From 6bb7d1afdc776be5edcc9a02dda20a0297046eff Mon Sep 17 00:00:00 2001
From: German Andryeyev <German.Andryeyev@amd.com>
Date: Fri, 18 Oct 2024 09:46:24 -0400
Subject: [PATCH] SWDEV-486602 - Fix Windows 32 bit build

Windows alings fields to 8 bytes even with 32bit builds.
Add BUG_CLR_SYSMEM_POOL to cotnrol sysmempool.

Change-Id: I8622aabc9f7391ed7dd8583b252ce9eb41d62293
---
 rocclr/platform/command.cpp | 12 ++++++++++--
 rocclr/platform/object.hpp  | 10 ++++++----
 rocclr/utils/flags.hpp      |  2 ++
 3 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/rocclr/platform/command.cpp b/rocclr/platform/command.cpp
index 6ad49c43e4..363012e607 100644
--- a/rocclr/platform/command.cpp
+++ b/rocclr/platform/command.cpp
@@ -321,12 +321,20 @@ Command::Command(HostQueue& queue, cl_command_type type, const EventWaitList& ev
 SysmemPool<ComputeCommand>* Command::command_pool_ = new SysmemPool<ComputeCommand>;
 // ================================================================================================
 void Command::operator delete(void* ptr) {
-  return command_pool_->Free(ptr);
+  if (DEBUG_CLR_SYSMEM_POOL) {
+    command_pool_->Free(ptr);
+  } else {
+    ::operator delete (ptr);
+  }
 }
 
 // ================================================================================================
 void* Command::operator new(size_t size) {
-  return command_pool_->Alloc(size);
+  if (DEBUG_CLR_SYSMEM_POOL) {
+    return command_pool_->Alloc(size);
+  } else {
+    return ::operator new (size);
+  }
 }
 
 // ================================================================================================
diff --git a/rocclr/platform/object.hpp b/rocclr/platform/object.hpp
index dece674ea5..f6f5f1a99f 100644
--- a/rocclr/platform/object.hpp
+++ b/rocclr/platform/object.hpp
@@ -256,14 +256,17 @@ public:
     // Save the chunk allocation
     obj->base_ = active_allocs_[chunk_idx]->base_;
     obj->base_->busy_++;
-    assert((reinterpret_cast<address>(&obj->object_) -
-            reinterpret_cast<address>(obj)) == sizeof(AllocChunk*) && "Incorrect mem obj offset!");
     return &obj->object_;
   }
 
   void Free(void* ptr) {
+#if IS_WINDOWS
     auto obj = reinterpret_cast<MemoryObject*>(
-      reinterpret_cast<address>(ptr) - sizeof(MemoryObject*));
+      reinterpret_cast<address>(ptr) - offsetof(MemoryObject, object_));
+#else
+    auto obj = reinterpret_cast<MemoryObject*>(
+      reinterpret_cast<address>(ptr) - sizeof(AllocChunk*));
+#endif
     auto freed = --obj->base_->free_;
     // If it's the last slot in the chunk, then release memory
     if (freed == 0) {
@@ -300,7 +303,6 @@ private:
     ~AllocChunk() { delete [] allocs_; }
   };
 
-
   std::atomic<uint64_t> current_alloc_ = 0; //!< Current allocation, global index
   std::atomic<size_t> max_chunk_idx_ = 0;   //!< Current max chunk index
   size_t  free_chunk_num_ = 0;              //!< The number of freed chunks
diff --git a/rocclr/utils/flags.hpp b/rocclr/utils/flags.hpp
index 5d39181b8f..9fc958809d 100644
--- a/rocclr/utils/flags.hpp
+++ b/rocclr/utils/flags.hpp
@@ -261,6 +261,8 @@ release(uint, DEBUG_HIP_BLOCK_SYNC, 50,                                       \
         "Blocks synchronization on CPU until the callback processing is done")\
 release(uint, DEBUG_CLR_MAX_BATCH_SIZE, 1000,                                 \
         "Forces the callback to clean-up CPU submission queue")               \
+release(bool, DEBUG_CLR_SYSMEM_POOL, true,                                    \
+        "Use sysmem pool implementation in runtime for amd commands")         \
 release(bool, DEBUG_HIP_KERNARG_COPY_OPT, true,                               \
         "Enable/Disable multiple kern arg copies")                            \
 release(bool, DEBUG_CLR_USE_STDMUTEX_IN_AMD_MONITOR, false,                   \