[SYCL][L0] Skip uninitialized buffer migration copy (#7168)

Signed-off-by: Sergey V Maslov <[email protected]>

Author: smaslov-intel

sycl/plugins/level_zero/pi_level_zero.cpp

@@ -3127,8 +3127,7 @@ pi_result piDeviceGetInfo(pi_device Device, pi_device_info ParamName,
 
     // intel extensions for GPU information
   case PI_DEVICE_INFO_DEVICE_ID:
-    return ReturnValue(
-        pi_uint32{Device->ZeDeviceProperties->deviceId});
+    return ReturnValue(pi_uint32{Device->ZeDeviceProperties->deviceId});
   case PI_DEVICE_INFO_PCI_ADDRESS: {
     if (getenv("ZES_ENABLE_SYSMAN") == nullptr) {
       zePrint("Set SYCL_ENABLE_PCI=1 to obtain PCI data.\n");
@@ -3958,8 +3957,12 @@ pi_result piMemBufferCreate(pi_context Context, pi_mem_flags Flags, size_t Size,
     if ((Flags & PI_MEM_FLAGS_HOST_PTR_USE) != 0 ||
         (Flags & PI_MEM_FLAGS_HOST_PTR_COPY) != 0) {
 
+      // We don't yet know which device needs this buffer, so make the first
+      // device in the context be the master, and hold the initial valid
+      // allocation.
       char *ZeHandleDst;
-      PI_CALL(Buffer->getZeHandle(ZeHandleDst, _pi_mem::write_only));
+      PI_CALL(Buffer->getZeHandle(ZeHandleDst, _pi_mem::write_only,
+                                  Context->Devices[0]));
       if (Buffer->OnHost) {
         // Do a host to host copy.
         // For an imported HostPtr the copy is unneeded.
@@ -4331,7 +4334,7 @@ pi_result piextMemCreateWithNativeHandle(pi_native_handle NativeHandle,
     // represent the buffer in this context) copy the data to a newly
     // created device allocation.
     char *ZeHandleDst;
-    PI_CALL(Buffer->getZeHandle(ZeHandleDst, _pi_mem::write_only));
+    PI_CALL(Buffer->getZeHandle(ZeHandleDst, _pi_mem::write_only, Device));
 
     // zeCommandListAppendMemoryCopy must not be called from simultaneous
     // threads with the same command list handle, so we need exclusive lock.
@@ -8757,9 +8760,16 @@ size_t _pi_buffer::getAlignment() const {
 pi_result _pi_buffer::getZeHandle(char *&ZeHandle, access_mode_t AccessMode,
                                   pi_device Device) {
 
+  // NOTE: There might be no valid allocation at all yet and we get
+  // here from piEnqueueKernelLaunch that would be doing the buffer
+  // initialization. In this case the Device is not null as kernel
+  // launch is always on a specific device.
   if (!Device)
     Device = LastDeviceWithValidAllocation;
-  PI_ASSERT(Device, PI_ERROR_INVALID_DEVICE);
+  // If the device is still not selected then use the first one in
+  // the context of the buffer.
+  if (!Device)
+    Device = Context->Devices[0];
 
   auto &Allocation = Allocations[Device];
 
@@ -8814,6 +8824,9 @@ pi_result _pi_buffer::getZeHandle(char *&ZeHandle, access_mode_t AccessMode,
       //       devices in the context have the same root.
       PI_CALL(getZeHandle(ZeHandle, AccessMode, Context->SingleRootDevice));
       Allocation.ReleaseAction = allocation_t::keep;
+      Allocation.ZeHandle = ZeHandle;
+      Allocation.Valid = true;
+      return PI_SUCCESS;
     } else { // Create device allocation
       if (enableBufferPooling()) {
         Allocation.ReleaseAction = allocation_t::free;
@@ -8836,13 +8849,27 @@ pi_result _pi_buffer::getZeHandle(char *&ZeHandle, access_mode_t AccessMode,
     if (Device == LastDeviceWithValidAllocation)
       die("getZeHandle: last used allocation is not valid");
 
-    Allocation.Valid = true;
     // For write-only access the allocation contents is not going to be used.
     // So don't do anything to make it "valid".
-    if (AccessMode != _pi_mem::write_only) {
-      char *ZeHandleSrc;
+    bool NeedCopy = AccessMode != _pi_mem::write_only;
+    // It's also possible that the buffer doesn't have a valid allocation
+    // yet presumably when it is passed to a kernel that will perform
+    // it's intialization.
+    if (NeedCopy && !LastDeviceWithValidAllocation) {
+      NeedCopy = false;
+    }
+    char *ZeHandleSrc = nullptr;
+    if (NeedCopy) {
       PI_CALL(getZeHandle(ZeHandleSrc, _pi_mem::read_only,
                           LastDeviceWithValidAllocation));
+      // It's possible with the single root-device contexts that
+      // the buffer is represented by the single root-device
+      // allocation and then skip the copy to itself.
+      if (ZeHandleSrc == ZeHandle)
+        NeedCopy = false;
+    }
+
+    if (NeedCopy) {
       // Copy valid buffer data to this allocation.
       // TODO: see if we should better use peer's device allocation used
       // directly, if that capability is reported with zeDeviceCanAccessPeer,
@@ -8851,7 +8878,7 @@ pi_result _pi_buffer::getZeHandle(char *&ZeHandle, access_mode_t AccessMode,
       //
       // zeCommandListAppendMemoryCopy must not be called from simultaneous
       // threads with the same command list handle, so we need exclusive lock.
-      ze_bool_t P2P;
+      ze_bool_t P2P = false;
       ZE_CALL(
           zeDeviceCanAccessPeer,
           (Device->ZeDevice, LastDeviceWithValidAllocation->ZeDevice, &P2P));
@@ -8895,9 +8922,10 @@ pi_result _pi_buffer::getZeHandle(char *&ZeHandle, access_mode_t AccessMode,
                  Size, nullptr, 0, nullptr));
       }
     }
+    Allocation.Valid = true;
+    LastDeviceWithValidAllocation = Device;
   }
 
-  LastDeviceWithValidAllocation = Device;
   // Invalidate other allocations that would become not valid if
   // this access is not read-only.
   if (AccessMode != _pi_mem::read_only) {

sycl/plugins/level_zero/pi_level_zero.hpp

@@ -1131,12 +1131,8 @@ struct _pi_buffer final : _pi_mem {
       }
     }
 
-    // Make first device in the context be the master. Mark that
-    // allocation (yet to be made) having "valid" data. And real
-    // allocation and initialization should follow the buffer
-    // construction with a "write_only" access copy.
-    LastDeviceWithValidAllocation = Context->Devices[0];
-    Allocations[LastDeviceWithValidAllocation].Valid = true;
+    // This initialization does not end up with any valid allocation yet.
+    LastDeviceWithValidAllocation = nullptr;
   }
 
   // Sub-buffer constructor