zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-10-19 16:48:10 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

[CUDA] Update calls to cudaMemAdvise and cudaGraphAddDependencies for CUDA 13 (#2525)

Browse Source 
* [CUDA] Update cudaMemAdvise and cudaGraphAddDependencies for CUDA 13

These functions' signatures changed in CUDA 13, so we differentiate
between CUDA 13 and preceding releases at compile time.

* Mention NVIDIA in ACKNOWLEDGMENTS.md
This commit is contained in:
Andrey Portnoy
2025-08-21 22:57:20 -04:00
committed by GitHub
parent f6819a1f26
commit 5722c147de
3 changed files with 20 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
mlx/backend/cuda/allocator.cpp
View File

@@ -30,8 +30,15 @@ SmallSizePool::SmallSizePool() {
next_free_ = buffer_;
CHECK_CUDA_ERROR(cudaMallocManaged(&data_, small_pool_size));
#if CUDART_VERSION >= 13000
cudaMemLocation loc;
loc.type = cudaMemLocationTypeDevice;
loc.id = 0;
#else
int loc = 0;
#endif // CUDART_VERSION >= 13000
CHECK_CUDA_ERROR(
cudaMemAdvise(data_, small_pool_size, cudaMemAdviseSetReadMostly, 0));
cudaMemAdvise(data_, small_pool_size, cudaMemAdviseSetReadMostly, loc));
auto curr = next_free_;
for (size_t i = 1; i < num_blocks; ++i) {

8
mlx/backend/cuda/device.cpp
View File

@@ -269,7 +269,13 @@ void CommandEncoder::commit() {
if (node_count_ > 0) {
if (!from_nodes_.empty()) {
CHECK_CUDA_ERROR(cudaGraphAddDependencies(
graph_, from_nodes_.data(), to_nodes_.data(), from_nodes_.size()));
graph_,
from_nodes_.data(),
to_nodes_.data(),
#if CUDART_VERSION >= 13000
nullptr, // edgeData
#endif // CUDART_VERSION >= 13000
from_nodes_.size()));
}
graph_key_ += ".";