[WIP][SYCL][CUDA] LIT XFAIL/UNSUPPORTED

Remove XFAIL from passing tests.

Change tests that should pass but fail as XFAIL, e.g., mark
unexpectedly failing sub-buffer test to fix later for SYCL CUDA.

Add `REQUIRES: opencl` for LIT tests requiring OpenCL.
LIT tests that `REQUIRES: opencl` do not need an `UNSUPPORTED: cuda`.

Mark tests of features not supported by SYCL CUDA with
`UNSUPPORTED: cuda`.

Tests that do not use the CUDA target triple do not require any
CUDA related LIT marker.

Tests that do not fail with CUDA do not require any CUDA specific
LIT marker.

Fix LIT test using `-Werror` to fail when the compiler driver warns
about an unknown CUDA version as CUDA has otherwise nothing to do with
the test.

Adapt test to moved `pi_cuda.hpp` header.

Some cleanup inside LIT configuration scripts.

Signed-off-by: Bjoern Knafla <[email protected]>

Author: bjoernknafla

sycl/test/aot/gpu.cpp

@@ -1,9 +1,11 @@
 // REQUIRES: ocloc, gpu
+// UNSUPPORTED: cuda
+// CUDA is not compatible with SPIR.
 
 // RUN: %clangxx -fsycl -fsycl-targets=spir64_gen-unknown-unknown-sycldevice -Xsycl-target-backend=spir64_gen-unknown-unknown-sycldevice "-device skl" %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
-// XFAIL: cuda
+
 //==----- gpu.cpp - AOT compilation for gen devices using GEN compiler  ------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/aot/multiple-devices.cpp

@@ -7,6 +7,8 @@
 //===------------------------------------------------------------------------===//
 
 // REQUIRES: opencl-aot, ocloc, aoc, cpu, gpu, accelerator
+// UNSUPPORTED: cuda
+// CUDA is not compatible with SPIR.
 
 // 1-command compilation case
 // Targeting CPU, GPU, FPGA

sycl/test/backend/cuda/primary_context.cpp

@@ -1,5 +1,5 @@
 // REQUIRES: cuda
-// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -I%opencl_include_dir -I%cuda_toolkit_include -o %t.out -lcuda -lsycl
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -I%opencl_include_dir -I%cuda_toolkit_include -I%sycl_plugins_dir/cuda -o %t.out -lcuda -lsycl
 // RUN: env SYCL_DEVICE_TYPE=GPU %t.out
 // NOTE: OpenCL is required for the runtime, even when using the CUDA BE.
 
@@ -11,24 +11,14 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "../../helpers.hpp"
 #include <CL/sycl.hpp>
-#include <CL/sycl/detail/pi_cuda.hpp>
 #include <cuda.h>
 #include <iostream>
+#include <pi_cuda.hpp>
 
 using namespace cl::sycl;
 
-void check(bool condition, const char *conditionString, const char *filename,
-           const long line) noexcept {
-  if (!condition) {
-    std::cerr << "CHECK failed in " << filename << "#" << line << " "
-              << conditionString << "\n";
-    std::abort();
-  }
-}
-
-#define CHECK(CONDITION) check(CONDITION, #CONDITION, __FILE__, __LINE__)
-
 bool isCudaDevice(const device &dev) {
   const platform platform = dev.get_info<info::device::platform>();
   const std::string platformVersion =

sycl/test/basic_tests/boolean.cpp

@@ -3,8 +3,6 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// XFAIL: cuda
-// TODO: investigate incorrect results on cuda backend
 #include <CL/sycl.hpp>
 
 #include <cassert>

sycl/test/basic_tests/buffer/buffer_full_copy.cpp

@@ -1,3 +1,6 @@
+// XFAIL: cuda
+// TODO: Fix fail for CUDA.
+//
 // RUN: %clangxx %s -o %t1.out -lsycl
 // RUN: env SYCL_DEVICE_TYPE=HOST %t1.out
 // RUN: %clangxx -fsycl  -fsycl-targets=%sycl_triple  %s -o %t2.out
@@ -6,9 +9,6 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t2.out
 // RUN: %ACC_RUN_PLACEHOLDER %t2.out
 
