diff --git a/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp b/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
index 158e3978e70e7..e24b1c034275f 100644
--- a/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
+++ b/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
@@ -211,7 +211,9 @@ Expected<StringRef> createOutputFile(const Twine &Prefix, StringRef Extension) {
   std::scoped_lock<decltype(TempFilesMutex)> Lock(TempFilesMutex);
   SmallString<128> OutputFile;
   if (SaveTemps) {
-    (Prefix + "." + Extension).toNullTerminatedStringRef(OutputFile);
+    // Generate a unique path name without creating a file
+    sys::fs::createUniquePath(Prefix + "-%%%%%%." + Extension, OutputFile,
+                              /*MakeAbsolute=*/false);
   } else {
     if (std::error_code EC =
             sys::fs::createTemporaryFile(Prefix, Extension, OutputFile))
diff --git a/llvm/include/llvm/Passes/CodeGenPassBuilder.h b/llvm/include/llvm/Passes/CodeGenPassBuilder.h
index 12f99f4d5a2ae..f140359f729ab 100644
--- a/llvm/include/llvm/Passes/CodeGenPassBuilder.h
+++ b/llvm/include/llvm/Passes/CodeGenPassBuilder.h
@@ -29,7 +29,6 @@
 #include "llvm/CodeGen/DwarfEHPrepare.h"
 #include "llvm/CodeGen/ExpandMemCmp.h"
 #include "llvm/CodeGen/ExpandReductions.h"
-#include "llvm/CodeGen/FPBuiltinFnSelection.h"
 #include "llvm/CodeGen/GCMetadata.h"
 #include "llvm/CodeGen/GlobalMerge.h"
 #include "llvm/CodeGen/IndirectBrExpand.h"
@@ -65,6 +64,7 @@
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Transforms/CFGuard.h"
 #include "llvm/Transforms/Scalar/ConstantHoisting.h"
+#include "llvm/Transforms/Scalar/FPBuiltinFnSelection.h"
 #include "llvm/Transforms/Scalar/LoopPassManager.h"
 #include "llvm/Transforms/Scalar/LoopStrengthReduce.h"
 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h"
diff --git a/llvm/include/llvm/CodeGen/FPBuiltinFnSelection.h b/llvm/include/llvm/Transforms/Scalar/FPBuiltinFnSelection.h
similarity index 74%
rename from llvm/include/llvm/CodeGen/FPBuiltinFnSelection.h
rename to llvm/include/llvm/Transforms/Scalar/FPBuiltinFnSelection.h
index 9a27dbe3da38f..228137cb38a75 100644
--- a/llvm/include/llvm/CodeGen/FPBuiltinFnSelection.h
+++ b/llvm/include/llvm/Transforms/Scalar/FPBuiltinFnSelection.h
@@ -1,4 +1,4 @@
-//===- FPBuiltinFnSelection.h - Pre-ISel intrinsic lowering pass ----------===//
+//===- FPBuiltinFnSelection.h - fpbuiltin intrinsic lowering pass ---------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -10,8 +10,8 @@
 // llvm.fpbuiltin.* intrinsics.
 //
 //===----------------------------------------------------------------------===//
-#ifndef LLVM_CODEGEN_FPBUILTINFNSELECTION_H
-#define LLVM_CODEGEN_FPBUILTINFNSELECTION_H
+#ifndef LLVM_TRANSFORMS_SCALAR_FPBUILTINFNSELECTION_H
+#define LLVM_TRANSFORMS_SCALAR_FPBUILTINFNSELECTION_H
 
 #include "llvm/IR/PassManager.h"
 
@@ -25,4 +25,4 @@ struct FPBuiltinFnSelectionPass : PassInfoMixin<FPBuiltinFnSelectionPass> {
 
