[SYCL] Add support for device UUID as a SYCL extension. (#3696)

This includes support for only level_zero. The
llvm/sycl/docs/extensions/IntelGPU/IntelGPUDeviceInfo.md doc has
been modified with a brief description of the extension. A new aspect
is added to indicate if the support is available.

Signed-off-by: rbegam <[email protected]>

Author: rbegam

sycl/doc/extensions/IntelGPU/IntelGPUDeviceInfo.md

@@ -9,9 +9,50 @@ This proposal details what is required to provide this information as a SYCL ext
 
 ## Feature Test Macro ##
 
-The Feature Test Macro will be defined as:
+The Feature Test Macro SYCL\_EXT\_INTEL\_DEVICE\_INFO will be defined as one of the values defined in the table below. The existence of this macro can be tested to determine if the implementation supports this feature, or applications can test the macro's value to determine which of the extension's APIs the implementation supports.
+
+| Value | Description |
+| ----- | ----------- |
+| 1     | Initial extension version\. Base features are supported |
+| 2     | Device UUID is supported |
+
+
+# Device UUID #
+
+A new device descriptor will be added which will provide the device Universal Unique ID (UUID).
+
+This new device descriptor is currently only available for devices in the Level Zero platform, and the matching aspect is only true for those devices. The DPC++ default behavior would be to expose the UUIDs of all supported devices which enables detection of total number of unique devices.
+
+
+## Version ##
+
+The extension supports this query in version 2 and later.
+
+
+## Device Information Descriptors ##
+
+| Device Descriptors | Return Type | Description |
+| ------------------ | ----------- | ----------- |
+| info\:\:device\:\:ext\_intel\_device\_info\_uuid | unsigned char | Returns the device UUID|
+
+
+## Aspects ##
+
+A new aspect, ext\_intel\_device\_info\_uuid, will be added.
+
+## Error Condition ##
+
+An invalid object runtime error will be thrown if the device does not support aspect\:\:ext\_intel\_device\_info\_uuid.
+
+
+## Example Usage ##
+
+The UUID can be obtained using the standard get\_info() interface.
+
+    if (dev.has(aspect::ext_intel_device_info_uuid)) {
+      auto UUID = dev.get_info<info::device::ext_intel_device_info_uuid>();
+    }
 
-    #define SYCL_EXT_INTEL_DEVICE_INFO 1
 
 
 # PCI Address #
@@ -23,6 +64,11 @@ This new device descriptor is only available for devices in the Level Zero platf
 **Note:** The environment variable SYCL\_ENABLE\_PCI must be set to 1 to obtain the PCI address.
 
 
+## Version ##
+
+All versions of the extension support this query.
+
+
 ## Device Information Descriptors ##
 
 | Device Descriptors | Return Type | Description |
@@ -56,6 +102,11 @@ A new device descriptor will be added which will provide the physical SIMD width
 This new device descriptor is only available for devices in the Level Zero platform, and the matching aspect is only true for those devices. The DPC++ default behavior is to expose GPU devices through the Level Zero platform.
 
 
+## Version ##
+
+All versions of the extension support this query.
+
+
 ## Device Information Descriptors ##
 
 | Device Descriptors | Return Type | Description |
@@ -91,6 +142,11 @@ This new device descriptor will provide the same information as "max\_compute\_u
 This new device descriptor is only available for devices in the Level Zero platform, and the matching aspect is only true for those devices. The DPC++ default behavior is to expose GPU devices through the Level Zero platform.
 
 
+## Version ##
+
+All versions of the extension support this query.
+
+
 ## Device Information Descriptors ##
 
 | Device Descriptors | Return Type | Description |
@@ -124,6 +180,11 @@ A new device descriptor will be added which will provide the number of slices on
 This new device descriptor is only available for devices in the Level Zero platform, and the matching aspect is only true for those devices. The DPC++ default behavior is to expose GPU devices through the Level Zero platform.
 
 
+## Version ##
+
+All versions of the extension support this query.
+
+
 ## Device Information Descriptors ##
 
 | Device Descriptors | Return Type | Description |
@@ -156,6 +217,11 @@ A new device descriptor will be added which will provide the number of subslices
 This new device descriptor is only available for devices in the Level Zero platform, and the matching aspect is only true for those devices. The DPC++ default behavior is to expose GPU devices through the Level Zero platform.
 
