diff --git a/SYCL/Basic/buffer/buffer_create.cpp b/SYCL/Basic/buffer/buffer_create.cpp
index 9509186370..f5c049320c 100755
--- a/SYCL/Basic/buffer/buffer_create.cpp
+++ b/SYCL/Basic/buffer/buffer_create.cpp
@@ -1,4 +1,4 @@
-// REQUIRES: gpu,level_zero
+// REQUIRES: cat,gpu,level_zero
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: env ZE_DEBUG=1 %GPU_RUN_PLACEHOLDER %t.out 2> %t1.out; cat %t1.out %GPU_CHECK_PLACEHOLDER
 
diff --git a/SYCL/Basic/buffer/native_buffer_creation_flags.cpp b/SYCL/Basic/buffer/native_buffer_creation_flags.cpp
index 68e912e3e4..3651ea7dd8 100644
--- a/SYCL/Basic/buffer/native_buffer_creation_flags.cpp
+++ b/SYCL/Basic/buffer/native_buffer_creation_flags.cpp
@@ -1,5 +1,5 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: %CPU_RUN_PLACEHOLDER SYCL_PI_TRACE=-1 %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER SYCL_PI_TRACE=-1 %t.out %CPU_CHECK_PLACEHOLDER
 
 #include <CL/sycl.hpp>
 
diff --git a/SYCL/Basic/enqueue_barrier.cpp b/SYCL/Basic/enqueue_barrier.cpp
index f5262f0adc..cb70a45586 100644
--- a/SYCL/Basic/enqueue_barrier.cpp
+++ b/SYCL/Basic/enqueue_barrier.cpp
@@ -1,7 +1,7 @@
 // RUN: %clangxx -fsycl %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %ACC_RUN_PLACEHOLDER %t.out 2>&1 %ACC_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %ACC_CHECK_PLACEHOLDER
 
 // The test is failing sporadically on Windows OpenCL RTs
 // Disabling on windows until fixed
diff --git a/SYCL/Basic/event_release.cpp b/SYCL/Basic/event_release.cpp
index 37483de619..20d4a19915 100644
--- a/SYCL/Basic/event_release.cpp
+++ b/SYCL/Basic/event_release.cpp
@@ -1,5 +1,5 @@
 // RUN: %clangxx -fsycl %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
 #include <CL/sycl.hpp>
 #include <cassert>
 #include <iostream>
diff --git a/SYCL/Basic/host-task-dependency.cpp b/SYCL/Basic/host-task-dependency.cpp
index 961d1f921c..12ec0e7732 100644
--- a/SYCL/Basic/host-task-dependency.cpp
+++ b/SYCL/Basic/host-task-dependency.cpp
@@ -1,7 +1,7 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out %threads_lib
-// RUN: %CPU_RUN_PLACEHOLDER SYCL_PI_TRACE=-1 %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: %GPU_RUN_PLACEHOLDER SYCL_PI_TRACE=-1 %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
-// RUN: %ACC_RUN_PLACEHOLDER SYCL_PI_TRACE=-1 %t.out 2>&1 %ACC_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER SYCL_PI_TRACE=-1 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER SYCL_PI_TRACE=-1 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER SYCL_PI_TRACE=-1 %t.out %ACC_CHECK_PLACEHOLDER
 //
 // TODO: Behaviour is unstable for level zero on Windows. Enable when fixed.
 // TODO: The test is sporadically fails on CUDA. Enable when fixed.
diff --git a/SYCL/Basic/info_ocl_version.cpp b/SYCL/Basic/info_ocl_version.cpp
index 46ea03fb33..53bfd40768 100644
--- a/SYCL/Basic/info_ocl_version.cpp
+++ b/SYCL/Basic/info_ocl_version.cpp
@@ -1,8 +1,8 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: %HOST_RUN_PLACEHOLDER %t.out
-// RUN: env %CPU_RUN_PLACEHOLDER %t.out
-// RUN: env %GPU_RUN_PLACEHOLDER %t.out
-// RUN: env %ACC_RUN_PLACEHOLDER %t.out
+// RUN: %CPU_RUN_PLACEHOLDER %t.out
+// RUN: %GPU_RUN_PLACEHOLDER %t.out
+// RUN: %ACC_RUN_PLACEHOLDER %t.out
 
