[Refactor] Remove dead cuda kernels (#42767)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-05-18 11:14:21 -07:00
committed by GitHub
parent 9758a6e5c5
commit 00e20e76f7
6 changed files with 1 additions and 605 deletions
-26
View File
@@ -60,32 +60,6 @@ void merge_attn_states(
const torch::Tensor& suffix_output, const torch::Tensor& suffix_lse,
const std::optional<int64_t> prefill_tokens_with_context,
const std::optional<torch::Tensor>& output_scale = std::nullopt);
#ifndef USE_ROCM
void convert_vertical_slash_indexes(
torch::Tensor& block_count, // [BATCH, N_HEADS, NUM_ROWS]
torch::Tensor& block_offset, // [BATCH, N_HEADS, NUM_ROWS, NNZ_S]
torch::Tensor& column_count, // [BATCH, N_HEADS, NUM_ROWS]
torch::Tensor& column_index, // [BATCH, N_HEADS, NUM_ROWS, NNZ_V]
torch::Tensor q_seqlens, // [BATCH, ]
torch::Tensor kv_seqlens, // [BATCH, ]
torch::Tensor vertical_indexes, // [BATCH, N_HEADS, NNZ_V]
torch::Tensor slash_indexes, // [BATCH, N_HEADS, NNZ_S]
int64_t context_size, int64_t block_size_M, int64_t block_size_N,
bool causal);
void convert_vertical_slash_indexes_mergehead(
torch::Tensor& block_count, // [BATCH, N_HEADS, NUM_ROWS]
torch::Tensor& block_offset, // [BATCH, N_HEADS, NUM_ROWS, NNZ_S]
torch::Tensor& column_count, // [BATCH, N_HEADS, NUM_ROWS]
torch::Tensor& column_index, // [BATCH, N_HEADS, NUM_ROWS, NNZ_V]
torch::Tensor q_seqlens, // [BATCH, ]
torch::Tensor kv_seqlens, // [BATCH, ]
torch::Tensor vertical_indexes, // [BATCH, N_HEADS, NNZ_V]
torch::Tensor slash_indexes, // [BATCH, N_HEADS, NNZ_S]
torch::Tensor vertical_indices_count, // [N_HEADS, ]
torch::Tensor slash_indices_count, int64_t context_size,
int64_t block_size_M, int64_t block_size_N, bool causal);
#endif
void rms_norm(torch::Tensor& out, torch::Tensor& input, torch::Tensor& weight,
double epsilon);