 } // end namespace llvm
 
-#endif // LLVM_CODEGEN_FPBUILTINFNSELECTION_H
+#endif // LLVM_TRANSFORMS_SCALAR_FPBUILTINFNSELECTION_H
diff --git a/llvm/lib/CodeGen/CMakeLists.txt b/llvm/lib/CodeGen/CMakeLists.txt
index 2658b70309674..77bf1b165d0cf 100644
--- a/llvm/lib/CodeGen/CMakeLists.txt
+++ b/llvm/lib/CodeGen/CMakeLists.txt
@@ -63,7 +63,6 @@ add_llvm_component_library(LLVMCodeGen
   ExpandVectorPredication.cpp
   FaultMaps.cpp
   FEntryInserter.cpp
-  FPBuiltinFnSelection.cpp
   FinalizeISel.cpp
   FixupStatepointCallerSaved.cpp
   FuncletLayout.cpp
diff --git a/llvm/lib/Transforms/Scalar/CMakeLists.txt b/llvm/lib/Transforms/Scalar/CMakeLists.txt
index ba09ebf8b04c4..abd119d43df26 100644
--- a/llvm/lib/Transforms/Scalar/CMakeLists.txt
+++ b/llvm/lib/Transforms/Scalar/CMakeLists.txt
@@ -14,6 +14,7 @@ add_llvm_component_library(LLVMScalarOpts
   EarlyCSE.cpp
   FlattenCFGPass.cpp
   Float2Int.cpp
+  FPBuiltinFnSelection.cpp
   GuardWidening.cpp
   GVN.cpp
   GVNHoist.cpp
@@ -97,4 +98,5 @@ add_llvm_component_library(LLVMScalarOpts
   InstCombine
   Support
   TransformUtils
+  TargetParser
   )
diff --git a/llvm/lib/CodeGen/FPBuiltinFnSelection.cpp b/llvm/lib/Transforms/Scalar/FPBuiltinFnSelection.cpp
similarity index 98%
rename from llvm/lib/CodeGen/FPBuiltinFnSelection.cpp
rename to llvm/lib/Transforms/Scalar/FPBuiltinFnSelection.cpp
index a4b6a34a2326a..dd423b42ab0ec 100644
--- a/llvm/lib/CodeGen/FPBuiltinFnSelection.cpp
+++ b/llvm/lib/Transforms/Scalar/FPBuiltinFnSelection.cpp
@@ -1,4 +1,4 @@
-//===- FPBuiltinFnSelection.cpp - Pre-ISel intrinsic lowering pass --------===//
+//===- FPBuiltinFnSelection.cpp - fpbuiltin intrinsic lowering pass -------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -11,7 +11,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/CodeGen/FPBuiltinFnSelection.h"
+#include "llvm/Transforms/Scalar/FPBuiltinFnSelection.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/CodeGen/Passes.h"
diff --git a/sycl/test-e2e/KernelFusion/cancel_fusion.cpp b/sycl/test-e2e/KernelFusion/cancel_fusion.cpp
index 9be8d8cbbd140..8141d22534fbb 100644
--- a/sycl/test-e2e/KernelFusion/cancel_fusion.cpp
+++ b/sycl/test-e2e/KernelFusion/cancel_fusion.cpp
@@ -1,8 +1,12 @@
-// RUN: %{build} %{embed-ir} -o %t.out
-// RUN: %{run} %t.out
+// RUN: %{build} -o %t.out
+// RUN: env SYCL_PI_TRACE=2 %{run} %t.out 2>&1 | FileCheck %s
 
 // Test cancel fusion
 
+// As fusion is cancelled, this test launches two kernels.
+// CHECK-COUNT-2: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
+
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/codeplay/experimental/fusion_wrapper.hpp>
 #include <sycl/properties/all_properties.hpp>
diff --git a/sycl/test-e2e/KernelFusion/complete_fusion.cpp b/sycl/test-e2e/KernelFusion/complete_fusion.cpp
index f49bb70fc6d9b..e9c79048c77cf 100644
--- a/sycl/test-e2e/KernelFusion/complete_fusion.cpp
+++ b/sycl/test-e2e/KernelFusion/complete_fusion.cpp
@@ -1,8 +1,12 @@
 // RUN: %{build} %{embed-ir} -o %t.out
-// RUN: %{run} %t.out
+// RUN: env SYCL_PI_TRACE=2 %{run} %t.out 2>&1 | FileCheck %s
 
