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

[hicache] Optimization for DMA copy (#8245)

Browse Source
This commit is contained in:
cctry
2025-08-09 17:16:07 -07:00
committed by GitHub
parent ef48d5547e
commit 5c31b35db2
2 changed files with 26 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/srt/managers/cache_controller.py
View File

@@ -433,7 +433,9 @@ class HiCacheController:
if self.io_backend == "kernel": if self.io_backend == "kernel":
return host_indices.to(self.mem_pool_device.device), device_indices return host_indices.to(self.mem_pool_device.device), device_indices
elif self.io_backend == "direct": elif self.io_backend == "direct":
return host_indices, device_indices.cpu() device_indices = device_indices.cpu()
host_indices, idx = host_indices.sort()
return host_indices, device_indices.index_select(0, idx)
else: else:
raise ValueError(f"Unsupported io backend") raise ValueError(f"Unsupported io backend")

28
sgl-kernel/csrc/kvcacheio/transfer.cu
View File

@@ -451,15 +451,33 @@ void transfer_kv_direct(
auto src_indices_cpu = src_indices.cpu(); auto src_indices_cpu = src_indices.cpu();
auto dst_indices_cpu = dst_indices.cpu(); auto dst_indices_cpu = dst_indices.cpu();
const int64_t num_pages = src_indices_cpu.size(0) / page_size; const auto num_indices = src_indices_cpu.numel();
const int64_t num_layers = src_layers.size(); const int64_t num_layers = src_layers.size();
int64_t* src_indices_ptr = src_indices_cpu.data_ptr<int64_t>();
int64_t* dst_indices_ptr = dst_indices_cpu.data_ptr<int64_t>();
for (int64_t i = 0; i < num_pages; ++i) { int64_t start_index = 0;
auto src_index = src_indices_cpu[i * page_size].item<int64_t>(); int64_t end_index = 0;
auto dst_index = dst_indices_cpu[i * page_size].item<int64_t>();
for (int64_t i = 0; i < num_indices; ++i) {
if (i < num_indices - 1) {
auto src_diff = src_indices_ptr[i + 1] - src_indices_ptr[i];
auto dst_diff = dst_indices_ptr[i + 1] - dst_indices_ptr[i];
if (src_diff == 1 && dst_diff == 1) {
continue;
}
end_index = i + 1;
} else { // last batch
end_index = num_indices;
}
auto src_index = src_indices_ptr[start_index];
auto dst_index = dst_indices_ptr[start_index];
auto num_tokens = end_index - start_index;
for (int64_t j = 0; j < num_layers; ++j) { for (int64_t j = 0; j < num_layers; ++j) {
transfer_page_direct(src_layers[j], dst_layers[j], src_index, dst_index, page_size); transfer_page_direct(src_layers[j], dst_layers[j], src_index, dst_index, num_tokens);
} }
start_index = end_index;
} }
} }