[CP V1 Deprecation 4/5] Canonicalize prefill CP API names (#36229)

python/sglang/kernels/ops/attention/__init__.py

@@ -181,7 +181,7 @@ for _mod, _fn in [
     ("dsa.transform_index", "transform_index_page_table_prefill"),
     ("dsa.transform_index", "transform_index_page_table_decode"),
     ("dsa.transform_index", "prepare_trtllm_nope_sparse_metadata"),
-    ("dsa.cp_split", "dsa_cp_round_robin_split_q_seqs_kernel"),
+    ("dsa.cp_split", "dsa_cp_interleave_q_seqs_kernel"),
     ("dsv4.fp4_indexer", "quantize_fp4_indexer_tensor"),
     ("dsv4.fp4_indexer", "store_fp4_index_k_cache"),
     ("dsv4.rms_normalize_hip", "rms_normalize_triton"),

python/sglang/kernels/ops/attention/dsa/cp_split.py

@@ -1,4 +1,4 @@
-"""Round-robin CP q-sequence split kernel for DSA prefill.
+"""Interleave CP q-sequence split kernel for DSA prefill.
 
 Migrated from ``sglang.srt.layers.attention.dsa.utils`` (RFC #29630, Phase 2.5).
 """
@@ -8,7 +8,7 @@ import triton.language as tl
 
 
 @triton.jit
-def dsa_cp_round_robin_split_q_seqs_kernel(
+def dsa_cp_interleave_q_seqs_kernel(
     in_seqs_ptr,
     out_seqs_ptr,
     bs_idx_ptr,

python/sglang/kernels/ops/attention/dsv4/metadata_kernel.py

@@ -106,7 +106,7 @@ def _init_compressed_attn_metadata_triton(
     Optional[torch.Tensor],
 ]:
     bs = seq_lens.shape[0]
-    # CP-v2 may add padding rows to the attention metadata, but those rows have
+    # CP may add padding rows to the attention metadata, but those rows have
     # no cache-write locations. Keep the write buffers unpadded and mask those
     # rows in the kernel.
     num_write_tokens = raw_out_loc.shape[0]

python/sglang/kernels/ops/layernorm/mhc.py

@@ -15,7 +15,7 @@ from sglang.srt.distributed.device_communicators.pynccl_allocator import (
 )
 from sglang.srt.distributed.parallel_state import get_tp_group
 from sglang.srt.environ import envs
-from sglang.srt.layers.attention.dsa.utils import is_dsa_prefill_cp_round_robin_split
+from sglang.srt.layers.attention.dsa.utils import is_dsa_prefill_cp_interleave
 from sglang.srt.layers.dp_attention import is_allocation_symmetric
 from sglang.srt.layers.utils.common import strict_contiguous
 from sglang.srt.utils.common import is_gfx1250_supported
@@ -1240,7 +1240,7 @@ def mhc_post(
     post_layer_mix: torch.Tensor,
     comb_res_mix: torch.Tensor,
 ) -> torch.Tensor:
-    if is_dsa_prefill_cp_round_robin_split():
+    if is_dsa_prefill_cp_interleave():
         x = strict_contiguous(x)
         residual = strict_contiguous(residual)
         post_layer_mix = strict_contiguous(post_layer_mix)