 //==--------info_ocl_version.cpp - SYCL objects get_info() test ------------==//
 //
diff --git a/SYCL/Basic/kernel_bundle/kernel_bundle_api.cpp b/SYCL/Basic/kernel_bundle/kernel_bundle_api.cpp
index 6cd72e1079..f9e3894606 100644
--- a/SYCL/Basic/kernel_bundle/kernel_bundle_api.cpp
+++ b/SYCL/Basic/kernel_bundle/kernel_bundle_api.cpp
@@ -3,7 +3,7 @@
 // involved leads to multiple definition of AssertHappened structure due each
 // device image is statically linked against fallback libdevice.
 // RUN: %clangxx -DSYCL_DISABLE_FALLBACK_ASSERT=1 -fsycl -fsycl-device-code-split=per_kernel -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
 //
 // -fsycl-device-code-split is not supported for cuda
 // UNSUPPORTED: cuda || rocm
diff --git a/SYCL/Basic/parallel_for_range_roundup.cpp b/SYCL/Basic/parallel_for_range_roundup.cpp
index 7ae3280d42..23adcf4384 100644
--- a/SYCL/Basic/parallel_for_range_roundup.cpp
+++ b/SYCL/Basic/parallel_for_range_roundup.cpp
@@ -1,5 +1,5 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
-// RUN: env SYCL_PARALLEL_FOR_RANGE_ROUNDING_TRACE=1 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PARALLEL_FOR_RANGE_ROUNDING_TRACE=1 %t.out %GPU_CHECK_PLACEHOLDER
 
 #include <CL/sycl.hpp>
 
diff --git a/SYCL/Basic/queue/release.cpp b/SYCL/Basic/queue/release.cpp
index 523c839a19..d44441f2f3 100644
--- a/SYCL/Basic/queue/release.cpp
+++ b/SYCL/Basic/queue/release.cpp
@@ -1,7 +1,7 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %ACC_RUN_PLACEHOLDER %t.out %ACC_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %ACC_CHECK_PLACEHOLDER
 
 #include <CL/sycl.hpp>
 int main() {
diff --git a/SYCL/Basic/use_pinned_host_memory.cpp b/SYCL/Basic/use_pinned_host_memory.cpp
index a596093313..99274d3614 100644
--- a/SYCL/Basic/use_pinned_host_memory.cpp
+++ b/SYCL/Basic/use_pinned_host_memory.cpp
@@ -1,5 +1,5 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
 
 #include <CL/sycl.hpp>
 
diff --git a/SYCL/Config/config.cpp b/SYCL/Config/config.cpp
index 86709fecc8..152b8a9c21 100644
--- a/SYCL/Config/config.cpp
+++ b/SYCL/Config/config.cpp
@@ -5,6 +5,7 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
+// REQUIRES: ls, grep
 // RUN: %clangxx %debug_option -O0 -fsycl %s -o %t.out
 // RUN: echo "SYCL_PRINT_EXECUTION_GRAPH=always" > %t.cfg
 // RUN: env SYCL_CONFIG_FILE_NAME=%t.cfg %t.out
diff --git a/SYCL/InorderQueue/in_order_property_trace.cpp b/SYCL/InorderQueue/in_order_property_trace.cpp
index db700174ee..9d0cad0336 100644
--- a/SYCL/InorderQueue/in_order_property_trace.cpp
+++ b/SYCL/InorderQueue/in_order_property_trace.cpp
@@ -1,6 +1,6 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
 // SYCL in order and default queue property trace test
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
diff --git a/SYCL/KernelAndProgram/basic.cpp b/SYCL/KernelAndProgram/basic.cpp
index 0a7b7b7211..3db156f257 100644
--- a/SYCL/KernelAndProgram/basic.cpp
+++ b/SYCL/KernelAndProgram/basic.cpp
@@ -2,12 +2,12 @@
 // REQUIRES: opencl || level_zero
 // RUN: rm -rf %t/cache_dir
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %ACC_RUN_PLACEHOLDER %t.out %ACC_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %ACC_RUN_PLACEHOLDER %t.out %ACC_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %CPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %CPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %GPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %GPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %ACC_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %ACC_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
 //
 // The test checks that caching works properly.
 #include "basic.hpp"
diff --git a/SYCL/KernelAndProgram/cache-build-result.cpp b/SYCL/KernelAndProgram/cache-build-result.cpp
index b9237eafe9..ca6b7b6dda 100644
--- a/SYCL/KernelAndProgram/cache-build-result.cpp
+++ b/SYCL/KernelAndProgram/cache-build-result.cpp
@@ -1,7 +1,7 @@
 // RUN: %clangxx -fsycl %s -o %t.out
-// RUN: env SYCL_CACHE_PERSISTENT=1 %CPU_RUN_PLACEHOLDER %t.out
-// RUN: env SYCL_CACHE_PERSISTENT=1 %GPU_RUN_PLACEHOLDER %t.out
-// RUN: env SYCL_CACHE_PERSISTENT=1 %ACC_RUN_PLACEHOLDER %t.out
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 %t.out
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 %t.out
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 %t.out
 // XFAIL: cuda
 #include <CL/sycl.hpp>
 
