[SYCL] Enable indirect access tracking for multi-threaded assertion tests (#538)

Memory pooling is being disabled by default:
intel/llvm#4799

This makes necessary to enable indirect access tracking feature for
multi-threaded assertion tests. Memory pooling was resulting in deferred
memory release that's why this issue was not reproducing before.

Author: againull

SYCL/Assert/assert_in_simultaneous_kernels.cpp

@@ -4,7 +4,13 @@
 // RUN: %clangxx -DSYCL_ENABLE_FALLBACK_ASSERT -fsycl -fsycl-targets=%sycl_triple %s -o %t.out %threads_lib
 // RUN: %CPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %CPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt
-// RUN: %GPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
+//
+// Since this is a multi-threaded application enable memory tracking and
+// deferred release feature in the Level Zero plugin to avoid releasing memory
+// too early. This is necessary because currently SYCL RT sets indirect access
+// flag for all kernels and the Level Zero runtime doesn't support deferred
+// release yet.
+// RUN: env SYCL_PI_LEVEL_ZERO_TRACK_INDIRECT_ACCESS_MEMORY=1 %GPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %GPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt
 // RUN: %ACC_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %ACC_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt

SYCL/Assert/assert_in_simultaneous_kernels_win.cpp

@@ -2,7 +2,13 @@
 // RUN: %clangxx -DSYCL_ENABLE_FALLBACK_ASSERT -fsycl -fsycl-targets=%sycl_triple %s -o %t.out %threads_lib
 // RUN: %CPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %CPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt
-// RUN: %GPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
+//
+// Since this is a multi-threaded application enable memory tracking and
+// deferred release feature in the Level Zero plugin to avoid releasing memory
+// too early. This is necessary because currently SYCL RT sets indirect access
+// flag for all kernels and the Level Zero runtime doesn't support deferred
+// release yet.
+// RUN: env SYCL_PI_LEVEL_ZERO_TRACK_INDIRECT_ACCESS_MEMORY=1 %GPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %GPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt
 // RUN: %ACC_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %ACC_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt

SYCL/Assert/assert_in_simultaneously_multiple_tus.cpp

@@ -3,7 +3,13 @@
 // RUN: %clangxx -DSYCL_ENABLE_FALLBACK_ASSERT -fsycl -fsycl-targets=%sycl_triple -I %S/Inputs %s %S/Inputs/kernels_in_file2.cpp -o %t.out %threads_lib
 // RUN: %CPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %CPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt
-// RUN: %GPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
+//
+// Since this is a multi-threaded application enable memory tracking and
+// deferred release feature in the Level Zero plugin to avoid releasing memory
+// too early. This is necessary because currently SYCL RT sets indirect access
+// flag for all kernels and the Level Zero runtime doesn't support deferred
+// release yet.
+// RUN: env SYCL_PI_LEVEL_ZERO_TRACK_INDIRECT_ACCESS_MEMORY=1 %GPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %GPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt
 // RUN: %ACC_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %ACC_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt

SYCL/Assert/assert_in_simultaneously_multiple_tus_one_ndebug.cpp

@@ -3,7 +3,13 @@
 // RUN: %clangxx -DSYCL_ENABLE_FALLBACK_ASSERT -fsycl -fsycl-targets=%sycl_triple -DDEFINE_NDEBUG_INFILE2 -I %S/Inputs %S/assert_in_simultaneously_multiple_tus.cpp %S/Inputs/kernels_in_file2.cpp -o %t.out %threads_lib
 // RUN: %CPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %CPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt
-// RUN: %GPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
+//
+// Since this is a multi-threaded application enable memory tracking and
+// deferred release feature in the Level Zero plugin to avoid releasing memory
+// too early. This is necessary because currently SYCL RT sets indirect access
+// flag for all kernels and the Level Zero runtime doesn't support deferred
+// release yet.
+// RUN: env SYCL_PI_LEVEL_ZERO_TRACK_INDIRECT_ACCESS_MEMORY=1 %GPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %GPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt
 // RUN: %ACC_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %ACC_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt