[SYCL] Take into account UR_RESULT_ERROR_OUT_OF_DEVICE_MEMORY error code in Program Manager (#15335)

Currently if program manager faces one of the errors -
UR_RESULT_ERROR_OUT_OF_RESOURCES or UR_RESULT_ERROR_OUT_OF_HOST_MEMORY -
during the program building/linking then it will clear the cache and
make another attempt.
This PR adds the following changes:
* Additionally take into account UR_RESULT_ERROR_OUT_OF_DEVICE_MEMORY
error which is also possible in addition to aforementioned error codes.
* Parameterize the existing unit test by error code (which allows to
avoid excessive code duplication) and add
UR_RESULT_ERROR_OUT_OF_DEVICE_MEMORY value to testing.

Author: againull

sycl/source/detail/kernel_program_cache.hpp

@@ -329,7 +329,8 @@ class KernelProgramCache {
         BuildResult->Error.Code = detail::get_ur_error(Ex);
         if (Ex.code() == errc::memory_allocation ||
             BuildResult->Error.Code == UR_RESULT_ERROR_OUT_OF_RESOURCES ||
-            BuildResult->Error.Code == UR_RESULT_ERROR_OUT_OF_HOST_MEMORY) {
+            BuildResult->Error.Code == UR_RESULT_ERROR_OUT_OF_HOST_MEMORY ||
+            BuildResult->Error.Code == UR_RESULT_ERROR_OUT_OF_DEVICE_MEMORY) {
           reset();
           BuildResult->updateAndNotify(BuildState::BS_Initial);
           continue;

sycl/source/detail/program_manager/program_manager.cpp

@@ -1605,7 +1605,8 @@ ProgramManager::ProgramPtr ProgramManager::build(
   };
   ur_result_t Error = doLink();
   if (Error == UR_RESULT_ERROR_OUT_OF_RESOURCES ||
-      Error == UR_RESULT_ERROR_OUT_OF_HOST_MEMORY) {
+      Error == UR_RESULT_ERROR_OUT_OF_HOST_MEMORY ||
+      Error == UR_RESULT_ERROR_OUT_OF_DEVICE_MEMORY) {
     Context->getKernelProgramCache().reset();
     Error = doLink();
   }
@@ -2427,7 +2428,8 @@ ProgramManager::link(const device_image_plain &DeviceImage,
   };
   ur_result_t Error = doLink();
   if (Error == UR_RESULT_ERROR_OUT_OF_RESOURCES ||
-      Error == UR_RESULT_ERROR_OUT_OF_HOST_MEMORY) {
+      Error == UR_RESULT_ERROR_OUT_OF_HOST_MEMORY ||
+      Error == UR_RESULT_ERROR_OUT_OF_DEVICE_MEMORY) {
     ContextImpl->getKernelProgramCache().reset();
     Error = doLink();
   }

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

@@ -35,28 +35,24 @@ static sycl::unittest::UrImageArray<2> ImgArray{Img};
 
 static int nProgramCreate = 0;
 static volatile bool outOfResourcesToggle = false;
-static volatile bool outOfHostMemoryToggle = false;
+static volatile ur_result_t ErrorCode = UR_RESULT_SUCCESS;
 
 static ur_result_t redefinedProgramCreateWithIL(void *) {
   ++nProgramCreate;
   if (outOfResourcesToggle) {
     outOfResourcesToggle = false;
-    return UR_RESULT_ERROR_OUT_OF_RESOURCES;
+    return ErrorCode;
   }
   return UR_RESULT_SUCCESS;
 }
 
-static ur_result_t redefinedProgramCreateWithILOutOfHostMemory(void *) {
-  ++nProgramCreate;
-  if (outOfHostMemoryToggle) {
-    outOfHostMemoryToggle = false;
-    return UR_RESULT_ERROR_OUT_OF_HOST_MEMORY;
-  }
-  return UR_RESULT_SUCCESS;
-}
+// Parameterized test fixture
+class OutOfResourcesTestSuite : public ::testing::TestWithParam<ur_result_t> {};
 
-TEST(OutOfResourcesTest, urProgramCreate) {
+TEST_P(OutOfResourcesTestSuite, urProgramCreate) {
+  nProgramCreate = 0;
   sycl::unittest::UrMock<> Mock;
+  ErrorCode = GetParam();
   mock::getCallbacks().set_before_callback("urProgramCreateWithIL",
                                            &redefinedProgramCreateWithIL);
 
