EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Remove annoying warnings in sgl kernel build (#9905)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-09-02 20:18:25 -07:00
committed by GitHub
parent 37565b7f21
commit d631290e32
5 changed files with 43 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
sgl-kernel/csrc/attention/cutlass_mla_kernel.cu
View File

@@ -162,7 +162,7 @@ typename T::Fmha::Arguments args_from_options(
// TODO(trevor-m): Change split_kv back to -1 when
// https://github.com/NVIDIA/cutlass/issues/2274 is fixed. Split_kv=1 will
// perform worse with larger context length and smaller batch sizes.
num_kv_splits, // split_kv
static_cast<int>(num_kv_splits), // split_kv
nullptr, // is_var_split_kv
};
// TODO(kaixih@nvidia): When split_kv=-1 and is_var_split_kv=false, we compute
@@ -259,7 +259,7 @@ int64_t cutlass_mla_get_workspace_size(int64_t max_seq_len, int64_t num_batches,
// Assumes device 0 when getting sm_count.
arguments.hw_info.sm_count =
sm_count <= 0 ? cutlass::KernelHardwareInfo::query_device_multiprocessor_count(/*device_id=*/0) : sm_count;
arguments.split_kv = num_kv_splits;
arguments.split_kv = static_cast<int>(num_kv_splits);
MlaSm100Type::Fmha::set_split_kv(arguments);
return MlaSm100Type::Fmha::get_workspace_size(arguments);