[SYCL][E2E] Use fine-grained includes in Plugin/* tests (#13851)

Author: aelovikov-intel

sycl/test-e2e/Plugin/cuda-max-local-mem-size.cpp

@@ -13,7 +13,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
 
 int main() {
   sycl::queue Q{};

sycl/test-e2e/Plugin/cuda_queue_priority.cpp

@@ -5,8 +5,10 @@
 //
 // Check that queue priority is passed to CUDA runtime
 #define SYCL_EXT_ONEAPI_BACKEND_CUDA_EXPERIMENTAL 1
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/experimental/backend/cuda.hpp>
-#include <sycl/sycl.hpp>
+#include <sycl/properties/all_properties.hpp>
 
 #include <cuda.h>
 

sycl/test-e2e/Plugin/enqueue-arg-order-image.cpp

@@ -14,7 +14,8 @@
 
 #include <iostream>
 
-#include <sycl/sycl.hpp>
+#include <sycl/accessor_image.hpp>
+#include <sycl/detail/core.hpp>
 
 using namespace sycl;
 

sycl/test-e2e/Plugin/interop-cuda-experimental.cpp

@@ -4,8 +4,9 @@
 // RUN: %{run} %t.out
 
 #define SYCL_EXT_ONEAPI_BACKEND_CUDA_EXPERIMENTAL 1
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/experimental/backend/cuda.hpp>
-#include <sycl/sycl.hpp>
 
 #include <cuda.h>
 

sycl/test-e2e/Plugin/interop-device.cpp

@@ -9,7 +9,8 @@
 // doesn't create a "new" SYCL device that isn't equally comparable to one of
 // the root devices in the pre-existin fixed hierarchy.
 
-#include <sycl/sycl.hpp>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 
 int main() {
   auto root_devices = sycl::device::get_devices();

sycl/test-e2e/Plugin/interop-experimental-single-TU-SYCL-CUDA-compilation.cpp

@@ -3,7 +3,7 @@
 // RUN: %{run} %t.out
 
 #include <cuda.h>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
 
 // ------------------------------------------------------------------------- //
 

sycl/test-e2e/Plugin/interop-l0-direct.cpp

@@ -6,8 +6,10 @@
 
 #include <iostream>
 #include <level_zero/ze_api.h>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/backend/level_zero.hpp>
-#include <sycl/sycl.hpp>
+#include <sycl/usm.hpp>
 #include <variant>
 #include <vector>
 

sycl/test-e2e/Plugin/interop-level-zero-buffer-helpers.hpp

@@ -1,11 +1,11 @@
 #pragma once
-#include "interop-level-zero-buffer-helpers.hpp"
 #include <iostream>
-#include <sycl/sycl.hpp>
-// clang-format off
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
+
 #include <level_zero/ze_api.h>
+
 #include <sycl/ext/oneapi/backend/level_zero.hpp>
-// clang-format on
 
 using namespace sycl;
 

sycl/test-e2e/Plugin/interop-level-zero-device.cpp

@@ -2,7 +2,10 @@
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 
-#include <sycl/sycl.hpp>
+#include <numeric>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
 
 template <typename FuncTy>
 void for_each_descendent_dev(sycl::device dev, FuncTy F) {

sycl/test-e2e/Plugin/interop-level-zero-get-native-mem.cpp

@@ -12,8 +12,9 @@
 
 // SYCL
 #include "interop-level-zero-buffer-helpers.hpp"
-#include <sycl/ext/oneapi/backend/level_zero.hpp>
 #include <sycl/detail/core.hpp>
+#include <sycl/detail/host_task_impl.hpp>
+#include <sycl/ext/oneapi/backend/level_zero.hpp>
 
 using namespace sycl;
 

sycl/test-e2e/Plugin/interop-level-zero-image-get-native-mem.cpp

@@ -21,8 +21,13 @@
 // interop-level-zero-image-get-native-mem.cpp
 
 #include <level_zero/ze_api.h>
-#include <sycl.hpp>
+#include <sycl/accessor_image.hpp>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/detail/host_task_impl.hpp>
 #include <sycl/ext/oneapi/backend/level_zero.hpp>
+#include <sycl/ext/oneapi/filter_selector.hpp>
+#include <sycl/stream.hpp>
 using namespace sycl;
 
