Skip to content

Add Sycl topk per row kernel #191

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wuxun-zhang wants to merge 5 commits into
base: main
Choose a base branch
from
Open

Add Sycl topk per row kernel #191

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
17369a2
[WIP] topk per row prefill/decode
wuxun-zhang Feb 2, 2026
1c0c8a5
Add topk per row prefill/decode impl
YangQun1 Feb 3, 2026
fcbdc17
Fix accuracy issue when context length > 200000
YangQun1 Feb 6, 2026
b0edecd
updates
wuxun-zhang Mar 12, 2026
239376b
remove vllm deps and rename cuda to xpu in test
YangQun1 Mar 12, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -366,7 +366,8 @@ if(BASIC_KERNELS_ENABLED)
"csrc/quantization/fp8/fp8_quant.cpp"
"csrc/xpu_view.cpp"
"csrc/tensor_utils.cpp"
"csrc/utils/mem_cpy.cpp")
"csrc/utils/mem_cpy.cpp"
"csrc/topk_per_row.cpp")
include_directories("/usr/include")

define_gpu_extension_target(
Expand Down
20 changes: 20 additions & 0 deletions csrc/ops.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -119,3 +119,23 @@ void swap_blocks(
torch::Tensor& dst,
int64_t block_size_in_bytes,
const torch::Tensor& block_mapping);

void top_k_per_row_decode(
const torch::Tensor& logits,
int64_t next_n,
const torch::Tensor& seqLens,
torch::Tensor& indices,
int64_t numRows,
int64_t stride0,
int64_t stride1,
int64_t topK);

void top_k_per_row_prefill(
const torch::Tensor& logits,
const torch::Tensor& rowStarts,
const torch::Tensor& rowEnds,
torch::Tensor& indices,
int64_t numRows,
int64_t stride0,
int64_t stride1,
int64_t topK);
Loading