Skip to content

auto-detect GPU arch and adapt XSched preemption level, remove hardco… #1

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

Open
kylin1019 wants to merge 1 commit into
base: xsched
Choose a base branch
from
+17 −2
Open

auto-detect GPU arch and adapt XSched preemption level, remove hardco… #1

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
17 changes: 16 additions & 1 deletion ggml/src/ggml-cuda/common.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -812,14 +812,29 @@ struct ggml_backend_cuda_context {

~ggml_backend_cuda_context();

int get_max_supported_preempt_level(int device_id) {
cudaDeviceProp prop;
CUDA_CHECK(cudaGetDeviceProperties(&prop, device_id));
int arch = prop.major * 10 + prop.minor;
switch (arch) {
case 35: // Kepler: K20, K40, GTX TITAN
return 2; // kPreemptLevelDeactivate
case 70: // Volta: V100
case 86: // Ampere: RTX 30系列
return 3; // kPreemptLevelInterrupt
default: // 其他架构（包括 A100 arch=80）
return 1; // kPreemptLevelBlock
}
}

cudaStream_t stream(int device, int stream) {
if (streams[device][stream] == nullptr) {
ggml_cuda_set_device(device);
CUDA_CHECK(cudaStreamCreateWithFlags(&streams[device][stream], cudaStreamNonBlocking));
HwQueueHandle hwqueue;
CudaQueueCreate(&hwqueue,streams[device][stream]);
XQueueHandle xqueue;
XQueueCreate(&xqueue, hwqueue, kPreemptLevelDeactivate, kQueueCreateFlagNone);
XQueueCreate(&xqueue, hwqueue, get_max_supported_preempt_level(device), kQueueCreateFlagNone);
XHintPriority(xqueue, priority); // In XSched, lower number means lower priority
}
return streams[device][stream];
Expand Down
2 changes: 1 addition & 1 deletion ggml/src/ggml-cuda/ggml-cuda.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2864,7 +2864,7 @@ static void ggml_backend_cuda_set_priority(ggml_backend_t backend, int prio) {
HwQueueHandle hwqueue;
CudaQueueCreate(&hwqueue,stream);
XQueueHandle xqueue;
XQueueCreate(&xqueue, hwqueue, kPreemptLevelDeactivate, kQueueCreateFlagNone);
XQueueCreate(&xqueue, hwqueue, cuda_ctx->get_max_supported_preempt_level(device), kQueueCreateFlagNone);
XHintPriority(xqueue, prio); // In XSched, lower number means lower priority
}
}
Expand Down