Support copying tensor from cpu to gpu without using copy engines (#10007)

2025-09-05 20:07:19 +08:00
parent 5e5c30d9ab
commit bd7f882142
6 changed files with 70 additions and 1 deletions
--- a/sgl-kernel/include/sgl_kernel_ops.h
+++ b/sgl-kernel/include/sgl_kernel_ops.h
@@ -750,3 +750,5 @@ std::vector<int64_t> create_greenctx_stream_by_value(int64_t smA, int64_t smB, i
 * From csrc/memory
 */
 void store_kv_cache(at::Tensor k_cache, at::Tensor v_cache, at::Tensor out_loc, at::Tensor k, at::Tensor v);
+
+void copy_to_gpu_no_ce(const at::Tensor& input, at::Tensor& output);