 // Test complete fusion without any internalization
 
+// The two kernels are fused, so only a single, fused kernel is launched.
+// CHECK-COUNT-1: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
+
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/codeplay/experimental/fusion_wrapper.hpp>
 #include <sycl/properties/all_properties.hpp>
diff --git a/sycl/test-e2e/KernelFusion/cooperative_kernel.cpp b/sycl/test-e2e/KernelFusion/cooperative_kernel.cpp
index 0faf05d84e2e0..ed9d4738b5c0d 100644
--- a/sycl/test-e2e/KernelFusion/cooperative_kernel.cpp
+++ b/sycl/test-e2e/KernelFusion/cooperative_kernel.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=2 %{run} %t.out 2>&1 | FileCheck %s
 
 // Test cooperative kernels are not fused
diff --git a/sycl/test-e2e/KernelFusion/different_nd_ranges.cpp b/sycl/test-e2e/KernelFusion/different_nd_ranges.cpp
index d9c99f4a9271f..ce7cbd0432b34 100644
--- a/sycl/test-e2e/KernelFusion/different_nd_ranges.cpp
+++ b/sycl/test-e2e/KernelFusion/different_nd_ranges.cpp
@@ -1,11 +1,13 @@
-// RUN: %{build} -o %t.out
-// RUN: env SYCL_RT_WARNING_LEVEL=1 SYCL_PARALLEL_FOR_RANGE_ROUNDING_PARAMS=16:32:64 \
-// RUN:   %{run} %t.out 2>&1 | FileCheck %s
+// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: env SYCL_PI_TRACE=2 env SYCL_RT_WARNING_LEVEL=1 \
+// RUN: SYCL_PARALLEL_FOR_RANGE_ROUNDING_PARAMS=16:32:64 %{run} %t.out 2>&1 \
+// RUN: | FileCheck %s --implicit-check-not "ERROR: JIT compilation for kernel fusion failed with message:"
 
 // Test complete fusion of kernels with different ND-ranges.
 
 // Kernels with different ND-ranges should be fused.
-// CHECK-NOT: Cannot fuse kernels with different offsets or local sizes
+// CHECK-COUNT-26: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
 
 #include <sycl/detail/core.hpp>
 
@@ -252,22 +254,6 @@ int main() {
   const auto R5 = {5ul};
   test({RangeDesc{{10}, R5}, RangeDesc{{20}, R5}, RangeDesc{{30}, R5}});
 
-  // Two 1-D kernels with different global sizes and a 2-D kernel with more
-  // work-items and specified (equal) local size.
-  test({RangeDesc{{10}, R2}, RangeDesc{{20}, R2}, RangeDesc{{10, 10}, {2, 1}}});
-
-  // Three 2-D kernels with different global sizes.
-  test({RangeDesc{{10, 15}, {2, 5}}, RangeDesc{{20, 10}, {2, 5}},
-        RangeDesc{{10, 5}, {2, 5}}});
-
-  // Three 3-D kernels with different global sizes.
-  test({RangeDesc{{10, 4, 2}, {5, 2, 1}}, RangeDesc{{20, 2, 4}, {5, 2, 1}},
-        RangeDesc{{10, 2, 4}, {5, 2, 1}}});
-
-  // 1-D, 2-D and 3-D kernels with different global sizes.
-  test({RangeDesc{{10}, R5}, RangeDesc{{10, 1}, {5, 1}},
-        RangeDesc{{10, 1, 1}, {5, 1, 1}}});
-
   // Test global sizes that trigger the rounded range kernel insertion.
   // Note that we lower the RR threshold when running this test.
   test({RangeDesc{67}, RangeDesc{87}, RangeDesc{64}});
diff --git a/sycl/test-e2e/KernelFusion/event_wait_cancel.cpp b/sycl/test-e2e/KernelFusion/event_wait_cancel.cpp
index 76b75d8c3a44d..1ad1b9cc5ba25 100644
--- a/sycl/test-e2e/KernelFusion/event_wait_cancel.cpp
+++ b/sycl/test-e2e/KernelFusion/event_wait_cancel.cpp
@@ -1,5 +1,5 @@
 // REQUIRES: aspect-usm_shared_allocations
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
 