 int main() {

sycl/test-e2e/Plugin/interop-level-zero-image-ownership.cpp

@@ -27,7 +27,9 @@
 // interop-level-zero-image-ownership.cpp
 
 #include <level_zero/ze_api.h>
-#include <sycl.hpp>
+#include <sycl/accessor_image.hpp>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/backend/level_zero.hpp>
 
 using namespace sycl;

sycl/test-e2e/Plugin/interop-level-zero-image.cpp

@@ -13,7 +13,9 @@
 // -lze_loader interop-level-zero-image.cpp
 
 #include <level_zero/ze_api.h>
-#include <sycl.hpp>
+#include <sycl/accessor_image.hpp>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/backend/level_zero.hpp>
 
 using namespace sycl;

sycl/test-e2e/Plugin/interop-level-zero-thread.cpp

@@ -30,7 +30,9 @@
 #include <level_zero/ze_api.h>
 #include <mutex>
 #include <sycl/backend/level_zero.hpp>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
+#include <sycl/usm.hpp>
 #include <thread>
 #include <vector>
 

sycl/test-e2e/Plugin/interop-opencl-make-kernel-bundle.cpp

@@ -8,7 +8,8 @@
 #include <cassert>
 #include <iostream>
 #include <string>
-#include <sycl/sycl.hpp>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 #include <vector>
 
 char const *source = R"OCL(kernel void do_nothing() {})OCL";

sycl/test-e2e/Plugin/interop-opencl-make-kernel.cpp

@@ -6,8 +6,10 @@
 // Test for OpenCL make_kernel
 
 #include <CL/opencl.h>
+#include <sycl/backend.hpp>
 #include <sycl/backend/opencl.hpp>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/usm.hpp>
 
 using namespace sycl;
 

sycl/test-e2e/Plugin/interop-opencl.cpp

@@ -12,8 +12,10 @@
 //===----------------------------------------------------------------------===//
 
 #include <CL/opencl.h>
+#include <sycl/backend.hpp>
 #include <sycl/backend/opencl.hpp>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/detail/host_task_impl.hpp>
 
 using namespace sycl;
 

sycl/test-e2e/Plugin/level-zero-event-leak.cpp

@@ -36,7 +36,11 @@
 // NOTE: The 1000 value below is to be larger than the "128" heuristic in
 // queue_impl::addSharedEvent.
 
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+
+#ifdef CHECK_INORDER
+#include <sycl/properties/all_properties.hpp>
+#endif
 
 int main(int argc, char **argv) {
   assert(argc == 2 && "Invalid number of arguments");

sycl/test-e2e/Plugin/level_zero_barrier_optimization.cpp

@@ -5,7 +5,8 @@
 // Test to check that we don't insert unnecessary L0 commands for
 // queue::ext_oneapi_submit_barrier() when we have in-order queue.
 
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
 
 class TestKernel;
 sycl::event submitKernel(sycl::queue &Q) {

sycl/test-e2e/Plugin/level_zero_batch_barrier.cpp

@@ -7,7 +7,8 @@
 // extra submissions.
 
 #include <level_zero/ze_api.h>
-#include <sycl/sycl.hpp>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 #include <vector>
 
 using namespace std;

sycl/test-e2e/Plugin/level_zero_batch_test.cpp

@@ -265,10 +265,12 @@
 // CKALL: Test Pass
 // CKALL: Test Pass
 
-#include "sycl/sycl.hpp"
 #include <chrono>
 #include <cmath>
 #include <iostream>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
+#include <sycl/usm.hpp>
 
 void validate(uint32_t *result, uint32_t *expect, size_t n) {
   int error = 0;

sycl/test-e2e/Plugin/level_zero_batch_test_copy_with_compute.cpp

@@ -226,7 +226,8 @@
 // CKALL: Test Pass
 // CKALL: Test Pass
 
-#include "sycl/sycl.hpp"
+#include "sycl/detail/core.hpp"
+#include "sycl/usm.hpp"
 #include <chrono>
 #include <cmath>
 #include <iostream>

sycl/test-e2e/Plugin/level_zero_device_free_mem.cpp

@@ -12,7 +12,7 @@
 // CHECK: Root-device free memory
 
 #include <iostream>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
 using namespace sycl;
 
 int main() {

sycl/test-e2e/Plugin/level_zero_dynamic_batch_test.cpp

@@ -40,10 +40,12 @@
 // CKALL: Test Pass
 // CKALL: Test Pass
 
-#include "sycl/sycl.hpp"
 #include <chrono>
 #include <cmath>
 #include <iostream>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
+#include <sycl/usm.hpp>
 
 void validate(uint32_t *result, uint32_t *expect, size_t n) {
   int error = 0;

sycl/test-e2e/Plugin/level_zero_eager_init.cpp

@@ -12,7 +12,7 @@
 // CHECK-NOT: ZE ---> zeFenceCreate
 //
 
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
 
 #include <array>
 #include <iostream>

sycl/test-e2e/Plugin/level_zero_events_caching.cpp

@@ -15,7 +15,7 @@
 
 // Check event caching modes in the L0 plugin.
 
