[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 <bjoern@codeplay.com>

Author: bjoernknafla

sycl/plugins/cuda/CMakeLists.txt

@@ -4,7 +4,7 @@ message(STATUS "Including the PI API CUDA backend.")
  # we only require the CUDA driver API to be used
  # CUDA_CUDA_LIBRARY variable defines the path to libcuda.so, the CUDA Driver API library.
 
-find_package(CUDA 10.0 REQUIRED)
+find_package(CUDA 10.1 REQUIRED)
 
 add_library(cudadrv SHARED IMPORTED)
 

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/basic_tests/buffer/subbuffer.cpp

@@ -1,11 +1,12 @@
-// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
+// XFAIL: cuda
+// TODO: Fix CUDA implementation.
+//
+// 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/handler/handler_copy_with_offset.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
@@ -48,7 +48,7 @@ int main() {
 
     vector_class<char> Expected{'x', 'x', '0', '1', '2', '3', 'x', 'x'};
     if (DataRaw != Expected)
-      throw std::runtime_error("Check of hadler.copy(ptr, acc) was failed");
+      throw std::runtime_error("Check of handler.copy(ptr, acc) was failed");
   }
 
   {
@@ -71,7 +71,7 @@ int main() {
     }
     vector_class<char> Expected{'2', '3', '4', '5', 'x', 'x', 'x', 'x'};
     if (DataRaw != Expected)
-      throw std::runtime_error("Check of hadler.copy(acc, ptr) was failed");
+      throw std::runtime_error("Check of handler.copy(acc, ptr) was failed");
   }
   return 0;
 }

sycl/test/basic_tests/handler/interop_task.cpp

@@ -1,7 +1,7 @@
+// REQUIRES: opencl
 // RUN: %clangxx -fsycl %s -o %t.out -L %opencl_libs_dir -lOpenCL
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
-// REQUIRES: opencl
 
 //==------- interop_task.cpp -----------------------------------------------==//
 //

sycl/test/basic_tests/kernel_info.cpp

@@ -12,28 +12,18 @@
 //===----------------------------------------------------------------------===//
 
 #include <CL/sycl.hpp>
+#include <cassert>
 
 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;
 
   buffer<int, 1> buf(range<1>(1));
   program prg(q.get_context());
 
   prg.build_with_kernel_type<class SingleTask>();
-  CHECK(prg.has_kernel<class SingleTask>());
+  assert(prg.has_kernel<class SingleTask>());
   kernel krn = prg.get_kernel<class SingleTask>();
 
   q.submit([&](handler &cgh) {
@@ -42,26 +32,26 @@ int main() {
   });
 
   const string_class krnName = krn.get_info<info::kernel::function_name>();
-  CHECK(!krnName.empty());
+  assert(!krnName.empty());
   const cl_uint krnArgCount = krn.get_info<info::kernel::num_args>();
-  CHECK(krnArgCount > 0);
+  assert(krnArgCount > 0);
   const context krnCtx = krn.get_info<info::kernel::context>();
-  CHECK(krnCtx == q.get_context());
+  assert(krnCtx == q.get_context());
   const program krnPrg = krn.get_info<info::kernel::program>();
-  CHECK(krnPrg == prg);
+  assert(krnPrg == prg);
   const cl_uint krnRefCount = krn.get_info<info::kernel::reference_count>();
-  CHECK(krnRefCount > 0);
+  assert(krnRefCount > 0);
   const string_class krnAttr = krn.get_info<info::kernel::attributes>();
-  CHECK(krnAttr.empty());
+  assert(krnAttr.empty());
 
   device dev = q.get_device();
   const size_t wgSize =
       krn.get_work_group_info<info::kernel_work_group::work_group_size>(dev);
-  CHECK(wgSize > 0);
+  assert(wgSize > 0);
   const size_t prefWGSizeMult = krn.get_work_group_info<
       info::kernel_work_group::preferred_work_group_size_multiple>(dev);
-  CHECK(prefWGSizeMult > 0);
+  assert(prefWGSizeMult > 0);
   const cl_ulong prvMemSize =
       krn.get_work_group_info<info::kernel_work_group::private_mem_size>(dev);
-  CHECK(prvMemSize == 0);
+  assert(prvMemSize == 0);
 }

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/basic_tests/sampler/sampler.cpp

