[SYCL] Fix multi-device support for persistent cache

sycl/cmake/modules/UnifiedRuntimeTag.cmake

@@ -1,7 +1,10 @@
-# commit cd92e72bbc4ebddef63c63c0f7e66a410f4b9552
-# Merge: 9a209aa5 b1222f08
-# Author: Callum Fare <[email protected]>
-# Date:   Wed Nov 13 09:57:16 2024 +0000
-#     Merge pull request #2254 from PietroGhg/pietro/events_rr
-#     [NATIVECPU] Implement events on Native CPU
-set(UNIFIED_RUNTIME_TAG cd92e72bbc4ebddef63c63c0f7e66a410f4b9552)
+# commit 3e1a0ea44842b10cfdc0e1e98af2eee8fcd8e937 (HEAD -> main)
+# Author: Artur Gainullin <[email protected]>
+# Date:   Wed Nov 6 16:05:50 2024 -0800
+#
+#     [L0] Fix binary sizes and binaries returned by urProgramGetInfo
+#
+#     Currently urProgramGetInfo will return UR_INVALID_PROGRAM is program is
+#     compiled only for a subset of associated devices, i.e. not all devices
+#     have level zero module and binaries. This PR fixes this behaviour.
+set(UNIFIED_RUNTIME_TAG 3e1a0ea44842b10cfdc0e1e98af2eee8fcd8e937)

sycl/source/detail/kernel_bundle_impl.hpp

@@ -415,20 +415,15 @@ class kernel_bundle_impl {
 
     std::vector<const uint8_t *> Binaries;
     std::vector<size_t> Lengths;
-    std::vector<std::vector<std::vector<char>>> PersistentBinaries;
-    for (size_t i = 0; i < Devices.size(); i++) {
-      std::vector<std::vector<char>> BinProg =
-          PersistentDeviceCodeCache::getCompiledKernelFromDisc(
-              Devices[i], UserArgs, SourceStr);
-
-      // exit if any device binary is missing
-      if (BinProg.empty()) {
-        return false;
-      }
-      PersistentBinaries.push_back(BinProg);
-
-      Binaries.push_back((uint8_t *)(PersistentBinaries[i][0].data()));
-      Lengths.push_back(PersistentBinaries[i][0].size());
+    std::vector<std::vector<char>> BinProgs =
+        PersistentDeviceCodeCache::getCompiledKernelFromDisc(Devices, UserArgs,
+                                                             SourceStr);
+    if (BinProgs.empty()) {
+      return false;
+    }
+    for (auto &BinProg : BinProgs) {
+      Binaries.push_back((uint8_t *)(BinProg.data()));
+      Lengths.push_back(BinProg.size());
     }
 
     ur_program_properties_t Properties = {};
@@ -564,11 +559,9 @@ class kernel_bundle_impl {
     // If caching enabled and kernel not fetched from cache, cache.
     if (PersistentDeviceCodeCache::isEnabled() && !FetchedFromCache &&
         SourceStrPtr) {
-      for (const auto &Device : Devices) {
-        PersistentDeviceCodeCache::putCompiledKernelToDisc(
-            Device, syclex::detail::userArgsAsString(BuildOptions),
-            *SourceStrPtr, UrProgram);
-      }
+      PersistentDeviceCodeCache::putCompiledKernelToDisc(
+          Devices, syclex::detail::userArgsAsString(BuildOptions),
+          *SourceStrPtr, UrProgram);
     }
 
     return std::make_shared<kernel_bundle_impl>(MContext, MDevices, DevImg,