Use MemoryAccessFlags instead of MemoryAccessType when binding (pytorch#5591)

Summary:
Pull Request resolved: pytorch#5591

## Context

Correct `ComputeGraph` functions to accept `MemoryAccessFlags` instead of `MemoryAccessType`. `MemoryAccessType` correspond to only a single bit, i.e. `READ` or `WRITE` but `MemoryAccessFlags` allows us to express a combination of bits which is the intended behaviour.
ghstack-source-id: 244431148
exported-using-ghexport

Reviewed By: jorgep31415

Differential Revision: D63327080

fbshipit-source-id: a4ca6b9f6dfb9a998932b934a2a230fbb3d7c06a

Author: SS-JIA

backends/vulkan/runtime/graph/ops/ExecuteNode.h

@@ -21,16 +21,16 @@ class ComputeGraph;
  * access permission.
  */
 struct ArgGroup {
-  ArgGroup(const ValueRef ref, const vkapi::MemoryAccessType access)
+  ArgGroup(const ValueRef ref, const vkapi::MemoryAccessFlags access)
       : refs{ref}, access(access) {}
 
   ArgGroup(
       const std::vector<ValueRef>& refs,
-      const vkapi::MemoryAccessType access)
+      const vkapi::MemoryAccessFlags access)
       : refs(refs), access(access) {}
 
   const std::vector<ValueRef> refs;
-  const vkapi::MemoryAccessType access;
+  const vkapi::MemoryAccessFlags access;
 };
 
 /*

backends/vulkan/runtime/graph/ops/utils/BindingUtils.cpp

@@ -13,7 +13,7 @@ namespace vkcompute {
 void bind_tensor_to_descriptor_set(
     api::vTensor& tensor,
     vkapi::PipelineBarrier& pipeline_barrier,
-    const vkapi::MemoryAccessType accessType,
+    const vkapi::MemoryAccessFlags accessType,
     vkapi::DescriptorSet& descriptor_set,
     const uint32_t idx) {
   if (tensor.buffer()) {

backends/vulkan/runtime/graph/ops/utils/BindingUtils.h

@@ -19,7 +19,7 @@ namespace vkcompute {
 void bind_tensor_to_descriptor_set(
     api::vTensor& tensor,
     vkapi::PipelineBarrier& pipeline_barrier,
-    const vkapi::MemoryAccessType accessType,
+    const vkapi::MemoryAccessFlags accessType,
     vkapi::DescriptorSet& descriptor_set,
     const uint32_t idx);