diff --git a/SYCL/KernelAndProgram/spec_consts.cpp b/SYCL/KernelAndProgram/spec_consts.cpp
index a1511a40a8..c55e7d8899 100644
--- a/SYCL/KernelAndProgram/spec_consts.cpp
+++ b/SYCL/KernelAndProgram/spec_consts.cpp
@@ -3,12 +3,12 @@
 // REQUIRES: opencl || level_zero
 // RUN: rm -rf %t/cache_dir
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %ACC_RUN_PLACEHOLDER %t.out %ACC_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
-// RUN: env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %ACC_RUN_PLACEHOLDER %t.out %ACC_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %CPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %CPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %GPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %GPU_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %ACC_CHECK_PLACEHOLDER --check-prefixes=CHECK-BUILD
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_CACHE_PERSISTENT=1 SYCL_CACHE_DIR=%t/cache_dir SYCL_PI_TRACE=-1 %t.out %ACC_CHECK_PLACEHOLDER --check-prefixes=CHECK-CACHE
 //
 // The test checks that caching works properly for SYCL application containing
 // specialization constant values.
diff --git a/SYCL/Plugin/enqueue-arg-order-buffer.cpp b/SYCL/Plugin/enqueue-arg-order-buffer.cpp
index a2d8d3e265..78905ded33 100644
--- a/SYCL/Plugin/enqueue-arg-order-buffer.cpp
+++ b/SYCL/Plugin/enqueue-arg-order-buffer.cpp
@@ -1,8 +1,8 @@
 // Temporarily disabled due to CUDA context related failures.
 // UNSUPPORTED: cuda
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
 
 #include <CL/sycl.hpp>
 #include <CL/sycl/accessor.hpp>
diff --git a/SYCL/Plugin/enqueue-arg-order-image.cpp b/SYCL/Plugin/enqueue-arg-order-image.cpp
index fe20bcb5d2..2d61ee5dc1 100644
--- a/SYCL/Plugin/enqueue-arg-order-image.cpp
+++ b/SYCL/Plugin/enqueue-arg-order-image.cpp
@@ -3,8 +3,8 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
 // Native images are created with host pointers only with host unified memory
 // support, enforce it for this test.
-// RUN: env SYCL_HOST_UNIFIED_MEMORY=1 SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_HOST_UNIFIED_MEMORY=1 SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_HOST_UNIFIED_MEMORY=1 SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_HOST_UNIFIED_MEMORY=1 SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
 
 #include <CL/sycl.hpp>
 #include <CL/sycl/accessor.hpp>
diff --git a/SYCL/Plugin/interop-opencl.cpp b/SYCL/Plugin/interop-opencl.cpp
index 579af94a06..6f2a9df7bd 100644
--- a/SYCL/Plugin/interop-opencl.cpp
+++ b/SYCL/Plugin/interop-opencl.cpp
@@ -1,9 +1,9 @@
 // REQUIRES: opencl
 
