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

[sgl-kernel] fix: fix cu118 compile error (#6123)

Browse Source 
Co-authored-by: zhyncs <me@zhyncs.com>
This commit is contained in:
PGFLMG
2025-05-09 05:26:51 +08:00
committed by GitHub
parent 2a936a841e
commit f6f96b0521
2 changed files with 22 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
sgl-kernel/csrc/attention/cutlass_mla_kernel.cu
View File

@@ -25,7 +25,21 @@ limitations under the License.
#include <device/sm100_mla.hpp>
#include <kernel/sm100_mla_tile_scheduler.hpp>
#if defined CUDA_VERSION && CUDA_VERSION >= 12040
// clang-format off
#if !defined(CUDA_VERSION) || CUDA_VERSION < 12040
void cutlass_mla_decode(
torch::Tensor const& out,
torch::Tensor const& q_nope_and_q_pe,
torch::Tensor const& kv_c_and_k_pe_cache,
torch::Tensor const& seq_lens,
torch::Tensor const& page_table,
torch::Tensor const& workspace) {
TORCH_CHECK(false, "CUDA version must be >= 12.4 for cutlass_mla_decode");
}
int64_t cutlass_mla_get_workspace_size(int64_t max_seq_len, int64_t num_batches, int64_t sm_count) {
TORCH_CHECK(false, "CUDA version must be >= 12.4 for cutlass_mla_get_workspace_size");
}
#else
#define CUTLASS_CHECK(status) \
{ \
@@ -209,3 +223,4 @@ int64_t cutlass_mla_get_workspace_size(int64_t max_seq_len, int64_t num_batches,
}
#endif
// clang-format on