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
 
 int main() {
   sycl::queue deviceQueue;

sycl/test-e2e/Plugin/level_zero_events_caching_leak.cpp

@@ -8,7 +8,7 @@
 // enabled/disabled.
 
 #include <array>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
 
 constexpr sycl::access::mode sycl_read = sycl::access::mode::read;
 constexpr sycl::access::mode sycl_write = sycl::access::mode::write;

sycl/test-e2e/Plugin/level_zero_ext_intel_cslice.cpp

@@ -23,7 +23,7 @@
 // RUN: %{setup_env} env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 SYCL_PI_LEVEL_ZERO_EXPOSE_CSLICE_IN_AFFINITY_PARTITIONING=1 \
 // RUN:  UR_L0_DEBUG=1 %{run} %t.out 2>&1 | FileCheck %s --check-prefixes=CHECK-PVC
 
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
 
 using namespace sycl;
 

sycl/test-e2e/Plugin/level_zero_ext_intel_queue_index.cpp

@@ -14,7 +14,8 @@
 // Same with Immediate CommandLists
 // RUN: %{setup_env} env SYCL_PI_LEVEL_ZERO_EXPOSE_CSLICE_IN_AFFINITY_PARTITIONING=1 env UR_L0_DEBUG=1 %{run} %t.out 2>&1 | FileCheck %s --check-prefixes=CHECK-PVC
 
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
 
 using namespace sycl;
 

sycl/test-e2e/Plugin/level_zero_imm_cmdlist.cpp

@@ -9,7 +9,7 @@
 // This test checks that immediate commandlists are used and not regular ones on
 // PVC Linux.
 
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
 
 int main(int argc, char **argv) {
   sycl::queue Q;

sycl/test-e2e/Plugin/level_zero_inorder.cpp

@@ -15,7 +15,9 @@
 #include <cassert>
 #include <iostream>
 #include <numeric>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
+#include <sycl/usm.hpp>
 
 static constexpr int MAGIC_NUM1 = 2;
 static constexpr int buffer_size = 100;

sycl/test-e2e/Plugin/level_zero_inorder_interleaving_kernel_copy.cpp

@@ -16,7 +16,9 @@
 #include <cassert>
 #include <iostream>
 #include <numeric>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
+#include <sycl/usm.hpp>
 
 static constexpr int MAGIC_NUM1 = 2;
 static constexpr int buffer_size = 100;

sycl/test-e2e/Plugin/level_zero_interop_memcpy.cpp

@@ -4,8 +4,10 @@
 // RUN: %{run} %t.out
 
 #include <level_zero/ze_api.h>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
 #include <sycl/ext/oneapi/backend/level_zero.hpp>
-#include <sycl/sycl.hpp>
+#include <sycl/usm.hpp>
 
 int main() {
   using namespace sycl;

sycl/test-e2e/Plugin/level_zero_memory_fill.cpp

@@ -9,7 +9,8 @@
 // CHECKL ZE ---> zeCommandListAppendMemoryFill
 
 #include <iostream>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/usm.hpp>
 
 using namespace sycl;
 

sycl/test-e2e/Plugin/level_zero_queue_priority.cpp

@@ -8,7 +8,8 @@
 //
 // With immediate command lists the command lists are recycled between queues in
 // a context.
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
 
 void test(sycl::context &C, sycl::device &D, sycl::property_list Props) {
   sycl::queue Q(C, D, Props);

sycl/test-e2e/Plugin/level_zero_queue_profiling.cpp

@@ -21,7 +21,8 @@
 //
 
 #include <iostream>
-#include <sycl/sycl.hpp>
+#include <sycl/detail/core.hpp>
+#include <sycl/properties/all_properties.hpp>
 using namespace sycl;
 
 int foo(queue &q, int n) {