS_TILELANG"):
     import tilelang
+
     from .tilelang import tilelang_nop_kernel, tilelang_nop_with_args_kernel
 
 with try_import("HAS_CUTEDSL"):
     import cutlass.cute as cute
+
     from .cutedsl import cutedsl_nop_kernel, cutedsl_nop_with_args_kernel
 
 from .kernels import get_trivial_add_kernel, nop_kernel, nop_with_args_kernel
@@ -105,7 +104,9 @@ def nop_cutedsl_tvm_ffi(self, *args):
                 cute_args.append(cute.runtime.from_dlpack(arg, enable_tvm_ffi=True))
             else:
                 cute_args.append(arg)
-        kernel = cute.compile(cutedsl_nop_with_args_kernel, *cute_args, options="--enable-tvm-ffi")
+        kernel = cute.compile(
+            cutedsl_nop_with_args_kernel, *cute_args, options="--enable-tvm-ffi"
+        )
         # remove constexpr args
         cute_args = cute_args[:-5]
         return lambda: kernel(*cute_args)

tritonbench/operators/launch_latency/tilelang.py

@@ -1,13 +1,14 @@
 import tilelang
 import tilelang.language as T
 
+
 @tilelang.jit()
 def tilelang_nop_kernel():
-
     @T.prim_func
     def nop_kernel():
         with T.Kernel(1) as bx:
             pass
+
     return nop_kernel
 
 
@@ -37,4 +38,5 @@ def nop_with_args_kernel(
     ):
         with T.Kernel(1) as bx:
             pass
+
     return nop_with_args_kernel