 // Test validity of events after cancel_fusion.
diff --git a/sycl/test-e2e/KernelFusion/internal_explicit_dependency.cpp b/sycl/test-e2e/KernelFusion/internal_explicit_dependency.cpp
index 693abaf37f916..67285414e2a26 100644
--- a/sycl/test-e2e/KernelFusion/internal_explicit_dependency.cpp
+++ b/sycl/test-e2e/KernelFusion/internal_explicit_dependency.cpp
@@ -1,10 +1,14 @@
 // REQUIRES: aspect-usm_shared_allocations
 // RUN: %{build} %{embed-ir} -o %t.out
-// RUN: %{run} %t.out
+// RUN: env SYCL_PI_TRACE=2 %{run} %t.out | FileCheck %s
 
 // Test complete fusion where one kernel in the fusion list specifies an
 // explicit dependency (via events) on another kernel in the fusion list.
 
+// The two kernels are fused, so only a single, fused kernel is launched.
+// CHECK-COUNT-1: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
+
 #include "fusion_event_test_common.h"
 
 #include <sycl/detail/core.hpp>
diff --git a/sycl/test-e2e/KernelFusion/jit_caching_multitarget.cpp b/sycl/test-e2e/KernelFusion/jit_caching_multitarget.cpp
index cfb192ecb276b..57c13be6907d6 100644
--- a/sycl/test-e2e/KernelFusion/jit_caching_multitarget.cpp
+++ b/sycl/test-e2e/KernelFusion/jit_caching_multitarget.cpp
@@ -1,5 +1,5 @@
 // REQUIRES: (gpu && (hip || cuda)), cpu
-// RUN: %{build} -fsycl-embed-ir -O2 -o %t.out
+// RUN: %{build} %{embed-ir} -O2 -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run-unfiltered-devices} %t.out 2>&1 | FileCheck %s --implicit-check-not "WRONG a VALUE" --implicit-check-not "WRONG b VALUE"
 
 // Test caching for JIT fused kernels when devices with different architectures
diff --git a/sycl/test-e2e/KernelFusion/math_function.cpp b/sycl/test-e2e/KernelFusion/math_function.cpp
index 49f2fd04c5b93..bb4a76f9fbfb7 100644
--- a/sycl/test-e2e/KernelFusion/math_function.cpp
+++ b/sycl/test-e2e/KernelFusion/math_function.cpp
@@ -1,8 +1,12 @@
 // RUN: %{build} %{embed-ir} -o %t.out
-// RUN: %{run} %t.out
+// RUN: env SYCL_PI_TRACE=2 %{run} %t.out | FileCheck %s
 
 // Test fusion of a kernel using a math function.
 
+// The two kernels are fused, so only a single, fused kernel is launched.
+// CHECK-COUNT-1: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
+
 #include <sycl/detail/core.hpp>
 
 #include <sycl/builtins.hpp>
diff --git a/sycl/test-e2e/KernelFusion/non-kernel-cg.cpp b/sycl/test-e2e/KernelFusion/non-kernel-cg.cpp
index 3ae4b5be72712..60a4fea7837df 100644
--- a/sycl/test-e2e/KernelFusion/non-kernel-cg.cpp
+++ b/sycl/test-e2e/KernelFusion/non-kernel-cg.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=2 %{run} %t.out 2>&1 | FileCheck %s
 
 // Test non-kernel device command groups are not fused
diff --git a/sycl/test-e2e/KernelFusion/non_unit_local_size.cpp b/sycl/test-e2e/KernelFusion/non_unit_local_size.cpp
index db369b8ee0f37..1cc03bc7ad0c3 100644
--- a/sycl/test-e2e/KernelFusion/non_unit_local_size.cpp
+++ b/sycl/test-e2e/KernelFusion/non_unit_local_size.cpp
@@ -1,9 +1,13 @@
 // RUN: %{build} %{embed-ir} -o %t.out
-// RUN: %{run} %t.out
+// RUN: env SYCL_PI_TRACE=2 %{run} %t.out | FileCheck %s
 
 // Test complete fusion with local internalization specified on the
 // accessors, where each work-item processes multiple data-items.
 