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_DEVICE_FILTER=opencl %CPU_RUN_PLACEHOLDER %t.out
-// RUN: env SYCL_DEVICE_FILTER=opencl %GPU_RUN_PLACEHOLDER %t.out
-// RUN: env SYCL_DEVICE_FILTER=opencl %ACC_RUN_PLACEHOLDER %t.out
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_DEVICE_FILTER=opencl %t.out
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_DEVICE_FILTER=opencl %t.out
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_DEVICE_FILTER=opencl %t.out
 
 //==-- interop-opencl.cpp - SYCL test for OpenCL interop API --------------===//
 //
diff --git a/SYCL/Plugin/sycl-ls.cpp b/SYCL/Plugin/sycl-ls.cpp
index eed3f820b4..3065d1d5e5 100644
--- a/SYCL/Plugin/sycl-ls.cpp
+++ b/SYCL/Plugin/sycl-ls.cpp
@@ -1,3 +1,4 @@
+// REQUIRES: diff, grep
 // RUN: sycl-ls --verbose | grep "Device \[" | wc -l >%t.verbose.out
 // RUN: sycl-ls | wc -l >%t.concise.out
 // RUN: diff %t.verbose.out %t.concise.out
diff --git a/SYCL/Regression/image_access.cpp b/SYCL/Regression/image_access.cpp
index 5ede32aba8..91792d1b60 100644
--- a/SYCL/Regression/image_access.cpp
+++ b/SYCL/Regression/image_access.cpp
@@ -1,7 +1,7 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: %HOST_RUN_PLACEHOLDER %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
 // No execution of FPGA because it does not support images
 //
 // UNSUPPORTED: cuda
diff --git a/SYCL/Regression/pi_release.cpp b/SYCL/Regression/pi_release.cpp
index 20a21fe06f..877ccdf19a 100644
--- a/SYCL/Regression/pi_release.cpp
+++ b/SYCL/Regression/pi_release.cpp
@@ -1,8 +1,8 @@
 // REQUIRES: opencl || level_zero || cuda
 // RUN: %clangxx -fsycl  -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_PI_TRACE=-1 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=-1 %ACC_RUN_PLACEHOLDER %t.out 2>&1 %ACC_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=-1 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=-1 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=-1 %t.out %ACC_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=-1 %t.out %GPU_CHECK_PLACEHOLDER
 
 #include <CL/sycl.hpp>
 
diff --git a/SYCL/Scheduler/HostAccDestruction.cpp b/SYCL/Scheduler/HostAccDestruction.cpp
index 3b2b9f7bf8..a3830646c2 100644
--- a/SYCL/Scheduler/HostAccDestruction.cpp
+++ b/SYCL/Scheduler/HostAccDestruction.cpp
@@ -1,7 +1,7 @@
 // RUN: %clangxx -fsycl -fsycl-dead-args-optimization %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %ACC_RUN_PLACEHOLDER %t.out 2>&1 %ACC_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %ACC_CHECK_PLACEHOLDER
 // UNSUPPORTED: cuda || rocm
 //==---------------------- HostAccDestruction.cpp --------------------------==//
 //
diff --git a/SYCL/Scheduler/InOrderQueueDeps.cpp b/SYCL/Scheduler/InOrderQueueDeps.cpp
index f1d6199a36..b340759ff5 100644
--- a/SYCL/Scheduler/InOrderQueueDeps.cpp
+++ b/SYCL/Scheduler/InOrderQueueDeps.cpp
@@ -1,8 +1,8 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env %HOST_RUN_PLACEHOLDER %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %ACC_RUN_PLACEHOLDER %t.out 2>&1 %ACC_CHECK_PLACEHOLDER
+// RUN: %HOST_RUN_PLACEHOLDER %t.out
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %ACC_CHECK_PLACEHOLDER
 
