[SYCL][E2E] Introduce %{sycl_target_opts} substitution (#16614)

aelovikov-intel · web-flow · commit df91a6bce0d3 · 2025-01-15T09:07:04.000-08:00
...and use it to pass `amd_arch` options to compilation only when
compiling for AMD triples instead of unconditional usage in `%clangxx`
substitution.

That would avoid passing the option unnecessary when running `llvm-lit`
with multiple target devices and a given test isn't supported for HIP.
We might not have such configuration presently, but the same logic will
be applied for "split" build/run of E2E tests and we'd definitely want
to build tests for all the possible targets at once.

Another use case for it is the "spirv backend" testing where we'd want
to pass extra option for any compilations targeting spirv triple.
diff --git a/sycl/test-e2e/Assert/assert_in_kernels_ndebug.cpp b/sycl/test-e2e/Assert/assert_in_kernels_ndebug.cpp
@@ -1,4 +1,4 @@
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} -DNDEBUG %S/assert_in_kernels.cpp -o %t.out
+// RUN: %clangxx -fsycl %{sycl_target_opts} -DNDEBUG %S/assert_in_kernels.cpp -o %t.out
 // RUN: %{run} %t.out | FileCheck %s
 //
 // CHECK-NOT: One shouldn't see this message
diff --git a/sycl/test-e2e/Assert/assert_in_multiple_tus_one_ndebug.cpp b/sycl/test-e2e/Assert/assert_in_multiple_tus_one_ndebug.cpp
@@ -9,7 +9,7 @@
 // XFAIL: (opencl && gpu)
 // XFAIL-TRACKER: https://github.com/intel/llvm/issues/11364
 
-// RUN: %clangxx -DSYCL_FALLBACK_ASSERT=1 -fsycl -fsycl-targets=%{sycl_triple} -DDEFINE_NDEBUG_INFILE2 -I %S/Inputs %S/assert_in_multiple_tus.cpp %S/Inputs/kernels_in_file2.cpp -o %t.out
+// RUN: %clangxx -DSYCL_FALLBACK_ASSERT=1 -fsycl %{sycl_target_opts} -DDEFINE_NDEBUG_INFILE2 -I %S/Inputs %S/assert_in_multiple_tus.cpp %S/Inputs/kernels_in_file2.cpp -o %t.out
 // Shouldn't fail on ACC as fallback assert isn't enqueued there
 // RUN: %{run} %t.out &> %t.txt ; FileCheck %s --input-file %t.txt %if fpga %{ --check-prefix=CHECK-ACC %}
 //
diff --git a/sycl/test-e2e/Assert/assert_in_multiple_tus_one_ndebug_win.cpp b/sycl/test-e2e/Assert/assert_in_multiple_tus_one_ndebug_win.cpp
@@ -1,5 +1,5 @@
 // REQUIRES: windows
-// RUN: %clangxx -DSYCL_FALLBACK_ASSERT=1 -fsycl -fsycl-targets=%{sycl_triple} -DDEFINE_NDEBUG_INFILE2 -I %S/Inputs %S/assert_in_multiple_tus.cpp %S/Inputs/kernels_in_file2.cpp -o %t.out
+// RUN: %clangxx -DSYCL_FALLBACK_ASSERT=1 -fsycl %{sycl_target_opts} -DDEFINE_NDEBUG_INFILE2 -I %S/Inputs %S/assert_in_multiple_tus.cpp %S/Inputs/kernels_in_file2.cpp -o %t.out
 // Shouldn't fail on ACC as fallback assert isn't enqueued there
 // RUN: %{run} %t.out &> %t.txt ; FileCheck %s --input-file %t.txt %if fpga %{ --check-prefix=CHECK-ACC %}
 //
diff --git a/sycl/test-e2e/Assert/assert_in_one_kernel_ndebug.cpp b/sycl/test-e2e/Assert/assert_in_one_kernel_ndebug.cpp
@@ -1,4 +1,4 @@
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} -DNDEBUG  %S/assert_in_one_kernel.cpp -o %t.out
+// RUN: %clangxx -fsycl %{sycl_target_opts} -DNDEBUG  %S/assert_in_one_kernel.cpp -o %t.out
 // RUN: %{run} %t.out | FileCheck %s
 //
 // CHECK-NOT: from assert statement
diff --git a/sycl/test-e2e/Assert/assert_in_simultaneously_multiple_tus_one_ndebug.cpp b/sycl/test-e2e/Assert/assert_in_simultaneously_multiple_tus_one_ndebug.cpp
@@ -5,7 +5,7 @@
 // XFAIL: (opencl && gpu)
 // XFAIL-TRACKER: https://github.com/intel/llvm/issues/11364
 //
-// RUN: %clangxx -DSYCL_FALLBACK_ASSERT=1 -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: %clangxx -DSYCL_FALLBACK_ASSERT=1 -fsycl %{sycl_target_opts} -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: %if cpu %{ %{run} %t.out &> %t.cpu.txt ; FileCheck %s --input-file %t.cpu.txt %}
 //
 // Since this is a multi-threaded application enable memory tracking and
diff --git a/sycl/test-e2e/BFloat16/bfloat16_builtins.cpp b/sycl/test-e2e/BFloat16/bfloat16_builtins.cpp
@@ -5,11 +5,11 @@
 // + below sm_80 always uses generic impls
 
