[OpenMP] Fix or disable NVPTX tests failing currently (#77844)
Summary:
This patch is an attempt to get a clean run of `check-openmp` running on
an NVPTX machine. I simply took the lists of tests that failed on my
`sm_89` machine and disabled them or fixed them. A lot of these tests
are disabled on AMDGPU already, so it makes sense that NVPTX fails. The
others are simply problems with NVPTX optimized debugging which will
need to be fixed. I opened an issue on one of them.
GitOrigin-RevId: ab02372c23d736390587aab141fe69c142373002
diff --git a/libomptarget/test/libc/assert.c b/libomptarget/test/libc/assert.c
index 9a10032..803a820 100644
--- a/libomptarget/test/libc/assert.c
+++ b/libomptarget/test/libc/assert.c
@@ -3,6 +3,8 @@
// REQUIRES: libc
+// NVPTX without LTO uses the implementation in OpenMP currently.
+// UNSUPPORTED: nvptx64-nvidia-cuda
// UNSUPPORTED: powerpc64-ibm-linux-gnu
// UNSUPPORTED: powerpc64-ibm-linux-gnu-LTO
// UNSUPPORTED: aarch64-unknown-linux-gnu
diff --git a/libomptarget/test/mapping/target_derefence_array_pointrs.cpp b/libomptarget/test/mapping/target_derefence_array_pointrs.cpp
index 7d4bc9e..a6dd406 100644
--- a/libomptarget/test/mapping/target_derefence_array_pointrs.cpp
+++ b/libomptarget/test/mapping/target_derefence_array_pointrs.cpp
@@ -2,7 +2,10 @@
// RUN: %libomptarget-run-generic 2>&1 \
// RUN: | %fcheck-generic
+// FIXME: This is currently broken on all GPU targets
// UNSUPPORTED: amdgcn-amd-amdhsa
+// UNSUPPORTED: nvptx64-nvidia-cuda
+// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
#include <stdio.h>
#include <stdlib.h>
diff --git a/libomptarget/test/mapping/target_uses_allocator.c b/libomptarget/test/mapping/target_uses_allocator.c
index 87b9406..eb20e96 100755
--- a/libomptarget/test/mapping/target_uses_allocator.c
+++ b/libomptarget/test/mapping/target_uses_allocator.c
@@ -1,6 +1,9 @@
// RUN: %libomptarget-compile-run-and-check-generic
+// FIXME: https://github.com/llvm/llvm-project/issues/77841
// UNSUPPORTED: amdgcn-amd-amdhsa
+// UNSUPPORTED: nvptx64-nvidia-cuda
+// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
#include <omp.h>
#include <stdio.h>
diff --git a/libomptarget/test/mapping/target_wrong_use_device_addr.c b/libomptarget/test/mapping/target_wrong_use_device_addr.c
index 6c7939e..7a5babd 100644
--- a/libomptarget/test/mapping/target_wrong_use_device_addr.c
+++ b/libomptarget/test/mapping/target_wrong_use_device_addr.c
@@ -2,6 +2,9 @@
// RUN: env LIBOMPTARGET_INFO=64 %libomptarget-run-fail-generic 2>&1 \
// RUN: | %fcheck-generic
+// FIXME: Fails due to optimized debugging in 'ptxas'
+// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
+
#include <stdio.h>
int main() {
diff --git a/libomptarget/test/offloading/bug64959.c b/libomptarget/test/offloading/bug64959.c
index 500911f..eddc553 100644
--- a/libomptarget/test/offloading/bug64959.c
+++ b/libomptarget/test/offloading/bug64959.c
@@ -1,8 +1,11 @@
-// RUN: %libomptarget-compilexx-run-and-check-generic
-// RUN: %libomptarget-compileoptxx-run-and-check-generic
+// RUN: %libomptarget-compile-run-and-check-generic
+// RUN: %libomptarget-compileopt-run-and-check-generic
// TODO: This requires malloc support for the threads states.
