[UR][OpenCL] Move banned platforms filtering to UR

Filter out banned OpenCl platforms earlier, in the UR OpenCL adapter
instead of the SYCL RT.

Author: PatKamin

sycl/source/detail/platform_impl.cpp

@@ -88,33 +88,6 @@ context_impl &platform_impl::khr_get_default_context() {
   return *It->second;
 }
 
-static bool IsBannedPlatform(platform Platform) {
-  // The NVIDIA OpenCL platform is currently not compatible with DPC++
-  // since it is only 1.2 but gets selected by default in many systems
-  // There is also no support on the PTX backend for OpenCL consumption,
-  // and there have been some internal reports.
-  // To avoid problems on default users and deployment of DPC++ on platforms
-  // where CUDA is available, the OpenCL support is disabled.
-  //
-  // There is also no support for the AMD HSA backend for OpenCL consumption,
-  // as well as reported problems with device queries, so AMD OpenCL support
-  // is disabled as well.
-  //
-  auto IsMatchingOpenCL = [](platform Platform, const std::string_view name) {
-    const bool HasNameMatch = Platform.get_info<info::platform::name>().find(
-                                  name) != std::string::npos;
-    const auto Backend = detail::getSyclObjImpl(Platform)->getBackend();
-    const bool IsMatchingOCL = (HasNameMatch && Backend == backend::opencl);
-    if (detail::ur::trace(detail::ur::TraceLevel::TRACE_ALL) && IsMatchingOCL) {
-      std::cout << "SYCL_UR_TRACE: " << name
-                << " OpenCL platform found but is not compatible." << std::endl;
-    }
-    return IsMatchingOCL;
-  };
-  return IsMatchingOpenCL(Platform, "NVIDIA CUDA") ||
-         IsMatchingOpenCL(Platform, "AMD Accelerated Parallel Processing");
-}
-
 // Get the vector of platforms supported by a given UR adapter
 // replace uses of this with a helper in adapter object, the adapter
 // objects will own the ur adapter handles and they'll need to pass them to
@@ -132,25 +105,13 @@ std::vector<platform> platform_impl::getAdapterPlatforms(adapter_impl &Adapter,
   for (const auto &UrPlatform : UrPlatforms) {
     platform Platform = detail::createSyclObjFromImpl<platform>(
         getOrMakePlatformImpl(UrPlatform, Adapter));
-    const bool IsBanned = IsBannedPlatform(Platform);
-    bool HasAnyDevices = false;
-
-    // Platform.get_devices() increments the device count for the platform
-    // and if the platform is banned (like OpenCL for AMD), it can cause
-    // incorrect device numbering, when used with ONEAPI_DEVICE_SELECTOR.
-    if (!IsBanned)
-      HasAnyDevices = !Platform.get_devices(info::device_type::all).empty();
+    bool HasAnyDevices = !Platform.get_devices(info::device_type::all).empty();
 
     if (!Supported) {
-      if (IsBanned || !HasAnyDevices) {
+      if (!HasAnyDevices) {
         Platforms.push_back(std::move(Platform));
       }
     } else {
-      if (IsBanned) {
-        continue; // bail as early as possible, otherwise banned platforms may
-                  // mess up device counting
-      }
-
       // The SYCL spec says that a platform has one or more devices. ( SYCL
       // 2020 4.6.2 ) If we have an empty platform, we don't report it back
       // from platform::get_platforms().

unified-runtime/source/adapters/opencl/platform.cpp

@@ -30,6 +30,35 @@ static cl_int mapURPlatformInfoToCL(ur_platform_info_t URPropName) {
   }
 }
 
+static bool isBannedOpenCLPlatform(cl_platform_id platform) {
+  size_t nameSize = 0;
+  cl_int res =
+      clGetPlatformInfo(platform, CL_PLATFORM_NAME, 0, nullptr, &nameSize);
+  if (res != CL_SUCCESS || nameSize == 0) {
+    return false;
+  }
+
+  std::string name(nameSize, '\0');
+  res = clGetPlatformInfo(platform, CL_PLATFORM_NAME, nameSize, name.data(),
+                          nullptr);
+  if (res != CL_SUCCESS) {
+    return false;
+  }
+
+  // The NVIDIA OpenCL platform is currently not compatible with DPC++
+  // since it is only 1.2 but gets selected by default in many systems.
+  // There is also no support on the PTX backend for OpenCL consumption.
+  //
+  // There is also no support for the AMD HSA backend for OpenCL consumption,
+  // as well as reported problems with device queries, so AMD OpenCL support
+  // is disabled as well.
+  bool isBanned =
+      name.find("NVIDIA CUDA") != std::string::npos ||
+      name.find("AMD Accelerated Parallel Processing") != std::string::npos;
+
+  return isBanned;
+}
+
 UR_DLLEXPORT ur_result_t UR_APICALL
 urPlatformGetInfo(ur_platform_handle_t hPlatform, ur_platform_info_t propName,
                   size_t propSize, void *pPropValue, size_t *pSizeRet) {
@@ -102,14 +131,21 @@ urPlatformGet(ur_adapter_handle_t, uint32_t NumEntries,
                              CLPlatforms.data(), nullptr);
       CL_RETURN_ON_FAILURE(Res);
 
+      // Filter out banned platforms
+      std::vector<cl_platform_id> FilteredPlatforms;
+      for (uint32_t i = 0; i < NumPlatforms; i++) {
+        if (!isBannedOpenCLPlatform(CLPlatforms[i])) {
+          FilteredPlatforms.push_back(CLPlatforms[i]);
+        }
+      }
+
       try {
-        for (uint32_t i = 0; i < NumPlatforms; i++) {
-          auto URPlatform =
-              std::make_unique<ur_platform_handle_t_>(CLPlatforms[i]);
+        for (auto &Platform : FilteredPlatforms) {
+          auto URPlatform = std::make_unique<ur_platform_handle_t_>(Platform);
           UR_RETURN_ON_FAILURE(URPlatform->InitDevices());
           Adapter->URPlatforms.emplace_back(URPlatform.release());
         }
-        Adapter->NumPlatforms = NumPlatforms;
+        Adapter->NumPlatforms = static_cast<uint32_t>(FilteredPlatforms.size());
       } catch (std::bad_alloc &) {
         return UR_RESULT_ERROR_OUT_OF_RESOURCES;
       } catch (...) {