Skip to content

[Contrib] Fix CUDA contrib build after FFI/header cleanups #19539

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

Merged
tlopex merged 1 commit into from
May 12, 2026
Merged

[Contrib] Fix CUDA contrib build after FFI/header cleanups #19539

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 2 additions & 0 deletions src/runtime/contrib/cutlass/fp16_group_gemm_runner_sm100.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,8 @@
* under the License.
*/

#include <tvm/runtime/logging.h>

#include <fstream>
#include <iostream>
#include <sstream>
Expand Down
2 changes: 2 additions & 0 deletions src/runtime/contrib/cutlass/fp16_group_gemm_runner_sm90.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,8 @@
* under the License.
*/

#include <tvm/runtime/logging.h>

#include <fstream>
#include <iostream>
#include <sstream>
Expand Down
1 change: 1 addition & 0 deletions src/runtime/contrib/cutlass/fp8_groupwise_scaled_gemm.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@
#include <float.h>
#include <tvm/ffi/extra/c_env_api.h>
#include <tvm/ffi/function.h>
#include <tvm/runtime/logging.h>
#include <tvm/runtime/tensor.h>

#include "cutlass/bfloat16.h"
Expand Down
2 changes: 2 additions & 0 deletions src/runtime/contrib/cutlass/gemm_runner.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,8 @@
* under the License.
*/

#include <tvm/runtime/logging.h>

#include <fstream>
#include <iostream>
#include <sstream>
Expand Down
1 change: 1 addition & 0 deletions src/runtime/contrib/nvshmem/init.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@
#include <tvm/ffi/function.h>
#include <tvm/ffi/reflection/registry.h>
#include <tvm/runtime/disco/disco_worker.h>
#include <tvm/runtime/logging.h>

#include "../../cuda/cuda_common.h"

Expand Down
1 change: 1 addition & 0 deletions src/runtime/contrib/thrust/thrust.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,7 @@
#include <tvm/ffi/extra/c_env_api.h>
#include <tvm/ffi/function.h>
#include <tvm/ffi/reflection/registry.h>
#include <tvm/runtime/logging.h>

#include <algorithm>
#include <functional>
Expand Down
49 changes: 37 additions & 12 deletions src/runtime/contrib/vllm/attention_kernels.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,6 +37,7 @@
#include <float.h>
#include <tvm/ffi/function.h>
#include <tvm/ffi/reflection/registry.h>
#include <tvm/runtime/logging.h>
#include <tvm/runtime/tensor.h>