 // DEFINE: %{mathflags} = %if cl_options %{/clang:-fno-fast-math%} %else %{-fno-fast-math%}
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %if any-device-is-cuda %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_80 %} %s -o %t.out %{mathflags}
+// RUN: %clangxx -fsycl %{sycl_target_opts} %if any-device-is-cuda %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_80 %} %s -o %t.out %{mathflags}
 // RUN: %{run} %t.out
 
 // Test "new" (ABI breaking) for all platforms ( sm_80/native if CUDA )
-// RUN:  %if preview-breaking-changes-supported %{  %clangxx -fsycl -fpreview-breaking-changes -fsycl-targets=%{sycl_triple} %if any-device-is-cuda %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_80 %} %s -o %t2.out %{mathflags} %}
+// RUN:  %if preview-breaking-changes-supported %{  %clangxx -fsycl -fpreview-breaking-changes %{sycl_target_opts} %if any-device-is-cuda %{ -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_80 %} %s -o %t2.out %{mathflags} %}
 // RUN:  %if preview-breaking-changes-supported %{  %{run} %t2.out  %}
 
 #include "bfloat16_builtins.hpp"
diff --git a/sycl/test-e2e/BFloat16/bfloat16_builtins_cuda_generic.cpp b/sycl/test-e2e/BFloat16/bfloat16_builtins_cuda_generic.cpp
@@ -7,7 +7,7 @@
 // DEFINE: %{mathflags} = %if cl_options %{/clang:-fno-fast-math%} %else %{-fno-fast-math%}
 
 // If CUDA, test "new" again for sm_75/generic
-// RUN:  %if any-device-is-cuda %{ %if preview-breaking-changes-supported %{  %clangxx -fsycl -fpreview-breaking-changes -fsycl-targets=%{sycl_triple}  -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75  %s -o %t3.out %{mathflags} %} %}
+// RUN:  %if any-device-is-cuda %{ %if preview-breaking-changes-supported %{  %clangxx -fsycl -fpreview-breaking-changes %{sycl_target_opts}  -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_75  %s -o %t3.out %{mathflags} %} %}
 // RUN:  %if any-device-is-cuda %{ %if preview-breaking-changes-supported %{  %{run} %t3.out  %} %}
 
 #include "bfloat16_builtins.hpp"
diff --git a/sycl/test-e2e/Basic/multisource.cpp b/sycl/test-e2e/Basic/multisource.cpp
@@ -9,14 +9,14 @@
 // Separate kernel sources and host code sources
 // RUN: %{build} -c -o %t.kernel.o -DINIT_KERNEL -DCALC_KERNEL
 // RUN: %{build} -c -o %t.main.o -DMAIN_APP
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t.kernel.o %t.main.o -Wno-unused-command-line-argument -o %t1.fat
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t.kernel.o %t.main.o -Wno-unused-command-line-argument -o %t1.fat
 // RUN: %{run} %t1.fat
 
 // Multiple sources with kernel code
 // RUN: %{build} -c -o %t.init.o -DINIT_KERNEL
 // RUN: %{build} -c -o %t.calc.o -DCALC_KERNEL
 // RUN: %{build} -c -o %t.main.o -DMAIN_APP
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t.init.o %t.calc.o %t.main.o -Wno-unused-command-line-argument -o %t2.fat
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t.init.o %t.calc.o %t.main.o -Wno-unused-command-line-argument -o %t2.fat
 // RUN: %{run} %t2.fat
 