+// FIXME: Flaky on all GPU targets.
// UNSUPPORTED: amdgcn-amd-amdhsa
+// UNSUPPORTED: nvptx64-nvidia-cuda
+// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
#include <omp.h>
#include <stdio.h>
diff --git a/libomptarget/test/offloading/info.c b/libomptarget/test/offloading/info.c
index 81300cc..da8e4c4 100644
--- a/libomptarget/test/offloading/info.c
+++ b/libomptarget/test/offloading/info.c
@@ -5,6 +5,9 @@
// RUN: env LIBOMPTARGET_INFO=63 %libomptarget-run-amdgcn-amd-amdhsa 2>&1 | \
// RUN: %fcheck-amdgcn-amd-amdhsa -allow-empty -check-prefixes=INFO,AMDGPU
+// FIXME: Fails due to optimized debugging in 'ptxas'.
+// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
+
#include <omp.h>
#include <stdio.h>
diff --git a/libomptarget/test/offloading/std_complex_arithmetic.cpp b/libomptarget/test/offloading/std_complex_arithmetic.cpp
index 44f87ad..641b849 100644
--- a/libomptarget/test/offloading/std_complex_arithmetic.cpp
+++ b/libomptarget/test/offloading/std_complex_arithmetic.cpp
@@ -6,6 +6,7 @@
// FIXME: This fails to link due to missing math symbols. We should provide the
// needed math functions in the GPU `libm` and require the GPU C library.
// UNSUPPORTED: amdgcn-amd-amdhsa
+// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
#include <cassert>
#include <complex>
diff --git a/libomptarget/test/unified_shared_memory/close_enter_exit.c b/libomptarget/test/unified_shared_memory/close_enter_exit.c
index ec292c8..71be4c9 100644
--- a/libomptarget/test/unified_shared_memory/close_enter_exit.c
+++ b/libomptarget/test/unified_shared_memory/close_enter_exit.c
@@ -4,7 +4,10 @@
// UNSUPPORTED: clang-6, clang-7, clang-8, clang-9
// Fails on amdgpu with error: GPU Memory Error
+// Fails on nvptx with error: an illegal memory access was encountered
// XFAIL: amdgcn-amd-amdhsa
+// XFAIL: nvptx64-nvidia-cuda
+// XFAIL: nvptx64-nvidia-cuda-LTO
#include <omp.h>
#include <stdio.h>
diff --git a/libomptarget/test/unified_shared_memory/close_modifier.c b/libomptarget/test/unified_shared_memory/close_modifier.c
index 47902a9..7ca815d 100644
--- a/libomptarget/test/unified_shared_memory/close_modifier.c
+++ b/libomptarget/test/unified_shared_memory/close_modifier.c
@@ -4,7 +4,10 @@
// UNSUPPORTED: clang-6, clang-7, clang-8, clang-9
// amdgpu runtime crash
+// Fails on nvptx with error: an illegal memory access was encountered
// UNSUPPORTED: amdgcn-amd-amdhsa
+// UNSUPPORTED: nvptx64-nvidia-cuda
+// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
#include <omp.h>
#include <stdio.h>
diff --git a/libomptarget/test/unified_shared_memory/shared_update.c b/libomptarget/test/unified_shared_memory/shared_update.c
index 3923b89..65db9e4 100644
--- a/libomptarget/test/unified_shared_memory/shared_update.c
+++ b/libomptarget/test/unified_shared_memory/shared_update.c
@@ -3,7 +3,10 @@
// REQUIRES: unified_shared_memory
// amdgpu runtime crash
+// Fails on nvptx with error: an illegal memory access was encountered
// UNSUPPORTED: amdgcn-amd-amdhsa
+// UNSUPPORTED: nvptx64-nvidia-cuda
+// UNSUPPORTED: nvptx64-nvidia-cuda-LTO
#include <omp.h>
#include <stdio.h>
