[libomptarget][cuda] Gracefully handle missing cuda library

[JonChesterfield]

Summary

[libomptarget][cuda] Gracefully handle missing cuda library

If using dynamic cuda, and it failed to load, it is not safe to call
cuGetErrorString.

[JonChesterfield]

Noticed when running offloading on a system with no cuda available. Failing to load cuda from the global constructor didn't matter, but the segv on calling cuGetErrorString did.

An existing hazard, this plugin doesn't record whether the constructor initialisation succeeded (aside from debug messages). If it failed, but there are nvptx offloading images present, I think libomptarget will call back into this library despite the constructor failing, and I'm not sure that will reliably report an error instead of crash. Reading the code suggests it'll deference nullptr for the StreamManager.

[jdoerfert]

LG

[JonChesterfield]

This has failed a CI run, libomp::nested.c, but I don't understand the test output. Going to revert this for now.

[JonChesterfield]

The failure was probably a symptom of the error fixed at D95467.

openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.h

 typedef struct CUmod_st *CUmodule;
 typedef struct CUctx_st *CUcontext;
 typedef struct CUfunc_st *CUfunction;
 typedef struct CUstream_st *CUstream;
 
 typedef enum cudaError_enum {
   CUDA_SUCCESS = 0,
   CUDA_ERROR_INVALID_VALUE = 1,
+  CUDA_ERROR_INVALID_HANDLE = 400,
 } CUresult;
 
 typedef enum CUstream_flags_enum {
   CU_STREAM_DEFAULT = 0x0,
   CU_STREAM_NON_BLOCKING = 0x1,
 } CUstream_flags;
 
 typedef enum CUdevice_attribute_enum {

openmp/libomptarget/plugins/cuda/dynamic_cuda/cuda.cpp

 //===--- cuda/dynamic_cuda/cuda.pp ------------------------------- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // Implement subset of cuda api by calling into cuda library via dlopen
 // Does the dlopen/dlsym calls as part of the call to cuInit
 //
 //===----------------------------------------------------------------------===//
 
 #include "cuda.h"
 #include "Debug.h"
 #include "dlwrap.h"
 
 #include <dlfcn.h>
 
 DLWRAP_INTERNAL(cuInit, 1);
 
 DLWRAP(cuCtxGetDevice, 1);
 DLWRAP(cuDeviceGet, 2);
static bool checkForCUDA() {
 
   return true;
 }
 
 CUresult cuInit(unsigned X) {
   // Note: Called exactly once from cuda rtl.cpp in a global constructor so
   // does not need to handle being called repeatedly or concurrently
   if (!checkForCUDA()) {
-    return CUDA_ERROR_INVALID_VALUE;
+    return CUDA_ERROR_INVALID_HANDLE;
   }
   return dlwrap_cuInit(X);
 }

openmp/libomptarget/plugins/cuda/src/rtl.cpp

 //===----RTLs/cuda/src/rtl.cpp - Target RTLs Implementation ------- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // RTL for CUDA machine
 //
 //===----------------------------------------------------------------------===//
 
 #include <cassert>
 #include <cstddef>
 #include <cuda.h>
 #include <list>
 #include <memory>
 #include <mutex>
 #include <string>
 #include <vector>
 
 #include "Debug.h"
 #include "omptargetplugin.h"
public:
 
   DeviceRTLTy()
       : NumberOfDevices(0), EnvNumTeams(-1), EnvTeamLimit(-1),
         RequiresFlags(OMP_REQ_UNDEFINED) {
 
     DP("Start initializing CUDA\n");
 
     CUresult Err = cuInit(0);
+    if (Err == CUDA_ERROR_INVALID_HANDLE) {
+      // Can't call cuGetErrorString if dlsym failed
+      DP("Failed to load CUDA shared library\n");
+      return;
+    }
     if (!checkResult(Err, "Error returned from cuInit\n")) {
       return;
     }
 
     Err = cuDeviceGetCount(&NumberOfDevices);
     if (!checkResult(Err, "Error returned from cuDeviceGetCount\n"))
       return;