dsv4.1: Top-k kernels and candidate selection (#39648)

Co-authored-by: BBuf <1182563586@qq.com>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: DarkSharpness <2040703891@qq.com>
Co-authored-by: DarkSharpness <76582120+DarkSharpness@users.noreply.github.com>
Co-authored-by: Ke Bao <ispobaoke@gmail.com>
Co-authored-by: Khoa Pham <khoa.pham@radixark.ai>
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
Co-authored-by: Xiaoyu Zhang <xiaoyu.zhang@radixark.ai>
Co-authored-by: Yuhao Yang <47235274+yhyang201@users.noreply.github.com>
Co-authored-by: Yuwei An <ayw.sirius19@gmail.com>
Co-authored-by: Zhichen Zeng <zczeng@uw.edu>
Co-authored-by: Ziyi Xu <ziyi.xu@radixark.ai>

python/sglang/test/kits/dsa_metadata_kit.py

@@ -5,6 +5,7 @@ from types import SimpleNamespace
 
 import torch
 
+from sglang.kernels.ops.attention.dsv4.topk import topk_v2_plan_is_written
 from sglang.srt.environ import envs
 from sglang.srt.layers.attention.dsa.dsa_topk_backend import DSATopKBackend
 from sglang.srt.layers.attention.dsa_backend import DeepseekSparseAttnBackend
@@ -101,6 +102,8 @@ def assert_metadata_equal(test, actual, expected):
     for name, value in actual_buffers.items():
         reference = expected_buffers[name]
         if name == "topk_v2_plan":
+            if not topk_v2_plan_is_written(expected.dsa_seqlens_expanded):
+                continue
             # Unused plan rows are intentionally uninitialized. Active rows are
             # compacted by atomicAdd, so compare them in request order.
             torch.testing.assert_close(value[0], reference[0])