[SYCL] Cleanup Windows testing (#17755)

This is a joined cherry-pick of #16286, #17046 and
#17100 made to align tests between `sycl` and `sycl-rel-6_1_0`
branches to make the pre-commit more stable on the latter.

Descriptions of cherry-picked commits:

---

Fix postcommit failure and move tests in right folder (#16286)

opencl cpu fails without a stack trace

Patch-By: Victor Lomuller <[email protected]>

---

[SYCL][E2E] Limit work group size in WorkGroupScratchMemory tests
(#17046)

Some devices don't support work group sizes of 1024, so use the maximum
size if it is smaller in the copy_dynamic_size.cpp and
dynamic_unused.cpp tests.

Patch-By: Ross Brunton <[email protected]>

---

[SYCL][E2E] Use size_t rather than unsigned long in
WorkGroupScratchMemory tests (#17100)

The max work group size is a `size_t`, rather than an `unsigned long`.
This trips up older versions of MSVC where `size_t` happens to be
`unsigned long long`.

Patch-By: Ross Brunton <[email protected]>

---

Author: AlexeySachkov

sycl/test-e2e/WorkGroupMemory/Dynamic/copy_dynamic_size.cpp renamed to sycl/test-e2e/WorkGroupScratchMemory/copy_dynamic_size.cpp

@@ -10,7 +10,6 @@
 #include <sycl/group_barrier.hpp>
 #include <sycl/usm.hpp>
 
-constexpr size_t Size = 1024;
 using DataType = int;
 
 namespace sycl_ext = sycl::ext::oneapi::experimental;
@@ -31,23 +30,27 @@ void copy_via_smem(DataType *a, DataType *b, sycl::nd_item<1> it) {
 
 int main() {
   sycl::queue queue;
-  DataType *a = sycl::malloc_device<DataType>(Size, queue);
-  DataType *b = sycl::malloc_device<DataType>(Size, queue);
-  std::vector<DataType> a_host(Size, 1.0);
-  std::vector<DataType> b_host(Size, -5.0);
+  auto size = std::min(
+      queue.get_device().get_info<sycl::info::device::max_work_group_size>(),
+      size_t{1024});
 
-  queue.copy(a_host.data(), a, Size).wait_and_throw();
+  DataType *a = sycl::malloc_device<DataType>(size, queue);
+  DataType *b = sycl::malloc_device<DataType>(size, queue);
+  std::vector<DataType> a_host(size, 1.0);
+  std::vector<DataType> b_host(size, -5.0);
+
+  queue.copy(a_host.data(), a, size).wait_and_throw();
 
   queue
       .submit([&](sycl::handler &cgh) {
-        cgh.parallel_for(sycl::nd_range<1>({Size}, {Size}),
+        cgh.parallel_for(sycl::nd_range<1>({size}, {size}),
                          sycl_ext::properties{sycl_ext::work_group_scratch_size(
-                             Size * sizeof(DataType))},
+                             size * sizeof(DataType))},
                          [=](sycl::nd_item<1> it) { copy_via_smem(a, b, it); });
       })
       .wait_and_throw();
 
-  queue.copy(b, b_host.data(), Size).wait_and_throw();
+  queue.copy(b, b_host.data(), size).wait_and_throw();
   for (size_t i = 0; i < b_host.size(); i++) {
     assert(b_host[i] == a_host[i]);
   }

sycl/test-e2e/WorkGroupMemory/Dynamic/dynamic_alloc_local_accessor.cpp renamed to sycl/test-e2e/WorkGroupScratchMemory/dynamic_alloc_local_accessor.cpp

@@ -2,7 +2,7 @@
 // RUN: %{run} %t.out
 //
 
-// UNSUPPORTED: gpu-intel-gen12
+// UNSUPPORTED: gpu-intel-gen12, cpu
 // UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/16072
 
 // Test work_group_dynamic extension with allocation size specified at runtime

sycl/test-e2e/WorkGroupMemory/Dynamic/dynamic_alloc_ptr_alias.cpp renamed to sycl/test-e2e/WorkGroupScratchMemory/dynamic_alloc_ptr_alias.cpp

@@ -2,7 +2,7 @@
 // RUN: %{run} %t.out
 //
 
-// UNSUPPORTED: gpu-intel-gen12
+// UNSUPPORTED: gpu-intel-gen12, cpu
 // UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/16072
 
 // Test work_group_dynamic extension with allocation size specified at runtime

sycl/test-e2e/WorkGroupMemory/Dynamic/dynamic_allocation.cpp renamed to sycl/test-e2e/WorkGroupScratchMemory/dynamic_allocation.cpp

@@ -2,7 +2,7 @@
 // RUN: %{run} %t.out
 //
 
-// UNSUPPORTED: gpu-intel-gen12
+// UNSUPPORTED: gpu-intel-gen12, cpu
 // UNSUPPORTED-TRACKER: https://github.com/intel/llvm/issues/16072
 
 // Test work_group_dynamic extension with allocation size specified at runtime.

sycl/test-e2e/WorkGroupMemory/Dynamic/dynamic_unused.cpp renamed to sycl/test-e2e/WorkGroupScratchMemory/dynamic_unused.cpp

@@ -9,33 +9,36 @@
 #include <sycl/ext/oneapi/work_group_scratch_memory.hpp>
 #include <sycl/usm.hpp>
 
-constexpr size_t Size = 1024;
 using DataType = int;
 
 namespace sycl_ext = sycl::ext::oneapi::experimental;
 
 int main() {
   sycl::queue queue;
-  DataType *a = sycl::malloc_device<DataType>(Size, queue);
-  DataType *b = sycl::malloc_device<DataType>(Size, queue);
-  std::vector<DataType> a_host(Size, 1.0);
-  std::vector<DataType> b_host(Size, -5.0);
+  auto size = std::min(
+      queue.get_device().get_info<sycl::info::device::max_work_group_size>(),
+      size_t{1024});
 
-  queue.copy(a_host.data(), a, Size).wait_and_throw();
+  DataType *a = sycl::malloc_device<DataType>(size, queue);
+  DataType *b = sycl::malloc_device<DataType>(size, queue);
+  std::vector<DataType> a_host(size, 1.0);
+  std::vector<DataType> b_host(size, -5.0);
+
+  queue.copy(a_host.data(), a, size).wait_and_throw();
 
   queue
       .submit([&](sycl::handler &cgh) {
-        cgh.parallel_for(sycl::nd_range<1>({Size}, {Size}),
+        cgh.parallel_for(sycl::nd_range<1>({size}, {size}),
                          sycl_ext::properties{sycl_ext::work_group_scratch_size(
-                             Size * sizeof(DataType))},
+                             size * sizeof(DataType))},
                          [=](sycl::nd_item<1> it) {
                            b[it.get_local_linear_id()] =
                                a[it.get_local_linear_id()];
                          });
       })
       .wait_and_throw();
 
-  queue.copy(b, b_host.data(), Size).wait_and_throw();
+  queue.copy(b, b_host.data(), size).wait_and_throw();
   for (size_t i = 0; i < b_host.size(); i++) {
     assert(b_host[i] == a_host[i]);
   }

sycl/test-e2e/WorkGroupMemory/lit.local.cfg renamed to sycl/test-e2e/WorkGroupScratchMemory/lit.local.cfg

@@ -0,0 +1,3 @@
+
+# https://github.com/intel/llvm/issues/16072
+config.unsupported_features += ['hip']