Skip to content

Detect NVSHMEM static heap fallback to enable symmetric heap size hints. #164

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
romerojosh merged 3 commits into from
Oct 7, 2026
Merged

Detect NVSHMEM static heap fallback to enable symmetric heap size hints. #164

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c1ef5c2
Detect NVSHMEM static heap fallback to enable symmetric heap size hints.
romerojosh Oct 7, 2026
7e3fb3b
Formatting.
romerojosh Oct 7, 2026
2b14c96
Add CUDA version guards.
romerojosh Oct 7, 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
22 changes: 22 additions & 0 deletions src/cudecomp.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -750,6 +750,28 @@ static void inspectNvshmemEnvVars(nvshmemRuntimeState& runtime) {
char* vmm_str = std::getenv("NVSHMEM_DISABLE_CUDA_VMM");
if (vmm_str) { runtime.nvshmem_vmm = std::strtol(vmm_str, nullptr, 10) == 0; }

if (runtime.nvshmem_vmm) {
#if CUDART_VERSION >= 11030
int dev;
CUdevice cu_dev;
CHECK_CUDA(cudaGetDevice(&dev));
CHECK_CUDA_DRV(cuDeviceGet(&cu_dev, dev));

int vmm_supported = 0;
int rdma_vmm_supported = 0;
CHECK_CUDA_DRV(
cuDeviceGetAttribute(&vmm_supported, CU_DEVICE_ATTRIBUTE_VIRTUAL_MEMORY_MANAGEMENT_SUPPORTED, cu_dev));
CHECK_CUDA_DRV(
cuDeviceGetAttribute(&rdma_vmm_supported, CU_DEVICE_ATTRIBUTE_GPU_DIRECT_RDMA_WITH_CUDA_VMM_SUPPORTED, cu_dev));
// NVSHMEM falls back to a static heap if either capability is unavailable.
// Passing these checks establishes VMM eligibility, not the active heap mode.
runtime.nvshmem_vmm = vmm_supported && rdma_vmm_supported;
#else
// NVSHMEM's VMM heap requires CUDA 11.3 or newer.
runtime.nvshmem_vmm = false;
#endif
}

// Check NVSHMEM_SYMMETRIC_SIZE
char* symmetric_size_str = std::getenv("NVSHMEM_SYMMETRIC_SIZE");
if (symmetric_size_str) {
Expand Down
Loading