@@ -6,9 +6,6 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
-// TODO: Image support in CUDA backend
-// XFAIL: cuda
-
 //==--------------- sampler.cpp - SYCL sampler basic test ------------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

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

@@ -5,7 +5,7 @@
 // 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 @@
-// RUN: %clangxx -Xclang -fsycl-allow-func-ptr -std=c++14 -fsycl %s -o %t.out -L %opencl_libs_dir -lOpenCL
+// 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
 // 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 @@
-// RUN: %clangxx -Xclang -fsycl-allow-func-ptr -std=c++14 -fsycl %s -o %t.out -L %opencl_libs_dir -lOpenCL
+// 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
 // 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/group-algorithm/all_of.cpp

@@ -1,9 +1,11 @@
+// UNSUPPORTED: cuda
+// OpenCL C 2.x alike work-group functions not yet supported by 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
-// UNSUPPORTED: cuda
 
 #include <CL/sycl.hpp>
 #include <algorithm>

sycl/test/group-algorithm/any_of.cpp

@@ -1,9 +1,11 @@
+// UNSUPPORTED: cuda
+// OpenCL C 2.x alike work-group functions not yet supported by 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
-// UNSUPPORTED: cuda
 
 #include <CL/sycl.hpp>
 #include <algorithm>

sycl/test/group-algorithm/broadcast.cpp

@@ -1,9 +1,11 @@
+// UNSUPPORTED: cuda
+// OpenCL C 2.x alike work-group functions not yet supported by 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
-// UNSUPPORTED: cuda
 
 #include <CL/sycl.hpp>
 #include <algorithm>

sycl/test/group-algorithm/exclusive_scan.cpp

@@ -1,9 +1,11 @@
+// UNSUPPORTED: cuda
+// OpenCL C 2.x alike work-group functions not yet supported by 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
-// UNSUPPORTED: cuda
 
 #include <CL/sycl.hpp>
 #include <algorithm>

sycl/test/group-algorithm/inclusive_scan.cpp

@@ -1,9 +1,11 @@
+// UNSUPPORTED: cuda
+// OpenCL C 2.x alike work-group functions not yet supported by 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
-// UNSUPPORTED: cuda
 
 #include <CL/sycl.hpp>
 #include <algorithm>

sycl/test/group-algorithm/leader.cpp

@@ -1,3 +1,6 @@
+// UNSUPPORTED: cuda
+// OpenCL C 2.x alike work-group functions not yet supported by 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

sycl/test/group-algorithm/none_of.cpp

@@ -1,9 +1,11 @@
+// UNSUPPORTED: cuda
+// OpenCL C 2.x alike work-group functions not yet supported by 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
-// UNSUPPORTED: cuda
 
 #include <CL/sycl.hpp>
 #include <algorithm>

sycl/test/group-algorithm/reduce.cpp

@@ -1,9 +1,11 @@
+// UNSUPPORTED: cuda
+// OpenCL C 2.x alike work-group functions not yet supported by 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
-// UNSUPPORTED: cuda
 
 #include <CL/sycl.hpp>
 #include <algorithm>

sycl/test/hier_par/hier_par_wgscope.cpp

@@ -18,9 +18,6 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
-// 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 -I %sycl_include -Xclang -verify-ignore-unexpected=note,warning -Wno-sycl-strict
 // RUN: %clangxx -include %t.h -g %s -o %t.out -lsycl -I %sycl_include -Xclang -verify-ignore-unexpected=note,warning

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

@@ -4,9 +4,10 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// UNSUPPORTED: cuda
+
 //==---------------- opencl-interop.cpp - SYCL linear id test --------------==//
 //
+//
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

sycl/test/lit.cfg.py

@@ -72,8 +72,8 @@
 
 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( ('%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 +3,15 @@
 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.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_dmemll.cpp

@@ -3,7 +3,6 @@
 // RUN: %clangxx -fsycl %s -o %t1.out -L %opencl_libs_dir -lOpenCL
 // RUN: %CPU_RUN_PLACEHOLDER %t1.out
 // RUN: %GPU_RUN_PLACEHOLDER %t1.out
-// XFAIL: cuda
 //==----------- ordered_dmemll.cpp - Device Memory Linked List test --------==//
 // It uses an ordered queue where explicit waiting is not necessary between
 // kernels

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 -----------------==//