Skip to content

Commit c73f76f

Browse files
authored
[SYCL][E2E] Document undefined symbol tests for CUDA/HIP (#17630)
1 parent 79a185a commit c73f76f

File tree

4 files changed

+21
-8
lines changed

4 files changed

+21
-8
lines changed

sycl/test-e2e/KernelAndProgram/build-log.cpp

+6-1
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,10 @@
1-
// for CUDA and HIP the failure happens at compile time, not during runtime
21
// UNSUPPORTED: target-nvidia || target-amd
2+
// UNSUPPORTED-INTENDED: The test looks for an exception thrown during the
3+
// compilation of the kernel, but for CUDA the failure is not thrown, but comes
4+
// from ptxas that crashes clang. The JIT part is not relevant, because the
5+
// flow is such that the AOT compilation still happens, it’s just that if we
6+
// request JIT, it will do the thing again at the run time.
7+
//
38
// TODO: rewrite this into a unit-test
49

510
// RUN: %{build} -DGPU -o %t_gpu.out

sycl/test-e2e/KernelAndProgram/cache-build-result.cpp

+8-2
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,11 @@
1-
// for CUDA and HIP the failure happens at compile time, not during runtime
2-
// UNSUPPORTED: target-nvidia || target-amd || ze_debug
1+
// UNSUPPORTED: target-nvidia || target-amd
2+
// UNSUPPORTED-INTENDED: The test looks for an exception thrown during the
3+
// compilation of the kernel, but for CUDA the failure is not thrown, but comes
4+
// from ptxas that crashes clang. The JIT part is not relevant, because the
5+
// flow is such that the AOT compilation still happens, it’s just that if we
6+
// request JIT, it will do the thing again at the run time.
7+
//
8+
// UNSUPPORTED: ze_debug
39

410
// RUN: %{build} -DSYCL_DISABLE_FALLBACK_ASSERT=1 -o %t.out
511
// RUN: %{build} -DSYCL_DISABLE_FALLBACK_ASSERT=1 -DGPU -o %t_gpu.out

sycl/test-e2e/KernelAndProgram/undefined-symbol.cpp

+6-2
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,10 @@
1-
// for CUDA and HIP the failure happens at compile time, not during runtime
21
// UNSUPPORTED: target-nvidia || target-amd
3-
2+
// UNSUPPORTED-INTENDED: The test looks for an exception thrown during the
3+
// compilation of the kernel, but for CUDA the failure is not thrown, but comes
4+
// from ptxas that crashes clang. The JIT part is not relevant, because the
5+
// flow is such that the AOT compilation still happens, it’s just that if we
6+
// request JIT, it will do the thing again at the run time.
7+
//
48
// RUN: %{build} -o %t.out
59
// RUN: %{run} %t.out
610
//

sycl/test/e2e_test_requirements/no-unsupported-without-info.cpp

+1-3
Original file line numberDiff line numberDiff line change
@@ -54,7 +54,7 @@
5454
// tests to match the required format and in that case you should just update
5555
// (i.e. reduce) the number and the list below.
5656
//
57-
// NUMBER-OF-UNSUPPORTED-WITHOUT-INFO: 282
57+
// NUMBER-OF-UNSUPPORTED-WITHOUT-INFO: 280
5858
//
5959
// List of improperly UNSUPPORTED tests.
6060
// Remove the CHECK once the test has been properly UNSUPPORTED.
@@ -208,7 +208,6 @@
208208
// CHECK-NEXT: InvokeSimd/Feature/popcnt.cpp
209209
// CHECK-NEXT: InvokeSimd/Regression/ImplicitSubgroup/call_vadd_1d_spill.cpp
210210
// CHECK-NEXT: InvokeSimd/Regression/call_vadd_1d_spill.cpp
211-
// CHECK-NEXT: KernelAndProgram/build-log.cpp
212211
// CHECK-NEXT: KernelAndProgram/cache-build-result.cpp
213212
// CHECK-NEXT: KernelAndProgram/free_function_apis.cpp
214213
// CHECK-NEXT: KernelAndProgram/free_function_kernels.cpp
@@ -218,7 +217,6 @@
218217
// CHECK-NEXT: KernelAndProgram/multiple-kernel-linking.cpp
219218
// CHECK-NEXT: KernelAndProgram/spec_constants_after_link.cpp
220219
// CHECK-NEXT: KernelAndProgram/spec_constants_after_link.cpp
221-
// CHECK-NEXT: KernelAndProgram/undefined-symbol.cpp
222220
// CHECK-NEXT: Matrix/SG32/element_wise_abc.cpp
223221
// CHECK-NEXT: Matrix/SG32/element_wise_all_ops.cpp
224222
// CHECK-NEXT: Matrix/SG32/element_wise_all_ops_half.cpp

0 commit comments

Comments
 (0)