SWDEV-456279 - Adding new hip flag to access contiguous memory and pass the flag to HSA API.

Change-Id: I1bafeaa3096395c729723af958d609bc41e7845c

rocclr/device/rocm/rocdevice.cpp

@@ -2331,7 +2331,8 @@ void Device::deviceVmemRelease(uint64_t mem_handle) const {
   }
 }
 
-void* Device::deviceLocalAlloc(size_t size, bool atomics, bool pseudo_fine_grain) const {
+void* Device::deviceLocalAlloc(size_t size, bool atomics, bool pseudo_fine_grain,
+                               bool contiguous) const {
   const hsa_amd_memory_pool_t& pool = (pseudo_fine_grain) ? gpu_ext_fine_grained_segment_
                                       : (atomics) ? gpu_fine_grained_segment_ : gpuvm_segment_;
 
@@ -2341,6 +2342,11 @@ void* Device::deviceLocalAlloc(size_t size, bool atomics, bool pseudo_fine_grain
     return nullptr;
   }
 
+  uint32_t hsa_mem_flags = 0;
+  if (contiguous) {
+    hsa_mem_flags = HSA_AMD_MEMORY_POOL_CONTIGUOUS_FLAG;
+  }
+
   void* ptr = nullptr;
   hsa_status_t stat = hsa_amd_memory_pool_allocate(pool, size, 0, &ptr);
   ClPrint(amd::LOG_DEBUG, amd::LOG_MEM, "Allocate hsa device memory %p, size 0x%zx", ptr, size);