 #include <sycl/detail/core.hpp>
diff --git a/sycl/test-e2e/Basic/multisource_spv_obj.cpp b/sycl/test-e2e/Basic/multisource_spv_obj.cpp
@@ -11,21 +11,21 @@
 // Separate kernel sources and host code sources
 // RUN: %{build} -fsycl-device-obj=spirv -c -o %t.kernel.o -DINIT_KERNEL -DCALC_KERNEL
 // RUN: %{build} -fsycl-device-obj=spirv -c -o %t.main.o -DMAIN_APP
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t.kernel.o %t.main.o -Wno-unused-command-line-argument -o %t1.fat
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t.kernel.o %t.main.o -Wno-unused-command-line-argument -o %t1.fat
 // RUN: %{run} %t1.fat
 
 // Multiple sources with kernel code
 // RUN: %{build} -fsycl-device-obj=spirv -c -o %t.init.o -DINIT_KERNEL
 // RUN: %{build} -fsycl-device-obj=spirv -c -o %t.calc.o -DCALC_KERNEL
 // RUN: %{build} -fsycl-device-obj=spirv -c -o %t.main.o -DMAIN_APP
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t.init.o %t.calc.o %t.main.o -Wno-unused-command-line-argument -o %t2.fat
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t.init.o %t.calc.o %t.main.o -Wno-unused-command-line-argument -o %t2.fat
 // RUN: %{run} %t2.fat
 
 // Multiple sources with kernel code, mixed SPIR-V and LLVM-IR objects
 // RUN: %{build} -fsycl-device-obj=spirv -c -o %t.init.o -DINIT_KERNEL
 // RUN: %{build} -fsycl-device-obj=llvmir -c -o %t.calc.o -DCALC_KERNEL
 // RUN: %{build} -c -o %t.main.o -DMAIN_APP
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t.init.o %t.calc.o %t.main.o -Wno-unused-command-line-argument -o %t3.fat
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t.init.o %t.calc.o %t.main.o -Wno-unused-command-line-argument -o %t3.fat
 // RUN: %{run} %t3.fat
 
 #include <sycl/detail/core.hpp>
diff --git a/sycl/test-e2e/DeviceImageDependencies/dynamic.cpp b/sycl/test-e2e/DeviceImageDependencies/dynamic.cpp
@@ -10,7 +10,7 @@
 // RUN: %clangxx %{dynamic_lib_options} %S/Inputs/b.cpp %if windows %{%T/libdevice_c.lib%} -o %T/libdevice_b.%{dynamic_lib_suffix}
 // RUN: %clangxx %{dynamic_lib_options} %S/Inputs/a.cpp %if windows %{%T/libdevice_b.lib%} -o %T/libdevice_a.%{dynamic_lib_suffix}
 
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} -fsycl-allow-device-image-dependencies -fsycl-device-code-split=per_kernel %S/Inputs/basic.cpp -o %t.out            \
+// RUN: %clangxx -fsycl %{sycl_target_opts} -fsycl-allow-device-image-dependencies -fsycl-device-code-split=per_kernel %S/Inputs/basic.cpp -o %t.out            \
 // RUN: %if windows                                                                       \
 // RUN:   %{%T/libdevice_a.lib%}                                                          \
 // RUN: %else                                                                             \
diff --git a/sycl/test-e2e/DeviceImageDependencies/objects.cpp b/sycl/test-e2e/DeviceImageDependencies/objects.cpp
@@ -6,5 +6,5 @@
 // RUN: %clangxx -fsycl %S/Inputs/b.cpp -I %S/Inputs -c -o %t_b.o
 // RUN: %clangxx -fsycl %S/Inputs/c.cpp -I %S/Inputs -c -o %t_c.o
 // RUN: %clangxx -fsycl %S/Inputs/d.cpp -I %S/Inputs -c -o %t_d.o
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} -fsycl-device-code-split=per_kernel -fsycl-allow-device-image-dependencies %t_a.o %t_b.o %t_c.o %t_d.o %S/Inputs/basic.cpp -o %t.out
+// RUN: %clangxx -fsycl %{sycl_target_opts} -fsycl-device-code-split=per_kernel -fsycl-allow-device-image-dependencies %t_a.o %t_b.o %t_c.o %t_d.o %S/Inputs/basic.cpp -o %t.out
 // RUN: %{run} %t.out
diff --git a/sycl/test-e2e/NewOffloadDriver/multisource.cpp b/sycl/test-e2e/NewOffloadDriver/multisource.cpp
@@ -11,23 +11,23 @@
 // Test with `--offload-new-driver`
 // RUN: %{build} --offload-new-driver -c -o %t.kernel.o -DINIT_KERNEL -DCALC_KERNEL
 // RUN: %{build} --offload-new-driver -c -o %t.main.o -DMAIN_APP
-// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl -fsycl-targets=%{sycl_triple} --offload-new-driver %t.kernel.o %t.main.o -o %t1.fat
+// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl %{sycl_target_opts} --offload-new-driver %t.kernel.o %t.main.o -o %t1.fat
 // RUN: %{run} %t1.fat
 
