[SYCL][CUDA] Add tests for CUDA-specific pi_mem_advice values (#602)

Add tests for CUDA-specific pi_mem_advice values are being added in intel/llvm#5090.

Author: t4c1

SYCL/DiscardEvents/discard_events_test_queue_ops.hpp

@@ -17,8 +17,9 @@ void TestQueueOperations(sycl::queue Q) {
   sycl::range<1> Range(BUFFER_SIZE);
   auto Dev = Q.get_device();
   auto Ctx = Q.get_context();
-  const int MemAdvice =
-      ((Dev.get_backend() == sycl::backend::ext_oneapi_cuda) ? 1 : 0);
+  const int MemAdvice = ((Dev.get_backend() == sycl::backend::ext_oneapi_cuda)
+                             ? PI_MEM_ADVICE_CUDA_SET_READ_MOSTLY
+                             : PI_MEM_ADVICE_UNKNOWN);
   int *x = sycl::malloc_shared<int>(BUFFER_SIZE, Q);
   assert(x != nullptr);
   int *y = sycl::malloc_shared<int>(BUFFER_SIZE, Q);
@@ -70,8 +71,9 @@ void TestQueueOperationsViaSubmit(sycl::queue Q) {
   sycl::range<1> Range(BUFFER_SIZE);
   auto Dev = Q.get_device();
   auto Ctx = Q.get_context();
-  const int MemAdvice =
-      ((Dev.get_backend() == sycl::backend::ext_oneapi_cuda) ? 1 : 0);
+  const int MemAdvice = ((Dev.get_backend() == sycl::backend::ext_oneapi_cuda)
+                             ? PI_MEM_ADVICE_CUDA_SET_READ_MOSTLY
+                             : PI_MEM_ADVICE_UNKNOWN);
   int *x = sycl::malloc_shared<int>(BUFFER_SIZE, Q);
   assert(x != nullptr);
   int *y = sycl::malloc_shared<int>(BUFFER_SIZE, Q);

SYCL/DiscardEvents/invalid_event.cpp

@@ -23,8 +23,9 @@ void QueueAPIsReturnDiscardedEvent(sycl::queue Q) {
   sycl::range<1> range(BUFFER_SIZE);
 
   auto Dev = Q.get_device();
-  const int MemAdvice =
-      ((Dev.get_backend() == sycl::backend::ext_oneapi_cuda) ? 1 : 0);
+  const int MemAdvice = ((Dev.get_backend() == sycl::backend::ext_oneapi_cuda)
+                             ? PI_MEM_ADVICE_CUDA_SET_READ_MOSTLY
+                             : PI_MEM_ADVICE_UNKNOWN);
   int *x = sycl::malloc_shared<int>(BUFFER_SIZE, Q);
   assert(x != nullptr);
   int *y = sycl::malloc_shared<int>(BUFFER_SIZE, Q);

SYCL/InorderQueue/in_order_usm_implicit.cpp

@@ -26,8 +26,7 @@ int main() {
   {
     queue Queue{property::queue::in_order()};
 
-    // optimize for read only
-    const int mem_advice = 1;
+    const int mem_advice = PI_MEM_ADVICE_CUDA_SET_READ_MOSTLY;
 
     const int dataSize = 32;
     const size_t numBytes = static_cast<size_t>(dataSize) * sizeof(int);

SYCL/USM/memadvise.cpp

@@ -30,8 +30,9 @@ int main() {
   queue q;
   auto dev = q.get_device();
   auto ctxt = q.get_context();
-  const int mem_advice =
-      ((dev.get_backend() == backend::ext_oneapi_cuda) ? 1 : 0);
+  const int mem_advice = ((dev.get_backend() == backend::ext_oneapi_cuda)
+                              ? PI_MEM_ADVICE_CUDA_SET_READ_MOSTLY
+                              : PI_MEM_ADVICE_UNKNOWN);
   if (!dev.get_info<info::device::usm_shared_allocations>())
     return 0;
 

SYCL/USM/memadvise_cuda.cpp

@@ -0,0 +1,57 @@
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t1.out
+// REQUIRES: cuda
+// RUN: %HOST_RUN_PLACEHOLDER %t1.out
+// RUN: %CPU_RUN_PLACEHOLDER %t1.out
+// RUN: %GPU_RUN_PLACEHOLDER %t1.out
+// RUN: %ACC_RUN_PLACEHOLDER %t1.out
+
+//==---------------- memadvise_cuda.cpp ------------------------------------==//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include <CL/sycl.hpp>
+#include <iostream>
+#include <vector>
+
+using namespace cl::sycl;
+
+int main() {
+  const size_t size = 100;
+  queue q;
+  auto dev = q.get_device();
+  auto ctx = q.get_context();
+  if (!dev.get_info<info::device::usm_shared_allocations>()) {
+    std::cout << "Shared USM is not supported. Skipping test." << std::endl;
+    return 0;
+  }
+
+  void *ptr = malloc_shared(size, dev, ctx);
+  if (ptr == nullptr) {
+    std::cout << "Allocation failed!" << std::endl;
+    return -1;
+  }
+
+  std::vector<int> valid_advices{
+      PI_MEM_ADVICE_CUDA_SET_READ_MOSTLY,
+      PI_MEM_ADVICE_CUDA_UNSET_READ_MOSTLY,
+      PI_MEM_ADVICE_CUDA_SET_PREFERRED_LOCATION,
+      PI_MEM_ADVICE_CUDA_UNSET_PREFERRED_LOCATION,
+      PI_MEM_ADVICE_CUDA_SET_ACCESSED_BY,
+      PI_MEM_ADVICE_CUDA_UNSET_ACCESSED_BY,
+      PI_MEM_ADVICE_CUDA_SET_PREFERRED_LOCATION_HOST,
+      PI_MEM_ADVICE_CUDA_UNSET_PREFERRED_LOCATION_HOST,
+      PI_MEM_ADVICE_CUDA_SET_ACCESSED_BY_HOST,
+      PI_MEM_ADVICE_CUDA_UNSET_ACCESSED_BY_HOST,
+  };
+  for (int advice : valid_advices) {
+    q.mem_advise(ptr, size, advice);
+  }
+
+  q.wait_and_throw();
+  std::cout << "Test passed." << std::endl;
+  return 0;
+}