Merge pull request oneapi-src#1906 from nrspruit/flex_gpu_copy_engine

[L0] Add check for Intel Flex/Arc for disabling use of copy engines.

Author: Omar Ahmed

source/adapters/level_zero/command_buffer.cpp

@@ -921,6 +921,13 @@ UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendUSMMemcpyExp(
 
   bool PreferCopyEngine = !IsDevicePointer(CommandBuffer->Context, Src) ||
                           !IsDevicePointer(CommandBuffer->Context, Dst);
+  // For better performance, Copy Engines are not preferred given Shared
+  // pointers on DG2.
+  if (CommandBuffer->Device->isDG2() &&
+      (IsSharedPointer(CommandBuffer->Context, Src) ||
+       IsSharedPointer(CommandBuffer->Context, Dst))) {
+    PreferCopyEngine = false;
+  }
   PreferCopyEngine |= UseCopyEngineForD2DCopy;
 
   return enqueueCommandBufferMemCopyHelper(

source/adapters/level_zero/device.hpp

@@ -189,6 +189,9 @@ struct ur_device_handle_t_ : _ur_object {
            (ZeDeviceProperties->deviceId & 0xff0) == 0xb60;
   }
 
+  // Checks if this GPU is an Intel Flex GPU or Intel Arc Alchemist
+  bool isDG2() { return (ZeDeviceProperties->deviceId & 0xff00) == 0x5600; }
+
   bool isIntegrated() {
     return (ZeDeviceProperties->flags & ZE_DEVICE_PROPERTY_FLAG_INTEGRATED);
   }

source/adapters/level_zero/kernel.cpp

@@ -613,6 +613,11 @@ ur_result_t ur_queue_handle_legacy_t_::enqueueDeviceGlobalVariableWrite(
   // Copy engine is preferred only for host to device transfer.
   // Device to device transfers run faster on compute engines.
   bool PreferCopyEngine = !IsDevicePointer(Queue->Context, Src);
+  // For better performance, Copy Engines are not preferred given Shared
+  // pointers on DG2.
+  if (Queue->Device->isDG2() && IsSharedPointer(Queue->Context, Src)) {
+    PreferCopyEngine = false;
+  }
 
   // Temporary option added to use copy engine for D2D copy
   PreferCopyEngine |= UseCopyEngineForD2DCopy;
@@ -663,6 +668,11 @@ ur_result_t ur_queue_handle_legacy_t_::enqueueDeviceGlobalVariableRead(
   // Copy engine is preferred only for host to device transfer.
   // Device to device transfers run faster on compute engines.
   bool PreferCopyEngine = !IsDevicePointer(Queue->Context, Dst);
+  // For better performance, Copy Engines are not preferred given Shared
+  // pointers on DG2.
+  if (Queue->Device->isDG2() && IsSharedPointer(Queue->Context, Dst)) {
+    PreferCopyEngine = false;
+  }
 
   // Temporary option added to use copy engine for D2D copy
   PreferCopyEngine |= UseCopyEngineForD2DCopy;

source/adapters/level_zero/memory.cpp

@@ -42,6 +42,19 @@ bool IsDevicePointer(ur_context_handle_t Context, const void *Ptr) {
   return (ZeMemoryAllocationProperties.type == ZE_MEMORY_TYPE_DEVICE);
 }
 
+// Helper function to check if a pointer is a shared pointer.
+bool IsSharedPointer(ur_context_handle_t Context, const void *Ptr) {
+  ze_device_handle_t ZeDeviceHandle;
+  ZeStruct<ze_memory_allocation_properties_t> ZeMemoryAllocationProperties;
+
+  // Query memory type of the pointer
+  ZE2UR_CALL(zeMemGetAllocProperties,
+             (Context->ZeContext, Ptr, &ZeMemoryAllocationProperties,
+              &ZeDeviceHandle));
+
+  return (ZeMemoryAllocationProperties.type == ZE_MEMORY_TYPE_SHARED);
+}
+
 // Shared by all memory read/write/copy PI interfaces.
 // PI interfaces must have queue's and destination buffer's mutexes locked for
 // exclusive use and source buffer's mutex locked for shared use on entry.
@@ -1191,6 +1204,12 @@ ur_result_t ur_queue_handle_legacy_t_::enqueueUSMMemcpy(
   // (versus compute engine).
   bool PreferCopyEngine = !IsDevicePointer(Queue->Context, Src) ||
                           !IsDevicePointer(Queue->Context, Dst);
+  // For better performance, Copy Engines are not preferred given Shared
+  // pointers on DG2.
+  if (Queue->Device->isDG2() && (IsSharedPointer(Queue->Context, Src) ||
+                                 IsSharedPointer(Queue->Context, Dst))) {
+    PreferCopyEngine = false;
+  }
 
   // Temporary option added to use copy engine for D2D copy
   PreferCopyEngine |= UseCopyEngineForD2DCopy;
@@ -1390,6 +1409,12 @@ ur_result_t ur_queue_handle_legacy_t_::enqueueUSMMemcpy2D(
   // (versus compute engine).
   bool PreferCopyEngine = !IsDevicePointer(Queue->Context, Src) ||
                           !IsDevicePointer(Queue->Context, Dst);
+  // For better performance, Copy Engines are not preferred given Shared
+  // pointers on DG2.
+  if (Queue->Device->isDG2() && (IsSharedPointer(Queue->Context, Src) ||
+                                 IsSharedPointer(Queue->Context, Dst))) {
+    PreferCopyEngine = false;
+  }
 
   // Temporary option added to use copy engine for D2D copy
   PreferCopyEngine |= UseCopyEngineForD2DCopy;

source/adapters/level_zero/memory.hpp

@@ -32,6 +32,7 @@ using ur_queue_handle_legacy_t = ur_queue_handle_legacy_t_ *;
 struct ur_device_handle_t_;
 
 bool IsDevicePointer(ur_context_handle_t Context, const void *Ptr);
+bool IsSharedPointer(ur_context_handle_t Context, const void *Ptr);
 
 // This is an experimental option to test performance of device to device copy
 // operations on copy engines (versus compute engine)