This is an automated email from the ASF dual-hosted git repository.
marin-ma pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/gluten.git
The following commit(s) were added to refs/heads/main by this push:
new 13ac7e49d3 [GLUTEN-11828][VL] Add cuda device detection (#11830)
13ac7e49d3 is described below
commit 13ac7e49d36503f347c405b45394c7a585bcb2f9
Author: Rong Ma <[email protected]>
AuthorDate: Wed Aug 5 16:34:34 2026 +0100
[GLUTEN-11828][VL] Add cuda device detection (#11830)
---
cpp/velox/compute/VeloxBackend.cc | 46 +++++++++++++++++++++++++--------------
1 file changed, 30 insertions(+), 16 deletions(-)
diff --git a/cpp/velox/compute/VeloxBackend.cc
b/cpp/velox/compute/VeloxBackend.cc
index 0d45559f0e..3b735601a1 100644
--- a/cpp/velox/compute/VeloxBackend.cc
+++ b/cpp/velox/compute/VeloxBackend.cc
@@ -120,6 +120,16 @@ ThreadManager* veloxThreadManagerFactory(const
std::string& kind, std::unique_pt
void veloxThreadManagerReleaser(ThreadManager* threadManager) {
delete threadManager;
}
+
+bool hasCudaDevice() {
+#ifdef GLUTEN_ENABLE_GPU
+ int count = 0;
+ cudaError_t err = cudaGetDeviceCount(&count);
+ return err == cudaSuccess && count > 0;
+#else
+ return false;
+#endif
+}
} // namespace
void VeloxBackend::init(
@@ -195,22 +205,26 @@ void VeloxBackend::init(
#ifdef GLUTEN_ENABLE_GPU
if (backendConf_->get<bool>(kCudfEnabled, kCudfEnabledDefault)) {
-
configureGpuTaskConcurrency(backendConf_->get<uint32_t>(kCudfConcurrentGpuTasks,
kCudfConcurrentGpuTasksDefault));
- std::unordered_map<std::string, std::string> options = {
- {velox::cudf_velox::CudfConfig::kCudfEnabled, "true"},
- {velox::cudf_velox::CudfConfig::kCudfDebugEnabled,
backendConf_->get(kDebugCudf, kDebugCudfDefault)},
- {velox::cudf_velox::CudfConfig::kCudfMemoryResource,
- backendConf_->get(kCudfMemoryResource, kCudfMemoryResourceDefault)},
- {velox::cudf_velox::CudfConfig::kCudfMemoryPercent,
- backendConf_->get(kCudfMemoryPercent, kCudfMemoryPercentDefault)},
- {velox::cudf_velox::CudfConfig::kCudfAllowCpuFallback,
- backendConf_->get(kCudfAllowCpuFallback,
kCudfAllowCpuFallbackDefault)}};
- auto& cudfConfig = velox::cudf_velox::CudfConfig::getInstance();
- cudfConfig.initialize(std::move(options));
- velox::cudf_velox::registerCudf();
-
velox::exec::Operator::registerOperator(std::make_unique<CudfVectorStreamOperatorTranslator>());
- velox::cudf_velox::registerSparkFunctions("");
- velox::cudf_velox::registerSparkAggregateFunctions("");
+ if (hasCudaDevice()) {
+
configureGpuTaskConcurrency(backendConf_->get<uint32_t>(kCudfConcurrentGpuTasks,
kCudfConcurrentGpuTasksDefault));
+ std::unordered_map<std::string, std::string> options = {
+ {velox::cudf_velox::CudfConfig::kCudfEnabled, "true"},
+ {velox::cudf_velox::CudfConfig::kCudfDebugEnabled,
backendConf_->get(kDebugCudf, kDebugCudfDefault)},
+ {velox::cudf_velox::CudfConfig::kCudfMemoryResource,
+ backendConf_->get(kCudfMemoryResource, kCudfMemoryResourceDefault)},
+ {velox::cudf_velox::CudfConfig::kCudfMemoryPercent,
+ backendConf_->get(kCudfMemoryPercent, kCudfMemoryPercentDefault)},
+ {velox::cudf_velox::CudfConfig::kCudfAllowCpuFallback,
+ backendConf_->get(kCudfAllowCpuFallback,
kCudfAllowCpuFallbackDefault)}};
+ auto& cudfConfig = velox::cudf_velox::CudfConfig::getInstance();
+ cudfConfig.initialize(std::move(options));
+ velox::cudf_velox::registerCudf();
+
velox::exec::Operator::registerOperator(std::make_unique<CudfVectorStreamOperatorTranslator>());
+ velox::cudf_velox::registerSparkFunctions("");
+ velox::cudf_velox::registerSparkAggregateFunctions("");
+ } else {
+ LOG(WARNING) << "No Cuda device found. Skip Cudf initialization.";
+ }
}
#endif
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
