Page first direct IO kernel (#10060)

Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
This commit is contained in:
huangtingwei
2025-09-10 13:35:34 +08:00
committed by GitHub
parent 737d73ed5b
commit 5be8c2f7f7
5 changed files with 358 additions and 2 deletions

View File

@@ -569,6 +569,21 @@ void transfer_kv_direct(
const at::Tensor dst_indices,
int64_t page_size);
void transfer_kv_per_layer_direct_pf_lf(
const std::vector<at::Tensor>& src_ptrs,
std::vector<at::Tensor> dst_ptrs,
const at::Tensor& src_indices,
const at::Tensor& dst_indices,
int64_t layer_id,
int64_t page_size);
void transfer_kv_all_layer_direct_lf_pf(
const std::vector<at::Tensor>& src_ptrs,
std::vector<at::Tensor> dst_ptrs,
const at::Tensor& src_indices,
const at::Tensor& dst_indices,
int64_t page_size);
/*
* From FlashInfer
*/