 
+## Version ##
+
+All versions of the extension support this query.
+
+
 ## Device Information Descriptors ##
 
 | Device Descriptors | Return Type | Description |
@@ -188,6 +254,11 @@ A new device descriptor will be added which will provide the number of EUs per s
 This new device descriptor is only available for devices in the Level Zero platform, and the matching aspect is only true for those devices. The DPC++ default behavior is to expose GPU devices through the Level Zero platform.
 
 
+## Version ##
+
+All versions of the extension support this query.
+
+
 ## Device Information Descriptors ##
 
 | Device Descriptors | Return Type | Description |
@@ -220,6 +291,11 @@ A new device descriptor will be added which will provide the maximum memory band
 This new device descriptor is only available for devices in the Level Zero platform, and the matching aspect is only true for those devices. The DPC++ default behavior is to expose GPU devices through the Level Zero platform.
 
 
+## Version ##
+
+All versions of the extension support this query.
+
+
 ## Device Information Descriptors ##
 
 | Device Descriptors | Return Type | Description |

sycl/include/CL/sycl/aspects.hpp

@@ -41,7 +41,8 @@ enum class aspect {
   ext_intel_mem_channel = 25,
   usm_atomic_host_allocations = 26,
   usm_atomic_shared_allocations = 27,
-  atomic64 = 28
+  atomic64 = 28,
+  ext_intel_device_info_uuid = 29,
 };
 
 } // namespace sycl

sycl/include/CL/sycl/detail/pi.h

@@ -52,6 +52,7 @@
 #include <CL/cl_ext.h>
 #include <CL/sycl/detail/cl.h>
 #include <CL/sycl/detail/export.hpp>
+
 #include <cstdint>
 
 #ifdef __cplusplus
@@ -278,6 +279,8 @@ typedef enum {
       CL_DEVICE_CROSS_DEVICE_SHARED_MEM_CAPABILITIES_INTEL,
   PI_DEVICE_INFO_USM_SYSTEM_SHARED_SUPPORT =
       CL_DEVICE_SHARED_SYSTEM_MEM_CAPABILITIES_INTEL,
+  // Intel UUID extension.
+  PI_DEVICE_INFO_UUID = CL_DEVICE_UUID_KHR,
   // These are Intel-specific extensions.
   PI_DEVICE_INFO_PCI_ADDRESS = 0x10020,
   PI_DEVICE_INFO_GPU_EU_COUNT = 0x10021,

sycl/include/CL/sycl/detail/type_traits.hpp

@@ -13,6 +13,7 @@
 #include <CL/sycl/detail/stl_type_traits.hpp>
 #include <CL/sycl/detail/type_list.hpp>
 
+#include <array>
 #include <type_traits>
 
 __SYCL_INLINE_NAMESPACE(cl) {
@@ -51,6 +52,11 @@ __SYCL_INLINE_CONSTEXPR bool is_group_v =
     detail::is_group<T>::value || detail::is_sub_group<T>::value;
 
 namespace detail {
+// Type for Intel device UUID extension.
+// For details about this extension, see
+// sycl/doc/extensions/IntelGPU/IntelGPUDeviceInfo.md
+using uuid_type = std::array<unsigned char, 16>;
+
 template <typename T, typename R> struct copy_cv_qualifiers;
 
 template <typename T, typename R>

sycl/include/CL/sycl/feature_test.hpp

@@ -13,7 +13,7 @@ namespace sycl {
 // Feature test macro definitions
 
 // TODO: Move these feature-test macros to compiler driver.
-#define SYCL_EXT_INTEL_DEVICE_INFO 1
+#define SYCL_EXT_INTEL_DEVICE_INFO 2
 #define SYCL_EXT_ONEAPI_MATRIX 1
 
 } // namespace sycl

sycl/include/CL/sycl/info/device_traits.def

@@ -94,3 +94,4 @@ __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_gpu_subslices_per_slice, pi_uint32)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_gpu_eu_count_per_subslice, pi_uint32)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_max_mem_bandwidth, pi_uint64)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_mem_channel, bool)
+__SYCL_PARAM_TRAITS_SPEC(device, ext_intel_device_info_uuid, detail::uuid_type)