 // The tested functionality is disabled with Level Zero until it is supported by
 // the plugin.
diff --git a/SYCL/Scheduler/MemObjRemapping.cpp b/SYCL/Scheduler/MemObjRemapping.cpp
index 4a08127bdf..8b28eafb74 100644
--- a/SYCL/Scheduler/MemObjRemapping.cpp
+++ b/SYCL/Scheduler/MemObjRemapping.cpp
@@ -1,6 +1,6 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_HOST_UNIFIED_MEMORY=1 SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_HOST_UNIFIED_MEMORY=1 SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_HOST_UNIFIED_MEMORY=1 SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_HOST_UNIFIED_MEMORY=1 SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
 #include <CL/sycl.hpp>
 #include <cassert>
 #include <cstddef>
diff --git a/SYCL/Scheduler/ReleaseResourcesTest.cpp b/SYCL/Scheduler/ReleaseResourcesTest.cpp
index ce92d9257a..668fcf89c6 100644
--- a/SYCL/Scheduler/ReleaseResourcesTest.cpp
+++ b/SYCL/Scheduler/ReleaseResourcesTest.cpp
@@ -1,8 +1,8 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -fsycl-dead-args-optimization %s -o %t.out
-// RUN: env %HOST_RUN_PLACEHOLDER %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %ACC_RUN_PLACEHOLDER %t.out 2>&1 %ACC_CHECK_PLACEHOLDER
+// RUN: %HOST_RUN_PLACEHOLDER %t.out
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %ACC_CHECK_PLACEHOLDER
 
 //==------------------- ReleaseResourcesTests.cpp --------------------------==//
 //
diff --git a/SYCL/SpecConstants/1.2.1/spec_const_redefine.cpp b/SYCL/SpecConstants/1.2.1/spec_const_redefine.cpp
index b983258fd0..03a691043c 100644
--- a/SYCL/SpecConstants/1.2.1/spec_const_redefine.cpp
+++ b/SYCL/SpecConstants/1.2.1/spec_const_redefine.cpp
@@ -4,8 +4,8 @@
 // program builds at run-time
 // RUN: %clangxx -DSYCL_DISABLE_FALLBACK_ASSERT -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: %HOST_RUN_PLACEHOLDER %t.out
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out 2>&1 %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out 2>&1 %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
 //
 //==----------- spec_const_redefine.cpp ------------------------------------==//
 //
diff --git a/SYCL/SpecConstants/1.2.1/spec_const_redefine_accelerator.cpp b/SYCL/SpecConstants/1.2.1/spec_const_redefine_accelerator.cpp
index 9621e2e5b8..2bd80172f9 100644
--- a/SYCL/SpecConstants/1.2.1/spec_const_redefine_accelerator.cpp
+++ b/SYCL/SpecConstants/1.2.1/spec_const_redefine_accelerator.cpp
@@ -16,4 +16,4 @@
 // XFAIL: opencl && accelerator
 // UNSUPPORTED: cuda || level_zero
 #include "spec_const_redefine_accelerator.cpp"
-// RUN: env SYCL_PI_TRACE=2 %ACC_RUN_PLACEHOLDER %t.out 2>&1 %ACC_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %ACC_CHECK_PLACEHOLDER
diff --git a/SYCL/Tracing/buffer_printers.cpp b/SYCL/Tracing/buffer_printers.cpp
index 5bbadf99be..99734f6d56 100644
--- a/SYCL/Tracing/buffer_printers.cpp
+++ b/SYCL/Tracing/buffer_printers.cpp
@@ -1,7 +1,7 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
-// RUN: env SYCL_PI_TRACE=2 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=2 %ACC_RUN_PLACEHOLDER %t.out %ACC_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=2 %t.out %ACC_CHECK_PLACEHOLDER
 
 #include <CL/sycl.hpp>
 #include <vector>
diff --git a/SYCL/Tracing/pi_tracing_test.cpp b/SYCL/Tracing/pi_tracing_test.cpp
index 8feb8af9c9..bd2ae88bbe 100644
--- a/SYCL/Tracing/pi_tracing_test.cpp
+++ b/SYCL/Tracing/pi_tracing_test.cpp
@@ -1,7 +1,7 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
-// RUN: env SYCL_PI_TRACE=-1 %CPU_RUN_PLACEHOLDER %t.out %CPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=-1 %GPU_RUN_PLACEHOLDER %t.out %GPU_CHECK_PLACEHOLDER
-// RUN: env SYCL_PI_TRACE=-1 %ACC_RUN_PLACEHOLDER %t.out %ACC_CHECK_PLACEHOLDER
+// RUN: %CPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=-1 %t.out %CPU_CHECK_PLACEHOLDER
+// RUN: %GPU_RUN_PLACEHOLDER env SYCL_PI_TRACE=-1 %t.out %GPU_CHECK_PLACEHOLDER
+// RUN: %ACC_RUN_PLACEHOLDER env SYCL_PI_TRACE=-1 %t.out %ACC_CHECK_PLACEHOLDER
 
