[SYCL] Disable reduction tests due to sporadic SegFaults in level_zero:gpu (#457)

This patch also eliminates compilation warnings caused by usage of deprecated
type ::half (replaced by sycl::half), and enables some tests for ACC.

Signed-off-by: Vyacheslav N Klochkov <[email protected]>

Author: v-klochkov

SYCL/Reduction/reduction_nd_range_scalar.hpp

@@ -30,6 +30,13 @@ int test(queue &Q, T Identity, T Init, BinaryOperation BOp,
     return 0;
   }
 
+  // TODO: Perhaps, this is a _temporary_ fix for CI. The test may run
+  // for too long when the range is big. That is especially bad on ACC.
+  if (GlobalRange.size() > 65536 && Q.get_device().is_accelerator()) {
+    std::cout << " SKIPPED due to risk of timeout in CI" << std::endl;
+    return 0;
+  }
+
   buffer<T, Dims> InBuf(GlobalRange);
   buffer<T, 1> OutBuf(1);
 

SYCL/Reduction/reduction_nd_s0_dw.cpp

@@ -3,6 +3,9 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(nd_range, reduction, func)
 // with reductions initialized with 0-dimensional discard_write accessor.
 

SYCL/Reduction/reduction_nd_s0_rw.cpp

@@ -3,6 +3,9 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(nd_range, reduction, func)
 // with reductions initialized with 0-dimensional read_write accessor.
 

SYCL/Reduction/reduction_nd_s1_dw.cpp

@@ -4,6 +4,9 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(nd_range, reduction, func)
 // with reductions initialized with 1-dimensional discard_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_nd_s1_rw.cpp

@@ -3,6 +3,9 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(nd_range, reduction, func)
 // with reductions initialized with 1-dimensional read_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_range_1d_s0_dw.cpp

@@ -6,6 +6,9 @@
 // implementation. Enable testing when implementation is fixed.
 // RUNx: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(range<1>, reduction, func)
 // with reductions initialized with 0-dimensional discard_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_range_1d_s0_rw.cpp

@@ -6,6 +6,9 @@
 // implementation. Enable testing when implementation is fixed.
 // RUNx: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(range<1>, reduction, func)
 // with reductions initialized with 0-dimensional read_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_range_1d_s1_dw.cpp

@@ -3,6 +3,9 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 #include "reduction_range_scalar.hpp"
 
 // This test performs basic checks of parallel_for(range<1>, reduction, func)

SYCL/Reduction/reduction_range_1d_s1_rw.cpp

@@ -3,6 +3,9 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(range<1>, reduction, func)
 // with reductions initialized with 1-dimensional read_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_range_2d_s1_dw.cpp

@@ -6,6 +6,9 @@
 // implementation. Enable testing when implementation is fixed.
 // RUNx: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(range<2>, reduction, func)
 // with reductions initialized with 1-dimensional discard_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_range_2d_s1_rw.cpp

@@ -6,6 +6,9 @@
 // implementation. Enable testing when implementation is fixed.
 // RUNx: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(range<2>, reduction, func)
 // with reductions initialized with 1-dimensional discard_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_range_3d_s1_dw.cpp

@@ -6,6 +6,9 @@
 // implementation. Enable testing when implementation is fixed.
 // RUNx: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(range<3>, reduction, func)
 // with reductions initialized with 1-dimensional discard_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_range_3d_s1_rw.cpp

@@ -6,6 +6,9 @@
 // implementation. Enable testing when implementation is fixed.
 // RUNx: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(range<3>, reduction, func)
 // with reductions initialized with 1-dimensional discard_write accessor
 // accessing 1 element buffer.

SYCL/Reduction/reduction_range_lambda.cpp

@@ -1,10 +1,7 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
-
-// TODO: accelerator may not suport atomics required by the current
-// implementation. Enable testing when implementation is fixed.
-// RUNx: %ACC_RUN_PLACEHOLDER %t.out
+// RUN: %ACC_RUN_PLACEHOLDER %t.out
 
 // This test performs basic checks of parallel_for(range, reduction, lambda)
 // with reductions initialized with 1-dimensional accessor accessing

SYCL/Reduction/reduction_range_scalar.hpp

@@ -32,7 +32,7 @@ int test(queue &Q, T Identity, T Init, BinaryOperation BOp,
 
   // TODO: Perhaps, this is a _temporary_ fix for CI. The test may run
   // for too long when the range is big. That is especially bad on ACC.
-  if (Range.size() > 65536) {
+  if (Range.size() > 65536 && Q.get_device().is_accelerator()) {
     std::cout << " SKIPPED due to risk of timeout in CI" << std::endl;
     return 0;
   }

SYCL/Reduction/reduction_range_usm_dw.cpp

@@ -1,10 +1,10 @@
 // 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
 
-// TODO: accelerator may not suport atomics required by the current
-// implementation. Enable testing when implementation is fixed.
-// RUNx: %ACC_RUN_PLACEHOLDER %t.out
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
 
 #include "reduction_range_scalar.hpp"
 

SYCL/Reduction/reduction_usm.cpp

@@ -3,6 +3,9 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
+// UNSUPPORTED: linux && level_zero
+
 // This test performs basic checks of parallel_for(nd_range, reduction, func)
 // with reductions initialized with USM var. It tests both
 // ext::oneapi::reduction and SYCL-2020 reduction (sycl::reduction) assuming

SYCL/Reduction/reduction_usm_dw.cpp

@@ -2,8 +2,8 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
-// TODO test disabled due to Jenkins testing failure on unrelated commit
-// failed on Linux level_zero only
+
+// TODO: test disabled due to sporadic fails in level_zero:gpu RT.
 // UNSUPPORTED: linux && level_zero
 
 // This test performs basic checks of parallel_for(nd_range, reduction, func)

SYCL/Reduction/reduction_utils.hpp

@@ -195,7 +195,7 @@ void printTestLabel(bool IsSYCL2020, const RangeT &Range, bool ToCERR = false) {
 }
 
 template <typename BOp, typename T> constexpr bool isPreciseResultFP() {
-  return (std::is_floating_point_v<T> || std::is_same_v<T, half>)&&(
+  return (std::is_floating_point_v<T> || std::is_same_v<T, sycl::half>)&&(
       std::is_same_v<ext::oneapi::minimum<>, BOp> ||
       std::is_same_v<ext::oneapi::minimum<T>, BOp> ||
       std::is_same_v<ext::oneapi::maximum<>, BOp> ||
@@ -209,7 +209,7 @@ int checkResults(queue &Q, bool IsSYCL2020, BinaryOperation,
   std::string ErrorStr;
   bool Passed;
 
-  if constexpr (std::is_floating_point_v<T> || std::is_same_v<T, half>) {
+  if constexpr (std::is_floating_point_v<T> || std::is_same_v<T, sycl::half>) {
     // It is a pretty simple and naive FP diff check here, which though
     // should work reasonably well for most of cases handled in reduction
     // tests.