sycl/include/CL/sycl/info/info_desc.hpp

@@ -134,7 +134,6 @@ enum class device : cl_device_info {
   usm_shared_allocations = PI_USM_SINGLE_SHARED_SUPPORT,
   usm_restricted_shared_allocations = PI_USM_CROSS_SHARED_SUPPORT,
   usm_system_allocator = PI_USM_SYSTEM_SHARED_SUPPORT,
-
   // intel extensions
   ext_intel_pci_address = PI_DEVICE_INFO_PCI_ADDRESS,
   ext_intel_gpu_eu_count = PI_DEVICE_INFO_GPU_EU_COUNT,
@@ -145,6 +144,7 @@ enum class device : cl_device_info {
       PI_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE,
   ext_intel_max_mem_bandwidth = PI_DEVICE_INFO_MAX_MEM_BANDWIDTH,
   ext_intel_mem_channel = PI_MEM_PROPERTIES_CHANNEL,
+  ext_intel_device_info_uuid = PI_DEVICE_INFO_UUID,
   atomic64 = PI_DEVICE_INFO_ATOMIC_64
 };
 

sycl/plugins/cuda/pi_cuda.cpp

@@ -1551,6 +1551,10 @@ pi_result cuda_piDeviceGetInfo(pi_device device, pi_device_info param_name,
   case PI_DEVICE_INFO_GPU_SUBSLICES_PER_SLICE:
   case PI_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE:
   case PI_DEVICE_INFO_MAX_MEM_BANDWIDTH:
+    // TODO: Check if Intel device UUID extension is utilized for CUDA.
+    // For details about this extension, see
+    // sycl/doc/extensions/IntelGPU/IntelGPUDeviceInfo.md
+  case PI_DEVICE_INFO_UUID:
     return PI_INVALID_VALUE;
 
   default:

sycl/plugins/level_zero/pi_level_zero.cpp

@@ -1655,6 +1655,11 @@ pi_result piDeviceGetInfo(pi_device Device, pi_device_info ParamName,
     return ReturnValue(Device->Platform);
   case PI_DEVICE_INFO_VENDOR_ID:
     return ReturnValue(pi_uint32{Device->ZeDeviceProperties.vendorId});
+  case PI_DEVICE_INFO_UUID:
+    // Intel extension for device UUID. This returns the UUID as
+    // std::array<std::byte, 16>. For details about this extension,
+    // see sycl/doc/extensions/IntelGPU/IntelGPUDeviceInfo.md.
+    return ReturnValue(Device->ZeDeviceProperties.uuid.id);
   case PI_DEVICE_INFO_EXTENSIONS: {
     // Convention adopted from OpenCL:
     //     "Returns a space separated list of extension names (the extension

sycl/plugins/opencl/pi_opencl.cpp

@@ -169,15 +169,19 @@ pi_result piDeviceGetInfo(pi_device device, pi_device_info paramName,
                           size_t paramValueSize, void *paramValue,
                           size_t *paramValueSizeRet) {
   switch (paramName) {
-    // Intel GPU EU device-specific information extensions.
     // TODO: Check regularly to see if support in enabled in OpenCL.
+    // Intel GPU EU device-specific information extensions.
   case PI_DEVICE_INFO_PCI_ADDRESS:
   case PI_DEVICE_INFO_GPU_EU_COUNT:
   case PI_DEVICE_INFO_GPU_EU_SIMD_WIDTH:
   case PI_DEVICE_INFO_GPU_SLICES:
   case PI_DEVICE_INFO_GPU_SUBSLICES_PER_SLICE:
   case PI_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE:
   case PI_DEVICE_INFO_MAX_MEM_BANDWIDTH:
+    // TODO: Check if device UUID extension is enabled in OpenCL.
+    // For details about Intel UUID extension, see
+    // sycl/doc/extensions/IntelGPU/IntelGPUDeviceInfo.md
+  case PI_DEVICE_INFO_UUID:
   case PI_DEVICE_INFO_ATOMIC_64:
     return PI_INVALID_VALUE;
 

sycl/source/detail/device_impl.cpp

@@ -298,6 +298,20 @@ bool device_impl::has(aspect Aspect) const {
                MDevice, PI_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE,
                sizeof(pi_device_type), &device_type,
                &return_size) == PI_SUCCESS;
+  case aspect::ext_intel_device_info_uuid: {
+    auto Result = getPlugin().call_nocheck<detail::PiApiKind::piDeviceGetInfo>(
+        MDevice, PI_DEVICE_INFO_UUID, 0, nullptr, &return_size);
+    if (Result != PI_SUCCESS) {
+      return false;
+    }
+
+    assert(return_size <= 16);
+    unsigned char UUID[16];
+
+    return getPlugin().call_nocheck<detail::PiApiKind::piDeviceGetInfo>(
+               MDevice, PI_DEVICE_INFO_UUID, 16 * sizeof(unsigned char), UUID,
+               nullptr) == PI_SUCCESS;
+  }
   case aspect::ext_intel_max_mem_bandwidth:
     // currently not supported
     return false;

sycl/source/detail/device_info.hpp

@@ -1101,6 +1101,14 @@ get_device_info_host<info::device::ext_intel_max_mem_bandwidth>() {
       PI_INVALID_DEVICE);
 }
 
