Add new flag

Signed-off-by: James Brodman <[email protected]>

Author: jbrodman

TESTS.md

@@ -52,7 +52,7 @@ PhysicalMemDestroy|measures time spent in zePhysicalMemDestroy on CPU.|<ul></ul>
 ResetCommandList|measures time spent in zeCommandListReset on CPU.|<ul><li>--CopyOnly Create copy only cmdlist (0 or 1)</li><li>--size Size of the buffer</li><li>--sourcePlacement Placement of the source buffer (Device or Host or Shared or non-USM-mapped or non-USMmisaligned or non-USM4KBAligned or non-USM2MBAligned or non-USMmisaligned-imported or non-USM4KBAligned-imported or non-USM2MBAligned-imported)</li></ul>|:heavy_check_mark:|:x:|
 SetKernelArgSvmPointer|measures time spent in clSetKernelArgSVMPointer on CPU.|<ul><li>--allocationSize Size of svm allocations, in bytes</li><li>--allocationsCount Number of allocations</li><li>--reallocate Allocations will be freed and allocated again between setKernelArgs (0 or 1)</li></ul>|:heavy_check_mark:|:heavy_check_mark:|
 SetKernelGroupSize|measures time spent in zeKernelSetGroupSize on CPU.|<ul><li>--asymmetricLocalWorkSize Use asymmetric local workSize (0 or 1)</li></ul>|:heavy_check_mark:|:x:|
-SubmitKernel|measures time spent in submitting a kernel to a SYCL (or SYCL-like) queue on CPU.|<ul><li>--DiscardEvents Create the queue with the discard_events property (0 or 1)</li><li>--Ioq Create the queue with the in_order property (0 or 1)</li><li>--KernelExecTime Approximately how long a single kernel executes, in us</li><li>--MeasureCompletion Measures time taken to complete the submission (default is to measure only submit calls) (0 or 1)</li><li>--NumKernels Number of kernels to submit to the queue</li><li>--Profiling Create the queue with the enable_profiling property (0 or 1)</li></ul>|:heavy_check_mark:|:heavy_check_mark:|
+SubmitKernel|measures time spent in submitting a kernel to a SYCL (or SYCL-like) queue on CPU.|<ul><li>--DiscardEvents Create the queue with the discard_events property (0 or 1)</li><li>--EnqueueFunctions Use the eventless SYCL enqueue functions (0 or 1)</li><li>--Ioq Create the queue with the in_order property (0 or 1)</li><li>--KernelExecTime Approximately how long a single kernel executes, in us</li><li>--MeasureCompletion Measures time taken to complete the submission (default is to measure only submit calls) (0 or 1)</li><li>--NumKernels Number of kernels to submit to the queue</li><li>--Profiling Create the queue with the enable_profiling property (0 or 1)</li></ul>|:heavy_check_mark:|:heavy_check_mark:|
 UsmMemoryAllocation|measures time spent in USM memory allocation APIs.|<ul><li>--measureMode Specifies which APIs to measure (Allocate or Free or Both)</li><li>--size Size to allocate</li><li>--type Type of memory being allocated (Device or Host or Shared)</li></ul>|:heavy_check_mark:|:x:|
 VirtualMemFree|measures time spent in zeVirtualMemFree on CPU.|<ul><li>--freeSize Size in bytes to be freed</li></ul>|:heavy_check_mark:|:x:|
 VirtualMemGetAccessAttrib|measures time spent in zeVirtualMemGetAccessAttribute on CPU.|<ul><li>--size Size in bytes to get the access attribute</li></ul>|:heavy_check_mark:|:x:|

source/benchmarks/api_overhead_benchmark/definitions/submit_kernel.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2022-2023 Intel Corporation
+ * Copyright (C) 2022-2025 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -13,6 +13,7 @@
 struct SubmitKernelArguments : TestCaseArgumentContainer {
     BooleanArgument useProfiling;
     BooleanArgument inOrderQueue;
+    BooleanArgument useEnqueueFunctions;
     BooleanArgument discardEvents;
     PositiveIntegerArgument numKernels;
     PositiveIntegerArgument kernelExecutionTime;
@@ -21,6 +22,7 @@ struct SubmitKernelArguments : TestCaseArgumentContainer {
     SubmitKernelArguments()
         : useProfiling(*this, "Profiling", "Create the queue with the enable_profiling property"),
           inOrderQueue(*this, "Ioq", "Create the queue with the in_order property"),
+          useEnqueueFunctions(*this, "EnqueueFunctions", "Use the eventless SYCL enqueue functions"),
           discardEvents(*this, "DiscardEvents", "Create the queue with the discard_events property"),
           numKernels(*this, "NumKernels", "Number of kernels to submit to the queue"),
           kernelExecutionTime(*this, "KernelExecTime", "Approximately how long a single kernel executes, in us"),

source/benchmarks/api_overhead_benchmark/gtest/submit_kernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2022-2023 Intel Corporation
+ * Copyright (C) 2022-2025 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -14,18 +14,19 @@
 
 [[maybe_unused]] static const inline RegisterTestCase<SubmitKernel> registerTestCase{};
 
-class SubmitKernelTest : public ::testing::TestWithParam<std::tuple<Api, bool, bool, bool, size_t, size_t, bool>> {
+class SubmitKernelTest : public ::testing::TestWithParam<std::tuple<Api, bool, bool, bool, bool, size_t, size_t, bool>> {
 };
 
 TEST_P(SubmitKernelTest, Test) {
     SubmitKernelArguments args{};
     args.api = std::get<0>(GetParam());
     args.useProfiling = std::get<1>(GetParam());
     args.inOrderQueue = std::get<2>(GetParam());
-    args.discardEvents = std::get<3>(GetParam());
-    args.numKernels = std::get<4>(GetParam());
-    args.kernelExecutionTime = std::get<5>(GetParam());
-    args.measureCompletionTime = std::get<6>(GetParam());
+    args.useEnqueueFunctions = std::get<3>(GetParam());
+    args.discardEvents = std::get<4>(GetParam());
+    args.numKernels = std::get<5>(GetParam());
+    args.kernelExecutionTime = std::get<6>(GetParam());
+    args.measureCompletionTime = std::get<7>(GetParam());
     SubmitKernel test;
     test.run(args);
 }
@@ -37,7 +38,8 @@ INSTANTIATE_TEST_SUITE_P(
         ::CommonGtestArgs::allApis(),
         ::testing::Values(false),         // useProfiling
         ::testing::Values(false, true),   // inOrderQueue
+        ::testing::Values(false, true),   // useEnqueueFunctions
         ::testing::Values(false, true),   // discardEvents
         ::testing::Values(10u),           // numKernels
         ::testing::Values(1u),            // kernelExecutionTime
-        ::testing::Values(false, true))); // measureCompletionTime
+        ::testing::Values(false, true))); // measureCompletionTime

source/benchmarks/api_overhead_benchmark/implementations/sycl/submit_kernel_sycl.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2023-2024 Intel Corporation
+ * Copyright (C) 2023-2025 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -72,7 +72,11 @@ static TestResult run(const SubmitKernelArguments &arguments, Statistics &statis
     for (auto i = 0u; i < arguments.iterations; i++) {
         timer.measureStart();
         for (auto iteration = 0u; iteration < arguments.numKernels; iteration++) {
-            queue.parallel_for(range, eat_time);
+            if (arguments.useEnqueueFunctions) {
+                sycl::ext::oneapi::experimental::nd_launch(queue, range, eat_time);
+            } else {
+                queue.parallel_for(range, eat_time);
+            }
         }
 
         if (!arguments.measureCompletionTime) {