From 244bd973cedb192dffc9f0cacb84767f4f91dbdb Mon Sep 17 00:00:00 2001
From: Anna Wellmann <a.wellmann@tu-braunschweig.de>
Date: Wed, 14 Jun 2023 10:41:53 +0200
Subject: [PATCH] Log device number for cuda devices
---
regression-tests/multigpu_test/rocket4GPU.yml | 3 +--
regression-tests/multigpu_test/rocket8GPU.yml | 3 +--
src/cuda/DeviceInfo.cpp | 12 ++++++------
3 files changed, 8 insertions(+), 10 deletions(-)
diff --git a/regression-tests/multigpu_test/rocket4GPU.yml b/regression-tests/multigpu_test/rocket4GPU.yml
index 2de1602cd..a7ece0557 100755
--- a/regression-tests/multigpu_test/rocket4GPU.yml
+++ b/regression-tests/multigpu_test/rocket4GPU.yml
@@ -45,8 +45,7 @@ collect:
overwrite: true
clean:
- - multigpu_test/output/*
- - multigpu_test/src/*
+ - multigpu_test/*
sbatch: multigpu_test/slurm4GPU.job
continue_if_job_fails: true
diff --git a/regression-tests/multigpu_test/rocket8GPU.yml b/regression-tests/multigpu_test/rocket8GPU.yml
index 29f06e919..76203eb9d 100755
--- a/regression-tests/multigpu_test/rocket8GPU.yml
+++ b/regression-tests/multigpu_test/rocket8GPU.yml
@@ -46,8 +46,7 @@ collect:
overwrite: true
clean:
- - multigpu_test/output/*
- - multigpu_test/src/*
+ - multigpu_test/*
sbatch: multigpu_test/slurm8GPU.job
continue_if_job_fails: true
diff --git a/src/cuda/DeviceInfo.cpp b/src/cuda/DeviceInfo.cpp
index 20ea2c4f6..e4fe15ed8 100644
--- a/src/cuda/DeviceInfo.cpp
+++ b/src/cuda/DeviceInfo.cpp
@@ -15,7 +15,7 @@ void verifyNumberOfDevices(int deviceId)
int device_count = 0;
cudaError_t errorId = cudaGetDeviceCount(&device_count);
if(errorId != cudaSuccess) {
- VF_LOG_CRITICAL("Error while accessing the device count: {}", cudaGetErrorString(errorId));
+ VF_LOG_CRITICAL("Device {}: Error while accessing the device count: {}", deviceId, cudaGetErrorString(errorId));
}
if (deviceId > device_count) {
throw std::runtime_error("chosen gpudevice >= device_count ... exiting\n");
@@ -28,13 +28,13 @@ void verifyComputeCapability(int deviceId)
cudaError_t errorId = cudaGetDeviceProperties(&deviceProp, deviceId);
if(errorId != cudaSuccess){
- VF_LOG_CRITICAL("Error while accessing the device properties occurs: {}", cudaGetErrorString(errorId));
+ VF_LOG_CRITICAL("Device {}: Error while accessing the device properties occurs: {}", deviceId, cudaGetErrorString(errorId));
}
VF_LOG_INFO("[compute capability] = [{}.{}]\n", deviceProp.major, deviceProp.minor);
if (deviceProp.major > 999) {
- throw std::runtime_error("warning, CUDA Device Emulation (CPU) detected, exiting\n");
+ throw std::runtime_error("Warning, CUDA Device Emulation (CPU) detected, exiting\n");
}
}
@@ -43,13 +43,13 @@ void setCudaDevice(int deviceId)
// choose a cuda device for kernel execution
cudaError_t errorId = cudaSetDevice(deviceId);
if (errorId != cudaSuccess) {
- VF_LOG_CRITICAL("Error while setting the device to {}: {}", deviceId, cudaGetErrorString(errorId));
+ VF_LOG_CRITICAL("Device {}: Error while setting the device to: {}", deviceId, cudaGetErrorString(errorId));
} else {
int device;
// double check that device was properly selected
errorId = cudaGetDevice(&device);
if(errorId != cudaSuccess) {
- VF_LOG_CRITICAL("Error while getting the device: {}", cudaGetErrorString(errorId));
+ VF_LOG_CRITICAL("Device {}: Error while getting the device: {}", deviceId, cudaGetErrorString(errorId));
}
}
}
@@ -70,7 +70,7 @@ void printCudaInformation(int deviceId)
cudaError_t errorId = cudaGetDeviceProperties(&prop, deviceId);
if(errorId != cudaSuccess){
- VF_LOG_CRITICAL("Error while accessing the device properties occurs: {}", cudaGetErrorString(errorId));
+ VF_LOG_CRITICAL("Device {}: Error while accessing the device properties for occurs: {}", deviceId, cudaGetErrorString(errorId));
}
printf(" --- General Information for device %d ---\n", deviceId);
--
GitLab