@@ -116,92 +112,21 @@ TEST(OutOfResourcesTest, urProgramCreate) {
   }
 }
 
-TEST(OutOfHostMemoryTest, urProgramCreate) {
-  // Reset to zero.
-  nProgramCreate = 0;
-
-  sycl::unittest::UrMock<> Mock;
-  mock::getCallbacks().set_before_callback(
-      "urProgramCreateWithIL", &redefinedProgramCreateWithILOutOfHostMemory);
-
-  sycl::platform Plt{sycl::platform()};
-  sycl::context Ctx{Plt};
-  auto CtxImpl = detail::getSyclObjImpl(Ctx);
-  queue q(Ctx, default_selector_v);
-
-  int runningTotal = 0;
-  // Cache is empty, so one urProgramCreateWithIL call.
-  q.single_task<class OutOfResourcesKernel1>([] {});
-  EXPECT_EQ(nProgramCreate, runningTotal += 1);
-
-  // Now, we make the next urProgramCreateWithIL call fail with
-  // UR_RESULT_ERROR_OUT_OF_HOST_MEMORY. The caching mechanism should catch
-  // this, clear the cache, and retry the urProgramCreateWithIL.
-  outOfHostMemoryToggle = true;
-  q.single_task<class OutOfResourcesKernel2>([] {});
-  EXPECT_FALSE(outOfHostMemoryToggle);
-  EXPECT_EQ(nProgramCreate, runningTotal += 2);
-  {
-    detail::KernelProgramCache::ProgramCache &Cache =
-        CtxImpl->getKernelProgramCache().acquireCachedPrograms().get();
-    EXPECT_EQ(Cache.size(), 1U) << "Expected 1 program in the cache";
-  }
-
-  // The next urProgramCreateWithIL call will fail with
-  // UR_RESULT_ERROR_OUT_OF_HOST_MEMORY. But OutOfResourcesKernel2 is in the
-  // cache, so we expect no new urProgramCreateWithIL calls.
-  outOfHostMemoryToggle = true;
-  q.single_task<class OutOfResourcesKernel2>([] {});
-  EXPECT_TRUE(outOfHostMemoryToggle);
-  EXPECT_EQ(nProgramCreate, runningTotal);
-
-  // OutOfResourcesKernel1 is not in the cache, so we have to
-  // build it. From what we set before, this call will fail,
-  // the cache will clear out, and will try again.
-  q.single_task<class OutOfResourcesKernel1>([] {});
-  EXPECT_FALSE(outOfHostMemoryToggle);
-  EXPECT_EQ(nProgramCreate, runningTotal += 2);
-  {
-    detail::KernelProgramCache::ProgramCache &Cache =
-        CtxImpl->getKernelProgramCache().acquireCachedPrograms().get();
-    EXPECT_EQ(Cache.size(), 1U) << "Expected 1 program in the cache";
-  }
-
-  // Finally, OutOfResourcesKernel1 will be in the cache, but
-  // OutOfResourceKenel2 will not, so one more urProgramCreateWithIL.
-  // Toggle is not set, so this should succeed.
-  q.single_task<class OutOfResourcesKernel1>([] {});
-  q.single_task<class OutOfResourcesKernel2>([] {});
-  EXPECT_EQ(nProgramCreate, runningTotal += 1);
-  {
-    detail::KernelProgramCache::ProgramCache &Cache =
-        CtxImpl->getKernelProgramCache().acquireCachedPrograms().get();
-    EXPECT_EQ(Cache.size(), 2U) << "Expected 2 program in the cache";
-  }
-}
-
 static int nProgramLink = 0;
 
 static ur_result_t redefinedProgramLink(void *) {
   ++nProgramLink;
   if (outOfResourcesToggle) {
     outOfResourcesToggle = false;
-    return UR_RESULT_ERROR_OUT_OF_RESOURCES;
-  }
-  return UR_RESULT_SUCCESS;
-}
-
-static ur_result_t redefinedProgramLinkOutOfHostMemory(void *) {
-  ++nProgramLink;
-  if (outOfHostMemoryToggle) {
-    outOfHostMemoryToggle = false;
-    return UR_RESULT_ERROR_OUT_OF_HOST_MEMORY;
+    return ErrorCode;
   }
   return UR_RESULT_SUCCESS;
 }
 
-TEST(OutOfResourcesTest, urProgramLink) {
+TEST_P(OutOfResourcesTestSuite, urProgramLink) {
+  nProgramLink = 0;
   sycl::unittest::UrMock<> Mock;
+  ErrorCode = GetParam();
   mock::getCallbacks().set_before_callback("urProgramLinkExp",
                                            &redefinedProgramLink);
 
@@ -236,41 +161,8 @@ TEST(OutOfResourcesTest, urProgramLink) {
   }
 }
 
-TEST(OutOfHostMemoryTest, urProgramLink) {
-  // Reset to zero.
-  nProgramLink = 0;
-
-  sycl::unittest::UrMock<> Mock;
-  mock::getCallbacks().set_before_callback(
-      "urProgramLinkExp", &redefinedProgramLinkOutOfHostMemory);
-
-  sycl::platform Plt{sycl::platform()};
-  sycl::context Ctx{Plt};
-  auto CtxImpl = detail::getSyclObjImpl(Ctx);
-  queue q(Ctx, default_selector_v);
-  // Put some programs in the cache
-  q.single_task<class OutOfResourcesKernel1>([] {});
-  q.single_task<class OutOfResourcesKernel2>([] {});
-  {
-    detail::KernelProgramCache::ProgramCache &Cache =
-        CtxImpl->getKernelProgramCache().acquireCachedPrograms().get();
-    EXPECT_EQ(Cache.size(), 2U) << "Expect 2 programs in the cache";
-  }
-
-  auto b1 = sycl::get_kernel_bundle<OutOfResourcesKernel1,
-                                    sycl::bundle_state::object>(Ctx);
-  auto b2 = sycl::get_kernel_bundle<OutOfResourcesKernel2,
-                                    sycl::bundle_state::object>(Ctx);
-  outOfHostMemoryToggle = true;
-  EXPECT_EQ(nProgramLink, 0);
-  auto b3 = sycl::link({b1, b2});
-  EXPECT_FALSE(outOfHostMemoryToggle);
-  // one restart due to out of resources, one link per each of b1 and b2.
-  EXPECT_EQ(nProgramLink, 3);
-  // no programs should be in the cache due to out of resources.
-  {
-    detail::KernelProgramCache::ProgramCache &Cache =
-        CtxImpl->getKernelProgramCache().acquireCachedPrograms().get();
-    EXPECT_EQ(Cache.size(), 0u) << "Expect no programs in the cache";
-  }
-}
+INSTANTIATE_TEST_SUITE_P(
+    OutOfResourcesParameterizedRun, OutOfResourcesTestSuite,
+    ::testing::Values(UR_RESULT_ERROR_OUT_OF_RESOURCES,
+                      UR_RESULT_ERROR_OUT_OF_HOST_MEMORY,
+                      UR_RESULT_ERROR_OUT_OF_DEVICE_MEMORY));