chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

Fix errors of page head kernels in sgl-kernel for ROCm (#12604)

Browse Source
This commit is contained in:
huangtingwei
2025-11-10 15:15:50 +08:00
committed by GitHub
parent ac19ce7efb
commit a5e5088dfb
1 changed files with 10 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
sgl-kernel/csrc/common_extension_rocm.cc
View File

@@ -165,6 +165,16 @@ TORCH_LIBRARY_EXPAND(sgl_kernel, m) {
"transfer_kv_all_layer_direct_lf_pf(Tensor[] src_ptrs, Tensor[] dst_ptrs, Tensor src_indices, "
"Tensor dst_indices, int page_size) ->() ");
m.impl("transfer_kv_all_layer_direct_lf_pf", torch::kCUDA, &transfer_kv_all_layer_direct_lf_pf);
m.def(
"transfer_kv_all_layer_lf_ph(Tensor src_k_layers, Tensor dst_k, Tensor src_v_layers, Tensor dst_v, "
"Tensor src_indices, Tensor dst_indices, int item_size, int dst_layout_dim, int num_layers, int page_size, int "
"head_num, int block_quota, int num_warps_per_block) -> ()");
m.impl("transfer_kv_all_layer_lf_ph", torch::kCUDA, &transfer_kv_all_layer_lf_ph);
m.def(
"transfer_kv_per_layer_ph_lf(Tensor src_k, Tensor dst_k, Tensor src_v, Tensor dst_v, Tensor src_indices, Tensor "
"dst_indices, int layer_id, int item_size, int src_layout_dim, int page_size, int head_num, int block_quota, int "
"num_warps_per_block) -> ()");
m.impl("transfer_kv_per_layer_ph_lf", torch::kCUDA, &transfer_kv_per_layer_ph_lf);
/*
* From csrc/grammar