+// The two kernels are fused, so only a single, fused kernel is launched.
+// CHECK-COUNT-1: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
+
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/codeplay/experimental/fusion_wrapper.hpp>
 #include <sycl/properties/all_properties.hpp>
diff --git a/sycl/test-e2e/KernelFusion/sync_acc_mem_op.cpp b/sycl/test-e2e/KernelFusion/sync_acc_mem_op.cpp
index 825b0818eb344..80a67c7d9dc47 100644
--- a/sycl/test-e2e/KernelFusion/sync_acc_mem_op.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_acc_mem_op.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Windows doesn't yet have full shutdown().
diff --git a/sycl/test-e2e/KernelFusion/sync_buffer_destruction.cpp b/sycl/test-e2e/KernelFusion/sync_buffer_destruction.cpp
index a221568e4e4f8..a0564b8d9adc1 100644
--- a/sycl/test-e2e/KernelFusion/sync_buffer_destruction.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_buffer_destruction.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Windows doesn't yet have full shutdown().
diff --git a/sycl/test-e2e/KernelFusion/sync_event_wait.cpp b/sycl/test-e2e/KernelFusion/sync_event_wait.cpp
index 9f93852dbcc82..eb70f197cde3d 100644
--- a/sycl/test-e2e/KernelFusion/sync_event_wait.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_event_wait.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Test fusion cancellation on event::wait() happening before
diff --git a/sycl/test-e2e/KernelFusion/sync_host_accessor.cpp b/sycl/test-e2e/KernelFusion/sync_host_accessor.cpp
index 7aa702767d93e..4e45646e8880b 100644
--- a/sycl/test-e2e/KernelFusion/sync_host_accessor.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_host_accessor.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Windows doesn't yet have full shutdown().
diff --git a/sycl/test-e2e/KernelFusion/sync_host_task.cpp b/sycl/test-e2e/KernelFusion/sync_host_task.cpp
index 5b1eaaac745b8..40754d4e7a8d9 100644
--- a/sycl/test-e2e/KernelFusion/sync_host_task.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_host_task.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Windows doesn't yet have full shutdown().
diff --git a/sycl/test-e2e/KernelFusion/sync_queue_destruction.cpp b/sycl/test-e2e/KernelFusion/sync_queue_destruction.cpp
index 24d50197571c3..a4063e40de041 100644
--- a/sycl/test-e2e/KernelFusion/sync_queue_destruction.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_queue_destruction.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Windows doesn't yet have full shutdown().
diff --git a/sycl/test-e2e/KernelFusion/sync_queue_wait.cpp b/sycl/test-e2e/KernelFusion/sync_queue_wait.cpp
index 7cf6197c7d1ec..f073e28933e4a 100644
--- a/sycl/test-e2e/KernelFusion/sync_queue_wait.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_queue_wait.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Windows doesn't yet have full shutdown().
diff --git a/sycl/test-e2e/KernelFusion/sync_second_queue.cpp b/sycl/test-e2e/KernelFusion/sync_second_queue.cpp
index 836d9984e824f..66ad1f7561f11 100644
--- a/sycl/test-e2e/KernelFusion/sync_second_queue.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_second_queue.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Windows doesn't yet have full shutdown().
diff --git a/sycl/test-e2e/KernelFusion/sync_two_queues_requirement.cpp b/sycl/test-e2e/KernelFusion/sync_two_queues_requirement.cpp
index 9bc209552c35a..969f45625df82 100644
--- a/sycl/test-e2e/KernelFusion/sync_two_queues_requirement.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_two_queues_requirement.cpp
@@ -1,5 +1,5 @@
 // For this test, complete_fusion must be supported.
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 
 // Test fusion cancellation for requirement between two active fusions.
diff --git a/sycl/test-e2e/KernelFusion/sync_usm_mem_op.cpp b/sycl/test-e2e/KernelFusion/sync_usm_mem_op.cpp
index fa0ae3f1b31d9..92fb80af18b44 100644
--- a/sycl/test-e2e/KernelFusion/sync_usm_mem_op.cpp
+++ b/sycl/test-e2e/KernelFusion/sync_usm_mem_op.cpp
@@ -1,4 +1,4 @@
-// RUN: %{build} %{embed-ir} -o %t.out
+// RUN: %{build} -o %t.out
 // RUN: env SYCL_RT_WARNING_LEVEL=1 %{run} %t.out 2>&1 | FileCheck %s
 // Windows doesn't yet have full shutdown().
 // UNSUPPORTED: ze_debug && windows
diff --git a/sycl/test-e2e/KernelFusion/usm_no_dependencies.cpp b/sycl/test-e2e/KernelFusion/usm_no_dependencies.cpp
index f3152cee4c413..6ad81d579226d 100644
--- a/sycl/test-e2e/KernelFusion/usm_no_dependencies.cpp
+++ b/sycl/test-e2e/KernelFusion/usm_no_dependencies.cpp
@@ -1,9 +1,13 @@
 // REQUIRES: aspect-usm_shared_allocations
 // RUN: %{build} %{embed-ir} -o %t.out
-// RUN: %{run} %t.out
+// RUN: env SYCL_PI_TRACE=2 %{run} %t.out | FileCheck %s
 