 // Test tracing of the Plugin Interface
 
diff --git a/SYCL/USM/usm_leak_check.cpp b/SYCL/USM/usm_leak_check.cpp
index c82d9cfb19..f0333e2337 100755
--- a/SYCL/USM/usm_leak_check.cpp
+++ b/SYCL/USM/usm_leak_check.cpp
@@ -1,12 +1,11 @@
-// REQUIRES: level_zero
 // TODO: ZE_DEBUG=4 produces no output on Windows. Enable when fixed.
 // UNSUPPORTED: windows
 
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 
-// RUN: env ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out u 2> %t1.out; cat %t1.out %GPU_CHECK_PLACEHOLDER --check-prefix CHECK-USM
-// RUN: env ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out s 2> %t1.out; cat %t1.out %GPU_CHECK_PLACEHOLDER --check-prefix CHECK-SMALL-BUF
-// RUN: env ZE_DEBUG=4 %GPU_RUN_PLACEHOLDER %t.out l 2> %t1.out; cat %t1.out %GPU_CHECK_PLACEHOLDER --check-prefix CHECK-LARGE-BUF
+// RUN: %GPU_RUN_PLACEHOLDER env ZE_DEBUG=4 %t.out u 2> %t1.out; cat %t1.out %GPU_CHECK_PLACEHOLDER --check-prefix CHECK-USM
+// RUN: %GPU_RUN_PLACEHOLDER env ZE_DEBUG=4 %t.out s 2> %t1.out; cat %t1.out %GPU_CHECK_PLACEHOLDER --check-prefix CHECK-SMALL-BUF
+// RUN: %GPU_RUN_PLACEHOLDER env ZE_DEBUG=4 %t.out l 2> %t1.out; cat %t1.out %GPU_CHECK_PLACEHOLDER --check-prefix CHECK-LARGE-BUF
 
 #include <CL/sycl.hpp>
 using namespace sycl;
diff --git a/SYCL/USM/usm_pooling.cpp b/SYCL/USM/usm_pooling.cpp
index 4b23cf9704..a673b6496b 100755
--- a/SYCL/USM/usm_pooling.cpp
+++ b/SYCL/USM/usm_pooling.cpp
@@ -1,4 +1,4 @@
-// REQUIRES: level_zero
+// REQUIRES: cat, level_zero
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 
 // Allocate 2 items of 2MB. Free 2. Allocate 3 more of 2MB.
diff --git a/SYCL/lit.cfg.py b/SYCL/lit.cfg.py
index 49c76a89c1..ad4855526a 100644
--- a/SYCL/lit.cfg.py
+++ b/SYCL/lit.cfg.py
@@ -6,6 +6,7 @@
 import subprocess
 import tempfile
 from distutils.spawn import find_executable
+from shutil import which
 
 import lit.formats
 import lit.util
@@ -199,8 +200,8 @@
 
 found_at_least_one_device = False
 
-host_run_substitute = "true"
-host_run_on_linux_substitute = "true "
+host_run_substitute = "echo "
+host_run_on_linux_substitute = "echo "
 host_check_substitute = ""
 host_check_on_linux_substitute = ""
 supported_device_types=['cpu', 'gpu', 'acc', 'host']
@@ -215,11 +216,11 @@
     found_at_least_one_device = True
     lit_config.note("Test HOST device")
     host_run_substitute = "env SYCL_DEVICE_FILTER=host "
-    host_check_substitute = "| FileCheck %s"
+    host_check_substitute = "2>&1 | FileCheck %s"
     config.available_features.add('host')
     if platform.system() == "Linux":
         host_run_on_linux_substitute = "env SYCL_DEVICE_FILTER=host "
-        host_check_on_linux_substitute = "| FileCheck %s"
+        host_check_on_linux_substitute = "2>&1 | FileCheck %s"
 else:
     lit_config.warning("HOST device not used")
 
@@ -228,20 +229,21 @@
 config.substitutions.append( ('%HOST_CHECK_PLACEHOLDER',  host_check_substitute) )
 config.substitutions.append( ('%HOST_CHECK_ON_LINUX_PLACEHOLDER',  host_check_on_linux_substitute) )
 
