[SYCL] [L0] Specify standard command lists explicitly when the test depends on it. (#1585)

* Make use of standard command lists explicit, when the test depends on it.
* Extended a test to work with imm cmdlists.

Author: rdeodhar

SYCL/Plugin/interop-level-zero-thread.cpp

@@ -168,6 +168,7 @@ sycl::event operation(sycl::queue q) {
       {l0_event, sycl::ext::oneapi::level_zero::ownership::keep},
       q.get_context());
 
+  zeEventHostSignal(l0_event);
   auto return_event = q.ext_oneapi_submit_barrier({sycl_event});
   struct operation new_op = {std::move(deps), l0_event, sycl_event,
                              return_event};

SYCL/Plugin/level_zero_device_scope_events.cpp

@@ -1,8 +1,8 @@
 // REQUIRES: gpu, level_zero
 
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
-// RUN: env SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=1 SYCL_PI_TRACE=-1 ZE_DEBUG=1 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=MODE1 %s
-// RUN: env SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=2 SYCL_PI_TRACE=-1 ZE_DEBUG=1 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=MODE2 %s
+// RUN: env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=1 SYCL_PI_TRACE=-1 ZE_DEBUG=1 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=MODE1 %s
+// RUN: env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=2 SYCL_PI_TRACE=-1 ZE_DEBUG=1 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=MODE2 %s
 // UNSUPPORTED: ze_debug-1,ze_debug4
 
 // Checks that with L0 device-scope events enabled the only host-visible L0

SYCL/Plugin/level_zero_events_caching.cpp

@@ -4,15 +4,20 @@
 // UNSUPPORTED: windows
 
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %level_zero_options %s -o %t.out
-// RUN: env SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=0 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CACHING-ENABLED %s
-// RUN: env SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=0 SYCL_PI_LEVEL_ZERO_DISABLE_EVENTS_CACHING=1 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CACHING-DISABLED %s
+// RUN: env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=0 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CACHING-ENABLED %s
+// RUN: env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=0 SYCL_PI_LEVEL_ZERO_DISABLE_EVENTS_CACHING=1 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CACHING-DISABLED-STD %s
+// RUN: env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=1 SYCL_PI_LEVEL_ZERO_DEVICE_SCOPE_EVENTS=0 SYCL_PI_LEVEL_ZERO_DISABLE_EVENTS_CACHING=1 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 | FileCheck --check-prefixes=CACHING-DISABLED-IMM %s
 
 // With events caching we should be reusing them and 9 should be enough.
 // Might require more than one if previous one hasn't been released by the time
 // we need a new one.
 
+// With immediate commandlists there is one additional event created
+// during queue sync, so the expected count increases by 1 to 257.
+
 // CACHING-ENABLED: zeEventCreate = {{[1-9]}}
-// CACHING-DISABLED: zeEventCreate = 256
+// CACHING-DISABLED-STD: zeEventCreate = 256
+// CACHING-DISABLED-IMM: zeEventCreate = 257
 
 // Check event caching modes in the L0 plugin.
 

SYCL/Plugin/level_zero_queue_create.cpp

@@ -3,7 +3,7 @@
 // UNSUPPORTED: windows
 
 // RUN: %clangxx -fsycl -fsycl-unnamed-lambda -fsycl-targets=%sycl_triple %level_zero_options %s -o %t.out
-// RUN: env ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
+// RUN: env SYCL_PI_LEVEL_ZERO_USE_IMMEDIATE_COMMANDLISTS=0 ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
 //
 // CHECK:  zeCommandQueueCreate = 1     \--->         zeCommandQueueDestroy = 1
 // The test is to check that there is only a single level zero queue created