 // Multiple sources with kernel code
 // Test with `--offload-new-driver`
 // RUN: %{build} --offload-new-driver -c -o %t.init.o -DINIT_KERNEL
 // RUN: %{build} --offload-new-driver -c -o %t.calc.o -DCALC_KERNEL
 // RUN: %{build} --offload-new-driver -c -o %t.main.o -DMAIN_APP
-// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl -fsycl-targets=%{sycl_triple} --offload-new-driver %t.init.o %t.calc.o %t.main.o -o %t2.fat
+// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl %{sycl_target_opts} --offload-new-driver %t.init.o %t.calc.o %t.main.o -o %t2.fat
 // RUN: %{run} %t2.fat
 
 // Multiple sources with kernel code with old-style objects
 // Test with `--offload-new-driver`
 // RUN: %{build} --no-offload-new-driver -c -o %t.init.o -DINIT_KERNEL
 // RUN: %{build} --no-offload-new-driver -c -o %t.calc.o -DCALC_KERNEL
 // RUN: %{build} --no-offload-new-driver -c -o %t.main.o -DMAIN_APP
-// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl -fsycl-targets=%{sycl_triple} --offload-new-driver %t.init.o %t.calc.o %t.main.o -o %t3.fat
+// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl %{sycl_target_opts} --offload-new-driver %t.init.o %t.calc.o %t.main.o -o %t3.fat
 // RUN: %{run} %t3.fat
 
 // Multiple sources with kernel code with old-style objects in a static archive
@@ -36,7 +36,7 @@
 // RUN: %{build} --no-offload-new-driver -c -o %t.calc.o -DCALC_KERNEL
 // RUN: %{build} --no-offload-new-driver -c -o %t.main.o -DMAIN_APP
 // RUN: llvm-ar r %t.a %t.init.o %t.calc.o
-// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl -fsycl-targets=%{sycl_triple} --offload-new-driver %t.main.o %t.a -o %t4.fat
+// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl %{sycl_target_opts} --offload-new-driver %t.main.o %t.a -o %t4.fat
 // RUN: %{run} %t4.fat
 
 #include <sycl/detail/core.hpp>
diff --git a/sycl/test-e2e/NewOffloadDriver/sycl-external-with-optional-features.cpp b/sycl/test-e2e/NewOffloadDriver/sycl-external-with-optional-features.cpp
@@ -1,7 +1,7 @@
 // Test with `--offload-new-driver`
 // RUN: %{build} -DSOURCE1 --offload-new-driver -c -o %t1.o
 // RUN: %{build} -DSOURCE2 --offload-new-driver -c -o %t2.o
-// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl -fsycl-targets=%{sycl_triple} --offload-new-driver %t1.o %t2.o -o %t.exe
+// RUN: %clangxx -Wno-error=unused-command-line-argument -fsycl %{sycl_target_opts} --offload-new-driver %t1.o %t2.o -o %t.exe
 // RUN: %{run} %t.exe
 // XFAIL: cuda
 // XFAIL-TRACKER: https://github.com/intel/llvm/issues/16413
diff --git a/sycl/test-e2e/OneapiDeviceSelector/illegal_input.cpp b/sycl/test-e2e/OneapiDeviceSelector/illegal_input.cpp
@@ -1,5 +1,5 @@
 
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %S/Inputs/trivial.cpp -o %t.out
+// RUN: %clangxx -fsycl %{sycl_target_opts} %S/Inputs/trivial.cpp -o %t.out
 // RUN: not --crash env ONEAPI_DEVICE_SELECTOR="macaroni:*" %{run-unfiltered-devices} %t.out
 // RUN: not --crash env ONEAPI_DEVICE_SELECTOR=":" %{run-unfiltered-devices} %t.out
 // RUN: not --crash env ONEAPI_DEVICE_SELECTOR="level_zero:." %{run-unfiltered-devices} %t.out
diff --git a/sycl/test-e2e/OptionalKernelFeatures/sycl-external-with-optional-features.cpp b/sycl/test-e2e/OptionalKernelFeatures/sycl-external-with-optional-features.cpp
@@ -1,6 +1,6 @@
 // RUN: %{build} -DSOURCE1 -c -o %t1.o
 // RUN: %{build} -DSOURCE2 -c -o %t2.o
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t1.o %t2.o -Wno-unused-command-line-argument -o %t.exe
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t1.o %t2.o -Wno-unused-command-line-argument -o %t.exe
 // RUN: %{run} %t.exe
 
 #ifdef SOURCE1
diff --git a/sycl/test-e2e/README.md b/sycl/test-e2e/README.md
@@ -67,7 +67,7 @@ is substituted with just `[Optional run_launcher if that is configured]`.
 Another little nuance is `%{sycl_triple}` substitution. It is constructed by
 concatenating triples for all the devices from `sycl_devices` supported by a
 given test. After that there is also a convenient `%{build}` substitution that
-is equivalent to `%clangxx -fsycl -fsycl-targets=%{sycl_triple} %s`.
+is equivalent to `%clangxx -fsycl %{sycl_target_opts} %s`.
 
 ## Prerequisites
 
diff --git a/sycl/test-e2e/Regression/DAE-separate-compile.cpp b/sycl/test-e2e/Regression/DAE-separate-compile.cpp
@@ -5,7 +5,7 @@
 // The test checks that the scenario works correctly.
 //
 // RUN: %{build} -O2 -c -o %t.o
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t.o %O0 -Wno-unused-command-line-argument -o %t.out
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t.o %O0 -Wno-unused-command-line-argument -o %t.out
 // RUN: %{run} %t.out
 
 // Failing on HIP AMD, enable after fixed
diff --git a/sycl/test-e2e/Regression/commandlist/gpu.cpp b/sycl/test-e2e/Regression/commandlist/gpu.cpp
@@ -1,4 +1,4 @@
 // REQUIRES: gpu
 
-// RUN: %clangxx -Wno-error=vla-cxx-extension -fsycl -fsycl-targets=%{sycl_triple} %S/Inputs/FindPrimesSYCL.cpp %S/Inputs/main.cpp -o %t.out %threads_lib
+// RUN: %clangxx -Wno-error=vla-cxx-extension -fsycl %{sycl_target_opts} %S/Inputs/FindPrimesSYCL.cpp %S/Inputs/main.cpp -o %t.out %threads_lib
 // RUN: %{run} %t.out
diff --git a/sycl/test-e2e/Regression/multiple-targets.cpp b/sycl/test-e2e/Regression/multiple-targets.cpp
@@ -4,16 +4,16 @@
 //
 // REQUIRES: cuda || hip || native_cpu
 // REQUIRES: build-and-run-mode
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple},spir64 -o %t1.out %s
+// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple},spir64 %if any-device-is-hip %{ %{hip_arch_opts} %} -o %t1.out %s
 // RUN: %{run} %t1.out
 //
-// RUN: %clangxx -fsycl -fsycl-targets=spir64,%{sycl_triple} -o %t2.out %s
+// RUN: %clangxx -fsycl -fsycl-targets=spir64,%{sycl_triple} %if any-device-is-hip %{ %{hip_arch_opts} %} -o %t2.out %s
 // RUN: %{run} %t2.out
 //
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple},spir64 -fsycl-device-code-split=per_kernel -o %t3.out %s
+// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple},spir64 %if any-device-is-hip %{ %{hip_arch_opts} %} -fsycl-device-code-split=per_kernel -o %t3.out %s
 // RUN: %{run} %t3.out
 //
-// RUN: %clangxx -fsycl -fsycl-targets=spir64,%{sycl_triple} -fsycl-device-code-split=per_kernel -o %t4.out %s
+// RUN: %clangxx -fsycl -fsycl-targets=spir64,%{sycl_triple} %if any-device-is-hip %{ %{hip_arch_opts} %} -fsycl-device-code-split=per_kernel -o %t4.out %s
 // RUN: %{run} %t4.out
 
 #include <sycl/detail/core.hpp>
diff --git a/sycl/test-e2e/SeparateCompile/same-kernel.cpp b/sycl/test-e2e/SeparateCompile/same-kernel.cpp
@@ -12,7 +12,7 @@
 // RUN: %{build} -DB_CPP=1 -c -o %t-same-kernel-b.o
 //
 // >> ---- link the full hetero app
-// RUN: %clangxx %t-same-kernel-a.o %t-same-kernel-b.o -Wno-unused-command-line-argument -o %t-same-kernel.exe -fsycl -fsycl-targets=%{sycl_triple}
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t-same-kernel-a.o %t-same-kernel-b.o -Wno-unused-command-line-argument -o %t-same-kernel.exe
 // RUN: %{run} %t-same-kernel.exe
 
 #include <sycl/detail/core.hpp>
diff --git a/sycl/test-e2e/SeparateCompile/sycl-external-within-staticlib.cpp b/sycl/test-e2e/SeparateCompile/sycl-external-within-staticlib.cpp
@@ -5,15 +5,15 @@
 // RUN: %{build} -O3 -DSOURCE3 -c -o %t3.o
 // RUN: rm -f %t.a
 // RUN: llvm-ar crv %t.a %t1.o %t2.o
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} -O3 %t3.o %t.a -Wno-unused-command-line-argument -o %t1.exe
+// RUN: %clangxx -fsycl %{sycl_target_opts} -O3 %t3.o %t.a -Wno-unused-command-line-argument -o %t1.exe
 // RUN: %{run} %t1.exe
 
 // Check the repacked case as it can behave differently.
 // RUN: echo create %t_repacked.a > %t.txt
 // RUN: echo addlib %t.a >> %t.txt
 // RUN: echo save >> %t.txt
 // RUN: cat %t.txt | llvm-ar -M
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} -O3 %t3.o %t_repacked.a -Wno-unused-command-line-argument -o %t2.exe
+// RUN: %clangxx -fsycl %{sycl_target_opts} -O3 %t3.o %t_repacked.a -Wno-unused-command-line-argument -o %t2.exe
 // RUN: %{run} %t2.exe
 
 #include <iostream>
diff --git a/sycl/test-e2e/SeparateCompile/sycl-external.cpp b/sycl/test-e2e/SeparateCompile/sycl-external.cpp
@@ -2,14 +2,14 @@
 // different object file.
 // RUN: %{build} -DSOURCE1 -c -o %t1.o
 // RUN: %{build} -DSOURCE2 -c -o %t2.o
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t1.o %t2.o -Wno-unused-command-line-argument -o %t1.exe
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t1.o %t2.o -Wno-unused-command-line-argument -o %t1.exe
 // RUN: %{run} %t1.exe
 //
 // Test2 - check that kernel can call a SYCL_EXTERNAL function defined in a
 // static library.
 // RUN: rm -f %t.a
 // RUN: llvm-ar crv %t.a %t1.o
-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} %t2.o %t.a -Wno-unused-command-line-argument -o %t2.exe
+// RUN: %clangxx -fsycl %{sycl_target_opts} %t2.o %t.a -Wno-unused-command-line-argument -o %t2.exe
 // RUN: %{run} %t2.exe
 
 #include <iostream>
diff --git a/sycl/test-e2e/format.py b/sycl/test-e2e/format.py
@@ -13,7 +13,7 @@
 import re
 
 
-def get_triple(test, backend):
+def get_triple(backend):
     if backend == "cuda":
         return "nvptx64-nvidia-cuda"
     if backend == "hip":
@@ -168,17 +168,27 @@ def execute(self, test, litConfig):
 
             for sycl_device in devices_for_test:
                 (backend, _) = sycl_device.split(":")
-                triples.add(get_triple(test, backend))
+                triples.add(get_triple(backend))
 
         substitutions = lit.TestRunner.getDefaultSubstitutions(test, tmpDir, tmpBase)
+
         substitutions.append(("%{sycl_triple}", format(",".join(triples))))
-        # -fsycl-targets is needed for CUDA/HIP, so just use it be default so
-        # -that new tests by default would runnable there (unless they have
-        # -other restrictions).
+
+        sycl_target_opts = "-fsycl-targets=%{sycl_triple}"
+        if get_triple("hip") in triples:
+            hip_arch_opts = (
+                " -Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch={}".format(
+                    test.config.amd_arch
+                )
+            )
+            sycl_target_opts += hip_arch_opts
+            substitutions.append(("%{hip_arch_opts}", hip_arch_opts))
+        substitutions.append(("%{sycl_target_opts}", sycl_target_opts))
+
         substitutions.append(
             (
                 "%{build}",
-                "%clangxx -fsycl -fsycl-targets=%{sycl_triple} %verbose_print %s",
+                "%clangxx -fsycl %{sycl_target_opts} %verbose_print %s",
             )
         )
         if platform.system() == "Windows":
diff --git a/sycl/test-e2e/lit.cfg.py b/sycl/test-e2e/lit.cfg.py
@@ -685,8 +685,6 @@ def open_check_file(file_name):
 # discovered already.
 config.sycl_dev_features = {}
 
-# Architecture flag for compiling for AMD HIP devices. Empty otherwise.
-arch_flag = ""
 # Version of the driver for a given device. Empty for non-Intel devices.
 config.intel_driver_ver = {}
 for sycl_device in config.sycl_devices:
@@ -838,9 +836,6 @@ def open_check_file(file_name):
                 )
             config.amd_arch = arch.replace(amd_arch_prefix, "")
         llvm_config.with_system_environment("ROCM_PATH")
-        arch_flag = (
-            "-Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=" + config.amd_arch
-        )
         config.substitutions.append(
             ("%rocm_path", os.environ.get("ROCM_PATH", "/opt/rocm"))
         )
@@ -856,10 +851,7 @@ def open_check_file(file_name):
     config.substitutions.append(("%clang", " true "))
 else:
     config.substitutions.append(
-        (
-            "%clangxx",
-            " " + config.dpcpp_compiler + " " + config.cxx_flags + " " + arch_flag,
-        )
+        ("%clangxx", " " + config.dpcpp_compiler + " " + config.cxx_flags)
     )
     config.substitutions.append(
         ("%clang", " " + config.dpcpp_compiler + " " + config.c_flags)
diff --git a/sycl/test-e2e/syclcompat/kernel/kernel_lin.cpp b/sycl/test-e2e/syclcompat/kernel/kernel_lin.cpp
@@ -2,6 +2,6 @@
 // TODO: Supported for ROCM 5. Further development required to support AMDGPU.
 // UNSUPPORTED: hip
 
-// RUN: %clangxx -fPIC -shared -fsycl -fsycl-targets=%{sycl_triple} %S/Inputs/kernel_module.cpp -o %t.so
-// RUN: %clangxx -DTEST_SHARED_LIB='"%t.so"' -ldl -fsycl -fsycl-targets=%{sycl_triple} %S/Inputs/kernel_function.cpp -o %t.out
+// RUN: %clangxx -fPIC -shared -fsycl %{sycl_target_opts} %S/Inputs/kernel_module.cpp -o %t.so
+// RUN: %clangxx -DTEST_SHARED_LIB='"%t.so"' -ldl -fsycl %{sycl_target_opts} %S/Inputs/kernel_function.cpp -o %t.out
 // RUN: %{run} %t.out
diff --git a/sycl/test-e2e/syclcompat/kernel/kernel_win.cpp b/sycl/test-e2e/syclcompat/kernel/kernel_win.cpp

Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,4 @@`
`1`		`-// RUN: %clangxx -fsycl -fsycl-targets=%{sycl_triple} -DNDEBUG %S/assert_in_kernels.cpp -o %t.out`
	`1`	`+// RUN: %clangxx -fsycl %{sycl_target_opts} -DNDEBUG %S/assert_in_kernels.cpp -o %t.out`
`2`	`2`	`// RUN: %{run} %t.out \| FileCheck %s`
`3`	`3`	`//`
`4`	`4`	`// CHECK-NOT: One shouldn't see this message`