Skip to content

respect GPU in exclusive process compute mode #74

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
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

respect GPU in exclusive process compute mode #74

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
30 changes: 28 additions & 2 deletions gpu.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,11 +3,11 @@
//

#include <stdlib.h>
#include <stdbool.h>
#include <nvml.h>
#include <string.h>

#include "main.h"

#define TS_VISIBLE_DEVICES "TS_VISIBLE_DEVICES"

static int free_percentage = 90;
Expand Down Expand Up @@ -90,6 +90,7 @@ int * getGpuList(int *num) {

nvmlMemory_t mem;
nvmlDevice_t dev;
nvmlComputeMode_t computeMode;
result = nvmlDeviceGetHandleByIndex_v2(visible[i], &dev);
if (result != 0) {
warning("Failed to get GPU handle for GPU %d: %s", visible[i], nvmlErrorString(result));
Expand All @@ -101,8 +102,33 @@ int * getGpuList(int *num) {
warning("Failed to get GPU memory for GPU %d: %s", visible[i], nvmlErrorString(result));
goto Error;
}

result = nvmlDeviceGetComputeMode(dev, &computeMode);
if (result != 0) {
warning("Failed to get GPU compute mode for GPU %d: %s", visible[i], nvmlErrorString(result));
goto Error;
}

// Check if there are any running processes on the GPU
bool anyRunningProcesses = false;
unsigned int infoCount = 0;
result = nvmlDeviceGetComputeRunningProcesses(dev, &infoCount, NULL);
if (result == NVML_ERROR_INSUFFICIENT_SIZE){
anyRunningProcesses = true;
} else if (result != 0) {
warning("Failed to get GPU compute processes for GPU %d: %s", visible[i], nvmlErrorString(result));
goto Error;
}

// Check if the GPU memory is free enough
bool isMemFree = mem.free > free_percentage / 100. * mem.total;
// When the compute mode is prohibited or in exclusive process mode with a
// running processes, the GPU is not free to use.
bool blockedByComputeMode = (computeMode == NVML_COMPUTEMODE_PROHIBITED) ||
(computeMode == NVML_COMPUTEMODE_EXCLUSIVE_PROCESS && anyRunningProcesses);

if (mem.free > free_percentage / 100. * mem.total)
bool isFree = isMemFree && !blockedByComputeMode;
if (isFree)
gpuList[count++] = visible[i];
}
free(visible);
Expand Down
Loading