-cpu_run_substitute = "true"
-cpu_run_on_linux_substitute = "true "
+cpu_run_substitute = "echo "
+cpu_run_on_linux_substitute = "echo "
 cpu_check_substitute = ""
 cpu_check_on_linux_substitute = ""
+cpu_err_redirect = ""
 
 if 'cpu' in config.target_devices.split(','):
     found_at_least_one_device = True
     lit_config.note("Test CPU device")
     cpu_run_substitute = "env SYCL_DEVICE_FILTER=cpu,host "
-    cpu_check_substitute = "| FileCheck %s"
+    cpu_check_substitute = "2>&1 | FileCheck %s"
     config.available_features.add('cpu')
     if platform.system() == "Linux":
         cpu_run_on_linux_substitute = "env SYCL_DEVICE_FILTER=cpu,host "
-        cpu_check_on_linux_substitute = "| FileCheck %s"
+        cpu_check_on_linux_substitute = "2>&1 | FileCheck %s"
 else:
     lit_config.warning("CPU device not used")
 
@@ -250,8 +252,8 @@
 config.substitutions.append( ('%CPU_CHECK_PLACEHOLDER',  cpu_check_substitute) )
 config.substitutions.append( ('%CPU_CHECK_ON_LINUX_PLACEHOLDER',  cpu_check_on_linux_substitute) )
 
-gpu_run_substitute = "true"
-gpu_run_on_linux_substitute = "true "
+gpu_run_substitute = "echo "
+gpu_run_on_linux_substitute = "echo "
 gpu_check_substitute = ""
 gpu_l0_check_substitute = ""
 gpu_check_on_linux_substitute = ""
@@ -260,15 +262,15 @@
     found_at_least_one_device = True
     lit_config.note("Test GPU device")
     gpu_run_substitute = " env SYCL_DEVICE_FILTER={SYCL_PLUGIN}:gpu,host ".format(SYCL_PLUGIN=config.sycl_be)
-    gpu_check_substitute = "| FileCheck %s"
+    gpu_check_substitute = "2>&1 | FileCheck %s"
     config.available_features.add('gpu')
 
     if config.sycl_be == "level_zero":
-        gpu_l0_check_substitute = "| FileCheck %s"
+        gpu_l0_check_substitute = "2>&1 | FileCheck %s"
 
     if platform.system() == "Linux":
         gpu_run_on_linux_substitute = "env SYCL_DEVICE_FILTER={SYCL_PLUGIN}:gpu,host ".format(SYCL_PLUGIN=config.sycl_be)
-        gpu_check_on_linux_substitute = "| FileCheck %s"
+        gpu_check_on_linux_substitute = "2>&1 | FileCheck %s"
 
 else:
     lit_config.warning("GPU device not used")
@@ -279,13 +281,13 @@
 config.substitutions.append( ('%GPU_L0_CHECK_PLACEHOLDER',  gpu_l0_check_substitute) )
 config.substitutions.append( ('%GPU_CHECK_ON_LINUX_PLACEHOLDER',  gpu_check_on_linux_substitute) )
 
-acc_run_substitute = "true"
+acc_run_substitute = "echo "
 acc_check_substitute = ""
 if 'acc' in config.target_devices.split(','):
     found_at_least_one_device = True
     lit_config.note("Tests accelerator device")
     acc_run_substitute = " env SYCL_DEVICE_FILTER=acc "
-    acc_check_substitute = "| FileCheck %s"
+    acc_check_substitute = "2>&1 | FileCheck %s"
     config.available_features.add('accelerator')
 else:
     lit_config.warning("Accelerator device not used")
@@ -324,6 +326,16 @@
     else:
         lit_config.warning("Couldn't find pre-installed AOT device compiler " + aot_tool)
 
+# Check for some CLI tools, that may not present on the system
+if which("ls") is not None:
+    config.available_features.add("ls")
+if which("grep") is not None:
+    config.available_features.add("grep")
+if which("cat") is not None:
+    config.available_features.add("cat")
+if which("diff") is not None:
+    config.available_features.add("diff")
+
 # Set timeout for test 1 min
 try:
     import psutil