 // Test complete fusion using USM pointers.
 
+// The two kernels are fused, so only a single, fused kernel is launched.
+// CHECK-COUNT-1: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
+
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/codeplay/experimental/fusion_wrapper.hpp>
 #include <sycl/properties/all_properties.hpp>
diff --git a/sycl/test-e2e/KernelFusion/work_group_barrier.cpp b/sycl/test-e2e/KernelFusion/work_group_barrier.cpp
index 39a5a4cdf57ca..fe09e4fb3050e 100644
--- a/sycl/test-e2e/KernelFusion/work_group_barrier.cpp
+++ b/sycl/test-e2e/KernelFusion/work_group_barrier.cpp
@@ -1,9 +1,13 @@
 // RUN: %{build} %{embed-ir} -o %t.out
-// RUN: %{run} %t.out
+// RUN: env SYCL_PI_TRACE=2 %{run} %t.out | FileCheck %s
 
 // Test complete fusion with a combination of kernels that require a work-group
 // barrier to be inserted by fusion.
 
+// The two kernels are fused, so only a single, fused kernel is launched.
+// CHECK-COUNT-1: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
+
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/codeplay/experimental/fusion_wrapper.hpp>
 #include <sycl/properties/all_properties.hpp>
diff --git a/sycl/test-e2e/KernelFusion/wrapped_usm.cpp b/sycl/test-e2e/KernelFusion/wrapped_usm.cpp
index 1254026223aa7..f011ad3b297be 100644
--- a/sycl/test-e2e/KernelFusion/wrapped_usm.cpp
+++ b/sycl/test-e2e/KernelFusion/wrapped_usm.cpp
@@ -1,9 +1,13 @@
 // REQUIRES: aspect-usm_shared_allocations
 // RUN: %{build} %{embed-ir} -o %t.out
-// RUN: %{run} %t.out
+// RUN: env SYCL_PI_TRACE=2 %{run} %t.out | FileCheck %s
 
 // Test complete fusion using an wrapped USM pointer as kernel functor argument.
 
+// The two kernels are fused, so only a single, fused kernel is launched.
+// CHECK-COUNT-1: piEnqueueKernelLaunch
+// CHECK-NOT: piEnqueueKernelLaunch
+
 #include <sycl/detail/core.hpp>
 #include <sycl/ext/codeplay/experimental/fusion_wrapper.hpp>
 #include <sycl/properties/all_properties.hpp>