-// TODO: cuda_piEnqueueMemBufferCopy not implemented
-// XFAIL: cuda
-
 //==------------- buffer_full_copy.cpp - SYCL buffer basic test ------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/basic_tests/buffer/subbuffer.cpp

@@ -1,11 +1,11 @@
-// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
+// XFAIL: cuda
+// TODO: Fix fail for CUDA.
+//
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// XFAIL: cuda
-// TODO: cuda fail due to unimplemented param_name 4121 in cuda_piDeviceGetInfo
-
 //==---------- subbuffer.cpp --- sub-buffer basic test ---------------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/basic_tests/buffer/subbuffer_interop.cpp

@@ -1,10 +1,10 @@
+// REQUIRES: opencl
+//
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
-// REQUIRES: opencl
-
 //==------------ subbuffer_interop.cpp - SYCL buffer basic test ------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/basic_tests/event.cpp

@@ -1,3 +1,5 @@
+// REQUIRES: opencl
+//
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 //==--------------- event.cpp - SYCL event test ----------------------------==//

sycl/test/basic_tests/image_api.cpp

@@ -1,3 +1,5 @@
+// REQUIRES: opencl
+//
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -I %sycl_source_dir %s -o %t1.out
 // RUN: %clangxx -I %sycl_source_dir %s -o %t3.out -lsycl
 // RUN: env SYCL_DEVICE_TYPE=HOST %t1.out

sycl/test/basic_tests/kernel_info.cpp

@@ -11,21 +11,11 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "../helpers.hpp"
 #include <CL/sycl.hpp>
 
 using namespace cl::sycl;
 
-void check(bool condition, const char *conditionString, const char *filename,
-           const long line) noexcept {
-  if (!condition) {
-    std::cerr << "CHECK failed in " << filename << "#" << line << " "
-              << conditionString << "\n";
-    std::abort();
-  }
-}
-
-#define CHECK(CONDITION) check(CONDITION, #CONDITION, __FILE__, __LINE__)
-
 int main() {
   queue q;
 
@@ -64,4 +54,4 @@ int main() {
   const cl_ulong prvMemSize =
       krn.get_work_group_info<info::kernel_work_group::private_mem_size>(dev);
   CHECK(prvMemSize == 0);
-}
+}

sycl/test/basic_tests/kernel_interop.cpp

@@ -1,10 +1,10 @@
+// REQUIRES: opencl
+//
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
-// REQUIRES: opencl
-
 //==--------------- kernel_interop.cpp - SYCL kernel ocl interop test ------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/basic_tests/parallel_for_range.cpp

@@ -1,8 +1,10 @@
-// RUN: %clangxx -fsycl %s -o %t.out
+// XFAIL: cuda
+// CUDA exposes broken hierarchical parallelism.
+
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple  %s -o %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// XFAIL: cuda
 
 #include <CL/sycl.hpp>
 

sycl/test/fpga_tests/fpga_pipes.cpp

@@ -3,7 +3,7 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// UNSUPPORTED: cuda
+
 //==------------- fpga_pipes.cpp - SYCL FPGA pipes test --------------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/fpga_tests/fpga_queue.cpp

@@ -1,9 +1,11 @@
+// REQUIRES: opencl
+//
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
-// UNSUPPORTED: cuda
+
 //==------------- fpga_queue.cpp - SYCL FPGA queues test -------------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/function-pointers/fp-as-kernel-arg.cpp

@@ -1,11 +1,13 @@
+// UNSUPPORTED: windows
+// UNSUPPORTED: cuda
+// CUDA does not support the function pointer as kernel argument extension.
+
 // RUN: %clangxx -Xclang -fsycl-allow-func-ptr -std=c++14 -fsycl %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // FIXME: This test should use runtime early exit once correct check for
 // corresponding extension is implemented
-// UNSUPPORTED: windows
-// XFAIL: cuda
 
 #include <CL/sycl.hpp>
 

sycl/test/function-pointers/pass-fp-through-buffer.cpp

@@ -1,11 +1,13 @@
+// UNSUPPORTED: windows
+// UNSUPPORTED: cuda
+// CUDA does not support the function pointer as kernel argument extension.
+
 // RUN: %clangxx -Xclang -fsycl-allow-func-ptr -std=c++14 -fsycl %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // FIXME: This test should use runtime early exit once correct check for
 // corresponding extension is implemented
-// UNSUPPORTED: windows
-// XFAIL: cuda
 
 #include <CL/sycl.hpp>
 

sycl/test/hier_par/hier_par_wgscope.cpp

@@ -18,8 +18,8 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// XFAIL: cuda
 // TODO: ptxas fatal   : Unresolved extern function '__spirv_ControlBarrier'
-// UNSUPPORTED: cuda
 
 // This test checks correctness of hierarchical kernel execution when there is
 // code and data in the work group scope.

sycl/test/kernel_from_file/hw.cpp

@@ -1,3 +1,6 @@
+// UNSUPPORTED: cuda
+// CUDA does not support SPIR-V.
+
 //-fsycl-targets=%sycl_triple
 // RUN: %clangxx -fsycl-device-only -fno-sycl-use-bitcode -Xclang -fsycl-int-header=%t.h -c %s -o %t.spv
 // RUN: %clangxx -include %t.h -g %s -o %t.out -lsycl
@@ -6,9 +9,6 @@
 
 // TODO: InvalidTargetTriple: Expects spir-unknown-unknown or spir64-unknown-unknown. Actual target triple is x86_64-unknown-linux-gnu
 
-// XFAIL: cuda
-// Currently unsupported on cuda as this test specifically tests a SPV path.
-
 #include <CL/sycl.hpp>
 #include <iostream>
 

sycl/test/linear_id/linear-sub_group.cpp

@@ -1,4 +1,4 @@
-// RUN: %clangxx -fsycl %s -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out

sycl/test/linear_id/opencl-interop.cpp

@@ -3,7 +3,7 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 // REQUIRES: opencl
-// UNSUPPORTED: cuda
+
 //==---------------- opencl-interop.cpp - SYCL linear id test --------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/lit.cfg.py

@@ -69,8 +69,9 @@
 
 config.substitutions.append( ('%sycl_libs_dir',  config.sycl_libs_dir ) )
 config.substitutions.append( ('%sycl_include',  config.sycl_include ) )
-config.substitutions.append( ('%opencl_libs_dir',  config.opencl_libs_dir) )
 config.substitutions.append( ('%sycl_source_dir', config.sycl_source_dir) )
+config.substitutions.append( ('%sycl_plugins_dir', config.sycl_plugins_dir) )
+config.substitutions.append( ('%opencl_libs_dir',  config.opencl_libs_dir) )
 config.substitutions.append( ('%opencl_include_dir',  config.opencl_include_dir) )
 config.substitutions.append( ('%cuda_toolkit_include',  config.cuda_toolkit_include) )
 

sycl/test/lit.site.cfg.py.in

@@ -3,15 +3,16 @@
 import sys
 
 config.llvm_tools_dir = "@LLVM_TOOLS_DIR@"
-config.sycl_tools_dir = lit_config.params.get('SYCL_TOOLS_DIR', "@LLVM_TOOLS_DIR@")
 config.lit_tools_dir = "@LLVM_LIT_TOOLS_DIR@"
+config.sycl_tools_dir = lit_config.params.get('SYCL_TOOLS_DIR', "@LLVM_TOOLS_DIR@")
 config.sycl_include = lit_config.params.get('SYCL_INCLUDE', "@SYCL_INCLUDE@")
 config.sycl_obj_root = "@SYCL_BINARY_DIR@"
 config.sycl_source_dir = "@SYCL_SOURCE_DIR@/source"
-config.opencl_libs_dir = os.path.dirname("@OpenCL_LIBRARIES@")
 config.sycl_libs_dir = lit_config.params.get('SYCL_LIBS_DIR', "@LLVM_LIBS_DIR@")
+config.sycl_plugins_dir = "@sycl_plugin_dir@"
 config.target_triple = "@TARGET_TRIPLE@"
 config.host_triple = "@LLVM_HOST_TRIPLE@"
+config.opencl_libs_dir = os.path.dirname("@OpenCL_LIBRARIES@")
 config.opencl_include_dir = "@OpenCL_INCLUDE_DIR@"
 config.cuda_toolkit_include = "@CUDA_TOOLKIT_INCLUDE@"
 

sycl/test/ordered_queue/ordered_queue.cpp

@@ -1,3 +1,5 @@
+// REQUIRES: opencl
+//
 // RUN: %clangxx -fsycl %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 //==---------- ordered_queue.cpp - SYCL ordered queue test -----------------==//

sycl/test/ordered_queue/prop.cpp

@@ -1,4 +1,6 @@
-// RUN: %clangxx -fsycl %s -o %t1.out -L %opencl_libs_dir -lOpenCL
+// REQUIRES: opencl
+//
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t1.out -L %opencl_libs_dir -lOpenCL
 // RUN: %CPU_RUN_PLACEHOLDER %t1.out
 // RUN: %GPU_RUN_PLACEHOLDER %t1.out
 

sycl/test/program_manager/env_vars.cpp

@@ -1,4 +1,7 @@
-// RUN: %clangxx -O0 -fsycl %s -o %t.out -lsycl
+// REQUIRES: opencl
+// Env vars are used to pass OpenCL-specific flags to PI compiling/linking.
+//
+// RUN: %clangxx -O0 -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -lsycl
 //
 // Deprecated SYCL_PROGRAM_BUILD_OPTIONS should work as an alias to
 // SYCL_PROGRAM_COMPILE_OPTIONS:

sycl/test/regression/image_access.cpp

@@ -5,11 +5,8 @@
 // TODO: For now PI checks are skipped for ACC device. To decide if it's good.
 // RUN: env %ACC_RUN_PLACEHOLDER %t.out
 
-// TODO: No CUDA image support
-// XFAIL: cuda
-
-// TODO: No CUDA image support
-// XFAIL: cuda
+// UNSUPPORTED: cuda
+// CUDA cannot support OpenCL spec conform images.
 
 //==-------------- image_access.cpp - SYCL image accessors test  -----------==//
 //

sycl/test/sub_group/attributes.cpp

@@ -1,4 +1,4 @@
-// RUN: %clangxx -fsycl %s -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUNx: %GPU_RUN_PLACEHOLDER %t.out

sycl/test/sub_group/barrier.cpp

@@ -3,7 +3,7 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// UNSUPPORTED: cuda
+
 //==---------- barrier.cpp - SYCL sub_group barrier test -------*- C++ -*---==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/sub_group/broadcast.cpp

@@ -1,10 +1,14 @@
+// XFAIL: cuda
+// CUDA compilation and runtime do not yet support sub groups.
+
+
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -D SG_GPU %s -o %t_gpu.out
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t_gpu.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// UNSUPPORTED: cuda
+
 //==--------- broadcast.cpp - SYCL sub_group broadcast test ----*- C++ -*---==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/sub_group/common.cpp

@@ -3,7 +3,7 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// UNSUPPORTED: cuda
+
 //==-------------- common.cpp - SYCL sub_group common test -----*- C++ -*---==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

sycl/test/sub_group/common_ocl.cpp

@@ -1,11 +1,13 @@
+// REQUIRES: opencl
+
 // RUN: %clang_cc1 -x cl -cl-std=CL2.0 %S/sg.cl -triple spir64-unknown-unknown -emit-llvm-bc -o %T/kernel_ocl.bc -include opencl-c.h
 // RUN: llvm-spirv %T/kernel_ocl.bc -o %T/kernel_ocl.spv
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: env SYCL_DEVICE_TYPE=HOST %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out %T/kernel_ocl.spv
 // RUN: %GPU_RUN_PLACEHOLDER %t.out %T/kernel_ocl.spv
 // RUN: %ACC_RUN_PLACEHOLDER %t.out %T/kernel_ocl.spv
-// UNSUPPORTED: cuda
+
 //==--- common_ocl.cpp - basic SG methods in SYCL vs OpenCL  ---*- C++ -*---==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.