+template <>
+inline detail::uuid_type
+get_device_info_host<info::device::ext_intel_device_info_uuid>() {
+  throw runtime_error(
+      "Obtaining the device uuid is not supported on HOST device",
+      PI_INVALID_DEVICE);
+}
+
 } // namespace detail
 } // namespace sycl
 } // __SYCL_INLINE_NAMESPACE(cl)

sycl/test/abi/sycl_symbols_linux.dump

@@ -4120,6 +4120,7 @@ _ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE4168EEENS3_12param_traitsIS4_XT_E
 _ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE4169EEENS3_12param_traitsIS4_XT_EE11return_typeEv
 _ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE4188EEENS3_12param_traitsIS4_XT_EE11return_typeEv
 _ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE4189EEENS3_12param_traitsIS4_XT_EE11return_typeEv
+_ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE4202EEENS3_12param_traitsIS4_XT_EE11return_typeEv
 _ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE65568EEENS3_12param_traitsIS4_XT_EE11return_typeEv
 _ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE65569EEENS3_12param_traitsIS4_XT_EE11return_typeEv
 _ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE65570EEENS3_12param_traitsIS4_XT_EE11return_typeEv

sycl/unittests/kernel-and-program/CMakeLists.txt

@@ -1,6 +1,7 @@
 add_sycl_unittest(KernelAndProgramTests OBJECT
   KernelRelease.cpp
   KernelInfo.cpp
+  DeviceInfo.cpp
   PersistentDeviceCodeCache.cpp
 )
 add_subdirectory(device)

sycl/unittests/kernel-and-program/DeviceInfo.cpp

@@ -0,0 +1,87 @@
+//==-------------- DeviceInfo.cpp --- device info unit test ----------------==//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include <CL/sycl.hpp>
+#include <detail/context_impl.hpp>
+#include <gtest/gtest.h>
+#include <helpers/PiMock.hpp>
+
+using namespace sycl;
+
+namespace {
+struct TestCtx {
+  TestCtx(context &Ctx) : Ctx{Ctx} {}
+
+  context &Ctx;
+  bool UUIDInfoCalled = false;
+};
+} // namespace
+
+static std::unique_ptr<TestCtx> TestContext;
+
+static pi_result redefinedDeviceGetInfo(pi_device device,
+                                        pi_device_info param_name,
+                                        size_t param_value_size,
+                                        void *param_value,
+                                        size_t *param_value_size_ret) {
+  if (param_name == PI_DEVICE_INFO_UUID) {
+    TestContext->UUIDInfoCalled = true;
+  }
+
+  return PI_SUCCESS;
+}
+
+class DeviceInfoTest : public ::testing::Test {
+public:
+  DeviceInfoTest() : Plt{default_selector()} {}
+
+protected:
+  void SetUp() override {
+    if (Plt.is_host()) {
+      std::clog << "This test is only supported on non-host platforms.\n";
+      std::clog << "Current platform is "
+                << Plt.get_info<info::platform::name>() << "\n";
+      return;
+    }
+
+    Mock = std::make_unique<unittest::PiMock>(Plt);
+
+    Mock->redefine<detail::PiApiKind::piDeviceGetInfo>(redefinedDeviceGetInfo);
+  }
+
+protected:
+  platform Plt;
+  std::unique_ptr<unittest::PiMock> Mock;
+};
+
+TEST_F(DeviceInfoTest, GetDeviceUUID) {
+  if (Plt.is_host()) {
+    return;
+  }
+
+  context Ctx{Plt};
+  TestContext.reset(new TestCtx(Ctx));
+
+  device Dev = Ctx.get_devices()[0];
+
+  if (!Dev.has(aspect::ext_intel_device_info_uuid)) {
+    std::clog
+        << "This test is only for the devices with UUID extension support.\n";
+    return;
+  }
+
+  auto UUID = Dev.get_info<info::device::ext_intel_device_info_uuid>();
+
+  EXPECT_EQ(TestContext->UUIDInfoCalled, true)
+      << "Expect piDeviceGetInfo to be "
+      << "called with PI_DEVICE_INFO_UUID";
+
+  EXPECT_EQ(sizeof(UUID), 16 * sizeof(unsigned char))
+      << "Expect device UUID to be "
+      << "of 16 bytes";
+}