[Kernel] Add KDA NVFP4 GEMM for Qwen3.x on SM120 (#36865)

Co-authored-by: Song Bian <biansonghz@gmail.com>
Co-authored-by: Cursor <cursoragent@cursor.com>

test/registered/kernels/ops/diffusion/test_import_surface.py

@@ -49,10 +49,15 @@ def _module_defines(module_path: str) -> set[str]:
     Importing would pull in Triton / CuTe-DSL / FlyDSL, none of which are
     installed on the CPU CI lane -- so this reads the source instead.
     """
-    path = _PACKAGE_DIR / (module_path.replace(".", "/") + ".py")
-    if not path.exists():
-        path = _PACKAGE_DIR / module_path.replace(".", "/") / "__init__.py"
-    assert path.exists(), f"{PACKAGE}.{module_path} does not exist"
+    if module_path.startswith("sglang."):
+        spec = importlib.util.find_spec(module_path)
+        assert spec is not None and spec.origin is not None, module_path
+        path = pathlib.Path(spec.origin)
+    else:
+        path = _PACKAGE_DIR / (module_path.replace(".", "/") + ".py")
+        if not path.exists():
+            path = _PACKAGE_DIR / module_path.replace(".", "/") / "__init__.py"
+        assert path.exists(), f"{PACKAGE}.{module_path} does not exist"
 
     names: set[str] = set()
     for node in ast.parse(path.read_text(encoding="utf-8")).body:
@@ -85,7 +90,12 @@ def _scan_root(root: str) -> tuple[frozenset[str], tuple[str, ...]]:
 
     for path in root_dir.rglob("*.py"):
         rel = path.relative_to(_REPO_ROOT).as_posix()
-        if rel.startswith("python/sglang/kernels/ops/diffusion/"):
+        if rel.startswith(
+            (
+                "python/sglang/kernels/ops/diffusion/",
+                "python/sglang/kernels/kda_kernels/",
+            )
+        ):
             continue
         try:
             source = path.read_text(encoding="utf-8")

test/registered/kernels/ops/layernorm/test_kernels_namespace.py

@@ -43,6 +43,7 @@ EXPECTED = {
     "diffusion.flux2_layernorm_modulate_fp8_quant": {"KDA"},
     "diffusion.flux2_qkv_epilogue": {"KDA"},
     "diffusion.flux2_token_cat_fp8": {"KDA"},
+    "gemm.qwen3x_nvfp4": {"KDA"},
 }
 
 _CPU = PlatformInfo(device_type="cpu")
@@ -120,6 +121,14 @@ def test_merged_diffusion_kda_provenance_backend(op, target_suffix):
     assert spec.target.endswith(target_suffix)
 
 
+def test_kda_backend_implementations_live_in_kda_home():
+    specs = [
+        spec for spec in K.registry.all_specs() if spec.backend is KernelBackend.KDA
+    ]
+    assert specs
+    assert all(spec.target.startswith("sglang.kernels.kda_kernels.") for spec in specs)
+
+
 def test_single_backend_resolves_without_backend():
     assert (
         K.select_kernel("kvcache.reshape_and_cache_flash").backend