#include <algorithm>
Expand Down Expand Up @@ -756,10 +757,10 @@ TVM_FFI_STATIC_INIT_BLOCK() {
namespace refl = tvm::ffi::reflection;
refl::GlobalDef().def(
"tvm.contrib.vllm.single_query_cached_kv_attention",
[](const DLTensor* query, const DLTensor* key_cache, const DLTensor* value_cache,
const DLTensor* block_tables, const DLTensor* context_lens, int block_size,
const DLTensor* max_context_len_tensor, // TODO(masahi): pass integer
DLTensor* exp_sums, DLTensor* max_logits, DLTensor* tmp_out, DLTensor* out) {
[](Tensor query, Tensor key_cache, Tensor value_cache, Tensor block_tables,
Tensor context_lens, int block_size,
Tensor max_context_len_tensor, // TODO(masahi): pass integer
Tensor exp_sums, Tensor max_logits, Tensor tmp_out, Tensor out) {
Comment thread
MasterJH5574 marked this conversation as resolved.
int num_seqs = query->shape[0];
int num_heads = query->shape[1];
int max_context_len = static_cast<int*>(max_context_len_tensor->data)[0];
Expand All @@ -768,13 +769,19 @@ TVM_FFI_STATIC_INIT_BLOCK() {
bool use_v1 =
max_context_len <= 8192 && (max_num_partitions == 1 || num_seqs * num_heads > 512);
if (use_v1) {
single_query_cached_kv_attention_v1(query, key_cache, value_cache, block_tables,
context_lens, block_size, max_context_len_tensor,
out);
single_query_cached_kv_attention_v1(
query.GetDLTensorPtr(), key_cache.GetDLTensorPtr(), value_cache.GetDLTensorPtr(),
block_tables.GetDLTensorPtr(), context_lens.GetDLTensorPtr(), block_size,
max_context_len_tensor.GetDLTensorPtr(), const_cast<DLTensor*>(out.GetDLTensorPtr()));
Comment thread
MasterJH5574 marked this conversation as resolved.
} else {
single_query_cached_kv_attention_v2(query, key_cache, value_cache, block_tables,
context_lens, block_size, max_context_len_tensor,
exp_sums, max_logits, tmp_out, out);
single_query_cached_kv_attention_v2(
query.GetDLTensorPtr(), key_cache.GetDLTensorPtr(), value_cache.GetDLTensorPtr(),
block_tables.GetDLTensorPtr(), context_lens.GetDLTensorPtr(), block_size,
max_context_len_tensor.GetDLTensorPtr(),
const_cast<DLTensor*>(exp_sums.GetDLTensorPtr()),
const_cast<DLTensor*>(max_logits.GetDLTensorPtr()),
const_cast<DLTensor*>(tmp_out.GetDLTensorPtr()),
const_cast<DLTensor*>(out.GetDLTensorPtr()));
}
});
}
Expand All @@ -784,9 +791,27 @@ TVM_FFI_STATIC_INIT_BLOCK() {
namespace refl = tvm::ffi::reflection;
refl::GlobalDef()
.def("tvm.contrib.vllm.single_query_cached_kv_attention_v1",
single_query_cached_kv_attention_v1)
[](Tensor query, Tensor key_cache, Tensor value_cache, Tensor block_tables,
Tensor context_lens, int block_size, Tensor max_context_len_tensor, Tensor out) {
Comment thread
MasterJH5574 marked this conversation as resolved.
single_query_cached_kv_attention_v1(
query.GetDLTensorPtr(), key_cache.GetDLTensorPtr(), value_cache.GetDLTensorPtr(),
block_tables.GetDLTensorPtr(), context_lens.GetDLTensorPtr(), block_size,
max_context_len_tensor.GetDLTensorPtr(),
const_cast<DLTensor*>(out.GetDLTensorPtr()));
})
.def("tvm.contrib.vllm.single_query_cached_kv_attention_v2",
single_query_cached_kv_attention_v2);
[](Tensor query, Tensor key_cache, Tensor value_cache, Tensor block_tables,
Tensor context_lens, int block_size, Tensor max_context_len_tensor, Tensor exp_sums,
Tensor max_logits, Tensor tmp_out, Tensor out) {
Comment thread
MasterJH5574 marked this conversation as resolved.
single_query_cached_kv_attention_v2(
query.GetDLTensorPtr(), key_cache.GetDLTensorPtr(), value_cache.GetDLTensorPtr(),
block_tables.GetDLTensorPtr(), context_lens.GetDLTensorPtr(), block_size,
max_context_len_tensor.GetDLTensorPtr(),
const_cast<DLTensor*>(exp_sums.GetDLTensorPtr()),
const_cast<DLTensor*>(max_logits.GetDLTensorPtr()),
const_cast<DLTensor*>(tmp_out.GetDLTensorPtr()),
const_cast<DLTensor*>(out.GetDLTensorPtr()));
});
}

} // namespace runtime
Expand Down
4 changes: 2 additions & 2 deletions src/runtime/contrib/vllm/cache_kernels.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -154,7 +154,7 @@ TVM_FFI_STATIC_INIT_BLOCK() {
static_cast<const int*>(slot_mapping->data), key_stride, value_stride, num_heads,
head_size, block_size, vec_size);

return Array{key_cache, value_cache};
return ffi::Array<Tensor>{key_cache, value_cache};
})
.def("tvm.contrib.vllm.reconstruct_from_cache",
[](Tensor key_cache, Tensor value_cache, Tensor slot_mapping) {
Expand Down Expand Up @@ -182,7 +182,7 @@ TVM_FFI_STATIC_INIT_BLOCK() {
static_cast<scalar_t*>(value->data), key_stride, value_stride, num_heads,
head_size, block_size, vec_size);

return Array{key, value};
return ffi::Array<Tensor>{key, value};
})
.def("tvm.contrib.vllm.copy_blocks", [](ffi::Array<Tensor> key_value_caches,
Tensor block_mapping) {
Expand Down
Loading